ISSN 0002-9920 (print)

ISSN 1088-9477 (online)

of the American Mathematical Society

August 2025 Volume 72, Number 7

The cover design is based on imagery

from “Apropos the Long-Lost Letter from
Levinson to Smale, Now Found!”, page 748.
 AMS
BACKSTOP GRANTS
For groups, societies, and programs affected by federal
funding cuts.

Together we do more to elevate mathematics.

Learn more:
www.ams.org/ams-backstop-grants
 August 2025
Cover Credit: The image used in the cover
design was taken by J. Koiller at MAST.

Feature: Quadratic Forms Beyond Arithmetic............711

Alexander Merkurjev and Raman Parimala
FROM THE
Feature: Topology Meets Machine Learning:
An Introduction Using the Euler Characteristic AMS SECRETARY
Transform.....................................................................719
Bastian Rieck
Calls for Nominations & Applications......................... 775
Career: Advice Column............................................... 728
Veselin Jungic State of the AMS...........................................................776
Memorial Tribute: Bernard Maskit Memorial
Tribute.......................................................................... 731
James W. Anderson, Ara Basmajian, Ruben Hidalgo,
Perry Susskind, and Edward C. Taylor
Education: Teaching Mathematics with Lean: INVITATIONS
Interactive Theorem Provers in the Classroom..........740 FROM THE AMS
Paola Iannone, Gihan Marasingha, and Athina Thoma

History: Apropos the Long-Lost Letter from

Levinson to Smale, Now Found!.................................748 Backstop Grants.................................inside front cover
Jair Koiller and Indika Rajapakse
Ergodic Theory............................................................... 727
Bookshelf..................................................................... 756
Emily J. Olson Rate Options for Unemployed Members................... 729

Conversation: Challenges and Opportunities Centennial Fellowship................................................ 730

in Peer Review............................................................. 757 Invariants of Links and 3-Manifolds from Graph
Lance Fortnow, Robert Harington, Igor Pak, Anne Schilling, Configurations................................................................747
and Mark C. Wilson
Members, Are You Relocating?................................... 755
Conversation: Guidelines for Writing and
Reviewing Mathematical Software..............................766 Student Travel Grants...................................................774
Jeroen Hanselman
Stefan Bergman Fellowship........................................ 778
AMS Communication: CASE Workshop Inspires
Students to Advocate for Mathematics...................... 779 Join the AMS................................................................ 804
Elaine Beebe
Join the Fiske Society..........................inside back cover
AMS Communication: Julia Robinson
Mathematics Festival a Hit in Providence.................. 782 Inequalities in Matrix Algebras..........................back cover
Elaine Beebe

News: AMS Updates.................................................... 784

News: Mathematics People......................................... 786

New Books Offered by the AMS.................................. 788

Meetings & Conferences of the AMS.......................... 795

 Notices
of the American Mathematical Society

INTERIM EDITOR-IN-CHIEF SUBSCRIPTION INFORMATION

Joseph H. Silverman For subscription information, please visit:
https://www.ams.org/publications/journals/subscriberinfo.
Subscription renewals are subject to late fees. See:
ASSOCIATE EDITORS https://www.ams.org/journal-faq
Daniela De Silva Boris Hasselblatt, ex officio for more journal subscription information.
Yvonne Lai Richard A. Levine
Haydee Lindo Reza Malek-Madani ADVERTISING
Han-Bom Moon Vidit Nanda Notices publishes situations wanted and classified advertising, and display
Emily J. Olson Emilie Purvine advertising for publishers and academic or scientific organizations.
Advertising requests, materials, and/or questions should be sent to:
Adrian Rice Jose Rodriguez
[email protected] (classified ads)
Laura P. Schaposnik Krystal Taylor [email protected] (display ads).

ASSISTANT TO THE EDITOR-IN-CHIEF PERMISSIONS

Karen Wargo All requests to reprint Notices articles should be sent to:
[email protected].

CONSULTANTS SUBMISSIONS
Travis Dillon Robert Harington Bryna Kra The editor-in-chief should be contacted about articles for
Ken Ono Kenneth A. Ribet Richard Stanley consideration after potential authors have reviewed the
Ada Stelzer Ravi Vakil Bianca Viray “Information for Notices Authors” tab at:
https://www.ams.org/noticesauthors.
MANAGING EDITOR The managing editor should be contacted for additions to our
Meaghan Healy news sections and for any questions or corrections.
Contact the managing editor at:
[email protected].
CONTRIBUTING WRITER
Elaine Beebe Letters to the editor should be sent to:
[email protected].

COMPOSITION, DESIGN, and EDITING To make suggestions for additions to other sections,
and for full contact information, see:
Brian Bartling John F. Brady Craig Dujon https://www.ams.org/noticescontact.
Teresa McClure Lori Nero Dan Normand
John C. Paul Miriam Schaerf Mike Southern
Peter Sykes

Supported by the AMS membership, this publication is freely available The print version is a privilege of Membership. Graduate students at member
electronically through the AMS website, the Society’s resource for delivering institutions can opt to receive the print journal by updating their individual
electronic products and services. Use the URL: https://www.ams.org/notices member profiles at: https://www.ams.org/member-directory. For questions
to access the Notices on the website. The online version of the Notices is the regarding updating your profile, please call 800-321-4267.
version of record, so it may occasionally differ slightly from the print version.
For back issues see: https://www.ams.org/backvols. Note: Single issues of
the Notices are not available after one calendar year.

The American Mathematical Society is committed to promoting and facilitating equity, diversity and inclusion throughout the mathematical sciences. For
its own long-term prosperity as well as that of the public at large, our discipline must connect with and appropriately incorporate all sectors of society. We
reaffirm the pledge in the AMS Mission Statement to “advance the status of the profession of mathematics, encouraging and facilitating full participation of
all individuals,” and urge all members to conduct their professional activities with this goal in mind. (as adopted by the April 2019 Council)

[Notices of the American Mathematical Society (ISSN 0002-9920) is published monthly except bimonthly in June/July by the American Mathematical Society
at 201 Charles Street, Providence, RI 02904-2213 USA, GST No. 12189 2046 RT****. Periodicals postage paid at Providence, RI, and additional mailing offices.
POSTMASTER: Send address change notices to Notices of the American Mathematical Society, PO Box 6248, Providence, RI 02904-6248 USA.] Publication
here of the Society’s street address and the other bracketed information is a technical requirement of the US Postal Service.

© Copyright 2025 by the American Mathematical Society. All rights reserved.

Printed in the United States of America. The paper used in this journal is acid-free and
falls within the guidelines established to ensure permanence and durability.
Opinions expressed in signed Notices articles are those of the authors and do not necessarily
reflect opinions of the editors or policies of the American Mathematical Society.
 Quadratic Forms
Beyond Arithmetic
Alexander Merkurjev and Raman Parimala
1. Introduction One of the earliest and most significant results in the
The concept of quadratic forms can be traced back to an- area of quadratic forms is Fermat’s theorem on sums of
cient civilizations such as the Babylonians and Greeks. The two squares. In the seventeenth century, Fermat stated that
Greeks, particularly Euclid in his famous work Elements, an odd prime number 𝑝 can be expressed as a sum of two
presented geometric methods for solving quadratic equa- squares if and only if 𝑝 is congruent to 1 modulo 4.
tions. The Greeks’ focus on geometry and their meth- Another milestone in the study of sums of squares is the
ods continued to influence mathematicians for centuries. four square theorem, proven by Lagrange in 1770. This the-
Rules for quadratic equations were also discussed in The orem states that every positive integer can be represented
Nine Chapters on the Mathematical Art, composed in China as the sum of four squares.
by 200 BCE. The general formula for solving a quadratic Euler’s sums of two squares identity
equation in one variable—equivalent to the modern sym-
bolic formula—was first stated by the Indian mathemati- (𝑥12 + 𝑥22 )(𝑦21 + 𝑦22 ) = (𝑥1 𝑦1 − 𝑥2 𝑦2 )2 + (𝑥1 𝑦2 + 𝑥2 𝑦1 )2
cian Brahmagupta in his treatise Brˉahmasphutasiddhˉanta in
628 CE.1 shows that the set of sums of two squares in a commuta-
A quadratic form over a commutative ring 𝑅 is a homo- tive ring is closed under multiplication; this statement was
geneous polynomial ∑ 𝑎𝑖𝑗 𝑥𝑖 𝑥𝑗 of degree 2 in 𝑛 variables generalized to the binary quadratic form 𝑥2 + 𝑛𝑦2 by Brah-
𝑥1 , … , 𝑥𝑛 with coefficients 𝑎𝑖𝑗 in 𝑅. In particular, the sum magupta. Similar formulas also exist for the sums of four
of squares 𝑥12 + 𝑥22 + … + 𝑥𝑛2 is a quadratic form defined over and eight squares.
any 𝑅. In the beginning of the nineteenth century Gauss com-
The problem of representing integers as sums of squares pleted the theory of composition of binary quadratic
dates back to ancient times. The Greeks, especially the forms over the integers. Far-reaching generalizations in
Pythagoreans, were interested in the properties of numbers the twenty-first century are the higher composition laws of
and their geometric interpretations. The concept of sums Bhargava leading to asymptotics for the number of num-
of squares is closely related to the Pythagorean theorem, ber fields with bounded discriminant of degree at most 5
conceived in Mesopotamia (1800 BC), first stated precisely (see [1]).
in the Shulbha Sutra of Baudhayana (800 BC) and a state- In the early twentieth century, the main focus of the
ment of proof from China.2 study of quadratic forms is the arithmetic theory over the
In the seventh century, the Indian mathematician Brah- rings of algebraic integers and number fields. Witt in the
magupta considered what is now called Pell’s equation, 30s laid the foundation of the algebraic theory of quadratic
𝑥2 − 𝑎𝑦2 = 1, and found a method for its solution. forms that deals with the study of quadratic forms over
arbitrary fields. There was great progress in this study initi-
ated by Pfister’s theory of multiplicative forms in the 60s.
Alexander Merkurjev is a professor of mathematics at the University of Califor-
nia, Los Angeles. His email address is [email protected].
Several open questions and conjectures were posed since
Raman Parimala is a professor in the department of mathematics at Emory then, a major one being the Milnor conjectures. Since the
University. Her email address is [email protected]. 80s, the introduction of powerful techniques from alge-
Communicated by Notices Associate Editor Han-Bom Moon. braic geometry transformed the study of quadratic forms,
For permission to reprint this article, please contact:
leading to great leaps like a solution of the Milnor conjec-
[email protected].
ture.
In this article we trace the progress in the algebraic the-
DOI: https://doi.org/10.1090/noti3192
1
ory of quadratic forms over the last four decades. For sim-
https://en.m.wikipedia.org/wiki/Brahmagupta
2https://m.rediff.com/news/special/did-india-discover plicity we only consider fields of characteristic different
-pythogoras-theorem-a-top-mathematician-answers/20150109 from 2, although the theory of quadratic forms in char-
.htm acteristic 2 is also well developed. We stay aside from the

AUGUST 2025 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY 711

theory of quadratic forms over commutative rings such as with 𝑥𝑖 ∈ 𝐹 is an arbitrary vector in 𝑉, we have
rings of algebraic integers etc.
In Section 2 we briefly recall basic definitions. In the 𝑞(𝑣) = ∑ 𝑎𝑖𝑗 𝑥𝑖 𝑥𝑗 ,
next section we introduce a method of the study of qua- 𝑖,𝑗
dratic forms in the last four decades based on algebraic ge-
ometry. We introduce the quadric hypersurface (quadric) where 𝑎𝑖𝑗 = 𝑎𝑗𝑖 = 𝑏𝑞 (𝑣 𝑖 , 𝑣𝑗 ). Thus 𝑞 is given by a quadratic
𝑋𝑞 associated with a quadratic form 𝑞 and a tool based homogeneous polynomial over 𝐹. We say that 𝑞 is nonde-
on the study of closed subvarieties (algebraic cycles) on generate if the (symmetric) 𝑛 × 𝑛 matrix 𝐴 = (𝑎𝑖𝑗 ) is non-
the products of quadrics. In Section 4 we state two theo- degenerate. This is equivalent to the nondegeneracy of the
rems on the proof of Milnor conjectures that compare the bilinear form 𝑏𝑞 . Below we will consider nondegenerate
graded Witt ring and the graded mod 2 Galois cohomol- quadratic forms only.
ogy via Milnor’s 𝐾-theory of fields. The determinant of 𝐴 is well defined modulo the sub-
In the next section some discrete invariants of fields (the group 𝐹 ×2 of squares in 𝐹 × . The discriminant of 𝑞 is
𝑢-invariant and the Pythagoras number) that are defined 𝑛(𝑛−1)
(−1) 2 det(𝐴) in 𝐹 × /𝐹 ×2 .
by means of quadratic forms are considered. Although
Two quadratic forms 𝑞 ∶ 𝑉 → 𝐹 and 𝑞′ ∶ 𝑉 ′ → 𝐹 are
these invariants were defined a while ago, the newly devel-
isomorphic if there is an 𝐹-linear isomorphism 𝑓 ∶ 𝑉 → 𝑉 ′
oped techniques allowed us to obtain recent results in this
such that 𝑞(𝑣) = 𝑞′ (𝑓(𝑣)) for all 𝑣 ∈ 𝑉. The forms 𝑞 and
direction. There are big open questions concerning these
𝑞′ are isomorphic if and only if they are given by the same
invariants. The determination of the 𝑢-invariant of func-
quadratic polynomial in some bases for 𝑉 and 𝑉 ′ .
tion fields of curves over totally imaginary number fields
Let 𝑎1 , 𝑎2 , … , 𝑎𝑛 be nonzero elements in 𝐹. We write
is the ultimate goal while even the finiteness for ℚ(√−1) re-
⟨𝑎1 , 𝑎2 , … , 𝑎𝑛 ⟩ for the quadratic form ∑ 𝑎𝑖 𝑥𝑖2 on the space
mains open. Another big open question is the determina-
of 𝑛-columns 𝑉 = 𝐹 𝑛 . It is called the diagonal form and
tion of the Pythagoras number of rational function fields
it has a diagonal matrix in the standard basis for 𝐹 𝑛 . Ev-
over the field of real numbers and the question is open
ery quadratic form over 𝐹 is isomorphic to a diagonal
even for ℝ(𝑥1 , 𝑥2 , 𝑥3 ).
form ⟨𝑎1 , 𝑎2 , … , 𝑎𝑛 ⟩ for some (not uniquely determined)
Another discrete invariant, the dimension of a quadratic
𝑎𝑖 ∈ 𝐹 × .
form, is discussed in Section 6. One of the old intrigu-
If 𝑞 ∶ 𝑉 → 𝐹 is a quadratic from over 𝐹 and 𝐿/𝐹 is a
ing questions on the determination of all possible split-
field extension, one can define a quadratic form 𝑞𝐿 on the
ting patterns of quadratic forms of given dimension is still
𝐿-space 𝐿 ⊗𝐹 𝑉 over 𝐿 with the associated bilinear form 𝑏
unsolved.
on 𝐿 ⊗𝐹 𝑉 defined by 𝑏(𝑥 ⊗ 𝑣, 𝑥′ ⊗ 𝑣′ ) = 𝑥𝑥′ 𝑏𝑞 (𝑣, 𝑣′ ) for
In the last section we introduce the Chow motives tech-
all 𝑥, 𝑥′ ∈ 𝐿 and 𝑣, 𝑣′ ∈ 𝑉.
nique. Indecomposable direct summands of the motives
The orthogonal sum of two forms 𝑞 ∶ 𝑉 → 𝐹 and 𝑞′ ∶
of quadrics can be visualized in the diagrams defined in
𝑉 → 𝐹 is the quadratic form 𝑞 ⟂ 𝑞′ on 𝑉 ⊕ 𝑉 ′ defined by
′
this section.
(𝑞 ⟂ 𝑞′ )(𝑣, 𝑣′ ) = 𝑞(𝑣) + 𝑞′ (𝑣′ ).
2. Definitions The form ℍ = ⟨1, −1⟩ is the hyperbolic plane. A quadratic
Basic references are [13] and [2]. form is hyperbolic if it is isomorphic to the orthogonal sum
Let 𝐹 be a field of characteristic not 2, that is 1 + 1 ≠ 0 𝑛ℍ of 𝑛 > 0 copies of ℍ.
in 𝐹. We write 𝐹 × for the multiplicative group of nonzero A quadratic form 𝑞 ∶ 𝑉 → 𝐹 is called isotropic if there is
elements in 𝐹. a nonzero vector 𝑣 ∈ 𝑉 such that 𝑞(𝑣) = 0; otherwise, 𝑞 is
A quadratic form on a finite dimensional vector space 𝑉 called anisotropic. Every form 𝑞 is isomorphic to 𝑞𝑎𝑛 ⟂ 𝑘ℍ,
over 𝐹 is a map 𝑞 ∶ 𝑉 → 𝐹 such that where 𝑞𝑎𝑛 is an anisotropic form and 𝑘 ≥ 0. The integer
𝑤(𝑞) = 𝑘 is called the Witt index of 𝑞. It is equal to 0 if and
1) 𝑞(𝑥𝑣) = 𝑥2 𝑞(𝑣) for all 𝑥 ∈ 𝐹 and 𝑣 ∈ 𝑉 and only if 𝑞 is anisotropic.
2) the map 𝑏𝑞 ∶ 𝑉 × 𝑉 → 𝐹 defined by the formula Two quadratic forms 𝑞 and 𝑞′ are Witt equivalent if the
1
𝑏𝑞 (𝑣, 𝑣′ ) = [𝑞(𝑣 + 𝑣′ ) − 𝑞(𝑣) − 𝑞(𝑣′ )] is a (symmet- forms 𝑞𝑎𝑛 and 𝑞′𝑎𝑛 are isomorphic. The set 𝑊(𝐹) of equiv-
2
ric) bilinear form. alence classes of quadratic forms over 𝐹 is the Witt ring
of 𝐹 with respect to the orthogonal sum and tensor prod-
Note that the bilinear form 𝑏𝑞 reconstructs 𝑞 by the equal- uct of forms. For example, 𝑊(ℝ) is isomorphic to ℤ: an
ity 𝑞(𝑣) = 𝑏𝑞 (𝑣, 𝑣). Thus, to give a quadratic form on a integer 𝑛 > 0 corresponds to the form 𝑛⟨1⟩ under the iso-
vector space 𝑉 is the same as to give a symmetric bilinear morphism.
form on 𝑉. One of the most important properties of the Witt ring:
The integer 𝑛 = dim(𝑉) is called the dimension of the two quadratic forms 𝑞 and 𝑞′ are isomorphic if and only if
form 𝑞. Let {𝑣 1 , 𝑣 2 , … , 𝑣 𝑛 } be a basis for 𝑉. If 𝑣 = ∑ 𝑥𝑖 𝑣 𝑖 𝑞 = 𝑞′ in 𝑊(𝐹) and dim(𝑞) = dim(𝑞′ ).

712 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY VOLUME 72, NUMBER 7

 The ideal 𝐼(𝐹) of even dimensional forms in 𝑊(𝐹) is The two quadrics 𝑋𝑞 and 𝑋ℎ are isomorphic if and only
called the fundamental ideal. The quotient 𝑊(𝐹)/𝐼(𝐹) is iso- if the quadratic forms 𝑞 and ℎ are similar, that is each of
morphic to ℤ/2ℤ. The powers 𝐼 𝑛 (𝐹) of the fundamental the forms 𝑞 and ℎ is an 𝐹-multiple of the other.
ideal form a filtration on 𝑊(𝐹). The discriminant yields We write 𝐹(𝑞) for the function field 𝐹(𝑋𝑞 ). Note that
an isomorphism between 𝐼(𝐹)/𝐼 2 (𝐹) and 𝐹 × /𝐹 ×2 . The de- 𝑞 is isotropic over 𝐹(𝑞). Indeed, if 𝑥1 , 𝑥2 , … , 𝑥𝑛 are ho-
termination of all the quotients 𝐼 𝑛 (𝐹)/𝐼 𝑛+1 (𝐹) is one of the mogeneous coordinates in ℙ(𝑉), we can view the ratio-
fundamental problems of the algebraic theory of quadratic nal functions 𝑥𝑖 /𝑥1 as the elements in 𝐹(𝑞)× satisfying
forms (see Section 4 below). 𝑞(1, 𝑥2 /𝑥1 , … , 𝑥𝑛 /𝑥1 ) = 0 in 𝐹(𝑞). Thus, the equation
Let 𝑎1 , 𝑎2 , … , 𝑎𝑛 ∈ 𝐹 × . The 2𝑛 -dimensional quadratic 𝑞(𝑣) = 0 has a nonzero solution over the field 𝐹(𝑞).
form The field 𝐹(𝑞) is a generic splitting field of 𝑞: the form 𝑞
is isotropic over a field extension 𝐿/𝐹 if and only if there is
⟨⟨𝑎1 , 𝑎2 , … , 𝑎𝑛 ⟩⟩ = ⟨1, −𝑎1 ⟩ ⊗ ⟨1, −𝑎2 ⟩ ⊗ ⋯ ⊗ ⟨1, −𝑎𝑛 ⟩
an 𝐹-place from 𝐹(𝑞) to 𝐿, that is an 𝐹-algebra homomor-
is called an 𝑛-fold Pfister form. The 𝑛th power 𝐼 𝑛 (𝐹) of the phism of a valuation 𝐹-subalgebra 𝑅 ⊂ 𝐹(𝑞) to 𝐿.
fundamental ideal in the Witt ring is generated by the 𝑛- Consider the following example of an application of the
fold Pfister forms as an abelian group. An 𝑛-fold Pfister theory of algebraic cycles. Let 𝑞 be an anisotropic quadratic
form is either anisotropic or hyperbolic. form over 𝐹 and let 𝐿/𝐹 be a field extension. We would
like to know when 𝑞 is isotropic over 𝐿, i.e., the equation
Example 2.1. 𝑞(𝑣) = 0 has a nonzero solution over 𝐿. Typically, the ex-
1) Let 𝐾 = 𝐹(𝑎1/2 ) be a quadratic field extension of 𝐹. tension 𝐿/𝐹 is finitely generated, so we can choose an inte-
The norm form 𝑞 ∶ 𝐾 → 𝐹 defined by 𝑞(𝑥 + 𝑦𝑎1/2 ) = gral variety 𝑌 over 𝐹 such that 𝐿 is isomorphic to the func-
𝑥2 − 𝑎𝑦2 is the 1-fold Pfister form ⟨⟨𝑎⟩⟩. tion field 𝐹(𝑌 ) of 𝑌 over 𝐹. By the main property of the
2) Let 𝑎, 𝑏 ∈ 𝐹 × and let 𝑄 be a quaternion 4-dimensional quadric, 𝑋𝑞 has a point over 𝐿 = 𝐹(𝑌 ) if and only if there
(associative noncommutative) 𝐹-algebra with basis is a morphism Spec(𝐿) → 𝑌 or, equivalently, a rational
{1, 𝑖, 𝑗, 𝑘} and multiplication table 𝑖2 = 𝑎, 𝑗2 = 𝑏 and morphism 𝑌 99K 𝑋𝑞 defined on a nonempty open subset
𝑘 = 𝑖𝑗 = −𝑗𝑖. The reduced norm form 𝑞 ∶ 𝑄 → 𝐹 de- 𝑈 ⊂ 𝑌 . The closure of the graph of 𝑈 → 𝑋𝑞 in the product
fined by 𝑞(𝑥 + 𝑦𝑖 + 𝑧𝑗 + 𝑡𝑘) = 𝑥2 − 𝑎𝑦2 − 𝑏𝑧2 + 𝑎𝑏𝑡2 is 𝑌 × 𝑋𝑞 yields a cycle class of dimension 𝑑 = dim(𝑌 ) in the
the 2-fold Pfister form ⟨⟨𝑎, 𝑏⟩⟩. Chow group CH𝑑 (𝑌 × 𝑋𝑞 ) of rational equivalence classes of
3) For a triple of elements 𝑎, 𝑏, 𝑐 ∈ 𝐹 × there is an octonian cycles of dimension 𝑑 on 𝑌 × 𝑋𝑞 .
(nonassociative noncommutative) 8-dimensional al- Conversely, a prime algebraic cycle on 𝑌 × 𝑋𝑞 of dimen-
gebra 𝐶 that admits a norm quadratic form 𝑞 ∶ 𝐶 → 𝐹 sion 𝑑 that is birational when projecting to 𝑌 yields a ratio-
that is the 3-fold Pfister form ⟨⟨𝑎, 𝑏, 𝑐⟩⟩. nal morphism 𝑌 99K 𝑋𝑞 and therefore a point of 𝑋𝑞 over
𝐿.
Note that norm forms 𝑞 in all the examples are multi- Thus, one can use the machinery of algebraic cycles. In
plicative, i.e., the product of two values of 𝑞 is a value of particular, one can obtain new cycles by intersecting with
𝑞. By the Hurwitz theorem, there are no algebras of di- other cycles (for example, with the Chern classes of the vec-
mensions not 1, 2, 4, and 8 with the multiplicative norm. tor bundles such as the tangent bundle), by considering
Nevertheless, the set of nonzero values of an 𝑛-fold Pfister pull-backs and push-forwards with respect to certain mor-
form is closed under multiplication for every 𝑛. In partic- phisms and also applying the Steenrod operations modulo
ular, the sum of 2𝑛 squares is a multiplicative quadratic 2.
form over any field.
4. Milnor Conjectures
3. Algebro-Geometric Methods
The most spectacular achievement in the algebraic theory
During the last three decades the methods of algebraic of quadratic forms is the solution of the Milnor conjec-
geometry interfaced algebraic theory of quadratic forms. tures.
One can associate to every quadratic form 𝑞 ∶ 𝑉 → 𝐹 over The Milnor 𝐾-theory 𝐾∗𝑀 (𝐹) of a field 𝐹 is the (graded)
𝐹 the quadric hypersurface 𝑋𝑞 (simply called the quadric of quotient ring of the tensor ℤ-algebra of the multiplicative
𝑞) given by the equation 𝑞(𝑣) = 0 in the projective space group 𝐹 × by the ideal generated by the tensors 𝑎⊗𝑏, where
ℙ(𝑉). The quadric 𝑋𝑞 has dimension dim(𝑞) − 2 and it is 𝑎, 𝑏 ∈ 𝐹 × with 𝑎 + 𝑏 = 1. In particular, 𝐾0𝑀 (𝐹) = ℤ and
smooth if 𝑞 is nondegenerate. The variety 𝑋𝑞 is integral if 𝐾1𝑀 (𝐹) = 𝐹 × . The group 𝐾𝑛𝑀 (𝐹) is generated by the sym-
dim(𝑞) ≥ 3. By definition, 𝑋𝑞 has a point over 𝐹 if and bols {𝑎1 , 𝑎2 , … , 𝑎𝑛 }, where 𝑎𝑖 ∈ 𝐹 × , that are multiplicative
only if the form 𝑞 is isotropic. More generally, 𝑋𝑞 has a with respect to every variable (when the other variables are
point over a field extension 𝐿/𝐹 if and only if 𝑞 is isotropic fixed) and satisfy the Steinberg relation {𝑎1 , 𝑎2 , … , 𝑎𝑛 } = 0 if
over 𝐿. For example, the quadric 𝑋𝑞 , where 𝑞 = 𝑥2 +𝑦2 +𝑧2 ,
has no points over ℝ, but it has points over ℂ.

AUGUST 2025 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY 713

𝑎𝑖 + 𝑎𝑗 = 1 for some 𝑖 ≠ 𝑗. The assignment are given two anisotropic quadratic forms 𝑓 and 𝑔 and we
{𝑎1 , 𝑎2 , … , 𝑎𝑛 } ↦ ⟨⟨𝑎1 , 𝑎2 , … , 𝑎𝑛 ⟩⟩ want to decide whether 𝑓 and 𝑔 are isomorphic, or, equiv-
alently, that the form 𝑞 = 𝑓 ⟂ (−𝑔) is hyperbolic. We
yields a well-defined graded ring homomorphism compute the cohomological invariants 𝑐𝑛𝐹 of 𝑞 one by one
𝑠𝐹 ∶ 𝐾∗𝑀 (𝐹) → 𝐼 ∗ (𝐹)/𝐼 ∗+1 (𝐹). starting with 𝑛 = 0. Note that the next invariant is defined
Let Γ𝐹 be the absolute Galois group of a field 𝐹, i.e., Γ𝐹 is if the previous one vanishes.
the Galois group of a separable closure of 𝐹 over 𝐹. We The invariant 𝑐0𝐹 takes 𝑞 to its dimension modulo 2 in
write 𝐻 ∗ (𝐹) for the graded cohomology ring of Γ𝐹 with co- ℤ/2ℤ = 𝐻 0 (𝐹). If this invariant vanishes, i.e., 𝑞 has even
efficients in ℤ/2ℤ. The multiplication in 𝐻 ∗ (𝐹) is given dimension, then 𝑞 ∈ 𝐼(𝐹). The invariant 𝑐1𝐹 (𝑞) is equal to
by the cup-product ∪. In particular, 𝐻 0 (𝐹) = ℤ/2ℤ and the discriminant of 𝑞 in 𝐹 × /𝐹 ×2 = 𝐻 1 (𝐹). If the discrimi-
𝐻 1 (𝐹) = 𝐹 × /𝐹 ×2 . For every 𝑎 ∈ 𝐹 × write (𝑎) for the corre- nant of 𝑞 is trivial, the form 𝑞 belongs to 𝐼 2 (𝐹) and we can
sponding cohomology class in 𝐻 1 (𝐹). The assignment compute 𝑐2𝐹 (𝑞). This is the class of the Clifford algebra of
𝑞 in the subgroup Br(𝐹)[2] = 𝐻 2 (𝐹) of the Brauer group of
{𝑎1 , 𝑎2 , … , 𝑎𝑛 } ↦ (𝑎1 ) ∪ (𝑎2 ) ∪ … ∪ (𝑎𝑛 ) classes of exponent 2. If this vanishes, then 𝑞 ∈ 𝐼 3 (𝐹) and
yields a well-defined graded ring homomorphism so on. If 𝑐𝑛𝐹 (𝑞) = 0, i.e., 𝑞 ∈ 𝐼 𝑛+1 (𝐹) and dim(𝑞) < 2𝑛+1 ,
then 𝑞 is hyperbolic ([19, p. 33]) and the form 𝑓 and 𝑔 are
ℎ𝐹 ∶ 𝐾∗𝑀 (𝐹) → 𝐻 ∗ (𝐹)
isomorphic.
called the norm residue homomorphism modulo 2. Thus,
we have the following diagram of graded ring homomor- 5. Field Invariants
phisms. We discuss two integer invariants of fields associated to
quadratic forms, namely the 𝑢-invariant and the Pythago-
𝐾∗𝑀 (𝐹)/2𝐾∗𝑀 (𝐹)
mm NNN ras numbers.
𝑠𝐹 mmm NNNℎ𝐹
mm NNN
mmmm NN' 5.1. 𝑢-invariant. The 𝑢-invariant 𝑢(𝐹) of a field 𝐹 is the
vm largest dimension of an anisotropic quadratic form over 𝐹.
∗ ∗+1
𝐼 (𝐹)/𝐼 (𝐹) 𝐻 ∗ (𝐹) For example 𝑢(ℝ) = ∞ and 𝑢(ℂ) = 1. The 𝑢-invariant of
Milnor has conjectured in the 70s that both maps 𝑠𝐹 and a finite field is 2. The 𝑢-invariant of a local field is equal
ℎ𝐹 are isomorphisms for all fields. This conjecture for to 4. The Hasse-Minkowski theorem implies that the 𝑢-
𝑛 = 2 was settled by the first author in the 80s leading to a invariant of a totally imaginary number field is 4. The for-
solution of the longstanding question on the generation of mula 𝑢(𝐹((𝑡))) = 2𝑢(𝐹) shows that every power of 2 is the
the 2-torsion in the Brauer group of a field by quaternion 𝑢-invariant of some field. Breaking the myth that all in-
algebras. variants of fields associated to quadratic forms are a power
Voevodsky in [24] proved one of the Milnor conjectures. of 2, it was shown in [15], that every even integer is the
𝑢-invariant of some field. On the other hand, it is known
Theorem 4.1. The graded ring homomorphism ℎ𝐹 is always
that the 𝑢-invariant is not equal to 3,5 or 7 ([19, Proposi-
an isomorphism.
tion 1.3, p. 111]). It is expected that every odd integer ≥ 9
In the proof Voevodsky introduces a number of revo- is the value of the 𝑢-invariant. Izhboldin has shown in [7]
lutionary ideas and tools. The main tool is the motivic that there are fields of 𝑢-invariant 9. In [22] Vishik proved
cohomology defined by Voevodsky. Another tool is the that every integer of the form 2𝑚 + 1 for 𝑚 ≥ 3 can be the
motivic Steenrod operations defined by Voevodsky in anal- 𝑢-invariant of a field.
ogy with the classical topological operations. The motivic The behavior of the 𝑢-invariant of finite field extensions
analogs of some of the basic operations, the Milnor oper- has been studied extensively. We have the following theo-
ations, played an essential role in the proof. rem of Leep for finite field extensions.
Enriching Voevodsky’s methods, Orlov, Vishik, and Vo-
evodsky proved in [16] another Milnor conjecture. Theorem 5.1 ([19, Theorem 3.1, p. 120]). Let 𝐿/𝐹 be an
𝑛+1
extension of degree 𝑛. Then 𝑢(𝐿) ≤ 𝑢(𝐹).
Theorem 4.2. The graded ring homomorphism 𝑠𝐹 is always 2

an isomorphism. There may be bounds for the 𝑢-invariants of finite ex-

In particular, for every 𝑛 ≥ 0, we get a group isomor- tensions independent of the degree of the extensions.
phism The behavior of the 𝑢-invariant under rational function
field extensions is not well understood.
𝑐𝑛𝐹 = ℎ𝐹𝑛 ∘ (𝑠𝑛𝐹 )−1 ∶ 𝐼 𝑛 (𝐹)/𝐼 𝑛+1 (𝐹) → 𝐻 𝑛 (𝐹).
Question 5.2. If 𝑢(𝐹) < ∞, is 𝑢(𝐹(𝑡)) < ∞?
We can call the maps 𝑐𝑛𝐹 the cohomological invariants of qua-
dratic forms. These invariants determine anisotropic qua- If this question has an affirmative answer, then it fol-
dratic forms up to isomorphism as follows. Suppose we lows that there are bounds independent of the degree for

714 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY VOLUME 72, NUMBER 7

the 𝑢-invariant of finite field extensions. In fact the above 𝑝(𝑘(𝑥)) = 𝑝(𝑘) + 1 ≤ 5 for real number fields 𝑘 ([19, The-
question is wide open if 𝐹 is a totally imaginary number orem 1.9, p. 100]). Conjecturally for a function field 𝐹
field. Even for the function fields of 𝑝-adic curves, the ques- in one variable over a number field, 𝑝(𝐹) ≤ 5 ([19, Con-
tion remained open until the late 90s when the first finite- jecture 1.10, p. 100]). The estimate 𝑝(𝐹) ≤ 7 is due to
ness results emerged from Merkurjev and Hoffmann-Van Colliot-Thélène and a sharper estimate 𝑝(𝐹) ≤ 6 is due
Geel. A theorem of Saltman on bounding indices of ele- to Pop ([20]). One has the following general result for a
ments in the 2-torsion of the Brauer group was pivotal to function field 𝐹 in 𝑑 variables over a number field, namely
this theorem. We have the following theorem for function 𝑝(𝐹) ≤ 2𝑑+1 (cf. [19, p. 100/101]). This combines the Mil-
fields of 𝑝-adic curves. nor conjecture and a local-global principle for the Galois
cohomology of higher-dimensional function fields due to
Theorem 5.3 ([17], [18]). Let 𝐾 be a 𝑝-adic field and 𝐹 the
Jannsen in [8].
function field of a curve over 𝐾. Then 𝑢(𝐹) = 8.
We have the following more general extension of the 6. Dimensions of Quadratic Forms
above theorem which fundamentally uses a theorem of In this section we discuss results on the dimension of qua-
Heath-Brown on the zeros of systems of 𝑝-adic quadratic dratic forms satisfying certain conditions.
forms. 6.1. Possible dimensions of anisotropic quadratic
Theorem 5.4 ([14]). Let 𝐾 be a 𝑝-adic field and 𝐹 the func- forms in 𝐼 𝑛 (𝐹). An anisotropic 𝑛-fold Pfister form over
tion field of a variety of dimension 𝑑 over 𝐾. Then 𝑢(𝐹) = 2𝑑+2 . 𝐹 belongs to 𝐼 𝑛 (𝐹) and has dimension 2𝑛 . By an Arason-
Pfister theorem, every nonzero anisotropic form in 𝐼 𝑛 (𝐹)
We also have the following theorem extending to gen- is of dimension at least 2𝑛 . Are there other restrictions on
eral complete discretely valued fields, which uses the patch- the dimensions of anisotropic quadratic forms in 𝐼 𝑛 (𝐹)?
ing techniques of Harbater-Hartmann-Krashen. Karpenko proved in [11] the following general result.
Theorem 5.5 ([4]). Let 𝐾 be a complete discretely valued field Theorem 6.1. Let 𝑞 be an anisotropic quadratic form such
with residue field 𝜅. Let 𝐹 be the function field of a curve over that 𝑞 ∈ 𝐼 𝑛 (𝐹) for some 𝑛 ≥ 1. If dim(𝑞) < 2𝑛+1 , then
𝐾. Suppose that char(𝜅) ≠ 2 and there exists an integer 𝑑 dim(𝑞) = 2𝑛+1 − 2𝑖+1 for some 𝑖 ∈ {0, 1, … , 𝑛}. All other even
such that 𝑢(𝐿) ≤ 𝑛 for all finitely generated extensions of 𝜅 of degrees at least 2𝑛+1 are possible.
transcendence degree at most 1. Then 𝑢(𝐹) ≤ 2𝑛.
For example, the possible dimensions of anisotropic
5.2. Pythagoras number. The Pythagoras number 𝑝(𝐹) of
quadratic forms in 𝐼 4 (𝐹) are
a field 𝐹 is the smallest integer 𝑛 such that every sum of
squares in 𝐹 is a sum of at most 𝑛 squares in 𝐹. 0, 16, 24, 28, 30, 32, 34, 36, … .
For example 𝑝(ℝ) = 𝑝(ℂ) = 1. If 𝐹 is a real number 6.2. Hoffmann’s separation theorem. Let 𝑞 and ℎ be two
field, then 3 ≤ 𝑝(𝐹) ≤ 4 (cf. [19, p. 95]). The study of the anisotropic quadratic forms over 𝐹. When is 𝑞 isotropic
Pythagoras number of fields which are formally real, i.e., over the field 𝐹(ℎ)? Equivalently, is there a rational mor-
−1 is not a sum of squares in the field, is interesting. The phism 𝑋ℎ 99K 𝑋𝑞 ? If 𝑞 is hyperbolic over 𝐹(ℎ), then
following theorem answers a question of Pfister. dim(𝑞) ≥ dim(ℎ) ([13, Ch. X, Theorem 4.5]). This inequal-
Theorem 5.6 ([6]). Every positive integer 𝑛 ≥ 1 is the ity does not hold in general if 𝑞 is just isotropic over 𝐹(ℎ).
Pythagoras number of a formally real field. Example 6.2. Let ℎ be a general 𝑛-fold Pfister form (i.e., ℎ
The determination of Pythagoras numbers of rational is similar to an 𝑛-fold Pfister form) and 𝑞 a subform of ℎ
1
function fields ℝ(𝑥1 , ⋯ 𝑥𝑛 ) has a long history. Pfister the- of dimension > dim(ℎ). Over the field 𝐹(ℎ) the Pfister
2
ory leads to 𝑝(ℝ(𝑥1 , ⋯ , 𝑥𝑛 )) ≤ 2𝑛 (cf. [19, p. 95]). We form ℎ is isotropic and hence hyperbolic. Therefore, 𝑞 is
have 𝑝(ℝ(𝑥)) = 2 and 𝑝(ℝ(𝑥1 , 𝑥2 )) = 4 ([19, p. 96]). We isotropic over 𝐹(ℎ).
have bounds 𝑛 + 2 ≤ 𝑝(ℝ(𝑥1 , ⋯ , 𝑥𝑛 )) ≤ 2𝑛 ([19, p. 97]).
One has the following sharper bound for 𝑝(𝐹) if 𝐹 is a for- Note that the numbers dim(𝑞) and dim(ℎ) in the exam-
mally real field of transcendence degree 𝑑 over a subfield, ple are not separated by a power of 2. The following result
namely 𝑝(𝐹) ≥ 𝑑 + 1 ([3]). proved by Hoffmann in [5] (known as the separation theo-
A major open question in this area is the following. rem) explains this observation.

Question 5.7. What is 𝑝(ℝ(𝑥1 , 𝑥2 , 𝑥3 ))? Theorem 6.3. Let 𝑞 and ℎ be two anisotropic quadratic forms
over 𝐹. Suppose that dim(𝑞) ≤ 2𝑛 < dim(ℎ) for some 𝑛 ≥ 0.
More on the arithmetic side, interesting questions arise
Then 𝑞 is anisotropic over 𝐹(ℎ).
concerning the Pythagoras number of a formally real func-
tion field over the field of rational numbers. The inequal- 6.3. The first Witt index. Let 𝑞 be an anisotropic qua-
ity 𝑝(ℚ(𝑥)) ≤ 8 goes back to Landau (1906) and we have dratic form of dimension at least 2 over 𝐹. The form 𝑞𝐹(𝑞)

AUGUST 2025 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY 715

is isotropic. Its Witt index is then positive, denoted 𝑖1 (𝑞), a general Pfister form. Quadratic forms of dimension up
and is called the first Witt index of 𝑞. The “typical” value of to 3 are excellent. A form of dimension 4 is excellent if and
the first Witt index is equal to 1. If 𝑞 is a “generic” form, only if it has trivial discriminant. The splitting pattern of
e.g., 𝑞 = ⟨𝑥1 , 𝑥2 , … , 𝑥𝑛 ⟩ over the field 𝐹(𝑥1 , 𝑥2 , … , 𝑥𝑛 ) of ra- an excellent form is determined inductively as follows: if
tional functions, then 𝑖1 (𝑞) = 1. On the other hand, if 𝑞 is in the definition, dim(𝑞) + dim(𝑞′ ) = 2𝑚 , then
an anisotropic 𝑚-fold Pfister form, then 𝑖1 (𝑞) = 2𝑚−1 since
SP(𝑞) = {0} ∪ (SP(𝑞′ ) + 𝑛 − 2𝑚−1 ).
𝑞 is hyperbolic over 𝐹(𝑞).
All possible values of the first Witt index of all quadratic In particular, the splitting pattern of an excellent form 𝑞
forms of given dimension 𝑛 were determined by Karpenko depends just on the dimension of 𝑞. For example, if 𝑞 is
in [9]. an excellent form of dimension 25, then SP(𝑞) = {0, 9, 12}.
Theorem 6.4. Let 𝑞 be an anisotropic quadratic form of di- 6.5. Izhboldin dimension. Let 𝑞 be a quadratic form
mension 𝑛. Write 𝑛 − 1 in base 2: over 𝐹. Izhboldin defined an integer associated with the
𝑛 − 1 = 2𝑘1 + 2𝑘2 + ⋯ + 2𝑘𝑟 quadric 𝑋𝑞 , called the Izhboldin dimension:

with 0 ≤ 𝑘1 < 𝑘2 < ⋯ < 𝑘𝑟 . Then dim𝐼𝑧ℎ (𝑋𝑞 ) = dim(𝑋𝑞 ) − 𝑖1 (𝑞) + 1 = dim(𝑞) − 𝑖1 (𝑞) − 1,
𝑖1 (𝑞) − 1 = 2𝑘1 + 2𝑘2 + ⋯ + 2𝑘𝑠 where 𝑖1 (𝑞) is the first Witt index of 𝑞.
for some 𝑠 = 0, 1, … , 𝑟 − 1. Example 6.7. Let ℎ and 𝑞 be as in Example 6.2. Then
In other words, 𝑖1 (𝜑) − 1 is the remainder upon dividing dim𝐼𝑧ℎ (𝑋ℎ ) = dim𝐼𝑧ℎ (𝑋𝑞 ) = 2𝑛−1 − 1. Note that 𝑞 is
dim(𝑞) − 1 by some power of 2 less than dim(𝑞). In fact, all isotropic over 𝐹(ℎ) and ℎ is isotropic over 𝐹(𝑞).
values of 𝑖1 (𝑞) given in the theorem are attained by some Theorem 6.8 ([10]). Let 𝑞 be a quadratic form and let 𝑌 be
forms over appropriate fields. In particular, the number of a complete (possibly singular) algebraic variety over 𝐹 with all
possible values of 𝑖1 (𝑞) is equal to the number of 1’s in the closed points of even degree and such that 𝑌 has a closed point
base 2 expression of the integer dim(𝑞) − 1. of odd degree over 𝐹(𝑞) (this holds, for example, if 𝑌 has a
Example 6.5. point over 𝐹(𝑞)). Then dim𝐼𝑧ℎ (𝑋𝑞 ) ≤ dim(𝑌 ) and in the case
dim𝐼𝑧ℎ (𝑋𝑞 ) = dim(𝑌 ) the form 𝑞 is isotropic over 𝐹(𝑌 ).
1) If dim(𝑞) = 2𝑚 + 1, then 𝑖1 (𝑞) = 1. This can also be
deduced from Hoffmann’s separation theorem. The following corollary can be viewed as a variant of the
2) All possible values of the first Witt index of an Separation Theorem.
anisotropic form of dimension 2𝑚 are the 2-powers
1, 2, 22 , … , 2𝑚−1 . The largest value 2𝑚−1 is the first Witt Corollary 6.9. Let ℎ and 𝑞 be anisotropic quadratic forms. If
index of an 𝑚-fold Pfister form. ℎ is isotropic over 𝐹(𝑞), then dim𝐼𝑧ℎ (𝑋𝑞 ) ≤ dim𝐼𝑧ℎ (𝑋ℎ ) and
in the case dim𝐼𝑧ℎ (𝑋𝑞 ) = dim𝐼𝑧ℎ (𝑋ℎ ) the form 𝑞 is isotropic
6.4. Splitting patterns of quadratic forms. Let 𝑞 be a qua- over 𝐹(ℎ).
dratic form of dimension 𝑛 over 𝐹 and let
7. Chow Motives of Quadrics
𝑖0 < 𝑖1 < ⋯ < 𝑖ℎ−1 < 𝑖ℎ = [𝑛/2]
The constructions and results in this section are due to
denote all Witt indices of quadratic forms 𝑞𝐿 over all field Vishik (see [21]). There is a functor from the category
extensions of 𝐿/𝐹. The tuple SP(𝑞) = (𝑖0 , 𝑖1 , … , 𝑖ℎ−1 , 𝑖ℎ ) of of smooth projective varieties to the additive category of
strictly increasing integers is called the splitting pattern of Chow Motives taking a variety 𝑋 to its motive 𝑀(𝑋) and a
𝑞. The smallest integer 𝑖0 is the Witt index of 𝑞. If 𝑞 is morphism 𝑓 ∶ 𝑋 → 𝑌 of varieties to the class of the graph
anisotropic, i.e., 𝑖0 = 0, the integer 𝑖1 is the first Witt index of 𝑓 in the Chow group of the classes of algebraic cycles in
𝑖1 (𝑞). 𝑋 × 𝑌 . The motive of the projective space ℙ𝑛 decomposes
All possible splitting patterns of quadratic forms of into a direct sum ℤ ⊕ ℤ(1) ⊕ ⋯ ⊕ ℤ(𝑛), where ℤ(𝑖) are the
small dimension are determined by Vishik in [21]. For ex- Tate motives.
ample, the splitting patterns of anisotropic 9-dimensional Let 𝑋 be a quadric of dimension 𝑛 over an algebraically
forms are (0, 1, 4) and (0, 1, 2, 3, 4). For arbitrary 𝑛, it is un- closed field. Then
known which splitting patterns of 𝑛-dimensional forms oc- 𝑀(𝑋) ≃ {
ℤ ⊕ ℤ(1) ⊕ ⋯ ⊕ ℤ(𝑛) = 𝑀(ℙ𝑛 ), if 𝑛 is odd;
ℤ ⊕ ℤ(1) ⊕ ⋯ ⊕ ℤ(𝑘 − 1) ⊕ ℤ(𝑘) ⊕ ℤ(𝑘) ⊕ ℤ(𝑘 + 1) ⊕ ⋯ ⊕ ℤ(𝑛), if 𝑛 = 2𝑘.
cur.
Note that the motive ℤ(𝑘) appears in the decomposition
Example 6.6. (Excellent form) We employ the following
twice if 𝑛 = 2𝑘.
inductive definition. An anisotropic quadratic form 𝑞 of
We introduce the set of symbols
dimension 𝑛 is called excellent if either 𝑞 = 0 or there is an
excellent form 𝑞′ such that dim(𝑞′ ) < dim(𝑞) and 𝑞 ⟂ 𝑞′ is Λ(𝑋) = Λ(𝑛) = {0, 1, … , 𝑘, 𝑘,̄ … , 1,̄ 0},
̄

716 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY VOLUME 72, NUMBER 7

where 𝑘 = [𝑛/2] and a bijection between Λ(𝑋) and the set
of Tate motives in the decomposition of 𝑀(𝑋) as follows:
𝑖 ↔ ℤ(𝑖) and 𝑖 ̄ ↔ ℤ(𝑛 − 𝑖) for 𝑖 = 0, 1, … , 𝑘.
Over an arbitrary field the motive 𝑀(𝑋) is a direct sum
𝑀1 ⊕ 𝑀2 ⊕ ⋯ 𝑀𝑠 of indecomposable motives. The col-
lection of indecomposable motives 𝑀𝑖 in the direct sum is The shape of the diagram depends only on dim(𝑞). All
unique up to isomorphism. Over an algebraic closure of 𝐹, connections in the diagram for an excellent form are called
every motive 𝑀𝑖 is the sum of certain Tate motives. Thus, the excellent connections.
the set Λ(𝑋) is the disjoint union of 𝑠 subsets Λ1 , Λ2 , … , Λ𝑠
Let 𝑞 be an anisotropic form with the splitting pattern
so that for every 𝑖 the elements of Λ𝑖 correspond to the Tate
(𝑖0 = 0, 𝑖1 , … , 𝑖ℎ−1 , 𝑖ℎ ) and 𝑋 = 𝑋𝑞 , 𝑛 = dim(𝑋). There are
motives in the decomposition of 𝑀𝑖 over an algebraic clo-
the following standard connections in the diagram of 𝑋.
sure. The motive 𝑀(𝑋) is indecomposable if and only if
𝑠 = 1. Proposition 7.3. Let 𝑖, 𝑗 = 0, 1, … , [𝑛/2] be such that 𝑖𝑟−1 ≤
We will consider the diagrams (graphs) with vertices 𝑖, 𝑗 < 𝑖𝑟 , where 1 ≤ 𝑟 ≤ ℎ, and 𝑖 + 𝑗 = 𝑖𝑟−1 + 𝑖𝑟 − 1. Then the
given by the set Λ(𝑋). A connection is an edge in the dia- symbols 𝑖 and 𝑗 ̄ are connected in the diagram of 𝑋.
gram such that both ends of the edge belong to the same
subset Λ𝑖 . (But we don’t necessarily connect every two ver- The standard connections for an excellent form coin-
tices in the same subset Λ𝑖 .) We will draw vertices hori- cide with excellent connections. Vishik proved in [23] that
zontally in the order they appear in the definition of Λ(𝑋) for an arbitrary form excellent connections also appear in
with one exception: if 𝑛 is even we draw the vertices 𝑘 and the diagram:
𝑘 ̄ one under the other to indicate that they correspond to Theorem 7.4. Let 𝑞 be an anisotropic quadratic form of di-
the same Tate motive ℤ(𝑘). Below are the diagrams for mension at least 2 and let 𝑓 be an excellent form of the same
a 6-dimensional hyperbolic (respectively, 7-dimensional dimension. Identify canonically Λ(𝑋𝑞 ) and Λ(𝑋𝑓 ). If two sym-
generic) quadrics. bols are connected in Λ(𝑋𝑓 ), then they are also connected in
Λ(𝑋𝑞 ).
The theorem shows that the diagram of 𝑋𝑓 is contained
in the diagram of 𝑋𝑞 . In particular, the diagram of an ex-
cellent form contains the smallest number of connections
Example 7.1. Let 𝑞 be a subform of dimension 2𝑚−1 + 1 among the diagram of forms of the same dimension.
of a general anisotropic 𝑚-fold Pfister form ℎ. Rost proved
that there is a direct summand 𝑀 of the motive 𝑀(𝑋𝑞 ) Example 7.5. Let 𝑞 be an anisotropic Albert form, i.e., 𝑞 is
such that over an algebraic closure 𝑀 is isomorphic to a 6-dimensional quadratic form with trivial discriminant.
ℤ ⊕ ℤ(2𝑚−1 − 1). This motive depends on ℎ only (not Its splitting pattern is (0, 1, 3). Below are the two diagrams
on the choice of 𝑞); it is called the Rost motive of ℎ and with the standard and excellent connections respectively.
denoted 𝑀ℎ .
The motive of the Pfister quadric 𝑀(𝑋ℎ ) is isomorphic
to 𝑀ℎ ⊕ 𝑀ℎ (1) ⊕ ⋯ ⊕ 𝑀ℎ (2𝑚−1 − 1). For example, if 𝑚 = 3,
the diagram of 𝑀(𝑋ℎ ) looks as follows.

Combining these diagrams we see that all vertices are

connected, hence the motive 𝑀(𝑋𝑞 ) is indecomposable.

Example 7.2. Let 𝑞 be an excellent quadratic form of di- Indecomposability of the motive of a quadric is used in
mension ≥ 2. The motive 𝑀(𝑋𝑞 ) is a direct sum of twists the following application due to Izhboldin and Karpenko
of Rost motives of the general Pfister forms appearing in in [12].
the definition of an excellent form. For example, the dia- Theorem 7.6. Let 𝑞 and ℎ be two anisotropic quadratic forms
gram of the motive of the 11-dimensional excellent form of the same odd dimension. Suppose that 𝑞 is isotropic over
⟨⟨𝑎, 𝑏, 𝑐⟩⟩ ⟂ 𝑑⟨𝑎, 𝑏, −𝑎𝑏⟩ is as follows. 𝐹(ℎ) and ℎ is isotropic over 𝐹(𝑞). If in addition at least one of
the two motives 𝑀(𝑋𝑞 ) and 𝑀(𝑋ℎ ) is indecomposable, then the
forms 𝑞 and ℎ are similar.

AUGUST 2025 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY 717

 218–224; English transl., Math. USSR-Izv. 38 (1992), no. 1,
ACKNOWLEDGMENTS. We are grateful to Nikita 215–221. MR1130036
Karpenko, Alexander Vishik, and Venapally Suresh for [16] D. Orlov, A. Vishik, and V. Voevodsky, An exact se-
the useful comments. We thank Manjul Bhargava for quence for 𝐾∗𝑀 /2 with applications to quadratic forms, Ann.
pointing out to us the contributions from ancient civi- of Math. (2) 165 (2007), no. 1, 1–13, DOI 10.4007/an-
lizations. nals.2007.165.1. MR2276765
[17] Raman Parimala and V. Suresh, The 𝑢-invariant of the
function fields of 𝑝-adic curves, Ann. of Math. (2) 172 (2010),
References no. 2, 1391–1405, DOI 10.4007/annals.2010.172.1397.
[1] Manjul Bhargava, Higher composition laws and applications, MR2680494
International Congress of Mathematicians. Vol. II, Eur. [18] R. Parimala and V. Suresh, Period-index and 𝑢-invariant
Math. Soc., Zürich, 2006, pp. 271–294. MR2275598 questions for function fields over complete discretely valued
[2] Richard Elman, Nikita Karpenko, and Alexander Merkur- fields, Invent. Math. 197 (2014), no. 1, 215–235, DOI
jev, The algebraic and geometric theory of quadratic forms, 10.1007/s00222-013-0483-y. MR3219517
American Mathematical Society Colloquium Publications, [19] Albrecht Pfister, Quadratic forms with applications to alge-
vol. 56, American Mathematical Society, Providence, RI, braic geometry and topology, London Mathematical Society
2008, DOI 10.1090/coll/056. MR2427530 Lecture Note Series, vol. 217, Cambridge University Press,
[3] David Grimm, Lower bounds for Pythagoras numbers of func- Cambridge, 1995, DOI 10.1017/CBO9780511526077.
tion fields, Comment. Math. Helv. 90 (2015), no. 2, 365– MR1366652
375, DOI 10.4171/CMH/356. MR3351748 [20] Florian Pop, On the Pythagoras number of function fields of
[4] David Harbater, Julia Hartmann, and Daniel Krashen, Ap- curves over number fields, Israel J. Math. 257 (2023), no. 2,
plications of patching to quadratic forms and central simple 561–574, DOI 10.1007/s11856-023-2548-y. MR4682929
algebras, Invent. Math. 178 (2009), no. 2, 231–263, DOI [21] Alexander Vishik, Motives of quadrics with applications
10.1007/s00222-009-0195-5. MR2545681 to the theory of quadratic forms, Geometric methods in
[5] Detlev W. Hoffmann, Isotropy of quadratic forms over the the algebraic theory of quadratic forms, Lecture Notes in
function field of a quadric, Math. Z. 220 (1995), no. 3, 461– Math., vol. 1835, Springer, Berlin, 2004, pp. 25–101, DOI
476, DOI 10.1007/BF02572626. MR1362256 10.1007/978-3-540-40990-8_2. MR2066515
[6] Detlev W. Hoffmann, Pythagoras numbers of fields, J. [22] Alexander Vishik, Fields of 𝑢-invariant 2𝑟 + 1, Algebra,
Amer. Math. Soc. 12 (1999), no. 3, 839–848, DOI arithmetic, and geometry: in honor of Yu. I. Manin. Vol.
10.1090/S0894-0347-99-00301-X. MR1670858 II, Progr. Math., vol. 270, Birkhäuser Boston, Boston, MA,
[7] Oleg T. Izhboldin, Fields of 𝑢-invariant 9, Ann. of Math. 2009, pp. 661–685, DOI 10.1007/978-0-8176-4747-6_22.
(2) 154 (2001), no. 3, 529–587, DOI 10.2307/3062141. MR2641206
MR1884616 [23] Alexander Vishik, Excellent connections in the motives of
[8] Uwe Jannsen, Hasse principles for higher-dimensional fields, quadrics (English, with English and French summaries),
Ann. of Math. (2) 183 (2016), no. 1, 1–71, DOI Ann. Sci. Éc. Norm. Supér. (4) 44 (2011), no. 1, 183–195,
10.4007/annals.2016.183.1.1. MR3432580 DOI 10.24033/asens.2142. MR2760197
[9] Nikita A. Karpenko, On the first Witt index of quadratic [24] Vladimir Voevodsky, Motivic cohomology with 𝐙/2-
forms, Invent. Math. 153 (2003), no. 2, 455–462, DOI coefficients, Publ. Math. Inst. Hautes Études Sci. 98 (2003),
10.1007/s00222-003-0294-7. MR1992018 59–104, DOI 10.1007/s10240-003-0010-6. MR2031199
[10] Nikita Karpenko and Alexander Merkurjev, Essential di-
mension of quadrics, Invent. Math. 153 (2003), no. 2, 361–
372, DOI 10.1007/s00222-003-0292-9. MR1992016
[11] Nikita A. Karpenko, Holes in 𝐼 𝑛 (English, with
English and French summaries), Ann. Sci. École
Norm. Sup. (4) 37 (2004), no. 6, 973–1002, DOI
10.1016/j.ansens.2004.07.002. MR2119244
[12] Nikita A. Karpenko, Upper motives of algebraic groups and
incompressibility of Severi-Brauer varieties, J. Reine Angew.
Math. 677 (2013), 179–198, DOI 10.1515/crelle.2012.011.
MR3039776
[13] T. Y. Lam, Introduction to quadratic forms over fields, Grad- Alexander Raman Parimala
uate Studies in Mathematics, vol. 67, American Mathemat- Merkurjev
ical Society, Providence, RI, 2005, DOI 10.1090/gsm/067. Credits
MR2104929
[14] David B. Leep, The 𝑢-invariant of 𝑝-adic function Photo of Alexander Merkurjev is courtesy of Alexander
fields, J. Reine Angew. Math. 679 (2013), 65–73, DOI Merkurjev.
10.1515/crelle.2012.029. MR3065154 Photo of Raman Parimala is courtesy of Raman Parimala.
[15] A. S. Merkur′ev, Simple algebras and quadratic forms
(Russian), Izv. Akad. Nauk SSSR Ser. Mat. 55 (1991), no. 1,

718 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY VOLUME 72, NUMBER 7

 Topology Meets Machine
Learning: An Introduction
Using the Euler
Characteristic Transform
Bastian Rieck
Machine learning is shaping up to be the transformative are finding a home in the nascent field of topological deep
technology of our times: Many of us have interacted with learning [PBB+ 24].
models like ChatGPT, new breakthroughs in applications
like healthcare are announced on an almost daily basis, 1. What is Machine Learning?
and novel avenues for integrating these tools into sci- Before diving into concrete examples, let us first take a
entific research are opening up, with some mathemati- step back and introduce the field of machine learning. Ma-
cians already using large language models as proof assis- chine learning is broadly construed as the art and science
tants [Ghr25, Won24]. of employing algorithms to solve tasks, but with the added
Written for an audience of mathematicians with limited restriction that the algorithm should—in a certain sense
prior exposure to machine learning, this article aims to dis- that will become clear below—adapt itself to the data at
pel some myths about the field, hoping to make it more hand. To substantiate this informal definition, we need
welcoming and open. Indeed, from the outside, machine to introduce some terms, beginning with the notion of
learning might look like a homogeneous entity, but in fact, a feature, which denotes a quantity derived from “raw”
the field is highly diverse, like mathematics itself. While input data. For instance, given a set of chemical struc-
the main thrust of the field arises from engineering ad- tures, the number of carbon atoms could be a feature.
vances, with bigger and better models, there is also plenty Adding other features—such as the number of oxygen
of space for mathematics and mathematicians. As a run- atoms, ring structures, or aromatic bounds—then yields
ning example of how mathematics—beyond linear algebra a high-dimensional representation of a chemical structure
and statistics, the classical drivers of machine learning— in the form of a feature vector, which, for convenience, we
may enrich the field, this article focuses on topology, which assume to live in some ℝ𝑑 .
recently started providing novel insights into the founda- Machine-learning algorithms may thus, on a very high
tions of machine learning: Point-set topology, harnessing level, be seen as functions of the form 𝑓 ∶ ℝ𝑑 → 𝒟, where
concepts like neighborhoods, can be used to extend ex- 𝒟 indicates some domain of interest. When 𝒟 = ℝ, we say
isting algorithms from graphs to cell complexes [HIZ20]. that this is a regression task,1 whereas when 𝒟 is a set, we are
Algebraic topology, making use of effective invariants like dealing with a classification task. As a running example, sup-
homology, improves the results of models for 3D shape re- pose we are interested in classifying chemical structures.
construction [WAM+ 22]. Finally, differential topology, pro- The domain 𝒟 could consist of the labels “toxic” and
viding tools to study smooth properties of data, results in “harmless.” Since we have access to labels, our classifica-
efficient methods for analyzing geometric simplicial com- tion task is an example of supervised machine learning.2 Not
plexes [MHR24]. These (and many more) research strands every function 𝑓 is equally useful in this context, though.
If 𝑓 would always predict “toxic,” it would not be a suitable
Bastian Rieck is a professor of computer science at University of Fribourg. His function for our task (even though its prediction might be
email address is [email protected]. the safest bet). The allure of machine learning lies in the
Communicated by Notices Associate Editor Emilie Purvine.
1This also generalizes to predicting more than one value.
For permission to reprint this article, please contact: 2In unsupervised machine learning, by contrast, input data are not labeled,
[email protected].
requiring models to “learn” characteristic properties or patterns of the data-
DOI: https://doi.org/10.1090/noti3193 generation process such as clusters.

AUGUST 2025 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY 719

fact that it provides a framework to learn a suitable func- goal of minimizing Equation (3). To this end, we repeat-
tion 𝑓 by presenting the algorithm with numerous exam- edly perform the prediction for a set of predefined samples,
ples of different chemical structures, in the hope that with the so-called training data set, which is presented to the neu-
sufficient data, the underlying mechanism(s) driving toxi- ral network in equal-sized batches (as opposed to using the
city can be derived. Thus, when we show this function a full data set as an input, which would often be prohibitive
new example, its guess as to whether it is toxic or not is in terms of memory requirements). Once we are satisfied
based on all previously seen examples. The field of ma- with the results, we evaluate the prediction on the test data
chine learning has developed many suitable models for set, which crucially, must be kept separate from the train-
finding or approximating such functions. Chief among ing data set.
those is the concept of deep neural networks.3 Deep-learning A common myth concerning deep learning is that it
models started a veritable revolution in some fields like “just” performs curve-fitting, arguably in a highly elabo-
computer vision, mostly because (i) they obviate the need rate way. While this article cannot possibly counter all
for “hand-crafted” features, as used at the beginning of this such claims, it should be pointed out that some neural
section, and (ii) they consist of standalone building blocks, networks can approximate functions in certain function
i.e., layers, that can be easily combined to build new mod- spaces arbitrarily well. This property is also known as
els for solving domain-specific problems. universal function approximation and implies that a certain
Initially, all layers of a deep neural network start with class of neural networks is dense (usually with respect to
a random set of parameters or weights, which are then the compact convergence topology) in the function space.
subsequently adjusted to minimize a loss function. A sim- For instance, feedforward neural networks are known to
ple deep neural network—a fully connected neural network— be able to approximate any Borel-measurable function be-
(𝑙) (𝑙−1) tween finite-dimensional spaces [HSW89]. Similar theo-
produces an output 𝑦𝑖 ∈ ℝ𝑑𝑙 based on an input 𝑦𝑖 ∈
ℝ 𝑑𝑙−1
via the recursion rems exist for other architectures and this property is often
(𝑙) (𝑙−1) invoked when discussing the merits of deep learning: A
𝑦𝑖 ≔ 𝜎 (𝑊 (𝑙) 𝑦𝑖 + 𝑏(𝑙) ) , (1) deep neural network with sufficient data may be the sim-
plest way to approximate functions that do not permit an-
where 𝑙 ∈ {1, … , 𝐿} ranges over the layers, 𝑊 (𝑙) denotes a
alytical solutions.
𝑑𝑙 × 𝑑𝑙−1 matrix of weights, 𝑏(𝑙) ∈ ℝ is a bias term, and 𝜎
The conceptual simplicity and modularity of deep neu-
refers to a nonlinear activation function (like a sigmoid or
(0) ral networks is not without its downsides, though. To
tanh). Any original input 𝑦𝑖 ∈ ℝ𝑑 is thus transformed perform well, deep neural networks typically require enor-
4
in a nonlinear fashion, resulting in a final prediction 𝑦 𝑖 ≔ mous amounts of data as well as a humongous number
(𝐿)
𝑦𝑖 ∈ ℝ𝑑𝐿 , which is often normalized to [0, 1] using a of parameters. Unless specific provisions are taken, deep
softmax function, i.e., neural networks are not energy efficient, and their data re-
exp ([𝑦 𝑖 ]𝑗 ) quirements still pose serious obstacles in many domains.
[𝑦 𝑖 ]𝑗 ↦ 𝑑𝐿
, (2) In addition, neural networks are still black box models
∑𝑘=1 exp ([𝑦 𝑖 ]𝑘 ) with opaque outputs. Despite research in interpretable ma-
where [𝑦 𝑖 ]𝑗 denotes the 𝑗th component of 𝑦 𝑖 . In our run- chine learning aiming to improve this state of affairs, there
ning example, binary cross-entropy (BCE) would then be a are still few models whose outputs can be checked eas-
suitable loss function. Mapping {“harmless,” “toxic”} to ily by human operators, for instance. Even large language
{0, 1}, the BCE loss for 𝑛 predictions is models (LLMs), arguably one of the most impressive feats
of the field, suffer from shortcomings and are incapable
𝑛
1 of assessing their own outputs. Like other deep neural
∑ 𝑦ˆ log(𝑦 𝑖 ) + (1 − 𝑦ˆ𝑖 ) log(1 − 𝑦 𝑖 ), (3)
𝑛 𝑖=1 𝑖 networks, they also remain vulnerable to adversarial at-
tacks, i.e., inputs that are generated to provoke a bad re-
where 𝑦ˆ𝑖 ∈ {0, 1} represents the true label of a sample sponse, often without the original user of the model be-
and 𝑦 𝑖 ∈ [0, 1] refers to the softmax-normalized predic- ing aware of it [CAD+ 21]. Adversarial examples for vision
tion of the neural network, which we interpret as the prob- models can appear innocuous to a human observer but
ability of the sample being in class 1, i.e., being toxic. The can cause incorrect predictions: For example, traffic signs
parameters of the neural network are now subsequently ad- can be slightly remodeled using adhesive tape, making a
justed using gradient descent or similar procedures, with the vision system unable to detect them. This dispels the com-
mon myths that (i) deep neural networks are always ro-
3See Schmidhuber [Sch22] for a “deep dive” into the respective algorithms and bust, and that (ii) deep-learning research is complete (in
their origin stories. the sense that, moving forward, there will not be any novel
4In practice, layers can also perform different operations or transformations to
conceptual insights). On the contrary—there are still
the input data.

720 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY VOLUME 72, NUMBER 7

 While a single number 𝜒(𝐾) is insufficient to fully char-
acterize a shape, it is possible to lift it to a multiscale sum-
mary known as the Euler Characteristic Transform (ECT), in-
troduced by Turner et al. [TMB14]. The ECT requires a geo-
Polyhedron 𝜒 Vertices Edges Faces metric simplicial complex 𝐾 in ℝ𝑑 . We can think of such a
complex as having an associated coordinate 𝑥𝑣 for each
Tetrahedron 2 4 6 4 vertex 𝑣 of a simplex 𝜎 such that every face of 𝜎 is in 𝐾 and
Cube 2 8 12 6 the intersection of two simplices in 𝐾 is either empty or
Octahedron 2 6 12 8 a face of both. Taking now any direction 𝑤 ∈ 𝕊𝑑−1 , i.e.,
Dodecahedron 2 20 30 12 any point on the (𝑑 − 1)-sphere, we obtain a real-valued
Icosahedron 2 12 30 20 function defined for all simplices of 𝐾 via
Figure 1. The five platonic solids (tetrahedron, cube, 𝑓𝑤 ∶ 𝐾 → ℝ
octahedron, dodecahedron, and icosahedron) all have Euler 𝜎 ↦ max⟨𝑥𝑣 , 𝑤⟩, (5)
characteristic 𝜒 = 2. 𝑣∈𝜍

where ⟨⋅, ⋅⟩ denotes the standard Euclidean inner prod-

fundamental challenges that necessitate an analysis of the uct. In the parlance of computational topology, this func-
very foundations of deep learning. It is here that topology tion 𝑓𝑤 can be used to obtain a filtration of 𝐾 in terms of
and mathematics in general can provide a new perspective. its subcomplexes: Given a threshold 𝑡 ∈ ℝ, we define
𝐾𝑡,𝑤 ≔ {𝜎 ∈ 𝐾 ∣ 𝑓𝑤 (𝜎) ≤ 𝑡}. Evaluating the Euler char-
2. Euler Characteristics Galore
acteristic of each 𝐾𝑡,𝑤 then yields the Euler Characteristic
We focus our discussion on the Euler Characteristic Trans- Curve (ECC) associated to the direction 𝑤. Referring to
form (ECT). Being an invariant that bridges geometry and the set of all finite geometric simplicial complexes as 𝒦,
topology, it is perfectly suited for such an overview article the ECC is an integer-valued function of the form
because it provides connection points for the largest num-
ber of researchers. The ECT is based on the concept of the ECC ∶ 𝒦 × 𝕊𝑑−1 × ℝ → ℤ
(6)
Euler characteristic. This integer-based quantity serves as a (𝐾, 𝑤, 𝑡) ↦ 𝜒(𝐾𝑡,𝑤 ).
summary statistic of the “shape” of a graph or simplicial The ECC thus maps a simplicial complex 𝐾, together with
complex. We define an (abstract) simplicial complex as a direction 𝑤 and a threshold 𝑡, to the Euler character-
a family of sets that is closed under taking subsets (also istic of its corresponding subcomplex. Figure 2 depicts
known as faces). A simplicial complex generalizes a graph the ECC calculation procedure for a mesh, i.e., a geomet-
by permitting more than mere dyadic relations. We refer to ric simplicial complex in ℝ3 , consisting of 276, 497 ver-
the elements (sets) of a simplicial complex as its simplices tices, 821, 475 edges, and 604, 393 faces (triangles). We con-
and, given a simplex {𝑣 0 , … , 𝑣 𝑖 }, we say that its dimension is 𝑖. sider a single filtration along the 𝑦-axis of the mesh, i.e.,
A 𝑝-dimensional simplicial complex thus consists of sim- 𝑤 = (0, 1, 0), sweeping the spaceship from bottom to top
plices of dimension up to and including 𝑝. For example, via 𝐾𝑡,𝑤 , where 𝑡 ∈ [2.93, 5.22] as a consequence of the
graphs can be considered 1-dimensional simplicial com- range of the vertex coordinates of the mesh. For each 𝑡,
plexes, consisting of edges (dimension 1) and vertices (di- we calculate the Euler characteristic 𝜒(𝐾𝑡,𝑤 ), leading to the
mension 0). Given a 𝑝-dimensional simplicial complex 𝐾, ECC associated to the direction 𝑤.
we write 𝐾 (𝑖) to refer to its simplices of dimension 𝑖. The While more “expressive” than a single Euler character-
Euler characteristic of 𝐾 is then defined as an alternating istic, the ECC still depends on the choice of direction. By
sum of simplex counts, i.e., using multiple directions, we may hope to obtain a bet-
𝑝 ter representation of 𝐾. This leads to the Euler Characteris-
𝜒(𝐾) ≔ ∑ (−1)𝑖 ||𝐾 (𝑖) || . (4) tic Transform, which is defined as the function that assigns
𝑖=0 each direction 𝑤 ∈ 𝕊𝑑−1 to its corresponding ECC, i.e.,
The Euler characteristic affords several equivalent defini- ECT ∶ 𝒦 × 𝕊𝑑−1 → ℤℝ
tions (for instance, in terms of the Betti numbers of 𝐾), gen- (7)
(𝐾, 𝑤) ↦ ECC(𝐾, 𝑤, ⋅),
eralizes to different objects (graphs, simplicial complexes,
polyhedra, . . . ), and is a topological invariant, meaning that where ℤℝ denotes the set of functions from ℝ to ℤ. Turner
if two spaces (objects) are homotopy equivalent, their Eu- et al. [TMB14] proved that the ECT serves as a sufficient
ler characteristic is the same. This is depicted in Figure 1 by “shape statistic,” yielding an injective mapping for dimen-
means of the classical example of the five platonic solids. sions 2 and 3 if an infinite number of directions is used.
Since all these spaces are homotopy-equivalent to the 2- Thus, given two simplicial complexes 𝐾 ≠ 𝐾 ′ , we have
sphere, they all share Euler characteristic 2. ECT(𝐾, ⋅) ≠ ECT(𝐾 ′ , ⋅). This result was later generalized to

AUGUST 2025 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY 721

 y
60 · 103
50 · 103
40 · 103

χ(Kt,w )
x 30 · 103
20 · 103
10 · 103
0 · 100
2.5 3 3.5 4 4.5 5 5.5
t
z

Geometric simplicial complex Filtration in y-direction Euler Characteristic Curve

Figure 2. An illustration of the Euler Characteristic Curve calculation procedure. Starting from a 3D model, specified as a mesh
with vertices, edges, and faces, we perform a single filtration in the direction of the 𝑦-axis. The filtration values are shown in the
cross-section of the shaded model, with warmer colors corresponding to larger filtration values. Finally, the Euler Characteristic
Curve is plotted for all thresholds. (The mesh depicts the “Enterprise NCC-1701-D,” from Star Trek: The Next Generation. It was
originally created by Moreno Stefanuto and made available under a “Free Standard” license. This paper depicts a modified
version by the author.)

arbitrary dimensions by Ghrist et al. [GLM18] and Curry considered a Radon transform, using 𝑋 = ℝ𝑑 and 𝑌 =
et al. [CMT22],5 who also showed that, somewhat surpris- 𝕊𝑑−1 × ℝ, with 𝑘(𝑥, (𝑤, 𝑡)) being the indicator function on
ingly, a finite number of directions is sufficient to guarantee {(𝑥, (𝑤, 𝑡)) ∣ ⟨𝑥, 𝑤⟩ ≤ 𝑡}. Since prior work already shows un-
injectivity. der which conditions one can recover the input to a Radon
We briefly recapitulate the proof idea underlying Ghrist transform, Ghrist et al. [GLM18] obtain a short, elegant
et al. [GLM18], which draws upon Euler calculus on o- proof of the fact that the ECT is not only injective but also
minimal structures to only permit inputs that are “well- invertible.
behaved” or “tame.” An o-minimal structure 𝒪 = {𝒪𝑑 }
over ℝ consists of a collection of subsets 𝒪𝑑 ⊂ ℝ𝑑 , 3. Using the Euler Characteristic Transform
which are closed under both intersection and complement. in a Machine-Learning Model
Moreover, 𝒪 needs to satisfy certain axioms, including6 Given the invertibility results and the fact that the ECT is
being closed under cross products and containing all al- highly efficient to implement, its integration into machine-
gebraic sets. In addition 𝒪1 must consist of finite unions learning models is only logical. Before we discuss how to
of open intervals and points. The sets in 𝒪 are called de- use the ECT with a deep-learning model, let us first ponder
finable or tame. Letting 𝑋 be a definable subset (intuitively, some of its practical uses with classical7 machine-learning
𝑋 can be seen as a generalized variant of a geometric sim- methods like support vector machines. Central to such al-
plicial complex) of ℝ𝑑 , the constructible functions on 𝑋 are gorithms is their use of handcrafted features; while often
integer-valued functions with definable level sets, denoted eschewed in deep learning,8 the power of such features
by ℎ ∶ 𝑋 → ℤ. Letting CF(𝑋) refer to the set of con- should not be underestimated—in particular when com-
structible functions on 𝑋, the Euler integral on 𝑋 is the func- bined with methods like the ECT.
tional ∫𝑋 ⋅d𝜒 ∶ CF(𝑋) → ℤ that maps 𝟙𝜍 ↦ (−1)dim 𝜍 for The results by Curry et al. [CMT22] indicate that one
each simplex 𝜎. Given definable subsets 𝑋, 𝑌 and a con- could use a finite set of directions to calculate the ECT.
structible function 𝑘 ∈ CF(𝑋 × 𝑌 ), the Radon transform is However, it is unclear how to find such directions. This
defined as leaves us with the option to sample directions 𝑊 ≔
{𝑤 1 , … , 𝑤 𝑘 } ⊂ 𝕊𝑑−1 , and consider the ECT restricted to
ℛ𝑘 ℎ(𝑦) ≔ ∫ ℎ(𝑥)𝑘(𝑥, 𝑦)d𝜒 (𝑥). (8) 𝑊. The hope is that the sample is sufficient to capture
𝑋

Ghrist et al. [GLM18] now show that the ECT can be 7While this is the common nomenclature as used in the machine-learning com-
munity, it should be pointed out that some of these methods predate deep learn-
5Despite their different publication dates, both works appeared concurrently as ing by a couple of years only. This indicates the rapid pace of the field.
preprints. 8It is a common myth that classical machine-learning techniques are always
6This expository article strives for clarity. Interested readers are invited to dive outperformed by deep learning; in fact, there are many applications in which
into Curry et al. [CMT22] and the references therein to learn more about o- such techniques exhibit strong performance while remaining computationally
minimal structures. efficient.

722 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY VOLUME 72, NUMBER 7

shape characteristics. Another discretization applied in
practice involves the choice of thresholds. For example,
50000
if each coordinate 𝑥𝑣 has at most unit norm, we know

Filtration value t

Filtration value t
that ⟨𝑥𝑣 , 𝑤⟩ ∈ [−1, 1]. Thus, we may sample thresholds 40000
𝑇 ≔ {𝑡1 , … , 𝑡 𝑙 } ⊂ [−1, 1] and evaluate each ECC for a spe- 30000
cific direction 𝑤 only at thresholds in 𝑇. This discretization
20000
allows us to represent the ECT as a matrix, where columns
are indexed by 𝑊 and rows are indexed by 𝑇. In this repre- 10000

sentation, each column corresponds to the ECC for some

Direction w Direction w
direction 𝑤 and all ECCs are aligned in the sense that a row (a) Raw (b) Column-normalized
index 𝑖 within any column corresponds to the same thresh- Figure 3. Visualizing the ECT of the model in Figure 2 as a
old 𝑡 𝑖 ∈ [−1, 1]. An alternative discretization involves pick- discretized image using either the raw 𝑡-values (left), or a
ing a set of 𝑙 thresholds 𝑇𝑤 for each direction 𝑤 individu- rescaled version (right) where each column is normalized
ally. While each column of the resulting matrix still repre- individually to [−1, 1]. Each 𝑥-axis shows 256 randomly
sents an ECC, the same row index 𝑖 within columns now selected directions, while each 𝑦-axis corresponds to the
filtration values 𝑡, with the lowest value starting at the top of
generally corresponds to different thresholds, depending
the image. Every column in the image corresponds to an
on the set of directions 𝑊. Both discretization strategies individual ECT, with the pixel color indicating the value for a
have their merits. The first strategy is preferable when the specific filtration threshold 𝑡.
input data is contained in (or normalized to) a unit sphere;
the second strategy can be useful in situations in which size olds and directions, but is seen to lead to good results in
differences between individual ECCs do not matter. practice [Mun25].
Regardless of the discretization strategy, the resulting How can we move beyond such handcrafted features
matrix can be viewed as an image, although care must be and use deep learning to find task-specific ECTs? The an-
taken to understand that the ordering of directions, i.e., swer lies in picking a different representation, which en-
the columns of the matrix/image, is entirely arbitrary. Fig- ables us to learn an appropriate set of directions 𝑊 as op-
ure 3 depicts this visualization for 256 randomly selected posed to sampling it. One obstacle to overcome here in-
directions. When using the “raw” values of 𝑡 to index volves the fact that the ECT is a discrete quantity based
the rows, we observe substantially more zeros. This is on step functions. Such functions are at odds with deep-
because we chose thresholds based on the global mini- learning models, which prefer their inputs to be smooth
mum and maximum across all filtrations, meaning that and differentiable. While different variants of the ECT ex-
any individual ECT, whose range is typically much smaller ist, for instance a smooth one obtained by mean-centering
than the global one, rises to its respective maximum very and integration [Mun25], we may also apply another
quickly. The column-normalized version, by contrast, ex- trick often found in machine-learning research: Instead of
hibits more variability, with different columns appear- working with the original definition of a function, we just
ing to have a smaller width—this is a perceptual illusion, work with a smooth approximation of it! While this tech-
though, arising from the fact that each column now grad- nically solves a different problem than what we originally
ually increases in intensity. Unlike the “raw” version, care set out to achieve, it often works surprisingly well. Regard-
must be taken when considering differences between indi- less of the specific approximation method, this procedure
vidual columns. will permit us to use the ECT as a differentiable building
Setting aside the dissimilarities between the two vari- block of deep-learning models. The ECT may thus be said
ants, it is possible to compare shapes by calculating an to constitute an inductive bias, a term that refers to the spe-
appropriate distance between their respective ECT images. cific assumptions built into a machine-learning model to
Such approaches work best when shapes are aligned to a affect its inner workings and results.9
shared coordinate system [Mun25] since any column per- To approximate the ECT, we first notice that any ECC
mutation of the resulting matrix describes the same ECT in can be written as a sum of indicator functions, i.e., we rewrite
the sense that there is no canonical ordering of different di- Equation (6) as
rections in higher dimensions. The matrix representation 𝑝
can also be used directly as an input to a machine-learning (𝐾, 𝑤, 𝑡) ↦ ∑ (−1)𝑖 ∑ 𝟙≤𝑡 (𝑓𝑤 (𝜎)), (9)
model by “flattening” the matrix, thus effectively turning 𝑖=0 (𝑖)
𝜍∈𝐾𝑡,𝑤
the matrix into a long feature vector. Similar to our initial
example with chemical structures, this process enables us
9The quintessential example of an inductive bias involves the locality assump-
to represent a complex shape as a fixed-size vector. Such a
representation crucially depends on the choices of thresh- tion of convolutional neural networks, i.e., the focus on small “patches” of an
input image.

AUGUST 2025 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY 723

 (𝑖)
where 𝐾𝑡,𝑤 denotes all 𝑖-simplices of 𝐾𝑡,𝑤 , and 𝑓𝑤 refers 1
to the filtration from Equation (5). Equivalently, Equa-
tion (7) permits a rephrasing via indicator functions. So
far, this is still an exact expression. The approximation
comes into play when we notice that an indicator function 0.5
can be replaced by a sigmoid function, i.e., 𝑆(𝑥) = 1/1+exp(−𝑥).
This lets us define an approximate ECC by rewriting Equa-
tion (9) as
𝑝 0
(𝐾, 𝑤, 𝑡) ↦ ∑ (−1)𝑖 ∑ 𝑆 (𝜆 (𝑡 − 𝑓𝑤 (𝜎))) , (10)
𝑖=0 (𝑖) −4 −2 0 2 4
𝜍∈𝐾𝑡,𝑤
Figure 4. An indicator function (dashed) and its
where 𝜆 ∈ ℝ denotes a scaling parameter that controls approximation using differently scaled sigmoid functions,
the tightness of the approximation. Figure 4 depicts this with 𝜆 ∈ {1, 2, 3, 4, 5}.
approximation for various values; we can see that 𝑆(𝜆𝑥)
is indeed a suitable (smooth) replacement for an indica-
directly on its own. As an example of the latter case, we
tor function. This minor modification has major advan-
can for instance compare two outputs of an ECT layer using
tages: Each of the summands in Equation (10) is differen-
a loss function. Treating the outputs as high-dimensional
tiable with respect to 𝑤, 𝑡, and 𝑓𝑤 (⋅). It is thus possible
vectors 𝑥 and 𝑦, we can use the squared 𝑙2 -norm of their
to use Equation (10) to transform a geometrical simplicial
difference, normalized by the dimension, as a criterion of
complex in a way that is fundamentally compatible with a
how well they are aligned. This loss function is also known
deep-learning model.
as the mean squared error (MSE). It is commonly used to
Lest we get lost in implementation details, it is bet-
solve problems in representation learning, so it is perfectly
ter to assume a more elevated position and take stock of
suited for a small experiment: Suppose we have a simple
what we already have: Using the approximation above, we
data set in ℝ2 whose (discretized) ECT we know. Start-
have turned the ECT from a discrete function into a func-
ing from a randomly initialized set of directions, can we
tion that continuously depends on its input parameters. In
learn the “best” set of directions to align the two ECTs? It
machine-learning terminology, we may thus treat this cal-
turns out that the approximation scheme defined in Equa-
culation as a layer. Assuming that all coordinates have at
tion (10) indeed permits solving such problems. In our
most unit norm so that ⟨𝑥𝑣 , 𝑤⟩ ∈ [−1, 1], our ECT layer has
toy example, we start with 32 uniformly sampled direc-
two hyperparameters, namely (i) the number 𝑙 of thresh-
tions, which we can parametrize using a single angle. We
olds to discretize [−1, 1], and (ii) the number 𝑘 of direc-
then sample the same number of angles from a normal
tions. The input to our layer is a geometric simplicial com-
distribution and minimize the MSE loss between the ECT
plex 𝐾, and the output is an 𝑙 × 𝑘 matrix (equivalently, we
based on the random directions to the “original” ECT. Fig-
may apply such a construction to all types of spaces that
ure 5 depicts this experiment, and we observe that only
afford an Euler characteristic, including cell complexes, for
a couple of optimization steps—using gradient descent, for
instance). Notice that when learning an ECT for a data set
example—are required to obtain a suitable approximation
of different objects, one matrix (i.e., the discretized image
to our input ECT. Moreover, this approximation gets better
of an ECT) for each sample in the input batch is returned,
with longer training times and more discretization steps.
resulting in a multidimensional array output, which is also
However, learning the directions could be considered
referred to as a tensor.10 Given the continuous dependence
somewhat solipsistic in that we derived a new represen-
on its input parameters, we refer to the construction above
tation (the ECT) and then showed that we can learn its
as the Differentiable Euler Characteristic Transform [RR24].
parameters. Indeed, this experiment is more of a “smoke
4. Two Applications of Differentiable Euler test,” since learning the parameters of a representation is
Characteristic Transforms the very basis of machine learning! To build a stronger ex-
The ECT layer we defined above gives rise to a fixed-size vec- periment showcasing the power of the ECT, let us consider
torial representation, which may either be integrated into learning coordinates instead. Recall that our approxima-
a larger neural network—thus effectively handing off the tion from Figure 4 permits us to also affect the coordinates
ECT results for deeper processing—or which can be used themselves: Any loss function that we evaluate will depend
on the directions and the coordinates; thus, if we make
10This terminology can be confusing for mathematicians at first since machine the coordinates a parameter of our layer, we can modify
learning does not (always) make use of any of the properties that would make them similarly to the example shown above. Using an
up a “mathematical” tensor. MSE loss, this time we search for input coordinates that

724 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY VOLUME 72, NUMBER 7

 Filtration value t

Filtration value t

Filtration value t
Direction w Direction w Direction w
(a) Original (b) Untrained (c) Trained

Figure 5. (a) We can learn the directions required to match the ECT of a simple data set. (b) Starting with a set of random
directions, our initial “guess” of the ECT looks nothing like the original. (c) After several optimization steps, which minimize the
dissimilarity between “our” ECT and the original one, we obtain a good approximation. We have thus learned a set of suitable
directions for calculating the ECT.

which employ a mechanism called message passing that

amounts to locally transporting information via the edges
of a graph.11 The ECT offers a new and surprisingly compet-
itive method for classifying such graphs [RR24]. Together
with its extremely small memory footprint—recall that the
ECT is essentially “just” counting the constituent parts of
an input object—this makes the ECT an interesting para-
digm to consider for new machine-learning applications.
Some of these applications are already discussed by Turner
et al. [TMB14], while others, including the experiments for
(a) Original (b) Untrained (c) Trained learning directions and coordinates, have been introduced
Figure 6. We can also learn the coordinates of a point cloud in our recent work [RR24]. As always, there is more work to
based on the ECT. (a) We sample 100 points from a double be done, some enticing directions being (i) an assessment
annulus. (b) The initial configuration of our coordinates is a
random sample from a noisy circle. (c) After minimizing the
of the theoretical expressivity of the ECT when it comes to
coordinates based on measuring the dissimilarity between distinguishing between geometric graphs, (ii) the analysis
the “desired” ECT and the current one, we are able to of the inverse problem, i.e., the reconstruction or generation
approximate the input point cloud closely (red points show of objects based on an ECT (in the discrete setting), and
the learned coordinates, black points the original ones). (iii) algorithmic aspects that enable the ECT to perform
efficiently in the context of high-volume geometry-based
minimize the differences of our ECT to a target ECT streaming data arising from LiDAR sensors, for instance.
(machine-learning researchers also like to use the term
ground truth). Figure 6 depicts an example, using 𝑘 = 256 5. The Future of Topology in Machine Learning
directions and 𝑙 = 256. We again manage to learn suitable The ECT served as the leitmotif of this article, demonstrat-
coordinates that align our two point clouds. This can be ing how to connect concepts from applied topology and
very helpful when working with coarser representations of modern machine-learning methods with relatively few
data, i.e., we may use this procedure to find the best ap- hitches. The reader is strongly recommended to check out
proximation of a downsampled point cloud to its original an excellent overview article by Munch [Mun25] to learn
version, leading to an ECT-based compression algorithm. more about it. However, toward the end, let us briefly
These tasks provide a glimpse of the versatility of the zoom out and consider the larger picture (knowing full
ECT. While learning directions or coordinates is not neces- well that predictions are hard, especially when concerning
sarily a machine-learning task sui generis, a computational the future). By design, this article could merely scratch the
layer based on the ECT permits more flexible applications. surface, but the author firmly believes that topology and
In conjunction with a loss term, we can employ the ECT topological concepts have a strong role to play in machine
in a variety of tasks. For instance, we can use it to clas-
sify geometric graphs, i.e., low-dimensional simplicial com- 11An expository article like this cannot possibly do justice to the large body of re-
plexes whose vertices and edges are embedded in some ℝ𝑛 . search available under that moniker. The reader is therefore invited to consult a
This task is typically dominated by graph neural networks, recent position paper for more details [Vel23].

AUGUST 2025 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY 725

learning. Many interesting directions are bound to fall into
ACKNOWLEDGMENTS. The author is indebted to
one of the following three areas:
Ernst Röell, Emily Simons, and the anonymous refer-
1. Learning functions on topological spaces such as sim- ees for their helpful comments, which served to sub-
plicial complexes or cell complexes. stantially improve this article.
2. Building hybrid models that imbue neural networks This work has received funding from the Swiss
with knowledge about topological structures in the State Secretariat for Education, Research, and Innova-
data. tion (SERI).
3. Analyzing qualitative properties of neural networks.
In the first area, topology can be used to generalize exist-
ing machine-learning paradigms to a larger variety of input References
data sets; this is appealing because not every data set “lives” [CAD+ 21] Anirban Chakraborty, Manaar Alam, Vishal Dey,
in a nice Euclidean space or a graph, and the inclusion of Anupam Chattopadhyay, and Debdeep Mukhopadhyay, A
survey on adversarial attacks and defences, CAAI Transactions
higher-order neighborhoods would provide a shift in per-
on Intelligence Technology 6 (2021), no. 1, 25–45.
spective, aiding knowledge discovery. Researchers in topo- [CMT22] Justin Curry, Sayan Mukherjee, and Katharine
logical data analysis (TDA) might feel particularly comfort- Turner, How many directions determine a shape and other
able in the second area since it is one of the declared aims sufficiency results for two topological transforms, Trans.
of TDA to understand topological features in data. Nev- Amer. Math. Soc. Ser. B 9 (2022), 1006–1043, DOI
ertheless, hybrid models do not necessarily have to draw 10.1090/btran/122. MR4505675
upon concepts from TDA; new and daring methods could [Ghr25] Robert Ghrist, AI & mathematics as a career (Jan-
for instance focus on computational aspects of Riemann- uary 2025), https://x.com/robertghrist/status
ian geometry like curvature or make use of new invariants /1883646365777236306.
[GLM18] Robert Ghrist, Rachel Levanger, and Huy Mai, Per-
like metric-space magnitude. Finally, the third area shifts
sistent homology and Euler integral transforms, J. Appl. Com-
the perspective and lets topology return to its roots in that put. Topol. 2 (2018), no. 1-2, 55–60, DOI 10.1007/s41468-
it can serve as a lens through which to study, for instance, 018-0017-1. MR3873179
the training behavior of neural networks. Understanding [HH19] Catherine F. Higham and Desmond J. Higham,
these training dynamics could lead to smaller, more effi- Deep learning: an introduction for applied mathemati-
cient models, but also shed some light on the soft under- cians, SIAM Rev. 61 (2019), no. 4, 860–891, DOI
belly of deep-learning models, namely their susceptibility 10.1137/18M1165748. MR4027841
to unstable training regimens, “adversarial” input data, or [HIZ20] Mustafa Hajij, Kyle Istvan, and Ghada Zamzmi,
their propensity for hallucinations. Cell complex neural networks, “Topological Data Analy-
Each of the three areas for new research has something sis and Beyond” Workshop at NeurIPS, 2020, https://
openreview.net/forum?id=6Tq18ySFpGU.
enticing to offer, not only for machine learning but also
[HSW89] Kurt Hornik, Maxwell Stinchcombe, and Halbert
for (applied) topology. There is vast potential for new White, Multilayer feedforward networks are universal approxi-
topology-aware models to serve as proof assistants or even mators, Neural Networks 2 (1989), no. 5, 359–366.
try to search for new conjectures and counterexamples. It [MHR24] Kelly Maggs, Celia Hacker, and Bastian Rieck, Sim-
is said that only those with their feet on rock can build cas- plicial representation learning with neural 𝑘-forms, Inter-
tles in the air. Topology can provide this rock upon which national Conference on Learning Representations, 2024,
robust machine-learning research can be built. Thinking https://openreview.net/forum?id=Djw0XhjHZb.
beyond topology, the author hopes that this article may [Mun25] Elizabeth Munch, An invitation to the Euler character-
also stir up the curiosity of mathematicians coming from istic transform, Amer. Math. Monthly 132 (2025), no. 1, 15–
25, DOI 10.1080/00029890.2024.2409616. MR4839811
other domains. Machine-learning research will only ben-
[PBB+ 24] Theodore Papamarkou, Tolga Birdal, Michael
efit from more inquisitive minds and there are countless Bronstein, Gunnar Carlsson, Justin Curry, Yue Gao,
things waiting to be discovered. To get a taste of potential Mustafa Hajij, Roland Kwitt, Pietro Liò, Paolo Di
directions, readers are invited to peruse an excellent exposi- Lorenzo, Vasileios Maroulas, Nina Miolane, Farzana Nas-
tory article on deep learning [HH19]. The author of this ar- rin, Karthikeyan Natesan Ramamurthy, Bastian Rieck, Si-
ticle also prepared additional scripts and literature on the mone Scardapane, Michael T. Schaub, Petar Veličković,
ECT, which are made available under https://topology Bei Wang, Yusu Wang, Guo-Wei Wei, and Ghada Za-
.rocks/ect. mzmi, Position: Topological deep learning is the new fron-
tier for relational learning, Proceedings of the 41st In-
ternational Conference on Machine Learning, 2024,
pp. 39529–39555, https://proceedings.mlr.press
/v235/papamarkou24a.html.

726 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY VOLUME 72, NUMBER 7

[RR24] Ernst Röell and Bastian Rieck, Differentiable Euler
characteristic transforms for shape classification, International
Conference on Learning Representations, 2024, https://
openreview.net/forum?id=MO632iPq3I.
[Sch22] Jürgen Schmidhuber, Annotated history of modern AI
and deep learning, 2022.
[TMB14] Katharine Turner, Sayan Mukherjee, and Doug M.
Boyer, Persistent homology transform for modeling shapes and
surfaces, Inf. Inference 3 (2014), no. 4, 310–344, DOI
10.1093/imaiai/iau011. MR3311455
[Vel23] Petar Veličković, Everything is connected: Graph neural Bastian Rieck
networks, Current Opinion in Structural Biology 79 (2023),
102538. Credits
[WAM+ 22] Dominik J. E. Waibel, Scott Atwell, Matthias All figures are courtesy of Bastian Rieck.
Meier, Carsten Marr, and Bastian Rieck, Capturing shape Part of Figure 2 is from public domain. Please find a copy of
information with multi-scale topological loss terms for 3D re- the license here: https://sketchfab.com/licenses.
construction, Medical Image Computing and Computer As- Photo of Bastian Rieck is courtesy of Bastian Rieck.
sisted Intervention, 2022, pp. 150–159.
[Won24] Matteo Wong, We’re entering uncharted territory
for math, The Atlantic (October 2024), https://www
.theatlantic.com/technology/archive/2024/10
/terence-tao-ai-interview/680153.

NEW FROM

Ergodic Theory
Simon Rubinstein-Salzedo, Euler Circle, Mountain View, CA
The Carus Mathematical Monographs, Volume 38; 2025; approximately 228 pages
• Softcover: List Price: US $79.00; Order Code: CAR/38
• eBook: List Price: US $75.00; Order Code: CAR/38.E
• Bundle (print & electronic): List Price: US $116.50;
Order Code: CAR/38.B
AMS and MAA member discounts apply.

Visit bookstore.ams.org/car-38

AUGUST 2025 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY 727

 CAREER

Advice Column
Veselin Jungic

I am a new visiting assistant professor and am teaching I’ll address your question in three steps. Firstly, when
large sections of an engineering calculus class. It is clear teaching engineering calculus classes, make clear to all
to me that a select few students in the class have mathe- your students that this is a math class and that your expec-
matical abilities well above the standard of the course and tations are that each student will use the correct mathemat-
enjoy the more theoretical aspects of the subject. How do ical notation and terminology consistently, will check all
I nurture the interests of these select few students while necessary conditions before applying a particular calculus
still having enough bandwidth to address the more rou- technique, and will fully justify their mathematical think-
tine concerns and types of questions posed by the rest of ing. Be your students’ role model when it comes to writing
my students? mathematics by writing your lecture notes and solutions
to assignments and exams in the same way you write your
—Open to ideas papers. Demonstrate in your lectures some of the calculus
nuisances, such as a limit in which L’Hôpital’s Rule is use-
Dear Visiting Professor, less, or an example of a function that is differentiable, but
You are right: In large engineering calculus classes, there whose derivative is not a continuous function. Explain the
are often students who could easily become successful importance of quantifiers in mathematical thinking. Prove
professional mathematicians if they choose that as a ca- things.
reer path. I still remember a conversation with one of Still, keep in mind that your engineering students come
my former engineering students. Impressed with the stu- from a different academic culture and that their academic
dent’s mathematical abilities, I asked why they do not values, jargon, and way of thinking may be different from
study mathematics. “I like to make things,” the student what you became used to as a student of mathematics.
answered. And yes, we must respect students’ choices and Be demanding, engineering students expect that from you,
decisions, but our responsibility as mathematics instruc- but do not be harsh.
tors is to create a learning environment where each of our Secondly, while the above approach to teaching calcu-
students is able to gain new mathematical knowledge to lus may resonate well with your best students, you may
the best of their abilities. In this light, your question is also wish to try to create additional opportunities for those
spot on. and other students to fulfill their mathematical potential.
But before I share with you some of my thoughts For example, you can add “The Problem of the Week” page
and practices on teaching engineering calculus classes, to your class website and invite students to submit their
a warning: Every engineering student has a demanding solutions to relatively challenging course-based questions.
course schedule. This makes engineering students, typi- To further spark students’ interest in the course, you may
cally smart, hardworking, and ambitious young people, wish to accompany each problem with a list of references
great time optimizers. Consequently, for many of them, and learning resources. Be imaginative: create calculus
adding more to their existing workload is out of the ques- tasks that require use of technology accessible to students,
tion. like Desmos, GeoGebra, or Python. Think about how to re-
ward each submission. You may also wish to add a bonus
Veselin Jungic is a professor emeritus at Simon Fraser University. His email question to each of your assignments and exams. The idea
address is [email protected]. is to create new learning opportunities without putting ad-
For permission to reprint this article, please contact: ditional time pressure on students.
[email protected]. But what I’d really like to recommend is that you invite
DOI: https://doi.org/10.1090/noti3187 your best students to meet you during your office hours.

728 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY VOLUME 72, NUMBER 7

This may create an opportunity for you to enrich the aca-
demic lives of some exceptional young people. Don’t miss
it! Some of the things that you may do during an ini-
tial meeting are to encourage students to learn LaTeX and AMS
MEMBERSHIP
to master common math software, like SAGE, Maple, or
Mathematica. Place calculus in the context of other math
courses and suggest to students to consider taking an intro-
duction to analysis class. Prepare a couple of appropriate
papers and invite your students to read and, as a follow
up, discuss them with you and among themselves. Great
is your key to unlocking the future
sources of such papers are, for example, Mathematics Maga-
Seeking your next career move? Let your AMS
zine, Crux Mathematicorum, The College Mathematics Journal,
and The American Mathematical Monthly. My personal fa- membership guide you. Choose the rate option
vorite paper to share with my first-year students is Richard that is comfortable for your budget. Then use
Guy’s “The strong law of small numbers,” published in The your benefits to assist your search.
Monthly, 95 (1988), no. 8, 697–712.
Share some of your own stories from the time when you
were a student. Talk to your students about their experi-
ences in your classroom. Listening to students’ opinions
and suggestions could be a valuable professional develop-
ment activity and may help you better understand your
own teaching practice.
Enjoy teaching and good luck!

Rate Options:
$0 $20 $57
† †

AMS Members
Veselin Jungic update your profile below to stay connected!
Credits
www.ams.org/account
Author photo is courtesy of Veselin Jungic.

www.ams.org/membership
*Annual statement of unemployed status is required.
†
Apply up to 20 AMS points to these rates. One point = $1 discount.

AUGUST 2025 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY 729

 Pursue your research
with the support of the
AMS

Centennial
FELLOWSHIP

This fellowship is awarded annually to

Provides an outstanding mathematical scientist
support of who has held a doctoral degree between

$75,000
three and twelve years.

Applications accepted between

July 15–September 30 each year.
Background pattern: StarLineArts / iStock / Getty Images Plus via Getty Images
Ribbon design: Fourleaflover / iStock / Getty Images Plus via Getty Images

Learn more and apply:

www.ams.org/centfellow
 MEMORIAL TRIBUTE

Bernard Maskit
Memorial Tribute
James W. Anderson, Ara Basmajian, Ruben Hidalgo,
Perry Susskind, and Edward C. Taylor

Mathematical Society in 2012. He had twelve doctoral stu-

His Life and Work dents and 45 descendants as of February 2025.1
Though the entirety of his mathematical work stands
Bernard Maskit, who we all called Bernie, was born on
on its own as serious and consequential, Maskit’s math-
May 27, 1935, and died on March 15, 2024, from exac-
ematical oeuvre is also, importantly, an essential bridge
erbations and complications of COVID-19. He grew up
on a long path that stretches back to the classification of
in New York City, attended DeWitt Clinton High School
the topological types of surfaces, beginning with Möbius
in the Bronx, and did his undergraduate studies at New
in 1861, along with the work of Klein, Fuchs, Koebe,
York University. Bernie completed his doctoral thesis, “On
and Poincaré, in which the uniformization theorem was
Klein’s Combination Theorem,” in 1964 at NYU under
proved. That is, any Riemann surface2 𝑆 has a regular holo-
famed complex analyst Lipman Bers ([MAC+ 95], [KM97]).
morphic cover 𝜋 ∶ 𝐷 → 𝐷/𝐺 = 𝑆, where 𝐷 is one of
Bernie’s professional life began with a postdoc at the Insti-
the Riemann sphere ℂ̂ = ℂ ∪ {∞}, the complex plane ℂ,
tute for Advanced Study (1964–1965), followed by an as-
or the upper half-plane, ℍ2 , and 𝐺 is a discrete group of
sistant professorship at MIT (1965–1972). He then moved
Möbius transformations acting properly discontinuously
to S.U.N.Y. at Stony Brook (now Stony Brook University)
on 𝐷. The other end of this long path is the geometriza-
where he remained until his retirement in 2008 as a pro-
tion conjecture of William P. Thurston, the proof of which
fessor emeritus. Bernie gave an invited talk about Kleinian
was completed by Grigori Perelman in 2003. The latter is a
groups at the 1974 International Congress of Mathemati-
result of celebrated significance in geometry and topology
cians, and became an inaugural fellow of the American
that occurred at the end of the twentieth century and at the
beginning of the twenty-first, and has as a consequence a
proof of the Poincaré conjecture.
James W. Anderson is a professor of mathematics at the University of Southamp- A Kleinian group 𝐺 is a discrete subgroup of 𝑃𝑆𝐿(2, ℂ),
ton, UK. His email address is [email protected]. regarded as the Möbius transformations acting on the Rie-
Ara Basmajian is a professor of mathematics at the Graduate Center and Hunter
mann sphere. The quotient by 𝐺 of that portion of the Rie-
College, CUNY. His email address is [email protected].
mann sphere on which 𝐺 acts properly discontinuously—
Ruben Hidalgo is a professor of mathematics at Universidad de La Frontera,
Temuco. His email address is [email protected].
the regular set—is a collection of Riemann surfaces. A Fuch-
Perry Susskind is a professor of mathematics at Connecticut College. His email sian group 𝐺 is (conjugate in 𝑃𝑆𝐿(2, ℂ) to) a discrete sub-
address is [email protected]. group of 𝑃𝑆𝐿(2, ℝ), and the latter acts on the upper half-
Edward C. Taylor was a retired professor of mathematics. He passed away on plane ℍ2 in the Riemann sphere. The quotient of the upper
March 10, 2025. half-plane by a Fuchsian group is a single Riemann surface.
A longer and more detailed version of this paper will appear on the arXiv.

For permission to reprint this article, please contact:

[email protected]. 1https://genealogy.math.ndsu.nodak.edu
2
A Riemann surface is a 1-dimensional complex manifold that may be thought
DOI: https://doi.org/10.1090/noti3202 of as a surface along with a conformal structure or a hyperbolic metric structure.

AUGUST 2025 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY 731

 MEMORIAL TRIBUTE

There is an extensive theory of Fuchsian groups, which may

be regarded as the analytic theory of Riemann surfaces.
Bernie’s main contributions in the areas of Kleinian
groups, low-dimensional topology, complex analysis,
and geometry of hyperbolic manifolds, include the
planarity theorem ([Mas65b]), the study of Schottky
groups ([Mas67]), the Klein-Maskit combination theo-
rems (([Mas65a, Mas68, Mas71a]), and the Poincaré poly-
hedron theorem ([Mas71b]). Maskit also was concerned
with moduli (deformation) spaces of Riemann surfaces,
including Teichmüller space, Schottky space, and more
generally the deformation theory of Kleinian groups. By
making use of his planarity theorem, Maskit’s early work
included three counterexamples to several reductions of Figure 1. Bernie and Lipman Bers.
the Poincaré conjecture to other conjectures that were pro-
posed by Papakyriakopoulos in the early 1960s ([Pap62], Maskit’s proof (in [Mas71b]), which he averred was the
[Mas63], [Pap63]). The planarity theorem has been of re- result of conversations he had with Ahlfors, Bers, Mag-
cent use in describing groups of homeomorphisms of pla- nus, and McMillan, is both correct and more general than
nar covers of the fundamental groups of certain orbifolds Poincaré’s. Poincaré and Maskit also provide a generaliza-
and geometrically finite Kleinian groups. Indeed, the con- tion of this result for polyhedra in ℍ3 (see below) in which
tinuing utility of the planarity theorem in this context is the group generated by the side pairing transformations is
evinced in there being a recent new proof by Bowditch Kleinian. (See the review of this paper in MathSciNet by J.
[Bow22]. A main thrust of Maskit’s career was to un- Lehner.)
derstand how one can construct Kleinian (or Fuchsian) In 1883, in an attempt to treat the theory of Kleinian
groups in different ways that produce collections of Rie- groups in roughly the same manner as the theory of Fuch-
mann surfaces. Below, we discuss Maskit’s combination sian groups, Poincaré extended the action of the Möbius
theorems, the Poincaré polyhedron theorem, and briefly, transformations from the Riemann sphere to the upper
joint work with Alan Beardon. half-space, ℍ3 = {(𝑧, 𝑡) | 𝑧 ∈ ℂ, 𝑡 > 0}.
The combination theorems are a collection of enhance- After Poincaré’s effort, the theory of Kleinian groups lan-
ments of the Klein combination theorem in which con- guished somewhat for a number of decades, but efforts
ditions are imposed on two Kleinian groups so that the to develop the theory picked up in 1964 with Ahlfors’s
group generated by both of them is also Kleinian, and alge- finiteness theorem ([Ahl64]). Maskit’s contributions to
braically is the free product of the groups. These enhance- Kleinian groups, beginning at that time, were generally of
ments include the cases of free products with amalgama- a geometric/topological nature and, as already discussed,
tion, and HNN extensions. Maskit’s program was to show included the Poincaré’s polyhedron theorem, the Klein-
that all Kleinian groups may be built by using the combi- Maskit combination theorems and the planarity theorem.
nation theorems. This approach is analogous to the ap- Nearly concurrently, Albert Marden’s aspiration was to re-
proach of Haken and Waldhausen in which a hyperbolic alize Poincaré’s original hope of treating the theory of
3-manifold may be cut up into a union of balls and by re- Kleinian groups acting on hyperbolic space in much the
versing the steps, the manifold may be reconstructed. (See same way as the theory of Fuchsian groups. This in part
[Mar90].) Maskit’s program was almost entirely successful motivated his work in ([Mar74]). To quote Marden:
and resulted in a large “zoo” of cases. (See chapters VIII–X In contrast to the analytic approach of Ahlfors
of [Mas88].) and Bers, Maskit has pioneered in the study of
The upper half-plane, ℍ2 , may be endowed with a met- Kleinian groups by purely geometric methods in
ric of constant negative curvature on which a Fuchsian the complex plane. His work complements the in-
group acts as isometries. The Poincaré polyhedron theo- sight provided by the analytic approach and yields
rem provides sufficient conditions for a polygon 𝑃 (with some very deep and fundamental knowledge con-
geodesic sides) in ℍ2 , along with isometries of ℍ2 that cerning certain special classes of groups. In par-
pair the sides of 𝑃, to generate a Fuchsian group for which ticular Maskit develops the concept that the class
𝑃 is a fundamental region. Poincaré’s proof was incom- of “nice”’ groups, that is the class for which gen-
plete, and the Kleinian and Fuchsian groups literature eralizations of the classical results for Fuchsian
was rife with a number of partial or inaccurate proofs. groups can be fruitfully sought, is the class of

732 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY VOLUME 72, NUMBER 7

 MEMORIAL TRIBUTE

Figure 3. Clifford Earle, William Harvey, and Bernie.

Figure 2. Bernie in Chile.

serve that the existence of a finite-sided fundamental poly-

constructible groups. These are the groups that
hedron for the action of 𝐺 on ℍ3 , a condition called geo-
arise from his far-reaching generalizations of the
metric finiteness, is equivalent to a characterization of the
Klein combination theorems. Although we have
set of accumulation points of the orbit of a point in ℍ3 un-
taken a different path in this paper, Maskit’s work
der 𝐺—the limit set of 𝐺—which is a subset of the Riemann
has been a great influence. Indeed, there are many
sphere on the boundary of ℍ3 .
points of contact between his work and ours al-
Characterizations by Beardon and Maskit of the limit
though the precise relation remains unclear.
set of a geometrically finite group comprise part of a collec-
Any expert in Kleinian groups will be familiar with the tion of five closely related equivalent notions of geometric
work of Bernard Maskit, but possibly anyone who has even finiteness, and all have been generalized in a number of
considered study in the area will be familiar with Maskit’s ways. For discrete groups of isometries of ℍ𝑛 , Bowditch,
1988 book, Kleinian groups ([Mas88]). In it he covers many Tukia, and others generalized the notion of geometric
essential topics—indeed, many of the topics chosen are finiteness ([Bow93]), and for discrete groups of isometries
those that Maskit had a large hand in developing. Al- of 𝐼𝑠𝑜𝑚(𝑋) where 𝑋 is an 𝑛-dimensional pinched Hadamard
though pains are taken in many cases to approach topics manifold, that is, 𝑋 is a complete, simply connected Rie-
in a way that allows for some generality (for example, dis- mannian manifold with all sectional curvatures bounded
crete subgroups of the isometries of hyperbolic 𝑛−space between two negative constants, Bowditch has carried out
are considered, rather than isometries of hyperbolic 2− or the generalization ([Bow95]). Other generalizations in-
3−space), the approach is largely 2-dimensional. That is, clude convergence groups.
the approach is largely oriented toward developing the pla- Each of his students had their own experience of Bernie
nar theory of Kleinian groups by exploring their action on as an advisor, though all would agree that Bernie was a
the Riemann sphere, rather than the 3-dimensional the- most important figure for his students. While Bernie did
ory of hyperbolic manifolds, which results by looking at not provide an overabundance of advice and guidance to
the action of a Kleinian group on the upper half-space. his students, he provided just what was necessary for each,
One should observe that this approach may make it appear and consequently allowed each of his students, according
that Maskit’s work is somewhat removed from the exciting to their individual abilities and mathematical maturity, to
work on geometrization and hyperbolic 3-manifolds that stand on their own two feet. The largely topological argu-
has occurred over the last 40 years, but this is mostly an ments of many of Bernie’s papers required highly demand-
issue of point of view, since these areas are either inextri- ing precision, rigor, and meticulous attention. Beyond just
cably intertwined or in some cases identical. the shepherding of students through the usual process of
This planar approach characterizes much of Maskit’s turning a particular but possibly vague mathematical prob-
work, though a significant and influential departure is the lem or idea into a viable thesis, Bernie quietly and mod-
paper [BM74] with Alan Beardon in which the authors estly modeled by example the kind of meticulous and care-
look at the action of a Kleinian group 𝐺 on the upper ful thinking that guided all of us in our subsequent math-
half-space. Even there, though, the main result is to ob- ematical (or not mathematical) careers.

AUGUST 2025 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY 733

 MEMORIAL TRIBUTE

One of us remarked at a conference banquet held in

his honor at NYU in March of 2008 that when teaching
classes, Bernie’s students benefited from, and appreciated
watching, listening, and learning, from instances in which
Bernie would think through a nontrivial argument on the
fly without having prepared it in advance. (As a response,
Bernie quipped that he did not realize he was being a good
teacher; rather, he claimed he thought he was just being
lazy about preparing his classes in advance.) At times, after
making an assertion in class, Bernie would ask, “Do I have
to worry?” Sometimes the answer was “Yes,” and signifi-
cant explication would follow. Other times, there ensued
a real-life instance of the apocryphal story of the student
challenging the professor’s statement that some observa-
tion made in class is trivial, wherein the professor stands
Figure 4. Bernie’s 60th birthday with students (from left to in silence for several minutes engaged in deep thought and
right) Perry Susskind, Ed Taylor, Bernie, Jim Anderson, Andy
then reiterates, “Yes, it’s trivial.” In this instance, Bernie
Haas, and Ara Basmajian.
would ask, “Do I have to worry,” stand for a minute or
two in silence, thinking deeply, and then proclaim, “No,
I don’t have to worry.” His students were usually relieved
that he—and we—didn’t have to worry.
On a personal level, Bernie was reserved with his stu-
dents but would invite us to gatherings at his house where
there was often excellent food and drink. Lastly, we men-
tion a final chapter in Bernie’s intellectual life. A few years
before he retired, Bernie became interested in an area of
computational psycholinguistics called multiple code the-
ory, developed by his wife and intellectual partner, Wilma
Bucci. Bernie made significant contributions in this area
(see [Sus25]) largely by introducing the use of the tech-
nique of mathematical smoothing for certain mathemat-
ical measures on spoken narratives. All of us agree, likely
due to Wilma’s influence, that Bernie became increasingly
content as time passed. They enjoyed good food and drink,
Figure 5. Bernie, Wilma, and Ruben Hidalgo.
and gathering with friends. Bernie enjoyed tennis for a
When stating theorems in his Kleinian Groups course, time—he was a handball player in his early years—and few
Bernie would often write just the conclusion, and an- might guess that Bernie and Wilma learned how to dance
nounce that he would be filling in the hypotheses as the tango.
needed during the proof, somehow mimicking a creative Our motivation for writing this memorial article began
process more engaging than just “theorem—proof.” Pe- with the impetus to honor our advisor and make others
ter Buser related a rather interesting criterion that Bernie aware of his work and role in important developments in
might employ in order to deem a paper finished. “It was 2- and 3-dimensional topology and geometry. Years ago,
in Oberwolfach, or perhaps at some other place. I was sit- as new PhDs, we were perhaps only partially equipped to
ting at a table with Bernie and we were discussing publish- fully understand the breadth and depth of Bernie’s contri-
ing mathematics. My concern was a paper that was done, butions in these areas. With the perspective of much time
but yet not finished, and I couldn’t get it done. ‘Hmm,’ he working in these areas ourselves, the process of revisiting
said, and after a while: ‘and then you improve a passage Bernie’s contributions to mathematics has been a highly
and change a notation, and later you feel that the earlier rewarding endeavor.
version was better, and you think it over again. You know,
when that happens, it’s the sign that the material has ma-
tured.’ I often remember this scene when I’m working on
a text.”

734 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY VOLUME 72, NUMBER 7

 MEMORIAL TRIBUTE

One time we met to discuss a question we were consid-

ering; to be honest, it was a shared annoyance. He had a
proof with a gap; I had a proof, the same proof in a differ-
ent setting, also having a gap. But the gaps were different.
Together, we had the proof.
While much of his work predated the widespread use of
computers in geometry, his constructions are computable.
In some cases, Thurston’s bounded image theorem is
paired with the constructions—this interdependence is key
to most of the proof of Thurston’s hyperbolization theo-
rem. Only manifolds fibering over the circle require a dif-
ferent approach. While Thurston was developing the ideas
for hyperbolization, he lectured quite publicly before he
had complete proofs. Thurston used at least one combi-
nation theorem that Bernie hadn’t yet proved. In one lec-
ture Bill gave at Columbia, Bernie informed him that the
Figure 6. Bernie and Wilma in Chile.
combination theorem he used in the lecture was not yet
proved. Felix Klein is credited with the first construction
(See the reference in [Mas65a].) In the eighty-year period
William Abikoff between Klein’s work and Bernie’s, there were few construc-
I first met Bernie at the Stony Brook Riemann Surfaces con- tivist contributions to the literature other than in the Fuch-
ference in 1969. I was an engineer at Bell Labs and had sian, i.e., invariant disk, case. The two that come quickly
recently passed mathematics doctoral exams at Brooklyn to mind are Lester Ford’s 1926 book and the mysterious
Poly. I had been talked out, actually laughed out, of my in- Fenchel-Nielsen manuscript; the latter was announced as
terest in a particular problem about sums of random func- appearing imminently in the late 1940s and was finally
tions and had come uninvited to that conference. The only published in 2003.
attendee I knew was Lesley Sibner. The excitement in the Bernie’s work has much contemporary interest in a va-
air was overwhelming. After dark, participants talked and riety of fields, including complex analysis, hyperbolic ge-
socialized while I, in desperation, drove to the Courant ometry, and knot theory. It had reenergized combinato-
Institute library, the only research library open at night, rial group theory and is a spiritual precursor of geometric
and tried to figure out what all those excited people were group theory. For me to write of its future influence would
talking about. A few of them not only welcomed me but require prophetic skills that I do not possess.
they made me feel that I belonged. Even now, over a half-
century later, I treasure the moments when I recall them—
Bernie, Lipa Bers, Ernie Rauch, and Bob Gunning. It didn’t
Wilma Bucci
take long for me to become an auxiliary member of the I met Bernie on Christmas Eve 1978, almost half a century
“Bers Mafia.” ago, when he was on a skiing trip with his two younger
Probably at Lipa’s suggestion, I visited Bernie at MIT sons, and staying with his sister who was a psychologist
around Christmas of 1971. We soon bonded both profes- and friend of mine. Back then, Bernie’s three sons all were
sionally and personally. At that time, the Kleinian groups under the age of 20; my son and daughter were in the same
crowd consisted of his contemporaries. And me. I became age range. We shared a strong wish to do the best we could
something of a kid brother. for our family. Bernie was truly passionate about math-
We shared nonstandard paths from the Bronx into and ematics, but also about his family. There were many in-
through mathematics. Bernie had started undergraduate stances of how conflict between family and work played
study at RPI, then withdrew because of family problems. out in our life together. As an example, about a year after
He then spent some time assisting in his family’s busi- we met, Bernie was involved in organizing a conference fo-
ness, while completing his undergraduate study at NYU. cused on Bill Thurston’s work, which was new at the time.
Starting in his graduate years, his main focus lay in the in- Bernie and I, while traveling in Maine, found the perfect
terplay and consequences of algebraic and geometric con- venue for the conference at Bowdoin College. We drove to
structions. Later on, his curiosity moved him to consider the meeting together, which took place in the summer of
questions raised in the work of his, in so many ways, life 1980. A young relative was left with my children in New
partner, Wilma Bucci. Jersey; Bernie’s sons were with their mother. Each day at

AUGUST 2025 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY 735

 MEMORIAL TRIBUTE

the meeting, Thurston presented his work. Each day af- also guided students working in this area on dissertations,
ter his talks, Bernie, Lipman Bers, and other colleagues papers, and posters. Research centers in the US, Canada,
met impromptu in sessions devoted to unraveling what Italy, Israel, and Argentina apply Bernie’s measures of the
Thurston had said. Bernie and I would then go back to referential process, and work by colleagues in these centers
our campsite near the meeting, and before settling down continues to extend the theory and applications.
for the evening call home to check on our children. (There
were no cell phones in those days so we had to find a pay
phone near the site.) About the third night, I got one of Some Memories
those responses that parents dread: “Everything is alright
now.” My daughter had been in a bicycle accident, had
been taken to the emergency room, and had broken her
Fred Gardiner
arm, though she was said to be okay. I told Bernie I had to Bernie leaves an extensive legacy of publications and a
go home right away; he said he was going with me. He told book that has become an essential reference in Kleinian
Lipa (Lipman Bers) and their colleagues. Lipa said they groups. But here I also want to focus on his generosity,
needed Bernie to stay. Bernie said he was going with me. fairness, and ability to come up with pointed comments.
Mary Bers said of course he has to go with Wilma. Bernie These qualities enabled him to draw together three gen-
drove me home and we dealt with that family problem erations of successful working mathematicians. It started
together, as we did with many problems on many levels in the early 1960s when he earned his PhD and began a
throughout the years. continuing collaboration with his thesis adviser, Lipman
As the children became more independent, we did ex- Bers, at NYU over many years. Riemann surface theory had
tensive traveling, went together to many conferences, and become of interest in the 1930s to nuclear physicists. In
to visit colleagues, in Europe and South America, includ- particular, Hermann Weyl had ten years before published
ing a wonderful year in France, based at IHES, just out- his monograph Die Idee der Riemannschen Fläche. This was
side Paris, where we went to market, cooked, and drank coupled with Schrödinger’s earlier use of complex analy-
wine. We always cooked together, wherever we were, sis to give discrete solutions to the wave equation which
Bernie mostly as prep chef. Bernie was an experienced described the discrete possibility for radii of orbitals in an
hiker and camper. We drove across the US twice, camping atom. But, in Bernie’s life it was the richness of the sub-
in national parks and forests and staying for the summer in ject of Kleinian groups and Riemann surfaces that drew
Berkeley where Bernie worked at MSRI and I commuted to so much attention. In my memory several special mo-
UCSF. Bernie played the recorder seriously and well, and ments stand out. The first occurred at Bers’s complex anal-
we did all kinds of dancing—folk dancing, contra dancing, ysis seminar, in which Bernie was an active participant, on
all types of ballroom dancing, and then became addicted the fourth floor of the Columbia University Math Build-
to Argentinian tango. ing. Dennis Sullivan, having just returned from IMPA in
While Bernie and I joined our families and our lives, we São Paulo, came to speak about a paper he wrote jointly
went our separate ways in our research for the first several with Mañé and Sad. That paper was about iteration of ra-
decades we were together. Around 2000, as he neared re- tional functions, and in the middle of working on it their
tirement from Stony Brook, Bernie moved from a full focus paper contained the idea of a holomorphic motion. It
on mathematics to an increasing interest in research ques- was a way to conclude that a certain curve of deforma-
tions associated with my work as a psychologist and psy- tions of a dynamical system acting injectively on a limit
cholinguist. Bernie brought his creative ways of thought to set would necessarily be realized by a quasiconformal con-
difficult questions associated with my research. When he jugacy. That turned out to be the beginning of the studies
looked at my work, which involved studying linguistic pro- of holomorphic motions. But at roughly the same time,
cesses using computer assisted measures, Bernie tended to there was what became known as Sullivan’s dictionary, a
first characterize the problems as impossible to solve, but collection of correspondences that tied the ideas and the-
he then proceeded to solve them, in the process introduc- orems of Kleinian groups, including the work of Ahlfors,
ing mathematical techniques that those in my field had Bers, Maskit, and others, with the burgeoning subject of
not considered. Bernie continued developing new ideas dynamical systems and the iterations of rational maps.
to the last days of his life and left us, as part of his legacy, Another moment happened in the mid-1980s when Bill
a number of ideas for projects that he planned to pursue. Thurston came to the same seminar to explain parts of his
Some of these new ideas are presented in the last paper extraordinary theory of measured foliations and lamina-
Bernie wrote, to be published in a special issue of the Jour- tions, which had begun in the mid 1970s publication of
nal of Psycholinguistic Research honoring his work. Bernie Travaux de Thurston sur les surfaces. All the while Bernie’s

736 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY VOLUME 72, NUMBER 7

 MEMORIAL TRIBUTE

expertise and pointed questions contributed immensely to Riemann surfaces and Kleinian groups, and the structure
this environment. of Kleinian groups was something into which Bernie had
When Bernie was at a seminar you would find other tremendous insight and understanding. Beginning with
mathematicians working seriously on related topics about his thesis, he became the world’s expert on the intrica-
which he was an expert and you would find collaborators. cies of Kleinian groups, for example, how complicated the
The wonderful Chapter VIII of his book on Kleinian groups limit sets could be. This was particularly important in
is titled “A Trip to the Zoo.” That’s where he took us and studying what happens as one moves to the boundary in
where we had a great time. these spaces. After finishing his thesis, Bernie spent two
years at the Institute for Advanced Study in Princeton, and
I joined him there the second year. That was a time when
Jane Gilman it looked very promising that one could prove the Poincaré
I met Bernie when I was a graduate student at Columbia conjecture for manifolds of dimension less than 5. It was
and he was already a well established mathematician. We the holy grail then, and attracted the best and the brightest,
were both students of Lipman Bers, but Bernie was reputed so it was a not little thing that a result of Bernie’s, his pla-
to be the best of Bers’s many students working on Teich- narity theorem, showed that one of the main approaches
müller theory. I was in awe of Bernie, but later, when I that would have proved the conjecture in dimension 3 was
collaborated with him, I learned that one did not have to false. It took another thirty-eight years before the conjec-
be in awe of Bernie, just of his mathematical powers. He ture was proved. Cliff Earle was also at the Institute that
was very approachable and very nice. I also got to know year and the three of us had a wonderful time talking about
some of Bernie’s students, especially those from his 1982– math together. We read one another’s work. I’m particu-
1995 Kleinian groups—Stony Brook period. In addition to larly grateful to Bernie for carefully reading my paper on
publishing important papers in hyperbolic geometry and moduli for Fuchsian groups and helping clarify its expo-
Kleinian groups, these students were all very nice and have sition. This was also the year of the first conference on
been a welcome addition to my mathematical world. My Kleinian groups organized by Ahlfors and Bers in which
collaboration with Bernie began at an MSRI semester on they showcased their students. It was in New Orleans
Kleinian groups and involved the two generator 𝑃𝑆𝐿(2, ℝ) where we not only talked math, but sampled the food and
discreteness problem: given a pair of hyperbolic elements wine ordered by Ahlfors at many of the famous restaurants.
with disjoint axes, determine whether or not they generate It was the first of many such conferences which grew as the
a discrete nonelementary group. This problem has a long numbers of their students multiplied. Bernie was always
history of publications with incorrect or incomplete solu- front and center in this community that launched so many
tions. Perhaps because the problem is easy to state, people careers. For over sixty years, Bernie was a good friend and
expected a simple closed form solution. Our joint 1991 pa- excellent colleague.
per gave a complete geometric algorithm for determining
discreteness or nondiscreteness. This one collaboration in-
fluenced much of my subsequent work on two generator The Day After and Memories
groups. I remember Bernie and his wife, Wilma, on an
overnight trip to Leningrad from a conference at the Uni-
versity of Joensuu in Finland in 1987. At the time laptop
Irwin Kra
computers were rare, but Bernie had one and brought it Yesterday, my good friend Bernie Maskit died. Although
with him on the trip. Soviet customs officers did not know not totally unexpected—we all die—a great shock never-
what to make of it. We were delayed for hours. Eventually theless. One more string cut. An end to our weekly Zoom
the laptop was allowed to enter the Soviet Union. meetings. He was my colleague (at MIT and Stony Brook
and was my first or second appointment to the SB math
department), collaborator (numerous joint papers), advi-
Remembrances of Bernie Maskit sor (in mathematics and administration), but most of all
friend (who else would drive me 60 miles, beginning at
3 a.m., to pick up a delivery from abroad?). How did I
Linda Keen ever reciprocate? He was a collaborative member of mathe-
Bernie and I were students of Lipman Bers in the early matics departments and a friend to many of his colleagues
1960s at NYU. It was an exciting time. Ahlfors and Bers and students. Hard to accept his loss, but nevertheless we
had just proved a major theorem that made it possible must.
to put a holomorphic structure on the moduli spaces of

AUGUST 2025 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY 737

 MEMORIAL TRIBUTE

Maskit and I worked on similar problems mostly in- well worth the wait. Our article constructed Riemannian
volving Kleinian groups and the Riemann surfaces asso- metrics that minimize the 𝐿2 -norm of the curvature tensor
ciated with them, he from a topological and group theo- on connected sums of five or more copies of the complex
retic approach and I as a complex analyst. Nevertheless projective plane, and this then allowed us to prove a clas-
we found a common language and rapidly developed joint sification result that I still find enormously satisfying.
programs, as is evident by the number of joint papers we However, soon after our paper was published, Bernie
produced. At the time we met, Kleinian groups and their decided to retire and move into Manhattan, and I saw him
moduli (deformations) were ready for a rebirth, mostly as only infrequently after that. But I still miss his laughter,
a result of two important, independent but related papers and I still miss our conversations. And I am sure that many
that had recently appeared, the first by Lipman Bers and of our colleagues must also be feeling the same sense of
the second by Bernard Maskit. This revival included strong loss.
contributions (some in joint papers) by L. V. Ahlfors, L.
Bers, C. J. Earle, I. Kra, A. Marden, and B. Maskit, among
others in the older generation. More important, perhaps,
Dennis P. Sullivan
was the number of young people attracted to this field. I first met Bernie at MIT when I was a postdoc and he was
Maskit was the authority on construction of complicated a junior professor, in 1970. I asked him what he worked
finitely generated Kleinian groups from simpler building on for which he answered, “I work on 2x2 matrices and
blocks. It was quite exciting for me to witness this act of surfaces.” I hid my wonder since I was part of the field
creation and his interactions with his students and follow- studying manifolds of dimension 5 or more, which did
ers. not include 3 and 4 because they were too hard, and did
not include two dimensions because they were too easy.
(cf corrigendum below)
Bernie Maskit Remembered Later in the 1970s, inspired by a question from Lipman
Bers, I learned about Kleinian groups and their limit sets
and I tried to use the Klein-Maskit combination construc-
Claude LeBrun tion to answer Bers’s question. During this period I vis-
I was shocked and deeply saddened when my friend, ited Stony Brook every summer and learned much more
collaborator, and colleague Bernie Maskit unexpectedly from Bernie and Irwin Kra about Schottky groups and au-
passed away last March. Bernie had first befriended me in tomorphic forms, respectively. I also learned how silly my
the early 1980s when I was an assistant professor at Stony first reaction at MIT was to Bernie’s modest answer, “I work
Brook. He later became chair of our department, and was on 2x2 matrices and surfaces,” because the automorphism
serving in that capacity when I came up for tenure in 1988. group of a surface of genus 𝑔 is a deep and rich mathemat-
Several other assistant professors in the department had ical object, actively studied, and central to several fields.
recently been denied tenure, so I was justifiably worried From Bernie I learned about the pleasant restaurants
about my own prospects. But Bernie’s gentle, avuncular and cultural diversions of his hometown, Huntington,
encouragement and generous practical advice helped me where he lived because his spouse worked in New York City
somehow navigate and endure the process without becom- and he worked in Stony Brook.
ing a total nervous wreck, even as he helped shepherd my Bernie was smart, modest, and likeable. I am glad to
case through to a successful outcome. have had the chance to know him.
After I returned from my first sabbatical, Bernie and I
started to meet for lunch from time to time, partly to dis-
cuss mathematics, but also just to chat about life, the uni- RIP Bernie Maskit
verse, and everything. My work on the scalar curvature of
self-dual 4-manifolds had convinced me of the importance
of locally conformally flat orbifolds, but this left me try-
Gadde Swarup
ing to puzzle out a host of subtle questions concerning A very kind man who [helped me to] revive my research. . . .
Kleinian groups and their limit sets. Fortunately, Bernie I had a brief collaboration with him and a longer collabo-
was a world-class expert on these matters, and he was only ration was lost in the mail. I had many pleasant moments
too happy to educate me about many of the fine points with him. . . Mladen Bestvina tells me that I told him about
I needed to understand. It took many years for our dis- the Maskit combination theorems, which led to his impor-
cussions to crystallize into a collaboration, but our joint tant work with Feighn on combination theorems for word
2008 paper, “On Optimal 4-Dimensional Metrics,” was hyperbolic groups.

738 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY VOLUME 72, NUMBER 7

 MEMORIAL TRIBUTE

References [MAC+ 95] Cathleen S. Morawetz, William Abikoff, Carol Co-

[Ahl64] Lars V. Ahlfors, Finitely generated Kleinian groups, rillon, Irwin Kra, Tilla Weinstein, and Jane Gilman, Remem-
Amer. J. Math. 86 (1964), 413–429, https://doi.org bering Lipman Bers, Notices Amer. Math. Soc. 42 (1995),
/10.2307/2373173. MR167618 no. 1, 8–25. MR1306867
[BM74] Alan F. Beardon and Bernard Maskit, Limit points [Pap62] C. D. Papakyriakopoulos, A reduction of the Poincaré
of Kleinian groups and finite sided fundamental polyhedra, conjecture to other conjectures, Bull. Amer. Math. Soc. 68
Acta Math. 132 (1974), 1–12, DOI 10.1007/BF02392106. (1962), 360–366, DOI 10.1090/S0002-9904-1962-10810-
MR333164 6. MR148039
[Bow93] B. H. Bowditch, Geometrical finiteness for hyperbolic [Pap63] C. D. Papakyriakopoulos, A reduction of the Poincaré
groups, J. Funct. Anal. 113 (1993), no. 2, 245–317, DOI conjecture to group theoretic conjectures, Ann. of Math. (2) 77
10.1006/jfan.1993.1052. MR1218098 (1963), 250–305, DOI 10.2307/1970216. MR145496
[Bow95] B. H. Bowditch, Geometrical finiteness with variable [Sus25] Perry Susskind, Maskit’s Mathematical Contributions:
negative curvature, Duke Math. J. 77 (1995), no. 1, 229– The Smoothing Operator and DAAP Measures, J. Psycholin-
274, DOI 10.1215/S0012-7094-95-07709-6. MR1317633 guistic Research 54 (2025).
[Bow22] Brian H. Bowditch, Notes on Maskit’s planarity the-
orem, Enseign. Math. 68 (2022), no. 1-2, 1–24, DOI
10.4171/lem/1019. MR4420863
[KM97] I. Kra and B. Maskit, Lipman Bers, complex analyst,
Lipa’s legacy (New York, 1995), Contemp. Math., vol. 211,
Amer. Math. Soc., Providence, RI, 1997, pp. 389–415, DOI
10.1090/conm/211/02831. MR1476998
[Mar74] Albert Marden, The geometry of finitely generated
Kleinian groups, Ann. of Math. (2) 99 (1974), 383–462,
https://doi.org/10.2307/1971059. MR349992
[Mar90] Albert Marden, Book Review: Kleinian groups, Bull.
Amer. Math. Soc. (N.S.) 22 (1990), no. 2, 310–315, DOI James W. Ara Basmajian
10.1090/S0273-0979-1990-15895-1. MR1567847 Anderson
[Mas63] Bernard Maskit, On a conjecture concerning pla-
nar coverings of surfaces, Bull. Amer. Math. Soc. 69
(1963), 396–398, DOI 10.1090/S0002-9904-1963-10942-
8. MR148040
[Mas65a] Bernard Maskit, On Klein’s combination theorem,
Trans. Amer. Math. Soc. 120 (1965), 499–509, https://
doi.org/10.2307/1994540. MR192047
[Mas65b] Bernard Maskit, A theorem on planar covering
surfaces with applications to 3-manifolds, Ann. of Math.
(2) 81 (1965), 341–355, https://doi.org/10.2307
/1970619. MR172252
[Mas67] Bernard Maskit, A characterization of Schottky Ruben Hidalgo Perry Susskind
groups, J. Analyse Math. 19 (1967), 227–230, DOI
10.1007/BF02788719. MR220929
[Mas68] Bernard Maskit, On Klein’s combination theorem. II,
Trans. Amer. Math. Soc. 131 (1968), 32–39, https://doi
.org/10.2307/1994678. MR223570
[Mas71a] Bernard Maskit, On Klein’s combination theorem. III,
Advances in the Theory of Riemann Surfaces (Proc. Conf.,
Stony Brook, N.Y., 1969), Ann. of Math. Stud., No. 66,
Princeton Univ. Press, Princeton, NJ, 1971, pp. 297–316.
MR289768
[Mas71b] Bernard Maskit, On Poincaré’s theorem for fundamen-
tal polygons, Advances in Math. 7 (1971), 219–230, DOI Edward C. Taylor
10.1016/S0001-8708(71)80003-8. MR297997
Credits
[Mas88] Bernard Maskit, Kleinian groups, Grundlehren der
mathematischen Wissenschaften [Fundamental Principles All photos are courtesy of the authors.
of Mathematical Sciences], vol. 287, Springer-Verlag,
Berlin, 1988. MR959135

AUGUST 2025 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY 739

 EDUCATION

Teaching Mathematics with

Lean: Interactive Theorem
Provers in the Classroom
Paola Iannone, Gihan Marasingha,
and Athina Thoma

Prologue One of the mathematicians we collaborated with is Gi-

In September 2018, I∗ (Paola) received an email from a han Marasingha,† who let us into his classroom for one of
mathematician at Imperial College London whom I had our projects. In this article, we discuss in turn our research
met a couple of years previously at a meeting. He asked findings and Gihan’s experience with using Lean for teach-
me whether I was interested in studying the impact of in- ing to outline some of the potential of this tool for learning
troducing Lean, an Interactive Theorem Prover (ITP), on mathematics at the undergraduate level.
students’ understanding of proof in his transition to proof
Introduction
course. I did not know at that time what an Interactive
Theorem Prover was, but it sounded interesting as it was In [BBB+ 24] several mathematicians have written about
linked to proof writing and mathematical reasoning—one the potential of Lean for the formalization of mathemat-
of my research interests—so I agreed to meet and discuss ics and how this tool is now—in their view—indispensable
the course. The mathematician was Kevin Buzzard and given the complexity of modern proofs. A more recent ar-
eventually I and Athina Thoma‡ designed and carried out ticle [Tao25] suggests that indeed the mathematics of the
that study. Since then we have collaborated with math- future will combine ITPs, machine learning, and large lan-
ematicians in Europe and the United States to study the guage models to allow for, among other things, a system-
impact that interacting with Lean (https://leanprover atic formalization of proofs. It seems, therefore, important
-community.github.io) has on students’ understanding that we train our students to use these tools and that we do
of mathematics, the development of their mathematics so in a way that contributes to their mathematical under-
epistemology, and the way in which students and, more standing. In this article, we write about the potential of
recently, teachers use this tool. Lean as a tool that can be used in classrooms to help stu-
dents to come to terms with proof and the requirement of
rigor necessary for university mathematics. We do so in
Paola Iannone∗ is a reader in mathematics education at the University of Edin- two ways. First, we present some results from educational
burgh. Her email address is [email protected]. studies concerning the potential of Lean for learning about
Gihan Marasingha† is an associate professor in mathematics at the University proof by focusing on the questions we asked and on our
of Exeter. His email address is [email protected].
findings. Then, Gihan discusses his experience of using
Athina Thoma‡ is a lecturer in mathematics education at the University of
Lean in a transition to proof course, sharing also from his
Southampton. Her email address is [email protected].
perspective the benefits and challenges. Finally, we draw
Communicated by Notices Associate Editor Yvonne Lai.
our writing together hoping to inspire more colleagues to
For permission to reprint this article, please contact: use this tool in their teaching. Before we start, we need to
[email protected]. make two disclaimers. The first concerns the implementa-
DOI: https://doi.org/10.1090/noti3184 tion of Lean for teaching that we are interested in: We will

740 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY VOLUME 72, NUMBER 7

 EDUCATION

focus on first-year courses of a degree in mathematics that error message window may display syntax errors or logic
have included learning to use Lean. We know that Lean is errors (e.g., the object in the code line does not belong to
now taught and used at the MSc and PhD levels, but these the right type). The top window also provides feedback to
courses are not the focus here. The second disclaimer is the user: It displays the context in which the user is work-
that we believe that this educational research is needed, ing as well as the goals in each line of written code. These
given the growing interest that undergraduate students are goals can be (logic) correct goals but not necessarily the
showing for Lean and other ITPs. At the University of Ed- right ones for the proof. In this game, Lean is used in tac-
inburgh (UK) there is a Lean club that is student-led and tic mode; tactics are automations that can close goals. For
is running with no input from the faculty. We know this example, the tactic 𝑟𝑒𝑓𝑙 (for reflexivity) applies this prop-
to be the case in other mathematics departments as well. erty to the goal of the proof (for example, 𝑟𝑒𝑓𝑙 can be used
Considering that students are actively involved in learning to close the goal presented in Figure 2). The research re-
and using Lean and contributing to its library, it is becom- ported in this article was carried out before the release of
ing important to see the impact Lean has on the mathe- Lean 4. However, the interface for the Natural Number
matical understanding of undergraduate mathematics stu- Game in Lean 4 contains elements similar to that of Lean
dents. 3—although with more customization features.

Interactive Theorem Provers The UK Context

Interactive Theorem Provers have been used in pure math- Before we discuss our studies and their findings, we intro-
ematics and computer science research since de Bruijn’s duce the UK higher education context. Students pursu-
creation in the 1960s of the Automath prover ([dB94]). ing mathematics degrees must have had mathematics as
ITPs are software tools which bring together proof au- one of their chosen subjects in high school, and admis-
tomation and human-machine proof interaction. They sion to a given university depends on that university’s en-
typically contain an interactive proof editor where it is trance requirements. Given the nature of the universities
possible to write lines of code following the ITP’s syn- we worked with, the students in our studies would have
tax, and draw on a verified symbolic library of math- achieved high marks in their final school exams. Degree
ematical objects. Among the most popular ITPs are courses last three years (four years in Scottish universities),
Rocq (https://rocq-prover.org), Isabelle (https:// and they are modularized: Students need to take a set num-
isabelle.in.tum.de), and Lean. Lean is the ITP ber of credits associated to courses in each year and to
which we discuss in this paper. Figures 1 and 2 show achieve marks that allow them to proceed to the follow-
the interface for the Natural Number Game in Lean ing year.
3 (https://www.ma.imperial.ac.uk/~buzzard/xena It is not unusual for mathematics departments in the
/natural_number_game/index2.html). This game was UK to offer an “introduction to proof” course in the first
designed by Kevin Buzzard and Mohammad Pedramfar in semester of the first year of the degree, in recognition of
2019 as an entry point to Lean for first-year students. The the difficulties students have in the transition to univer-
underlying idea of the design was that students could fo- sity mathematics. The courses that students in our stud-
cus on the Lean programming aspect as the mathematics of ies took, and the course where Gihan implemented the
the game would be familiar to them. In the game, the stu- use of Lean, were of this type. These courses typically in-
dents progress by coding their way through the construc- clude a combination of topics such as number systems,
tion of the natural numbers completing levels which con- sets, permutations and combinations; the binomial the-
tain statements regarding addition, multiplication, and orem; equivalence relations and arithmetic modulo 𝑛; Eu-
more. It was updated by Kevin Buzzard and Jon Eugster clid’s algorithm; and an introduction to limits. Teaching is
for Lean 4, and it can be found at https://adam.math conducted in lectures with a large number of students and
.hhu.de. smaller group tutorials, depending on the size of the co-
For the Natural Number Game in Lean 3 each of the hort and on the university. Assessment is typically based
levels of the game is preceded by a short description of the by coursework set during the teaching period and a final
statement to prove and of the available tools (e.g., tactics closed-book exam.
and previously proved theorems). On the left-hand side of
the interface is the coding space, where proofs are written Our Starting Point—Key Considerations
after the word “begin.” On the right-hand side, there are and Questions
two windows: The top window displays the context and Educational research on the challenges that students face
the goal that needs to be proved next; the bottom window with proof when joining a mathematics degree has high-
presents error messages. Once a line of code is written, the lighted several areas of concern. We list them below, and

AUGUST 2025 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY 741

 EDUCATION

Figure 1. This is the interface of the Natural Number Game in Lean 3 (Level 9 of the Multiplication World). On the left-hand side
is the space for writing Lean code, while the right-hand side contains two windows: The top window displays the context and
the goal, and the bottom window shows any error messages.

Figure 2. This figure shows the continuation of the proof from Figure 1, reflecting the updated goal after applying different
tactics.

to each of them we add one or two references to one of the key ideas of a proof but at best check line-by-line
many papers that reports research in that area: reasoning [MRFW+ 12].
• Proof writing: Several studies show that students
• Epistemological difficulties: It is well docu- and their lecturers do not share the same under-
mented that students arriving at a university often standing on what constitutes an “academic style”
fail to realize the role that proof has in pure math- of proof writing or of the difference between nat-
ematics and why it is necessary [GBD+ 16]. ural and technical language [LMR19].
• Proof comprehension: How well students un- • Proof appraisal: Students are often unable to as-
derstand proofs which are presented to them has certain the validity of proofs when they read them,
been a preoccupation of researchers for many often trusting the “authority” of the person who
years. The overwhelming message is that students presented the proof to them (e.g., their teacher)
often do not have a general understanding of the [SS08].

742 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY VOLUME 72, NUMBER 7

 EDUCATION

• Proof composition: Several students’ difficulties Our Next Step—Students’ Interaction with Lean
are found regarding proof composition. Among One of the outcomes of our first study, as also noted in
these there is the inability to use the definitions [BBB+ 24], was that teaching mathematical content that
in the context of a proof, the uncertainty of how students found difficult and teaching how to use Lean at
to start a proof, and how to partition a proof into the same time was not helping the majority of students
goals and subgoals [Pol14], [Moo94]. with either task, especially in 2018/2019 when there were
In the first study, we explored whether interaction with no Lean resources but the manual which is—by mathe-
Lean could help students overcome at least some of the dif- maticians’ own admission—difficult to use. At this time,
ficulties mentioned above. Therefore, the focus was on as- and with these difficulties in mind, games such as the Nat-
certaining the impact of engagement with Lean on writing ural Number Game were constructed (here we refer to the
proofs with pen and paper. We collected task-based inter- Natural Number Game in Lean 3). This game has become
views with students in an introduction to proof course at a popular with students as an entry point for Lean; for exam-
university in the UK and two surveys focusing on students’ ple, this was the first task that students in the Lean club at
use of and perspectives on Lean. For the analysis of the Edinburgh University engaged with. One of Lean’s key fea-
task-based interviews, we considered 36 proofs of an un- tures, also present in the Natural Number Game(s), is its
seen statement the students were asked to prove. We then ability to provide immediate feedack on each line of code
asked a mathematician external to the project to score the the user types.
36 proofs we had collected from 0 (no attempt made) to 4 We decided to investigate whether just engaging with
(a proof that would get full marks) so that we could com- the Natural Number Game—not supported by any input
pare proofs that were scored at the same level. Through the by a teacher—could help first-year students with mathe-
comparison between proofs written by Lean users and non- matics learning, especially with using mathematical lan-
Lean users which scored equally, we found that Lean users guage correctly, and composing and understanding proofs.
used mathematical language more precisely (proof writing We thought about this in terms of studying the way in
[TI22]). Lean users also structured proofs in “chunks” by which a human (a student) interacts with a tool (Lean) in
explicitly reporting goals and subgoals of the proof (proof order to complete a goal (write a proof). We adopted an
composition [Pol14]). For some students the need of for- approach that takes into account the actions performed by
malizing results they were familiar with helped the shift the human while using the tool, the reasons why such ac-
from naïve to more advanced views of mathematics (math- tions are performed, and the concepts (mathematical or
ematics epistemology [IT23]). However, we also noticed otherwise) which appear relevant to the human in that
that for some students who were Lean users the perceived situation [Tro05]. During the summer of 2022, we inter-
need for formalization got in the way of writing a relatively viewed four students for a total of approximately 70 hours
simple proof, and sometimes they deployed a very heavy while they made their way though the Natural Number
logical language, which may have hampered their mathe- Game. We found that even with four students the ways
matical intuition. As engaging with Lean was not a com- of engaging with the game could be very different. The
pulsory part of the course, after the third week of the course first observation we made was that these students, who
only 18 out of 300 students said they still worked with had just finished the first year of a mathematics degree
Lean and attended the optional Lean workshops. Reasons in a university in the UK, already had their own ways of
for nonengagement the students gave included difficulties tackling mathematics problems which they developed at
with the Lean syntax, competing demands on their time school and then at the university. They had also prior pro-
from other compulsory parts of their degree, and—more gramming experiences which shaped the way they worked
worryingly—the lack of appreciation of how what they saw within the programming context. Now they were asked
as a programming activity could relate to learning pure to somehow link these “ways of reasoning and acting” to-
mathematics and proof. The latter issue was unexpected gether while programming proofs in Lean. This interac-
and could be the outcome of the fact that programming tion between these “ways of reasoning and acting” gave rise
in university mathematics is generally linked to computa- to very different outcomes. One of the students, for exam-
tional tasks, statistical analysis, and mathematical model- ple, always referred to what they would do in a pen-and-
ing. So whereas on the one hand this implementation of paper proof before they wrote lines of code in Lean. Their
Lean was not successful in engaging a high number of stu- actions (writing the lines of code) were justified by the be-
dents, we were hopeful that a differently organized interac- lief that the proof should follow the outline of the proof
tion with Lean could have more benefits in the transition that they would write if they were not interacting with the
to proof. tool (the ITP). In this way the student did not perhaps de-
ploy the full power of the ITP, but they were able to check

AUGUST 2025 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY 743

 EDUCATION

and refine the proof they were writing. They also admitted Beginning students may say that it is the set of integers be-
that at times the ITP helped them to see inconsistencies tween 1 and 𝑛.
and mistakes in the pen-and-paper proof they would have These are type errors. To help give students feedback on
not otherwise seen. This back and forth between program- type errors, I used the strongly typed functional program-
ming in Lean and thinking about how to write the corre- ming language Haskell to run outreach courses with high
sponding pen-and-paper proof helped them clarify their school students in the 2010s. As in Lean, certain mathemat-
proofs and occasionally find problems in them and—we ical objects can be expressed in the Haskell language. If a
argue—helped this student with proof writing and proof student uses a value 𝑓(𝑥) ∈ 𝑌 where a function 𝑓 ∶ 𝑋 → 𝑌
composition. On the other end of the spectrum, one of is expected, Haskell responds with an appropriate error
the students interviewed consistently “played the game.” message. However, Haskell is not designed to deal with
Their actions were reactive, trying out lines of code when proofs.
they didn’t know what to do and reacting to the feedback Proofs, like functions and sets, are mathematical ob-
provided by Lean. They also often utilized lines of code jects, albeit objects of far greater complexity. As dis-
that closed multiple goals at the same time. We argue that cussed in the previous section, this complexity presents
this interaction may not be helpful for learning how to challenges for students when understanding and writing
write proofs as it depends on the reactive engagement with proofs. For many beginning students, the logical internal
the Lean feedback. We concluded that, in order for Lean structure of a proof is not immediately apparent. Students
to be helpful to students to overcome at least some of the are often unsure of how to begin a proof and cannot tell if
difficulties they encounter in the first year of university re- their proofs are correct.
lated to proof, it must be carefully integrated into teaching. This is where Lean is most effective. To write a proof
In the next part of this article, our colleague Gihan Maras- in Lean is to construct a (dependent) function of an ap-
ingha will write about how he used Lean in his teaching. propriate type. The input to the function is the hypothesis
and the output is the conclusion on the theorem. This de-
One Experience of Teaching Mathematics scription may seem baffling to the uninitiated, but Lean’s
with Lean syntactic sugar ensures that the body of such a function
I (Gihan) have taught an introductory mathematics class resembles, at least in structure, a traditional proof.
of over 200 students at the University of Exeter in Eng- In representing propositional logic, Lean treats logical
land for many years. The topics taught include sets, func- connectives via rules of inference. Thus, the implication
tions, logic, and the basics of group theory, linear al- 𝑝 → 𝑞 is defined by two rules: (1) To prove 𝑝 → 𝑞 is to
gebra, number theory, and analysis. In 2019, I intro- assume 𝑝 and deduce 𝑞 and (2) 𝑞 follows given proofs of
duced Lean-based teaching resources and have tried dif- 𝑝 → 𝑞 and 𝑝. This is not a new idea and is familiar to math-
ferent approaches since then. These include delivery via ematical logicians, but this approach was absent from my
interactive lecture notes, Lean “games,” and standard lec- undergraduate education. It is more intuitive than truth ta-
ture notes supplemented via Lean environments hosted bles and is a more accurate model of the natural deductive
on cloud-computing platforms; and they include chang- processes of real mathematics.
ing between noncredit-bearing and credit-bearing student Moreover, Lean reifies the narrative of a proof by pre-
submissions. senting the user, at each stage, with one or more goals. Each
This experience, and my engagement with mathematics goal consists of a target, that which is to be proved, and the
education research, has made clear the issues faced by be- context, the set of initial premises together with the vari-
ginning undergraduates. In what follows, I discuss my in- ables and facts that have been introduced.
troduction of Lean in that class as a way to address some of For example, consider the task of proving that for every
these issues, commenting on the benefits and challenges. integer 𝑥, if 𝑥 > 5, then there exists an integer 𝑦 such that
The transition from high school to university can be dif- 𝑥𝑦 > 5. The first step of the proof is to assume 𝑥 is an inte-
ficult. High school mathematics focuses on processes for ger. This introduces the integer variable 𝑥 into the context
manipulating algebraic expressions, rather than develop- and changes the target to one of proving that if 𝑥 > 5, then
ing a conceptual understanding of the underlying mathe- there exists an integer 𝑦 such that 𝑥𝑦 > 5.
matical objects. This leaves students confused regarding The next step is to assume 𝑥 > 5. This new assumption
the difference between a function 𝑓 ∶ 𝑋 → 𝑌 and a value is introduced into the context and the target changes to one
𝑓(𝑥) in 𝑌 , for 𝑥 ∈ 𝑋. Likewise, students can struggle with of proving that there exists an integer 𝑦 such that 𝑥𝑦 > 5.
the distinction between a set, its elements, and related con- All these changes are presented to students as they pro-
structions. A classic example is the confusion students en- ceed with their proof. By working with the goal(s), a stu-
counter when asked: “What is the symmetric group 𝑆𝑛 ?” dent is led more concretely through the proof journey.

744 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY VOLUME 72, NUMBER 7

 EDUCATION

Moreover, they come to make a connection between each cannot be expected to cope with error messages without
line of the proof and its effect on the goal state. instruction.
The two biggest improvements in my students’ proof un- The skeptic could argue that any benefit attributable to
derstanding that I can attribute to Lean are (1) their use the use of Lean is outweighed by the time and effort in-
of the language of goals and targets, even in later courses, volved in dealing with the formal aspects of Lean and error
to self-explain when working through a proof and (2) the messages, all of which seem superfluous to mathematics.
ability to identify what is required to prove any given kind However, the “coding” skills needed to write correct Lean
of statement. For instance, knowing that to prove a state- code are directly transferable to traditional proof writing
ment ∃𝑥, 𝑃(𝑥) is to exhibit a value 𝑥 together with a proof of and have the benefit that students get immediate feedback
𝑃(𝑥). By combining these skills, students are able to make on proof correctness. One such coding skill is the art of de-
progress with some proofs entirely mechanistically. This composing a complicated function into smaller functions.
improvement in mathematical thinking and communica- But this is precisely what professional mathematicians do
tion through the use of Lean agrees with the mathematics when expressing a (hoped-for) theorem in terms of lem-
education findings presented in the previous section. mas. Lean will even check that the skeleton proof is correct,
Lean is by no means limited to simple results in logic. assuming the truth of conjectured lemmas.
Its foundation is rich, allowing easy representation of stan- One may legitimately question the extent to which fa-
dard mathematical objects such as group, vector space, cility with Lean transfers over to facility with traditional
and topological space. Indeed, most topics in undergradu- proof. Certainly some theorems can be proved by referring
ate mathematics are currently represented in Lean’s math- to Lean’s output of the goal state at every step and throwing
ematical library, Mathlib. For instance, students can use Lean tactics at the problem until something works. If the
Lean to prove Lagrange’s theorem on the order of a sub- student sees no connection between the Lean proof and
group of a finite group. their handwritten work, this is indeed a problem. It is less
This richness and the game-like nature of Lean mean of a concern for more advanced theorems, where “tactic
that many students find it fun and addictive, going on to bashing” simply will not work. This reiterates the need, es-
work with Lean and other Interactive Theorem Provers on pecially for beginners, to require them to give both Lean
their own time, either for further study of formal mathe- and handwritten proofs of their arguments.
matics or for applications in computer science. There are many options for running Lean. The lowest-
Teaching with Lean requires thought and planning. In friction option is to use the Lean 4 web editor, running
practical terms, learning mathematics and Lean simultane- over the internet. This requires no installation and no
ously is too overwhelming for beginning undergraduates. preparation. This doesn’t permit working with multifile
Part of my approach is to run Lean computer labs on the projects, but it is a great way to get started and presents a
topics students have encountered at least a week previously similar interface to a standard Visual Studio Code installa-
in lectures. tion.
Beginning students will not automatically see the con- Another no-install option is to use a cloud computing
nection between Lean and mathematical proofs. I find this service like CoCalc, GitHub CodeSpaces, or GitPod. I
is often because they have not yet developed a strong un- have used two of these services successfully with my un-
derstanding of what a proof is, as referred to in the previ- dergraduates. Some cloud-computing services offer class-
ous section. Just as some students will present logically room management, making it possible for students to sub-
nonsensical arguments that look like traditional proofs, mit automatically graded Lean coursework. I have used
they will fail to recognize that a Lean proof is structurally graded coursework in my class. This improves engagement
identical to a traditional proof, simply because it looks dif- with Lean, but you should be aware of the potential for pla-
ferent. For Lean to be useful, a Lean workshop should in- giarism.
clude time writing traditional proofs and translating be- Lean “games,” running on the Lean Game Engine, are
tween them and Lean. an excellent alternative for self-study. The latest version of
A major hurdle for teaching with Lean is that, being a the engine permits users to enter proofs a line at a time,
formal computer language, it is unforgiving. A misplaced provides hints, and maintains a store of useful Lean tactics
colon or unmatched parenthesis will result in an angry red and theorems. A game designer can restrict which tactics
error message. Students, especially those without coding and theorems are available to the user at any stage in the
experience, find this frustrating. Their original challeng- game.
ing task, that of learning mathematical proof, has been Lean Verbose (https://github.com/PatrickMassot
doubled by the task of learning to code in Lean. Students /verbose-lean4), led by Patrick Massot, is a controlled
natural language interface to Lean. It allows the user to

AUGUST 2025 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY 745

 EDUCATION

write Lean proofs in a formal language that looks much tiated by mathematicians and involves a tool now used
more like a natural language than standard Lean. for research in mathematics. Furthermore, the affordances
Most of my Lean-using colleagues will tell you that us- and drawbacks of the use of Lean in terms of students’
ing Lean has changed how they think about mathematics. learning are being collaboratively investigated by mathe-
It has certainly changed how I teach mathematics. Early maticians and mathematics educators.
material in my lecture notes focuses on developing the abil- The Lean community is now very active and constantly
ity to parse expressions, building mental models of math- growing, which also means that the mathematical library
ematical objects, including proofs. Students then instan- is expanding and more mathematical topics are formal-
tiate and check these mental models in Lean. Making a ized. At the same time, there are several conferences about
strong connection between traditional mathematics and using Lean (or Interactive Theorem Provers) in research
Lean makes proof much more concrete and accessible. (https://lftcm2023.github.io) and mathematics
teaching (https://sites.google.com/view/learning
Some Concluding Remarks -math-with-lean-events/home). Our next steps, as edu-
In this article, we discussed the possible benefits as well cational researchers, will be to investigate the implementa-
as some of the drawbacks of using Lean as part of a first- tion of a series of resources designed for a first-year course
year introduction to proof course. We presented both re- that is supported by Lean and the impact that those re-
sults from current research in mathematics education and sources have on students’ learning about proof. For those
the experience of using Lean in teaching. The picture that who wish to incorporate this tool in their classrooms, the
emerges suggests that introducing this tool in teaching next steps involve being informed by the experience of oth-
needs to be done with caution: There is the need to cre- ers and engaging with considering the potential outcomes
ate suitable resources for the students, and there are some of using the tool within their specific contexts.
initial hurdles to overcome for both students and teachers.
However, the results may be worth the effort not only in References
terms of enhancing students’ learning about proof but also [BBB+ 24] Jonas Bayer, Christoph Benzmüller, Kevin Buz-
to improve their programming skills, which are highly val- zard, Marco David, Leslie Lamport, Yuri Matiyasevich,
Lawrence Paulson, Dierk Schleicher, Benedikt Stock, and
ued in the workplace.
Efim Zelmanov, Mathematical proof between generations,
There is one more interesting aspect of the current pop-
Notices Amer. Math. Soc. 71 (2024), no. 1, 79–92.
ularity of Lean (and of Interactive Theorem Provers in gen- MR4693604
eral) with students and with mathematicians who want to [DW23] Paul Christian Dawkins and Keith Weber, Identifying
use this tool in their teaching. This is one of the very few minimally invasive active classroom activities to be developed
examples of a tool that mathematicians may use for their in partnership with mathematicians, Mathematicians’ reflec-
research entering their teaching practice. Of course there tions on teaching: A symbiosis with mathematics educa-
are other computer-based tools used in teaching mathe- tion theories, 2023, pp. 103–121.
matics (e.g., Python), but these are not designed specif- [dB94] N. G. de Bruijn, A survey of the project Automath
ically for mathematics research. The way in which we [MR0592822 (81m:03017)], Selected papers on Automath,
Stud. Logic Found. Math., vol. 133, North-Holland,
teach mathematics at a university has often been criticized
Amsterdam, 1994, pp. 141–161, DOI 10.1016/S0049-
for being detached from the way in which mathemati- 237X(08)70203-9. MR1429401
cians do mathematics, and this has had undesirable con- [GBD+ 16] Ghislaine Gueudet, Marianna Bosch, Andrea A
sequences for how students think about and learn mathe- DiSessa, Oh Nam Kwon, and Lieven Verschaffel, Tran-
matics. Moreover, [DW23] discusses the viability of ped- sitions in mathematics education, Springer Nature, 2016,
agogical interventions in university mathematics teaching https://library.oapen.org/bitstream/handle
and explores why so many pedagogical interventions are /20.500.12657/27698/1/1002308.pdf.
started by enthusiastic colleagues but very few survive in [IT23] Paola Iannone and Athina Thoma, Interactive theorem
the long term. The authors conclude that pedagogical in- provers for university mathematics: an exploratory study of stu-
dents’ perceptions, Int. J. Math. Educ. Sci. Technol. (2023),
terventions are unlikely to endure beyond their initial im-
1–23.
plementation if they are disconnected from the practices [LMR19] Kristen Lew and Juan Pablo Mejia-Ramos, Linguistic
of doing mathematics and are designed by “others” (re- conventions of mathematical proof writing at the undergraduate
searchers in education) without a strong collaborative ap- level: Mathematicians’ and students’ perspectives, J. Res. Math.
proach. Using Lean—and Interactive Theorem Provers in Educ. 50 (2019), no. 2, 121–155.
general—to support the teaching of transition to proof
courses could be an example of a lasting pedagogical in-
novation. Its adoption originates from course designs ini-

746 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY VOLUME 72, NUMBER 7

 NEW FROM THE
+
[MRFW 12] Juan Pablo Mejia-Ramos, Evan Fuller, Keith We-
ber, Kathryn Rhoads, and Aron Samkoff, An assessment
model for proof comprehension in undergraduate mathematics,
Educ. Stud. Math. 79 (2012), 3–18.
[Moo94] Robert C Moore, Making the transition to formal proof,
Educ. Stud. Math. 27 (1994), no. 3, 249–266.
[Pol14] G. Polya, How to solve it: A new aspect of mathe-
matical method, Princeton Science Library, Princeton Uni-
versity Press, Princeton, NJ, 2014. With a foreword by
John H. Conway; Reprint of the second (2004) edition
[MR2183670]. MR3289212
[SS08] Annie Selden and John Selden, Overcoming students’
difficulties in learning to understand and construct proofs, Mak-
ing the connection: Research and teaching in undergradu-
ate mathematics (2008), 95–110.
[Tao25] Terence Tao, Machine-assisted proof, Notices Amer.
Math. Soc. 72 (2025), no. 1, 6–13. MR4845676
[TI22] Athina Thoma and Paola Iannone, Learning about proof
with the theorem prover LEAN: the abundant numbers task, Int.
J. Res. Undergrad. Math. Educ. (2022), 64–93.
[Tro05] Luc Trouche, An instrumental approach to mathemat-
ics learning in symbolic calculator environments, The didacti-
cal challenge of symbolic calculators: Turning a computa-
tional device into a mathematical instrument (2005), 137–
162.

Invariants of Links and

3-Manifolds from Graph
Configurations
Christine Lescop, CNRS and Université Grenoble
Alpes, France
This self-contained book explains how to count graph
configurations to obtain topological invariants for
Paola Iannone Gihan Marasingha 3-manifolds and links in these 3-manifolds, and it
investigates the properties of the obtained invariants.
A publication of the European Mathematical Society (EMS). Distributed within the Americas by
the American Mathematical Society.

EMS Monographs in Mathematics, Volume 12; 2024; 587 pages

Hardcover: List Price: US $109.00; Order Code: EMSMONO/12

AMS and MAA member discounts apply.

Explore more titles at

Athina Thoma bookstore.ams.org/EMSMONO
A publication of the European Mathematical Society (EMS).
Credits Distributed within the Americas
by the American Mathematical Society.
Figures 1 and 2 and photo of Paola Iannone are courtesy of
Paola Iannone.
Photo of Gihan Marasingha is courtesy of Gihan Marasingha.
Photo of Athina Thoma is courtesy of Athina Thoma.

AUGUST 2025 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY 747

 HISTORY

Apropos the Long-Lost Letter

from Levinson to Smale,
Now Found!
Jair Koiller and Indika Rajapakse

A pivotal moment in the emergence of modern dynami- on dynamical systems [Sma60], which had claimed that
cal systems theory was a famous letter written by Norman Morse-Smale systems, as they are now called, are dense.
Levinson to Stephen Smale in early 1960. Although men- (In hindsight, if Smale had been correct, it would imply
tioned frequently in the literature, this document had been that chaos does not exist.)
thought to be lost for many years. But some blunders are blessings. We now know that
As recounted in his article “Finding a horseshoe on the Levinson’s letter ultimately led Smale to discover the now-
beaches of Rio” [Sma98], Smale had been introduced to famous horseshoe map,1 dynamical systems theory got
Mauricio Peixoto by Elon Lima while the three of them an enormous boost, and chaos theory became one of the
were visiting the Institute for Advanced Study (IAS) at paradigms of modern science [AD02].
Princeton in the late 1950s. Lima and Peixoto subse- But what happened to the long-lost letter?
quently invited Smale to visit the recently established In- And did Smale reply?
stituto de Matemática Pura e Aplicada (IMPA) in Rio de The answers—along with an additional discovery—are
Janeiro for six months before he joined Berkeley in the fall contained in this article.
of 1960. Levinson’s letter found him soon after he arrived Not long after Mauricio Peixoto passed away in 2019
in Rio. at the age of 98, his widow, Alciléa,2 donated about fifteen
In the letter, Levinson pointed to his own Annals of boxes of personal documents to the Museu de Astronomia
Mathematics paper from 1949 [Lev49] and an earlier study e Ciências Afins (MAST), located in Rio de Janeiro.3 By a
by Mary Cartwright and J. E. Littlewood [CL45] to argue stroke of luck, on November 15, 2022, Jair Koiller discov-
that smooth ODEs which arise in practical applications ered the famous letter from Levinson to Smale during his
may have complicated solutions, also having an infinite first visit to MAST, when he went to take a look at Peixoto’s
number of periodic orbits that could not be perturbed Nachlass. It was dated February 20, 1960, and contained
away. This was both shocking and embarrassing to Smale, Levinson’s suggestion, in his characteristically neat hand-
since it contradicted a key conjecture in his very first paper writing, that Smale’s conjecture “may not be valid” (see
Figure 1).
Jair Koiller is a visiting professor at the Instituto de Fı́sica da Universidade do But that was not all. Along with this important docu-
Estado do Rio de Janeiro, R. São Francisco Xavier, Brazil. His email address is ment, hidden inside a notebook on differential equations,
[email protected]. was a second letter, dated March 22, 1960, this time also
Indika Rajapakse is a professor in the department of computational medicine
and bioinformatics and mathematics department at the University of Michigan. 1
We assume our readers know about the horseshoe map. If not, we recommend
His email address is [email protected]. the very nice article by Michael Shub [Shu05].
2
Communicated by Notices Associate Editor Adrian C. Rice. Alciléa Augusto Homem de Mello is a prominent figure in mathematics educa-
For permission to reprint this article, please contact: tion in Brazil.
3
MAST archives are open to researchers. Jair Koiller, together with Ronaldo
[email protected].
Garcia, Mario Jorge Carneiro, and Antonio Roberto da Silva, plans to look at
DOI: https://doi.org/10.1090/noti3167 materials and correspondence in Mauricio’s papers.

748 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY VOLUME 72, NUMBER 7

 HISTORY

life story told by his wife, Zipporah (“Fagi”), should be

more widely known [Seg09].
Morse-Smale systems. Conjecture (A). The class of
Morse-Smale systems (so named by René Thom) was intro-
duced in Smale’s first work on dynamical systems [Sma60].
The idea was to extend for arbitrary-dimensional closed
manifolds 𝑀 the planar systems 𝑋 considered by Peixoto
in his 1959 Annals of Mathematics paper [Pei59]. The aim
was to provide Morse inequalities like those in gradient
systems.
Morse-Smale systems are characterized by having (1)
only a finite number of fixed points and (2) periodic or-
bits, of hyperbolic type (the name did not seem to exist at
that time). For the former, this means that all the eigenval-
ues of the linear part have nonzero real part; for the latter,
no characteristic exponent has absolute value 1. In addi-
tion, (3) the 𝛼- and 𝜔-limit sets (i.e, for 𝑡 → ±∞) of all
orbits are in their union; (4) there is transversality4 of the
stable and unstable manifolds; and (5) no closed orbit is
an 𝛼- and 𝜔-limit of a point.
Smale began the paper by remarking that such systems
may be very important because of the following possibili-
ties:

(A) It seems at least plausible that systems satisfy-

ing (1)–(5) form an open dense set in the space
(with the 𝐶 1 topology) of all vectorfields in 𝑀.
Figure 1. Levinson’s first letter, referring to his 1949 paper
(B) It seems likely that conditions (1)–(5) are nec-
[Lev49], casting doubt about “possibility (A).” essary and sufficient for 𝑋 to be structurally stable
in the sense of Andronov and Pontrjagin.5
intended for the eyes of Mauricio Peixoto (see Figures 2 But Smale’s confidence was dealt a substantial blow
and 3). when he received the letter of February 1960, in which
In an interview in June 2024, Smale told us that he had Levinson described a result of his from 1949 which effec-
replied to Levinson and even met with him briefly at one tively contained a counterexample to conjecture (A). Af-
point. It would seem that Smale gave the letters to Mauri- ter receiving Levinson’s letter, Smale read Birkhoff, Morse,
cio, who had kept them all these years. Why did he not Hedlund, and, of course, Poincaré, and he realized “how
publicize them? Had he forgotten about them? When we crazy the idea was” (his own words).
asked Smale about this, his answer was: “Look, this was 63 But what was the development in the second letter that
years ago!” We hypothesize that Smale had a similar mem- Levinson asked Peixoto and Smale to verify? Just to say
ory lapse about the letters after giving them to Peixoto. “the estimate (2.4) extends readily to all perturbations”
In drawing attention to the discovery of these letters, our [for system (2.1)], as it appears in the first letter, would
goal is to highlight: be incomprehensible to the reader.
(i) the impact of the horseshoe map’s début on So- Our point is exactly this: Pursuing that estimate was not
viet mathematicians after Smale went to Kiev in interesting to Smale. Indeed, we quote Dmitrii Anosov be-
1961 and low on the point in Levinson’s paper that caught Smale’s
attention.
(ii) the friendship of Smale with Elon Lima and
Mauricio Peixoto and his legacy in Brazil.
We also wish to note Levinson’s admirable moral char- 4Here’s Thom’s influence—Smale attended his lectures in Chicago.
acter. In addition to being a life-long opponent of all 5Structurally stable, as defined by Andronov and Pontrjagin, means that small
forms of racism and discrimination, Levinson was praised changes in the system’s parameters do not lead to significant changes in its over-
by his colleagues for his great generosity [Lev98]; and his all behavior.

AUGUST 2025 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY 749

 HISTORY

Figure 2. Levinson’s second letter contains the masterly Figure 3. Levinson’s second letter, second page.
understatement: “Let me sketch the simple argument so that
Peixoto or you can point out the hole if I have a blind spot.”
that, prior to seeing the paper by Cartwright-Littlewood,
But perhaps Smale could interest Peixoto in pursuing Levinson’s intuition was the same as Smale’s: How could
Levinson’s arguments? This is why we think Smale handed smooth ODEs “behave badly”? So, ten years before he
the letters over to him. wrote his famous letter, Levinson had also been shocked!
What was Levinson’s counterexample? During the Sec- Having reviewed their paper for Math Reviews, Levinson
ond World War, the time-dependent second-order equa- decided to give a simpler example with the same “bizarre”
tion 𝑥̈ + 𝜅𝑃(𝑥)𝑥̇ + 𝑥 = 𝑐𝜅 sin 𝑡, 𝑃 an even polynomial, with phenomena. His model was equation (2.0) in [Lev49]: 𝜖𝑥+ ̈
𝜅 large, had appeared in the study of high-frequency ra- 𝜙(𝑥)𝑥+𝜖𝑥
̇ = 𝑏 sin 𝑡, 𝜖 > 0 small, 0 < 𝑏 < 1. Levinson’s trick
dio waves (radar); in our current terminology it is a time- was first to take 𝜙(𝑥) piecewise constant, = 1 for |𝑥| > 1
dependent “slow-fast” planar system with relaxation oscil- and = −1 for |𝑥| < 1, mimicking van der Pol’s 𝑃. Each
lations. In [CL45], Cartwright and Littlewood studied this piece, being linear, could be solved in closed form. But one
differential equation taking van der Pol’s 𝑃 = 𝑥2 − 1 and needed to track the interchanges to construct the planar
found that solutions exhibited unexpected and unusual be- map for (2.0) (or equivalently (2.1)–(2.2)). This is how
havior.6 the “singular solutions” appeared.
In his appraisal of Levinson’s work on nonlinear os- Estimate (2.4), referred to in Levinson’s first letter,
cillations (Selected Papers, [Lev98]), Jürgen Moser wrote means that the behavior of these singular solutions is ro-
bust vis-a-vis polynomial approximations. The suggestion in
6A short account on why Mary Cartwright was commissioned to study this par- the second letter was just to verify that arbitrary 𝐶 1 pertur-
ticular kind of nonlinear differential equation during World War II can be bations could also be allowed. Smale took the validity of
found in https://www.bbc.com/news/magazine-21713163. In 1969 she this fact for granted. Being a topologist, he preferred to
became Dame Mary Cartwright, Commander of the Order of the British Em-
pire. It is known that Wiener and his students also worked on radar during the
distill the essence of the phenomena.
Second World War; Levinson may also have participated in this and other clas- But for Smale to understand Levinson’s counterexample
sified work, but we did not find any evidence. fully, it was necessary to translate the analytic arguments

750 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY VOLUME 72, NUMBER 7

 HISTORY

Smale, who was already famous in topology, but

not yet in dynamical systems, wrote to Novikov
that he was coming to Kiev and wanted to stop in
Moscow and meet him, and maybe other young
Moscow mathematicians. . . . I went to Kiev mainly
to assure him that Novikov and other advanced
young people would be in Moscow. While stand-
ing in the registration line and looking over some-
one’s shoulders ahead, I read the title of one of
them: ‘S. Smale. A structurally stable, differen-
tiable homeomorphism with infinitely many pe-
riodic points.’ At this moment the world turned
over for me and a new life started.
In addition to his formal lecture, Smale was
kind enough to explain his discovery (‘Smale’s
horseshoe’) in more detail to a group of inter-
ested participants. There were several people from
Gorky; from Moscow I remember only myself and
M. M. Postnikov. Then he went to Moscow, met
Novikov, Arnold, Sinai and myself at the Steklov
Mathematical Institute and gave us still more de-
tails.8
Anosov guessed Smale’s train of thoughts as follows:
He felt that he was a God who had to create a
world with certain phenomena: how could he
do it? He extracted from the papers of the three
authors (or maybe from Levinson’s letter?) that
Figure 4. The notebook with the letters. strong friction and a large perturbing force are
present in the corresponding examples, and at the
into geometric form, which is what led him to the discov- same time, the trajectories stay in a bounded re-
ery of the horseshoe map. gion in the phase space.
From Rio to Kiev and Moscow, 1961. The horseshoe map The first can be modeled geometrically by as-
received its international début during the First Symposium suming that trajectories approach fast in one direc-
on Non-Linear Vibrations under the auspices of the In- tion and diverge fast in another. We can consider
ternational Union of Theoretical and Applied Mechanics not a continuous motion, but rather iterations of a
(IUTAM), September 12–18, 1961, in Kiev (then in the map (in this case a certain natural map—time shift
USSR). The symposium was chaired by Yu. Mitropolsky by the period is present), so that we can imagine a
and N. N. Bogolyubov, leaders of the Ukrainian school of square becoming a long and narrow rectangle.
nonlinear ODEs. But since the trajectories stay in a bounded re-
The scientific committee also included Mary Cartwright gion in the phase space, we should bend the rec-
(UK), C. Hayashi (Japan), and Solomon Lefschetz (USA). tangle so that it does not leave the region.
Interestingly, in a commemoration of 50 years of ICNOs,7
Anosov also said that Smale should have known the
in the section on the 1961 symposium, Smale’s name ap-
“baker’s map,” a basic example of symbolic dynamics. If
pears among the senior scholars. He was 31.
so, his attention would have been caught already at the
Smale’s impact on the young Soviet mathematicians is
very beginning (i.e., the second page) of Levinson’s 1949
perhaps best described by Dmitrii Anosov in an interview
paper, where the singular solutions are described in terms
with Rostislav Grigorchuk for the European Mathematical
Society Newsletter [Gri03]): 8
An English version with the title “A Structurally Stable Differentiable Home-
omorphism with an Infinite Number of Periodic Points” is in Volume 2 of
7International Conferences on Nonlinear Oscillations: The first such gather- [Sma00]. Its Russian translation is in Nelin. Dinam. 3 (2007), no. 4,
ing was the Kiev symposium of 1961. http://www.mi.sanu.ac.rs/novi 445–446. For Smale’s papers in Russian: https://www.mathnet.ru/php
_sajt/news/an_doc/icnoenoc.pdf /person.phtml?option_lang=eng&personid=24381.

AUGUST 2025 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY 751

 HISTORY

We know from Smale that he answered Levinson, but he

was not sure if it was before or after finding the horseshoe.
Our guess is that it was no later than the talk he gave at
Berkeley in the summer of 1960.
But, given that Levinson was unsuccessful in luring
Smale for a joint paper, perhaps there was a different main
purpose for his second letter: a trial balloon.
Was there an intention to follow up with a job offer
from MIT? After all, Levinson was the “eminence grise” be-
hind the chair of MIT’s Department of Mathematics, Ted
Martin.10 (In our June 2024 interview, when asked about
the speculations contained in this article, Smale said in his
ironic but tactful way: “I abstain. I do not interfere in other
people’s work.”)
Figure 5. “My best-known work was done on the beaches of Smale’s mathematical legacy in Brazil. Levinson’s letters
Rio de Janeiro” [Sma90]. prompted Smale to devote over a decade to promoting
a “rebranding” of dynamical systems, culminating with
of arbitrary ±1 sequences. Following a remark about “pre- the outstanding paper [Sma67] which paved the way for
vious examples of discontinuous recurrent motion, in the several generations. But his most famous achievement
sense of Birkhoff for analytic differential equations,” Levin- was arguably his proof of the generalized Poincaré con-
son gave references to work by Hadamard, Birkhoff, and jecture for 𝑛 ≥ 5, for which he was awarded the Fields
Morse, in particular for “certain systems of two degrees Medal in 1966. As he related in “The story of the higher-
of freedom in classical mechanics including the restricted dimensional Poincaré conjecture” [Sma90], he completed
problem of three bodies.” the proof in that semester mirabilis in Rio. Gradient systems,
But why a second letter? What then were Levinson’s mo- actually a particular case of Morse-Smale systems, were
tivations, to write not just once, but twice? There would among the tools that he used.
have been little time to receive a response. And he could Incidentally, when asked about stories of Smale doing
have simply just pointed out his 1949 paper. We summon math at the beach, Mauricio Peixoto would answer with
the nerve to offer a speculation. his well-humored style:
Perhaps Levinson’s purpose in writing the second let-
Baloney! Steve was coming in to IMPA with dark
ter was to propose a collaboration between the three math-
circles and red eyes.11
ematicians. After all, his generous spirit is well attested
by his friends and colleagues. Moreover, Levinson and On more than one occasion, Smale wrote about Elon
Peixoto shared a common academic grounding, having Lima, whom he first met at the Symposium Internacional
both come from engineering backgrounds. And Peixoto de Topologia Algebraica, held in Mexico City in August
was only ten years younger than Levinson. 1956.12 Smale had just finished his PhD with Raoul Bott,
But Levinson could not have imagined realistically that and Lima was halfway through his thesis supervised by
a direct collaboration with Smale would materialize. With Spanier. They were 26–27 years old and immediately re-
his spectacular results in differential topology that revolu- alized they were like-minded in many ways.
tionized immersion theory [Spr05], Smale was a rising star Two years later Smale and Lima were postdocs at the In-
and looking to do more on his own. stitute for Advanced Study in Princeton. Mauricio Peixoto
Perhaps Levinson believed that it would have been good came there at the invitation of Solomon Lefschetz. At the
for both Smale and Peixoto to be able to add “nuance”: time, Peixoto was polishing his now-famous paper about
structural stability versus the very singular solutions in
three dimensions of Cartwright-Littlewood and himself.
Proposing a collaboration was not common at that time, 10Unfortunately, the MIT archives do not have Levinson’s correspondence from

but even if it had occurred, it would probably have been this period.
11Jair Koiller heard him talking about this tale a couple of times in the beautiful
in the form of a short note, since it required making only
coffee room of IMPA’s new building.
a rather simple, additional estimate.9 12That meeting became infamous after Witold Hurewicz sadly died after
falling from the top of a Mayan step pyramid. One can download the Pro-
9Smale and Peixoto never published together, but Smale does have joint works ceedings from https://paginas.matem.unam.mx/matematicos/images
with Jacob Palis. /sXX/mexico.pdf.

752 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY VOLUME 72, NUMBER 7

 HISTORY

structural stability in two dimensions that appeared in the

Annals of Mathematics [Pei59].13
Peixoto and Lima were able to lure Smale into a mathe-
matical adventure in Rio—although it seems to us quite ex-
traordinary that, as a recent PhD and untenured, he would
trade IAS for an unknown institute that had been in exis-
tence for only a few years. At the time Peixoto was almost
40. But in spite of their differences in age and political
views, Smale and Peixoto became friends for life.
In 1964 Peixoto sent Smale a very promising student
named Jacob Palis,14 who quickly finished a PhD thesis pre-
cisely on the subject of Morse-Smale systems (1967). De-
spite job invitations in the USA, Palis stubbornly decided
to return to Brazil to help IMPA, which became a very pres-
tigious international center and a hub for Latin American Figure 6. At Peixoto’s dacha in Petropolis, north of Rio,
students. mid-seventies. Left to right: Peixoto, Smale, and Palis.
Palis was very influential, not only in dynamical sys-
tems, but also as a leader in Brazilian and international sci- Aftermath (pun intended). In examining the newly dis-
ence. He was president of the Third World Academy of Sci- covered letters, we have uncovered a potential critical
ences (TWAS) and the International Mathematical Union point, a transition between generations. For Steve Smale,
(IMU). According to the Math Genealogy Project, he has the question was not only one of mathematical taste, but
330 descendants. also about his mathematical ambition. Throughout his ca-
Mauricio Peixoto [Pei93] and Jacob Palis [Pal93] gave reer, Smale consistently looked at the big picture, the foun-
testimonies in the Smalefest proceedings [HMS93] about dations, and opening up new pathways.
Smale’s importance for the development of the Brazilian Around 1960, Norman Levinson, a leading expert in
School of Dynamical Systems. To give just one recent nonlinear differential equations, made a surprising career
example, out of a miriad: Lorenzo Diaz, from Rio de shift to start anew in the hardest area of all mathematics,
Janeiro Catholic University, is one of the co-inventors of analytic number theory. In 1974, just a year before his un-
blenders, which are compact invariant sets where stable timely death from a brain tumor, he published a striking
and unstable manifolds, as the name suggests, blend to- paper on the Riemann Hypothesis. His finding that more
gether. It is therefore a natural descendent of the horse- than a third of the roots of the Riemann zeta function lie
shoe [BCDW16]. on the critical line was the most advanced result up to that
Smale’s vigorous influence continues today not only via time.
his former Brazilian students, but also through many other Could Smale’s work on dynamical systems have in-
Brazilian mathematicians working in dynamical systems, spired Levinson to take this bold step, at a mature age? By
mechanics, economics, complexity, information theory, transitioning to a new field, Levinson may have recognized
and biomathematics. that a new generation was emerging with a fresh perspec-
Internationally, “nonlinear science and chaos” became tive, focused on the global, qualitative aspects of dynami-
very common keywords in grant applications by experi- cal systems. This new wave of mathematicians would su-
mentalists. Recollections by some of the contributors to persede all previous generations, to which he, Levinson,
chaos theory from the pioneering time of the 1970s can also belonged—all of them following Newton’s advice to
be found in [LAS+ 21].15 look for “little pebbles and shells.”16
And Smale? While he continued to engage with foun-
13
The idea came from the Andronov school at the Gorky Physico-Technical Insti- dational questions in dynamical systems [Sma91], after a
tute, Nizhny Novgorod, 500 km from Moscow. Peixoto learned it from Lefchetz decade of work in the field he shifted most of his focus
and Smale learned from Peixoto.
14 to other areas. In a recent talk at IMPA, Smale discussed
The mathematical community mourns the loss of Jacob Palis, who passed away
during preparation of this article on May 7, 2025.
15For the record, chaos had already been noticed, but disregarded, by van der ceivers before the frequency jumps to the next lower value. However, this is a
Pol and van der Mark, as reported in a short letter in Nature in 1927 [vv27]. subsidiary phenomenon.” (The emphasis is ours.) Cartwright and Little-
The title was “Frequency demultiplication”—meaning periodic subharmonics. wood wrote that this phenomenon “gives faith in our results.” But as Smale
The following excerpt is worthy of notice (and was also mentioned in [Hol90]): commented, “The world wasn’t ready to listen.” See [LAS+ 21] for recollections
“In some recent experiments it was possible to obtain a frequency demultiplica- from some of the early contributors to chaos theory.
16
tion up to the ratio 1:1/200. Often an irregular noise is heard in telephone re- Always the beach!

AUGUST 2025 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY 753

 HISTORY

his work in bioinformatics, a topic he had been explor- [Hol90] Philip Holmes, Poincaré, celestial mechanics,
ing in Hong Kong. In a playful nod to his audience of dynamical-systems theory and “chaos”, Phys. Rep. 193 (1990),
dynamicists, he suggested that the chaotic nature of ODEs no. 3, 137–163, DOI 10.1016/0370-1573(90)90012-Q.
limited their predictive power. But we are sure this was just MR1076345
[LAS+ 21] Christophe Letellier, Ralph Abraham, Dima L. She-
a “blague.”
pelyansky, Otto E. Rössler, Philip Holmes, René Lozi, Leon
More recently, Smale has turned his attention to con- Glass, Arkady Pikovsky, Lars F. Olsen, Ichiro Tsuda, Celso
crete ODEs, particularly those related to synchronization Grebogi, Ulrich Parlitz, Robert Gilmore, Louis M. Pecora,
phenomena in cell biology. Currently, he is collaborating and Thomas L. Carroll, Some elements for a history of the dy-
with Dr. Indika Rajapakse to study synchronization in beat- namical systems theory, Chaos 31 (2021), no. 5, Paper No.
ing heart cells using Kuramoto oscillators and applying 053110, 20, DOI 10.1063/5.0047851. MR4254255
Morse-Smale theory. Thus, even today, Smale’s dynamic [Lev49] Norman Levinson, A second order differential equation
career continues to evolve.17 with singular solutions, Ann. of Math. (2) 50 (1949), 127–
153, https://doi.org/10.2307/1969357. MR30079
ACKNOWLEDGMENTS. Thanks to José Benito Abel- [Lev98] Norman Levinson, Selected papers of Norman Levinson.
Vol. 1, Contemporary Mathematicians, Birkhäuser Boston,
las and Assis Gonçalves from MAST Archives for their
Inc., Boston, MA, 1998. With a foreword by Henry P. McK-
invaluable help. Thanks to Marcio Rangel and Heloisa ean, Edited and with a preface by John A. Nohel and David
Bertol Domingues, MAST current and former directors. H. Sattinger. MR1491093
JK wants to thank Flavia Soares and Antonio Roberto [Pal93] J. Palis, On the contribution of Smale to dynamical sys-
da Silva for sharing afternoons at MAST, and Peixoto’s tems, From Topology to Computation: Proceedings of the
wife, Alciléa, for encouraging us to look at his Nachlass. Smalefest (Berkeley, CA, 1990), Springer, New York, 1993,
We also thank Lindsey Muir for the critical reading of pp. 165–178. MR1246117
our manuscript, and we sincerely appreciate the assis- [Pei59] M. M. Peixoto, On structural stability, Ann. of Math.
tance of Adrian Rice in editing the article into its present (2) 69 (1959), 199–222, https://doi.org/10.2307
/1970100. MR101951
form.
[Pei93] M. M. Peixoto, Some recollections of the early work of
Steve Smale, From Topology to Computation: Proceedings
References of the Smalefest (Berkeley, CA, 1990), Springer, New York,
1993, pp. 73–75. MR1246109
[AD02] David Aubin and Amy Dahan Dalmedico, Writing the
[Seg09] Joel Segel, Recountings: Conversations with MIT mathe-
history of dynamical systems and chaos: longue durée and rev-
maticians, CRC Press, 2009.
olution, disciplines and cultures (English, with English and
[Shu05] Michael Shub, What is . . . a horseshoe?, Notices Amer.
French summaries), Historia Math. 29 (2002), no. 3, 273–
Math. Soc. 52 (2005), no. 5, 516–517. MR2140094
339, DOI 10.1006/hmat.2002.2351. MR1917131
[Sma00] Stephen Smale, The collected papers of Stephen Smale.
[Bat00] Steve Batterson, Stephen Smale: the mathematician who
Vol. 1–3, Singapore University Press, Singapore; World Sci-
broke the dimension barrier, American Mathematical Society,
entific Publishing Co., Inc., River Edge, NJ, 2000. Edited
Providence, RI, 2000. MR1737026
by F. Cucker and R. Wong. MR1781696
[BCDW16] Ch. Bonatti, S. Crovisier, L. J. Dı́az, and A.
[Sma60] Stephen Smale, Morse inequalities for a dynamical
Wilkinson, What is. . . a blender?, Notices Amer. Math. Soc.
system, Bull. Amer. Math. Soc. 66 (1960), 43–49, DOI
63 (2016), no. 10, 1175–1178, DOI 10.1090/noti1438.
10.1090/S0002-9904-1960-10386-2. MR117745
MR3559345
[Sma67] S. Smale, Differentiable dynamical systems, Bull. Amer.
[CL45] M. L. Cartwright and J. E. Littlewood, On non-linear
Math. Soc. 73 (1967), 747–817, DOI 10.1090/S0002-9904-
differential equations of the second order. I. The equation 𝑦 ̈ −
1967-11798-1. MR228014
𝑘(1 − 𝑦2 )𝑦 + 𝑦 = 𝑏𝜆𝑘cos(𝜆𝑡 + 𝑎), 𝑘 large, J. London Math.
[Sma90] Steve Smale, The story of the higher-dimensional
Soc. 20 (1945), 180–189, DOI 10.1112/jlms/s1-20.3.180.
Poincaré conjecture (what actually happened on the beaches
MR16789
of Rio), Math. Intelligencer 12 (1990), no. 2, 44–51, DOI
[Gri03] R. Grigorchuk, Interview with D. V . Anosov, part 2, Eu-
10.1007/BF03024004. MR1044929
ropean Math. Society Newsletter 48 (2003), 15–20.
[Sma91] Steve Smale, Dynamics retrospective: great problems, at-
[HMS93] M. W. Hirsch, J. E. Marsden, and M. Shub (eds.),
tempts that failed, Phys. D 51 (1991), no. 1-3, 267–273,
From Topology to Computation: Proceedings of the Smale-
DOI 10.1016/0167-2789(91)90238-5. Nonlinear science:
fest, Springer-Verlag, New York, 1993. Held at the Uni-
the next decade (Los Alamos, NM, 1990). MR1128817
versity of California, Berkeley, California, August 5–9,
[Sma98] Steve Smale, Finding a horseshoe on the beaches of
1990, https://doi.org/10.1007/978-1-4612-2740
Rio, Math. Intelligencer 20 (1998), no. 1, 39–44, DOI
-3. MR1246102
10.1007/BF03024399. MR1601831
17
For those interested in reading a biography of Steve Smale, we recommend
[Bat00].

754 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY VOLUME 72, NUMBER 7

 MEMBERS,
?

AMS
[Spr05] David Spring, The golden age of immersion theory in

RELOCATING
topology: 1959–1973. A mathematical survey from a histor-
ical perspective, Bull. Amer. Math. Soc. (N.S.) 42 (2005), ARE
no. 2, 163–180, DOI 10.1090/S0273-0979-05-01048-7. YOU
MR2133309
[vv27] B. van der Pol and J. van der Mark, Frequency demul-
tiplication, Nature 120 (1927/09/01), no. 3019, 363–364, Please make sure that Notices
https://doi.org/10.1038/120363a0.
and Bulletin find their new home.

Jair Koiller Indika Rajapakse

Credits
Figure 1, Figure 2, Figure 3, and Figure 4 photos were taken Update your address at
by J. Koiller at MAST.
Figure 5 is courtesy of Márcia Foletto (Agência O Globo). www.ams.org/member-directory.
Figure 6 is courtesy of IMPA. You can also send address changes
Photo of Jair Koiller is courtesy of Jair Koiller. to [email protected] or:
Photo of Indika Rajapakse is courtesy of Lindsey Muir.
Customer Service Department
American Mathematical Society
201 Charles Street
Providence, RI 02904-2213 USA

AUGUST 2025 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY 755

 BOOKSHELF
New and Noteworthy Titles on our Bookshelf
August 2025

Navigating the Math Major: Making Democracy Count:

Charting Your Course How Mathematics Improves
MAA Press: An Imprint of the American Voting, Electoral Maps,
Mathematical Society, 2024, 172 pp.
and Representation
By Carrie Diaz Eaton, Princeton University Press, 2024, 408 pp.
Allison Henrich, Steven Klee,
and Jennifer Townsend By Ismar Volić

One of the issues of equity in You may already know about

education is who has access to ranked voting and various ways
the “secret syllabus” or “hidden to count ballots to determine a
curriculum” of college. Such winner. You may already know
a document would explain the about fairness criteria and Ken-
important steps it takes to be successful, many of which are neth Arrow’s impossibility theorem (if not, I won’t give
never explicitly discussed in a college’s orientation. Those it away right now). You may understand approval voting,
who accomplish the steps hidden within the secret syl- gerrymandering, apportionment, electoral maps, or repre-
labus often have more impressive resumes, do better in sentation. Like Volić in this book, you may have a com-
their coursework, and have more opportunities to partici- pelling argument for the “best,” although not mathemati-
pate in mathematical community events like attending a cally perfect, voting system. You may also have plenty of
conference, doing an REU, or going to graduate school. examples fabricated to show a certain paradox or violate
Consider this book to be the “secret syllabus” for math- a fairness criterion. If this is the case, what this book can
ematics majors. It contains tips about how to be a math- offer you is a plethora of examples, many from real elec-
ematics (or related) major, including suggestions of com- tions and not conjured on a computer, to illustrate each
munities to join, courses they will likely take, how to give of the topics above. I was utterly impressed with the num-
a talk or poster, what types of jobs to apply for, and more. ber of examples that come from elections as recent as the
This book is an important addition to any department, 2020s, the analysis of current district maps, and historical
whether you have a course that introduces your major or context for why we have the representation we do in the
not. While I believe it has something to offer every student, United States. Of course, if any of the topics above are
it seems most likely to benefit those who do not have ac- new to you, this book offers a very digestible presentation
cess to a mathematics-specific mentor or advisor. I believe of the material.
this book embodies part of the mission of the AMS, which With just enough history to create an understandable
seeks to “advance the status of the profession of mathemat- story, the author reveals the many ways mathematics
ics, encouraging and facilitating full participation of all in- shows up in democracy in the United States. This book
dividuals.” Certainly, this book not only encourages but contains focused chapters on each topic using terminol-
also provides explanations for how undergraduate mathe- ogy suitable for any interested reader. While it is not a
matics majors can fully engage with our discipline. textbook, it could be adopted for units on voting theory in
a course. I found the book to offer thought-provoking dis-
cussions of very relevant points about the structure of our
This Bookshelf was prepared by Notices Associate Editor Emily J. Olson. government (such as the size of the House of Representa-
Appearance of a book in the Notices Bookshelf does not represent an endorse- tives) and the problematic ways our votes count (or don’t).
ment by the Notices or by the AMS. While there may be no perfect system, we can use math to
Suggestions can be sent to [email protected]. do better in our democracy; learn how in this incredibly
DOI: https://doi.org/10.1090/noti3197
up-to-date book.

756 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY VOLUME 72, NUMBER 7

 CONVERSATION

Challenges and Opportunities

in Peer Review
Lance Fortnow, Robert Harington, Igor Pak,
Anne Schilling, and Mark C. Wilson

This is the edited transcript of a panel discussion held at William Whewell, who also invented the terms “physicist”
JMM2025 in Seattle. The participants were Anne Schilling, and “scientist,” deployed for the new journal Proceedings
Robert Harington, Igor Pak, and Lance Fortnow, and the panel of the Royal Society what we now think of as peer review.
was moderated by Mark C. Wilson. I hope that this article Originally, those peer reviews were published alongside
induces readers to submit their good ideas to the conversation. the actual articles themselves. That ebbed away and they
—Ed. were used in the main to prevent embarrassment to the
MW: (Why) do we still need peer review as we have Royal Society. That leads through to recent decades. I
known it for the last several decades? think funders started to embrace it in the 1960s and 1970s.
AS: A very short answer is that we get evaluated every I think there was some controversy, as they were forced
couple of years for promotion, and papers only count if into the idea that it was necessary to deploy peer review
they have been published in a peer-reviewed journal. If toward grants, and there became a much greater emphasis
you want to get tenure or if you want to get hired, peo- on anonymized peer review.
ple look at whether your papers were published or not. It I think peer review should be a rigorous evaluation and
does give some credit to the paper if you can say that it was report on a paper. Will it prevent fraud? I don’t think it
published and didn’t appear only on arXiv. will. I don’t think we can expect peer review to prevent
RH: The most widely accepted story about the origin fraud. Will a reviewer receive due credit for their peer re-
of peer review goes back to the seventeenth century and view? I don’t think that’s actually true at all. Certainly
the Royal Society. Henry Oldenburg, for the Philosophical toward tenure, it’s a very minor part depending on the in-
Transactions of the Royal Society, first wanted to use the idea stitution you’re at. Is it a way to improve and validate an
of peer review, though in reality, I don’t think he consulted article? I think you could argue that that’s true. Preprints,
anyone, and it wasn’t until the nineteenth century when currently at least, are not peer reviewed, but they’re very
much used, though in the end most researchers will refer
Lance Fortnow is dean of the College of Computing at Illinois Institute of Tech-
to the final peer-reviewed published version.
nology. His email address is [email protected].
I think the real question we need to answer is “what is
Robert Harington is chief publishing officer of the AMS. His email address is
[email protected].
the purpose of peer review?”
Igor Pak is professor of mathematics at UCLA. His email address is pak@math IP: I think peer review is absolutely important and vital,
.ucla.edu. but the reasons why are completely different from those
Anne Schilling is professor of mathematics at UC Davis. Her email address is stated by previous speakers. There is a way to look at peer
[email protected]. review as a kind of cultural phenomenon. It’s not that old,
Mark C. Wilson was the recent editor-in-chief of the Notices of the AMS. His and it’s a culture developed by the mathematical commu-
email address is [email protected]. nity. That’s what allowed mathematics to grow and be-
Communicated by Notices Associate Editor Vidit Nanda. come as big and as deep as it is. One can look at areas
For permission to reprint this article, please contact: adjacent to pure mathematics, and you’ll see that the pa-
[email protected]. pers are shorter. There is a fast turnaround which doesn’t
DOI: https://doi.org/10.1090/noti3194

AUGUST 2025 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY 757

 CONVERSATION

allow in-depth thinking about problems. This is a byprod- sor “how do you judge a candidate if you don’t have these
uct of lacking serious rigorous peer review. It’s not that peer reviews?” He said “well, we actually read the paper.”
they don’t have peer review because the papers are short You look at recommendations from other people. Maybe
so you can just read them quickly. So in a sense, you don’t we put too much emphasis on the review process. Often
need peer review. One can change university structures, it depends a lot, maybe too much, on who the reviewer is,
after all, some of us become vice chancellors and so forth. whether they had a good day. There is a lot of randomness
The problem is if we want to preserve mathematics the way in the process. So I think maybe focusing more on other
it is with its depth and diversity, we need to keep rigorous measures of quality, but not on the peer review, might be
peer review. If we’re okay with arXiv being the standard- important.
bearer, then it’s fine. Then we don’t need peer review at all. IP: I think that you might be right—in my area peer re-
So it depends on how much of a small-c conservative you view is pretty terrible. And it’s terrible because it’s not a
are. community type of relationship. It’s an adversarial rela-
LF: My first question is what is the purpose not only of tionship. There are editors who basically want to stop the
peer review, but of journals, because that’s really what’s flood of papers. There are people who are submitting pa-
changed. Journals used to be a way to disseminate re- pers, and because of various career pressures, really need
search, but these days, mostly since the beginning of the those papers to get in. And somehow we put up a lot of ob-
internet, that’s no longer the case. It’s very rare now that stacles for them. For example, you cannot submit to two
you’re first hearing about a result, especially an important journals simultaneously. For some reason, referees take six
result, by seeing it in a journal. You’re probably getting it months—sometimes two years—to start to read the paper,
via email or social media or a magazine article. So then which is insane. This creates terrible incentives for every-
what is the purpose of journals? It is to give a stamp of body. What you can learn from, for example, physics, is
approval in some sense. It’s to say, this paper has been a completely different framework. The Phys. Review series
judged by the journal to be good for the journal, what- has a rather communal aspect to it: lots of papers, pub-
ever that means. At least it should be correct, it should be lished very quickly, they don’t claim to make a major re-
novel, there should be something interesting in the proof, sult, people judged not just on publications, just like eco-
the model, something like that. It’s hard to do that with- nomics.
out peer review. I don’t think peer review is as critical as it Now in mathematics we can have 500 applicants for
used to be, because having your paper out there can often one position. Even if you have a committee of ten peo-
create a conversation about the paper. ple and divide the work, we can’t read all the papers, and
[audience; David Saltman]: Peer review tends to im- we don’t have the expertise. I don’t think that approach
prove the exposition. You get your paper refereed quicker of reading the papers would work, but you can try to es-
if it is well written, and sometimes it’s not published at all sentially remove job pressures by having journals accept
if it’s unreadable. Back when they were printed on papyrus, way more papers, maybe ten times as many as now, and
I was the editor and managing editor of Transactions of the speed up the process. We could remove these incentives
AMS. In my experience, you’re improving the exposition to become the best journal ever by basically getting rid of
by requiring a referee. 99% of submissions. That’s not how you actually become
MW: What, if anything, can mathematics journals learn friends. It is a major change in publishing, and I’m saying
about peer review from journals in other fields? that with almost certain knowledge that this is not going
LF: I’ve talked to people in lots of different fields, and to happen.
nobody thinks their field does peer review well! I’m not RH: I am not a mathematician—my PhD is in biochem-
a fan of how we deal with it in CS, via conferences, espe- istry. I’ve been in publishing for 30 years, in different
cially theoretical computer science, where we have to re- fields, for the last 11 years at the AMS. So I’ve seen a lot
view lots of papers by program committee very quickly, of different things. Even though I hear what Igor says, I
and I don’t think we get full reviews very often at all. One am not in full agreement. What you’re talking about in
field that’s interested me is theoretical economics, where terms of sheer volume and the way that acceptance and
I have worked a little. There, when someone goes up to rejection is treated, is something that’s true across many
be an assistant professor, usually they don’t have any pub- different disciplines. It’s more to do with the culture of
lished papers yet, because the publication process is very that particular journal or perhaps a society’s approach. I
slow. So peer reviews are not used to determine who to do think there are some significant differences in fields. I
hire as a new professor in an economics department. They have learnt, over the last 11 years or so, that in mathemat-
come in with what’s called a job market paper, a single ics, in some ways we’re more like the humanities than we
paper that they talk about. I asked an economics profes- are another scientific discipline. We’re not reporting on a

758 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY VOLUME 72, NUMBER 7

 CONVERSATION

result, and certainly in pure mathematics we’re not report- not saying I’m advocating it. I’m just saying that might be
ing on an experiment. The intellectual property is in the a way of looking at a new model in peer review.
article that you write, the way you express it, the elegance of AS: One thing I want to reiterate is that referees don’t
a proof. And so it’s a very different type of article than you really get any benefit from refereeing, because they are
would have if you were reporting on an experiment. So anonymous. I think that’s one of the big problems. Just
that’s something to bear in mind. Many of the articles that trying to get referees to actually do the work can often be
appear in some of the AMS journals can run to 70 pages a problem. We talked about more open referee processes.
or more for a proof. A reviewer has to review a mathemat- There’s of course, also the opposite to that, a double-blind
ics paper proof by rigorously going through it. If we look referee process, which we are actually currently trying out
at the biomedical sciences, you really are reporting on an in Combinatorial Theory, the new diamond open access
experiment, there’s a much faster rate of publishing, partly journal. The authors can’t see who the referees are, but the
because there’s a lot of competition between researchers to referees are also not supposed to see who the authors are.
get that result out quickly. Of course the referees can often find out who the authors
There are models that exist within biomedical publish- are because the papers are actually on arXiv.
ing, for example, of peer review, that are emerging faster [audience; name unknown]: I am chief editor of a jour-
than they are within mathematics. And that’s partly for nal that is for students, called Asian American Voices, and I
those pressures that I talk about. These are things we could had to collaborate with a lot of people from humanities.
learn from, I think. You see transparent peer review, which They have a completely different way of thinking, organiz-
refers to publishing the peer review alongside the papers ing things. At the same time, I’m a managing editor of
themselves. Then you have open peer review. I don’t know a math education journal. I learned a lot about how to
if any of you followed the story around the journal eLife gather information, how to give feedback to students, so
funded by Howard Hughes Medical Institute, but they re- that was a very good experience.
cently changed to a model where they no longer use peer MW: There are a lot of different types of peer review. If
review to actually recommend publishing or rejecting a pa- you haven’t heard about it, you might want to check out
per. They use peer review in an open environment, so you peer-witnessed review. I just realized that there’s a missing
know who the peer reviewers are, the author and the re- type of review, which is where the author knows who the
viewers are transparent there to each other. At most, what reviewers are, but not vice versa. That’s a joke—I don’t
they might do is suggest that this paper is worth looking at think we’re going to go for that.
more. Those models are very controversial for a number MW: What scares you about the future of journal peer
of reasons. review?
Another emerging model is peer review of preprints AS: Well, what scares me is that we put a lot of effort
themselves. This is happening in the biomedical sciences into the peer review process. Editors work hard, referees
first, and certainly I know Howard Hughes and Gates are work hard, and then the journals come and they put the
thinking about this kind of thing. But there the idea is articles behind a paywall and you can’t access the papers
that you would have peer review on the preprints and these anymore. We as mathematicians do a lot of work and at
would be used to validate the papers at an early stage be- the end of the day, we don’t actually get the benefits. So
fore they actually go to a journal. It requires that every- that’s why I’m very much in favor of diamond open access.
thing to be open and transparent. I think there are some Also, just the amount of papers that are out there is scary.
real problems there, because in the world of peer review, A lot of people do math, that’s great, but then if the papers
you have to find the time to do those peer reviews. If you’re have to be refereed, it takes a lot of time. Doing a good job
peer reviewing preprints, you know, where are you going is hard, so that scares me a little bit. I’m not sure whether
to find that extra time to do that? If you are an early career this scares me, but I’ve been, as was mentioned, involved
researcher, do you really want your name attached to a re- in making Combinatorial Theory, which is diamond open
view? Is that going to cause some fear about your career access. In the process, there are a lot of things that have
progress, if you think a more senior person may judge you to happen. For example, when you start out, the journal
for your opinions? What is the reward in terms of your ca- doesn’t have any reputation. And in some countries peo-
reer progress and what you offer to the field in doing peer ple are judged by in which tier of journals you have pub-
review? lished. There are lists of journals and to get tenure, you
I potentially see a place where we could have peer re- have to have published in journals from a particular list.
view of, say, arXiv preprints that might be used as a way to If you start a new journal, you have to get it accredited by
submit your article with those early reviews to a journal, Scopus or Web of Science. And that is a very long process.
and that might then be used in rigorous peer review. I’m Change is always very slow.

AUGUST 2025 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY 759

 CONVERSATION

RH: I was wondering if I might ask you a question, too, is no way this person read the paper. So what am I sup-
because we at AMS have a diamond open access journal posed to do with this report? Do you ask for somebody
where incidentally we’re using double anonymous peer re- else and wait for another six months, or just accept it as
view. Are you scared about how sustainable your journal is? Those types of pressure on editors become somewhat
is going to be in terms of funding? unbearable and the quality of the peer review goes down.
AS: Actually we were lucky. We work with Lyrasis, which With that the quality of all mathematics essentially is go-
is a nonprofit organization that helps find funding for di- ing down because from time to time, papers with major
amond open access journals. They actually go to libraries mistakes start to pop up. We don’t know who the referee
and say, can you give us a small amount of money to sus- was, who screwed up, there is no way to find out. The ed-
tain these journals? So I’m currently not scared about that. itor might lose a job, but the editor didn’t read the paper.
RH: I’m really worried about the erosion of research in- They just chose the referee. So what scares me is that even-
tegrity. By that I mean that if peer review is so difficult tually we’re going to converge to the point when peer re-
to do and people don’t have time to do it, where is the view is a joke. Nobody’s going to take it seriously. People
incentive for early career researchers to embrace peer re- will judge “that paper’s published in such and such jour-
view as part of their culture as they come through the re- nal, that’s basically a superficial opinion about the paper.
search ecosystem? I’m scared that the notion of peer review That’s not serious peer review.”
ebbs away and that leads to an erosion of research integrity. LF: It’s hard to answer this question without jumping
That’s only being exacerbated by what’s happening in the into the next one. I’m old enough to remember when
AI world. LATEX first came out. When it first came out, all of a sudden
IP: I basically have the same answer. I spent eight years all our papers looked amazing. I mean, it looked like they
at Discrete Mathematics, which may have been too long. were already published in a journal, no matter how much
You could see over eight years the erosion happening, from garbage there was. It made it a lot easier and a lot faster to
all directions. It is absolutely terrific that mathematics is write papers, which I think led to more papers. And I feel
now internationalized and there are communities from all we’re about to hit that situation again, where AI is going to
over the world of various kinds that submit to the journal. make it a lot easier to write papers and not necessarily of
However, there are various cultural differences and vari- high quality. So we’re going to see an even larger growth
ous incentives that different countries place on researchers in papers. In fact, I’ve already gotten several AI-generated
both in teaching and publishing. For example, various cri- P versus NP proofs in the last few months. It’s pretty clear
teria for what level of journals they have to publish. You garbage now, but at some point they’re going to start look-
notice that for many, many people that you ask to be re- ing good. I feel that there’s going to be a huge explosion a
viewers, those who just published an article in your jour- number of papers, and we can’t even handle the ones we
nal, you ask them to referee a paper and they say yes. They have now.
say it very quickly because they want this person to publish [audience; Doug Allen]: How did you deal with the
that paper. It doesn’t matter whether it is a good paper or step-up in volume of papers when LATEX came in?
bad paper. And that’s, of course, a completely useless ref- LF: It came at the same time as a huge growth in the
eree report. So you have to start all over. Because of the field. We multiplied conferences like crazy. That caused a
way incentives are aligned it’s getting harder to find people lot of other issues, because now no one ever sees anyone
who actually want to take a lot of time and do this. Once I any more.
had to referee a 114-page paper, and that took two weeks of [audience; name unknown]: The quality of reviews is
my life, which I’m never going to get back. I would never getting lower, but also there’s outright fraud because it’s
agree to do it right now because I have other things I want such an important thing to get published, so that scares
to do. This was a long time ago when I didn’t have family me.
or kids. I could see it over and over when I was an edi- IP: I didn’t have to deal with outright fraud. I don’t
tor in Transactions of the AMS. I think I lost a lot of friends. think the mathematics has as much at stake, so there’s no
Keep asking people who are responsible, and eventually reason to be fraudulent. However, I remember I had to
they say, okay, maybe you should stop emailing me. As deal with an interesting case of self-plagiarism, where a
this progresses, reviews are getting shallower. There is no person basically took a published paper, because that was
way for me as an editor to directly check if they read the his LATEX file, just removed a large part of it and put it in an-
whole paper and understood it. You can see it from the other paper and put some other stuff, different title, some
report that there are some commas need to be changed other application, but 50% of the paper was a previous pa-
and there was some definition that it has to be fixed, but per. I didn’t figure it out—technology probably can help
that’s it. I’ve seen it on a really long technical paper. There these days. But the referee looked up the previous paper of

760 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY VOLUME 72, NUMBER 7

 CONVERSATION

the author’s and found it. I tried to ban the author for five this many articles for these journals, the editors judged
years from submitting, but the editor-in-chief was against them to be high quality.” Of course, in the end, you need
that. There is no control over that. I’m one hundred per- promotion and hiring communities to value them. But at
cent sure people do this all the time, maybe not in the top the moment right now, that data is not really there. There’s
journals, but in lower-tier journals. no way to give evidence that I write good reviews versus
There is a curious case1 you can look at on Wikipedia. bad reviews.
He published about 3-400 papers by basically taking pa- MW: Or even do them at all.
pers by other people and LATEXing them, changing the title IP: Some physics journals have an award, given to some
a little, and submitting to very obscure journals. They got number of people once a year. They have a very rigorous
published because those are good papers and the results process for choosing those people. Various computer sci-
are relatively obscure, and it’s very hard to search for a par- ence conferences just publish a list of people who reviewed.
ticular theorem. There are so many papers and so many subreviewers that
AS: I think arXiv actually checks whether there’s overlap. you can’t really tell who wrote for what. Various journals
It can be detected. Some of the editorial tools that we use also once a year publish the referees. I’m sort of against
also can say there’s overlap with some other paper. that because it loses some anonymity.
IP: Wouldn’t the double-blind thing defeat this, if you I want to mention about paper mills. There is a different
cannot actually know who the author is? How are you kind of fraud that’s also been perpetuated, but it’s a very
going to look up the previous paper of that person? low level fraud. So if you actually have a paper that’s going
MW: I think the point is that it just checks for overlap to be appearing in a fancy journal, there is a way to sell a
with any paper irrespective of author. coauthorship. There are people in other countries, who, es-
AS: And this is not a person—software is actually check- pecially if they have children who are about to finish high
ing. school, are willing to pay a lot of money to get this child
RH: The real issues around fraud haven’t hit math yet to appear on a fancy paper. Because that increases chance
in a big way. Certainly in the biomedical sciences, you see substantially of acceptance to a fancy university and we’re
an immense amount of fraud, especially around images. talking about a lot of money. So we’re all missing out, at
There seems to be no end of people who are willing to sort least I am, on a steady stream of income! This level of
of perpetrate that kind of fraud. And the thing we are be- fraud will only be increasing. That’s my prediction.
ginning to see really is on the paper mill front. There are MW: What challenges and opportunities for peer review
companies out there who essentially ask for payment for do you see coming from AI?
you to publish your paper and make up a journal, make LF: I’d like to talk about some of the positives. I think
up a series of papers that might be in this fake journal and that when we’re not there yet, but I really believe we’re go-
and then flood publishers with those papers. It’s remark- ing to get to a point where AI is going to be able to take
able how much of a business there is actually to be had a well-written proof and verify it as well as a good author
there. That stems a little bit from the culture of paying could. You don’t have to come up with novel ideas, you
article processing charges for open access. That’s a whole just have to go through the proof. I think AI will do a good
separate discussion. job, and apparently it has been done already, in checking
MW: I guess one of the things protecting mathematics novelty, whether a result has appeared in the literature be-
to some extent is, as Igor said, the small potatoes nature of fore. So that means you could really now separate the ob-
it. The other one, I guess, is the methodology of the field. jective pieces—is the proof correct, is the proof novel—-
It’s relatively easy to make up a fictitious experiment and from the subjective pieces. You know, how important is
have a whole career out of it. In fact, some people seem to the theore, the proof, the model. So, and I think that could
have done that. But it’s hard to do that in math, making up really help a lot because now you don’t even let the pa-
a completely fictitious theorem with a wrong proof. It gets per get submitted to a journal until they’ve shown that, it’s
found quite quickly. Still, it’s something to be concerned been fully verified in some sort of process. You only have
about. to worry about whether it is reasonably well written and
[audience]: It seems like there are things that journals are the results important? So I think that will help a lot.
and publishers could do in terms of having an editor judge It’s not going to eliminate all these problems, but I think
the quality of reviews and publicly attaching that on the you’re not going to get a 150 page paper and say “my God,
journal’s homepage, for example, and then that could be I have to check this proof.”
something that reviewers would put on their CV. “I wrote AI is going to cause other challenges. You’re going to see
a lot of people trying to abuse it—try to make papers look
1https://en.wikipedia.org/wiki/D%C4%83nu%C8%9B_Marcu really, really good, that just aren’t correct at all. And then

AUGUST 2025 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY 761

 CONVERSATION

at some point, an AI is going to make a real breakthrough, have guidelines on our website for our journals. We follow
find a step in a proof that’s really significant. How does a what’s called COPE (the Committee on Publishing Ethics)
journal handle that scenario? You can’t really give an AI approach to those guidelines. So you can’t have a large
coauthorship, so is it the person who just came up with language model author a paper. But if you’re an author
the prompt the right one to get the credit? I don’t know— whose first language is not English, for example, there’s
that’s something we have to think about as well. no reason at all why you shouldn’t use those tools to help
IP: I have a little less rosy view. Every time we have clean up your submission. In fact, that happens now, ex-
a we have a new technology that brings advantages. For cept it’s a more manual approach. And as long as you’re
example, I remember when I first time had to submit to saying that you are doing that, I don’t think there’s any is-
STOC, I had to submit 15 copies of my paper. That was sue there and that’s a really beneficial idea that leads to
ridiculous—you had to put them all in the mail and send more equity amongst those who are submitting, where it’s
them somewhere and who knows what they’re doing with really the math that matters, not where you’re from, your
that. They are still probably lying somewhere in the base- native language and so on. We have also seen an example
ment. But often enough when you have new technology, of someone who decided “well, we need to find a reviewer.
it removes some timewasting, but it also makes you do I’m going to upload this submission I have to a large lan-
stuff. I’m a little worried that in the future some journals guage model to help me find a reviewer.” Well, that’s not
will require me to provide some kind of Lean code so that okay, because as soon as you put a paper on a large lan-
it can verify my result. I don’t want to do that. And the guage model, you are risking both privacy, confidentiality,
journals will make my life basically miserable by making and potentially copyright as well if it becomes published.
sure it will pass through some AI checks. On the positive So we put a guideline up there to say, you really can’t do
side, it might be a good idea for journals to develop their that, but I think people don’t understand necessarily what
own basic AI and train it on all the papers that they ac- is appropriate.
cepted and rejected. And not make it as a requirement of Also, AI is not just, of course, generative AI. There are
any kind. Just make it public, because there are a lot of tools that we can use to automate the backend systems
cranks who submit papers, so if they submit their paper more effectively. Most of those are language-based and
to this AI and it gives a really low score, maybe they know that’s an opportunity where we probably will at the AMS
that they shouldn’t submit to the journal. go. Within our production processes and our peer review
LF: You may get in an AI arms race. system processes, we work with EditFlow. I suspect that’s
IP: Eventually some people will train an AI to improve where AI will be, in the short term, mostly used.
their score and so forth, but at least it will give some idea But these guidelines are really important. I really worry
to some people if what they’re doing makes any sense at about privacy, copyright, and the research integrity parts.
all. That’s my only very weak positive aspects of AI. MW: G. H. Hardy said that when you’re a reviewer of a
AS: I went to a conference where somebody showed paper, you should ask yourself three questions. Is it true?
software such as ProverX that can actually check proofs. It Is it new? Is it interesting? And from what Lance has said,
still seems very clunky. Who should then be responsible two of those can soon be delegated. I’ve thought for 30
for making sure that the proof that the author provides can years, why do we not delegate some of the correctness part
actually be checked by computer? As Igor said, is that the to grad students? They’ve got to start their reviewing ca-
author that now has to provide the code or is it the journal reer somewhere. Maybe the novelty and importance needs
that takes the proof and puts it into this format? to be done by an expert. But the line-by-line correctness
LF: We’re not there yet, but I guess we’ll get to the point is something that grad students and postdocs may have
where a well-written proof should be able to be checked more time to do. In computer science, they do this be-
by AI. If it’s not a well-written proof, then it’s the responsi- cause they have subreviewers. I just wonder why mathe-
bility of the author to write it well. maticians haven’t been doing that more. The other thing
RH: I think that we’re not used to the rate of change we I wanted to mention was about when Lance said you can’t
are going to see in the evolution of these tools, so I don’t have AI as a coauthor. Doron Zeilberger long ago was list-
think we’re far away from actually being in that proof cor- ing his computer as a coauthor on his papers.
rection situation. That is the one big positive that I think [audience; name unknown]: What is the average aca-
may come out. AI appears to offer tools for situations demic level of your peer reviewers for journals? Are they
where you may want to see greater efficiency around large mostly well-established professors, or are they mostly post-
quantities of information. So that, as a general principle, docs?
seems to apply to peer review. The issue, though, right now AS: When I send out a paper, I often have problems
is that we are also seeing abuses. So for example, we now finding somebody who would say yes. When I send out

762 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY VOLUME 72, NUMBER 7

 CONVERSATION

quick opinion requests, usually people say yes. Those can people to use the sense that they’ve been a reviewer in their
be established researchers because they can say whether hunt for tenure or in their career in some form, to help
they think it’s interesting or not. For actually doing all the them progress. That has been tried in different ways and
checking, I often use postdocs or grad students. there’s no perfect way of doing it yet.
RH: One of the things I’ve noticed about mathematics is On the doubly anonymous piece of it, we do the same
that you have so many small fields, and you have to pretty as you do in the sense that it’s a very light touch. We’re not
much be an expert to do a review. I think most of our forcing people to do certain things because we fully know
reviewers are experts, which tends to make them more se- that you can go on arXiv if you wish and find out how
nior. an author may be. What it does do is create a pause and
IP: When I was an editor, I never used students. I tried as a reviewer, or at least have to consider your potential
to use postdocs as little as possible, not because I don’t and implicit biases. We feel that’s worth it, and it certainly
trust them—I don’t trust grad students. However, post- hasn’t caused any harm in terms of reviewers saying no to
docs are under enough pressure themselves, and they need review, or in the way that reviews are done, or in the speed
to write papers. They don’t have time to commit to actu- of publication. By the criterion of “do no harm,” double
ally read papers. And also, they’re somewhat vulnerable anonymous is working. Further down the line we would
in a sense that there is a relationship between referee and like to build evidence demographically that more authors
editor, and they’re sometimes afraid to say no. You want from different parts of the world are able to publish in our
them to avoid having to make this choice. So I think the journals, so it’s all about the high-quality research rather
optimum would be early career tenure-track faculty who than who they are.
have a little more room to grow and are junior enough to AS: I want to comment one more time on using grad
actually agree to referee the paper. But there is such a stark students or postdocs as referees. To respond to Igor, the
difference in levels, and there is so much pressure on grad- way that I’ve done it is like this. I sent a quick review re-
uate students and postdocs, I think it is inhumane to use quest to a senior mathematician and they said “oh, this
them these days. is a great paper and this particular postdoc is actually cur-
MW: How can peer review be improved? rently reading it.” So then asking this postdoc, I feel like
RH: It would be great to involve early career researchers it’s a good use of their time because they are reading the
more in the review process, but for the reasons we’ve paper anyway, and then they can actually make comments
brought up, it’s difficult. I know that, for example, the and have actually worked through all the details. You said
American Psychological Association has what’s called a we shouldn’t abuse postdocs—I agree with that. But in this
mentoring or buddy approach. That’s something I’m look- case, I feel they will get the credit of having done the review
ing at. We use double anonymous peer review now—it’s and they have already sort of done the work.
a policy of the AMS and we’re rolling it out across all our IP: We talked a lot about small and serious issues. My
journals. Even within a doubly anonymous setting, you big issue is this adversarial relationship and the slow-
could have a senior reviewer and an early career reviewer walking of everything. I’m going to say something revo-
who is perhaps a member of our society—we haven’t fig- lutionary: I personally think the system where the author
ured it all out yet—a mentoring opportunity. That would submits the paper to just one journal and there is an ethi-
instill the idea of peer review at an earlier stage, it would cal obligation not to submit to other journals is not ideal.
train you how to do peer review, so I love the idea of it, ex- I understand why historically that happened. You don’t
cept that you are up against the problem again that it takes want different journals to ask different people who would
time to mentor. How much time do I as a senior researcher be both spending, or maybe wasting, time by doing the
have to actually not only do the review but also work with same. But that’s not what actually happens in real life. In
a junior person to mentor them? I’m not saying it’s insur- real life, the referee sits on the paper for six months or a
mountable, but that is probably something we would need year and a half, and then when there is some kind of holi-
to consider. day writes the review in a week. That’s my experience. Ev-
I think it’s not so much about providing material incen- ery time as an editor, you ask the referee “Are you done? Is
tives. In fact, I would argue that trying to pay someone to there any news?” And they say, “I need another week, or
do a review is actually a negative thing. By the time you’ve a month” and basically they’re looking at the calendar to
actually had to then put in your employer identification find when is the next holiday. Ideally what we should do is
number and pay taxes on a $500 fee or something, you that authors should be able to submit to multiple journals
might as well forget it. But the idea of providing those in- and the journals will look for some referees and they’re not
centives, even within a double anonymous environment, going to do any work anyway. But it’s really easy to set up a
I think is important. What you really want to do is allow system basically built on top of arXiv when the journal will

AUGUST 2025 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY 763

 CONVERSATION

just basically mark “I have somebody who’s looking at the a way to share referee reports, then it might be faster for
paper.” That would create an incentive to the journal to papers to actually get published. If the result is correct,
do a faster job. As soon as somebody actually produces a but doesn’t quite make the cut for a particular journal, it
referee report, there can be different colors on top of that. could go to the next journal.
There is not going to be any redundancy when it’s done. MW: I’m pretty sure I’ve read somewhere the idea that
It’s not like there are people who are spending six months you just submit to arXiv and let journals bid on it. You’re
reading a paper. And once the journals do get into this not submitting to anything. People look at it, someone
kind of friendly competition, there will be just a few jour- says “oh, that looks interesting. I’m interested enough
nals looking at the paper. Hopefully the process will speed based on the abstract and a couple of lines of it to give a
up because what happens right now is bad. I’m one of the referee report for this journal that I like.” Now, of course,
very few people who (probably) has done this. I had pa- the problem would be that a lot of them are just going to
pers which are sitting in a journal for over a year, and then I sit there, whereas anything that Terry Tao put out, even if
would tell the journal, “OK, I’m just going to withdraw the half asleep, would still get looked at.
paper. You’re done.” And they say the referee is promising [audience; name unknown]: I had an experience
they’re going to start reading the paper soon, but I thought where I refereed a paper and then suggested rejection. A
“No, you’re doing a bad job and you should be punished. month later, I got the same paper back from a different
This is not how it’s supposed to be.” Most people don’t do journal, and the authors had made none of the changes I
that and the journals feel a lot of power over the authors, suggested. Sharing reports in some way between journals
and they shouldn’t. So this would at least level the field a would be helpful, to also shape authors’ behavior.
little bit. But I understand it’s controversial. [audience; name unknown]: I think we need to think
RH: I have one word with two letters. No. a bit more from the author side, particularly in a world
[audience; name unknown]: I want to say, I don’t where we have a whole regions that are clearly underrepre-
agree with it. If everybody is sending the same paper to sev- sented. There are talented people, young researchers that
eral journals, we are going to be as editors in a real night- are starting to get a hold in the world. Of course, there’s
mare to find a good referee for it. Maybe the thing that we longstanding disparity and races and genders even in the
have to change is the way of evaluation and promotion of US. Editors perhaps need to establish a code of what are
researchers based on how many papers they publish, and proper characteristics of a review and what is improper
this is happening worldwide. (such as ridicule or hyperbole). Every review should at
IP: I’m sure we all agree with you. However, that is a least start out with some positive characteristics, and not
system that cannot be changed. People have to be eval- just immediately launch into invective, because you may
uated somehow. There is a departmental committee on be permanently turning off a promising young researcher.
promotion in every university and they have people who Anonymous peer review, when it’s just like anonymous
can read maybe one paper. But they don’t have enough chat on the Internet, can very easily get out of control.
expertise everywhere. From there, it goes up to adminis- MW: I understand that well. One PhD student’s first
tration. I am a bureaucrat now. I am administrative vice paper was met with a totally unfair criticism and I really
chair, whatever that means. It goes up to the dean, who let the editor have it, because the editor should never have
doesn’t know anything about mathematics at all, and then even allowed that review to come back. Part of it is edito-
it goes to the committee of appointments, which doesn’t rial judgment, and another is that you should never have
know anything about anything. The way university admin- unstructured feedback in any format. Having a checklist
istration is structured, that’s impossible to change. And I and structured feedback is always the way to go. I wonder
agree with you. However, the way refereeing is structured whether COPE (Committee on Publishing Ethics) has any-
right now is basically making giant obstacles for young thing to say about this? [in fact their guidelines say “Reviews
people to get promoted, because they’re waiting for a year should be objective and constructive, ensuring feedback is clear
or two until their paper gets refereed. Something needs to and helpful to authors.” —Ed.]
be done to help junior people. LF: I’ve noticed that a big discussion point is how hard
AS: I want to say one thing about how we could possibly it is to find referees. I think there’s a lot of people out there
speed up the refereeing process. For example, in Forum of who might want to be referees who aren’t as well known.
Math, we have Pi and also Sigma, where Pi is supposed to Right now it’s editors finding referees. It’s not referees say-
be the really stellar journal and then Sigma is also really ing “I’m not that busy right now.”
good, but slightly below. If a paper gets submitted to Pi MW: There may be a lot of underemployed referees, like
and it doesn’t quite make the cut, then you can actually scholarly people at teaching institutions.
use the referee reports from Pi for Sigma. So if there was

764 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY VOLUME 72, NUMBER 7

 LF: Right. People you might not think of, that would
maybe not do as good a job as a super-expert, but the super-
expert’s not going to do it anyway. Finding a way to have
those people be able to come out of the woodwork and
make themselves available, might be interesting.
RH: I not sure this is completely related. At the AMS,
I’m responsible for the financial aspects of Math Reviews
and MathSciNet. We have some 19,000 reviewers. That’s
often more junior people. There is a stable there of people
who want to engage with us that are often more junior. I
don’t know how to translate that into journal peer review.

Lance Fortnow Robert Harington

Igor Pak Anne Schilling

Mark C. Wilson
Credits
Photo of Lance Fortnow is courtesy of Scott Pfeiffer.
Photo of Robert Harington is courtesy of the AMS.
Photo of Igor Pak is courtesy of Olga Petrova.
Photo of Anne Schilling is courtesy of Anne Schilling.
Photo of Mark C. Wilson is courtesy of Mark C. Wilson.

AUGUST 2025 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY 765

 CONVERSATION

Guidelines for
Writing and Reviewing
Mathematical Software
Jeroen Hanselman

1. Why Proper Coding is Important come without a proof. This is not a problem that only oc-
for Mathematics curs in mathematics, however. The replication crisis is a
The advent of computers has allowed mathematicians to problem affecting many areas of science, as discussed, for
do increasingly more difficult computations that used to example, in [Bak16]. In computer science, people have
be practically impossible. We now have databases of math- started to do reviews for research artifacts that accompany
ematical objects, measuring multiple terabytes in size, as a publication. It is discussed in articles such as [LHH+ 24]
well as huge computer algebra packages, proof assistants, and [Eit22], and put into practice at a conference like
and numerous other computational tools that aid us in CHES, https://ches.iacr.org/2024/artifacts.php,
doing mathematics. where authors could have their code tested on Availability,
Despite the usefulness of computation in mathemat- (Re)usability and Reproducibility.
ics, it is not always used in a scientifically responsible As the mathematical objects we are dealing with be-
way. Mathematicians have spent decades refining the way come more complex, we increasingly rely on computa-
one should write down mathematics. Everything must be tional software, and this trend will most likely continue.
proven and written down clearly so that others can under- It is therefore crucial that we start thinking of guidelines
stand the logic that was used and agree that the results and general practices that improve the reproducibility of
were correct. Peer reviewers will try their best to see if the our code. Otherwise, we might run into serious problems
arguments and results look plausible and correct. When in the future.
a paper relies on a computational part, however, the in- Although we might prefer to have higher-quality code
cluded code does not always get treated with the same care. and a better peer-reviewing process for software, there are
On the one hand, reviewers might not always know how some good reasons why it has not happened yet. When do-
to properly review source code and, on the other hand, ing research, software is mostly used as a tool. Although
authors might not always know how to write good code. we are trained to transform the ideas we scribbled down
Because of this, the code might be badly structured, dif- somewhere into a rigorously written proof, we are not al-
ficult to install and run, and sometimes even impossible ways taught to turn the code we wrote for ourselves into
to find. As a consequence, there are published papers software that can be used and reused by the wider scien-
floating around that contain proofs relying on code that tific community. A peer reviewer also does not always ex-
no one can run. This essentially means that these results amine the code accompanying a paper. This could be due
to a lack of time, lack of familiarity with the software used,
Jeroen Hanselman is a postdoc in computational arithmetic geometry at the or a variety of other reasons. The code may also be in bad
RPTU Kaiserslautern-Landau working for the Mathematical Research Data Ini- shape, making it a hassle for a reviewer to review.
tiative. His email address is [email protected]. Furthermore, when applying for jobs, hiring commit-
For permission to reprint this article, please contact: tees do not always value the code in your repository as
[email protected]. much as the number of publications you have. As this
DOI: https://doi.org/10.1090/noti3210 problem was recognized across many disciplines, it led to

766 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY VOLUME 72, NUMBER 7

 CONVERSATION

the creation of journals like JOSS (Journal of Open Source amount of source code available compared to previous it-
Software) https://joss.theoj.org/, which enables re- erations of the ANTS conference. Authors were also happy
searchers to publish their software as a separate citable about receiving feedback on the code that they wrote. A
publication. software reviewing process will most likely also be in-
Finally, writing good, reproducible software is hard and cluded in the next iterations of ANTS and LuCaNT.
time-consuming. It is not just about writing legible code.
One needs to think about long-term storage solutions, in- 2. Some Statistics About the Impact
stallation instructions for different operating systems, ex- of Software Reviewing
amples, etc. The process is a lot more involved than just To give an impression of the impact of software reviewing,
writing down a proof. I have compiled some statistics based on the reviews my
In summary, one sees that writing good reproducible MaRDI colleagues and I conducted for the ANTS and Lu-
software is an important task, but we, as a commu- CaNT conferences. Note that the conferences explicitly an-
nity, still need to work to integrate it into the scien- nounced that software peer-reviewing would be part of the
tific publication process. There are various initiatives process, and authors were given the opportunity to submit
concerned with this topic. For example, the European their code if they had not done so initially.
Open Science Cloud (https://eosc.eu/) has a Task Henceforth, we will distinguish between papers and
Force on “Infrastructure for Quality Research Software”; pieces of software. We will clarify the reason for this now.
the United States Research Software Engineer Associa- To begin with, it is possible for a single paper to come
tion (https://us-rse.org/) has a working group fo- with multiple pieces of software. One example of this is
cused on Code Review; the Helmholtz Centre (https:// when a paper is accompanied by different implementa-
helmholtz.software/) promotes reuse and citation of tions of the same algorithms, written in different program-
research software; the EU’s TIER2 project aims to improve ming languages. Secondly, it was not always possible to
reproducibility in science (https://tier2-project.eu) get access to the code despite the paper clearly requiring an
and https://fairsharing.org is a resource on data and implementation to illustrate or prove the results discussed
metadata standards. The German government also recog- within. Either no link was given or the link that was given
nized the need to adapt to the increasing reliance on com- was broken. If the paper was more theoretical and the al-
putation and established the NFDI (https://www.nfdi gorithms described did not really need an implementation
.de), an organization aimed at improving research data to show their usefulness, I did not count the paper as being
infrastructure across all sciences. I am part of the MaRDI accompanied by a piece of software.
project (Mathematical Research Data Initiative, https:// The papers were reviewed based on a certain list of cri-
www.mardi4nfdi.de), which is a subbranch of the NFDI teria we will discuss in more detail in the rest of the article.
specialized in mathematics. To read more about MaRDI’s For our statistical analysis we group these criteria together
goals, one can read the MaRDI white paper [Con23]. As in the following categories:
part of this project, I have been writing code reviews to try • Lic: The software came with a license.
and figure out the best way to improve the quality of soft- • Rdme: The repository had a Readme or a landing page
ware written in mathematics. explaining the contents.
The following guidelines are based on my experiences • Repo: The code was hosted on a repository for long-
with reviewing code, and discussions with various authors term storage.
and reviewers about best practices. I would like to thank • Inst(E): The code was easy to install (within 15 min-
John Cremona, Claus Fieker, Christiane Görgen, Lars Kast- utes).
ner, Jennifer Paulhus, and Drew Sutherland for their help- • Inst(H): It was possible to install the code after
ful feedback. I am writing down what I have learned in putting in more effort.
the hopes of improving the community standards for han- • Specs: The specifications necessary for allowing re-
dling code in mathematics. At this point, I have written ap- producibility were included in either the paper or the
proximately 100 software reviews for various conferences, repository. (Specifications included: the version of
most notably, for ANTS XVI (Algorithmic Number The- software used, the CPU type, and the required amount
ory Symposium, https://antsmath.org/) and LuCaNT of memory to run the code.) Including specifica-
(LMFDB, Computation, and Number Theory, https:// tions was deemed essential for reproducibility when
lucant.org/). The introduction of a software reviewing the algorithms were benchmarked and compared with
process was done as an experiment, but reactions have other algorithms.
been positive. The inclusion of a software reviewing pro-
cess seems to have led to a significant increase in the

AUGUST 2025 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY 767

 CONVERSATION

• Rdble: The code in the software had proper inden- A possible reason for the improvement could be that
tation, well-chosen variable names, sufficient annota- having well-written software correlates with having a
tion, etc. better-written paper. Accepted papers tend to have higher-
• Ran: The code ran without any issues. (Regardless of quality code, which leaves a better impression on peer
whether the output actually matched with the claims reviewers and editors. Another reason is that many au-
made in the paper.) thors took the feedback they received to heart and made
• Correct: The reviewer judged the results or data that improvements to their code and their repositories. This
were part of the publication were judged to be “verifi- ranged from small things, like adding a short Readme, to
able” or “likely correct.” bug-fixing and the addition of significant documentation
• Ran and Correct: If the code was judged to be Correct and/or examples. I saw positive changes for 19 out of the
it did not necessarily mean that the code ran flawlessly. 35 pieces of software that were accepted. Overall, I believe
Sometimes it was still possible to verify the results in the software reviewing process had a positive impact on
a different way. (When it was possible to circumvent the reproducibility of the code.
a bug, for example, or if there was separate code that
verified the results produced by the code that gave er- 3. Peer Reviewing Mathematics
rors.) On the flip side, just because the code ran, it and Research Software
does not necessarily mean that the output was correct. We will now discuss which goals a software peer-reviewing
This is why we have a separate category for code that process should aim to achieve. To start, we examine some
ran and also produced the correct results. aspects of traditional peer reviews.
Greiffenhagen [Gre24] has written an article discussing
Two rounds of reviewing were performed. In the first
the process and the role of peer reviewing in mathematics.
round, the pieces of software that came with papers that
We will shortly summarize some of his findings:
were submitted to the conferences were reviewed thor-
oughly. Of the 75 pieces of software mentioned in the sub- 1. Checking a mathematical proof for correctness is diffi-
mitted papers, 67 pieces of software had code available for cult and the peer-reviewing process cannot ensure that
review, and the reviewers were able to install 41 pieces of what is written down is 100% correct. What it can do
software. Only 30 pieces of software had no problems at is increase confidence in the correctness of the result.
all. These numbers were actually not that bad. If I were to 2. Peer reviewing a mathematical paper is generally more
increase the sample size and do software reviews for simi- difficult and time-intensive than it is in other disci-
lar papers released 1–2 years ago, the numbers would most plines. As a result, the average number of reviewers
likely look much worse. Additionally, the problems that that can look at a single paper is lower compared to
occur when trying to run the code can generally be fixed by those disciplines.
an author within a reasonably short time. This also high- 3. The primary responsibility for ensuring that the stated
lights the advantages of a software review process. It in- results are correct lies with the author, but the review-
centivizes authors to put more effort into their code, and ers share partial responsibility.
it points out flaws in the authors’ code that can be fixed 4. Reviewers generally do not read an article in detail.
before publication. They skim the article to get a rough understanding of
The second round of reviewing was done to study the it and to see whether they believe that the approach
effects the reviewing process had on the published articles. used could work, and then zoom in on the key argu-
All accepted papers were reviewed a second time to see ments or parts that they do not understand to verify
whether the authors took the criticism they got to heart their correctness.
and whether it was still possible to install and run the code. 5. Real mistakes are rarely found. It happens more often
Of the 39 pieces of software mentioned in the accepted that reviewers ask for clarification for parts they do not
papers, 36 pieces had code available for review, and the understand.
reviewers were able to install 35 pieces of software. The re- 6. It is a widely held belief that it is impossible to find
sults of this analysis can be found in Table 1 and Table 2. all mistakes, but, if there are any, the community will
Upon comparing the numbers, we find that the statis- eventually find them.
tics of the accepted papers are significantly better than the Most of these points (with the possible exception of
submitted ones. For 30 out of the 75 submitted pieces of (2)) will likely also apply to any kind of software peer-
software (40%), we were able to get the software to run reviewing process. It is impossible to ensure that the code
and to verify the results, whereas this was true for 27 out will be completely bug-free (1), the authors of the code
of the 39 pieces of software (69.2%) that were accepted for should bear the brunt of the responsibility to ensure that
publication. the code they published is correct (3), reviewers cannot be

768 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY VOLUME 72, NUMBER 7

 CONVERSATION

Total Lic Rdme Repo Inst(E) Inst(H) Specs Rdble

Pieces of Software (Subm) 67 29 48 64 52 59 34 49
Percentage 100% 43.3% 71.6% 95.5% 77.6% 88% 50.7% 73.1%
Pieces of Software (Publ) 36 23 33 35 31 35 24 32
Percentage 100% 63.8% 91.7% 97.2% 86.1% 97.2% 61.5% 82%
Table 1. Comparing reproducibility statistics of submitted papers and published papers. Percentages are taken with respect to
the total number of pieces of software for which code was available.

Total Ran Correct Ran and Correct

Pieces of Software (Subm) 59 41 35 30
Percentage 100% 69.5% 59.3% 50.8%
Pieces of Software (Publ) 35 32 29 27
Percentage 100% 91.4% 82.8% 77.1%
Table 2. Comparing reproducibility statistics of submitted papers and published papers. Percentages are taken with respect to
the total number of pieces of software that the reviewer was able to install.

expected to check every line of code (4), it will be challeng- study mathematics. To mirror the above list:
ing for anyone to find mistakes in the code (5), and ideally • We use algorithms to tabulate properties of math-
most bugs will be discovered over time as others reuse the ematical objects. (See e.g. the atlas of finite
code, reimplement the algorithms, and so on (6). groups [WCN85], the LMFDB [LMF24] or https://
In light of this, the key goals of any peer-reviewing pro- mathbases.org/.)
cess should be the following: • We write code snippets to compute examples that are
A) The reviewing process should increase confidence that too difficult to compute by hand.
the results are correct. • We perform computations to study conjectures and
B) The reviewing process should try to ensure that other formulate new hypotheses.
mathematicians can follow the reasoning and repro- • We develop algorithms to compute things we were not
duce the results by following the same steps as the au- able to compute before.
thors. • We write algorithms that improve on previous algo-
We point out a few differences between reviewing code rithms, making them more efficient.
and reviewing a mathematical paper. Consequently, to achieve goals A) and B), the peer-
First of all, the purpose of the code accompanying a pa- reviewing process will need to differ slightly from the clas-
per may not necessarily be to prove a theorem. It is cer- sical approach. To help a reviewer accomplish A) authors
tainly common to outsource a part of a proof to a script could, for example, add test files to the code that will help
written using a computer. But there are more types of data a reviewer check that there are no bugs in the code. Alterna-
and code that can come with a paper. A good way to think tively, the authors could add separate files that verify prop-
about it is to compare it to an experimental science. People erties of the computed data. To help achieve B), authors
do experiments to get data. These experiments and data must provide adequate installation instructions, hardware
are used as part of the publication in different ways. The specifications, and documentation, ensuring that other
goal could be: mathematicians can reproduce their computations.
• To construct some sort of database for future use. (For Secondly, not all mathematicians will be able to un-
example, the periodic table and all of the properties derstand the code. Even if they are familiar with pro-
of the elements.) gramming, they might not understand the specific pro-
• To illustrate a part of the theory. gramming languages used. Understanding a big repository
• To formulate or support new theories by using the filled with C code is far more difficult than executing a Sage
data gathered from the experiments. script in a Jupyter Notebook, for example. This makes it
• To explain the setup of a new kind of experiment. difficult to find good reviewers. When looking for a re-
• To explain how a certain kind of experiment can be viewer of a traditional math paper, an editor tries to find
improved to give faster/better results. an expert who understands the topic. If it is also required
that this expert understands the code and software used,
In computational mathematics we do very similar
the pool of potential reviewers will be extremely small.
things. Only in our case, the experiments we conduct are
performed using the algorithms that we write and use to

AUGUST 2025 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY 769

 CONVERSATION

A solution might be that an editor splits the peer-review The paper presents an algorithm that is claimed to
process into two tasks: be better than existing algorithms. Here, providing hard-
• Reviewing the mathematical content of the paper: ware specifications and run times becomes crucial. If you
This should be done by an expert in the particular claim your algorithm performs better than existing ones,
topic discussed by the paper. you will need to provide benchmarking tools to support
• Reviewing the software that accompanied the paper: those claims.
This should be done by someone who is familiar with The paper discusses a database of mathematical ob-
programming. This person should also be familiar jects. It is difficult to provide proper documentation for
with the field of research but does not need to be an a database of mathematical objects. Ideally, it should be
expert on the specific topic discussed in the paper. possible to find out which people used which algorithms
to produce each individual object in the database. One
The reviewer and the code reviewer could potentially be
would also like to know whether the data was computed
the same person, and, if not, they could communicate if
rigorously or heuristically, and whether it was conditional
need be. This split will make it easier for editors to find
on any conjectures or independent of them, as well as if
reviewers, as they do not necessarily need someone who is
the results have been double-checked using sanity checks
an expert in two things at the same time. As a consequence,
or other reliability tests. Other things to consider are: How
the code reviewer does not need to check the correctness
easy is it for users to access the database? Is there a user-
of the mathematics in the code. Their main focus should
friendly interface? Are there instructions for how to query
be on the reproducibility. The average user should be able
the database? Is it possible for users to verify the data in
to find the software, install it, run it, and check that the
the database themselves using some algorithms?
results are as the authors claim they are. A high level of
The code is used to compute examples included in
reproducibility will ensure that other mathematicians will
the paper. If the examples are not particularly hard to
be able to use the code, check whether it is correct or not,
compute, just having a separate, clearly named file for each
and figure out potential mistakes that were made. This ap-
example that runs the needed code will suffice. If comput-
proach has another advantage: it will reduce the time re-
ing the examples will take a long time, it will be helpful to
quired to do a code review, making it easier to find willing
provide an estimate of the expected duration.
reviewers.
The code is used to do computations as part of a
We propose four categories that should be part of a soft-
proof. If the code is used as part of a proof, it will be crucial
ware review:
to cross-reference between the code and the paper, as well
1. Role of the code in the paper as vice versa. Furthermore, the code should be properly
2. Reproducibility (Installation and rerunning the code) commented, allowing readers of the proof to understand
3. Reliability and Correctness the software component. The code is an actual part of the
4. Readability proof and should be treated as such.
We will discuss these categories in more detail in the fol- Now here are some general guidelines that may be help-
lowing sections, which will deal with mathematical soft- ful to keep in mind when writing research software.
ware from the point of view of an author and from the 4.1. Metadata and the role of the code in the paper.
point of view of a reviewer.
Describe the purpose of the code. Write a few sentences in the
4. Guidelines for Writing Code paper describing what kind of code accompanies the paper,
what it was used for, and where it can be found. It is often
The focus of an author seeking to make his research repro-
difficult to find this information, as it is usually scattered
ducible or a reviewer tasked with reviewing the code heav-
throughout the paper or missing altogether. Most people
ily depends on the role the code plays in the paper. We
declare what the intent of their paper is in the abstract or
will briefly discuss some common scenarios before outlin-
the introduction, but it is helpful to do something similar
ing the guidelines:
for the accompanying code.
The paper explains a new algorithm capable of com-
puting things we were not able to compute before. The Refer to sections of the paper in your code. Conversely, it can
main goal of the paper is to explain why this new algo- be helpful to reference your paper in your code. This way,
rithm works. The key part is to demonstrate its correct- if someone does not understand what is happening, they
ness. The theory behind the algorithm will primarily be can easily find the relevant sections in the paper.
explained in the paper, but examples and proper commen- 4.2. Reproducibility.
tary in the repository will be useful for understanding the Make sure your code has a license. The software you write
algorithm and verifying its correctness. Listing run times should come with a license. Without a license, no one
and system specifications will be less relevant.

770 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY VOLUME 72, NUMBER 7

 CONVERSATION

is allowed to reuse your code. If you have co-authors, a colleague or a friend try to install the software and see
they could even sue you if you reuse your own code, if they manage. Sometimes the issue lies with the installa-
as they have partial ownership. For more information tion instructions someone else wrote for a package you de-
about what kind of license to use, check out: https:// pend on. In that case, you could try to contact the original
choosealicense.com/. For academia, it is best to pick an authors of that code and ask them to improve their instruc-
open-source license. Common ones are the GPL license, tions or provide better instructions on your own page.
the MIT license, and the BSD licenses. One difference is Think about what you could do to make the installation as easy
that if someone wants to reuse code published under a GPL as possible. Complicated installations are sometimes un-
license, the new work also needs to be open source and be avoidable, but there are steps that can be taken to make
published under the same license. It can be used for com- the process easier for users. It may be possible to write an
mercial purposes, but everyone who receives a copy of the installation script that automatizes a couple of steps. Try
product must have access to the source code. The MIT and to minimize the need to manually change files as much
BSD licenses (which are relatively similar) do not enforce as possible. Sometimes installation instructions require
that derivative works are published under the same license. users to install a package from a different author and man-
The main condition is that the new work contains proper ually modify certain lines in that package. This is a bad
attribution to the original author. The software may then practice that can cause a lot of issues. Will the library
also be turned into commercial closed-source software. If still work with your changes? What if the package gets up-
your code depends on software written by someone else, dated?
ensure the license you choose is compatible with theirs. Provide documentation for the code. Every function you write
Long-term repository. Ensure the code is hosted on a plat- should include an explanation of the input and output,
form where it can remain for multiple years. A bad place as well as a description of its purpose. Make sure to also
to host would be on your own website. When people include examples of how to use your code. Ideally, your
change jobs or leave academia, these websites may disap- code should include enough explanatory text to help users
pear. Even if you move the code elsewhere, papers that understand what it does. A Jupyter Notebook would also
cite the code will still reference the old location, making it work well for something like this.
difficult to find the code. It is recommended to host your Mention the specifications of the systems used. Be sure to list
code on a public platform like GitHub, GitLab, or Zenodo. key hardware specifications (CPU, RAM), the OS used, and
GitHub has a lot of features and is the most popular plat- the exact versions of the programming languages and pack-
form. One possible negative of GitHub is that it is a sub- ages involved. For large or difficult computations, it is use-
sidiary of Microsoft. However unlikely, they might choose ful to record the amount of data produced and the time it
to monetize GitHub or make other unwanted changes at took for the computations to finish.
some point. Zenodo is hosted by the CERN data center
Citation. Make sure to give credit to the work you are build-
and geared toward science. Every upload also has a DOI
ing on. Cite the software and the packages you are using
(a persistent identifier), which makes them more easily
in your repository and in the accompanying paper.
citable.
4.3. Correctness and reliability.
Ensure your code has a Readme file. The goal of the Readme
file should be to be the visiting card of your repository. It Tests. It is good practice to write test code to boost con-
should give a quick impression of what the purpose of the fidence in your algorithms. The test code does not nec-
repository is. It is also a good place to provide an overview essarily need to test your examples. Its goal could be to
of the files contained in the repository and their purpose. test certain functions or submodules of the entire package,
The Readme could also include installation instructions, perform sanity checks, verify the code is working in certain
basic documentation, and some small examples. Finally, edge cases, or test randomly generated examples.
it can be used to reference the accompanying paper and Running the code on a real example could take a long
cite works that your software depends on. time, but well-chosen tests can make a convincing case that
the code works in principle despite taking less time to run.
Provide clear installation instructions. Users need to be able
When writing test code, aim to cover as much code as pos-
to run your code to verify your results, gain a deeper un-
sible. There also exist tools measuring code coverage that
derstanding of your work, or reuse it and improve upon
can help with this.
it. Many repositories currently do not include any instruc-
Test code can also help prevent bugs from sneaking in.
tions at all or only very vague ones. The average user will
If changing the code causes tests to fail, you know that
have difficulties figuring out what they need to do, so make
something went wrong. Tools for continuous integration
the instructions as explicit as possible. A good test is to let

AUGUST 2025 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY 771

 CONVERSATION

can help you run tests automatically every time you change documentation, Docker files/virtual machines, and
something in the code. test files.
Verification files. If you want to prove to someone that your • The version of the code that was reviewed. (If version
computations are correct, you might not want to redo the information is not available, the download date of the
computation, as it could take a lot of time. However, veri- code can be listed.)
fying that something is correct can sometimes be done in • The link that was used to download the code.
a different and potentially quicker way. Computing the • A short description of the relationship between the
inverse of a matrix 𝐴, for example, is different from mul- code and the paper. What does the code add to the
tiplying 𝐴 with the matrix you computed and checking if paper? How important is it for the results?
the result is the identity matrix. These verification tests also After the first phase of the software review, you should
increase confidence that your algorithm is correct. have a good idea of what to expect to find in the code. The
Multiple implementations. To check whether the data you next step will be to take a look at the repository and its
computed is correct, it can be helpful to implement your structure.
code in multiple software systems. In some cases, differ- 5.2. Reproducibility. A quick scan of the repository can
ent systems actually produce different results (due to bugs already reveal a lot:
in the software systems). Testing in different systems in-
• Does the repository have a license?
creases the chances of your results being correct.
• Where is the code hosted? The author’s own personal
4.4. Readability. website, for example, is a bad idea, as it is not guaran-
Use clear naming conventions. Thinking a bit longer about teed to be there for a long time. A platform like Zen-
proper names for files, functions, and variables will do odo would be ideal, but a GitHub repository is still
wonders for making code more comprehensible for other fine.
people (and even for yourself). No one likes to see names • Is the code publicly available?
like helperfunctions3 or mmhm. • Is the code open source?
• Make your naming conventions meaningful so people • Does the repository have a DOI that links to it?
who have read the paper can immediately understand • Does the repository have a Readme?
what each term represents. • What is the quality of this Readme?
• Stick to common mathematical notation such as 𝑓 for Now, the reviewer should attempt to install the code
function, 𝑣 for vector, etc. and run some of the examples or test code included. Write
• Use as few abbreviations as possible. Longer names down all the steps you have taken when trying to install
might help with legibility even if it takes a bit longer the code in the review. If the installation fails, the author
to type. will be able to use this information to help identify what
• Remain consistent with your naming scheme. caused the error.
Ensure your repository is clearly structured and that files are
• How difficult was it to install the software? Is it doable
easy to find. This is especially important if the repository is
for the average user?
big. If there are hundreds of files, you don’t want people
• Can the installation instructions be improved?
to spend hours trying to figure out what each one does and
• Check what information the authors gave about the
where to find what they need.
setup they used. This includes CPU, RAM, what op-
5. Guidelines for Reviewing Code erating system was used, what software was used, and
The first thing to do is to read the introduction to get a version information. It can also include benchmark-
rough impression of what the article is about. After this, ing information and running times used by the au-
skim through the rest of the article and take a look at the thors.
files provided. This should give you sufficient information • See if the authors properly cited all the software they
to determine all the necessary details for the first part of used in their program.
the review. To help the authors process the feedback they get from
5.1. Metadata. The review should contain the following the review, be sure to list:
information about the paper and its accompanying code: • Which operating system the reviewer used for the re-
• The title of the paper. view.
• The authors’ names (if given). • Which programming languages or standard software
• The date on which the review was written. packages were used for the review. (Including version
• An overview of the files accompanying the paper, such information.)
as: code, notebooks, computed data, example files,
772 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY VOLUME 72, NUMBER 7
 CONVERSATION

Once the software is installed and runs successfully on braist might not be as important to an applied mathemati-
some basic examples, the reviewer should attempt to re- cian. These guidelines are a step toward getting people to
produce the results described in the paper. think and discuss what kind of solutions would work best.
5.3. Reliability and correctness. There are various ways A discussion is what we need. I hope anyone who
to increase confidence in the correctness of the results thinks this is important will talk to their colleagues; talk
stated in the paper. The main way for the reviewer to test to their students; talk to editors of journals, organizers of
this is to try to run the code itself and see if the results conferences, and whoever can make a change in how we
match the authors’ claims. However, this might not always deal with research data and research software. If we want
be feasible. For instance, if the computations the authors to improve the situation, achieve higher-quality reusable
did took months, it is unrealistic to expect the reviewer code and data, and gain recognition for our coding efforts,
to rerun them. Nevertheless, there are some solutions that the change must come from within our community. We
can increase confidence that running the code on these big need to decide that we want this, and we need to actively
examples will also produce the correct results. start working toward it.
• The repository could include smaller examples and References
show the code works correctly there. [Bak16] Monya Baker, 1,500 scientists lift the lid on reproducibil-
• The repository could contain test files that show that ity, Nature 533 (2016), 452 –454, https://doi.org/10
parts of the code are working correctly. .1038/533452a.
• Sometimes verifying that the results are correct is eas- [Con23] The MaRDI Consortium, Research data management
ier and less time-intensive than doing the computa- planning in mathematics, 2023.
tions. The repository could include files that verify the [Eit22] Frachtenberg Eitan, Research artifacts and citations in
correctness of, or perform some sanity checks on, the computer systems papers, PeerJ Computer Science 8:e887
(2022).
computed results.
[Gre24] Christian Greiffenhagen, Checking correctness in
In any case, if an error arises or the results are different mathematical peer review, Social Studies of Science 54
from what was expected, the reviewer should describe ex- (2024), no. 2, 184–209, https://doi.org/10.1177
actly what they did and what went wrong. Sometimes it is /03063127231200274. PMID: 37776165.
easy to identify the issue, and the reviewer could attempt [LHH+ 24] Mugeng Liu, Xiaolong Huang, Wei He, Yibing Xie,
a quick fix, but it is not the reviewer’s task to debug the Jie M. Zhang, Xiang Jing, Zhenpeng Chen, and Yun Ma,
authors’ code. Research artifacts in software engineering publications: Status
and trends, Journal of Systems and Software 213 (2024),
5.4. Readability. The final thing a reviewer should check 112032, https://www.sciencedirect.com/science
is whether basic good coding practices are being respected. /article/pii/S016412122400075X.
• Do functions have a description of what they do? Are [LMF24] The LMFDB Collaboration, The L-functions and mod-
the input and the output clearly described? ular forms database, 2024. [Online; accessed 29 August
2024].
• Is the code properly annotated?
[WCN85] Robert Arnott Wilson, John H. Conway, and Si-
• Does the indentation look okay? mon P. Norton, Atlas of finite groups, 1985, https://api
• Are files, functions, and variables given distinctive and .semanticscholar.org/CorpusID:117473588.
descriptive names?
• Is the repository structured sensibly?
A software review that takes the four categories dis-
cussed above into account should provide a comprehen-
sive overview of the quality of the code without being too
time-consuming for the reviewer.
An example of the reviewing template I have used for
reviewing code can be found on https://github.com
/JHanselman/code-review-template.

6. Future Outlook Jeroen Hanselman

I personally consider it important for us as a mathematical
Credits
community to start implementing some form of quality
Photo of Jeroen Hanselman is courtesy of Jeroen Hanselman.
control for research software and research data in math-
ematics. There will probably not be a solution that fits
everyone. What might be important for a computer alge-

AUGUST 2025 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY 773

 AMS Student

The AMS offers partial travel support to

the annual Joint Mathematics Meetings

• Grow your network • Find graduate programs

• Connect with employers • Develop lasting connections
• Present your research

Submission deadlines and eligibility info:

www.ams.org/travel-grants
 FROM THE AMS SECRETARY

AMS Fellowships candidate’s research and will try to award the fellowship
to those for whom the award would make a real difference
in the development of their research careers. Work in all
Centennial Research Fellowship
areas of mathematics, including interdisciplinary work is
The AMS Centennial Fellowship Program makes an award eligible.
annually to an outstanding mathematician to help further
Application period. Applications will be collected via
their career in research. One award will be made for the
MathPrograms.org July 15, 2025–September 30, 2025
2026–2027 academic year in the amount of US$75,000.
(11:59 p.m. ET). Find more information at https://www
Acceptance of the fellowship cannot be postponed. AMS
.ams.org/centfellow. For questions, contact the Pro-
membership will also be offered to the recipient for the
grams Department at [email protected].
duration of the fellowship.
About this fellowship. Eligibility: The eligibility rules are Stefan Bergman Fellowship
as follows: The Stefan Bergman Fellowship was established in 2023
The primary selection criterion for the Centennial Fel- with the proceeds of the Stefan Bergman Trust to support
lowship is the excellence of the candidate’s research. the advancement of the research portfolio of a mathemati-
• Preference will be given to candidates who have not cian who specializes in the areas of real analysis, complex
had extensive fellowship support in the past. analysis, or partial differential equations. One award will
• Recipients may not hold the Centennial Fellowship be made for the 2026–2027 academic year in the amount
concurrently with another research fellowship such of US$25,000. AMS membership will also be offered to
as a Sloan, NSF postdoctoral fellowship, or CAREER the recipient for the duration of the fellowship.
award. About this fellowship. Applications will be accepted
• Under normal circumstances, the fellowship cannot from mathematicians at a US institution who have not re-
be deferred. ceived tenure or comparable (at the discretion of the selec-
• A recipient of the fellowship shall have held his or her tion committee) and have not held significant fellowship
doctoral degree for at least three years and not more support.
than twelve years at the inception of the award (that is, Awardees may use the fellowship in any way that most
received between September 1, 2014, and September effectively enables their research—for instance, for release
1, 2023, for the award beginning in August 2026). time, participation in special research programs, travel sup-
• Applications will be accepted from mathematicians port, childcare, etc. The award is issued through the re-
currently holding a tenured, tenure track, postdoc- cipient’s institution, and no part of it may be utilized for
toral, or comparable (at the discretion of the selection indirect costs.
committee) position at a US institution. Application period. Applications will be collected via
Applications should include a detailed research plan MathPrograms.org July 15, 2025–September 30, 2025
for the fellowship period that is contextualized by the re- (11:59 p.m. ET). Find more information at https://www
search statement. The plan should include a description of .ams.org/bergman-fellow. For questions, contact the
how the fellowship will support the applicant’s success. It Programs Department at [email protected].
should be no more than one page. The selection commit-
tee will consider the plan in addition to the quality of the

DOI: https://doi.org/10.1090/noti3219

AUGUST 2025 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY 775

 FROM THE AMS SECRETARY

In reviewing the many activ- tures. On the practical side, the 2024 JMM lost a significant
ities of the American Mathe- amount of money, at a level that is unsustainable, and we
matical Society in 2024 and have been actively working to identify opportunities for
selecting particular items to enhancing future meetings while also constraining costs.
highlight in this commu- In terms of book sales in 2024, the AMS had our
nication, it became appar- strongest revenue year since the pandemic, supported
ent that the varied efforts in part by continued strong eBook sales to libraries,
across the AMS all tie back particularly in Asia. Some of our newest publica-
to our mission, one that is tions demonstrate the breadth of titles that the AMS
so nicely encapsulated by publishes, including two personal favorites: Richard
the tagline “Advancing Re- Evan Schwartz’s latest book, Playing with Shape and
search. Creating Connec- Form: A Glimpse of Topology (https://bookstore.ams
tions.” The year 2024 in- .org/view?productcode=MBK/152), and Max Dehn:
volved a number of signif- Polyphonic Portrait (https://bookstore.ams.org/view?
Figure 1. John Meier, Chief
Executive Officer icant staff transitions, with ProductCode=HMATH/46), a curated collection of essays
in aggregate well over a cen- describing Max Dehn’s remarkable professional life and
tury (or two!) of experience contributions. The year 2024 also saw a milestone accom-
moving on and the resulting need to search, hire, and in- plishment with the publication of the 250th volume in
tegrate new faces into several key roles. I have pulled to- the Graduate Studies in Mathematics series, David Eisen-
gether this summary with admiration and gratitude for bud and Joe Harris’s The Practice of Algebraic Curves: A
the way that, amid these transitions, the Society stayed Second Course in Algebraic Geometry (https://bookstore
mission-focused and successfully continued its work. .ams.org/view?productcode=GSM/250). This is the
As I will note at the end of this report, all of the items first book in a pilot initiative, where the AMS offers a freely
mentioned within provide examples that demonstrate the available eBook along with a print version for purchase.
answer to the question “Why should someone choose to The AMS’s partnership with the MAA also continues,
be a member of the AMS?” The answer, in short, is that including the publication of Alissa S. Crans and Glen T.
each individual’s membership and engagement enable our Whitney’s The Mathematical Playground: People and Prob-
work to support the mathematical sciences, provide av- lems from 31 Years of Math Horizons (https://bookstore
enues for professional growth for individuals, and make .ams.org/view?ProductCode=PRB/38).
it possible for the AMS to amplify the voices of our com- In terms of journals, the AMS signed an agreement as
munity. More on that toward the end! part of India’s new “One Nation, One Subscription” pro-
The core functions of the Society are publishing mathe- gram, providing access to our journal content to thousands
matics and organizing meetings. We opened 2024 with the of institutions of higher education. I am also delighted to
Joint Mathematics Meetings in San Francisco. It was a very report that the Memoirs and Transactions of the AMS transi-
successful JMM, with thousands of presentations, from stu- tioned to double-anonymous peer review.
dent posters through Terence Tao’s AMS Colloquium Lec- In September 2024, the Mathematical Reviews moved
to new offices in a building located across the street from
DOI: https://doi.org/10.1090/noti3201 its former location. The move was remarkably smooth,

776 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY VOLUME 72, NUMBER 7

 FROM THE AMS SECRETARY

and the new location retains much of the charm of the for- Tomastik Prize in Differential Equations will be awarded
mer location but with significantly improved functional- every three years for notable work in differential equa-
ity. (And a working HVAC system.) The AMS continues to tions. The Mohammad K. Azarian Prizes for Mathemati-
provide access to MathSciNet to mathematicians around cal Reviews will be awarded annually on a rotating basis,
the globe, including providing this resource at deeply dis- highlighting and celebrating the contributions of early ca-
counted rates for some countries where the cost is prohib- reer, lifetime achievement, and MathSciNet for Develop-
itive. As one example, we continue to provide access free ing Countries reviewers.
of charge to mathematicians in Ukraine. While some activities of the AMS were not financially fa-
For those who would like some numbers, in 2024, vorable, other AMS activities and programs generated posi-
the Mathematical Reviews added 129,514 items to Math- tive net revenue and in total, it all balanced out reasonably
SciNet; verified and saved 324,257 author profiles; created well, with the AMS ending 2024 with a modest operating
49,794 new author profiles; and created 9,165 new insti- contribution.
tution codes. Let me now return to the topic of how your membership
In 2024 the Office of Government Relations continued and engagement allow the Society to support our mission
their advocacy efforts on behalf of the Society. They ar- and contribute positively to the mathematical community.
ranged 87 Hill visits to congressional offices and contin- The AMS is a leading mathematics society, and we are able
ued their suite of fellows programs that provide mathe- to have outsized, positive impact because of our members.
maticians with experience working on policy issues and Being a member of the AMS signals that you are invested
communication about the mathematical sciences. in our research journals and the mathematics they help
In January 2024, the AMS Council created a Task Force to disseminate and in the publication of research mathe-
to Support Mathematics Degree Programs in response to matics in a venue that is not profit-centered. It shows that
closures and threatened closures of mathematics programs. you care about developing the mathematics community
Dedicated task force members and AMS staff from the and making it a welcoming place where we value every-
Offices of Government Relations and Programs & Profes- one’s contributions. And for those who are engaged be-
sional Services took up the important work of this task yond paying the annual dues, your time and insight make
force. Recognizing that supporting departments and pro- the AMS more effective at achieving our mission. It is the
grams is more effective when it is done before a threat membership that makes it possible for us to stand up com-
emerges, the task force quickly moved from a defensive mittees to support mathematics programs, to offer confer-
stance toward a proactive stance. In January 2025, the ences even when expenses exceed revenues, and to publish
AMS Council supported modifying the charge and estab- a range of books where we can focus on their impact and
lished them as a standing committee. You might think not just on sales figures. If you have read this far then I
that a Society that has over 100 committees doesn’t need doubt I need to be telling you any of this, but perhaps I
another, but you would most definitely be wrong! The have suggested some good talking points you can use the
work of this committee is perfectly aligned with the AMS’s next time a colleague wonders why AMS membership is
mission to advance research and create connections, and I important.
want to thank the inaugural chairs of this committee, Jim Finally, I mentioned that 2024 was a year of transitions,
Brown (Occidental) and Christine Berkesch (University of with some adjustments to the org chart and retirements
Minnesota), for their work as this committee continues for- leading to vacancies. I myself was part of one of those
ward. transitions, as I stepped into the CEO role, which had for-
AMS members and donors were generous in 2024, merly been titled the executive director, in July 2024. I
allowing us to continue and sustain growth in travel have benefited greatly from the conversations I have had
grants and research enhancement grants for mathemati- with John Ewing, Donald McClure, Catherine Roberts, and
cians throughout their careers. There was a dramatic ex- Lucy Maddock who served during the interim period while
pansion of the Publication Support Fund—an endowment a search for the CEO was occurring. Their efforts have
directed toward the sustainability of AMS books, journals, served the Society well, and their gracious welcoming of
and databases. These texts can be technically sophisticated me is warmly appreciated.
and complicated to create, are often highly specialized in John Meier
subject matter and audience, and are at times made avail- Chief Executive Officer
able free of charge to the public as a service to the mathe- April 2025
matical community.
Through philanthropic support, the AMS was also able Credits
to establish two new prizes. The Edmond and Nancy Photo of John Meier is courtesy of Adam Atkinson, Lafayette
College.

AUGUST 2025 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY 777

Real Analysis | Complex Analysis | Partial Differential Equations | Real Analysis | Complex
Analysis | Partial Differential Equations | Real Analysis | Complex Analysis | Partial Differential
Advance
Equations your
| Real Analysis | Complex Analysis | Partial Differential Equations | Real Analysis |
research
Complex portfolio
Analysis | Partial with| Real
Differential Equations theAnalysis | Complex Analysis | Partial

STEFAN BERGMAN
Differential Equations | Real Analysis | Complex Analysis | Partial Differential Equations | Real
Analysis | Complex Analysis | Partial Differential Equations | Real Analysis | Complex Analysis |
Partial Differential Equations | Real Analysis | Complex Analysis | Partial Differential Equations
Real Analysis | Complex Analysis | Partial Differential Equations | Real Analysis | Complex
F E L LO W S H I P
Analysis | Partial Differential Equations | Real Analysis | Complex Analysis | Partial Differential
Equations | Real Analysis | Complex Analysis | Partial Differential Equations | Real Analysis |
Research mathematicians who meet the following criteria
Complex Analysis | Partial Differential Equations | Real Analysis | Complex Analysis | Partial
are encouraged to apply for this fellowship:
Differential• Equations | Real Analysis | Complex Analysis | Partial Differential Equations | Real
Engaged in real analysis, complex analysis,
Analysis | Complex Analysis
or partial | Partial
differential Differential Equations | Real Analysis | Complex
equations r t o Analysis |
f

0
o
pp
Partial Differential Equations | Real Analysis | Complex Analysis | Partial Differential
• At a US institution
e s
s u

,0 0 Equations
vid
5
• Have not received tenure
Real Analysis | Complex Analysis | Partial Differential Equations | Real oAnalysis | Complex
2
Pr

$
• Have not held significant fellowship support
Analysis | Partial Differential Equations

APPLICATIONS ACCEPTED BETWEEN

JULY 15–SEPTEMBER 30 EACH YEAR.
Learn more and apply: www.ams.org/bergman-fellow
 AMS COMMUNICATION

CASE Workshop Inspires

Students to Advocate
for Mathematics
Elaine Beebe

Since 2019, the American Mathematical Society (AMS) has “I am so proud of these students and have confidence
sponsored graduate students to participate in the Catalyz- that they will continue to serve the math community
ing Advocacy in Science and Engineering (CASE) Work- throughout their careers,” Saxe said.
shop as fellows. In April 2025, the AMS sponsored three CASE Fellows:
Organized by the Amer- Ann Clifton is a PhD candidate in the Computational
ican Association for the Analysis and Modeling program at Louisiana Tech Univer-
Advancement of Science sity, planning to graduate this summer. Her research is in
(AAAS), the annual CASE discrete math, in particular structural graph theory. “My
Workshop introduces STEM dissertation work concerns characterizing which graphs
students to the federal pol- have an equitable dissection—that is, graphs that can be re-
icymaking process. Span- cursively partitioned into balanced, connected subgraphs,”
ning three and a half days Clifton said.
in Washington, DC, the The CASE Fellowship seemed like a great opportunity
workshop empowers grad- to learn more about how to be an effective advocate for
uate students and upper- higher education, she said. “Because my route to the PhD
level undergraduates to be- has been nontraditional—I’ve held a full-time instructor
come advocates for basic re- role while completing my degree—I’ve seen firsthand how
search. uncertainty in federal and state funding shapes campus
Figure 1. (l-r) Ann Clifton, Sam “Math students often are life,” Clifton said.
Glickman, and Travis Dillon. eager to engage in advocacy Travis Dillon is a fourth-year graduate student in math-
and push for change but ematics at the Massachusetts Institute of Technology (MIT).
few have experience doing so or knowing how to start,” “My research is in combinatorial and discrete geometry,
said Karen Saxe, AMS senior vice president for government which I describe to nonmathematicians as ‘thinking about
relations. “The CASE workshop is a boot camp during shapes in many dimensions,’” he said.
which AMS-sponsored CASE Fellows learn about federal Dillon has done outreach, teaching at Canada/USA
policymaking and legislative processes. They learn from Mathcamp in the summer, but CASE was his introduction
the best in DC. The final day is spent on Capitol Hill advo- to science advocacy. “I love thinking about pure mathe-
cating for research funding in Congress. matics, but it can feel very removed from the actual world,”
he said. “As I’ve thought about what to do after defending
Elaine Beebe is the communication and outreach content specialist at the Amer- my thesis, I’ve realized I want to spend more of my time
ican Mathematical Society. Her email address is [email protected]. on projects that have an impact beyond the mathematical
For permission to reprint this article, please contact: research community.”
[email protected].

DOI: https://doi.org/10.1090/noti3208

AUGUST 2025 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY 779

 AMS COMMUNICATION

Sam Glickman is a fifth- a scientist/mathematician/engineer. He also advised us to

year mathematics PhD stu- do what we’re good at. Not everyone is going to be a great
dent at the University of science communicator, and that’s fine, but if that’s what
Hawai’i at Mˉanoa. “I study you enjoy—get out there!
the mathematical founda- Tuesday was focused on how to make the most of our
tions of AI, computational visits to Capitol Hill. Unlike how we’re trained to write pa-
science, and data science, pers, in order to effectively engage, you need to start with
and my dissertation focuses the conclusion/impact/”why should I care?” first. They
on optimal control theory also emphasized the “dialogue model” over the “deficit
and mathematical model- model” of scientific communication. We oftentimes think
ing,” he said. if we provide a better argument or more evidence that we
In order to blend his can sway someone to our side but that’s just not the case.
strong interest in mathe- It’s important to have a story to tell. Passion and human
matics with hands-on pol- connection to your field is what will stick.
Figure 2. Sam Glickman visits icy work, Glickman applied Dillon: The opening remarks by Sudip Parikh set the
the Capitol Hill office of his for the CASE workshop, his tone for the meeting for me. He was warm and pragmatic,
senator, Mazie Hirono (D-HI).
first experience with advo- explaining what AAAS has focused its efforts on in the last
cacy on the national level. few months and describing practical action amid the in-
He grew up exposed to public service; Glickman’s father cessant noise of the media. And he emphasized that, in
was a state representative in the Florida House and his the long term, one of the most important actions we can
mother, a physician, was a medical director in the VA Hos- all take is to build relationships in our community. As
pital system. Glickman has collaborated with his advisor he pointed out, most people don’t know anyone working
and research colleagues on advocacy and outreach projects, in science, so showing up in your community, not to ad-
communicating the value of STEM research and education vocate, just as a scientist or mathematician, is one of the
to the Hawai’i state government and business community most important things we can do to build public connec-
through the Hawai’i Chamber of Commerce’s Young Pro- tion with and support of science.
fessionals Program. Glickman: The Monday panel titled “How Congres-
AMS: How much did you know about CASE in ad- sional Offices and Committees Operate” definitely had the
vance? Was the fellowship as you expected? greatest impact on me. We heard from mathematician Dr.
Dillon: I knew the rough schedule from previous years, Duncan Wright [the 2022–2023 AMS Congressional Fel-
but the program was so much better than I expected. The low] about his work as a House Oversight and Govern-
days were pretty full, but the talks were engaging and infor- ment Reform Committee staffer and heard an inspiring,
mative. I feel like I just downloaded information for two motivational story from Julia Pan, a Senate Committee
days—it was a really effective boot camp! on Commerce, Science, and Transportation staffer. I con-
Glickman: I had read articles about prior AMS- nected with both of them on LinkedIn after!
sponsored CASE attendees before applying, and their expe- AMS: Tell us about your Wednesday visit to Capitol
riences stoked my interest. The workshop was much more Hill.
dynamic than I expected, blending practical information Glickman: It went very well! I was fortunate enough
about communicating science with detailed looks into the to meet with Dr. Robby Goldman from Senator Mazie
structure of the federal government and the current federal Hirono’s office and Alan Muneoka from Representative
landscape. Ed Case’s office, both of whom were incredibly attentive
Clifton: It was pretty much what I expected from my and receptive, alongside Kirsten Kau, MPH, a veterinary
background research. One thing that was surprising was medicine student from Hawai’i who studies at Oregon
just how much information they fit into two days. State University. Kirsten and I both discussed our research
AMS: Of the Monday and Tuesday group sessions, and how it relates to national priorities, and we all talked
which had the greatest impact on you and why? about how scientists can advocate for STEM research and
Clifton: I don’t think I can choose between the two days. education in our communities.
They both felt equally impactful and important. Clifton: I joined a group from Tulane University to visit
On Monday, I really enjoyed hearing from Sudip Parikh, representatives from Louisiana. We met with the offices of
CEO of AAAS, about how to be effective members of the Congresswoman Julia Letlow, Congressman Troy Carter,
community. He said the best thing we can do is engage and Senator Bill Cassidy. During the first visit, I think we
locally, build relationships, and be willing to call yourself were all a little nervous but by the third we were confident

780 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY VOLUME 72, NUMBER 7

 AMS COMMUNICATION

in sharing our backgrounds and research interests and ex- Not everyone has to be a “science communicator.” It’s
plaining why sustained and increased support for federal important to find ways to get involved that play to your
funding is crucial. I felt like the visits were productive and strengths.
would not hesitate to participate in a Hill or state visit in AMS: What are your future plans with respect to ad-
the future. vocacy?
Dillon: It was great! I had joint Hill visits with the other Clifton: I’m looking forward to bringing what I learned
students from Massachusetts, and I was nervous as we were back to campus. I’m hoping to coordinate a group visit to
preparing the night before—Congressional offices are so our representatives at the state capitol and also spread the
big and important, a far cry from the relaxed atmosphere word about the CASE Workshop and Fellowship.
of mathematics departments. But after the first meeting, I Dillon: I’m going to keep exploring the area! I got a
realized that even if the offices are big and important, hav- lot of book and news source recommendations at the con-
ing a meeting with a Congressional staffer is nothing more ference that I’m going to start reading. I’m considering ap-
and nothing less than having a conversation, and I know plying to the AAAS Science & Technology Policy Fellow-
how to do that—I’ve been doing it all my life! Of course, ship next year. And I’m thinking about how I can have an
the AAAS had been telling us as much the whole week, but impact while I’m in grad school, by visiting my represen-
having the meetings really made me realize how feasible tatives and by building community, in my university, my
this is for me. After these visits, I can easily imagine coor- neighborhood, and my home town.
dinating a grad student trip to Congressional offices in the Glickman: In the short term, I intend to keep in touch
future. with my legislators’ offices about my work and ways my
AMS: For you, what were the big takeaways from colleagues and I can keep advocating for STEM. It turns out
CASE? that Senator Hirono is cosponsoring a bill that involves
Glickman: Anyone can be an advocate for STEM re- an application of my research, so I want to stay up-to-date
search and education. Legislators represent their con- and in touch with her office on that. In the long term, I
stituents, so if a constituent wants to see something em- intend to pursue a career in public service and maybe run
phasized, they can simply give their legislator a call, send for office myself one day!
them an email, or, if they are willing to brave the morn-
ing lines outside of the Congressional office buildings, go Credits
right up to them and tell them in person! Figure 1 and Figure 2 are courtesy of Sam Glickman.
Dillon: Advocacy is really possible, not just in the fu-
ture, but right now. My Hill visits really inspired a sense
of agency and possibility.
My other takeaway is that in the long term, one of the
most important things we can all do is to build commu-
nity: meet your neighbors, attend city council meetings,
talk with people you don’t know, and talk with people
you might disagree with. We’re all in this together, after
all, and most people, regardless of their specific political
beliefs, understand that.
Clifton: How you engage in your community is impor-
tant. As educators and mathematicians, sometimes we’re
tempted to teach or talk “at” people but there has to be
two-way dialogue. The first step is to just get involved and
meet people and then be willing to casually share what
you do professionally. Showing up and saying “I’m a sci-
entist/mathematician and you should listen to me because
I’m an expert” won’t be as effective.
There is more bipartisanship on the Hill than what we
are able to see as the public. I found that encouraging.
I also feel like I have a better understanding of opportu-
nities to be engaged in this process that might even involve
career paths outside of what are typical for a PhD mathe-
matician.

AUGUST 2025 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY 781

 AMS COMMUNICATION

Julia Robinson Mathematics

Festival a Hit in Providence
Elaine Beebe

Flanked by parents, kids aged mostly 5 to 10 swarmed

the nine stations, from the Jumpin’ Julia hopscotch game
(a logic puzzle) to ever-popular Pentominoes to the afore-
mentioned Mosaics.
“The object of the game is to put at least one tile of each
of three colors—green, yellow, and blue—into a 2x2, 2x3,
or larger rectangle, while also following additional rules,
such as that in the solution no tiles of certain colors—e.g.,
green and yellow or yellow and blue–may touch on the
long edge, or there must be the same number of two col-
ors,” explained Sarah Klyberg of the AMS Programs depart-
ment, who volunteered at the JRMF with her son, William,
Figure 1. Arvisha Dooda, 5 (center), plays the Changing a seventh-grader.
Colors shapes game at the Julia Robinson Math Festival in
Providence. Nine-year-old Arush Dooda spent almost two hours
playing Mosaics. Starting with level one, he completed
The American Mathematical Society hosted its first local the tile puzzles sequentially, as they increased in difficulty,
Julia Robinson Mathematics Festival (JRMF) on April 10, with intense concentration.
and the Mosaics puzzle station was the place to be. Even Meanwhile, his sister, Arvisha, 5, experienced the smor-
Rhode Island’s governor thought so. gasbord of puzzles that is a Julia Robinson Math Festi-
About 130 parents and children attended the JRMF, val by trying almost every game. Zipping from station to
held in the airy cafeteria of Pleasant View Elementary in station with her long braids sailing behind her, Arvisha
Providence from 5 to 7 p.m. on a Thursday night. showed similar focus in the tasks the puzzles set forth.
At cafeteria tables, 25 volunteers ran stations with Their mother, Keerdhi Dooda, said that both her children
brightly colored puzzles and games for all ages, designed enjoy math, which drew the Providence family to the JRMF.
with a low floor so that anyone can engage and a high “When he plays video games, they are math,” she said of
ceiling so that everyone can find a meaningful challenge. her son. “And when we have families over to our house, I
Wearing matching red jackets, most volunteers were from hear him talking about math with the other children.”
City Year Providence, an education-focused nonprofit ser- On hand to highlight the Math Matters RI campaign,
vice organization that places a corps of young mentors in a statewide math education initiative, Governor Dan Mc-
public schools. Kee sat at the cafeteria tables and tried his hand at a few
puzzles before settling down at Mosaics. “I was thrilled
to see the level of family engagement at the Pleasant View
Elaine Beebe is the communication and outreach content specialist at the Amer- Elementary event, which ties in beautifully with my latest
ican Mathematical Society. Her email address is [email protected]. education initiative, Math Matters RI,” McKee said. “One
For permission to reprint this article, please contact: of my administration’s education goals is to meet or ex-
[email protected]. ceed Massachusetts students’ test scores in Math and Eng-
DOI: https://doi.org/10.1090/noti3204
lish Language Arts, and programs like this one are an

782 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY VOLUME 72, NUMBER 7

 AMS COMMUNICATION

Figure 2. William Klyberg (left) teaches RI Governor Dan

McKee (right) the Mosaics game.

effective and fun way to build math skills. I want to

thank the American Mathematical Society, City Year, and Figure 3. Representatives of City Year Providence (left), the
Providence Public Schools for organizing such a successful Rhode Island Department of Education (RIDE) (right), John
evening.” Meier, AMS CEO (center left), and Rhode Island Governor Dan
Meanwhile at the table, Arush and another fourth-grade McKee (center right).
boy started sharing hints for each other’s puzzles. “It was math together,” said Rachel Turner, City Year development
fun to see them collaborate, and I shared with them that operations manager and volunteer at the JRMF. “It’s always
a big part of what research mathematicians do is collabo- powerful to create opportunities where young people can
rate on problems, which is exactly what they were doing!” explore their sparks—and there was so much of that hap-
said Sarah Klyberg. “The boys looked at each other with pening. The joy across the community and age groups was
little grins after I told them they were doing what research truly inspiring.”
mathematicians do.” Kline concurred. “I thought the event was a really big
Over at the Jumpin’ Julia game, made of floor mats with success! There was great turnout with over 130 attendees,
grids of numbers, Paloma Reyes, age 4, explored while especially for the first AMS-sponsored Math Festival in
her mother, Yajaira, recalled an unpleasant math experi- Rhode Island. It was also great to see the whole commu-
ence in high school. “It’s best if she’s exposed to different nity turn out for the event.”
ways of learning, and math especially,” she said of Paloma. “I’m looking forward to continuing to work with the
“There’s something she can gain here even at her age.” AMS to spread the joy of math,” Kline said.
“A Math Festival is a time for the whole community to “Sometimes we highlight the many ways that mathe-
come together and celebrate the joy of math,” said Daniel matics is useful and a key to unlocking professional op-
Kline, executive director of the California nonprofit Julia portunities,” Meier said. “That’s true and important, but
Robinson Mathematics Festival. “Our mission with host- sometimes I worry that it can make it sound like mathe-
ing Math Festivals is to help all students build confidence, matics is an ‘eating icky vegetables’ activity.”
joy, and a sense of belonging around math.” “We had well over a hundred people who came away
Kline flew to Rhode Island to train the volunteers and at- knowing that math is useful, important, and absolutely
tend the state’s first festival. The AMS has organized JRMF fun.”
events at the Joint Mathematics Meetings in 2023, 2024, A girl of about seven at the festival would agree. When
and 2025, but this was the first festival near the AMS Prov- asked “How much do you like math?” she did visible men-
idence headquarters. tal computations, then said, “Ninety-two percent.”
“It was terrific to see the AMS partnering with City Year
Providence and Pleasant View Elementary to bring a Julia Credits
Robinson Mathematics Festival to life here in Providence,” All figures are courtesy of Drew Lagace, AMS Communica-
said AMS CEO John Meier. He loved watching Dr. Javier tions.
Montañez, the Providence Public School District super-
intendent, sitting down to work on mathematics puzzles
with the families that came to the festival.
“It was a wonderful event with students of all ages and
their families teaming up with City Year members to tackle

AUGUST 2025 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY 783

 NEWS

AMS Updates
where support is allocated via a rigorous peer-review pro-
AMS Signs “Call for cess and continues to advocate quite strongly for this sys-
Constructive Engagement” tem.”
The statement continues:
The American Mathematical Society (AMS) joined more Our colleges and universities share a commitment to serve as
than 500 members of the educational community in sign- centers of open inquiry where, in their pursuit of truth, faculty,
ing a public statement, “A Call for Constructive Engage- students, and staff are free to exchange ideas and opinions across
ment,” the first instance where current presidents have spo- a full range of viewpoints without fear of retribution, censorship,
ken out collectively in large numbers on federal matters. or deportation.
At the time of release on April 22, 2025, more Because of these freedoms, American institutions of higher
than 150 current leaders of America’s colleges, univer- learning are essential to American prosperity and serve as pro-
sities, and scholarly societies had signed the letter, ac- ductive partners with government in promoting the common
cording to a press release. The full statement and good.
an up-to-date list of signatories—numbering 568 as of
April 30—are available online at https://www.aacu.org —AMS Communications
/newsroom/a-call-for-constructive-engagement.
“This letter appropriately highlights the importance of
open inquiry and the ways that research done at colleges The Axiom of a Sonnet:
and universities contributes to the common good,” said
John Meier, AMS CEO. Hart Gives the 2025
The public statement was developed by university and
college presidents and other educational leaders nation-
Einstein Public Lecture
wide after national meetings convened by the American Mathematics is often referred to as the language of the sci-
Association of Colleges and Universities (AAC&U) and the ences. Sarah Hart sees its language present within the arts
American Academy of Arts & Sciences. as well, especially literature. Hart was at Clemson Univer-
As leaders of America’s colleges, universities, and scholarly sity on March 8, 2025, to deliver the 2025 AMS Einstein
societies, we speak with one voice against the unprecedented gov- Public Lecture in Mathematics as part of the AMS South-
ernment overreach and political interference now endangering eastern Spring Sectional Meeting.
American higher education, the statement begins. Before Hart gave her lecture, “A Mathematical Journey
We are open to constructive reform and do not oppose legit- Through Literature,” she was approached by a fan bran-
imate government oversight. However, we must oppose undue dishing a book. He asked Hart to sign his copy of her 2023
government intrusion in the lives of those who learn, live, and book, Once Upon a Prime: The Wondrous Connections Be-
work on our campuses. We will always seek effective and fair tween Mathematics and Literature (Flatiron Books), which
financial practices, but we must reject the coercive use of public he said with delight that he had read all the way through
research funding. without stopping. Hart, who is a professor emerita of
“The letter speaks about the full range of scholarly in- mathematics at Birkbeck, University of London, and a Fel-
quiry and is not focused on mathematics,” Meier said. low of Gresham College, signed happily, with a flourish.
“However, having a stable, well-supported infrastructure A New York Times Book Review Editors’ Choice, Once
for creative investigation is vital to mathematics and all Upon a Prime won the Mathematical Association of Amer-
aspects of mathematicians’ professional lives. The AMS ica’s 2024 Euler Book Prize. Witty and wise, the book has
knows the value of curiosity-fueled, fundamental research been translated into Chinese, Italian, Korean, and Spanish.
DOI: https://doi.org/10.1090/noti3216

784 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY VOLUME 72, NUMBER 7

 NEWS

In it, Hart bridges the worlds of mathematics and literature commitment to public engagement, and I feel honored to
with grace and ease. have been invited to deliver it.”
She did the same for an enthusiastic audience at Clem-
son. “We talk about mathematics as being the language —AMS Communications
of the universe, a vital tool for science,” Hart said. But,
she noted, mathematics also provides the rhythm of music,
symmetries in art, poetry rhyme schemes, and symbolism Deaths of AMS Members
in literature.
“We set up some ground rules—like the axioms of Eu- Bezalel Peleg, of Israel, died on May 9, 2019. Born on Feb-
clidean geometry—and within those rules, we have a play- ruary 25, 1936, he was a member of the Society for 53
ground,” she said. “Literature itself has an inherent struc- years.
ture much like geometry.” John E. Wetzel, of Champaign, Illinois, died on Novem-
Not only are mathematics and literature inextricably ber 29, 2021. Born on March 6, 1932, he was a member
linked, Hart said, but understanding these links can en- of the Society for 65 years.
hance the enjoyment of both. From haiku to sonnets to Mark D. Pankin, of Alexandria, Virginia, died on De-
sestinas, she explained the mathematics of poetry. Explor- cember 17, 2021. Born on January 11, 1945, he was a mem-
ing fiction’s hidden mathematical structure, she unveiled ber of the Society for 50 years.
mathematical imagery and symbolism in works from fairy Dinu D. Amzar, of Germany, died on March 30, 2022.
tales to classic novels such as Moby Dick, and even popular Born on March 11, 1943, he was a member of the Society
fiction such as Jurassic Park. for 24 years.
Hart unraveled the construction of a sestina, a form of Joseph Neggers, of Tuscaloosa, Alabama, died on Au-
poetry that has six stanzas of six lines apiece and complex gust 10, 2022. Born on January 10, 1940, he was a member
rules in terms of repeating words within. Could this poem of the Society for 59 years.
structure be based on other numbers as well? George W. Morgan, of Providence, Rhode Island, died
Using four doesn’t work, as Hart showed the audience: on February 4, 2023. Born on May 11, 1924, he was a
You can’t write a quartina. But, she declared, “We mathe- member of the Society for 69 years.
maticians like to solve for n.” Jackson L. Sedwick, of Chesterfield, Missouri, died on
Hart explicated her discovery. The n-ina is guaranteed March 26, 2023. Born on March 14, 1930, he was a mem-
to work if n is a Sophie Germain prime, where n and 2n ber of the Society for 41 years.
+1 must both be prime. So go home and write an 11-ina, R. Sherman Lehman, of Berkeley, California, died on
or try your hand at a 23-ina, she joked with the audience: July 28, 2023. Born on January 25, 1930, he was a member
prompting laughter, not for the first time. of the Society for 71 years.
Hart’s Einstein Public Lecture was ebullient, joyful, and Stephen S. Shatz, of Philadelphia, Pennsylvania, died in
well received by 110 attendees at Clemson. Her vast expe- August 2023. Born on April 27, 1937, he was a member of
rience made her a natural choice to deliver the annual lec- the Society for 65 years.
ture, which the AMS has hosted at one of its eight Sectional Alan E. Scrivner, of Loveland, Colorado, died on No-
Meetings since 2005. “[Einstein] Lecturers are chosen in vember 14, 2023. Born on September 1, 1957, he was a
the tradition of sharing fundamental understandings with member of the Society for 44 years.
broad appeal in mind,” according to the AMS. Surinder K. Sehgal, of Powell, Ohio, died on February
From 2020 to 2024 Hart delivered six public lectures 26, 2024. Born on April 22, 1938, he was a member of the
per year as the Gresham Professor of Geometry, the oldest Society for 50 years.
mathematics professorship in Britain. She was the thirty- Allen Stenger, of Boulder, Colorado, died on December
third person and first woman to be appointed to the post 6, 2024. Born on March 11, 1952, he was a member of the
since the position’s inception in 1597, and her lectures Society for 23 years.
have been viewed on YouTube up to 300,000 times apiece. David C. Kay, of Asheville, North Carolina, died on Jan-
“I believe it’s important for the mathematics commu- uary 29, 2025. Born on July 26, 1933, he was a member
nity to share with a broad audience just what a wonder- of the Society for 64 years.
fully exciting and creative discipline mathematics is—it’s Victoria Powers, of Atlanta, Georgia, died on February
not about hard sums or boring calculations, just as litera- 2, 2025. Born on July 28, 1958, she was a member of the
ture isn’t about spelling or grammar,” Hart said. Society for 45 years.
“The fact that the AMS has this prestigious public lecture
as part of its annual calendar is an important signal of its

AUGUST 2025 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY 785

 NEWS

Mathematics People
Mathematicians Elected to The latest class is comprised of 471 scientists, engineers,
and innovators across 24 AAAS disciplinary sections. The
American Academy of Arts new Fellows were to be celebrated at a forum in Washing-
ton, DC, on June 7, 2025.
and Sciences Section on Mathematics:
Solomon Friedberg, Boston College (AMS Fellow, AMS
Seven mathematicians were among the nearly 250 mem- life member);
bers elected in 2025 to the American Academy of Arts and Overtoun M. Jenda, Auburn University (AMS member);
Sciences: Jonathan C. Mattingly, Duke University (AMS Fellow,
Ian Agol, University of California, Berkeley (AMS Fel- AMS life member);
low, AMS life member); Linda Preiss Rothschild, University of California San
Frank Calegari, University of Chicago (AMS Fellow, Diego (emeritus) (AMS Fellow, AMS life member);
AMS member); Frank Sottile, Texas A&M University (AMS Fellow, AMS
Jonathan Christopher Mattingly, Duke University (AMS life member).
Fellow, AMS life member); Section on Statistics:
Christopher Skinner, Princeton University (AMS Fellow, Alexander Aue, University of California, Davis;
AMS member); Jinbo Chen, University of Pennsylvania;
Philip B. Stark, University of California, Berkeley; Scott H. Holan, University of Missouri;
Michael I. Weinstein, Columbia University (AMS Fel- Lexin Li, University of California, Berkeley;
low, AMS member); Xihong Lin, Harvard University;
Lauren K. Williams, Harvard University (AMS Fellow, Daniel S. Nettleton, Iowa State University;
AMS life member). Cynthia Rudin, Duke University.
“Since 1780, the American Academy of Arts and Sci- “This year’s class of Fellows are the embodiment of
ences has honored excellence and convened leaders from scientific excellence and service to our communities,”
across disciplines and divides to examine new ideas, ad- said Sudip S. Parikh, AAAS chief executive officer, in a press
dress issues of importance, and work together ‘to advance release. “At a time when the future of the scientific enter-
the interest, honor, dignity, and happiness of a free, in- prise in the US and around the world is uncertain, their
dependent, and virtuous people,’” according to a press work demonstrates the value of sustained investment in
release. “The newest members have distinguished them- science and engineering.”
selves in academia, the arts, industry, policy, research, and
science.” —American Association for the Advancement of Science

—American Academy of Arts and Sciences

Wallenberg Foundation Names
Twelve Elected to 2024 2025 Awardees
Class of AAAS Fellows Sixteen mathematicians shared SEK 35 million (US$3.6
million) in research funding as the Knut and Alice Wallen-
Five mathematicians and seven statisticians have been berg Foundation, in cooperation with the Royal Swedish
elected to the 2024 class of Fellows of the American As- Academy of Sciences, named its 2025 awardees.
sociation for the Advancement of Science (AAAS). Eight researchers received international postdoctoral
positions and funding for two years after they return to
DOI: https://doi.org/10.1090/noti3215
Sweden:

786 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY VOLUME 72, NUMBER 7

 Mathematics People
NEWS

Lukas Nakamura, Uppsala University (Institut des Shen was honored “for his deep contributions to the
Hautes Études Scientifiques); theory of stochastic partial differential equations and the
Axel Ljungström, Stockholm University (University of furthering of our understanding of the stochastic Yang-
Nottingham); Mills dynamics.”
Simon Cooper, Stockholm University (King’s College Subag was honored “for his deep and influential results
London); on the geometry of spin glasses.”
Joakim Cronvall, Lund University (KU Leuven); Sellke was honored “for his wide-ranging contribu-
Federica Milinanni, KTH Royal Institute of Technol- tions to applications of probability, especially in the de-
ogy (Northwestern University); velopment and understanding of algorithms for high-
Mateusz Stroinski, Uppsala University (University of dimensional optimization.”
Hamburg); Since 1976, the Rollo Davidson Trust has awarded an
Sven Sandfeldt, KTH Royal Institute of Technology (Uni- annual prize to one or more young probabilists. The
versity of Chicago); Trust was founded in 1975 in memory of Davidson, an ac-
Danai Deligeorgaki, KTH Royal Institute of Technol- complished mathematician of remarkable potential and
ogy (Max Planck Institute for Molecular Cell Biology and Fellow-elect of Churchill College, Cambridge, who died
Genetics). on the Piz Bernina in 1970. In addition to the annual
Five researchers received grants to recruit a foreign re- prize for young probabilists, the Trust awards an annual
searcher for a postdoctoral position in Sweden: prize for master’s students.
Daniel Ahlberg, Stockholm University;
Simon Larson, University of Gothenburg; —Rollo Davidson Trust
Magnus Goffeng, Lund University;
David Witt Nyström, University of Gothenburg;
Simon L. Rydin Myerson, Chalmers University of Tech- Goda Wins Traub Prize
nology.
Three established researchers from outside Sweden will for Achievement in
be visiting professors at Swedish universities:
Stefano De Marchi, University of Padua (Uppsala Uni-
Information-Based Complexity
versity);
Takashi Goda, University of Tokyo, is the recipient of
Adrianna Gillman, University of Colorado Boulder
the 2025 Joseph F. Traub Prize for Achievement in
(KTH Royal Institute of Technology);
Information-Based Complexity (IBC), as announced by
Johannes Rau, University of the Andes (University of
the Journal of Complexity.
Gävle).
Goda works in numerical analysis and uncertainty
The awards by the Wallenberg Foundation, in coopera-
quantification, with a particular focus on quasi-Monte
tion with the Royal Swedish Academy of Sciences, will sup-
Carlo methods, uniform distribution theory, and tractabil-
port prominent researchers in mathematics during 2014–
ity of high-dimensional problems, such as integration and
2030. The funding totals SEK 650 million (US$67.5 mil-
approximation of smooth functions.
lion). Including this year’s grants, 168 researchers have
The annual prize is for outstanding achievement in
received funding since 2014.
information-based complexity. The achievement can be
based on work done in a single year, a number of years, or
—Knut and Alice Wallenberg Foundation
over a lifetime. It can be published in any journal, num-
ber of journals, or monographs. Goda will receive $3,000,
sponsored by Elsevier, and a plaque.
Shen, Subag, Sellke Awarded Nominations for the 2026 Traub Prize are due March
Rollo Davidson Prize 31, 2026, to the award committee. Email submissions to
Erich Novak at [email protected].
Hao Shen (University of Wisconsin), Eliran Subag (Weiz-
mann Institute), and Mark Sellke (Harvard University) are —Journal of Complexity
joint recipients of the 2025 Rollo Davidson Prize, as an-
nounced by the Rollo Davidson Trust, Churchill College,
University of Cambridge.

AUGUST 2025 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY 787

 NEW BOOKS

New Books Offered by the AMS

Algebra and Algebraic Applications
Geometry Experimental
Mathematics
Mathematical
Surveys
Iwasawa Theory and Its A Computational Perspective
Perspective,Volume 3
and
Monographs

Volume 291
Matthew P. Richey, St. Olaf
Tadashi Ochiai, Institute of Sci- College, Northfield, MN, and
Iwasawa Theory
and Its Perspective ence Tokyo, Japan Matthew L. Wright, St. Olaf
Volume 3 Iwasawa theory began in the late College, Northfield, MN
1950s with a series of papers
Tadashi Ochiai
This textbook is designed for
by Kenkichi Iwasawa on ideal an undergraduate course in
class groups in the cyclotomic computational or experimental

relation to 𝑝 ​-adic 𝐿-functions.

tower of number fields and their mathematics. Many of the most
interesting and important topics
The theory was later generalized in theoretical mathematics are both motivated and illu-
by putting it in the context of minated by computation. By using a computer to explore
elliptic curves and modular forms. The main motivation and investigate mathematical phenomena, students are
for writing this book was the need for a total perspective encouraged to conduct experiments and make their own
of Iwasawa theory that includes the new trends of general- discoveries.
ized Iwasawa theory. Another motivation is to update the The explorations are motivated by rich and interesting
classical theory for class groups, taking into account the mathematical examples drawn from diverse areas such as
changed point of view on Iwasawa theory. number theory, complex analysis, and probability. In this
The goal of this third part of the three-part publication setting, students are exposed to key concepts from pro-

theory of 𝑝​-adic Galois representations.

is to present additional aspects of the cyclotomic Iwasawa gramming such as algorithms, recursion, and probabilistic
sampling, allowing them to develop practical computa-
This item will also be of interest to those working in number tional skills. Over 450 practice problems, exercises, and
theory. explorations are distributed regularly throughout the text,
reinforcing the material and encouraging further investi-
Mathematical Surveys and Monographs, Volume 291
gations. A code repository, providing implementations of
July 2025, approximately 110 pages, Softcover, ISBN: 978-
these examples, is available at the AMS webpage for the text.
1-4704-7732-5, LC 2022056149, 2020 Mathematics Subject
Thoroughly classroom tested and designed to show
Classification: 11F33, 11F67, 11F80, 11G40, 11R23, 11R34,
students how to combine mathematical and computa-
11R42, List US$135, AMS members US$108, MAA mem-
tional thinking, the book requires prior knowledge of only
bers US$121.50, Order code SURV/291
calculus and introductory linear algebra. Only minimal
bookstore.ams.org/surv-291 prior experience with computer programming is necessary,
provided the reader has a computing environment, access
to online documentation, and a willingness to experiment.
Notably, the volume has been carefully prepared so that
it may be used with almost any programming language.
This item will also be of interest to those working in number
theory and probability and statistics.

788 Notices of the American Mathematical Society Volume 72, Number 7

 NEW BOOKS

AMS/MAA Textbooks, Volume 74

August 2025, approximately 364 pages, Softcover, ISBN: Discrete Mathematics and
978-1-4704-7397-6, 2020 Mathematics Subject Classifica-
tion: 97N80, 00A35, List US$85, AMS Individual member
Combinatorics
US$63.75, AMS Institutional member US$68, MAA mem-
bers US$63.75, Order code TEXT/74 Milestones in Graph
bookstore.ams.org/text-74 Theory
A Century of Progress
Lowell W. Beineke, Purdue Uni-
Calculus versity Fort Wayne, IN, Bjarne
Toft, University of Southern Den-
PRACTIS mark, Odense, Denmark, and
Robin J. Wilson, The Open Uni-
Precalculus Review and
versity, Milton Keynes, England
Calculus Topics in Sync
Diana McGinnis, Georgia State This book gives an engaging
University, Atlanta, Georgia, and overview of the advances in
Marilyn Reba, Georgia State Uni- graph theory during the 20th
versity, Atlanta, Georgia century. The authors, all subject experts, considered hun-
dreds of original papers, picking out key developments and
PRACTIS (Precalculus Review and some of the notable milestones in the subject. This carefully
Calculus Topics In Sync) provides researched volume leads the reader from the struggles of the
just-in-time resources to support early pioneers, through the rapid expansion of the subject
Calculus I students. This volume in the 1960s and 1970s, up to the present day, with graph
contains worksheets which may theory now a part of mainstream mathematics.
be assigned to students for targeted remediation of the After an opening chapter giving an overview of graph
necessary material to be successful in Calculus. theory and its legacy from the 18th and 19th centuries, the
Prepared by two highly-experienced instructors, the book is organized thematically into seven chapters, each
twenty-eight worksheets cover topics broadly divided into covering the developments made in a specified area. Topics
four categories: covered in these chapters include map colorings, planarity,
• limits, Hamiltonian graphs, matchings, extremal graph theory, and
• differentiation, complexity. Each chapter is supplemented with copious
• applications of derivatives, endnotes, providing additional comments, bibliographic
• integration. details, and further context.
In addition, each worksheet comes with an answer key. Written as an accessible account of the history of the
The convenience of the worksheets is enhanced by subject, this book is suitable not only for graph theorists,
• a table showing how the resources align with pop- but also for anyone interested in learning about the history
ular Calculus textbooks, of this fascinating subject. Some basic knowledge of linear
• guidelines and suggestions for using the work- algebra and group theory would be helpful, but is certainly
sheets, not essential.
• a handy table summarizing the topics of each
worksheet. This item will also be of interest to those working in general
Presentation slides, covering the precalculus/calculus interest.
topics from each worksheet, are also available for use by Spectrum, Volume 108
those instructors who wish to present these topics in the July 2025, approximately 158 pages, Softcover, ISBN: 978-
classroom, or who want to share them with students on 1-4704-6431-8, 2020 Mathematics Subject Classification:
their learning management system. These can be found at 01A60, 05C10, 05C15, 05C20, 05C35, 05C45, 05C50,
www.ams.org/bookpages/clrm-76. 05C55, 05C70, 05C85, List US$69, AMS Individual mem-
Classroom Resource Materials, Volume 76 ber US$51.75, AMS Institutional member US$55.20, MAA
approximately 142 pages, Softcover, 2020 Mathematics members US$51.75, Order code SPEC/108
Subject Classification: 97U30, 26–01, 26A03, List US$0, bookstore.ams.org/spec-108
Order code CLRM/76
bookstore.ams.org/clrm-76

August 2025 Notices of the American Mathematical Society 789

 NEW BOOKS

General Interest New in Contemporary

A Mathematician’s Angle Mathematics
on School Math
Essays from the First 25
Years of the MAA Online Analysis
Column, Devlin’s Angle,
1996 –2020
Classical Hypergeometric
Keith Devlin, Stanford Univer-
sity, CA C ONTEMPORARY Functions and
M ATHEMATICS Generalizations
First published in January 1996,
8 18

Howard S. Cohl, National In-

Devlin’s Angle is a popular on- Classical Hypergeometric stitute of Standards and Technol-
line monthly feature on the Functions and ogy, Gaithersburg, MD, Roberto
MAA Math Values website. In Generalizations S. Costas-Santos, Universidad
this book, Keith Devlin has celebrated the first quarter Howard S. Cohl Loyola Andalucía, Sevilla, Spain,
century of the MAA’s web presence by curating a collection Roberto S. Costas-Santos
and Robert S. Maier, University
Robert S. Maier
of 46 of the 288 posts from that period, chosen for their Editors of Arizona, Tucson, AZ, Editors
relevance to K–12 mathematics teaching.
The posts are organized into nine themed chapters, each This is the first volume of a
beginning with its own introduction regarding the history two-volume collection of recent
and nature of the posts presented. Topics covered include research results related to hypergeometric functions. The
the teaching of multiplication, teaching for conceptual second volume (Contemporary Mathematics, Volume
understanding, and a discussion of mathematical creativity. 819) is titled Applications and q-Extensions of Hypergeomet-
The book closes with a final chapter touching on teaching ric Functions. This volume contains the proceedings of a
at the college level. Due to the nature of mathematics, many minisymposium and two AMS special sessions in three

ric, 𝑞 ​-series and Generalizations at the 16th International

of the columns contain observations that remain relevant conferences: Minisymposium on All Things Hypergeomet-
in the present day.
Devlin’s lively, conversational style is encapsulated in Symposium on Orthogonal Polynomials, Special Functions
this informative and thought-provoking collection. It will and Applications (OPSFA-16), June 13–17, 2022, Centre
appeal to mathematics teachers at all levels, as well as any- de Recherches Mathématiques, Montréal, Québec, Canada;

𝑞 ​-series at the 2022 AMS Fall Western Sectional Meeting,

one interested in mathematics education at the K–12 level. AMS Special Session on Hypergeometric Functions and

This item will also be of interest to those working in math October 22–23, 2022, University of Utah, Salt Lake City,

Functions, 𝑞 ​-series and Generalizations, at the 2023 AMS

education. Utah; and the AMS Special Session on Hypergeometric
Spectrum, Volume 107
July 2025, 272 pages, Softcover, ISBN: 978-1-4704-7933-6, Spring Eastern Virtual Sectional Meeting, April 1–2, 2023.
2020 Mathematics Subject Classification: 00–XX, 97–00, List This book provides a sampling of current mathematical
US$39, AMS Individual member US$29.25, AMS Institu- research related to the Gauss hypergeometric function, and
tional member US$31.20, MAA members US$29.25, Order as well, its immediate generalizations and extensions. This
code SPEC/107 includes the generalized hypergeometric functions that
originated with Kummer, as well as such classical special
bookstore.ams.org/spec-107 functions as Lamé and Heun functions. It also includes
certain functions relevant to algebraic geometry, such as
hypergeometric functions over finite fields. All research
articles come with extensive bibliographies and can serve
as entry points to the current literature.

790 Notices of the American Mathematical Society Volume 72, Number 7

 NEW BOOKS

Contemporary Mathematics, Volume 818 Contemporary Mathematics, Volume 819

April 2025, 288 pages, Softcover, ISBN: 978-1-4704-7653- July 2025, 299 pages, Softcover, ISBN: 978-1-4704-7654-
3, LC 2024039112, 2020 Mathematics Subject Classification: 0, 2020 Mathematics Subject Classification: 33C05, 33C20,
33C05, 33C20, 33C45, 33C90, 33E05, 11G20, 11G55, 33C45, 33C90, 33D15, 33E05, 14H52, 81Q05, List US$135,
14H52, List US$135, AMS members US$108, MAA mem- AMS members US$108, MAA members US$121.50, Order
bers US$121.50, Order code CONM/818 code CONM/819
bookstore.ams.org/conm/818 bookstore.ams.org/conm-819

q ​-Extensions of
Applications and
CONTEMPORARY
MATHEMATICS
Mathematical Physics
819
Hypergeometric Functions
Howard S. Cohl, National In-
Applications and From Representation
stitute of Standards and Technol-
q-Extensions of
Hypergeometric ogy, Gaithersburg, MD, Roberto CONTEMPORARY Theory to Mathematical
M
ATHEMATICS Physics and Back
Functions S. Costas-Santos, Universidad 817

Loyola Andalucía, Sevilla, Spain, Mikhail Khovanov, Johns Hop-

Howard S. Cohl
Roberto S. Costas-Santos
and Robert S. Maier, University From Representation kins University, Baltimore, MD,
Robert S. Maier
Editors of Arizona, Tucson, AZ, Editors Theory to Mathematical Joshua Sussan, CUNY Medgar
Physics and Back Evers, Brooklyn, NY, and Anton
This is the second volume of a Mikhail Khovanov Zeitlin, Georgia Institute of Tech-
two-volume collection of recent Joshua Sussan
nology, Atlanta, GA, Editors
Anton Zeitlin
research results related to hypergeometric functions. The Editors

first volume (Contemporary Mathematics, Volume 818) is This volume is a proceedings of a

titled Classical Hypergeometric Functions and Generalizations. workshop at the Simons Center
This volume contains the proceedings of a minisympo- for Geometry and Physics from

Minisymposium on All Things Hypergeometric, 𝑞 ​-series

sium and two AMS special sessions in three conferences: May 30– June 4, 2022. The workshop highlighted progress
in the areas of vertex operator algebras, conformal field
and Generalizations at the 16th International Symposium theory, categorification, low dimensional topology and
on Orthogonal Polynomials, Special Functions and Ap- representation theory of affine Lie algebras, loop groups,
plications (OPSFA-16), June 13–17, 2022, Centre de Re- and quantum groups.

Special Session on Hypergeometric Functions and 𝑞 ​-series

cherches Mathématiques, Montréal, Québec, Canada; AMS In the past 40 years, string theory gave rise to the math-
ematical theory of vertex operator algebras, which led to
at the 2022 AMS Fall Western Sectional Meeting, October the construction of representations of affine Lie algebras
22–23, 2022, University of Utah, Salt Lake City, Utah; and and the Moonshine module of the Monster group. These

𝑞 ​-series and Generalizations, at the 2023 AMS Spring East-

the AMS Special Session on Hypergeometric Functions, mathematical constructions have in turn led to ideas about
3-dimensional quantum gravity. In another direction, the
ern Virtual Sectional Meeting, April 1–2, 2023. discovery of the Jones polynomial led to a physical con-
This book provides a sampling of recent research on struction of 3-dimensional topological quantum field the-
applications of classical hypergeometric and related special ories (TQFTs), which in turn advanced many mathematical

where, and on 𝑞 ​-extensions of hypergeometric functions

functions to problems in mathematical physics and else- developments in quantum groups and low dimensional

and other topics in 𝑞-​calculus. The problems in mathemat-

topology.
Louis Crane and Igor Frenkel introduced the categorifi-
ical physics include the explicit integration of the station- cation program with the goal of upgrading 3-dimensional
ary Schrödinger equation with many potentials, and the TQFTs coming from representation theory of quantum

mass in terms of elliptic integrals. The 𝑞 ​-calculus topics

computation of the gravitational potential of an ellipsoidal groups to 4-dimensional TQFTs. This idea gave rise to the
development of link homologies constructed from repre-
include a study of Ramanujan’s q-continued fractions, new sentation-theoretic, algebraic-geometric, combinatorial,
q-identities, and important limits of basic hypergeometric and physical structures.
orthogonal polynomials. All research articles come with Articles in this volume present both classical and new
extensive bibliographies and can serve as entry points to results related to these topics. They will be interesting to
the current literature. researchers and graduate students working in mathematical
aspects of modern quantum field theory.

August 2025 Notices of the American Mathematical Society 791

 NEW BOOKS

Contemporary Mathematics, Volume 817 Rigid Local Systems and Sporadic Simple
June 2025, 377 pages, Softcover, ISBN: 978-1-4704-7339- Groups
6, 2020 Mathematics Subject Classification: 20D08, 17B37, Nicholas Michael Katz, Princeton University, New Jersey,
13F60, 57K20, 83C45, 18G70, 15B48, 20H99, 57K16, Antonio Rojas-León, Universidad de Sevilla, Spain, and
57K18, List US$135, AMS members US$108, MAA mem- Pham Huu Tiep, Rutgers University, Piscataway, New Jersey
bers US$121.50, Order code CONM/817
Memoirs of the American Mathematical Society, Volume
bookstore.ams.org/conm-817 308, Number 1559
May 2025, 185 pages, Softcover, ISBN: 978-1-4704-
New in Memoirs 7342-6, 2020 Mathematics Subject Classification: 11T23;
20C15, 20C33, 20C34, 20D06, 20D08, List US$85, AMS
of the AMS members US$68, MAA members US$76.50, Order code
MEMO/308/1559
bookstore.ams.org/memo-308-1559
Algebra and Algebraic
Geometry Analysis
The AKSZ Construction in Derived Stability in Gagliardo–Nirenberg–Sobolev
Algebraic Geometry as an Extended Inequalities: Flows, Regularity and the Entropy
Topological Field Theory Method
Damien Calaque, Université Montpellier, France, Rune Matteo Bonforte, Universidad Autónoma de Madrid, Spain,
Haugseng, Norwegian University of Science and Technology, and Institute of Mathematical Sciences, Madrid, Spain, Jean
Trondheim, Norway, and Claudia Scheimbauer, Technische Dolbeault, Université Paris-Dauphine, France, Bruno Naza-
Universität München, Munich, Germany ret, Université Paris, France, and Mokaplan Inria, Paris, France,
Memoirs of the American Mathematical Society, Volume and Nikita Simonov, Sorbonne Université, Paris, France
308, Number 1555 This item will also be of interest to those working in differential
May 2025, 173 pages, Softcover, ISBN: 978-1-4704- equations.
7272-6, 2020 Mathematics Subject Classification: 14A30,
Memoirs of the American Mathematical Society, Volume
18N65, 57R56; 14D21, 14J42, 53D30, List US$85, AMS
308, Number 1554
members US$68, MAA members US$76.50, Order code
May 2025, 166 pages, Softcover, ISBN: 978-1-4704-7181-
MEMO/308/1555
1, 2020 Mathematics Subject Classification: 26D10, 46E35,
bookstore.ams.org/memo-308-1555 35K55; 49J40, 35B40, 49K20, 49K30, 35J20, List US$85,

Differentials for Calabi-Yau-𝕏 Categories

AMS members US$68, MAA members US$76.50, Order
q-Stability Conditions via q-Quadratic code MEMO/308/1554

Akishi Ikeda, Josai University, Saitama, Japan, and Yu Qiu, bookstore.ams.org/memo-308-1554

Tsinghua University, Beijing, People’s Republic of China, and
Beijing Institute of Mathematical Sciences and Applications,
People’s Republic of China
Differential Equations

Invariant Φ​4 ​-Measure and Associated Flow on ℝ3

Memoirs of the American Mathematical Society, Volume
308, Number 1557
Construction of a Non-Gaussian and Rotation-
May 2025, 92 pages, Softcover, ISBN: 978-1-4704-7312-
9, 2020 Mathematics Subject Classification: 18G80, 30F30;
Through Stochastic Quantization
16E10, 16G20, List US$85, AMS members US$68, MAA Sergio Albeverio, University of Bonn, Germany, and Seiichiro
members US$76.50, Order code MEMO/308/1557 Kusuoka, Kyoto University, Japan

bookstore.ams.org/memo-308-1557 This item will also be of interest to those working in mathemat-

ical physics.
Memoirs of the American Mathematical Society, Volume
308, Number 1558
May 2025, 114 pages, Softcover, ISBN: 978-1-4704-
7318-1, 2020 Mathematics Subject Classification: 81S20;

792 Notices of the American Mathematical Society Volume 72, Number 7

 NEW BOOKS

81T08, 60H15, 35Q40, 35R60, 35K58, List US$85, AMS A publication of the European Mathematical Society (EMS). Distributed
members US$68, MAA members US$76.50, Order code within the Americas by the American Mathematical Society.
MEMO/308/1558 Memoirs of the European Mathematical Society, Volume
bookstore.ams.org/memo-308-1558 19
March 2025, 314 pages, Softcover, ISBN: 978-3-98547-088-
Regularity for Orlicz Phase Problems 4, 2020 Mathematics Subject Classification: 13J07; 14F30,
Sumiya Baasandorj, Seoul National University, Korea, and 14G22, List US$75, AMS members US$60, Order code
University of Ottawa, Ontario, Canada, and Sun-Sig Byun, EMSMEM/19
Seoul National University, Korea bookstore.ams.org/emsmem-19
Memoirs of the American Mathematical Society, Volume
308, Number 1556
May 2025, 131 pages, Softcover, ISBN: 978-1-4704-7304-
Differential Equations
4, 2020 Mathematics Subject Classification: 49N60; 35B27,
35B65, 35J20, List US$85, AMS members US$68, MAA FBI Transform in Gevrey
members US$76.50, Order code MEMO/308/1556 Classes and Anosov Flows
bookstore.ams.org/memo-308-1556 Yannick Guedes Bonthonneau,
LAGA, CNRS, Université Sorbonne

New AMS-Distributed Paris Nord, France, and Malo

Jézéquel, CNRS, Université Brest

Publications Bretagne Occidentale, France

This monograph has a triple
purpose. The first chapter is ded-
Algebra and Algebraic icated to the study of semi-classi-
cal pseudo-differential operators
Geometry whose symbols are Gevrey in all
variables. In the second one, the theory of I-lagrangian func-
tional spaces is presented, in the case that the lagrangian is
Almost Coherent Modules not a compactly supported deformation of the cotangent.
and Almost Coherent The relation with the operators presented before is also
Sheaves analyzed. In the last chapter, the theory is applied to the
Bogdan Zavyalov, Princeton Uni- study of Gevrey Anosov flows: in particular, the growth of
versity and Institute for Advanced their dynamical determinants is investigated.
Study, NJ
This item will also be of interest to those working in probability
The author extends the theory of and statistics and number theory.
almost coherent modules that
A publication of the Société Mathématique de France, Marseilles (SMF),
was introduced in Almost Ring distributed by the AMS in the U.S., Canada, and Mexico. Orders from
Theory, by Gabber and Ramero other countries should be sent to the SMF. Members of the SMF receive
(2003). Then he globalizes it by a 30% discount from list.
developing a new theory of al- Astérisque, Number 456
most coherent sheaves on schemes and on a class of ‘nice’ April 2025, 233 pages, Softcover, ISBN: 978-2-37905-209-
formal schemes. He shows that these sheaves satisfy many 5, 2020 Mathematics Subject Classification: 37C30, 35A22,
properties similar to usual coherent sheaves, i.e., the amost List US$89, AMS members US$71.20, Order code AST/456
proper mapping theorem, the formal GAGA, etc. He also
bookstore.ams.org/ast-456
image functor 𝑓 !​ and verifies its properties. Lastly, he studies
constructs an almost version of the Grothendieck twisted

sheaves of 𝑝​-adic nearby cycles on admissible formal mod-

els of rigid-analytic varieties and shows that these sheaves
provide examples of almost coherent sheaves. This gives
a new proof of the finiteness result for étale cohomology

Scholze’s 𝑝​-adic Hodge theory for rigid-analytic varieties

of proper rigid-analytic varieties obtained before in Peter

(2013).

August 2025 Notices of the American Mathematical Society 793

 NEW BOOKS

Geometry and Topology Geometric Methods in

Group Theory
Papers Dedicated to Ruth
Global Nonlinear Stability Charney
of Minkowski Space for Rachel Skipper, University of
Spacelike-Characteristic Utah, Salt Lake City, UT, and
Initial Data Indira Chatterji, Laboratoire
O. Graf, Université Grenoble Dieudonné, Université Cote d’Azur,
Alpes, France Nice, France, Editors
In this book the author proves This volume contains the pro-
the nonlinear global stability of ceedings of one of the first
Minkowski space in the frame- hybrid conferences in geometric
work of the spatial-characteristic group theory, which brought together two conferences
Cauchy problem for the Einstein cancelled following the COVID-19 pandemic. One was in
equations in vacuum. The spa- honor of Ruth Charney, and the other was a workshop on
tial-characteristic initial data are prescribed on a 3-disk generalizations of hyperbolicity, a theme that benefited
and on the future-complete zero hypersurface emanating greatly from the influence of Ruth Charney.
from the boundary of this disk. The author’s result extends The articles in this volume reflect the organizers’ goals
the original result proved by Christodoulou and Klainer- of making geometric group theory accessible and open to
man for which the initial data are prescribed on a spatial all. This volume can be seen as a capsule of questions and
hyperplane. answers that this part of the community was asking itself
The proof is based on the vector field method and the in the summer of 2021.
bootstrap argument introduced in Christodoulou-Klain-
This item will also be of interest to those working in number
erman. The main novelty is the introduction and control
theory.
of new geometric constructions adapted to the character-
istic-space framework. In particular, the author uses ver- A publication of the Société Mathématique de France, Marseilles (SMF),
distributed by the AMS in the U.S., Canada, and Mexico. Orders from
tex-prescribed light cones, boundary-prescribed maximal other countries should be sent to the SMF. Members of the SMF receive
spatial hypersurfaces, and global harmonic coordinates on a 30% discount from list.
Riemannian 3-disks.
Séminaires et Congrès, Number 34
This item will also be of interest to those working in differential April 2025, 316 pages, Softcover, ISBN: 978-2-85629-998-2,
equations. 2020 Mathematics Subject Classification: 20F65, 57M07,
A publication of the Société Mathématique de France, Marseilles (SMF), 57M60, List US$98, AMS members US$78.40, Order code
distributed by the AMS in the U.S., Canada, and Mexico. Orders from SECO/34
other countries should be sent to the SMF. Members of the SMF receive
a 30% discount from list. bookstore.ams.org/seco-34

Mémoires de la Société Mathématique de France, Num-

ber 184
April 2025, 212 pages, Softcover, ISBN: 978-2-37905-207-
1, 2020 Mathematics Subject Classification: 83C05, 35L05,
List US$108, AMS members US$86.40, Order code SMF-
MEM/184
bookstore.ams.org/smfmem-184

794 Notices of the American Mathematical Society Volume 72, Number 7

 Meetings & Conferences of the AMS
August Table of Contents

The Meetings and Conferences section of the Notices gives

information on all AMS meetings and conferences ap-
proved by press time for this issue. Please refer to the page Meetings in this Issue
numbers cited on this page for more detailed information
on each event. Paid meeting registration is required to submit 2025
an abstract to a sectional meeting.
August 23–24 Denver, Colorado p. 796
Invited Speakers and Special Sessions are listed as soon
as they are approved by the cognizant program committee; October 3–5 New Orleans, Louisiana p. 797
the codes listed are needed for electronic abstract sub- October 18–19 St. Louis, Missouri p. 799
mission. For some meetings the list may be incomplete. October 25–26 Fall Eastern Virtual p. 801
Information in this issue may be dated.
The most up-to-date meeting and conference informa- 2026
tion can be found online at www.ams.org/meetings.
January 4–7 Washington, DC
Important Information About AMS Meetings: Potential
organizers, speakers, and hosts should refer to https:// (JMM 2026) p. 802
www.ams.org/meetings/meetings-general for general March 7–8 Boise, Idaho p. 803
information regarding participation in AMS meetings and March 28–29 Boston, Massachusetts p. 803
conferences. March 28–29 Savannah, Georgia p. 803
Abstracts: Speakers should submit abstracts on the April 18–19 Fargo, North Dakota p. 804
easy-to-use interactive Web form. No knowledge of LaTeX
October 3–4 Washington, DC p. 804
is necessary to submit an electronic form, although those
who use LaTeX may submit abstracts with such coding, and
all math displays and similarly coded material (such as ac-
cent marks in text) must be typeset in LaTeX. Visit www.ams
The AMS strives to ensure that participants in its activities
.org/cgi-bin/abstracts/abstract.pl . Questions
enjoy a welcoming environment. Please see our full Policy
about abstracts may be sent to [email protected]. Close
on a Welcoming Environment at https://www.ams
attention should be paid to specified deadlines in this issue.
.org/welcoming-environment-policy.
Unfortunately, late abstracts cannot be accommodated.

Associate Secretaries of the AMS

Central Section: Betsy Stovall, University of Wisconsin–
Madison, 480 Lincoln Drive, Madison, WI 53706; email:
[email protected]; telephone: (608) 262-2933.
Eastern Section: Benjamin Lotto, Department of Math-
ematics and Statistics, Vassar College, Poughkeepsie, NY
12604-0004; email: [email protected]; telephone: (845)
437-7437.
Southeastern Section: Brian D. Boe, Department of Math-
ematics, University of Georgia, 220 D W Brooks Drive,
Athens, GA 30602-7403; email: [email protected];
telephone: (706) 542-2547.
Western Section: Michelle Manes, University of Hawaii,
Department of Mathematics, 2565 McCarthy Mall, Keller
401A, Honolulu, HI 96822; email: [email protected];
telephone: (808) 956-4679.

August 2025 Notices of the American Mathematical Society 795

 MEETINGS & CONFERENCES

Meetings & Conferences

of the AMS
IMPORTANT information regarding meetings programs: AMS Sectional Meeting programs do not appear in the print
version of the Notices. However, comprehensive and continually updated meeting and program information with links
to the abstract for each talk can be found on the AMS website. See https://www.ams.org/meetings.
Final programs for Sectional Meetings will be archived on the AMS website accessible from the stated URL.
New: Sectional Meetings Require Registration to Submit Abstracts. In an effort to spread the cost of the sectional
meetings more equitably among all who attend and hence help keep registration fees low, starting with the 2020 fall
sectional meetings, you must be registered for a sectional meeting in order to submit an abstract for that meeting.
You will be prompted to register on the Abstracts Submission Page. In the event that your abstract is not accepted or
you have to cancel your participation in the program due to unforeseen circumstances, your registration fee will be
reimbursed.

Denver, Colorado
Denver, CO - University of Denver
August 23–24, 2025 Program first available on AMS website: To be announced
Saturday – Sunday Issue of Abstracts: Volume 46, Issue 4

Meeting #1209 Deadlines

For organizers: Expired
Western Section
For abstracts: Expired
Associate Secretary for the AMS: Michelle Ann Manes

The scientific information listed below may be dated. For the latest information, see https://www.ams.org/amsmtgs
/sectional.html.

Invited Addresses
Alexandra Florea, University of California Irvine, Title to be announced.
Connor Mooney, University of California Irvine, Title to be announced.
Bobby L. E. Wilson, University of Washington, Title to be announced.

Special Sessions
If you are volunteering to speak in a Special Session, you should send your abstract as early as possible via the abstract submission
form found at https://www.ams.org/cgi-bin/abstracts/abstract.pl.

Algebraic Logic, Nikolaos Galatos, University of Denver, and Peter Jipsen, Chapman University, California, USA.
Algebras and Algorithms, Peter Mayr, Keith Kearnes, and Charlotte Aten, University of Colorado Boulder.
Analytic and Probabilistic Themes in Combinatorics, Paul Horn and Mei Yin, University of Denver.
Analytic Number Theory and Related Topics, Alexandra Florea, University of California Irvine, and Anurag Sahay, Uni-
versity of Rochester.

796 Notices of the American Mathematical Society Volume 72, Number 7

 MEETINGS & CONFERENCES

Chromatic Homotopy Theory, Agnes Beaudry, University of Colorado Boulder, XiaoLin Danny Shi, University of Wash-
ington, and Zhouli Xu, University of California, San Diego.
Geometric methods in Combinatorics, Jianping Pan, Arizona State University, and Esther Banaian, University of Cali-
fornia, Riverside.
Geometry, Integrability, Symmetry, and Physics, Andrew R. Linshaw, University of Denver, Sasha Tsymbaliuk, Purdue
University, and Shashank Kanade, University of Denver.
Logarithmic geometry and Gromov–Witten theory, Leo Sherwood Herr, Virginia Tech.
Nonassociative Algebraic Structures, Petr Vojtechovsky and Sujoy Mukherjee, University of Denver.
Nonlinear Elliptic PDEs and Geometry associated with Invited Address by, Connor Mooney, University of California Irvine,
and Ravi Shankar, Princeton University.
Nonlinear Waves and Integrable Systems, Barbara Prinari and Nicholas James Ossi, University at Buffalo.
Recent Advances in Dynamical Systems, Anh N Le and Ronnie Pavlov, University of Denver.
Recent Advances in Nonlinear PDEs and Geometry, Fernando Charro, Wayne State University, and Padi Fuster Aguilera,
University of Colorado at Boulder.
Recent Developments in Areas Influenced by Ramanujan’s Mathematics, Shashank Kanade, University of Denver, Aritram
Dhar, IISER Mohali, and Avi N Mukhopadhyay and Sroyon Sengupta, University of Florida.
Recent Progress in Frame Theory and Harmonic Analysis, Dongwei Chen, Emily J King, and Clayton Shonkwiler, Colo-
rado State University.
Recent Progress in PDE Models of Incompressible Fluids, Neel Patel, University of Maine, and Siddhant Agrawal, University
of Colorado Boulder.
Representations of Finite and Algebraic Groups, Daniel K Nakano, University of Georgia, John McHugh, University of
California Santa Cruz, and Mandi A. Schaeffer Fry, University of Denver.
Solitons, Turbulence and Singularities in Fluid Dynamics, Optics, and Plasmas, Pavel M Lushnikov, University of New Mexico.

New Orleans, Louisiana

Tulane University
October 3–5, 2025 Program first available on AMS website: To be announced
Friday – Sunday Issue of Abstracts: Volume 46, Issue 4

Meeting #1210 Deadlines

For organizers: Expired
Southeastern Section
For abstracts: August 12, 2025
Associate Secretary for the AMS: Brian D. Boe

The scientific information listed below may be dated. For the latest information, see https://www.ams.org/amsmtgs
/sectional.html.

Invited Addresses
Marcelo Mendes Disconzi, Vanderbilt University, To be announced.
Boris Hanin, Princeton University, To be announced.
Akos Magyar, University of Georgia, Exponential Sums, Polynomial Ergodic Theorems and Nilpotent Groups.

Special Sessions
If you are volunteering to speak in a Special Session, you should send your abstract as early as possible via the abstract submission
form found at https://www.ams.org/cgi-bin/abstracts/abstract.pl.

Actuarial Mathematics and Actuarial Education (Code: SS 12A), Jiacheng Cai, Salisbury University, and Fang Yang,
Georgia State University.
Advances and challenges in the study of nonlinear PDEs (Code: SS 22A), Prerona Dutta, Xavier University of Louisiana,
and Son Tu, Michigan State University.
Advances in Applied Topology and Topological Data Analysis (Code: SS 24A), Rafal Komendarczyk, Tulane University, and
Sushovan Majhi, George Washington University.

August 2025 Notices of the American Mathematical Society 797

 MEETINGS & CONFERENCES

Advances in High and Infinite Dimensional Stochastic Analysis (Code: SS 5A), Nathan Glatt-Holtz and Giulia Carigi,
Indiana University.
Advances in Integration Methods (Code: SS 18A), Kristina VanDusen, Southeastern Louisiana University, and Zachary
Patrick Bradshaw, Advanced Processing Branch, Naval Surface Warfare Center, Panama City Division.
Advances in mathematical modeling and analysis of biological dynamics (Code: SS 49A), Amy Veprauskas and Azmy S
Ackleh, University of Louisiana at Lafayette.
Algebraic combinatorics and combinatorial commutative algebra (Code: SS 34A), Sean Grate, Auburn University, Anton
Dochtermann, Texas State University, and Thiago Holleben, Dalhousie University.
Analysis of Mathematical Models in Life Sciences (Code: SS 14A), Xiang-Sheng Wang, University of Louisiana at Lafayette,
and Guihong Fan, Columbus State University.
Automorphic Forms and Representation Theory: current trends (Code: SS 39A), Karol Koziol, Baruch College, CUNY, and
Anantharam Raghuram, Fordham University.
Coarse Geometry of Nonpositively Curved Complexes and Groups with an Emphasis on Coxeter and Artin Groups (Code: SS
45A), Annette Ulrike Karrer, The Ohio State University, and Pallavi Dani, Louisiana State University.
Combinatorics and Geometry Related to Representation Theory (Code: SS 37A), William Erickson and Markus Hunziker,
Baylor University.
Commutative Algebra (Code: SS 6A), Kuei-Nuan Lin, Pennsylvania State University, Greater Allegheny, and Dipendranath
Mahato and Tai Huy Ha, Tulane University.
Computational Topology and Geometry in Data Science (Code: SS 35A), Sayoni Chakraborty, The University of Texas at
Dallas, Baris Coskunuzer, UT Dallas, and Surbhi Kumar and Brighton Nuwagira, The University of Texas at Dallas.
Cryptography, Coding Theory, and their Intersections with Algebraic Geometry & Combinatorics (Code: SS 47A), Emily Mc-
Millon, Rice University, Mahir Can, Tulane University, and Henry Chimal-Dzul, University of Texas at San Antonio.
Discontinuous Galerkin Finite Element Methods for Partial Differential Equations: Theory and applications (Code: SS 17A),
Mahboub Baccouch, University of Nebraska At Omaha, and Slimane Adjerid, Virginia Tech.
Engaging Mathematics Learners with AI (Code: SS 13A), Michael Todd Edwards, Miami University of Ohio, Carlos A
Lopez Gonzalez, Technology Educator Alliance, and Zheng Yang, Sichuan University - Pittsburgh Institute.
Enhancing Mathematical Accuracy: Precision in Differential Equations Through Block Hybrid Methods with Applications from
Cancer and Addiction Models (Code: SS 50A), Maduabuchi Gabriel Orakwelu, Clinton School of Public Service, University
of Arkansas.
Ergodic Theory and Discrete Analysis (Code: SS 2A), Neil Lyall and Tomasz Szarek, University of Georgia.
Experimental Mathematics (Code: SS 10A), Armin Straub, University of South Alabama, Luis A. Medina, University of
Puerto Rico, and Brandt Kronholm, University of Texas Rio Grande Valley.
Finite Rings and Their Applications (Code: SS 7A), Steve Szabo, Eastern Kentucky University.
Fractal Geometry and Dynamical Systems (Code: SS 1A), Mrinal Kanti Roychowdhury, School of Mathematical and
Statistical Sciences, University of Texas Rio Grande Valley, and William Ott, University of Houston.
Function spaces and Operators in Harmonic Analysis and their applications (Code: SS 38A), Liding Yao, The Ohio State
University, Chian Yeong Chuah, Ohio State University, and Jan Lang, The Ohio State University.
Homological and categorical methods in commutative algebra (Code: SS 11A), David A Jorgensen, University of Texas,
Arlington, and Katie Grebel, University of Texas at Arlington.
Homological methods in commutative algebra and algebraic geometry (Code: SS 30A), Alessandra Costantini and Tái Huy
Há, Tulane University.
Hopf algebras and tensor categories (Code: SS 20A), Xingting Wang and Siu-Hung Ng, Louisiana State University, and
Qing Zhang, UC Santa Barbara.
Hyperplane Arrangements and Polytopes (Code: SS 8A), Galen Dorpalen-Barry, Texas A&M, and Christin Bibby, Loui-
siana State University.
Integrability and randomness (Code: SS 19A), Katerina Gkogkou, Kenneth McLaughlin, and Guido Mazzuca, Tulane
University.
Interactions, Discrepancies, Approximations: From Energy Optimization to Dynamics (Code: SS 41A), Ihsan Topaloglu,
Virginia Commonwealth University, and Ryan W Matzke, Vanderbilt University.
Linearity in Tropical geometry (Code: SS 40A), Netanel Friedenberg and Kalina Mincheva, Tulane University.
Mathematical Modeling in Ecology and Epidemiology (Code: SS 46A), Zhisheng Shuai, University of Central Florida, and
King-Yeung Lam, The Ohio State University.
Mathematical modeling of viscous flows (Code: SS 26A), Hongfei Chen and Adnan Morshed, Tulane University.

798 Notices of the American Mathematical Society Volume 72, Number 7

 MEETINGS & CONFERENCES

Modular forms in combinatorics and number theory (Code: SS 3A), Kalani Thalagoda, Tulane University, Michael Hanson,
University of North Texas, and Olivia Beckwith, Tulane University.
Orthogonal Polynomials and Special Functions: Asymptotics and Applications (Code: SS 29A), Andrei Martinez Finkelshtein,
Baylor University, Kenneth McLaughlin, Tulane University, and Brian Simanek, Baylor University.
p-adic Valuations and Related Topics (Code: SS 27A), Jane H. Long, Stephen F. Austin State University, Olena Kozhush-
kina, Ursinus College, and Justin G. Trulen, Kentucky Wesleyan College.
Partial differential equations at the intersection of gravity, fluids, and waves (Code: SS 21A), Marcelo Mendes Disconzi,
Vanderbilt University.
Quantitative Topology and Analysis (Code: SS 42A), Piotr Hajlasz and Armin Schikorra, University of Pittsburgh.
Random structures in combinatorics and number theory (Code: SS 44A), Charles D Burnette, Xavier University of Louisiana,
and Diego Villamizar, Tulane University.
Recent advances at the interface of dynamics and fluids (Code: SS 31A), Vincent R Martinez, CUNY Hunter College &
Graduate Center, Kyle Liss, Duke University, and Samuel Punshon-Smith, Tulane University.
Recent Advances in Control and Inverse Problems for Partial Differential Equations (Code: SS 16A), Shitao Liu, Clemson
University, and Louis Roder Tcheugoue Tebou, Florida International University.
Recent Advances in Graph Theory (Code: SS 32A), Lina Li, University of Mississippi, and Emily Heath, California State
Polytechnic University, Pomona.
Recent Advances in Model-Based and Data-Driven Methods for Inverse Problems and Imaging (Code: SS 36A), Yiran Wang,
Emory University, and Yimin Zhong, Auburn University.
Recent Advances in Numerical Methods for Ordinary/Partial Differential Equations and Their Applications (Code: SS 28A),
Thoa Thi Kim Thieu, The University of Texas Rio Grande Valley, Liet Vo, University of Texas Rio Grande Valley, and Khoa
Vo, Texas Tech University.
Recent Advances in the Mathematics of Cryptography (Code: SS 23A), Sarah Arpin, Virginia Polytechnic Institute and State
University, Ryann Cartor, Clemson University, and Jason LeGrow, Virginia Tech.
Recent advances on the studies of qualitative and quantitative properties of solutions of PDEs (Code: SS 4A), Jiuyi Zhu, Lou-
isiana State University, and Yuanzhen Shao, The University of Alabama.
Recent Theoretical and Numerical Advances in Relativistic Hydrodynamics (Code: SS 15A), Teerthal Patel, Henry Hirvonen,
Marcelo Mendes Disconzi, and Jean-François Paquet, Vanderbilt University.
Singularity formation in fluid dynamics (Code: SS 25A), Xinyu Zhao, New Jersey Institute of Technology, and Jia Shi,
Massachusetts Institute of Technology.
Special session on emerging interactions between representation theory, geometry, and combinatorics (Code: SS 48A), Daniel
Orr and Milo James Bechtloff Weising, Virginia Tech.
Statistics and AI in Precision Health and Biology (Code: SS 33A), Farzana Nasrin, University of Hawaii at Manoa, and G
M Fahad Bin Mostafa, Arizona State University.
Syzygies and Local Cohomology in Commutative Algebra (Code: SS 43A), Jiamin Li, University of Oklahoma, Michael
Perlman, University of Minnesota, and Ritvik Ramkumar, Cornell University.
Topics in Combinatorics and Graph Theory (Code: SS 9A), Xiangqian Joseph Zhou, Wright State University, and Rong
Luo, West Virginia University.

St. Louis, Missouri

Saint Louis University
October 18–19, 2025 Program first available on AMS website: To be announced
Saturday – Sunday Issue of Abstracts: Volume 46, Issue 4

Meeting #1211 Deadlines

For organizers: Expired
Central Section
For abstracts: August 26, 2025
Associate Secretary for the AMS: Betsy Stovall

The scientific information listed below may be dated. For the latest information, see https://www.ams.org/amsmtgs
/sectional.html.

August 2025 Notices of the American Mathematical Society 799

 MEETINGS & CONFERENCES

Invited Addresses
Nicole R Looper, University of Illinois Chicago, AMS Invited Address, Nicole Looper, University of Illinois at Chicago.
Pablo Raúl Stinga, Iowa State University, AMS Invited Address, Pablo Raúl Stinga, Iowa State University.

Special Sessions
If you are volunteering to speak in a Special Session, you should send your abstract as early as possible via the abstract submission
form found at https://www.ams.org/cgi-bin/abstracts/abstract.pl.

Algebraic and Arithmetic Dynamics I (Code: SS 16A), Nicole R Looper, University of Illinois Chicago, and Benjamin A
Hutz, St. Louis University.
Algorithms and Number Theory: Celebrating the Career of Eric Bach I (Code: SS 25A), Jonathan P Sorenson, Butler Uni-
versity, and Andrew Shallue, Illinois Wesleyan University.
Applied and Computational Algebra I (Code: SS 26A), Timothy Duff, University of Missouri.
Categorical, Homological and Combinatorial Methods in Noncommutative Algebra I (Code: SS 2A), Ashish K. Srivastava,
Cody Gilbert, and Greg Marks, Saint Louis University, and Pedro Guil Asensio, University of Murcia, Spain.
Commutative Algebra, Hodge Theory, Singularities I (Code: SS 4A), Uli Walther, Purdue University, Claudia Miller, Syr-
acuse University, and Daniel Bath, University of Leuven.
Commutative Algebra I (Code: SS 20A), Paolo Mantero, University of Arkansas, and Liana Şega, University of Missouri
Kansas City.
Computability I (Code: SS 6A), Wesley Calvert, Southern Illinois University, Johanna N Y Franklin, Hofstra University,
and Matthew Harrison-Trainor, Victoria University of Wellington.
Convexity, Probability, and Analysis I (Code: SS 14A), Michael Roysdon, Case Western Reserve University, Dylan Lang-
harst, Sorbonne University, and Elisabeth M Werner, Case Western Reserve University.
Dynamic Equations on Time Scales: Theory, Methods, and Applications I (Code: SS 10A), Shalmali Bandyopadhyay, Uni-
versity of Tennessee at Martin, Tom Cuchta, Marshall University, and F. Ayca Cetinkaya, The University of Tennessee
Chattanooga.
Finite Rings and Their Applications I (Code: SS 23A), Steve Szabo, Eastern Kentucky University.
Foliations and 3-manifolds I (Code: SS 11A), Nathan M Dunfield, University of Illinois Urbana-Champaign, Steven
Frankel, Washington University in St. Louis, Sarah Dean Rasmussen, University of Illinois Urbana-Champaign, and
Rachel Roberts, Washington University in St Louis.
Geometric Variational Problems and Applications I (Code: SS 32A), William Myers Feldman and Raghavendra Venkatra-
man, University of Utah.
Geometry, Representation Theory and Noncommutative Algebra I (Code: SS 29A), Daniele Rosso, Indiana University
Northwest, and Jason Gaddis, Miami University.
Groups and Representation Theory I (Code: SS 8A), Julianne Rainbolt, Saint Louis University, and Mandi A. Schaeffer
Fry, University of Denver.
Harmonic Analysis and Partial Differential Equations I (Code: SS 17A), Brandon Sweeting, Andrew Walton Green, Henri
Martikainen, and Brett Wick, Washington University in St. Louis.
Harmonic Analysis, Frame Theory, and Tilings I (Code: SS 27A), Darrin M Speegle, Saint Louis University, Marcin M.
Bownik, University of Oregon, and Daniel Freeman and Dorsa Ghoreishi, Saint Louis University.
Harmonic Analysis, Geometric Measure Theory and Fractals I (Code: SS 3A), Alan Chang, Washington University in St.
Louis, Caleb Z Marshall, University of British Columbia, and Jianhui Li, Northwestern University.
Hodge theory, Algebraic cycles, and Arithmetic I (Code: SS 13A), Matt Kerr, Devin Akman, and Wanlin Li, Washington
University in St. Louis.
Homotopy Theory I (Code: SS 12A), David Chan, Teena Meredith Gerhardt, and Maximilien Péroux, Michigan State
University.
Interactions between Geometry, Combinatorics, and Flag Varieties I (Code: SS 9A), Martha Precup, Washington University
in St. Louis, and Ozlem Ugurlu, Saint Louis University.
Inverse Problems I (Code: SS 7A), Robert M. Owczarek, University of New Mexico, and Hanna E. Makaruk, Los Alamos
National Laboratory, Los Alamos, NM.
Mathematical Finance, Environmental Science, and Data Science I (Code: SS 19A), Indranil SenGupta, City University
of New York (CUNY)- Hunter College, Semere Kidane Gebresilasie, Wentworth Institute of Technology, and Shantanu
Awasthi, Missouri Southern State University.

800 Notices of the American Mathematical Society Volume 72, Number 7

 MEETINGS & CONFERENCES

Modeling Experiences for Undergraduate Students in Differential Equations I (Code: SS 18A), Brody Dylan Johnson, Saint
Louis University, Tova Brown, Wisconsin Lutheran College, and Brian Winkel, SIMIODE.
Noncommutative Geometry and Applications I (Code: SS 15A), Yanli Song and Xiang Tang, Washington University in
St.Louis.
Numerical PDEs and Geometry I (Code: SS 24A), Ari Stern, Washington University in St. Louis, and Yakov Berchen-
ko-Kogan, Florida Institute of Technology.
Parking Functions and their Generalizations I (Code: SS 21A), Jennifer Elder, Missouri Western State University, Pamela
Estephania Harris, University of Wisconsin-Milwaukee, and Juan Carlos Martinez Mori, Georgia Institute of Technology.
Real and Complex Function Theory with Application to Operator Theory I (Code: SS 5A), Elodie Pozzi, Saint Louis University,
and Constanze Liaw, University of Delaware.
Recent Developments in Mathematical Biology: Current and Future Directions. I (Code: SS 30A), Nalin Fonseka, University
of Central Missouri, and Rana D Parshad, Iowa State University.
Recent Trends in Discrete and Continuous Probability I (Code: SS 33A), Hugo Panzo, Saint Louis University, and Phanuel
A. Mariano, Union College.
Recent Trends in Nonlocal and Local PDEs I (Code: SS 1A), Pablo Raúl Stinga, Iowa State University, and Animesh
Biswas, Missouri State University.
Riemannian Geometry and Symmetry I (Code: SS 22A), William Wylie, Syracuse University, and Michael Roch Jablonski,
University of Oklahoma.
Tensor Categories and Representation Theory I (Code: SS 31A), Benjamin Spencer, Indiana University, Thibault Decoppet,
Harvard University, and Julia Yael Plavnik, Indiana University.
Topology and Group Actions I (Code: SS 28A), Lvzhou Chen, Purdue University, Michael P. Landry, Saint Louis Univer-
sity, Yash Lodha, University of Hawaii at Manoa, Samuel J. Taylor, Temple University, and David Ben McReynolds and
Sam Nariman, Purdue University.
Saint Louis University Poster Session (Code: SLUPOS), Elodie Pozzi, Saint Louis University.

Fall Eastern Virtual Sectional Meeting

Virtual
October 25–26, 2025 Program first available on AMS website: To be announced
Saturday – Sunday Issue of Abstracts: Volume 46, Issue 4

Meeting #1212 Deadlines

Eastern Section For organizers: Expired
Associate Secretary for the AMS: Benjamin A. Lotto For abstracts: September 9, 2025

The scientific information listed below may be dated. For the latest information, see https://www.ams.org/amsmtgs
/sectional.html.

Special Sessions
If you are volunteering to speak in a Special Session, you should send your abstract as early as possible via the abstract submission
form found at https://www.ams.org/cgi-bin/abstracts/abstract.pl.

Advances and Challenges in Hyperbolic Conservation Laws (Code: SS 1A), Prerona Dutta, Xavier University of Louisiana.
Advances in Non-Associative Rings and Algebras (Code: SS 2A), Layla H.M. Sorkatti, Southern Illinois University.
Advances in Quantum Topology, Tensor Categories, and Diagrammatic Algebras (Code: SS 3A), Melody Molander, UC Santa
Barbara, Anup Poudel, The Ohio State University, and Rhea Palak Bakshi, UC Santa Barbara.
Bridging Scales and Uncertainties: Advances in Inverse Problems, Data Assimilation, and Multiscale Modeling (Code: SS 5A),
Souvik Roy, The University of Texas at Arlington, and Anuj Abhishek, Case Western Reserve University.
Combinatorics and Graph Theory: The Audience Counts (Code: SS 6A), Rachel Kirsch, George Mason University, and
Aleyah Dawkins, Carnegie Mellon University.
Commutative Algebra & its Mathematical Interactions (Code: SS 7A), Dipendranath Mahato, Tulane University, and
Souvik Dey, Charles University, Prague.

August 2025 Notices of the American Mathematical Society 801

 MEETINGS & CONFERENCES

Continuous and Complex Dynamics (Code: SS 8A), Krystyna Kuperberg, Auburn University, and John C Mayer, Univ.
of Alabama - Birmingham.
Current Developments in Harmonic Analysis (Code: SS 9A), Brandon Sweeting and Andrew Walton Green, Washington
University in St. Louis, Cody B. Stockdale, Clemson University, and Nathan A. Wagner, Brown University.
Deterministic and Random Multivariate Analyticity (Code: SS 10A), Andriy I. Bandura, Ivano-Frankivsk National Technical
University of Oil and Gas, Oleh Skaskiv, Myroslav Sheremeta, and Oksana Holovata, Ivan Franko National University of
Lviv, Sergey Favorov, V.N.Karazin Kharkiv National University, Alexandre Eremenko, Purdue University, Mihail Sodin,
Tel Aviv University, and Alexander Rashkovskii, University of Stavanger.
Gauge Theory and Low-Dimensional Topology (Code: SS 11A), Paul M. N Feehan, Rutgers University, New Brunswick, and
Thomas Gibbs Leness, Florida International University.
LS-category and Topological Complexity: Theory and Applications (Code: SS 4A), Ekansh Jauhari, University of Florida, and
Jesús González, Departamento de Matematicas, Cinvestav.
Markovian and Queueing Models (Code: SS 12A), Alan Krinik and Randall J. Swift, California State Polytechnic Uni-
versity, Pomona, and Gerardo Rubino, INRIA, France.
Microlocal Analysis and Applications in Imaging (Code: SS 13A), Raluca Felea, Rochester Institute of Technology.
New Approaches to Mathematical General Relativity (Code: SS 14A), Christina Sormani, CUNY, Lehman College, James
Isenberg, University of Oregon, Emeritus, Eric Lang, University of Copenhagen, and Edward Thompson Bryden, Uni-
versiteit Antwerpen.
New Horizons for Quantum Error Correction: Categorical Approaches to Topological Codes (Code: SS 15A), Shanna Dobson,
University of California, Riverside, Tushar Pandey, DataRobot, and Sarah Chehade, Oak Ridge National Laboratory.
Nonlocal Differential Equations and Their Application (Code: SS 16A), Christopher Steven Goodrich, University of New
South Wales.
Notions of Positivity, Preserving Transforms, and Inequalities (Code: SS 17A), Daniel Soskin, Institute for Advanced Study,
Princeton, USA, and Prateek Kumar Vishwakarma, Universite Laval, QC, Canada.
Partial Differential Equations of Mathematical Physics and Others Areas of Applications (Code: SS 18A), Carmen J. Vanegas,
Universidad Técnica de Manabí.
Quantum Supergroups in Low-Dimensional Topology (Code: SS 19A), Matthew Harper, Michigan State University, Micah
Chrisman, Ohio State University, and Anup Poudel, The Ohio State University.
Recent Advances in Variational Methods and Nonlinear Elliptic PDEs (Code: SS 20A), Florin Catrina, St. John’s University,
Rushun Tian, Capital Normal University, and Zhi-Qiang Wang, Utah State University.
Recent Trends in Graph Theory (Code: SS 21A), Michael Santana, Lauren Keough, and Taylor Short, Grand Valley State
University.
Representations of p-adic Groups and Noncommutative Geometry (Code: SS 22A), Igor V. Nikolaev, St. John’s University,
Staten Island, NY, and Shanna Dobson, University of California, Riverside.
Research in Analysis and PDEs by Early Career Mathematicians (Code: SS 23A), Joshua Flynn, Massachusetts Institute
of Technology, Andrew Miller, Bridgewater State University, and Hyun Chul Jang, California Institute of Technology.
Unveiling Connections: Number Theory Meets Algebraic Geometry (Code: SS 24A), Pankaj Singh, Shreya Sharma, and Ms.
Swati, University of South Carolina.

Contributed Paper Sessions

AMS Contributed Paper Session (Code: AMSCP1A), Benjamin A. Lotto, Vassar.

Washington, District of Columbia

Walter E. Washington Convention Center and Marriott Marquis Washington DC
January 4–7, 2026 Deadlines
Sunday – Wednesday For organizers: Expired
For abstracts: September 9, 2025
Meeting #1213
Associate Secretary for the AMS: Betsy Stovall
Program first available on AMS website: To be announced
Issue of Abstracts: To be announced

802 Notices of the American Mathematical Society Volume 72, Number 7

 MEETINGS & CONFERENCES

Boise, Idaho
Boise State University
March 7–8, 2026 Program first available on AMS website: Not applicable
Saturday – Sunday Issue of Abstracts: Not applicable

Meeting #1214 Deadlines

Western Section For organizers: August 12, 2025
Associate Secretary for the AMS: Michelle Ann Manes, AIM For abstracts: January 13, 2026

Boston, Massachusetts
Boston College
March 28–29, 2026 Program first available on AMS website: Not applicable
Saturday – Sunday Issue of Abstracts: Not applicable

Meeting #1215 Deadlines

Eastern Section For organizers: September 2, 2025
Associate Secretary for the AMS: Benjamin A. Lotto For abstracts: February 3, 2026

Savannah, Georgia
Savannah, GA - Georgia Southern University-Armstrong Campus
March 28–29, 2026 Program first available on AMS website: Not applicable
Saturday – Sunday Issue of Abstracts: Not applicable

Meeting #1216 Deadlines

Southeastern Section For organizers: September 2, 2025
Associate Secretary for the AMS: Brian D. Boe For abstracts: February 3, 2026

The scientific information listed below may be dated. For the latest information, see https://www.ams.org/amsmtgs
/sectional.html.

Invited Addresses
Jennifer Hom, Georgia Institute of Technology, To be announced.
John W Shareshian, Washington University in St. Louis, To be announced.

Special Sessions
If you are volunteering to speak in a Special Session, you should send your abstract as early as possible via the abstract submission
form found at https://www.ams.org/cgi-bin/abstracts/abstract.pl.

Algebraic, combinatorial and geometric aspects of representation theory, Cornelius Pillen, Department of Mathematics and
Statistics, Aparna Upadhyay, University at Buffalo, SUNY, and Arik Wilbert, University of South Alabama.
Deterministic and Stochastic PDEs: Theoretical and Numerical Analyses, Pelin Guven Geredeli, Clemson University, and
Xiang Wan, Loyola University Chicago.
Modular Forms, Erick Ross and Hui Xue, Clemson University.
Recent advances in theory and practice of data science, Divine Wanduku, Georgia Southern University, and Ionut E. Iacob,
Statistical Consulting Unit (SCU), Georgia Southern University.
Recent Advances of PDEs in Modern Mathematical Physics: Theory and Applications, Yuanzhen Shao, The University of
Alabama, and Yi Hu and Shijun Zheng, Georgia Southern University.

August 2025 Notices of the American Mathematical Society 803

 MEETINGS & CONFERENCES

Trees in many contexts, Hua Wang, Department of Mathematical Sciences, Georgia Southern University, and Heather
Smith Blake, Davidson College.

Fargo, North Dakota

North Dakota State University
April 18–19, 2026 Program first available on AMS website: Not applicable
Saturday – Sunday Issue of Abstracts: Not applicable

Meeting #1217 Deadlines

For organizers: September 23, 2025
Central Section
For abstracts: February 24, 2026
Associate Secretary for the AMS: Betsy Stovall, University
of Wisconsin-Madison

Washington, District of Columbia

George Washington University
October 3–4, 2026 Issue of Abstracts: To be announced
Saturday – Sunday
Deadlines
Eastern Section
For organizers: March 3, 2026
Associate Secretary for the AMS: Benjamin A. Lotto
For abstracts: August 25, 2026
Program first available on AMS website: To be announced

AMS members are committed to

the vitality and sustainability of
the mathematics profession.
Thank you!
YOUR MEMBERSHIP Join or Renew Today!

Thank you!
Join
WWW.AMS.ORG/JOIN
makes great things
happen for the entire Renew
mathematical community WWW.AMS.ORG/RENEW

As a member, you also

have access to exclusive
resources and discounts.
Learn more here:

W W W.A M S .O R G / M E M B E R S H I P

804 Notices of the American Mathematical Society Volume 72, Number 7

 August is National
Make-a-Will Month.

Make a Will,
Leave a Legacy

C re
dit: S
Why should I make a will?

usan Sc
hwartz Wildstro
Making a will can give you the peace of mind that your
assets will be distributed according to your wishes and

m.
Susan with former student
values, and it can be tax-efficient for your heirs. Jess Freeman, who went on
to earn a PhD in mathematics
and now works in finance.

Support mathematics:
Make a bequest to the AMS When I was making my will some
years ago, I decided that I should put
When you make a gift in your will to the American
a few bequests in for the organizations
Mathematical Society, you become a member of the
that have meant a lot to me over the
Thomas S. Fiske Society. Fiske Society members years. The AMS was one of those
care deeply about the future of mathematics and are groups.
committed to supporting mathematical scholarship, —Susan Schwartz Wildstrom
research, and community for years to come. Fiske Society Member

Join the Fiske Society today!

Questions? Please contact Sophia Mackenzie

at [email protected] or (401) 455-4119,
or visit www.ams.org/legacy for more information.
American Mathematical Society
Distribution Center

35 Monticello Place,
Pawtucket, RI 02861 USA

NEW IN THE
Graduate Studies in Mathematics Series
GRADUATE STUDIES
I N M AT H E M AT I C S 251
Inequalities in Matrix Algebras
Eric Carlen, Rutgers University, New Brunswick, NJ
Inequalities in The theory of positive or completely positive maps from
one matrix algebra to another is the mathematical theory
Matrix Algebras
underlying the quantum mechanics of finite systems, as
well as much of quantum information and computing. This
book covers:
Eric Carlen
• Core Concepts — Covers inequalities, strong
subadditivity of quantum entropy, and entanglement
in tensor products of Hilbert spaces.
• Mathematical Progress — Explores the evolution
of entanglement theory and its role in recent
advancements.
• Prerequisites — Requires a good knowledge of linear
algebra and the basics of analysis and probability.
Graduate Studies in Mathematics, Volume 251; 2025; approximately 458
pages
• Hardcover: List Price: US $135.00; Order Code: GSM/251
• Softcover: List Price: US $89.00; Order Code: GSM/251.S
• eBook: List Price: US $85.00; Order Code: GSM/251.E
• Bundle (print & electronic): List Price: US $131.50; Order Code:
GSM/251.S.B
AMS and MAA member discounts apply.

bookstore.ams.org/gsm/251