MATHEMATICA

4/2015
 STUDIA
UNIVERSITATIS BABEŞ-BOLYAI
MATHEMATICA

4/2015
 EDITORIAL BOARD OF
STUDIA UNIVERSITATIS BABE Ş -BOLYAI MATHEMATICA

EDITORS:
Radu Precup, Babeş-Bolyai University, Cluj-Napoca, Romania (Editor-in-Chief)
Octavian Agratini, Babeş-Bolyai University, Cluj-Napoca, Romania
Simion Breaz, Babeş-Bolyai University, Cluj-Napoca, Romania
Csaba Varga, Babeş-Bolyai University, Cluj-Napoca, Romania

MEMBERS OF THE BOARD:

Ulrich Albrecht, Auburn University, USA
Francesco Altomare, University of Bari, Italy
Dorin Andrica, Babeş-Bolyai University, Cluj-Napoca, Romania
Silvana Bazzoni, University of Padova, Italy
Petru Blaga, Babeş-Bolyai University, Cluj-Napoca, Romania
Wolfgang Breckner, Babeş-Bolyai University, Cluj-Napoca, Romania
Teodor Bulboacă, Babeş-Bolyai University, Cluj-Napoca, Romania
Gheorghe Coman, Babeş-Bolyai University, Cluj-Napoca, Romania
Louis Funar, University of Grenoble, France
Ioan Gavrea, Technical University, Cluj-Napoca, Romania
Vijay Gupta, Netaji Subhas Institute of Technology, New Delhi, India
Nicolae Jităraşu, State University of Moldova, Chişinău, Moldova
Gábor Kassay, Babeş-Bolyai University, Cluj-Napoca, Romania
Mirela Kohr, Babeş-Bolyai University, Cluj-Napoca, Romania
Iosif Kolumbán, Babeş-Bolyai University, Cluj-Napoca, Romania
Alexandru Kristály, Babeş-Bolyai University, Cluj-Napoca, Romania
Andrei Mărcuş, Babeş-Bolyai University, Cluj-Napoca, Romania
Waclaw Marzantowicz, Adam Mickiewicz, Poznan, Poland
Giuseppe Mastroianni, University of Basilicata, Potenza, Italy
Mihail Megan, West University of Timişoara, Romania
Gradimir V. Milovanović, Megatrend University, Belgrade, Serbia
Petru Mocanu, Babeş-Bolyai University, Cluj-Napoca, Romania
Boris Mordukhovich, Wayne State University, Detroit, USA
András Némethi, Rényi Alfréd Institute of Mathematics, Hungary
Rafael Ortega, University of Granada, Spain
Adrian Petruşel, Babeş-Bolyai University, Cluj-Napoca, Romania
Cornel Pintea, Babeş-Bolyai University, Cluj-Napoca, Romania
Patrizia Pucci, University of Perugia, Italy
Ioan Purdea, Babeş-Bolyai University, Cluj-Napoca, Romania
John M. Rassias, National and Capodistrian University of Athens, Greece
Themistocles M. Rassias, National Technical University of Athens, Greece
Ioan A. Rus, Babeş-Bolyai University, Cluj-Napoca, Romania
Grigore Sălăgean, Babeş-Bolyai University, Cluj-Napoca, Romania
Mircea Sofonea, University of Perpignan, France
Anna Soós, Babeş-Bolyai University, Cluj-Napoca, Romania
András Stipsicz, Rényi Alfréd Institute of Mathematics, Hungary
Ferenc Szenkovits, Babeş-Bolyai University, Cluj-Napoca, Romania
Michel Théra, University of Limoges, France

BOOK REVIEWS:
Ştefan Cobzaş, Babeş-Bolyai University, Cluj-Napoca, Romania

SECRETARIES OF THE BOARD:

Teodora Cătinaş, Babeş-Bolyai University, Cluj-Napoca, Romania
Hannelore Lisei, Babeş-Bolyai University, Cluj-Napoca, Romania

TECHNICAL EDITOR:
Georgeta Bonda, Babeş-Bolyai University, Cluj-Napoca, Romania
YEAR (LX) 2015
MONTH DECEMBER
ISSUE 4

STUDIA
UNIVERSITATIS BABEŞ-BOLYAI
MATHEMATICA
4
Redacţia: 400084 Cluj-Napoca, str. M. Kogălniceanu nr. 1
Telefon: 0264 405300

CONTENTS

Ulrich Albrecht, ℵ1 -A-coseperable groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 493

Bo-Yan Xi and Feng Qi, Some new integral inequalities of
Hermite-Hadamard type for (log, (α, m))-convex functions
on co-ordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 509
Silvestru Sever Dragomir and Ian Gomm, Some Hermite-Hadamard
type inequalities for functions whose exponentials are convex . . . . . . . . . . . . . . 527
Virgil Pescar, On the univalence of an integral operator . . . . . . . . . . . . . . . . . . . . . 535
Saurabh Porwal, Sanjay Kumar Ghai and Kaushal Kumar,
Some new subclasses of bi-univalent functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 543
Fuad Al Sarari and Latha Satyanarayana, A new class of
(j, k)-symmetric harmonic starlike functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 553
Sukhwinder Singh Billing, On close-to-convex functions satisfying
a differential inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 561
Fatma Kanca and Irem Baglan, Generalized and numerical solution
for a quasilinear parabolic equation with nonlocal conditions . . . . . . . . . . . . . . 567
Zoltán Gábos and Ágnes Mester, Lines in the three-dimensional
Bolyai-Lobachevskian hyperbolic geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 583
Masoud Aminizadeh and Mina Ghotbaldini, Multisymplectic
connections on supermanifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 597
Igor Milovanović, Emina Milovanović and Edin Glogić,
On applications of Andrica-Badea and Nagy inequalities in
spectral graph theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 603
Flavius Pătrulescu and Teodor Groşan, Conjugate free convection
in a vertical channel filled with nanofluid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 611
Book reviews . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 623
Stud. Univ. Babeş-Bolyai Math. 60(2015), No. 4, 493–507

ℵ1-A-coseperable groups
Ulrich Albrecht

Abstract. Let A be a countable self-small Abelian group with a right Noetherian

right hereditary endomorphism ring. We show that the question whether strongly-
ℵ1 -A-generated groups are ℵ1 -A-coseparable is undecidable in ZFC. Our main
focus is on the algebraic aspect of the proof, not on the underlying set-theory.
Mathematics Subject Classification (2010): 20K20, 20K40.
Keywords: ℵ1 -a-coseparable groups, Martin’s axiom, endomorphism rings.

1. Introduction
Let A be an Abelian group with endomorphism ring E = E(A). Associated with
A are the functors HA (.) = Hom(A, .) and TA (.) = . ⊗E A which induce natural
maps θG : TA HA (G) → G and φM : M → HA TA (M ) defined by θG (α ⊗ a) = α(a)
and [φM (x)](a) = x ⊗ a for all α ∈ HA (G), x ∈ M and a ∈ A. The A-solvable
groups are the Abelian groups G such that θG is an isomorphism. Finally, a sequence
0 → G → H → L → 0 of Abelian group is A-balanced if the induced sequence
0 → HA (G) → HA (H) → HA (L) → 0 of right E-modules is exact
An important class of A-solvable groups are the (finitely) A-projective groups,
i.e. groups which are isomorphic to a direct summand of ⊕I A for some (finite) index-
set I. Finitely A-projective groups are always A-solvable [8], and the same holds for
arbitrary A-projective groups [9] if A is self-small, i.e. if HA preserves direct sums of
copies of A. Arnold and Murley showed in [9, Corollary 2.3] that a countable Abelian
group is self-small if and only if E is countable.
Epimorphic images of A-projective groups are called A-generated, but need not
be A-solvable. It is easy to see that a group G is A-generated if and only if θG is onto.
Moreover, if A is self-small, then a group G is A-solvable if and only if there is an
A-balanced exact sequence 0 → U → F → G → 0 in which F is A-projective and U
is A-generated [3]. Finally, G is A-torsion-free if every finitely A-generated subgroup
of G is isomorphic to a subgroup of a finitely A-projective group, and an A-generated
subgroup U of an A-torsion-free group G is A-pure if (U + P )/U is A-torsion-free for
all finitely A-generated subgroups P of G. If A is flat as an E-module, then A-torsion-
free groups are A-solvable [4]. We want to remind the reader that a right E-module M
494 Ulrich Albrecht

is non-singular if xI 6= 0 for all non-zero x in M and all essential right ideals I of E.

The ring R is right non-singular if RR is a non-singular module. If U is a submodule
of a non-singular right E-module M , then the S-closure of U in M consists of all
x ∈ M such that xI ⊆ M for some essential right ideal I of E [14]. Non-singularity is
closely related to A-torsion-freeness whenever A is a self-small Abelian group whose
endomorphism ring is right non-singular [5]:
a) If an A-generated group G is A-torsion-free, then HA (G) is non-singular.
b) An A-generated subgroup U of an A-torsion-free group G is contained in a
smallest A-pure subgroup V of G which is obtained as θG (TA (W )) where W is
the S-closure of HA (U ) in HA (G).
The focus of this paper are A-torsion-free groups G such that all A-generated
subgroups U of G with |U | < |G| are A-projective. Since A-generated subgroups of
A-projective groups need not be A-projective in general ([4] and [8]), some immediate
restrictions on A are needed to guarantee the existence of non-trivial groups with the
above property.

2. Hereditary Endomorphism Rings and κ-A-projective groups

An Abelian group is κ-A-generated, where κ is an infinite cardinal, if it is an
epimorphic image of ⊕I A for some index-set I with |I| < κ. The ℵ0 -A-generated
groups are referred to as finitely A-generated groups. An A-generated group G is κ-
A-projective if every κ-A-generated subgroup U of G is A-projective. If |A| < κ, then
this is equivalent to the condition that all A-generated subgroups U with |U | < κ
are A-projective. Since every finitely A-generated subgroup of a κ-A-projective group
G is A-projective, G is A-solvable. In particular, an A-generated group G is ℵ0 -A-
projective if every finitely A-generated subgroup is A-projective. If A is faithfully
flat as a left E-module, then finitely A-generated A-projective groups are finitely
A-projective [4].
Theorem 2.1. The following conditions are equivalent for a self-small torsion-free
Abelian group A:
a) i) A-projective groups are κ-A-projective for all infinite cardinals κ.
ii) Every exact sequence 0 → U → G → H → 0, in which G and H is κ-A-
projective for some infinite cardinal κ, is A-balanced.
b) E is a right hereditary ring.
In this case, A is faithfully flat as an E-module.
Proof. a) ⇒ b): To see that A is flat as an E-module, observe that An is ℵ0 -A-
projective for all n < ω, from which we obtain that G = α(An ) is A-projective for all
α : An → A. By a.ii), the exact sequence 0 → U → An → G → 0 with U = ker α is
A-balanced which yields the commutative diagram
0 −−−−→ TA HA (U ) −−−−→ TA HA (An ) −−−−→ TA HA (G) −−−−→ 0
  
oyθAn oyθG
θ  
yU
0 −−−−→ U −−−−→ An −−−−→ G −−−−→ 0.
 ℵ1 -A-coseperable groups 495

Thus, θU is an isomorphism. By Ulmer’s Theorem [17], A is E-flat.

Consider a right ideal I of E. Because A is E-flat, TA (I) ∼
= IA ⊆ A. Since IA is
an A-generated subgroup of A, and A is |IA|+ -A-projective by a.i), IA is A-projective.
Thus, HA TA (I) is a projective module fitting into the commutative diagram
0 −−−−→ HA TA (I) −−−−→ HA TA (E)
x x
oφE
φ 
 I
0 −−−−→ I −−−−→ E
from which we obtain that φI is one-to-one.
On the other hand, consider an exact sequence 0 → V → F → I → 0 where F
is a free right E-module. It induces the exact sequence
0 → TA (V ) → TA (F ) → TA (I) → 0.
The latter sequence is A-balanced by a.ii). Hence, the top-row in the commutative
diagram
HA TA (F ) −−−−→ HA TA (I) −−−−→ 0
x x
oφF
 φ
 I
F −−−−→ I −−−−→ 0
is exact, which yields that φI is onto. Consequently, I is projective, and E is right
hereditary.
b) ⇒ a): Let M be a right E-module. Since E is right hereditary, we can find an
exact sequence 0 → P → F → M → 0 in which P and F are projective. It induces
exact sequence
0 = TorR R
1 (F, A) → Tor1 (M, A) → TA (P ) → TA (F ) → TA (M ) → 0.

We obtain the commutative diagram

0 −−−−→ HA (TorR
1 (M, A)) −−−−→ HA TA (P ) −−−−→ HA TA (F )
x x
oφP oφF
 

0 −−−−→ P −−−−→ F.
Therefore, HA (TorR
1 (M, A)) = 0 for all right R-modules M .
+
If M is torsion-free, then it is isomorphic to a submodule of QM = Q ⊗Z M . Since
TorR R
1 (QM, A) is torsion-free and divisible, HA (Tor1 (QM, A)) = 0 is only possible if
TorR1 (QM, A) = 0. However, because E is right hereditary, we have the exact sequence
0 → TorR R R
1 (M, A) → Tor1 (QM, A) = 0, and Tor1 (M, A) = 0.
+
If M is torsion, then we select an exact sequence 0 → U → F1 → A → 0 in
which F1 is a free left E-module. It induces
0 = TorR R
1 (M, F1 ) → Tor1 (M, A) → M ⊗E V.

Since M ⊗E V is torsion, the same holds for TorR 1 (M, A). But, the latter also is
isomorphic to a subgroup of the torsion-free group TA (P ). Thus, TorR
1 (M, A) = 0.
496 Ulrich Albrecht

For an arbitrary M , we consider the exact sequence

0 = TorR R R
1 (tM, A) → Tor1 (M, A) → Tor1 (M/tM, A) = 0
where the first and the last term vanish but what has already been shown. Thus, A
is E-flat.
To show that A is faithful as a left E-module, suppose that TA (M ) = 0. The
sequence 0 → P → F → M → 0 yields the exact sequence
0 → TA (P ) → TA (F ) → TA (M ) = 0
since A is flat as an E-module. Hence, the top-row of the commutative diagram
0 −−−−→ HA TA (P ) −−−−→ HA TA (F ) −−−−→ 0
x x
oφP oφF
 

0 −−−−→ P −−−−→ F −−−−→ M −−−−→ 0.

is exact. A simple diagram chase shows M = 0.
Finally, A-generated subgroups of A-projective groups are A-projective if A is
faithfully flat and E is right hereditary [4], and a.i) holds. Finally, a.ii) is a direct
consequence of [6] since κ-A-projective groups are A-solvable. 
In particular, the last result shows that A-generated subgroups of self-small
groups with right hereditary endomorphism ring are A-projective. Our next results
summarizes other properties of such groups which we use frequently in this paper:
Corollary 2.2. Let A be a self-small torsion-free Abelian group whose endomorphism
ring is right hereditary:
a) Every exact sequence G → P → 0 such that G is A-generated and P is A-
projective splits.
b) An A-generated group is A-torsion-free if and only if it is ℵ0 -A-projective.
c) An A-generated subgroup of an A-torsion-free group is A-pure if and only if U
is a direct summand of U + V for all finitely A-generated subgroups V of G.
Proof. a) follows directly from the fact that A is faithfully flat which was established
in Theorem 2.1.
b) It remains to show that A-torsion-free groups are ℵ0 -A-projective. Suppose
that G is A-torsion-free, and let U be a finitely A-generated subgroup of G. Then U
can be embedded into an A-projective group, and thus is A-projective by Theorem
2.1.
c) Let U be an A-pure subgroup of an A-torsion-free group G. If V is a finitely
A-generated subgroup of G, then (U + V )/U can be embedded into an A-projective
group by Theorem 2.1. Thus, (U +V )/U is A-projective. By a), U is a direct summand
of U + V . 
However, the S-closure of a countable submodule of a non-singular module does
not need to be countable even if R is countable. For instance, if Q = Qω and R =
Z1S + Z(ω) , then Q is the maximal ring of quotients of R and |Q| > |R| although Q
is an essential extension of R. We want to remind the reader that a right E-module
M has Goldie-dimension m < ∞ if it contains an essential submodule which is the
 ℵ1 -A-coseperable groups 497

direct sum of m non-zero uniform submodules where a module X 6= 0 is uniform if

all its non-zero submodules are essential.

Proposition 2.3. Let R be a countable right non-singular ring which has finite right
Goldie-dimension. The S-closure of a countable submodule U of a non-singular right
R-module M is countable.

Proof. Let V be S-closure of U , and assume that V is uncountable. Let

X
F = {X ⊆ R | |X| < ∞ and xR is an essential right ideal}.
x∈X

Then, V = {y ∈ M | yX ⊆ U for some x ∈ F} since R has finite right Goldie-

dimension. Since V is uncountable and F is countable, we can find X0 ∈ F such
that yX0 ⊆ U for uncountably many y ∈ V . Let Y0 = {y ∈ V | yX0 ⊆ U }. Write
X0 = {x1 , . . . , xn }, and consider Y0 x1 ⊆ U . There is an uncountable subset Y1 of Y0
such that yx1 = y 0 x1 for all y, y 0 ∈ Y1 since U is countable. Repeating this argument
with x2 and Y1 yields an uncountable subset Y2 of Y1 such that yx2 = y 0 x2 for
all y, y 0 ∈ Y2 . By induction, we can find an uncountable subset Yn of Y0 such that
yxi = y 0 xi for all i = 1, . . . , n and all y, y 0 ∈ Yn . Thus, (y−y 0 )(x1 R+. . .+xn R) = 0 for
all y, y 0 ∈ Yn which contradicts the fact that M is non-singular because x1 R+. . .+xn R
is essential. Thus V has to be countable. 

By Sandomierski’s Theorem [11], a right finite dimensional, right hereditary ring

is right Noetherian.

Corollary 2.4. The following conditions are equivalent for a self-small torsion-free
Abelian group A whose endomorphism ring is right hereditary:
a) E is right Noetherian.
b) An A-generated subgroup U of a finitely A-projective group G is finitely A-
projective.

Proof. a) ⇒ b): Suppose that U is an A-generated subgroup of a finitely A-projective

group P . Then HA (U ) is a submodule of HA (P ), and hence a finitely generated
projective module. By Theorem 2.1, U is A-solvable, and U ∼ = TA HA (U ) is finitely
A-projective.
b) ⇒ a): Let I be a right ideal of E. Arguing as in the proof of Theorem 2.1, φI
is an isomorphism. Moreover TA (I) ∼ = IA since A is flat as an E-module. By b), IA
is finitely A-projective, from which we obtain that I is finitely generated. 

In view of the results of this section, we assume from this point on that A is a
self-small torsion-free group with a right Noetherian right hereditary endomorphism
ring. Huber and Warfield showed in [16] that E is a right and left Noetherian ring
whenever A is a torsion-free reduced group of finite rank with a right hereditary
endomorphism ring. Moreover, no generality is lost if we restrict our discussion to
the case that κ is a regular cardinal because Shelah’s singular compactness theorem
applies to A-projective groups [2].
498 Ulrich Albrecht

3. ℵ1 -A-Coseparable Groups
Let κ > ℵ0 be a regular cardinal, and A be a torsion-free Abelian group with
|A| < κ. An A-projective subgroup U of an ℵ0 -A-projective group G is κ-A-closed
provided that (U + V )/U is A-projective for all κ-A-generated subgroups V of G. If
|U | < κ, then this is equivalent to saying that G/U is κ-A-projective. The group G
is strongly κ-A-projective if it is κ-A-projective and every κ-A-generated subgroup U
of G is contained in an κ-A-generated, κ-A-closed subgroup V of G. By [1], SA (AI )
is ℵ1 -A-projective, but not strongly ℵ1 -A-projective since ⊕I A is not an ℵ1 -A-closed
subgroup.
In the following we focus on the case κ = ℵ1 since we are mainly interested in
the algebraic aspects instead of the underlying set-theory. However, most results of
this section carry over to the general case. In order to avoid immediate difficulties,
we restrict our discussion to the case that A is countable.
Lemma 3.1. Let A be a self-small countable torsion-free group with a right Noetherian
right hereditary endomorphism ring.
a) If G is ℵ1 -A-projective, then G/U is ℵ1 -A-projective for all ℵ1 -A-closed sub-
groups U of G.
b) If G is strongly ℵ1 -A-projective, then G/U is strongly ℵ1 -A-projective for all
countable ℵ1 -A-closed subgroups U of G.
Proof. a) Let {φn |n < ω} ⊆ HA (G/U ). Since Σn<ω φn (A) is countable, there is a
countable subgroup K of G such that Σn<ω φn (A) ⊆ (K + U )/U . Because A is count-
able, we can choose K to be A-generated. Since U is ℵ1 -A-closed in G, the group
(K + U )/U is U -projective, and the same holds Σn<ω φn (A). Therefore, G/U is ℵ1 -
A-projective.
b) Let V /U be a countable A-generated subgroup of G/U . Without loss of gen-
erality, we may assume that V is A-generated. Then, V is contained in an ℵ1 -A-closed
subgroup W is a ℵ1 -A-closed subgroup of G. Since U is countable this means that
G/W is ℵ1 -A-projective. Since G/W ∼ = (G/U )/(W/U ) and G/U is ℵ1 -A-projective,
we obtain that G/U is strongly ℵ1 -A-projective. 
An A-generated group G ℵ1 -A-coseparable if it is ℵ1 -A-projective and every A-
generated subgroup U of G such that G/U is countable contains a direct summand V
of G such that G/V is countable. Our next results describes ℵ1 -A-coseparable group.
Although our arguments follow the general outline of [13], significant modifications
are necessary in our setting.
Theorem 3.2. Let A be a self-small countable torsion-free group with a right Noether-
ian right hereditary endomorphism ring. A group G is ℵ1 -A-coseparable if and only if
G is A-solvable and every exact sequence
0→P →X→G→0
with P a direct summand of ⊕ω A and X A-generated splits.
Proof. Suppose that G is ℵ1 -A-coseparable, and consider an exact sequence
α β
0→P →X→G→0
 ℵ1 -A-coseperable groups 499

with P a direct summand of ⊕ω A and X A-generated. Since A is faithfully flat, X is

A-generated and G is A-solvable, the induced sequence
α HA (β)
0 → HA (P ) −→ HA (X) −→ HA (G) → 0
of right E-modules is exact by Theorem 2.1. Since HA (G) is non-singular by the re-
marks in the introduction, the same holds for HA (X). Observe that HA (P ) is count-
able since it is a direct summand of HA (⊕ω A), and the latter is countable because A
is self-small. We choose a complement W of im(HA (α)) in HA (X), and observe that
HA (X)/W is nonsingular. Since
M = HA (X)/(im(HA (α) ⊕ W ) ∼
= [HA (X)/W ][(im(HA (α) ⊕ W )/W ]
is singular and (im(HA (α) ⊕ W )/W is countable, HA (X)/W is countable as the S-
closure of a countable submodule by Proposition 2.3 because E is right Noetherian
and countable. Applying TA yields the commutative diagram
TA HA (α) TA HA (β)
0 −−−−→ TA HA (P ) −−−−−−→ TA HA (X) −−−−−−→ TA HA (G) −−−−→ 0
  
oyθP oyθG
 θ 
yX
α β
0 −−−−→ P −−−−→ X −−−−→ G −−−−→ 0.
Therefore, X is A-solvable, and U = θX (TA (W )) is an A-generated subgroup of X
such that α(P ) ∩ X = 0 and
X/[α(P ) ⊕ U ] ∼
= TA (M )
is countable. If H = β(U ), then β|U is one-to-one. Since β(U ) ∼ = U ∼ = TA (W )
is A-generated and G/β(U ) is countable, there is a subgroup K of U such that
G = β(K) ⊕ B for some countable subgroup B of G using the fact that G is ℵ1 -
A-coseparable. Select a subgroup V of X containing α(P ) such that β(V ) = B.
Clearly, V is countable.
To show X = K ⊕ V , take x ∈ X and write β(x) = β(k) + β(v) for some k ∈ K
and v ∈ V . Then x − k − b ∈ α(P ) ⊆ V . On the other hand, suppose that y ∈ K ∩ V .
Then β(y) ∈ β(K) ∩ B = 0, from which we obtain
y ∈ α(P ) ∩ K ⊆ α(P ) ∩ U = 0.
Moreover, V is A-generated since it is a direct summand of X, while β(V ) ∼ = V /α(P )
is A-projective as a countable subgroup of G. Therefore, α(P ) is a direct summand
of V .
Conversely, assume that G is an A-solvable group such that every exact sequence
0 → P → X → G → 0 with P a direct summand of ⊕ω A and X A-generated
splits. Suppose that G contains a countable A-generated subgroup U which is not
A-projective. Since U is A-solvable because A is E-flat by Theorem 2.1, HA (U ) is
not projective. Looking at projective resolutions of HA (U ), we can find a countable
projective module P with Ext1E (HA (U ), P ) 6= 0. Since E is right hereditary, we have
an exact sequence
Ext1E (HA (G), P ) → Ext1E (HA (U ), P ) → 0.
500 Ulrich Albrecht

Thus, there is a non-splitting sequence 0 → P → M → HA (G) → 0 which induces

0 → TA (P ) → TA (M ) → TA HA (G) → 0 which splits since G ∼= TA HA (G). We
therefore obtain the commutative diagram
0 −−−−→ HA TA (P ) −−−−→ HA TA (M ) −−−−→ HA TA HA (G) −−−−→ 0
x x x
oφP oφHA (G)
 φ 
 M
0 −−−−→ P −−−−→ M −−−−→ HA (G) −−−−→ 0
in which φM is an isomorphism by the 3-Lemma. Since the top-row splits, the same has
to hold for the bottom, which contradicts its choice. Therefore, G is ℵ1 -A-projective.
Consider an A-generated subgroup C of G such that G/C is countable. We can
find a countable subgroup B such that G = C + B, and no generality is lost if we
assume in addition that B is A-generated. By what was shown in the last paragraph,
π
B is A-projective. We consider the exact sequence 0 → K → B ⊕ C → G → 0 with
π(b, c) = b + c. Since G is A-solvable, and C is an A-generated subgroup of G, the
group B ⊕ C is A-solvable. By Theorem 2.1, K = {(b, −b) | b ∈ B ∩ C} is A-generated,
and hence A-solvable since A is E-flat. Since K is isomorphic to a subgroup of the
countable A-projective group B, another application of Theorem 2.1 yields that K is
a countable A-projective group. By our hypotheses, the map π splits, say πδ = 1G for
some homomorphism δ : G → B ⊕C. Let ρ : B ⊕C → B be the projection onto B with
kernel C, and consider D = ker(ρδ). Since G/D is A-generated and isomorphic to a
subgroup of the countable A-projective group B, it is A-projective itself. By Theorem
2.1, D is a direct summand of G. Moreover, every d ∈ D satisfies δ(d) = (0, c) for
some c ∈ C since ρδ(d) = 0 yields δ(d) ∈ ker ρ = C. Then d = πδ(d) = π(0, c) = c,
and D ⊆ C. 

A group W is an A-Whitehead group if it admits an A-balanced exact sequence

0 → U → F → W → 0 in which F is A-projective and U is A-generated with the
property that

0 → Hom(W, A) → Hom(F, A) → Hom(U, A) → 0

is exact.

Corollary 3.3. Let A be a self-small countable torsion-free group with a right Noether-
ian right hereditary endomorphism ring.
a) Every ℵ1 -A-coseparable group W is an A-Whitehead group.
b) It is consistent with ZFC that there exists a strongly ℵ1 -A-projective group G
which is not ℵ1 -A-coseparable.

Proof. a) By [7], an A-solvable group W is an A-Whitehead group if every exact

sequence 0 → A → X → W → 0 with SA (X) = X splits which is satisfied by W
because of Theorem 3.2.
b) If we assume V = L, then all A-Whitehead groups are A-projective. However,
there exist strongly ℵ1 -A-projective group G with Hom(G, A) = 0 [7]. 
 ℵ1 -A-coseperable groups 501

4. Strongly ℵ1 -A-Projective Groups and Martin’s Axiom

We use the formulation of Martin’s Axiom given in [13, Definition VI.4.2]. A
partially ordered set (P, ≤) satisfies the countable chain condition (ccc) if any an-
tichain in (P, ≤) is countable. An antichain is a subset A of P such that any two
distinct members of A are incompatible, i.e., whenever p, q ∈ A, then there does not
exist r ∈ P such that r ≥ p and r ≥ q. A subset D of P is dense if, for every p ∈ P
there is q ∈ D such that p ≤ q. Finally, a subset F of P is directed, if, for all p, q ∈ F,
there is r ∈ F such that r ≥ p and r ≥ q.
For a cardinal κ, let MA(κ) denote the statement:

Let (P, ≤) be a partially ordered set satisfying the countable chain condition
(ccc). For every family D = {Dα | α < κ} of dense subsets of P , there is a directed
subset F of P such that F ∩ Dα 6= ∅ for all α, i.e. F is D-generic.

Martin’s axiom (MA) stipulates that MA(κ) holds for every κ < 2ℵ0 [13].
Theorem 4.1. (M A + ℵ1 < 2ℵ0 ) Let A be a self-small countable torsion-free group
with a right Noetherian right hereditary endomorphism ring. If G is a strongly ℵ1 -A-
projective group and 0 → U → ⊕I A → G → 0 is an A-balanced exact sequence such
that SA (U ) = U and |I| < 2ℵ0 , then the induced sequence
0 → Hom(G, B) → Hom(⊕I A, B) → Hom(G, B) → 0
is exact for all countable A-solvable group B.
Proof. We consider an A-balanced exact sequence 0 → U → ⊕I A → G → 0 where
U → ⊕I A is the inclusion map. Let P(U ) be the collection of A-generated A-pure
subgroups V of F = ⊕I A containing U such that V /U is finitely A-projective. Since
V is A-generated and A is faithfully flat, U is a direct summand of V by Corollary
2.2, say V = U ⊕ RV for some finitely A-projective group RV .
To show that the sequence Hom(⊕I A, B) → Hom(G, B) → 0 is exact whenever
B is a countable A-solvable group, let φ ∈ Hom(U, B), and consider
P = {(V, ψ) | V ∈ P(U ), ψ ∈ Hom(V, B), and ψ|U = φ}.
We partially order P by (V1 , ψ1 ) ≥ (V2 , ψ2 ) if and only if V2 ⊆ V1 and ψ1 |V2 = ψ2 .
Once we have shown that P and D = {D(J)|J ⊆ I finite} satisfy the hypotheses
of Martin’s Axiom, then we can find a D-generic directed directed subset F of P .
Define a map ψ : ⊕I A → B as follows. For x ∈ ⊕I A, choose a finite subset J of I
such that x ∈ ⊕J A. Since F is D-generic, there is (V, δ) ∈ D(J) ∩ F with x ∈ V .
Define ψ(x) = δ(x). Moreover, if (V1 , δ1 ) and (V2 , δ2 ) are two choices, then there
is (V3 , δ3 ) ∈ F such that (Vi , δi ) ≤ (V3 , δ3 ) for i = 1, 2 since F is directed. Thus,
δ1 (x) = δ3 (x) = δ2 (x). 
The key towards showing that (P, ≤) satisfies the countable chain condition is
Theorem 4.2. Every uncountable subset P 0 of P contains an uncountable subset P 00
for which we can find an A-pure A-projective subgroup X of F containing U as a
direct summand such that V ⊆ X whenever (V, ψ) ∈ P 00 .
502 Ulrich Albrecht

Proof. We may assume that P 0 = {(Vν , ψν )|ν < ω1 }. Since U is a direct summand
of Vν , we obtain that HA (Vν /U ) ∼ = HA (Vν )/HA (U ) is a finitely generated right E-
module. In particular, it has finite right Goldie dimension since E is right Noetherian.
Therefore, no generality is lost if we assume that there is m < ω such that HA (Vν /U )
has Goldie dimension m for all ν < ω1 .
Let 0 ≤ k ≤ m be maximal with respect to the property that there exists an
A-pure A-projective subgroup T of F containing U such that HA (T /U ) has Goldie
dimension k and T is contained in uncountable many Vν . This k exists since U is the
choice for T in the case k = 0. Observe that T /U is A-solvable as an A-generated
subgroup of the A-solvable group G = F/U . Thus, 0 → U → T → T /U → 0 is
A-balanced, and HA (T /U ) ∼ = HA (T )/HA (U ) has finite Goldie-dimension and is non-
singular. Thus, it contains a finitely generated essential submodule. Since E is right
Noetherian and countable, any essential extension of a non-singular finite dimensional
right E-module is countable by Proposition 2.3. In particular, HA (T /U ) is countable,
and hence T /U ∼ = TA HA (T /U ) is countable. Since G is ℵ1 -A-projective, T /U is A-
projective, and T = U ⊕ W because A is faithfully flat by Corollary 2.2.
Suppose that T 0 is an A-generated subgroup of F containing T such that T 6= T 0 .
There exists α ∈ HA (T 0 ) with α(A) 6⊆ T . Since T is A-pure in F , we obtain T +α(A) =
T ⊕ C with C 6= 0. Thus, the Goldie-dimension of HA (T 0 ) is at least k + 1, and T 0 is
contained in only countably many of the Vν . No generality is lost if we assume that T
is contained in Vν for all ν. Since T is A-pure in F and Vν = U ⊕ RVν = T + RVν for
some finitely A-projective subgroup RVν of F , we obtain decompositions Vν = T ⊕Wν .
Observe that Wν is finitely A-projective.
We construct X as the union of a smooth ascending chain {Xν |ν < ω1 } of A-
pure A-projective subgroups of F containing T and an ascending chain of ordinals
{σν |ν < ω1 } such that Xν+1 /Xν is A-projective, Wσν+1 ⊆ Xν+1 , and Xν /U is an
image of ⊕ω A for all ν < ω1 . We set X0 = T , and Xα = ∪ν<α Xν if α is a limit
ordinal. Then, Xα /U is a countable subgroup of F/U , and hence A-projective. Set
σα = sup(σν |ν < α).
If α = ν + 1, then there exists a subgroup Cν of F containing Xν such that the
group Cν /U is an A-projective countable ℵ1 -A-closed subgroup of F/U since F/U is
strongly ℵ1 -A-projective. In particular, F/Cν ∼ = (F/U )/(Cν /U ) is A-solvable. Since A
is flat, Cν is A-generated by Theorem 2.1. If K is a countable A-generated subgroup
of F , then (K + U )/U is a countable subgroup of F/U . Hence,
(K + Cν )/Cν ∼
= [(K + Cν )/U ]/[Cν /U ]
is A-projective.
To construct σα , assume Wµ ∩ Cν 6= 0 for all µ > σν . Then,
Wµ /(Wµ ∩ Cν ) ∼
= (Wµ + Cν )/Cν
is A-projective by the last paragraph. Since A is faithfully flat, Wµ ∩Cν is A-generated,
and there is a map 0 6= αµ ∈ HA (Wµ ∩ Cν ) ⊆ HA (Cν ). Since Cν /U is a countable
subgroup of F/U , it is A-projective, and Cν = U ⊕ Pν since A is faithfully flat.
Observe that Pν is countable and A-projective. Write αµ = βµ + µ with βµ ∈ HA (U )
and µ ∈ HA (Pν ). Since E is countable, the same holds for HA (Pν ), and there is
 ℵ1 -A-coseperable groups 503

 ∈ HA (Pν ) such that µ =  for uncountably many µ. For all these µ, we have
(A) ⊆ Wµ + U ⊆ Vµ . Hence, T + (A) ⊆ Vµ for uncountably many µ. However, this
is only possible if (A) ⊆ T . But then, αµ (A) ⊆ T ∩ Wµ = 0, a contradiction.
Therefore, we can find an ordinal σα > σν with Cν ∩ Wσα = 0. In particular,
Xν ⊆ Cν yields Xν ∩ Wσα = 0. Let Y be the S-closure of
HA (Xν ⊕ Wσα ) = HA (Xν ⊕ HA (Wσα ) ⊇ HA (U )
in HA (F ) and let Xα = θF (Y ⊗ A) = Y A. As an A-generated subgroup of F , Xα is
A-solvable. Then, the inclusion Y ⊆ HA (Xα ) induces the commutative diagram
0 −−−−→ TA (Y ) −−−−→ TA HA (Xα ) −−−−→ TA (HA (Xα )/Y ) −−−−→ 0
 
oyθF |TA (Y ) oyθXα
 

1
YA
0 −−−−→ Y A −−− −→ YA −−−−→ 0
from which we get TA (HA (Xα )/Y ) = 0. Since A is faithfully flat, Y = HA (Xα ), and
HA (Xα )/[HA (Xν ) ⊕ HA (Wσα )] is singular.
Observe that Y /HA (U ) is the S-closure of [HA (Xν ) + HA (Wσα )]/HA (U ) in
HA (F )/HA (U ) because
HA (F )/Y ∼= [HA (F )/HA (U )]/[Y /HA (U )]
is non-singular and
Y /HA (Xν ⊕ Wσα ) ∼
= [Y /HA (U )]/[HA ((Xν ⊕ Wσα )/HA (U )]
is singular. Since F/U is A-solvable, and Xν /U is countable, HA (Xν )/HA (U ) is count-
able. Moreover, Wµ is finitely A-projective. Hence, the E-module HA (Wσα ) is count-
able too, and
[HA (Xν ) + HA (Wσα )]/HA (U )
is countable. Thus, Y /HA (U ) is an essential extension of a countable non-singular
right E-module. By Proposition 2.3, we obtain that Y /HA (U ) is countable. Thus,
there is a countable submodule Y 0 of Y with Y = Y 0 + HA (U ). Then Xα /U is
countable and Xα = Y 0 A + Xν , and. Consequently, Xα /U has to be A-projective,
and the same holds for Xα ∼ = Xα /U ⊕ U .
It remains to show that Xα /Xν is A-projective. For this, observe that the group
Xα /(Xα ∩ Cν ) =∼ (Xα + Cν )/Cν
is countable since it is an epimorphic image of (Xα +Cν )/U which is countable because
Xα and Cν /U are countable. Since Cν /U is ℵ1 -A-closed in F/U , we have that
Xα /(Xα ∩ Cν ) ∼
= [(Xα + Cν )/U ]/[Cν /U ]
is A-projective. Since A is flat, Xα ∩ Cν is A-generated as in Theorem 2.1. For τ in
HA (Xα ∩ Cν ), choose a regular element c ∈ E such that τ c ∈ HA (Xν ) ⊕ HA (Wσα ),
say τ c = β + γ for some β ∈ HA (Xν ) and γ ∈ HA (Wσα ). Then
γ = τ c − β ∈ HA (Wσα ) ∩ HA (Cν ) = 0.
Hence, τ c ∈ HA (Xν ). Since Xν is A-pure in F , we obtain τ ∈ HA (Xν ). Therefore,
HA (Xα ∩ Cν ) ⊆ HA (Xν ), and Xα ∩ Cν ⊆ Xν . Since Xν is contained in Xα and in
504 Ulrich Albrecht

Cν , we obtain Xα ∩ Cν = Xν . Then Xα /Xν =∼ (Xα + Cν )/Cν is A-projective by what

we have already shown. In particular, Xν is a direct summand of Xα .
Consequently, X = ∪ν<ω1 Xν is A-pure and A-projective. Because
Xν+1 /Xν ∼ = [Xν+1 /T ]/[Xν /T ]
is A-projective for all ν, the group X/T is A-projective. This yields X = T ⊕ S.
However, T = U ⊕ W , so that X = U ⊕ W ⊕ S. Finally,
Vσν+1 = T ⊕ Wσν+ ⊆ Xν+1 ⊆ X
00
for all ν < ω1 . Let P = {(Vσν+1 , ψσν+1 )|ν < ω1 }. 
Corollary 4.3. P satisfies the countable chain condition.
Proof. Since B is a countable A-solvable group, there is an exact sequence
0 → V → ⊕ω A → B → 0
which is A-balanced by Theorem 2.1. Thus, HA (B) is countable as an epimorphic
image of HA (⊕ω A) ∼ = ⊕ω E using the self-smallness of A.
Let P 0 be an uncountable subset of P . By the previous Lemma, we may as-
sume P 0 = {(Vν , ψν )|ν < ω1 } such that there is an A-pure A-projective subgroup X
containing U as a direct summand satisfying Vν ⊆ X for all ν < ω1 . We can write
X = U ⊕ Y and Y = ⊕J Yj where each Yj is isomorphic to a subgroup of A. This is
possible since E is hereditary.
For ν < ω1 , we have Vν = U ⊕(Y ∩Vν ). Since Y ∩Vν is finitely A-projective, there
is a finite subset Jν of J such that HA (Y ∩ Vν ) ⊆ HA (⊕Jν Yj ), and Y ∩ Vν ⊆ ⊕Jν Yj .
Therefore, Vν is an A-pure subgroup of
Vν + (⊕Jν Yj ) = U ⊕ (⊕Jν Yj ).
Because ⊕Jν Yj is finitely A-generated, Vν is a direct summand of U ⊕ (⊕Jν Yj ), say
Vν + (⊕Jν Yj ) = Vν ⊕ Xν . Since Vν + (⊕Jν Yj ) is A-projective, the same holds for
Xν . Thus, Xν is isomorphic to a direct summand of ⊕Jν Yj . Moreover, ψν : Vν → B
extends to a map λν : U ⊕ (⊕Jν Yj ) → B. By the Adjoint-Functor-Theorem,
Hom(⊕J Yj , B) = ∼ HomE (HA (⊕J Yj ), HA (B))
ν ν

is countable since HA (B) is countable as was shown in the first paragraph of the proof
and Jν is finite. Consequently, there are at most countably many different extensions
of φ to U ⊕ (⊕Jν ).
If there are only countably many different Jν ’s, then there is ν0 such that Jν0 =
Jµ for uncountable µ. Thus, there are µ1 and µ2 with Jν0 = Jµ1 = Jµ2 and λµ1 = λµ2 .
Thus, ψµ1 and ψµ2 have a common extension. Therefore, P 0 = {(Vν , ψν )|ν < ω1 }
cannot be an antichain. On the other hand, if there are uncountably many Jν ’s, then
we may assume without loss of generality that Jν 6= Jµ for µ 6= ν. Finally, we can
impose the requirement that all the Jν have the same order. Thus, Jν cannot be
contained in Jµ for µ 6= ν. Since (Vν , ψν ) ≤ (Vν ⊕ Xν , λν ), we may assume that
Vν = U ⊕ (⊕Jν Yj ) and λν = ψν .
There is a subset T of J which is maximal with respect to the property that
it is contained in uncountably many of the Jν . We may assume that T is actually
 ℵ1 -A-coseperable groups 505

contained in all of the Jν . Observe that T is finite and a proper subset of all the Jν .
Otherwise, all the Jν would have to coincide with T since they have the same finite
order. Since Hom(⊕T Yj , B) ∼ = HomE (HA (⊕T Yj ), HA (B)) is countable by the Adjoint-
Functor-Theorem, there are uncountably many ψν which have the same restriction to
W = U ⊕ (⊕T Yj ). Without loss of generality, we may assume that this happens for
all ν.
Let j ∈ J0 \ T . The maximality of T guarantees that j is contained in only
countably many of the Jν . Since J0 \ T is finite, there is µ < ω1 with Jµ ∩ J0 = T .
The maps ψµ and ψ0 have a common extension σ : U ⊕ (⊕J0 ∪Jν Yj ) → B since they
coincide on W . Since U ⊕ (⊕J0 ∪Jν Yj ) is a direct summand of X, and X is A-pure in
F , we have that U ⊕ (⊕J0 ∪Jν Yj ) is A-pure in F . Because J0 ∪ Jν is finite,
(U ⊕ (⊕J0 ∪Jν Yj ), σ) ∈ P.
Thus,
(U ⊕ (⊕J0 ∪Jν Yj ), σ) ≥ (Vµ , ψµ ), (V0 , ψ0 ).
Consequently, P 0 cannot be an anti-chain. 
For every finite subset J of I, let D(J) = {(V, ψ) ∈ P | ⊕J A ⊆ V }.
Proposition 4.4. P and D = {D(J)|J ⊆ I finite} satisfy the hypotheses of Martin’s
Axiom.
Proof. By Corollary 4.3, it remains to show that D(J) is dense in P . For this, let
(V, ψ) ∈ P . We have to find (W, α) ∈ P such that ⊕J A and V are contained in W
and α|V = ψ. Since V /U is finitely A-projective and G ∼ = F/U is strongly ℵ1 -A-
projective, there is a subgroup X of F containing V and ⊕J A such that X/U is a
ℵ1 -A-closed, A-projective countable subgroup of F/U . Since
[F/U ]/[X/U ] ∼
= F/X
is ℵ1 -A-projective, it is A-solvable by Theorem 2.1. Using the same result once more,
we obtain that the sequence 0 → X → F → F/X → 0 is A-balanced. In particular,
SA (X) = X and X is A-projective. Moreover,
HA (F )/HA (X) ∼ = HA (F/X) ∼ = HA ([F/U ]/[X/U ]) ∼
= HA (F/U )/HA (X/U )
since X in F and X/U in F/U are A-balanced by the faithful flatness of A. But the
latter is non-singular, since [F/U ]/[X/U ] is ℵ1 -Aprojective. Therefore, X is A-pure
in F .
Since the group X/U is A-projective, we have a decomposition X = U ⊕ P .
Hence, V = U ⊕ (V ∩ P ) and V ∩ P is finitely A-projective. In the same way,
(⊕J A) + U = U ⊕ [((⊕J A) + U ) ∩ P ]
yields that ((⊕J A) + U ) ∩ P is A-generated and an image of ⊕J A.
Therefore, ((⊕J A) + U ) ∩ P and V ∩ P are finitely A-projective subgroups of P . Thus,
HA (((⊕J A) + U ) ∩ P ) and HA (V ∩ P ) are finitely generated submodule of HA (P ).
Since E is right hereditary, HA (P ) is a direct sum of right ideals of E, which yields
that HA (((⊕J A) + U ) ∩ P ) and HA (V ∩ P ) are contained in a finitely generated
direct summand of HA (P ). Hence, there is a finitely A-projective summand D of P
506 Ulrich Albrecht

which contains V ∩ P and ((⊕J A) + U ) ∩ P . Since U ⊕ D = V + D and V is A-pure

in F , we obtain that V is a direct summand of U ⊕ D. Thus, ψ extends to a map
α : U ⊕ D → B. Clearly, (U ⊕ D, α) ∈ P and (U ⊕ D, α) ≥ (V, ψ). 

An A-generated group G ℵ1 -A-separable if every countable subset of G is con-

tained in an A-projective direct summand of G.
Corollary 4.5. (M A + ℵ1 < 2ℵ0 ) If A is a self-small countable torsion-free group
with a right Noetherian right hereditary endomorphism ring, then every strongly ℵ1 -
A-projective group is ℵ1 -A-separable and ℵ1 -A-coseparable.
Proof. By Theorem 3.2 and Theorem 4.1, a strongly ℵ1 -A-projective group G is ℵ1 -
A-coseparable. It remains to show that is ℵ1 -A-separable too. For a countable subset
X of G select a countable ℵ1 -A-closed subgroup U of G containing X. Since G/U is
strongly ℵ1 -A-projective, the sequence 0 → U → G → G/U → 0 splits by Theorem
4.1. 

References
[1] Albrecht, U., Endomorphism rings and A-projective torsion-free Abelian Groups, Abelian
Group Theory, Proceedings Honolulu 1982/83, Springer Lecture Notes in Mathematics
1006, Springer Verlag, Berlin, New York, Heidelberg, 1983, 209-227.
[2] Albrecht, U., A-projective groups of large Cardinality, Abelian Group Theory and Mod-
ules, Proceedings Udine 1984, CISM Courses and Lectures 287, Springer Verlag, Wien,
New York, 1984, 233-243.
[3] Albrecht, U., Abelian groups A, such that the category of A-solvable groups is preabelian,
Contemporary Mathematics 87(1989), 117-131.
[4] Albrecht, U., Baer’s lemma and Fuchs’ problem 84a, Transactions of the American
Mathematical Society, 293(1986), 565-582.
[5] Albrecht, U., Endomorphism rings and a generalization of torsion-freeness and purity,
Communications in Algebra, 17(5)(1989), 1101-1135.
[6] Albrecht, U., Endomorphism rings, tensor products, and Fuchs’ problem 47, Warfield
Memorial Volume, Contemporary Mathematics, 130(1992), 17-31.
[7] Albrecht, U., A-Whitehead groups, accepted for publication in Studia UBB Math.
[8] Arnold, D.M., Lady, L., Endomorphism rings and direct sums of torsion-free Abelian
groups, Trans. Amer. Math. Soc., 211(1975), 225-237.
[9] Arnold, D.M., Murley, C.E., Abelian groups A, such that Hom(A,) preserves direct sums
of copies of A, Pac. J. Math., 56(1975), 7-20.
[10] Arnold, D. M., Finite Rank Torsion Free Abelian Groups and Rings, Lecture Notes in
Mathematics Vol. 931, Springer-Verlag Berlin-Heidelberg-New York, 1982.
[11] Chatters, A.W., Hajarnavis, C.R., Rings with Chain Conditions, Pitman, 1980.
[12] Corner, A.L.S., Every countable reduced torsion-free ring is an endomorphism ring, Proc.
London Math. Soc., 13(1963), 687-710.
[13] Eklof, P.C., Mekler, A.H., Almost Free Modules, Vol. 46, North Holland Mathematical
Library, 1990.
 ℵ1 -A-coseperable groups 507

[14] Goodearl, K.R., Ring Theory, Pure and Applied Mathematics No. 33, Marcel Dekker,
New York-Basel, 1976.
[15] Göbel, R., Trilifaj, J., Approximations and Endomorphism Algebras of Modules, de
Gruyter Expositions in Mathematics 41, W. de Gruyter, Berlin - New York, 2006.
[16] Huber, M., Warfield, R.B., Homomorphisms between Cartesian powers of an Abelian
group, Lecture Notes in Mathematics, 874(1981), 202-227.
[17] Ulmer, F., A flatness criterrion in Groethendick categories, Inventiones Math., 19(1973),
331-336.

Ulrich Albrecht
Department of Mathematics
Auburn University
Auburn, AL 36849, U.S.A.
e-mail: [email protected]
Stud. Univ. Babeş-Bolyai Math. 60(2015), No. 4, 509–525

Some new integral inequalities of

Hermite-Hadamard type for
(log, (α, m))-convex functions on co-ordinates
Bo-Yan Xi and Feng Qi

Abstract. In the paper, the authors introduce a new concept “(log, (α, m))-convex
functions on the co-ordinates on the rectangle of the plane” and establish some
new integral inequalities of Hermite-Hadamard type for (log, (α, m))-convex func-
tions on the co-ordinates on the rectangle from the plane.
Mathematics Subject Classification (2010): 26A51, 26D15, 26D20, 26E60, 41A55.
Keywords: Co-ordinates, (log, (α, m))-convex functions on co-ordinates, Hermite-
Hadamard’s inequality.

1. Introduction
The following definitions are well known in the literature.

Definition 1.1. A function f : I ⊆ R = (−∞, +∞) → R is said to be convex if

f (λx + (1 − λ)y) ≤ λf (x) + (1 − λ)f (y)
holds for all x, y ∈ I and λ ∈ [0, 1].

Definition 1.2. If a positive function f : I ⊆ R → (0, ∞) satisfies

f (λx + (1 − λ)y) ≤ f λ (x)f 1−λ (y),
for all x, y ∈ I and λ ∈ [0, 1], then we call f a logarithmically convex function on I.

This work was partially supported by the National Natural Science Foundation of China under
Grant No. 11361038 and by the Foundation of the Research Program of Science and Technology at
Universities of Inner Mongolia Autonomous Region under Grant No. NJZY14192 and by the Inner
Mongolia Autonomous Region Natural Science Foundation Project under Grant No. 2015MS0123,
China.
510 Bo-Yan Xi and Feng Qi

Definition 1.3 ([8]). For f : [0, b] → R and m ∈ (0, 1], if

f (tx + m(1 − t)y) ≤ tf (x) + m(1 − t)f (y)
is valid for all x, y ∈ [0, b] and t ∈ [0, 1], then we say that f is an m-convex function
on [0, b].
Definition 1.4. [(9)] For f : [0, b] → R and (α, m) ∈ (0, 1] × (0, 1], if
f (tx + m(1 − t)y) ≤ tα f (x) + m(1 − tα )f (y)
is valid for all x, y ∈ [0, b] and t ∈ [0, 1], then we say that f is an (α, m)-convex
function on [0, b].
Definition 1.5 ([4, 5]). A function f : ∆ = [a, b] × [c, d] ⊆ R2 → R, with a < b and
c < d, is said to be convex on the co-ordinates on ∆ if the partial functions
fy : [a, b] → R, fy (u) = f (u, y) and fx : [c, d] → R, fx (v) = f (x, v)
are convex for all x ∈ (a, b) and y ∈ (c, d).
Definition 1.6 ([4, 5]). A function f : ∆ = [a, b] × [c, d] ⊆ R2 → R, with a < b and
c < d, is said to be convex on the co-ordinates on ∆ if the partial functions
f (tx + (1 − t)z, λy + (1 − λ)w)
≤ tλf (x, y) + t(1 − λ)f (x, w) + (1 − t)λf (z, y) + (1 − t)(1 − λ)f (z, w)
holds for all t, λ ∈ [0, 1], (x, y), (z, w) ∈ ∆.
Definition 1.7 ([3]). For some (α1 , m1 ), (α2 , m2 ) ∈ (0, 1]2 , a function f : [0, b]×[0, d] →
R is said to be (α1 , m1 )-(α2 , m2 )-convex on the co-ordinates on [0, b] × [0, d], if
f (ta + m1 (1 − t)b, λc + m2 (1 − λ)d) ≤ tα1 λα2 f (a, c) + m2 tα1 (1 − λα2 )f (a, d)
+ m1 (1 − tα1 )λα2 f (b, c) + m1 m2 (1 − tα1 )(1 − λα2 )f (b, d) (1.1)
holds for all t, λ ∈ [0, 1] and (x, y), (z, w) ∈ [0, b] × [0, d].
Now we recite some integral inequalities of Hermite-Hadamard type for the
above-mentioned convex functions.
Theorem 1.1 ([6]). Let f : R0 → R be m-convex and m ∈ (0, 1]. If f ∈ L([a, b]) for
0 ≤ a < b < ∞, then
Z b  
1 f (a) + mf (b/m) mf (a/m) + f (b)
f (x) d x ≤ min , .
b−a a 2 2
Theorem 1.2 ([7, Theorem 3.1]). Let I ⊇ R0 be an open real interval and let f : I → R
be a differentiable function on I such that f 0 ∈ L([a, b]) for 0 ≤ a < b < ∞. If [f 0 (x)]q
is (α, m)-convex on [a, b] for some given numbers α, m ∈ (0, 1] and q ≥ 1, then
Z b  1−1/q 
f (a) + f (b) 1 b−a 1
− f (x) d x ≤ min v1 [f 0 (a)]q
2 b−a a 2 2
  q 1/q    q 1/q 
b a
+ v2 m f 0 , v2 m f 0 + v1 [f 0 (b)]q ,
m m
 Some new inequalities of Hermite-Hadamard type 511

where  
1 1
v1 = α+ α
(α + 1)(α + 2) 2
and  2 
1 α +α+2 1
v2 = − α .
(α + 1)(α + 2) 2 2
Theorem 1.3 ([4, 5, Theorem 2.2]). Let f : ∆ = [a, b] × [c, d] be convex on the co-
ordinates on ∆ with a < b and c < d. Then
   Z b  
a+b c+d 1 1 c+d
f , ≤ f x, dx
2 2 2 b−a a 2
Z d   
1 a+b
+ f ,y dy
d−c c 2
Z bZ d
1
≤ f (x, y) d y d x
(b − a)(d − c) a c
 Z b Z b 
1 1
≤ f (x, c) d x + f (x, d) d x
4 b−a a a
Z d Z d 
1
+ f (a, y) d y + f (b, y) d y
d−c c c
1 
≤ f (a, c) + f (b, c) + f (a, d) + f (b, d) .
4
For more information on this topic, please refer to [1, 2, 10, 11, 12, 13, 14, 15]
and closely related references therein.
In this paper, we will introduce a new concept “(log, (α, m))-convex function on
the co-ordinates” and establish some integral inequalities of Hermite-Hadamard type
for functions whose derivatives are of “co-ordinated (log, (α, m))-convexity”.

2. A definition and a lemma

Motivated by Definitions 1.2 to 1.4, we introduce the notion “co-ordinated
(log, (α, m))-convex function”.
Definition 2.1. A mapping f : [0, b] × [c, d] → R+ = (0, +∞) is called co-ordinated
(log, (α, m))-convex on [0, b] × [c, d] for b > 0 and c, d ∈ R with c < d, if
f (tx + (1 − t)z, λy + m(1 − λ)w) ≤ [λα f (x, y)
+ m(1 − λα )f (x, w)]t [λα f (z, y) + m(1 − λα )f (z, w)]1−t (2.1)
holds for all t, λ ∈ [0, 1], for all (x, y), (z, w) ∈ [0, b] × [c, d], and for all m, α ∈ (0, 1].
Remark 2.1. It is clear that, for all t, λ ∈ [0, 1] and (x, y), (z, w) ∈ [0, b] × [c, d] and
for some m, α ∈ (0, 1],
[λα f (x, y) + m(1 − λα )f (x, w)]t [λα f (z, y) + m(1 − λα )f (z, w)]1−t
≤ tλα f (x, y) + mt(1 − λα )f (x, w) + (1 − t)λα f (z, y) + m(1 − t)(1 − λα )f (z, w).
512 Bo-Yan Xi and Feng Qi

If the function f is co-ordinated (log, (α, m))-convex on [0, b] × [c, d], then, by
taking (α1 , m1 ) = (1, 1) and (α2 , m2 ) = (α, m) in Definition 1.7, we easily see that it
is also co-ordinated (1, 1)-(α, m)-convex on [0, b] × [c, d].
In order to prove our main results, we need the following lemma.
Lemma 2.1. Let f : ∆ = [a, b] × [c, d] ⊆ R2 → R have partial derivatives of the second
∂2f
order. If ∂x∂y ∈ L(∆), then

4 9f (a, c) − 3f (a, d) − 3f (b, c) + f (b, d)
S(f ) ,
(b − a)(d − c) 4
Z b Z d
1 1
− [3f (x, c) − f (x, d)] d x − [3f (a, y) − f (b, y)] d y
2(b − a) a 2(d − c) c
Z bZ d 
1
+ f (x, y) d y d x
(b − a)(d − c) a c
Z 1Z 1
00
= (1 + 2t)(1 + 2λ)fxy (ta + (1 − t)b, λc + (1 − λ)d) d t d λ. (2.2)
0 0

Proof. By integration by parts, we have

Z 1Z 1
00
(1 + 2t)(1 + 2λ)fxy (ta + (1 − t)b, λc + (1 − λ)d) d t d λ
0 0
Z 1 
1 t=1
=− (1 + 2λ) (1 + 2t)fy0 (ta + (1 − t)b, λc + (1 − λ)d) t=0
b−a 0
Z 1 
−2 fy0 (ta + (1 − t)b, λc + (1 − λ)d) d t d λ
0
Z 1 h
1
=− 3(1 + 2λ)fy0 (a, λc + (1 − λ)d)
b−a 0
i
−(1 + 2λ)fy0 (b, λc + (1 − λ)d) d λ
Z 1 Z 1 
−2 (1 + 2λ)fy0 (ta + (1 − t)b, λc + (1 − λ)d) d t d λ
0 0

1
= 3(1 + 2λ)f (a, λc + (1 − λ)d)
(b − a)(d − c)
λ=1
−(1 + 2λ)f (b, λc + (1 − λ)d) λ=0
Z 1 Z 1
−6 f (a, λc + 1 − λ)d) d λ + 2 f (b, λc + (1 − λ)d) d λ
0 0
Z 1
λ=1
−2 (1 + 2λ)f (ta + (1 − t)b, λc + (1 − λ)d) λ=0
dt
0
Z 1 Z 1 
+4 f (ta + (1 − t)b, λc + (1 − λ)d) d t d λ
0 0
 Some new inequalities of Hermite-Hadamard type 513

1
= 9f (a, c) − 3f (b, c) − 3f (a, d) + f (b, d)
(b − a)(d − c)
Z 1 Z 1
−6 f (a, λc + (1 − λ)d) d λ + 2 f (b, λc + (1 − λ)d) d λ
0 0
Z 1 Z 1
−6 f (ta + (1 − t)b, c) d t + 2 f (ta + (1 − t)b, d) d t
0 0
Z 1Z 1 
+4 f (ta + (1 − t)b, λc + (1 − λ)d) d t d λ .
0 0
After further making use of the substitutions x = ta + (1 − t)b and y = λc + (1 − λ)d)
for t, λ ∈ [0, 1], we obtain (2.2). Lemma 2.1 is thus proved. 

3. Some integral inequalities of Hermite-Hadamard type

Now we turn our attention to establish inequalities of Hermite-Hadamard type
for (log, (α, m))-convex functions on the co-ordinates.
Theorem 3.1. Let f : R0 × R → R be a partial differentiable mapping on R0 × R and
00 b 00 q
fxy ∈ L1 ([a, m ] × [c, d]) with 0 ≤ a < b and c < d for some fixed m ∈ (0, 1]. If fxy
b
is co-ordinated (log, (α, m))-convex on [0, m ] × [c, d] for q ≥ 1 and α ∈ (0, 1], then

22(1−1/q)

00 q
|S(f )| ≤ 7(3α + 4) fxy (a, c) + 7mα(2α
[6(α + 1)(α + 2)]1/q
 q   q 1/q
00 d 00 q 00 d
+ 3) fxy a, + 5(3α + 4) fxy (b, c) + 5m(2α + 3) fxy b, .
m m
Proof. By Lemma 2.1, Hölder’s integral inequality, the (log, (α, m))-convexity of
00 q
fxy , and the GA-inequality, we obtain
Z 1Z 1
00
|S(f )| ≤ (1 + 2t)(1 + 2λ) fxy (ta + (1 − t)b, λc + (1 − λ)d) d t d λ
0 0
Z 1 Z 1 1−1/q Z 1 Z 1
≤ (1 + 2t)(1 + 2λ) d t d λ (1 + 2t)(1 + 2λ)
0 0 0 0
1/q
00 q
× fxy f (ta + (1 − t)b, λc + (1 − λ)d) d t d λ
Z 1 Z 1 
00 q
≤ 22(1−1/q) (1 + 2t)(1 + 2λ) λα fxy (a, c)
0 0
  q t 
00 d 00 q
+m(1 − λα ) fxy a, λα fxy (b, c)
m
  q 1−t 1/q
α 00 d
+m(1 − λ ) fxy b, dtdλ
m
Z 1 Z 1 
00 q
≤ 22(1−1/q) (1 + 2t)(1 + 2λ) tλα fxy (a, c)
0 0
514 Bo-Yan Xi and Feng Qi
  q
α 00 d 00 q
+mt(1 − λ ) fxy a, + (1 − t)λα fxy (b, c)
m
  q 1/q
d α 00
+m(1 − t)(1 − λ ) b, fxy
dtdλ
m
22(1−1/q)

00 q
= 7(3α + 4) fxy (a, c)
[6(α + 1)(α + 2)]1/q
 q
00 b
+7mα(2α + 3) fxy ,c
m
  q 1/q
00 q 00 b
+5(3α + 4) fxy (a, d) + 5m(2α + 3) fxy ,d .
m
This completes the proof of Theorem 3.1. 

Corollary 3.1.1. Under the assumptions of Theorem 3.1, if q = 1, we have

  
1 00 00 d
|S(f )| ≤ 7(3α + 4) fxy (a, c) + 7mα(2α + 3) fxy a,
6(α + 1)(α + 2) m
 
00 00 d
+ 5(3α + 4) fxy (b, c) + 5m(2α + 3) fxy b, .
m
Corollary 3.1.2. Under the assumptions of Corollary 3.1.1,
1. if m = 1, then
1 h
00 00
|S(f )| ≤ 7(3α + 4) fxy (a, c) + 7α(2α + 3) fxy (a, d)
6(α + 1)(α + 2)
i
00 00
+ 5(3α + 4) fxy (b, c) + 5(2α + 3) fxy (b, d) ;

2. if α = 1, then
  
1 00 00 d
|S(f )| ≤ 49 fxy (a, c) + 35m fxy a,
36 m
 
00 00 d
+ 35 fxy (b, c) + 25m fxy b, ;
m
3. if m = α = 1, then
1h 00 00 00 00
i
|S(f )| ≤ 49 fxy (a, c) + 35 fxy (a, d) + 35 fxy (b, c) + 25 fxy (b, d) .
36
Corollary 3.1.3. Under the assumptions of Theorem 3.1,
1. if m = 1, then

22(1−1/q) h
00 q
|S(f )| ≤ 7(3α + 4) fxy (a, c)
[6(α + 1)(α + 2)]1/q
q 1/q
i
00 q 00 00 q
+ 7α(2α + 3) fxy (b, c) + 5(3α + 4) fxy (a, d) + 5(2α + 3) fxy (b, d) ;
 Some new inequalities of Hermite-Hadamard type 515

2. if α = 1, then
  q
4 00 q 00 b
|S(f )| ≤ 49 fxy (a, c) + 35m f xy , c
122/q m
  q 1/q
00 q 00 b
+ 35 fxy (a, d) + 25m fxy ,d .
m
Theorem 3.2. Let f : R0 × R → R be a partial differentiable mapping on R0 × R and
00 b 00 q
fxy ∈ L1 ([a, m ] × [c, d]) for 0 ≤ a < b, c < d and some fixed m ∈ (0, 1]. If fxy is
b
co-ordinated (log, (α, m))-convex on [0, m ] × [c, d] for q > 1 and some α ∈ (0, 1] with
q ≥ r > −1, then
 (2q−r−1)/(q−1) 1−1/q
(3 − 1)(q − 1)
|S(f )| ≤
2q − r − 1
 1/q 
1
× (2r3r+1 + 3r+1 + 1)
4(α + 1)(α + 2)(r + 1)(r + 2)
   q
00 q 00 d
× (3α + 4) fxy (a, c) + mα(2α + 3) fxy a, + (3r+2 − 5
m
   q 1/q
00 q 00 d
− 2r) (3α + 4) fxy (b, c) + m(2α + 3) fxy b, .
m
Proof. By Lemma 2.1, Hölder’s integral inequality, the (log, (α, m))-convexity of
00 q
fxy , and the well known GA-inequality, we obtain
Z 1Z 1
00
|S(f )| ≤ (1 + 2t)(1 + 2λ) fxy (ta + (1 − t)b, λc + (1 − λ)d) d t d λ
0 0
Z 1Z 1 1−1/q Z 1 Z 1
(q−r)/(q−1)
≤ (1 + 2t) (1 + 2λ) d t d λ (1 + 2t)r
0 0 0 0
1/q
00 q
×(1 + 2λ) fxy f (ta + (1 − t)b, λc + (1 − λ)d) d t d λ
1−1/q Z 1 Z 1
(3(2q−r−1)/(q−1) − 1)(q − 1)

≤ (1 + 2t)r (1 + 2λ)
2q − r − 1 0 0
   q t
00 q 00 d
× λα fxy (a, c) + m(1 − λα ) fxy a,
m
   q 1−t 1/q
00 q 00 d
× λα fxy (b, c) + m(1 − λα ) fxy b, dtdλ
m
 (2q−r−1)/(q−1) 1−1/q Z 1 Z 1
(3 − 1)(q − 1)
≤ (1 + 2t)r
2q − r − 1 0 0
  q
00 q 00 d
×(1 + 2λ) tλα fxy (a, c) + mt(1 − λα ) fxy a,
m
516 Bo-Yan Xi and Feng Qi

  q 1/q
00 q 00 d
+(1 − t)λα fxy (b, c) + m(1 − t)(1 − λα ) fxy b, dtdλ
m
 (2q−r−1)/(q−1) 1−1/q
(3 − 1)(q − 1)
=
2q − r − 1
 1/q 
1
× (2r3r+1 + 3r+1 + 1)
4(α + 1)(α + 2)(r + 1)(r + 2)
   q
00 q 00 d
× (3α + 4) fxy (a, c) + mα(2α + 3) fxy a, + (3r+2 − 5
m
   q 1/q
00 q 00 d
−2r) (3α + 4) fxy (b, c) + m(2α + 3) fxy b, .
m
The proof of Theorem 3.2 is complete. 

Corollary 3.2.1. Under the conditions of Theorem 3.2, if r = 0, we have

 (2q−1)/(q−1) 1−1/q  1/q
(3 − 1)(q − 1) 1
|S(f )| ≤
2q − 1 2(α + 1)(α + 2)
  q
00 q 00 d
× (3α + 4) fxy (a, c) + mα(2α + 3) fxy a,
m
  q 1/q
00 q 00 d
+ (3α + 4) fxy (b, c) + m(2α + 3) fxy b, .
m
Corollary 3.2.2. Under the conditions of Theorem 3.2,
1. if m = 1, then
 (2q−r−1)/(q−1) 1−1/q
(3 − 1)(q − 1)
|S(f )| ≤
2q − r − 1
 1/q 
1 h
× (2r3r+1 + 3r+1 + 1) (3α
4(α + 1)(α + 2)(r + 1)(r + 2)
i
00 q 00 q
+ 4) fxy (a, c) + α(2α + 3) fxy (a, d)
h i1/q
00 q 00 q
+ (3r+2 − 5 − 2r) (3α + 4) fxy (b, c) + (2α + 3) fxy (b, d) ;

2. if α = 1, then
 (2q−r−1)/(q−1) 1−1/q  1/q
(3 − 1)(q − 1) 1
|S(f )| ≤
2q − r − 1 24(r + 1)(r + 2)
    q
00 q 00 b
× (2r3r+1 + 3r+1 + 1) 7 fxy (a, c) + 5m fxy ,c
m
   q 1/q
00 q 00 b
+ (3r+2 − 5 − 2r) 7 fxy (a, d) + 5m fxy ,d ;
m
 Some new inequalities of Hermite-Hadamard type 517

3. if m = α = 1, then
 (2q−r−1)/(q−1) 1−1/q  1/q
(3 − 1)(q − 1) 1
|S(f )| ≤
2q − r − 1 24(r + 1)(r + 2)
 h i
00 q 00 q
× (2r3r+1 + 3r+1 + 1) 7 fxy (a, c) + 5 fxy (b, c)
h i1/q
00 q 00 q
+ (3r+2 − 5 − 2r) 7 fxy (a, d) + 5 fxy (b, d) .

Theorem 3.3. Let f : R0 × R → R be a partial differentiable mapping on R0 × R and

00 b 00 q
fxy ∈ L1 ([a, m ] × [c, d]) for 0 ≤ a < b, c < d and some fixed m ∈ (0, 1]. If fxy is
b
co-ordinated (log, (α, m))-convex on [0, m ] × [c, d] for q > 1 and some α ∈ (0, 1], then
 (2q−1)/(q−1) 2(1−1/q)  1/q 
(3 − 1)(q − 1) 1 00 q
|S(f )| ≤ fxy (a, c)
2q − 1 2(α + 1)
 q   q 1/q
00 b 00 q 00 d
+ mα fxy a, + fxy (b, c) + mα fxy b, .
m m
Proof. By Lemma 2.1, Hölder’s integral inequality, the (log, (α, m))-convexity of
00 q
fxy , and the GA-inequality, we obtain
Z 1Z 1
00
|S(f )| ≤ (1 + 2t)(1 + 2λ) fxy (ta + (1 − t)b, λc + (1 − λ)d) d t d λ
0 0
Z 1 Z 1 1−1/q
≤ (1 + 2t)q/(q−1) (1 + 2λ)q/(q−1) d t d λ
0 0
Z 1 Z 1 1/q
00 q
× fxy f (ta + (1 − t)b, λc + (1 − λ)d) d t d λ
0 0
2(1−1/q)
(3(2q−1)/(q−1) − 1)(q − 1)

≤
2q − 1
Z 1 Z 1   q t
α 00 q α 00 d
× λ fxy (a, c) + m(1 − λ ) fxy a,
0 0 m
   q 1−t 1/q
00 q 00 d
× λα fxy (b, c) + m(1 − λα ) fxy b, dtdλ
m
 (2q−1)/(q−1) 2(1−1/q)
(3 − 1)(q − 1)
≤
2q − 1
Z 1 Z 1   q
α 00 q α 00 d
× tλ fxy (a, c) + mt(1 − λ ) fxy a,
0 0 m
  q 1/q
α 00 q α 00 d
+(1 − t)λ fxy (b, c) + m(1 − t)(1 − λ ) fxy b, dtdλ
m
 (2q−1)/(q−1) 2(1−1/q)  1/q 
(3 − 1)(q − 1) 1 00 q
= fxy (a, c)
2q − 1 2(α + 1)
518 Bo-Yan Xi and Feng Qi

 q   q 1/q
00 b 00 q 00 d
+mα fxy a, + fxy (b, c) + mα fxy b, .
m m
The proof of Theorem 3.3 is complete. 

Corollary 3.3.1. Under the conditions of Theorem 3.3, if m = α = 1, then

2(1−1/q)  1/q h
(3(2q−1)/(q−1) − 1)(q − 1)

1 00 q
|S(f )| ≤ fxy (a, c)
2q − 1 4
q 1/q
i
00 q
00 q
00
+ fxy (a, d) + fxy (b, c) + fxy (b, d) .

Theorem 3.4. Let f : R0 × R → R+ be integrable on [0, mb2 ] × [c, d] for 0 ≤ a < b,

c < d, and some m ∈ (0, 1]. If f is co-ordinated (log, (α, m))-convex on [0, mb2 ] × [c, d]
for α ∈ (0, 1], then
Z dZ b
1
f (x, y) d x d y
(b − a)(d − c) c a
Z b  
1 α d 1
≤ f (x, c) + m(2 − 1)f x, d x + α+1
2(α + 1)(b − a) a m 2 (d − c)
Z d     
y y
× L f (a, y) + m(2α − 1)f a, , f (b, y) + m(2α − 1)f b, dy
c m m
Z b  
1 d 1
≤ f (x, c) + m(2α − 1)f x, d x + α+2
2(α + 1)(b − a) a m 2 (d − c)
Z d     
α y y
× f (a, y) + f (b, y) + m(2 − 1) f a, + f b, dy
c m m
   
1 c
≤ α L f (a, c) + m(2α − 1)f a, , f (b, c)
2 (α + 1) m
      
α c α d α d
+ m(2 − 1)f b, + m(2 − 1)L f a, + m(2 − 1)f a, 2 ,
m m m
   
d d 1
f b, + m(2α − 1)f b, 2 ≤ α+1
m m 2 (α + 1)
     
α c c
× f (a, c) + f (b, c) + m(2 − 1) f a, + f b, + m(2α − 1)
m m
         
d d α d d
× f a, + f b, + m(2 − 1) f a, 2 + f b, 2 ,
m m m m

where L(u, v) is the logarithmic mean defined by

 v−u ,

u 6= v,
L(u, v) = ln v − ln u
u, u = v.
 Some new inequalities of Hermite-Hadamard type 519

Proof. Putting y = λc + (1 − λ)d for 0 ≤ λ ≤ 1 and using the (log, (α, m))-convexity
of f , we have
 
α α d
f (x, y) = f (x, λc + (1 − λ)d) ≤ λ f (x, c) + m(1 − λ )f x, (3.1)
m
for all (x, y) ∈ [a, b] × [c, d], t = 12 , and 0 ≤ λ ≤ 1.
Similarly, setting x = ta + (1 − t)b for 0 ≤ t ≤ 1 and using the (log, (α, m))-
convexity of f with 0 ≤ t ≤ 1 and λ = 12 in (2.1), we obtain

f (x, c) = f (ta + (1 − t)b, c)

  t   1−t
1 c c
≤ α f (a, c) + m(2α − 1)f a, f (b, c) + m(2α − 1)f b,
2 m m
and
      
d d 1 d
f x, = f ta + (1 − t)b, ≤ α f a,
m m 2 m
 t     1−t
d d d
+ m(2α − 1)f a, 2 f b, + m(2α − 1)f b, 2 . (3.2)
m m m
From inequalities (3.1) to (3.2), we have
Z dZ b Z 1 Z b
1 1
f (x, y) d x d y ≤ λα f (x, c) + m(1
(b − a)(d − c) c a b−a 0 a
  Z b
α d 1
− λ )f x, dxdλ = f (x, c) + m(2α − 1)
m (α + 1)(b − a) a
  Z 1   t
d 1 α c
× f x, dx ≤ α f (a, c) + m(2 − 1)f a, (3.3)
m 2 (α + 1) 0 m
  1−t   
c d
× f (b, c) + m(2α − 1)f b, + m(2α − 1) f a, + m(2α
m m
 t     1−t 
d d α d
− 1)f a, 2 × f b, + m(2 − 1)f b, 2 d t.
m m m
It is obvious that
Z 1
u+v
ut v 1−t d t = L(u, v) and L(u, v) ≤ . (3.4)
0 2
By (3.3) and (3.4), we acquire
Z d Z b
1
f (x, y) d x d y
(b − a)(d − c) c a
     
1 α c α c
≤ α L f (a, c) + m(2 − 1)f a, , f (b, c) + m(2 − 1)f b,
2 (α + 1) m m
        
d d d d
+m(2α − 1)L f a, + m(2α − 1)f a, 2 , f b, + m(2α − 1)f b, 2
m m m m
520 Bo-Yan Xi and Feng Qi
     
1 α c c
≤ f (a, c) + f (b, c) + m(2 − 1) f a, + f b,
2α+1 (α + 1) m m
         
α d d α d d
+m(2 − 1) f a, + f b, + m(2 − 1) f a, 2 + f b, 2 .
m m m m
Similarly, one has
Z dZ b Z dZ 1
1 1
f (x, y) d x d y = f (ta + (1 − t)b, y) d t d y
(b − a)(d − c) c a d−c c 0
Z d Z 1  t 
1 y
≤ α f (a, y) + m(2α − 1)f a, f (b, y) + m(2α − 1)
2 (d − c) c 0 m
 1−t Z d   
y 1 α y
× f b, dtdy = α L f (a, y) + m(2 − 1)f a, ,
m 2 (d − c) c m
  Z d 
y 1
f (b, y) + m(2α − 1)f b, d y ≤ α+1 f (a, y) + f (b, y)
m 2 (d − c) c
     Z 1 
y y 1
+ m(2α − 1) f a, + f b, d y ≤ α+1 λα f (a, c)
m m 2 0
   
d d
+ m(2α − 1)f a, + λα f (b, c) + m(1 − λα )f b, + m(2α − 1)
m m
      
c d c
× λα f a, + m(1 − λα )f a, 2 + λα f b, + m(1 − λα )
m m m
  
d 1
× f b, 2 d λ = α+1 f (a, c) + f (b, c) + m(2α
m 2 (α + 1)
         
c c d d
− 1) f a, + f b, + m(2α − 1) f a, + f b,
m m m m
    
d d
+ m(2α − 1) f a, 2 + f b, 2 .
m m
Theorem 3.4 is thus proved. 
b
Theorem 3.5. Let f : [0, m ] × [c, d] ⊆ R0 × R → R+ be integrable on [0, mb2 ] × [c, d] for
0 ≤ a < b, c < d, and some fixed m ∈ (0, 1]. If f is co-ordinated (log, (α, m))-convex
on [0, mb2 ] × [c, d] for α ∈ (0, 1], then
  Z b    1/2
a+b c+d 1 c+d c+d
f , ≤ α+1 f x, + m(2α − 1)f x,
2 2 2 (b − a) a 2 2m
    1/2
c+d c+d
× f a + b − x, +m(2α − 1)f a + b − x, dx
2 2m
Z d    
1 a+b a+b y
+ α+1 f , y + m(2α − 1)f , dy
2 (d − c) c 2 2 m
Z b    
1 c+d c+d
≤ α+1 f x, + m(2α − 1)f x, dx
2 (b − a) a 2 2m
 Some new inequalities of Hermite-Hadamard type 521

Z d    
1 a+b α a+b y
+ α+1 f , y + m(2 − 1)f , dy
2 (d − c) c 2 2 m
Z d Z b  
1 y
≤ 2α+1 f (x, y) + 2m(2α − 1)f x,
2 (b − a)(d − c) c a m
    1/2
y y
+m2 (2α − 1)2 f x, 2 + f (x, y) + m(2α − 1)f x,
m m
  1/2
y
× f (a + b − x, y) + m(2α − 1)f a + b − x,
m
    1/2
y y
+m(2α − 1) f x, + m(2α − 1)f x, 2
m m
    1/2 
y α y
× f a + b − x, + m(2 − 1)f a + b − x, 2 dxdy
m m
Z d Z b  
1 y
≤ 2α f (x, y) + 2m(2α − 1)f x,
2 (b − a)(d − c) c a m
 
y
+m2 (2α − 1)2 f x, 2 d x d y.
m
Proof. Using the (log, (α, m))-convexity of f , we have
   
a+b c+d 1 c+d
f , =f (ta + (1 − t)b + (1 − t)a + tb),
2 2 2 2
    1/2
1 c+d c+d
≤ α f ta + (1 − t)b, + m(2α − 1)f ta + (1 − t)b,
2 2 2m
    1/2
c+d c+d
× f (1 − t)a + tb, +m(2α − 1)f (1 − t)a + tb,
2 2m
for all t ∈ [0, 1].
Integrating on both sides of the above inequality on [0, 1], from the GA-
inequality, and by the (log, (α, m))-convexity of f , we reveals
  Z 1  
a+b c+d 1 c+d
f , = f (ta + (1 − t)b + (1 − t)a + tb), dt
2 2 0 2 2
Z 1    1/2
1 c+d α c+d
≤ α f ta + (1 − t)b, + m(2 − 1)f ta + (1 − t)b,
2 0 2 2m
    1/2
c+d c+d
× f (1 − t)a + tb, +m(2α − 1)f (1 − t)a + tb, dt
2 2m
Z b    1/2
1 c+d c+d
= α f x, + m(2α − 1)f x,
2 (b − a) a 2 2m
    1/2
c+d c+d
× f a + b − x, +m(2α − 1)f a + b − x, dx
2 2m
522 Bo-Yan Xi and Feng Qi

Z b    
1 c+d α c+d
≤ α f x, + m(2 − 1)f x, dx
2 (b − a) a 2 2m
Z 1 Z b  
1 1
= α f x, [λc + (1 − λ)d + (1 − λ)c + λd]
2 (b − a) 0 a 2
 
α 1
+m(2 − 1)f x, [λc + (1 − λ)d + (1 − λ)c + λd] d x d λ
2m
Z 1 Z b  
1 (1 − λ)c + λd
≤ 2α f (x, λc + (1 − λ)d) + m(2α − 1)f x,
2 (b − a) 0 a m
    
λc + (1 − λ)d (1 − λ)c + λd
+m(2α − 1) f x, + m(2α − 1)f x, dxdλ
m m2
Z d Z b
1
= 2α f (x, y)
2 (b − a)(d − c) c a
   
y y
+2m(2α − 1)f x, + m2 (2α − 1)2 f x, 2 d x d y.
m m
Similarly, we obtain
  Z 1  
a+b c+d 1 a+b
f , ≤ α f , λc + (1 − λ)d
2 2 2 0 2
 
α a + b λc + (1 − λ)d
+m(2 − 1)f , dλ
2 m
Z d     
1 a+b α a+b y
= α f , y + m(2 − 1)f , dy
2 (d − c) c 2 2 m
Z d Z 1   1/2
1 α y
≤ 2α f (ta + (1 − t)b, y) + m(2 − 1)f ta + (1 − t)b,
2 (d − c) c 0 m
  1/2
y
× f ((1 − t)a + tb, y) + m(2α − 1)f (1 − t)a + tb,
m
    1/2
α y α y
+m(2 − 1) f ta + (1 − t)b, + m(2 − 1)f ta + (1 − t)b, 2
m m
    1/2 
y y
× f (1 − t)a + tb, + m(2α − 1)f (1 − t)a + tb, 2 dtdy
m m
Z d Z b   1/2
1 y
= 2α f (x, y) + m(2α − 1)f x,
2 (b − a)(d − c) c a m
  1/2
y
× f (a + b − x, y) + m(2α − 1)f a + b − x,
m
    1/2
y y
+m(2α − 1) f x, + m(2α − 1)f x, 2
m m
    1/2 
y y
× f a + b − x, + m(2α − 1)f a + b − x, 2 dxdy
m m
 Some new inequalities of Hermite-Hadamard type 523

Z d Z b  
1 αy
≤ 2α f (x, y) + 2m(2 − 1)f x,
2 (b − a)(d − c) c a m
 
y
+m2 (2α − 1)2 f x, 2 d x d y.
m
The proof of Theorem 3.5 is complete. 
Corollary 3.5.1. Under the conditions of Theorems 3.4 and 3.5, if m = 1, then
   Z b    1/2
a+b c+d 1 1 c+d c+d
f , ≤ f x, f a + b − x, dx
2 2 2 b−a a 2 2
Z d   
1 a+b
+ f ,y dy
d−c c 2
 Z b   Z d   
1 1 c+d 1 a+b
≤ f x, dx + f ,y dy
2 b−a a 2 d−c c 2
Z dZ b
1
≤ f (x, y) d x d y
(b − a)(d − c) c a
Z b
1
f (x, c) + (2α − 1)f (x, d) d x
 
≤
2(α + 1)(b − a) a
Z d
1

+ L f (a, y), f (b, y) d y
2(d − c) c
Z b
1
f (x, c) + (2α − 1)f (x, d) d x
 
≤
2(α + 1)(b − a) a
Z d
1  
+ f (a, y) + f (b, y) d y
4(d − c) c
1 n
L f (a, c), f (b, c) + (2α − 1)L f (a, d), f (b, d)



≤
2(α + 1)
o
+ f (a, c) + f (b, c) + (2α − 1) f (a, d) + f (b, d)


1 n o
f (a, c) + f (b, c) + (2α − 1) f (a, d) + f (b, d) .

≤
2(α + 1)
If m = α = 1, then
   Z b    1/2
a+b c+d 1 1 c+d c+d
f , ≤ f x, f a + b − x, dx
2 2 2 b−a a 2 2
Z d   
1 a+b
+ f ,y dy
d−c c 2
 Z b   Z d   
1 1 c+d 1 a+b
≤ f x, dx + f ,y dy
2 b−a a 2 d−c c 2
Z dZ b
1
≤ f (x, y) d x d y
(b − a)(d − c) c a
524 Bo-Yan Xi and Feng Qi

Z b Z d
1   1

≤ f (x, c) + f (x, d) d x + L f (a, y), f (b, y) d y
4(b − a) a 2(d − c) c
 Z b Z d 
1 1   1  
≤ f (x, c) + f (x, d) d x + f (a, y) + f (b, y) d y
4 b−a a d−c c
1 n
≤ f (a, c) + f (a, d) + f (b, c) + f (b, d)
4


o
+L f (a, c), f (b, c) + L f (a, d), f (b, d)
1 
≤ f (a, c) + f (a, d) + f (b, c) + f (b, d) .
4

References
[1] Bai, R.-F., Qi, F., Xi, B.-Y., Hermite-Hadamard type inequalities for the m- and (α, m)-
logarithmically convex functions, Filomat, 27(2013), no. 1, 1-7.
[2] Bai, S.-P., Wang, S.-H., Qi, F., Some Hermite-Hadamard type inequalities for n-time
differentiable (α, m)-convex functions, J. Inequal. Appl., 2012:267, 11 pages.
[3] Chun, L., Some new inequalities of Hermite-Hadamard type for (α1 , m1 )-(α2 , m2 )-convex
functions on coordinates, J. Funct. Spaces, 2014(2014), Article ID 975950, 7 pages.
[4] Dragomir, S.S., On the Hadamard’s inequality for convex functions on the co-ordinates
in a rectangle from the plane, Taiwanese J. Math., 5(2001), no. 4, 775-788.
[5] Dragomir, S.S., Pearce, C.E.M., Selected Topics on Hermite-Hadamard Inequalities and
Applications, RGMIA Monographs, Victoria University (2000).
[6] Dragomir, S.S., Toader, G., Some inequalities for m-convex functions, Studia Univ.
Babeş-Bolyai Math., 38(1993), no. 1, 21-28.
[7] Bakula, M.K., Özdemir, M.E., Pečarić, J., Hadamard type inequalities for m-convex and
(α, m)-convex functions, J. Inequal. Pure Appl. Math., 9(2008), no. 4, Art. 96, 12 pages.
[8] Toader, G., Some generalizations of the convexity, Proceedings of the Colloquium on
Approximation and Optimization, Cluj-Napoca, 1985, 329-338.
[9] Miheşan, V.G., A generalization of the convexity, Seminar on Functional Equations,
Approx. Convex, Cluj-Napoca, 1993 (Romanian).
[10] Qi, F., Wei, Z.-L., Yang, Q., Generalizations and refinements of Hermite-Hadamard’s
inequality, Rocky Mountain J. Math., 35(2005), no. 1, 235-251.
[11] Qi, F., Xi, B.-Y., Some integral inequalities of Simpson type for GA-ε-convex functions,
Georgian Math. J., 20(2013), no. 4, 775-788.
[12] Xi, B.-Y., Bai, R.-F., Qi, F., Hermite-Hadamard type inequalities for the m- and (α, m)-
geometrically convex functions, Aequationes Math., 84(2012), no. 3, 261-269.
[13] Xi, B.-Y., Qi, F., Some Hermite-Hadamard type inequalities for differentiable convex
functions and applications, Hacet. J. Math. Stat., 42(2013), no. 3, 243-257.
[14] Xi, B.-Y., Qi, F., Some integral inequalities of Hermite-Hadamard type for convex func-
tions with applications to means, J. Funct. Spaces Appl., 2012(2012), Article ID 980438,
14 pages.
[15] Xi, B.-Y., Qi, F., Integral inequalities of Simpson type for logarithmically convex func-
tions, Adv. Stud. Contemp. Math., Kyungshang, 23(2013), no. 4, 559-566.
 Some new inequalities of Hermite-Hadamard type 525

Bo-Yan Xi
College of Mathematics
Inner Mongolia University for Nationalities
Tongliao City, 028043
Inner Mongolia Autonomous Region, China
e-mail: [email protected], [email protected], [email protected]
Feng Qi
Department of Mathematics, School of Science
Tianjin Polytechnic University
Tianjin City, 300387, China;
Institute of Mathematics
Henan Polytechnic University
Jiaozuo City, 454010,
Henan Province, China
e-mail: [email protected], [email protected], [email protected]
Stud. Univ. Babeş-Bolyai Math. 60(2015), No. 4, 527–534

Some Hermite-Hadamard type inequalities

for functions whose exponentials are convex
Silvestru Sever Dragomir and Ian Gomm

Abstract. Some inequalities of Hermite-Hadamard type for functions whose ex-

ponentials are convex are obtained.
Mathematics Subject Classification (2010): 26D15, 25D10.
Keywords: Convex functions, Hermite-Hadamard inequality, special means.

1. Introduction
The following integral inequality
  Z b
a+b 1 f (a) + f (b)
f ≤ f (t) dt ≤ , (1.1)
2 b−a a 2
which holds for any convex function f : [a, b] → R, is well known in the literature as
the Hermite-Hadamard inequality.
There is an extensive amount of literature devoted to this simple and nice result
which has many applications in the Theory of Special Means and in Information
Theory for divergence measures, from which we would like to refer the reader to the
monograph [1] and the references therein.
We denote by Expconv (I) the class of all functions defined on the interval I
of real numbers such that exp (f ) is convex on I. If Conv (I) is the class of convex
functions defined on I then we have the following fact:
Proposition 1.1. We have the strict inclusion
Conv (I) Expconv (I) .
Proof. If f is convex , then exp(f ) is log-convex and therefore convex on I and the
inclusion is proved.
For r ≥ 1 the function fr (x) = r ln x, x > 0 is concave on (0, ∞) . We have
exp (fr (x)) = xr is a convex function, therefore fr ∈ Expconv (I) \ Conv (I) . 
528 Silvestru Sever Dragomir and Ian Gomm

We observe that for twice differentiable functions g on ˚ I, the interior of I we

have that  
00 2
(exp (g (x))) = [g 0 (x)] + g 00 (x) exp g (x) , x ∈ ˚I,

therefore g ∈ Expconv (I) if and only if

2
[g 0 (x)] + g 00 (x) ≥ 0 for any x ∈ ˚
I.

2. Some Hermite-Hadamard type inequalities

Now, if g ∈ Expconv (I), then by the Hermite-Hadamard inequality for exp (g)
we have for a, b ∈ I with a < b that
  Z b
a+b 1 1
exp g ≤ exp g (t) dt ≤ [exp g (a) + exp g (b)] . (2.1)
2 b−a a 2
By Jensen’s integral inequality for the exp function we also have for any integrable
function h : [a, b] → R that
Z b ! Z b
1 1
exp h (t) dt ≤ exp h (t) dt. (2.2)
b−a a b−a a

We define the logarithmic mean as


 a if a = b,
L = L (a, b) := a, b > 0.
 b−a
ln b−ln a if a 6= b,

We can improve the inequality (2.1) for convex functions as follows:

Theorem 2.1. Let f : I → R be a convex function on I and a, b ∈ I with a < b. Then

we have for f (b) 6= f (a) the inequalities
  Z b ! Z b
a+b 1 1
exp f ≤ exp f (t) dt ≤ exp f (t) dt (2.3)
2 b−a a b−a a
 
exp f (b) − exp f (a) 1
≤ ≤ [exp f (a) + exp f (b)] .
f (b) − f (a) 2
Proof. The first inequality follows by Hermite-Hadamard inequality for the convex
function f . The second inequality follows by (2.2).
It is know that if g is log convex, then by [2]
Z b
1
g (t) dt ≤ L (g (a) , g (b)) . (2.4)
b−a a
Since f is convex, then g = exp (f ) is log-convex and by (2.4) we get the third
inequality in (2.3). 
 Some Hermite-Hadamard type inequalities 529

A recent paper connected with such results is [4].

Consider the identric mean of two positive numbers

 a if a = b,

I = I (a, b) := 1 a, b > 0.
 1  bb  b−a

e aa if a 6= b,

We observe that
Z b
1
ln I (a, b) = ln udu
b−a a
for a, b > 0, a 6= b.
The following result holds:
Theorem 2.2. Assume that f ∈ Expconv (I) and a, b ∈ I with a < b. Then we have
Z b !
1
exp f (t) dt ≤ I (exp f (a) , exp f (b)) (2.5)
b−a a
and
 
a+b
exp f (2.6)
2
Z b   !
1 exp f (x) + exp f (a + b − x)
≤ exp ln dx
b−a a 2
Z b
1
≤ exp f (x) dx.
b−a a

Proof. Since f ∈ Expconv (I) , then

exp f ((1 − λ) a + λb) ≤ (1 − λ) exp f (a) + λ exp f (b)
for any λ ∈ [0, 1] , which is equivalent to
f ((1 − λ) a + λb) ≤ ln [(1 − λ) exp f (a) + λ exp f (b)] (2.7)
for any λ ∈ [0, 1] .
Integrating (2.7) on [0, 1] we get
Z b Z 1
1
f (t) dt = f ((1 − λ) a + λb) dλ (2.8)
b−a a 0
Z 1
≤ ln [(1 − λ) exp f (a) + λ exp f (b)] dλ
0
Z exp f (b)
1
= ln udu
exp f (b) − exp f (a) exp f (a)
= ln I (exp f (a) , exp f (b))
and the inequality in (2.5) is proved.
530 Silvestru Sever Dragomir and Ian Gomm

From (2.7) we have

   
x+y exp f (x) + exp f (y)
f ≤ ln (2.9)
2 2
for any x, y ∈ I.
From (2.9) we have
   
a+b exp f (x) + exp f (a + b − x)
f ≤ ln (2.10)
2 2

for any x ∈ [a, b] .

Integrating the inequality (2.10) over x on [a, b] we get the first inequality in
(2.6).
By the Jensen’s inequality for the concave function ln we have
Z b  
1 exp f (x) + exp f (a + b − x)
ln dx (2.11)
b−a a 2
Z b  !
1 exp f (x) + exp f (a + b − x)
≤ ln dx
b−a a 2
Z b !
1
= ln [exp f (x) + exp f (a + b − x)] dx
2 (b − a) a
Z b !
1
= ln exp f (x) dx
b−a a

and the second inequality in (2.6) is proved. 

If we consider Toader’s mean defined as (see for instance [5] and [7] for many
relations of this mean with other means)

 a if a = b,
E = E (a, b) := a, b ∈ R.
log I (exp a, exp b) if a 6= b,


we can write (2.5) in an equivalent form as

Z b
1
f (t) dt ≤ E (exp f (a) , exp f (b)) . (2.12)
b−a a

Remark 2.3. If the function g : I → (0, ∞) is convex on I, then f = ln g ∈ Expconv (I)

and for a, b ∈ I with a < b we have, by (2.5) and (2.6), the following inequalities
Z b !
1
exp ln g (t) dt ≤ I (g (a) , g (b)) (2.13)
b−a a
 Some Hermite-Hadamard type inequalities 531

and
  Z b   !
a+b 1 g (x) + g (a + b − x)
g ≤ exp ln dx (2.14)
2 b−a a 2
Z b
1
≤ g (x) dx.
b−a a

3. Related results
The following related result also holds:
Theorem 3.1. Assume that f ∈ Expconv (I) and a, b ∈ I with a < b. Then we have
Z b
f (a) (x − a) + f (b) (b − x) 1
− f (y) dy (3.1)
b−a b−a a
" Z b #
1
≥ exp f (x) exp (−f (x)) − exp [−f (y)] dy
b−a a
for any x ∈ [a, b] .
In particular, we have
Z b
f (a) + f (b) 1
− f (y) dy (3.2)
2 b−a a
  "    Z b #
a+b a+b 1
≥ exp f exp −f − exp [−f (y)] dy .
2 2 b−a a
Proof. Since the function exp (f ) is convex, it has lateral derivatives in each point of
(a, b) and f = ln (exp f ) does the same. Then for any x, y ∈ (a, b) we have
0
exp f (x) − exp f (y) ≥ f− (y) (x − y) exp f (y)
and dividing by exp f (y) > 0 we get
0
exp f (x) exp [−f (y)] − 1 ≥ f− (y) (x − y) (3.3)
for any x, y ∈ (a, b) .
Integrating (3.3) over y on [a, b] and dividing by b − a we get
Z b
1
exp f (x) exp [−f (y)] dy − 1 (3.4)
b−a a
Z b
1
≥ f 0 (y) (x − y) dy
b−a a −
" Z b #
1 b
= f (y) (x − y)|a + f (y) dy
b−a a
"Z #
b
1
= f (y) dy − f (a) (x − a) − f (b) (b − x)
b−a a
for any x ∈ [a, b] , which is equivalent to the desired inequality (3.1). 
532 Silvestru Sever Dragomir and Ian Gomm

Corollary 3.2. With the assumptions of Theorem 3.1 we have

Z b
f (a) + f (b) 1
− f (y) dy (3.5)
2 b−a a
" Z b #
exp f (a) + exp f (b) 1
≥ 1− exp [−f (y)] dy .
2 b−a a

Proof. If we take x = a and x = b in (3.4) we get

Z b Z b
1 1
exp f (a) exp [−f (y)] dy − 1 ≥ f (y) dy − f (b)
b−a a b−a a
and
Z b Z b
1 1
exp f (b) exp [−f (y)] dy − 1 ≥ f (y) dy − f (a) .
b−a a b−a a
Adding these inequalities and dividing by two we get
" Z b #
exp f (a) + exp f (b) 1
exp [−f (y)] dy − 1
2 b−a a
Z b
1 f (a) + f (b)
≥ f (y) dy − ,
b−a a 2
which is equivalent to the desired inequality (3.5). 

Corollary 3.3. With the assumptions of Theorem 3.1 and if

Rb
f (b) b − f (a) a − a f (y) dy
x0 := ∈ [a, b] , (3.6)
f (b) − f (a)
where f (b) 6= f (a) , then we have
Z b Rb !!
1 f (b) b − f (a) a − a f (y) dy
exp [−f (y)] dy ≥ exp −f . (3.7)
b−a a f (b) − f (a)

Proof. Follows by (3.1) by taking x = x0 defined in (3.6). 

The inequality (3.7) can be found in Sándor’s paper [3] where x0 considered in
(3.6) is in fact a mean called by him as “generated by derivatives of functions”. This
mean is extended in [9] (see also [6]), and generalized many results related to integral
inequalities. See also [8] for more results.
Remark 3.4. Since
Rb
f 0 (y) ydy
x0 = Rab ,
f 0 (y) dy
a
then a sufficient condition for (3.6) to hold is that f is monotonic nondecreasing or
nonincreasing on the whole interval [a, b] .
 Some Hermite-Hadamard type inequalities 533

Remark 3.5. If the function g : I → (0, ∞) is convex on I, then f = ln g ∈ Expconv (I)

and for a, b ∈ I with a < b we have, by (3.1), (3.2) and (3.5), the following inequalities
Z b
 x−a b−x
 1
ln [g (a)] b−a
[g (b)] b−a
− ln g (y) dy (3.8)
b−a a
" Z b #
1 1 1
≥ g (x) − dy ,
g (x) b − a a g (y)

Z b
p  1
ln g (a) g (b) − ln g (y) dy (3.9)
b−a a
 " Z b #
a+b 1 1 1
≥g
− dy ,
2 g a+b
2
b − a a g (y)
and
Z b
p  1
ln g (a) g (b) − ln g (y) dy (3.10)
b−a a
" Z b #
g (a) + g (b) 1 1
≥ 1− dy .
2 b − a a g (y)
If  
[g(b)]b Rb
ln [g(a)]a − a
ln g (y) dy
x0 :=   ∈ [a, b] , (3.11)
g(b)
ln g(a)

where g (b) 6= g (a) , then we have

Z b
1 1 1
dy ≥ !. (3.12)
b − a a g (y) [g(b)]b
  R
ln [g(a)]a
− ab ln g(y)dy
g g(b)
ln( g(a) )

Acknowledgement. The authors would like to warmly thank the anonymous referee
for pointing out some essential references and making valuable comments that have
been implemented in the final version of the paper.

References
[1] Dragomir, S.S., Pearce, C.E.M., Selected Topics on Hermite-Hadamard Inequalities and
Applications, RGMIA Monographs, 2000.
[2] Gill, P.M., Pearce, C.E.M., Pečarić, J., Hadamard’s inequality for r−convex functions,
J. of Math. Anal. and Appl., 215(1997), 461-470.
[3] Sándor, J., On means generated by derivatives of functions, Inter. J. Math. Educ. Sci.
Technol., 28(1)(1997), 146-148.
[4] Sándor, J., On upper Hermite-Hadamard inequalities for geometric-convex and log-
convex functions, Notes Number Th. Discr. Math., 20(2014), no. 5, 25-30.
534 Silvestru Sever Dragomir and Ian Gomm

[5] Sándor, J., Toader, Gh., On some exponential means, Preprint, Babeş-Bolyai Univ.,
Cluj, 1990, 35-40.
[6] Sándor, J., Toader, Gh., Some general means, Czechoslovak Math. J., 49(124)(1999),
53-62.
[7] Sándor, J., Toader, Gh., On some exponential means. Part II, Intern. J. Math. Math.
Sci., 2006, ID 051937.
[8] Song, Y., Long, B., Chu, Y., On Toader-Sandor mean, Intern. Math. Forum, 8(2013),
no. 22, 1057-1067.
[9] Toader, Gh., Sándor, J. Inequalities for general integral means, J. Inequal. Pure & Appl.
Math., 7(2006), no. 1, Art. 13.

Silvestru Sever Dragomir

Mathematics, College of Engineering & Science
Victoria University, PO Box 14428
Melbourne City, MC 8001, Australia
School of Computational & Applied Mathematics
University of the Witwatersrand, Private Bag 3
Johannesburg 2050, South Africa
e-mail: [email protected]
Ian Gomm
Mathematics, College of Engineering & Science
Victoria University, PO Box 14428
Melbourne City, MC 8001, Australia
e-mail: [email protected]
Stud. Univ. Babeş-Bolyai Math. 60(2015), No. 4, 535–542

On the univalence of an integral operator

Virgil Pescar

Abstract. In this paper we introduce a new general integral operator for analytic
functions in the open unit disk and we derive some criteria for univalence of this
integral operator.
Mathematics Subject Classification (2010): 30C45.
Keywords: Analytic, Schwarz lemma, integral operator, univalence.

1. Introduction
Let P be the class of functions p of the form
∞
X
p(z) = 1 + bk z k ,
k=1

which are analytic in the open unit disk U = {z ∈ C : |z| < 1}, with positive real part
in U. We denote by A be the class of functions f of the form
∞
X
f (z) = z + an z n ,
n=2

which are analytic in the open unit disk U and we consider S the subclass of A
consisting of functions f ∈ A, which are univalent in U.
In this work we introduce a new integral operator, which is defined by
n
Z zY
γ
Kγ1 ,...,γn (z) = (pj (u)) j du, (1.1)
0 j=1

for functions pj ∈ P and γj be complex numbers, j = 1, n.

2. Preliminary results
In order to prove our main results we will use the lemmas.
536 Virgil Pescar

Lemma 2.1. [1]. If the function f is analytic in U and

zf 00 (z)
(1 − |z|2 ) ≤ 1, (2.1)
f 0 (z)
for all z ∈ U, then the function f is univalent in U.
Lemma 2.2. [4]. Let α be a complex number, Re α > 0 and f ∈ A. If
1 − |z|2Re α zf 00 (z)
≤ 1, (2.2)
Re α f 0 (z)
for all z ∈ U, then for any complex number β, Reβ ≥ Reα, the function
 Z z  β1
Fβ (z) = β uβ−1 f 0 (u)du (2.3)
0

is regular and univalent in U.

Lemma 2.3. (Schwarz [2]). Let f be the function regular in the disk
UR = {z ∈ C : |z| < R}
with |f (z)| < M , M fixed. If f (z) has in z = 0 one zero with multiply ≥ m, then
M
|f (z)| ≤ |z|m , (z ∈ UR ) , (2.4)
Rm
the equality (in the inequality (2.4) for z 6= 0) can hold if
M m
f (z) = eiθ z ,
Rm
where θ is constant.
Lemma 2.4. [3]. If the function f is regular in U and |f (z)| < 1 in U, then for all
ξ ∈ U and z ∈ U the following inequalities hold
f (ξ) − f (z) |ξ − z|
≤ , (2.5)
1 − f (z)f (ξ) |1 − zξ|

1 − |f (z)|2
|f 0 (z)| ≤ , (2.6)
1 − |z|2
(z+u)
the equalities hold only in the case f (z) = 1+uz , where || = 1 and |u| < 1.

Remark 2.5. [3]. For z = 0, from inequality (2.5)

f (ξ) − f (0)
≤ |ξ| (2.7)
1 − f (0)f (ξ)
and, hence
|ξ| + |f (0)|
|f (ξ)| ≤ . (2.8)
1 + |f (0)||ξ|
 On the univalence of an integral operator 537

Considering f (0) = a and ξ = z, we have

|z| + |a|
|f (z)| ≤ , (2.9)
1 + |a||z|
for all z ∈ U.

3. Main results
Theorem 3.1. Let γj be complex numbers, Mj positive real numbers, pj ∈ P,
pj (z) = 1 + b1j z + b2j z 2 + . . . , j = 1, n.
If
zp0j (z)
≤ Mj , (j = 1, n; z ∈ U) (3.1)
pj (z)
and
√
3 3
|γ1 |M1 + |γ2 |M2 + . . . + |γn |Mn ≤ , (3.2)
2
then the integral operator Kγ1 ,...,γn defined by (1.1), is in the class S.
Proof. The function Kγ1 ,...,γn is regular in U and
Kγ1 ,...,γn (0) = Kγ0 1 ,...,γn (0) − 1 = 0.
We have
n
zKγ001 ,...,γn (z) X zp0j (z)
= γ j , (z ∈ U), (3.3)
Kγ0 1 ,...,γn (z) j=1
pj (z)

and hence, we obtain

n
zKγ001 ,...,γn (z) X zp0j (z)
(1 − |z|2 ) ≤ (1 − |z|2
) |γj | , (3.4)
Kγ0 1 ,...,γn (z) j=1
pj (z)

for all z ∈ U.
From (3.1) and Lemma 2.3 we get
zp0j (z)
≤ Mj |z|, (j = 1, n; z ∈ U) (3.5)
pj (z)
and by (3.4) we have
n
zKγ001 ,...,γn (z) X
(1 − |z|2 ) ≤ (1 − |z|2
)|z| |γj |Mj , (3.6)
Kγ0 1 ,...,γn (z) j=1

for all z ∈ U.
Since
2
max(1 − |z|2 )|z| = √ ,
|z|≤1 3 3
538 Virgil Pescar

from (3.2) and (3.6) we obtain that

zKγ001 ,...,γn (z)
(1 − |z|2 ) ≤ 1, (3.7)
Kγ0 1 ,...,γn (z)
for all z ∈ U and by Lemma 2.1, it results that the integral operator Kγ1 ,...,γn belongs
to the class S. 
Theorem 3.2. Let α, γj be complex numbers, j = 1, n, 0 < Re α ≤ 1 and pj ∈ P,
pj (z) = 1 + b1j z + b2j z 2 + . . ., j = 1, n.
If
2Re α+1
zp0j (z) (2Re α + 1) 2Re α
≤ , (j = 1, n; z ∈ U) (3.8)
pj (z) 2
and
|γ1 | + |γ2 | + . . . | + γn | ≤ 1, (3.9)
then the integral operator Kγ1 ,...,γn , defined by (1.1), is in the class S.
Proof. From (3.3) we obtain
n
1 − |z|2Re α zKγ001 ,...,γn (z) 1 − |z|2Re α X zp0j (z)
≤ |γ j | , (3.10)
Re α Kγ0 1 ,...,γn (z) Re α j=1
pj (z)

for all z ∈ U.
By (3.8) and Lemma 2.3, we get
2Re α+1
zp0j (z) (2Re α + 1) 2Re α
≤ |z|, (j = 1, n; z ∈ U) (3.11)
pj (z) 2
and hence, by (3.10) we have

1 − |z|2Re α zKγ001 ,...,γn (z)

· ≤
Re α Kγ0 1 ,...,γn (z)
2Re α+1 n
1 − |z|2Re α (2Re α + 1) 2Re α X
≤ |z| |γj |, (3.12)
Re α 2 j=1

for all z ∈ U.
We have
(1 − |z|)2Re α |z|
 
2
max = 2Re α+1
|z|≤1 Re α (Re α + 1) 2Re α
and from (3.9) and (3.12) we get
1 − |z|2Re α zKγ001 ,...,γn (z)
≤ 1, (3.13)
Re α Kγ0 1 ,...,γn (z)
for all z ∈ U. By (3.13) and Lemma 2.2, for β = 1, f = Kγ1 ,...,γn , it results that the
integral operator Kγ1 ,...,γn is in the class S. 
 On the univalence of an integral operator 539

Theorem 3.3. Let γj be complex numbers,

pj ∈ P, pj (z) = 1 + b1j z + b2j z 2 + . . . , j = 1, n.
If
1
|γ1 | + |γ2 | + . . . | + γn | ≤
, (3.14)
2
then the integral operator Kγ1 ,...,γn defined by (1.1) belongs to the class S.
Proof. Since pj ∈ P, j = 1, n we have
zp0j (z) 2|z|
≤ , (z ∈ U; j = 1, n), (3.15)
pj (z) 1 − |z|2
by (3.3) we obtain
n
zKγ001 ,...,γn (z) X
(1 − |z|2 ) ≤ 2 |γj |, (z ∈ U). (3.16)
Kγ0 1 ,...,γn (z) j=1

From (3.14) and (3.16) we get

zKγ001 ,...,γn (z)
(1 − |z|2 ) ≤ 1, (3.17)
Kγ0 1 ,...,γn (z)
for all z ∈ U.
By (3.17) and Lemma 2.1 we obtain that the integral operator Kγ1 ,...,γn belongs
to the class S. 
Theorem 3.4. Let α, γj be complex numbers, j = 1, n, 0 < Re α ≤ 1, Mj positive real
numbers and pj ∈ P, pj (z) = 1 + b1j z + b2j z 2 + . . ., j = 1, n.
If
p0j (z)
< Mj , (z ∈ U; j = 1, n), (3.18)
pj (z)
1
M1 |γ1 | + M2 |γ2 | + . . . + Mn |γn | ≤ h i, (3.19)
1−|z|2Re α |z|+|c|
max|z|≤1 Re α |z| 1+|c||z|
where
b11 γ1 + b12 γ2 + . . . + b1n γn
c= , (3.20)
M1 |γ1 | + M2 |γ2 | + . . . + Mn |γn |
then the integral operator Kγ1 ,γ2 ,...,γn defined by (1.1) is in the class S.
Proof. We have
n
1 − |z|2Re α zKγ001 ,...,γn (z) 1 − |z|2Re α X p0j (z)
≤ |z| |γ j | , (3.21)
Re α Kγ0 1 ,...,γn (z) Re α j=1
pj (z)
for all z ∈ U. We consider the function
1 Kγ001 ,...,γn (z)
fn (z) = , (z ∈ U) (3.22)
M1 |γ1 | + M2 |γ2 | + . . . + Mn |γn | Kγ0 1 ,...,γn (z)
540 Virgil Pescar

and from (1.1) we obtain

γ1 p0 (z)
fn (z) = · 1 + ...+
M1 |γ1 | + M2 |γ2 | + . . . + Mn |γn | p1 (z)

γn p0 (z)
+ · n , (3.23)
M1 |γ1 | + M2 |γ2 | + . . . + Mn |γn | pn (z)
for all z ∈ U.
From (3.18) and (3.23) we obtain |fn (z)| < 1, z ∈ U.
We have
b11 γ1 + . . . + b1n γn
fn (0) = =c
M1 |γ1 | + . . . + Mn |γn |
and by Remark 2.5 we get
|z| + |c|
|fn (z)| ≤ , (z ∈ U), (3.24)
1 + |c||z|
where
|b11 γ1 + b12 γ2 + . . . + b1n γn |
|c| = .
M1 |γ1 | + M2 |γ2 | + . . . + Mn |γn |
From (3.22) and (3.24) we obtain
1 − |z|2Re α zKγ001 ,...,γn (z)
≤
Re α Kγ0 1 ,...,γn (z)

1 − |z|2Re α
 
|z| + |c|
≤ (M1 |γ1 | + M2 |γ2 | + . . . + Mn |γn |) max |z| , (3.25)
|z|≤1 Re α 1 + |c||z|
for all z ∈ U.
By (3.19) and (3.25) we have
1 − |z|2Re α zKγ001 ,...,γn (z)
≤ 1, (z ∈ U). (3.26)
Re α Kγ0 1 ,...,γn (z)
From (3.26) and Lemma 2.2 for β = 1, it results that the integral operator
Kγ1 ,...,γn ∈ S. 
Corollary 3.5. Let α, γj be complex numbers, j = 1, n, 0 < Re α ≤ 1, Mj positive real
numbers and pj ∈ P, pj (z) = 1 + b1j z + b2j z 2 + . . ., j = 1, n.
If
p0j (z)
< Mj , (z ∈ U; j = 1, n), (3.27)
pj (z)
2Re α+1
(2Re α + 1) 2Re α
|b11 γ1 + b12 γ2 + . . . + b1n γn | ≤ , (3.28)
2
|b11 γ1 + b12 γ2 + . . . + b1n γn | = M1 |γ1 | + M2 |γ2 | + . . . + Mn |γn |, (3.29)
then the integral operator Kγ1 ,...,γn ∈ S.
 On the univalence of an integral operator 541

Proof. From (3.29) and (3.20) we obtain |c| = 1. Using the inequality (3.19) we have
1
M1 |γ1 | + M2 |γ2 | + . . . + Mn |γn | ≤ h i, (3.30)
1−|z|2Re α
max|z|≤1 Re α |z|

Since
1 − |z|2Re α
 
2
max|z|≤1 |z| = 2Re α+1 , (3.31)
Re α (2Re α + 1) 2Re α
from (3.30) and (3.29) we obtain (3.28).
The conditions of Theorem 3.4 are satisfied. 

Corollary 3.6. Let α, γj be complex numbers, j = 1, n, 0 < Re α ≤ 1, Mj positive real

numbers and pj ∈ P, pj (z) = 1 + b1j z + b2j z 2 + . . ., j = 1, n,
b11 γ1 + b12 γ2 + . . . + b1n γn = 0.
If
p0j (z)
< Mj , (z ∈ U; j = 1, n), (3.32)
pj (z)

Re α+1
M1 |γ1 | + M2 |γ2 | + . . . + Mn |γn | ≤ (Re α + 1) Re α , (3.33)
then the integral operator Kγ1 ,...,γn ∈ S.
Proof. From Theorem 3.4, by (3.20), we obtain c = 0 and using the inequality (3.19)
we get
1
M1 |γ1 | + M2 |γ2 | + . . . + Mn |γn | ≤ h i. (3.34)
1−|z|2Re α
max|z|≤1 Re α |z|2

We have
1 − |z|2Re α 2
 
1
max|z|≤1 |z| = Re α+1
Re α (Re α + 1) Re α
and from (3.34) we obtain the inequality (3.33). Since the conditions of Theorem 3.4
are verified it results that Kγ1 ,...,γn belongs to S. 

References
[1] Becker, J., Löwnersche Differentialgleichung und quasikonform fortsetzbare schlichte
Funktionen, J. Reine Angew. Math., 255(1972), 23-43.
[2] Mayer, O., The Functions Theory of One Variable Complex, Bucureşti, 1981.
[3] Nehari, Z., Conformal Mapping, Mc Graw-Hill Book Comp., New York, 1952 (Dover.
Publ. Inc., 1975).
[4] Pascu, N.N., An improvement of Becker’s univalence criterion, Proceedings of the Com-
memorative Session Simion Stoilow, Braşov, (1987), 43-48.
542 Virgil Pescar

Virgil Pescar
”Transilvania” University of Braşov
Faculty of Mathematics and Computer Science
Department of Mathematics
500091 Braşov, Romania
e-mail: [email protected]
Stud. Univ. Babeş-Bolyai Math. 60(2015), No. 4, 543–552

Some new subclasses of bi-univalent functions

Saurabh Porwal, Sanjay Kumar Ghai and Kaushal Kumar

Abstract. The purpose of the present paper is to obtain the initial coefficients for
normalized analytic functions f in the open unit disk U with its inverse g = f −1
belonging to the classes Hσn (φ), STσn (α, φ), Mσn (α, φ) and Lnσ (α, φ). Relevant
connections of the results presented here with various known results are briefly
indicated. Finally, we give an open problem for the readers.
Mathematics Subject Classification (2010): 30C45.
Keywords: Univalent functions, bi-univalent functions, subordination, Salagean
derivative.

1. Introduction
Let A denote the class of functions f of the form
∞
X
f (z) = z + ak z k , (1.1)
k=2
which are analytic in the open unit disk U = {z ∈ C : |z| < 1} and satisfy the nor-
malization condition f (0) = f 0 (0) − 1 = 0. Let S be the subclass of A consisting
of functions of the form (1.1) which are also univalent in U . The Koebe one-quarter
theorem [4] ensures that the image of U under every univalent function f ∈ A con-
tains a disk of radius 41 . Thus every univalent function f has an inverse f −1 satisfying
f −1 (f (z)) = z, (z ∈ U ) and f (f −1 (w)) = w, (|w| < r0 (f ), r0 (f ) ≥ 1/4). A function
f ∈ A is said to be bi-univalent in U if both f and f −1 are univalent in U . Let σ
denote the class of bi-univalent functions defined in the unit disk U .
A domain U ⊂ C is convex if the line segment joining any two points in U lies
entirely in U , while a domain is starlike with respect to a point w0 ∈ U if the line
segment joining any point of U to w0 lies inside U . A function f ∈ A is starlike if f (U )
is a starlike domain with respect to origin, and
o convex if f (U ) is convex. Analytically
0
n
f ∈ A is starlike if and only if < zff (z)(z)
> 0, whereas f ∈ A is convex if and
zf 00 (z)
n o
only if < 1 + f 0 (z) > 0. The classes consisting of starlike and convex functions
are denoted by ST and CV respectively. The classes ST (α) and CV (α) of starlike
544 Saurabh Porwal, Sanjay Kumar Ghai and Kaushal Kumar

and
n convex o functions nof order αo, 0 ≤ α < 1, are respectively characterized by
zf 0 (z) zf 00 (z)
< f (z) > α and < 1 + f 0 (z) > α. Ma and Minda [8] unified various subclasses
of starlike and convex functions by using subordination. Now we recall the definition
of subordination
An analytic function f is subordinate to an analytic function g, written f (z) ≺
g(z), provided there is an analytic function w defined on U with w(0) = 0 and
|w(z)| < 1 satisfying f (z) = g(w(z)).
Lewin [7] investigated the class σ of bi-univalent functions and obtained the
bound for the second coefficient. Several researchers have subsequently studied similar
problems in this direction (see [2], [5], [6], [10], [12], [13]). Brannan and Taha [2] con-
sidered certain subclasses of bi-univalent functions, similar to the familiar subclasses
of univalent functions consisting of strongly starlike, starlike and convex functions.
They introduced bi-starlike functions and bi-convex functions and obtained estimates
on the initial coefficients. Recently, Srivastava et al. [12] introduced and investigated
subclasses of the bi-univalent functions and obtained bounds for the initial coefficients.
The results of [12] were generalized in [5], [6], [10] and [13].
Very recently Ali et al. [1] estimates on the initial coefficients for bi-starlike of
Ma-Minda type and bi-convex of Ma-Minda type functions are obtained. In this paper,
we generalized these results by using Salagean operator and obtain sharp estimates
on coefficient for function classes Hσn (φ), STσn (α, φ), Mσn (α, φ) and Lnσ (α, φ).

2. Coefficient estimates
In the sequel, it is assumed that φ is an analytic function with positive real part
in the disk U , satisfying φ(0) = 1, φ0 (0) > 0 and φ(U ) is symmetric with respect to
the real axis. Such a function has a series expansion of the form

φ (z) = 1 + B1 z + B2 z 2 + B3 z 3 + . . . , (B1 > 0) . (2.1)

A function f ∈ σ is said to be in the class Hσn (φ) if the following subordination hold:

Dn f (z)
≺ φ (z)
z
and
Dn g (w)
≺ φ (w) , g (w) = f −1 (w) ,
w
where Dn stands for the Salagean operator introduced by Salagean [11] for function
f of the form
∞
X
f (z) = z + ak z k
k=2
 Some new subclasses of bi-univalent functions 545

analytic in the open unit disk U as follows

D0 f (z) = f (z)
D1 f (z) = zf 0 (z)
......
Dn f (z) = D(Dn−1 f (z))
Thus
∞
X
Dn f (z) = z + k n ak z k .
k=2
For functions in the class Hσn (φ), we obtain the following result.
Theorem 2.1. If f ∈ Hσn (φ) is given by
∞
X
f (z) = z + ak z k , (2.2)
k=2
then √
B1 B1
|a2 | ≤ p (2.3)
|3n B12 − 22n B2 + 22n B1 |
and  
1 B1
|a3 | ≤ + 2n B1 . (2.4)
3n 2
Proof. Let f ∈ Hσn (φ) and g = f −1 . Then there are analytic functions u, v : U → U ,
with u(0) = v(0) = 0 , satisfying
Dn f (z) Dn g (w)
= φ (u (z)) and = φ (v (w)) . (2.5)
z w
Define the functions p1 (z) and p2 (z) by
1 + u (z)
p1 (z) = = 1 + c1 z + c2 z 2 + . . .
1 − u (z)
and
1 + v (z)
p2 (z) = = 1 + b1 z + b2 z 2 + . . .
1 − v (z)
or, equivalently,
c2
   
p1 (z) − 1 1
u (z) = = c1 z + c2 − 1 z 2 + . . . (2.6)
p1 (z) + 1 2 2
and
b21
   
p2 (z) − 1 1 2
v (z) = = b1 z + b2 − z + ... . (2.7)
p2 (z) + 1 2 2
Then p1 (z) and p2 (z) are analytic in U with p1 (0) = 1 = p2 (0). Since u, v : U → U ,
the functions p1 (z) and p2 (z) have a positive real part in U , and |bi | ≤ 2 and |ci | ≤ 2.
In view of (2.5)-(2.7), clearly
Dn f (z)
 
p1 (z) − 1
=φ (2.8)
z p1 (z) + 1
546 Saurabh Porwal, Sanjay Kumar Ghai and Kaushal Kumar

and
Dn g (w)
 
p2 (w) − 1
=φ . (2.9)
w p2 (w) + 1
Using (2.5) and (2.7) together with (2.1), it is evident that
c21
     
p1 (z) − 1 1 1 1
φ = 1 + B 1 c1 z + B 1 c2 − + B2 c1 z 2 + . . .
2
(2.10)
p1 (z) + 1 2 2 2 4
and
b2
     
p2 (w) − 1 1 1 1
φ = 1 + B 1 b1 w + B1 b2 − 1 + B2 b21 w2 + . . . (2.11)
p2 (w) + 1 2 2 2 4
Since f ∈ σ has the Maclaurin series given by (2.2), a computation shows that its
inverse g = f −1 has the expansion
g (w) = f −1 (w) = w − a2 w2 + 2a22 − a3 w3 + . . .

Since
Dn f (z)
= 1 + 2n a2 z + 3n a3 z 2 + . . .
z
and
Dn g (w)
= 1 − 2n a2 w + 2a22 − a3 3n w2 + . . . ,


w
it follows from (2.8)-(2.11) that
1
2n a2 = B1 c1 , (2.12)
2
c21
 
n 1 1
3 a3 = B1 c2 − + B2 c21 (2.13)
2 2 4
1
−2n a2 = B1 b1 (2.14)
2
and
b2
 

1 1
3n 2a22 − a3 = B1 b2 − 1 + B2 b21 . (2.15)
2 2 4
From (2.12) and (2.14), it follows that
c1 = −b1 . (2.16)
Now (2.13)-(2.16) yield
B13 (b2 + c2 )
a22 =
4 (3n B12 − 22n B2 + 22n B1 )
which, in view of the well-known inequalities |b2 | ≤ 2 and |c2 | ≤ 2 for functions
with positive real part, gives us the desired estimate on |a2 | as asserted in (2.3). By
subtracting (2.15) from (2.13), further computations using (2.12) and (2.16) lead to
B1 (c2 − b2 ) B12 c21
a3 = + ,
4.3n 4.22n
and this yields the estimates given in (2.4). 
 Some new subclasses of bi-univalent functions 547

Remark 2.2. If we put n = 1 in Theorem 2.1, then we obtain the corresponding result
of Ali et al. [1].
 γ
1+z
Remark 2.3. If we put n = 1 with φ (z) = 1−z in Theorem 2.1, then we obtain
the corresponding result of Srivastava et al. [12].

Remark 2.4. If we put n = 1 with φ (z) = 1+(1−2γ)z

1−z in Theorem 2.1, then we obtain
the corresponding result of Srivastava et al. [12].
A function f ∈ σ is said to be in the class STσn (α, φ), n ∈ N0 , α ≥ 0, if the
following subordinations hold:
(1 − α) Dn+1 f (z) + αDn+2 f (z)
≺ φ (z)
Dn f (z)
and
(1 − α) Dn+1 g (w) + αDn+2 g (w)
≺ φ (w) ; g(w) = f −1 (w).
Dn g (w)
Note that STσn (φ) ≡ STσn (0, φ). For the functions in the class STσn (α, φ), the following
coefficient estimates are obtained.
Theorem 2.5. Let f given by (2.2) be in the class STσn (α, φ). Then
√
B1 B1
|a2 | ≤ r
2
B12 (3n 2 (1 + 3α) − 22n (1 + 2α)) + (B1 − B2 ) 22n (1 + 2α)

and
B1 + |B2 − B1 |
|a3 | ≤ .
3n · 2 (1 + 3α) − 22n (1 + 2α)
Proof. Let f ∈ STσn (α, φ). Then there are analytic functions u, v : U → U , with
u(0) = v(0) = 0, satisfying
(1 − α) Dn+1 f (z) + αDn+2 f (z)
= φ (u (z)) (2.17)
Dn f (z)
and
(1 − α) Dn+1 g (w) + αDn+2 g (w)
g = f −1 .


= φ (v (w)) , (2.18)
Dn g (w)
Since
(1 − α) Dn+1 f (z) + αDn+2 f (z)
= 1 + (1 + 2α) 2n a2 z
Dn f (z)
+ 3n · 2 (1 + 3α) a3 − 22n (1 + 2α) a22 z 2 + . . .

and
(1 − α) Dn+1 g (w) + αDn+2 f (w)
= 1 − (1 + 2α) 2n a2 w
Dn g (w)
3n · 4 (1 + 3α) − 22n (1 + 2α) a22 − 3n · 2 (1 + 3α) a3 w2 + . . . ,



+
548 Saurabh Porwal, Sanjay Kumar Ghai and Kaushal Kumar

then (2.10),(2.11), (2.17) and (2.18) yield

1
2n (1 + 2α)a2 =
B1 c1 , (2.19)
2
c2
 
1 1
3n 2(1 + 3α)a3 − 22n (1 + 2α)a22 = B1 c2 − 1 + B2 c21 , (2.20)
2 2 4
1
−2n (1 + 2α)a2 = B1 b1 , (2.21)
2
and
b2
 
1 1
3n · 4 (1 + 3α) − 22n (1 + 2α) a22 − 3n · 2 (1 + 3α) a3 = B1 b2 − 1 + B2 b21 .


2 2 4
(2.22)
Now, the required result follows by using the techniques as used in Theorem 2.1. 

Remark 2.6. If we put n = 0 in Theorem 2.5, then we obtain the corresponding result
of Ali et al. [1].   γ
1+z
Next, if we put n = 0, φ (z) = 1−z with α = 0, then we obtain corresponding
result of Brannan and Taha [2].
Next, a function f ∈ σ belongs to the class Mσn (α, φ), n ∈ N0 , α ≥ 0, if the
following subordinations hold:
Dn+1 f (z) Dn+2 f (z)
(1 − α) + α ≺ φ (z)
Dn f (z) Dn+1 f (z)
and
Dn+1 g (w) Dn+2 g (w)
(1 − α) + α ≺ φ (w) , g (w) = f −1 (w) .
Dn g (w) Dn+1 g (w)
For function in the class Mσn (α, φ), the following coefficient estimates hold.
Theorem 2.7. Let f given by (2.2) be in the class Mσn (α, φ). Then
√
B1 B1
|a2 | ≤ r   (2.23)
2
B12 2 · 3n (1 + 2α) − 22n (1 + 3α) + 22n (1 + α) (B1 − B2 )

and
B1 + |B2 − B1 |
|a3 | ≤ . (2.24)
2 (1 + 2α) 3n − (1 + 3α) 22n
Proof. If f ∈ Mσn (α, φ) , then there are analytic functions u, v : U → U , with u (0) =
v (0) = 0, such that
Dn+1 f (z) Dn+2 f (z)
(1 − α) n
+ α n+1 = φ (u (z)) (2.25)
D f (z) D f (z)
and
Dn+1 g (w) Dn+2 g (w)
(1 − α) + α = φ (v (w)) . (2.26)
Dn g (w) Dn+1 g (w)
 Some new subclasses of bi-univalent functions 549

Since
Dn+1 f (z) Dn+2 f (z)
(1 − α) n
+ α n+1 = 1 + (1 + α) 2n a2 z
D f (z) D f (z)
+ 2 (1 + 2α) 3n a3 − (1 + 3α) 22n a22 z 2 + . . .

and
Dn+1 g (w) Dn+2 g (w)
(1 − α) n
+ α n+1 = 1 − (1 + α) 2n a2 w
D g (w) D g (w)
4 (1 + 2α) 3n − (1 + 3α) 22n a22 − 2 (1 + 2α) 3n a3 w2 + . . . .



+
From (2.10), (2.11), (2.25) and (2.26) it follows that
1
(1 + α) 2n a2 = B1 c1 (2.27)
2
c21
 
n 2n 2 1 1
2 (1 + 2α) 3 a3 − (1 + 3α) 2 a2 = B1 c2 − + B2 c21 (2.28)
2 2 4
1
− (1 + α) 2n a2 = B1 b1 (2.29)
2
and
b2
 
1 1
4 (1 + 2α) 3n − (1 + 3α) 22n a22 −2 (1 + 2α) 3n a3 = B1 b2 − 1 + B2 b21 . (2.30)


2 2 4
Equation (2.27) and (2.29) yield
c1 = −b1 . (2.31)
From (2.28), (2.30) and (2.31), it follows that
B13 (b2 + c2 )
a22 =  
2
4 B12 (2 · 3n (1 + 2α) − 22n (1 + 3α)) + 22n (1 + α) (B1 − B2 )

which yields the describe estimate on as describe in (2.23). As in the earlier proofs,
use of (2.28)-(2.31) shows that
(B1 /2) ((4 (1 + 2α) 3n − (1 + 3α) 2n ) c2 + (1 + 3α) 2n b2 ) + b21 (1 + 2α) (B2 −B1 )
a3 = .
4 · 3n (1 + 2α) (2 (1 + 2α) 3n − (1 + 3α) 22n )
Thus the proof of Theorem 2.7 is complete. 
Next, a function f ∈ σ is said to be in the class Lnσ (α, φ) n ∈ N0 , α ≥ 0, if the
following subordinations hold:
 n+1 α  n+2 1−α
D f (z) D f (z)
≺ φ (z)
Dn f (z) Dn+1 f (z)
and α  1−α
Dn+1 g (w) Dn+2 g (w)

≺ φ (w)
Dn g (w) Dn+1 g (w)
g (w) = f −1 (w) .
For function in this class, the following coefficient estimates are obtained
550 Saurabh Porwal, Sanjay Kumar Ghai and Kaushal Kumar

Theorem 2.8. Let f given by (2.2) be in the class Lnσ (α, φ). Then
√
2B1 B1
|a2 | ≤ r ,
n 2 2n 2 2n 2
2 (4 (3 − α) 3 + (α + 5α − 8) 2 B1 ) + 4 · 2 (α − 2) (B1 − B2 )
(2.32)
and
2 (3 − 2α) 3n (B1 + |B1 − B2 |)
|a3 | ≤ . (2.33)
|3n (3 − 2α) (4 (3 − 2α) 3n + (α2 + 5α − 8) 22n )|
Proof. Let f ∈ Lnσ (α, φ). Then there are analytic functions u, v : U → U , with
u (0) = v (0) = 0 , such that
 n+1 α  n+2 (1−α)
D f (z) D f (z)
= φ (u (z)) (2.34)
Dn f (z) Dn+1 f (z)
and
α  (1−α)
Dn+1 g (w) Dn+2 g (w)

= φ (v (w)) . (2.35)
Dn g (w) Dn+1 g (w)
Since
α  n+2 (1−α)
Dn+1 f (z)

D f (z)
= 1 + 2n (2 − α) a2 z
Dn f (z) Dn+1 f (z)
  2  
α − 5α + 8
+ 3n · 2 (3 − 2α) a3 + 22n a22 z 2 + . . .
2
and
α  (1−α)
Dn+1 g (w) Dn+2 g (w)

= 1 − 2n (2 − α) a2 w
Dn g (w) Dn+1 g (w)
α2 + 5α − 8
  
n
+ 4 · (3 − 2α) 3 + 2 n
a2 − 3 · 2 (3 − 2α) a3 w2 + . . .
2
from (2.10), (2.11), (2.34) and (2.35) it follows that
1
2n · (2 − α) a2 = B1 c1 (2.36)
2

2n a22 c21
 
n 2 1 1
3 2 (3 − 2α) a3 + α + 5α − 8 2 · = B1 c2 − + B2 c21 (2.37)
2 2 2 4
1
−2n (2 − α) a2 = B 1 b1 (2.38)
2
and

!
α2 + 5α − 8 b21
 
n n 2 n 1 1
4 (3 − 2α) 3 + 2 a2 −3 ·2 (3 − 2α) a3 = B1 b2 − + B2 b21 .
2 2 2 4
(2.39)
Now (2.36) and (2.38) clearly yield
c1 = −b1 . (2.40)
 Some new subclasses of bi-univalent functions 551

Equation (2.37), (2.39) and (2.40) lead to

B13 (b2 + c2 )
a22 = 2
2 (4 (3 − 2α) 3n + (α2 + 5α − 8) 22n ) B12 + 4 · 22n (α − 2) (B1 − B2 )
which yields the desired estimate on |a2 | as asserted in (2.32). Proceeding similarly
as in the earlier proof, using (2.37)-(2.40), it following that
(B1 /2)((8(3 − 2α)3n + 22n (α2 + 5α − 8))c2 − 22n (α2 + 5α − 8)b2 ) + 3n 2b21 (3 − 2α)(B1 − B2 )
a3 =
4 · 3n (3 − 2α)(4(3 − 2α)3n + (α2 + 5α − 8)22n )
which yields the estimate (2.33). 

Remark 2.9. If we put n = 0 in Theorem 2.7-2.8 then we obtain the corresponding

result of Ali et al. [1].
Remark 2.10. Sharp estimates for the coefficients |a2 |, |a3 | and other coefficients
of functions belonging to the classes investigated in this paper are yet open prob-
lems. Indeed it would be of interest even to find estimates (not necessarily sharp) for
|an |, n ≥ 4.

References
[1] Ali, R.M., Lee, S.K., Ravichandran, V., Supramaniam, S., Coefficient estimates for bi-
univalent Ma-Minda starlike and convex functions, Appl. Math. Lett., 25(2012), 344-351.
[2] Branna, D.A., Taha, T.S., On some classes of bi-univalent functions, Studia Univ.
Babes-Bolyai Math., 31(1986), no. 2, 70-77.
[3] Brannan, D.A., Clunie, J., Kirwan, W.E., Coefficient estimates for a class of starlike
functions, Canad. J. Math., 22(1970), 476-485.
[4] Duren, P.L., Univalent functions, In: Grundlehred der Mathematischen Wissenschaften,
vol. 259, Springer, New York, 1983.
[5] Frasin, B.A., Aouf, M.K., New subclasses of bi-univalent functions, Appl. Math. Lett.,
24(2011), no. 9, 1569-1573.
[6] Goyal, S.P., Goswami, P., Estimate for initial Maclaurin coefficients of bi-univalent
functions for a class defined by fractional derivatives, J. Egypt. Math. Soc., 20(2012),
no. 3, 179-182.
[7] Lewin, M., On a coefficient problem for bi-univalent functions, Proc. Amer. Math. Soc.,
18(1967), 63-68.
[8] Ma, W.C., Minda, D., A unified treatment of some special classes of univalent functions,
In: Proceedings of the Conference on Complex Analysis, Tianjin, 1992, 157-169, Conf.
Proc. Lecture Notes Anal. L. Int. Press, Cambridge, MA, 1994.
[9] Netanyahu, E., The minimal distance of the image boundary from the origin and the
second coefficient of a univalent function in |z| < 1, Arch. Ration. Mech. Anal., 32(1969),
100-112.
[10] Porwal, S., Darus, M., On a new subclass of Bi-univalent functions, J. Egypt. Math.
Soc., 21(2013), 190-193.
[11] Salagean, G.S., Subclasses of univalent functions, Complex Analysis-Fifth Romanian
Finish Seminar, Bucharest, 1(1983), 362-372.
552 Saurabh Porwal, Sanjay Kumar Ghai and Kaushal Kumar

[12] Srivastava, H.M., Mishra, A.K., Gochhayat, P., Certain subclasses of analytic and bi-
univalent functions, Appl. Math. Lett., 23(2010), no. 10, 1188-1192.
[13] Xu, Q.H., Gui, Y.C., Srivastava, H.M., Coefficient estimates for a certain subclass of
analytic and bi-univalent functions, Appl. Math. Lett., 25(2012), 990-994.

Saurabh Porwal
Department of Mathematics
U.I.E.T., C.S.J.M. University
Kanpur-208024, (U.P.), India
e-mail: [email protected]
Sanjay Kumar Ghai
Principal, BVM College of Technology and Management
Gwalior-474001, (M.P.), India
e-mail: [email protected]
Kaushal Kumar
Department of Mathematics
U.I.E.T., C.S.J.M. University
Kanpur-208024, (U.P.), India
Stud. Univ. Babeş-Bolyai Math. 60(2015), No. 4, 553–559

A new class of (j, k)-symmetric harmonic starlike

functions
Fuad Al Sarari and Latha Satyanarayana

Abstract. Using the concepts of (j, k)-symmetrical functions we define the class
of sense-preserving harmonic univalent functions SHj,ks (β), and prove certain
interesting results.
Mathematics Subject Classification (2010): 30C45.
Keywords: (j, k)-Symmetric functions, harmonic functions.

1. Introduction
A continuous function f = u + iv is a complex valued harmonic function in a
complex domain C if both u and v are real harmonic in C. In any simply connected
domain D ∈ C we can write f (z) = h + g, where h and g are analytic in D. We call h
the analytic part and g the co-analytic part of f . A necessary and sufficient condition
for f to be locally univalent and sense-preserving in D is that |h0 (z)| > |g 0 (z)| in D,
[see 3].
Denote by SH the class of functions f (z) = h + g that are harmonic univalent and
orientation preserving in the open unit disk U = {z : z ∈ C and |z| < 1}, for which
f (0) = fz (0) − 1 = 0. Then for f (z) = h + g ∈ SH , we may express the analytic
functions f and g as

∞
X ∞
X
h(z) = z + an z n , g(z) = bn z n , |b1 | < 1. (1.1)
n=2 n=1

Note that SH reduces to the class S of normalized analytic univalent functions if

the coanalytic part of its members is zero. For this class the function f (z) may be
expressed as
X∞
f (z) = z + an z n , (1.2)
n=2
554 Fuad Al Sarari and Latha Satyanarayana

A function f (z) = h + g with h and g given by (1.1) is said to be harmonic starlike

of order β for 0 ≤ β < 1, for |z| = r < 1 if
( ) ( )
∂ iθ
∂ iθ ∂θ f (re ) zh0 (z) − zg 0 (z)
(arg f (re )) = = =< ≥ β.
∂θ f (reiθ ) h(z) + g(z)
∗
The class of all harmonic starlike functions of order β is denoted by SH (β) and
extensively studied by Jahangiri [4]. The cases β = 0 and b1 = 1 were studied by
Silverman and Silvia [8] and Silverman [7].
Definition 1.1. Let k be a positive integer. A domain D is said to be k-fold symmetric if
a rotation of D about the origin through an angle 2πk carries D onto itself. A function
f is said to be k-fold symmetric in U if for every z in U
2πi 2πi
f (e k z) = e k f (z).
The family of all k-fold symmetric functions is denoted by S k and for k = 2 we get
class of odd univalent functions.
The notion of (j, k)-symmetrical functions (k = 2, 3, ... ; j = 0, 1, 2, ..., k − 1) is a
generalization of the notion of even, odd, k-symmetrical functions and also general-
ize the well-known result that each function defined on a symmetrical subset can be
uniquely expressed as the sum of an even function and an odd function.
The theory of (j, k) symmetrical functions has many interesting applications, for in-
stance in the investigation of the set of fixed points of mappings, for the estimation
of the absolute value of some integrals, and for obtaining some results of the type of
Cartan’s uniqueness theorem for holomorphic mappings [5].
2πi
Definition 1.2. Let ε = (e k ) and j = 0, 1, 2, .., k −1 where k ≥ 2 is a natural number.
A function f : D 7→ C and D is a k-fold symmetric set, f is called (j, k)-symmetrical
if
f (εz) = εj f (z), z ∈ U.

We note that the family of all (j, k)-symmetric functions is denoted be S (j,k) .
Also, S (0,2) , S (1,2) and S (1,k) are called even, odd and k-symmetric functions respec-
tively. We have the following decomposition theorem.
Theorem 1.3. [5] For every mapping f : D 7→ C, and D is a k-fold symmetric set,
there exists exactly the sequence of (j, k)- symmetrical functions fj,k ,
k−1
X
f (z) = fj,k (z)
j=0

where
k−1
1 X −vj
fj,k (z) = ε f (εv z), (1.3)
k v=0

(f ∈ A; k = 1, 2, ...; j = 0, 1, 2, ..., k − 1)
 A new class of (j, k)-symmetric harmonic starlike functions 555

From (1.3) we can get

∞
k−1 k−1
!
1 X −vj 1 X −vj X
fj,k (z) = ε f (εv z) = ε v
an (ε z) n
,
k v=0 k v=0 n=1

then
∞ k−1
(
X
n 1 X (n−j)v 1, n = lk + j;
fj,k (z) = δn,j an z , a1 = 1, δn,j = ε = , (1.4)
n=1
k v=0 0, n 6= lk + j;
Ahuja and Jahangiri [2] discussed the class SH(β) which denotes the class of complex-
valued, sense-preserving, harmonic univalent functions f of the form (1.1) and satis-
fying ( )
∂
2 ∂θ f (reiθ )
= ≥ β, 0 ≤ β < 1.
f (reiθ ) − f (−reiθ )
Al-Shaqsi and and Maslina Darus [1] discussed the class SHks (β) which denotes the
class of complex-valued, sense-preserving, harmonic univalent functions f of the form
(1.1) and satisfying ( )
∂ iθ
∂θ f (re )
= ≥ β, 0 ≤ β < 1.
fk (reiθ )
Now using the concepts of (j, k)−symmteric points we define the following
Definition 1.4. For 0 ≤ β < 1 and k = 1, 2, 3, ..., j = 0, 1, ..., k−1, let SHj,k
s (β) denote
the class of sense-preserving, harmonic univalent functions f of the form (1.1) which
satisfy the condition ( )
∂ iθ
∂θ f (re )
= ≥ β. (1.5)
fj,k (reiθ )
Where z = reiθ , 0 ≤ r < 1, 0 ≤ θ < 2π and fj,k = hj,k + gj,k , where hj,k , gj,k given
by
k−1 k−1
1 X −vj 1 X −vj v
hj,k (z) = ε h(εv z), gj,k (z) = ε g(ε z). (1.6)
k v=0 k v=0

The following special cases are of interest

(i) SH1,k k
s (β) =SHs (β) the class introduced by Al-Shaqsi and Darus [1].
1,2
(ii) SHs (β) =SH(β) the class introduced by Ahuja and Jahangiri [2].
∗
(iii) SH1,1
s (β) =SH (β) the class introduced by Jahangiri [4].
1,1 ∗
(iv) SHs (0) =SH the class introduced by Silverman and Silvia [8].
We need the following lemma to prove our main results.
Lemma 1.5. [4] Let f = f + g with h and g are given by (1.1). If
∞  
X n−β n+β
|an | + |bn | ≤ 2, a1 = 1, 0 ≤ β < 1.
n=1
1−β 1−β
Then f is sense-preserving, harmonic univalent and starlike of order β in U.
556 Fuad Al Sarari and Latha Satyanarayana

2. Main result
Theorem 2.1. Let f ∈ SH(j,k)
s (β) where f given by (1.1), then fj,k (z) is in SH ∗ (β),
where fj,k given by (1.6).
Proof. Suppose that f ∈ SH(j,k)
s (β). Then we get
( )
∂ iθ
∂θ f (re )
= ≥ β. (2.1)
fj,k (reiθ )
replacing reiθ by εv reiθ in (2.1), we get
( )
∂ v iθ
∂θ f (ε re )
= ≥ β.
fj,k (εv reiθ )
According to the definition of fj,k and εk = 1, we know fj,k (εv reiθ ) = εvj fj,k (reiθ ),
we get ( )
ε−vj ∂θ
∂
f (εv reiθ )
= ≥ β, (2.2)
fj,k (reiθ )
letting (v = 0, 1, 2, ....k − 1) in (2.2) and summing them we can get
( Pk−1 ) ( )
1 −vj ∂ v iθ ∂ iθ
k v=0 ε ∂θ f (ε re ) ∂θ fj,k (re )
= == > β, (2.3)
fj,k (reiθ ) fj,k (reiθ )
that is fj,k (z) ∈ SH ∗ (β). 
Theorem 2.2. If f = h + g with h and g given by (1.1) and fj,k = hj,k + gj,k with hj,k
and gj,k given by (1.6). Let
∞  
X (n − 1)k + j − β (n − 1)k + j + β
|a(n−1)k+j | + |b(n−1)k+j | (2.4)
n=1
1 − δ1,j β 1 − δ1,j β
∞
X n
+ {|an | + |bn |} ≤ 2,
1 − δ1,j β
n6=lk+j
where a1 = 1, 0 ≤ β < 1, k = 1, 2, 3, ..., j = 0, 1, ..., k − 1, l ∈ N and δ1,j is defined
by (1.4). Then f is sense-preserving harmonic univalent in U and f ∈ SH(j,k) s (β).
Proof. Since
∞   X ∞  
X n−β n+β n − δn,j β n + δn,j β
|an | + |bn | ≤ |an | + |bn | ,
n=1
1−β 1−β n=1
1 − δ1,j β 1 − δ1,j β
where δn,j is given by (1.4),
∞  
X (n − 1)k + j − β (n − 1)k + j + β
= |a(n−1)k+j | + |b(n−1)k+j |
n=1
1 − δ1,j β 1 − δ1,j β
∞
X n
+ {|an | + |bn |} ≤ 2.
1 − δ1,j β
n6=lk+j
 A new class of (j, k)-symmetric harmonic starlike functions 557

By Lemma 1.5, we conclude that f is sense-preserving, harmonic univalent and starlike

in U. To prove f ∈ SH(j,k)
s (β), according to condition (1.5), we need to show that
( ) ( )
∂ iθ
zh0 (z) − zg 0 (z)
 
∂θ f (re ) A(z)
= =< =< ≥ β.
fk (reiθ ) hj,k (z) + gj,k (z) B(z)
Where z = reiθ , 0 ≤ r < 1, 0 ≤ θ < 2π, 0 ≤ β < 1 and k = 1, 2, 3, ..., j =
0, 1, ..., k − 1.
∞
X ∞
X
A(z) = zh0 (z) − zg 0 (z) = z + nan z n − nbn z n (2.5)
n=2 n=1
and
∞
X ∞
X
B(z) = fj,k (z) = an δn,j z n + δn,j bn z n , (2.6)
n=1 n=1
where δn,j is defined by (1.4), and εk = 1.
Using the fact that <{w} ≥ β if and only if |1 − β + w| ≥ |1 + β − w|, it suffices to
show that
|A(z) + (1 − β)B(z)| − |A(z) − (1 + β)B(z)| ≥ 0.
For A(z) and B(z) as given in (2.5) and (2.6) respectively, we get
|A(z) − (1 − β)B(z)| − |A(z) − (1 − β)B(z)|
= |(1−β)hj,k +zh0 (z)+(1 − β)gj,k − zg 0 (z)|−|(1+β)hj,k −zh0 (z)+(1 + β)gj,k + zg 0 (z)|
∞
X ∞
X
= [1 + (1 − β)δ1,j ]z + [n + (1 − β)δn,j ]an z n − [n − (1 − β)δn,j ]bn z n
n=2 n=1
∞
X ∞
X
− [1 − (1 + β)δ1,j ]z + [n − (1 + β)δn,j ]an z n − [n + (1 + β)δn,j ]bn z n
n=2 n=1
∞
X ∞
X
≥ [1 + (1 − β)δ1,j ]|z| − [n + (1 − β)δn,j ]|an ||z|n − [n − (1 − β)δn,j ]|bn ||z|n
n=2 n=1
∞
X ∞
X
−[1 − (1 + β)δ1,j ]|z| − [n − (1 + β)δn,j ]|an ||z|n − [n + (1 + β)δn,j ]|bn ||z|n
n=2 n=1
∞ ∞
( )
X n − βδn,j n + βδn,j
X
= 2(1 − βδ1,j )|z| 1 − |an ||z|n−1 − |bn ||z|n−1
n=2
1 − βδ 1,j n=1
1 − βδ 1,j
∞ ∞
( )
X n − βδn,j X n + βδn,j
≥ 2(1 − δ1,j )|z| 1 − |an | − |bn | .
n=2
1 − βδ1,j n=1
1 − βδ1,j
From the definition of δn,j in (1.4), we have
|A(z) + (1 − β)B(z)| − |A(z) − (1 + β)B(z)|
∞ ∞
( )
X nk + j − β X nk + j + β
≥ 2(1 − βδ1,j )|z| 1 − |ank+j | − |bnk+j |
n=1
1 − βδ1,j n=1
1 − δ1,j β
558 Fuad Al Sarari and Latha Satyanarayana
 
∞ ∞
 X n X n 1+β 
−(1 − βδ1,j )|z| |an | + |bn | + |b1 |
 1 − δ1,j β 1 − δ1,j β 1 − δ1,j β 
n6=lk+j n6=lk+j

∞ 
( )
X (n−1)k+j −β (n−1)k+j +β
≥ 2(1 − βδ1,j )|z| 1− |a(n−1)k+j | − |b(n−1)k+j |
n=1
1 − δ1,j β 1 − δ1,j β
 
∞
 X n 
−(1 − βδ1,j )|z| [|an | + |bn |] ≥ 0,
 1 − δ1,j β 
n6=lk+j

we note that in (2.4). This concludes the proof of the theorem. 

For j = 1 we get the result introduced by Al-Shaqsi and Darus in [1].

Corollary 2.3. If f = h + g with h and g given by (1.1) and fk = hk + gk . Let

∞  
X (n − 1)k + 1 − β (n − 1)k + 1 + β
|a(n−1)k+1 | + |b(n−1)k+1 | (2.7)
n=1
1−β 1−β

∞
X n
+ {|an | + |bn |} ≤ 2,
1−β
n6=lk+1

where a1 = 1, 0 ≤ β < 1, k = 1, 2, 3, ..., l ∈ N. Then f is sense-preserving harmonic

univalent in U and f ∈ SH(k)
s (β).

References
[1] Al Shaqsi, K., Darus, M., On subclass of harmonic starlike functions with respect to
k-symmetric points, Int. Math. Forum, 2(2007), 2799-2805.
[2] Ahuja, O.P., Jahangiri, J.M., Sakaguchi-type harmonic univalent functions, Scientiae
Mathematicae Japonicae, 59(1)(2004), 239-244.
[3] Clunie, J., Shell-Small, T., Harmonic univalent functions, Ann. Acad. Aci. Fenn. Ser. A
I Math., 9(1984), 3-25.
[4] Jahangiri, J.M, Harmonic functions starlike in the unit disk, Journal of Mathematical
Analysis and Applications, 235(2)(1999), 470-477.
[5] Liczberski, P., Polubinski, J., On (j, k)-symmtrical functions, Mathematica Bohemica,
120(1995), 13-25.
[6] Shell-Small, T., Constants for planar harmonic mapping, J. London. Math. Soc.,
42(2)(1990), 237-248.
[7] Silverman, H., Harmonic univalent functions with negative coefficients, Journal of Math-
ematical Analysis and Applications, 220(1)(1998), 283-289.
[8] Silverman, H., Silvia, E.M., Subclasses of harmonic univalent functions, The New
Zealand Journal of Mathematics, 28(2)(1999), 275-284.
 A new class of (j, k)-symmetric harmonic starlike functions 559

Fuad Al Sarari
Department of Studies in Mathematics
Manasagangotri, University of Mysore
Mysore 570 006, India
e-mail: [email protected] or [email protected]
Latha Satyanarayana
Department of Mathematics
Yuvaraja’s College, University of Mysore
Mysore 570 005, INDIA
e-mail: [email protected]
Stud. Univ. Babeş-Bolyai Math. 60(2015), No. 4, 561–565

On close-to-convex functions satisfying

a differential inequality
Sukhwinder Singh Billing

Abstract. Let Cα (β) denote the class of normalized functions f, analytic in the
open unit disk E which satisfy the condition
 
zf ′ (z) (zf ′ (z))′
ℜ (1 − α) +α > β, z ∈ E,
φ(z) φ′ (z)
f (z)f ′ (z)
where 6= 0, z ∈ E, φ is starlike and α, β are pre-assigned real numbers.
z
In 1977, Chichra, P. N. [1] introduced and studied the class Cα = Cα (0). He
proved the members of class Cα are close-to-convex forα ≥ 0. We  here prove that
α zφ′ (z)
functions in class Cα (β) are close-to-convex for − ℜ ≤ β < 1, α ≥ 0
2 φ(z)
and the result is sharp in the sense  that the constant β cannot be replaced by
α φ(z)
a real number smaller than − ℜ . We claim that our result improves
2 zφ′ (z)
the result of Chichra, P. N. [1].
Mathematics Subject Classification (2010): 30C80, 30C45.
Keywords: Analytic function, convex function, starlike function, close-to-convex.

1. Introduction
Let A be the class of functions f, analytic in E = {z : |z| < 1} and normalized
by the conditions f (0) = f ′ (0) − 1 = 0. Let S ∗ and K denote the classes of starlike
and convex functions respectively analytically defined as follows:
  ′  
∗ zf (z)
S = f ∈A:ℜ > 0, z ∈ E ,
f (z)
and    
zf ′′ (z)
K= f ∈A:ℜ 1+ ′ > 0, z ∈ E .
f (z)
This is well-known that
f (z) ∈ K ⇔ zf ′ (z) ∈ S ∗ . (1.1)
562 Sukhwinder Singh Billing

A function f ∈ A is said to be close to convex if there is a real number α, −π/2 <

α < π/2 and a convex function g (not necessarily normalized) such that
 
f ′ (z)
ℜ eiα ′ > 0, z ∈ E.
g (z)
In view of the relation (1.1), the above definition takes the following form in case g is
normalized. A function f ∈ A is said to be close to convex if there is a real number
α, −π/2 < α < π/2, and a starlike function φ such that
 
zf ′ (z)
ℜ eiα > 0, z ∈ E.
φ(z)
It is well-known that every close-to-convex function is univalent. In 1934/35, Noshiro
[3] and Warchawski [4] obtained a simple but elegant criterion for univalence of ana-
lytic functions. They proved that if an analytic function f satisfies ℜf ′ (z) > 0 for all
z in E, then f is close-to-convex and hence univalent in E.
Let Cα (β) denote the class of normalized analytic functions f which satisfy the con-
dition  
zf ′ (z) (zf ′ (z))′
ℜ (1 − α) +α ′ > β, z ∈ E,
φ(z) φ (z)
f (z)f ′ (z)
where 6= 0, z ∈ E, φ is starlike and α, β are pre-assigned real numbers.
z
The class Cα = Cα (0) was introduced and studied by Chichra, P. N. [1] in 1977. He
called the members of class Cα as α− close-to-convex functions. Infact, he proved the
following result.
Theorem 1.1. Let f (z) ∈ Cα and α ≥ 0. Then f (z) is close-to-convex in E.
In the present
 ′paper, we establish the result that functions in Cα (β) are close-to-convex
α zφ (z)
for − ℜ ≤ β < 1, α ≥ 0. Our result is the best possible in the sense that
2 φ(z)  
α φ(z)
the constant β cannot be replaced by a real number smaller than − ℜ .
2 zφ′ (z)
We also claim that our result improves the result of Chichra, P. N. [1]. To prove our
main result, we shall use the following lemma of Miller [2].
Lemma 1.2. Let D be a subset of C × C (C is the complex plane) and let φ : D → C
be a complex function. For u = u1 + iu2 , v = v1 + iv2 (u1 , u2 , v1 , v2 are reals), let φ
satisfy the following conditions:
(i) φ(u, v) is continuous in D
(ii) (1, 0) ∈ D and ℜ[φ(1, 0)] > 0 and
(iii) ℜ[φ(iu2 , v1 )] ≤ 0 for all (iu2 , v1 ) ∈ D such that v1 ≤ −(1 + u22 )/2.
Let p(z) = 1 + p1 z + p2 z 2 + · · · be regular in the open unit disk E, such that
(p(z), zp′ (z)) ∈ D for all z ∈ E. If
ℜ[φ(p(z), zp′ (z))] > 0, z ∈ E,
then ℜp(z) > 0, z ∈ E.
 On close-to-convex functions satisfying a differential inequality 563

2. Main result
Theorem 2.1. Let α and β be real numbers such that α ≥ 0 and
 
α φ(z)
− ℜ ≤β<1
2 zφ′ (z)
for a starlike function φ. Assume that f ∈ A satisfies
 
zf ′ (z) (zf ′ (z))′
ℜ (1 − α) +α ′ > β, z ∈ E, (2.1)
φ(z) φ (z)
 ′ 
zf (z)
then ℜ > 0 in E and hence f is close-to-convex and hence univalent in E.
φ(z)
The result is sharp in the sense that the constant β onthe right  hand side of (2.1)
α φ(z)
cannot be replaced by a real number smaller than − ℜ .
2 zφ′ (z)
Proof. Let p(z) = 1 + p1 z + p2 z 2 + . . . be analytic in E such that for all z ∈ E, we
write
zf ′ (z)
= p(z).
φ(z)
Then,
zf ′ (z) (zf ′ (z))′ φ(z)
(1 − α) +α ′ = p(z) + αzp′ (z) .
φ(z) φ (z) zφ(z)
Therefore, condition (2.1) is equivalent to
 
1 α φ(z) β
ℜ p(z) + zp′ (z) ′ − > 0, z ∈ E. (2.2)
1−β 1−β zφ (z) 1 − β
For D = C × C, define Φ(u, v) : D → C as under:
1 α φ(z) β
Φ(u, v) = u+ v − , z ∈ E.
1−β 1 − β zφ′ (z) 1 − β
Then Φ(u, v) is continuous in D, (1, 0) ∈ D and ℜ (Φ(1, 0)) = 1 > 0. Further, in view
of (2.2), we get, ℜ[Φ(p(z), zp′ (z))] > 0, z ∈ E. Let u = u1 + iu2 , v = v1 + iv2 where
1 + u22
u1 , u2 , v1 and v2 are all real numbers. Then, for (iu2 , v1 ) ∈ D, with v1 ≤ − , we
2
have
 
1 α φ(z) β
ℜΦ(iu2 , v1 ) = ℜ u2 i + v1 −
1−β 1 − β zφ′ (z) 1 − β
α 1 + u22
   
φ(z) β
≤ − ℜ +
1−β 2 zφ′ (z) 1−β
   
α φ(z) β
≤ − ℜ +
2(1 − β) zφ′ (z) 1−β
≤ 0.
In view of (2.2) and Lemma 1.2, proof now follows.
564 Sukhwinder Singh Billing

Figure 2.1

Figure 2.2

To show that the constant βon the right

 hand side of (2.1) cannot be replaced by
α φ(z)
a real number smaller than − ℜ , we consider the function f (z) = z ez ∈ A
2 zφ′ (z)
and φ(z) = z ∈ S ∗ . Using Mathematica 9.0, we plot, in Figure 2.1, the image of the
zf ′ (z) (zf ′ (z))′
unit disk under the operator (1 − α) +α ′ taking α = 2. From this
φ(z) φ (z)
′
zf (z) (zf ′ (z))′
figure, we notice that minimum real part of (1 − α) +α is smaller
φ(z) φ′ (z)
 On close-to-convex functions satisfying a differential inequality 565
 
α φ(z)
than −1 (the calculated value of − ℜ for α = 2 and φ(z) = z). In Figure
2 zφ′ (z)
zf ′ (z)
2.2, we plot the image of unit disk under the operator . It is obvious that
 ′  φ(z)
zf (z) 1 π
ℜ ≯ 0 for all z in E. For example, the point z = − + i is an interior
φ(z)  ′  2 4
zf (z) π−2
point of E, but at this point ℜ = − √ = −0.1224 · · · < 0. This justifies
φ(z) 4 2e
our claim. 
Remark 2.2. We claim that our result improves the result of Chichra, P. N. [1]. In
fact, when we take f (z) = −z − 2 log(1 − z) ∈ A, φ(z) = z and α = 2 in Theorem
2.1, we notice that at z = −1,
 
zf ′ (z) (zf ′ (z))′
ℜ (1 − α) +α = −1.
φ(z) φ′ (z)
Thus the function f does not satisfy
 ′ the hypothesis
 ofTheorem 1.1 due to Chichra,
zf (z) 1+z
P. N. [1] i.e. f ∈
/ Cα although ℜ =ℜ > 0 in E. Hence the result of
φ(z) 1−z
Chichra, P. N. [1] fails to conclude the close-to-convexity in this case whereas Theorem
2.1 concludes the same.

References
[1] Chichra, P.N., New subclasses of the class of close-to-convex functions, Proc. Amer.
Math. Soc., 62(1)(1977), 37-43.
[2] Miller, S.S., Differential inequalities and Carathéodory functions, Bull. Amer. Math.
Soc., 81(1975), 79-81.
[3] Noshiro, K., On the theory of schlicht functions, J. Fac. Sci., Hokkaido Univ., 2(1934-35),
129-155.
[4] Warchawski, S.E., On the higher derivatives at the boundary in conformal mappings,
Trans. Amer. Math. Soc., 38(1935), 310-340.

Sukhwinder Singh Billing

Department of Mathematics
Sri Guru Granth Sahib World University
Fatehgarh Sahib-140 406, Punjab, India
e-mail: [email protected]
Stud. Univ. Babeş-Bolyai Math. 60(2015), No. 4, 567–581

Generalized and numerical solution for

a quasilinear parabolic equation with
nonlocal conditions
Fatma Kanca and Irem Baglan

Abstract. In this paper we study the one dimensional mixed problem with non-
local boundary conditions, for the quasilinear parabolic equation. We prove an
existence, uniqueness of the weak generalized solution and also continuous depen-
dence upon the data of the solution are shown by using the generalized Fourier
method. We construct an iteration algorithm for the numerical solution of this
problem. We analyze computationally convergence of the iteration algorithm, as
well as on test example.
Mathematics Subject Classification (2010): 35K55.
Keywords: Quasilinear parabolic equation, nonlocal boundary condition, finite
difference method.

1. Introduction
In this study, we consider the following mixed problem
∂u ∂ 2 u
− 2 = f (x, t, u), D := {0 < x < 1, 0 < t < T } (1.1)
∂t ∂x
u(0, t) = u(1, t), t ∈ [0, T ] (1.2)
ux (1, t) = 0, t ∈ [0, T ] (1.3)
u(x, 0) = ϕ(x), x ∈ [0, 1] (1.4)
for a quasilinear parabolic equation with the nonlinear source term f = f (x, t, u).
The functions ϕ(x) and f (x, t, u) are given functions on [0, 1] and D̄ × (−∞, ∞) ,
respectively.
Denote the solution of the problem (1.1)-(1.4) by u = u(x, t).
This problem was investigated with different boundary conditions by various
researchers by using Fourier or different methods [2, 4].
568 Fatma Kanca and Irem Baglan

In this study, we consider the initial-boundary value problem (1.1)-(1.4) with

nonlocal boundary conditions (1.2 )-(1.3). The periodic nature of (1.2 )-(1.3) type
boundary conditions is demonstrated in [10]. In this study, we prove the existence,
uniqueness, convergence of the weak generalized solution continuous dependence upon
the data of the solution and we construct an iteration algorithm for the numerical
solution of this problem. We analyze computationally convergence of the iteration
algorithm, as well as on test example. We demonstrate a numerical procedure for this
problem on concrete examples, and also we obtain numerical solution by using the
implicit finite difference algorithm [11].
We will use the weak solution approach from [3] for the considered problem
(1.1)-(1.4).
According to [1, 5] assume the following definitions.
Definition 1.1. The function v(x, t) ∈ C 2 (D̄) is called test function if it satisfies the
following conditions:
v(x, T ) = 0, v(0, t) = v(1, t), vx (1, t) = 0, ∀t ∈ [0, T ] and ∀x ∈ [0, 1].
Definition 1.2. The function u(x, t) ∈ C(D̄) satisfying the integral identity
ZT Z1 
∂2v
 
∂v
− 2 u − f (x, t, u)v dxdt
∂t ∂x
0 0

ZT Z1
− [u(0, t)vx (0, t) − v(0, t)ux (0, t)] dt − ϕ(x)v(x, 0)dx = 0, (1.5)
0 0
for arbitrary test function v = v(x, t), is called a generalized (weak) solution of the
problem (1)-(4).

2. Reducing the problem to countable system of integral equations

Consider the following system of functions on the interval [0, 1] :
X0 (x) = 2, X2k−1 (x) = 4 cos(2πkx), X2k (x) = 4(1 − x) sin(2πkx), k = 1, 2, . . . ,
(2.1)
Y0 (x) = x, Y2k−1 (x) = x cos(2πkx), Y2k (x) = sin(2πkx), k = 1, 2, . . . (2.2)
The system of functions (2.1) and (2.2) arise in [6] for the solution of a nonlocal
boundary value problem in heat conduction.
It is easy to verify that the system of function (2.1) and (2.2) are biorthonormal
on [0, 1]. They are also Riesz bases in L2 [0, 1] (see [7, 8]).
We will use the Fourier series representation of the weak solution to transform
the initial-boundary value problem to the infinite set of nonlinear integral equations.
Any solution of the equation (1.1)-(1.4) can be represented as
∞
X
u(x, t) = uk (t)Xk (x), (2.3)
k=1
 Numerical solution for a quasilinear parabolic equation 569

where the functions uk (t), k = 0, 1, 2, . . . satisfy the following system of equations:

Zt
u0 (t) = ϕ0 + f0 (τ )dτ,
0
Zt
2 2
u2k (t) = ϕ2k e−(2πk) t + f2k (τ )e−(2πk) (t−τ )
dτ, (2.4)
0
2
t
u2k−1 (t) = (ϕ2k−1 − 4πkϕ2k )e−(2πk)
Zt
2
(t−τ )
+ e−(2πk) [f2k−1 (τ ) − 4πk(t − τ )f2k (τ )]dτ,
0

where
Z 1
ϕk = ϕ(x)Yk (x)dx,
0
Z 1
fk (x) = f (x, t, u)Yk (x)dx.
0

Definition 2.1. Denote the set

{u(t)} = {u0 (t), u2k (t), u2k−1 (t), k = 1, 2, . . . , } ,
of continuous on [0, T ] satisfying the following condition
X∞  
underset0 ≤ t ≤ T max2 |u0 (t)| + 4 max |u2k (t)| + max |u2k−1 (t)| < ∞,
0≤t≤T 0≤t≤T
k=1

by B. Let
∞ 
X 
ku(t)k = max 2 |u0 (t)| + 4 max |u2k (t)| + max |u2k−1 (t)| ,
0≤t≤T 0≤t≤T 0≤t≤T
k=1

be the norm in B. It can be shown that B is the Banach space [9].

We denote the solution of the nonlinear system (2.4) by {u(t)} .

Theorem 2.2. a) Let the function f (x, t, u) is continuous with respect to all arguments
in D̄ × (−∞, ∞) and satisfies the following condition
|f (x, t, u) − f (x, t, ũ)| ≤ b(x, t) |u − ũ| ,
where b(x, t) ∈ L2 (D), b(x, t) ≥ 0,
b) f (x, t, 0) ∈ C 2 [0, 1], t ∈ [0, 1],
c) ϕ(x) ∈ C 2 [0, 1].
Then the system (2.4) has a unique solution in D.
570 Fatma Kanca and Irem Baglan

Proof. For N = 0, 1, . . . let’s define an iteration for the system (2.4) as follows:
Zt Z1
(N +1) (0)
u0 (t) = u0 (t) + f (ξ, τ, Au(N ) (ξ, τ ))ξdξdτ,
0 0
Zt Z1
(N +1) (0) 2
(t−τ )
u2k (t) = u2k (t) + e−(2πk) f (ξ, τ, Au(N ) (ξ, τ )) sin 2πkξdξdτ,
0 0 (2.5)
Zt Z1
(N +1) (0) 2
(t−τ )
u2k−1 (t) = u2k−1 (t) + e−(2πk) f (ξ, τ, Au(N ) (ξ, τ ))ξ cos 2πkξdξdτ
0 0
Zt Z1
2
(t−τ )
−4πk e−(2πk) f (ξ, τ, Au(N ) (ξ, τ ))(t − τ ) sin 2πkξdξdτ,
0 0

where, for simplicity, let

∞  
(N ) (N ) (N )
X
Au(N ) (ξ, τ ) = 2u0 (τ ) + 4 u2k (τ )(1 − ξ) sin 2πkξ + u2k−1 (τ ) cos 2πkξ .
k=1

where,
(0) (0) 2 (0) 2
u0 (t) = ϕ0 , u2k (t) = ϕ2k e−(2πk) t , u2k−1 (t) = (ϕ2k−1 − 4πk ϕ2k ) e−(2πk) t .

From the condition of the theorem we have u(0) (t) ∈ B. We will prove that the other
sequentially approximations satisfy this condition.
Let us write N = 0 in (2.5).
Zt Z1
(1) (0)
u0 (t) = u0 (t) + f (ξ, τ, Au(0) (ξ, τ ))dξdτ.
0 0

Zt Z1
Adding and subtracting f (ξ, τ, 0)dξdτ, applying Cauchy inequality, Lipschitz
0 0
condition, taking the maximum of both sides of the last inequality yields the following:
(1)
√ √
max u0 (t) ≤ |ϕ0 | + T kb(x, t)kL2 (D) u(0) (t) + T kf (x, t, 0)kL2 (D) .
0≤t≤T

Zt Z1
(1) −(2πk)2 t 2
(t−τ )
u2k (t) = ϕ2k e + e−(2πk) f (ξ, τ, Au(N ) (ξ, τ )) sin 2πkξdξdτ.
0 0

Zt Z1
2
(t−τ )
Adding and subtracting e−(2πk) f (ξ, τ, 0) sin 2πkξdξdτ, applying Cauchy in-
0 0
equality, taking the summation of both sides respect to k and using Hölder inequality,
 Numerical solution for a quasilinear parabolic equation 571

Bessel inequality, Lipschitz condition and taking maximum of both sides of the last
inequality yields the following:
∞ ∞
X (1)
X 1 1
max u2k (t) ≤ |ϕ2k | + √ kb(x, t)kL2 (D) u(0) (t) + √ kf (x, t, 0)kL2 (D) .
k=1
0≤t≤T
k=1
4 3 4 3
In the same way, we obtain:
∞ ∞ ∞
X (1)
X 1 X ′′
max u2k−1 (t) ≤ |ϕ2k−1 | + √ ϕ2k
k=1
0≤t≤T
k=1
6 k=1
1 1
+ √ kb(x, t)kL2 (D) u(0) (t) + √ kf (x, t, 0)kL2 (D)
4 3 4 3
√ (0)
√
+ 2 |T | kb(x, t)kL2 (D) u (t) + 2 |T | kf (x, t, 0)kL (D) .
2

Finally we have the following inequality:

X∞  
(1) (1) (1) (1)
u (t) = 2 max u0 (t) + 4 max u2k (t) + max u2k−1 (t)
B 0≤t≤T 0≤t≤T 0≤t≤T
k=1
∞ √ ∞
X 2 6 |T | X ′′
≤ 2 |ϕ0 | + 4 (|ϕ2k | + |ϕ2k−1 |) + ϕ2k
3
k=1 k=1
√ !
√ 2 3 √  
+ 2 T+ + 4 2 |T | kb(x, t)kL2 (D) u(0) (t)
3 B
√ !
√ 2 3 √
+ 2 T+ + 4 2 |T | kf (x, t, 0)kL2 (D) .
3
Hence u(1) (t) ∈ B. In the same way, for a general value of N we have
X∞  
(N ) (N ) (N ) (N )
u (t) = 2 max u0 (t) + 4 max u2k (t) + max u2k−1 (t)
B 0≤t≤T 0≤t≤T 0≤t≤T
k=1
∞ √ ∞
X 2 6 |T | X ′′
≤ 2 |ϕ0 | + 4 (|ϕ2k | + |ϕ2k−1 |) + ϕ2k
3
k=1 k=1
√ !
√ 2 3 √  
+ 2 T+ + 4 2 |T | kb(x, t)kL2 (D) u(N −1) (t)
3 B
√ !
√ 2 3 √
+ 2 T+ + 4 2 |T | kf (x, t, 0)kL2 (D) ,
3
u(N −1) (t) ∈ B, we deduce that u(N ) (t) ∈ B, we obtain
{u(t)} = {u0 (t), u2k (t), u2k−1 (t), k = 1, 2, . . .} ∈ B.
Now we prove that the iterations u(N +1) (t) converge in B, as N → ∞.
∞
X
(1) (0) (1) (0) (1) (0)
u(1) (t) − u(0) (t) = 2(u0 (t) − u0 (t)) + 4 [(u2k (t) − u2k (t)) + (u2k−1 (t) − u2k−1 (t))]
k=1
572 Fatma Kanca and Irem Baglan

Zt Z1 h i
=2 f (ξ, τ, Au(0) (ξ, τ )) − f (ξ, τ, 0) ξdξdτ
0 0
Zt Z1 h i 2
+4 f (ξ, τ, Au(0) (ξ, τ )) − f (ξ, τ, 0) e−(2πk) (t−τ ) sin 2πkξdξdτ
0 0
Zt Z1 h i 2
+4 f (ξ, τ, Au(0) (ξ, τ )) − f (ξ, τ, 0) e−(2πk) (t−τ ) ξ cos 2πkξdξdτ
0 0
Zt Z1 h i 2
−16πk (t − τ ) f (ξ, τ, Au(0) (ξ, τ )) − f (ξ, τ, 0) e−(2πk) (t−τ ) sin 2πkξdξdτ.
0 0
Applying Cauchy inequality, Hölder inequality, Lipschitz condition and Bessel inequal-
ity to the right side of u(1) (t) − u(0) (t) respectively, we obtain:
(1) (0)
u(1) (t) − u(0) (t) ≤ 2 u0 (t) − u0 (t)
∞  
(1) (0) (1) (0)
X
+4 u2k (t) − u2k (t) + u2k−1 (t) − u2k−1 (t)
k=1
 21
√ ! Zt Z1

√ √ 2 3 
≤ 2 T + 4 2 |T | + b2 (ξ, τ )dξdτ  u(0) (t)
3
0 0
 12
√ ! Zt Z1

√ √ 2 3 
+ 2 T + 4 2 |T | + f 2 (ξ, τ, 0)dξdτ  ,
3
0 0
  21
√ ! Zt Z1

√ √ 2 3
AT =  2 T + 4 2 |T | + b2 (ξ, τ )dξdτ  u(0) (t)
 
3
0 0
 12 
√ ! Zt Z1

√ √ 2 3 
+ 2 T + 4 2 |T | + f 2 (ξ, τ, 0)dξdτ   .

3
0 0
Applying Cauchy unequality, Hölder inequality, Lipschitz condition and Bessel in-
equality to the right hand side of u(2) (t) − u(1) (t) respectively, we obtain:
(2) (1)
u(2) (t) − u(1) (t) ≤ 2 u0 (t) − u0 (t)
∞ 
X 
(2) (1) (2) (1)
+4 u2k (t) − u2k (t) + u2k−1 (t) − u2k−1 (t)
k=1
 12
√ ! Zt Z1

√ √ 2 3 
≤ 2 T + 4 2 |T | + b2 (ξ, τ )dξdτ  AT .
3
0 0
 Numerical solution for a quasilinear parabolic equation 573

In the same way, for a general value of N we have

(N +1) (N )
u(N +1) (t) − u(N ) (t) ≤ 2 u0 (t) − u0 (t)
∞ 
X 
(N +1) (N ) (N +1) (N )
+4 u2k (t) − u2k (t) + u2k−1 (t) − u2k−1 (t)
k=1
 2  N2
√ Zt Z1
√ √ 2 3 N AT 
≤ (2 T + 4 2 |T | + ) √  b2 (ξ, τ )dξdτ  

3 N!
0 0
√
√ √ 2 3 N 1 N
≤ (2 T + 4 2 |T | + ) AT √ kb(x, t)kL2 (D) . (2.6)
3 N!
Then the last inequality shows that the u(N +1) (t) convergence in B.
Now let us show lim u(N +1) (t) = u(t). It follows that if we prove
N →∞

lim u(τ ) − u(N ) (τ ) = 0,

N →∞ B

then we may deduce that u(t) satisfies (2.4). For this aim we estimate the difference
u(t) − u(N +1) (t) B , after some transformation we obtain:
(N +1)
u(t) − u(N +1) (t) = 2 u0 (t) − u0 (t)

∞  
(N +1) (N +1)
X
+4 u2k (t) − u2k (t) + u2k−1 (t) − u2k−1 (t)
k=1

Zt Z1 n o
≤2 f [ξ, τ, Au(ξ, τ )] − f [ξ, τ, Au(N ) (ξ, τ )] ξdξdτ
0 0

∞ Z t Z1 n o
X 2
(t−τ )
+4 e−(2πk) f [ξ, τ, Au(ξ, τ )] − f [ξ, τ, Au(N ) (ξ, τ )] sin 2πkξdξdτ
k=1 0 0

∞ Zt Z1 n o
X 2
(t−τ )
+4 e−(2πk) f [ξ, τ, Au(ξ, τ )] − f [ξ, τ, Au(N ) (ξ, τ )] ξ cos 2πkξdξdτ
k=1 0 0

Zt Z1 h i
+16πk (t − τ ) f (ξ, τ, Au(ξ, τ )) − f (ξ, τ, Au(N ) (ξ, τ ))
0 0

2
(t−τ )
·e−(2πk) sin 2πkξdξdτ .
574 Fatma Kanca and Irem Baglan

Adding and subtracting f (ξ, τ, Au(N +1) (ξ, τ )) under appropriate integrals to the right
hand side of the inequality we obtain
 12
√ !Zt Z1

√ √ 2 3 2 
u(t)−u(N +1) (t) ≤ 2 T +4 2|T |+ b2 (ξ, τ ) u(τ )−u(N +1) (τ ) dξdτ
3  
0 0
 12
√ ! Zt Z1

√ √ 2 3  
+ 2 T + 4 2 |T | + b2 (ξ, τ )dξdτ u(N +1) (t) − u(N ) (t) .
3   B
0 0
Applying Gronwall’s inequality to the last inequality and using inequality (2.6), we
have
√ !(N +1)
√ √
r
(N +1) 2 2 3 (N +1)
u(t) − u (t) ≤ AT 2 T + 4 2 |T | + kb(x, t)kL2 (D)
B N! 3
√ !2
√ √ 2 3 2
× exp 2 T + 4 2 |T | + kb(x, t))kL2 (D) . (2.7)
3
For the uniqueness, we assume that the problem (1.1)-(1.4) has two solutions u, v.
Applying Cauchy inequality, Hölder inequality, Lipschitz condition and Bessel inequal-
ity to the right hand side of |u(t) − v(t)| respectively, we obtain:
√ !2 Zt Z1
2
√ √ 2 3 2
|u(t) − v(t)| ≤ 2 T + 4 2 |T | + b2 (ξ, τ ) |u(τ ) − v(τ )| dξdτ,
3
0 0
applying Gronwall’s inequality to the last inequality we have u(t) = v(t).
The theorem is proved. 

3. Solution of the problem (1.1)-(1.4)

Using the solution of the system (2.4) we compose the series
∞
X
2u0 (t) + 4 [u2k (t)(1 − x)sin2πkx + u2k−1 (t)cos2πkx].
k=1
It is evident that these series convergence uniformly on D. Therefore the sum
∞
X
u(ξ, τ ) = 2u0 (τ ) + 4 [u2k (τ )(1 − ξ)sin2πkξ + u2k−1 (τ )cos2πkξ],
k=1
continuous on D.
l
X
ul (ξ, τ ) = 2u0 (τ ) + 4 [u2k (τ )(1 − ξ)sin2πkξ + u2k−1 (τ )cos2πkξ]. (3.1)
k=1
From the conditions of Theorem 2.2 and from
lim ul (ξ, τ ) = u(ξ, τ ),
l→∞
 Numerical solution for a quasilinear parabolic equation 575

it follows
lim f (ξ, τ, ul (τ, ξ)) = f (ξ, τ, u(ξ, τ )).
l→∞
Using ul (ξ, τ ) and
l
X
ϕl (x) = 2ϕ0 + 4 [ϕ2k (1 − x)sin2πkx + ϕ2k−1 cos2πkx], x ∈ [0, 1]
k=1

on the left hand side of (1.5) we denote the obtained expression by Jl :

ZT Z1 
∂2v
 
∂v
Jl = + 2 u(l) (x, t) + f (x, t, u(l) (x, t))v(x, t) dxdt
∂t ∂x
0 0
Z1
+ ϕ(l) (x)v(x, 0)dx. (3.2)
0

Applying the integration by part formula to the right hand side the last equation and
using the conditions of Theorem 2.2 , we can show that
lim Jl = 0.
l→∞

This shows that the function u(x, t) is a generalized(weak) solution of the problem
(1.1)-(1.4).
The following theorem shows the existence and uniqueness results for the generalized
solutions of problem (1.1)-(1.4).
Theorem 3.1. Under the assumptions of Theorem 2.2, Problem (1.1)-(1.4) possesses
a unique generalized solution u = u(x, t) ∈ C(D) in the following form
∞
X
u(x, t) = 2u0 (t) + 4 [u2k (t)(1 − x)sin2πkx + u2k−1 (t)cos2πkx].
k=1

4. Continuous dependence upon the data

In this section, we shall prove the continuous dependence of the solution
u = u(x, t)
using an iteration method.
Theorem 4.1. Under the conditions of Theorem 2.2, the solution u = u(x, t) depends
contiunously upon the data.

Proof. Let φ = {ϕ, f } and φ = ϕ, f be two sets of data which satisfy the conditions
of Theorem 1.Let u = u(x, t) and v = v(x, t) be the solutions of the problem (1.1)-(1.4)
corresponding to the data φ and φ respectively and
f (t, x, 0) − f (t, x, 0) ≤ ε, f or ε ≥ 0.
576 Fatma Kanca and Irem Baglan

The solution v = v(x, t) is in the following form

Zt
v0 (t) = ϕ0 + f 0 (τ )dτ,
0
Zt
2 2
v2k (t) = ϕ2k e−(2πk) t + f 2k (τ )e−(2πk) (t−τ )
dτ,
0
2
t
v2k−1 (t) = (ϕ2k−1 − 4πktϕ2k )e−(2πk)
Zt
2
(t−τ )
+ e−(2πk) [f 2k−1 (τ ) − 4πk(t − τ )f 2k (τ )]dτ,
0
where, for simplicity, let
∞ 
X 
(N ) (N ) (N ) (N )
Av (ξ, τ ) = 2v0 (τ ) +4 v2k (τ )(1 − ξ) sin 2πkξ + v2k−1 (τ ) cos 2πkξ
k=1
Zt Z1
(N +1) (0)
v0 (t) = v0 (t) + f (ξ, τ, Av (N ) (ξ, τ ))ξdξdτ,
0 0
Zt Z1
(N +1) (0) 2
(t−τ )
v2k (t) = v2k (t) + e−(2πk) f (ξ, τ, Av (N ) (ξ, τ )) sin 2πkξdξdτ,
0 0
Zt Z1
(N +1) (0) 2
(t−τ )
v2k−1 (t) = v2k−1 (t) + e−(2πk) f (ξ, τ, Av (N ) (ξ, τ ))ξ cos 2πkξdξdτ
0 0
Zt Z1
2
(t−τ )
−4πk e−(2πk) f (ξ, τ, Av (N ) (ξ, τ ))(t − τ ) sin 2πkξdξdτ,
0 0
where
(0) (0) 2 (0) 2
v0 (t) = ϕ0 , v2k (t) = ϕ2k e−(2πk) t , u2k−1 (t) = (ϕ2k−1 − 4πkϕ2k ) e−(2πk) t .
From the condition of the theorem we have v (0) (t) ∈ B. We will prove that the other
sequentially approximations satisfy this condition.
First of all , we consider u(1) (t) − v (1) (t), applying Cauchy inequality, Hölder
inequality, Lipschitz condition and Bessel inequality to the u(1) (t) − v (1) (t) respec-
tively, we obtain:
(1) (1)
u(1) (t) − v (1) (t) ≤ 2 u0 (t) − v0 (t)
∞ 
X 
(1) (1) (1) (1)
+4 u2k (t) − v2k (t) + u2k−1 (t) − v2k−1 (t) ≤ 2 max |ϕ0 − ϕ0 |
k=1
∞ √ ∞
X 2 6 |T | X ′′ ′′
+4 max |ϕ2k − ϕ2k | + max |ϕ2k−1 − ϕ2k−1 | + max ϕ2k − ϕ2k
3
k=1 k=1
 Numerical solution for a quasilinear parabolic equation 577

 21
√ ! Zt Z1

√ √ 2 3 
2 T + 4 2 |T | + b2 (ξ, τ )dξdτ  u(0) (t)
3
0 0
 12
√ ! Zt Z1

√ √ 2 3  2
+ 2 T + 4 2 |T | + b (ξ, τ )dξdτ  v (0) (t)
3
0 0
 21
√ ! Zt Z1

√ √ 2 3  2
+ 2 T + 4 2 |T | + f 2 (ξ, τ, 0) − f (ξ, τ, 0)dξdτ  ,
3
0 0
" √ !
√ √ 2 3
AT = kϕ − ϕk + 2 T + 4 2 |T | + kb(x, t)k u(0) (t)
3
√ ! # √ !
√ √ 2 3 (0)
√ √ 2 3
+ 2 T + 4 2 |T | + kb(x, t)k v (t) + 2 T + 4 2 |T | + f −f .
3 3
kϕ − ϕk = 2 max |ϕ0 − ϕ0 |
∞ √ ∞
X 2 6 |T | X ′′ ′′
+4 max |ϕ2k − ϕ2k | + max |ϕ2k−1 − ϕ2k−1 | + max ϕ2k − ϕ2k .
3
k=1 k=1
Applying Cauchy inequality, Hölder inequality, Lipschitz condition and Bessel inequal-
ity to the right hand side of u(2) (t) − v (2) (t) respectively, we obtain:
(2) (2)
u(2) (t) − v (2) (t) ≤ 2 u0 (t) − v0 (t)
∞ 
X 
(2) (2) (2) (2)
+4 u2k (t) − v2k (t) + u2k−1 (t) − v2k−1 (t)
k=1
 21
√ ! Zt Z1

√ √ 2 3 
≤ 2 T + 4 2 |T | + b2 (ξ, τ )dξdτ  AT
3
0 0
 21
√ ! Zt Z1
√ √ 2 3 2
+ 2 T + 4 2 |T | +  b (ξ, τ )dξdτ  AT .
3
0 0
In the same way, for a general value of N we have
(N +1) (N +1)
u(N +1) (t) − v (N +1) (t) ≤ 2 u0 (t) − v0 (t)
∞ 
X 
(N +1) (N +1) (N +1) (N +1)
+4 u2k (t) − v2k (t) + u2k−1 (t) − v2k−1 (t)
k=1
 2  N2
√ !N t Z1
√ √ 2 3 AT 
Z
≤ 2 T + 4 2 |T | + √  b2 (ξ, τ )dξdτ  

3 N!
0 0
578 Fatma Kanca and Irem Baglan

  2  N2
√ !N Zt Z1
√ √ 2 3 A  2
+ 2 T + 4 2 |T | + √ T  b (ξ, τ )dξdτ  

3 N!
0 0

√ !N
√ √ 2 3 1
≤ 2 T + 4 2 |T | + AT √ kb(x, t)kN
L2 (D)
3 N!

√ !N
√ √ 2 3 1 N
+ 2 T + 4 2 |T | + AT √ b(x, t) L2 (D)
3 N!

≤ AT · aN = aN (kϕ − ϕk + C(t) + M1 f − f )
where
 2  N2
√ !N Zt Z1
√ √ 2 3 1 
aN = 2 T + 4 2 |T | + √  b2 (ξ, τ )dξdτ  

3 N!
0 0
 2  N2
√ !N Zt Z1
√ √ 2 3 1 2
+ 2 T + 4 2 |T | + √ b (ξ, τ )dξdτ   .
 
3

N!
0 0

and
√
√ √ 2 3 N
M1 = (2 T + 4 2 |T | + ) .
3
(The sequence aN is convergent then we can write aN ≤ M, ∀N ). It follows from
the estimation ([2], page 76-77) that lim u(N +1) (t) = u(t), then let N → ∞ for last
N →∞
equation
|u(t) − v(t)| ≤ M kϕ − ϕk + M2 f − f )
where M2 = M.M1 . If f − f ≤ ε and kϕ − ϕk ≤ ε then |u(t) − v(t)| ≤ ε. 

5. Numerical procedure for the nonlinear problem (1.1)-(1.4)

We construct an iteration algorithm for the linearization of the problem (1.1)-(1.4):

∂u(n) ∂ 2 u(n)
− = f (x, t, u(n−1) ), (x, t) ∈ D (5.1)
∂t ∂x2
u(n) (0, t) = u(n) (1, t), t ∈ [0, T ] (5.2)
u(n)
x (1, t) = 0, t ∈ [0, T ] (5.3)
(n)
u (x, 0) = ϕ(x), x ∈ [0, 1] . (5.4)
 Numerical solution for a quasilinear parabolic equation 579

Let u(n) (x, t) = v(x, t) and f (x, t, u(n−1) ) = fe(x, t). Then the problem (5.1)-(5.4)
can be written as a linear problem:
∂v ∂2v
− 2 = fe(x, t) (x, t) ∈ D (5.5)
∂t ∂x
v(0, t) = v(1, t), t ∈ [0, T ] (5.6)
vx (1, t) = 0, t ∈ [0, T ] (5.7)
v(x, 0) = ϕ(x), x ∈ [0, 1] . (5.8)
We use the finite difference method to solve (5.5)-(5.8).
We subdivide the intervals [0, 1] and [0, T ] into M and N subintervals of equal lengths
1 T
h= M and τ = N , respectively. Then, we add a line x = (M + 1) h to generate the
fictitious point needed for the second boundary condition.
We choose the implicit scheme, which is absolutely stable and has a second order
accuracy in h and a first order accuracy in τ.
The implicit monotone difference scheme for (5.5)-(5.8) is as follows:
vi,j+1 − vi,j a2
= 2 (vi−1,j+1 − 2vi,j+1 + vi+1,j+1 ) + fei,j+1
τ h
vi,0 = ϕi , v0,j = vM,j , vx,Mj = 0
where 0 ≤ i ≤ M and 1 ≤ j ≤ N are the indices for the spatial and time steps,
respectively, vi,j is the approximation to v(xi , tj ), fi,j = f (xi , tj ), vi = v(xi ), xi = ih,
tj = jτ . [12]
At the t = 0 level, adjustment should be made according to the initial condition
and the compatibility requirements.

6. Numerical example
In this section, we will consider an example of numerical solution of the problem
(1.1)-(1.4).
These problems were solved by applying the iteration scheme and the finite
difference scheme which were explained in the Section 5. The condition
(n+1) (n)
error(i, j) := ui,j − ui,j
∞
−3
with error(i, j) := 10 was used as a stopping criteria for the iteration process.
Example 6.1. Consider the problem
∂u ∂ 2 u 
− 2 = 1 − (2π)2 (cos 2πx + (sin 2πx)2 ) u

∂t ∂x
u(x, 0) = exp(− cos 2πx), x ∈ [0, 1]
u(0, t) = u(1, t), t ∈ [0, T ] , ux (1, t) = 0, t ∈ [0, T ] .
It is easy to see that the analytical solution of this problem is
u(x, t) = exp(t − cos 2πx).
The comparisons between the analytical solution and the numerical finite difference
solution f when T = 1 are shown in Figure 1 for the step sizes h = 0.0025, τ = 0.0025.
580 Fatma Kanca and Irem Baglan

5
u(1,t)

0
0 0.2 0.4 0.6 0.8 1
x

Figure 1. The exact and numerical solutions of u(x, 1), the exact
solution is shown with dashes line.

References
[1] Chandirov, G.I., On mixed problem for a class of quasilinear hyperbolic equation, Ph.D.
Thesis, Tibilisi, 1970.
[2] Ciftci (Baglan), I., Halilov, H., On Fourier Method for a Qusilinear Parabolic Equation
with Periodic Boundary Condition, Hacettepe Journal of Mathematics and Statistics,
37(2008), no. 2, 69-79.
[3] Conzalez-Velasco, E.A., Fourier Analysis and Boundary Value Problems, Academic
Press, Newyork, 1995.
[4] Hasanov, K.K., On solution of mixed problem for a quasilinear hiperbolic and parabolic
equation, Ph.D. Thesis, Baku, 1961.
[5] IL’in, V.A., Solvability of mixed problem for hyperbolic and parabolic equation, Uspekhi
Math. Nauk., 15(2)(1960), 97-154.
[6] Ionkin, N.I., Solution of a boundary-value problem in heat conduction with a nonclassical
boundary condition, Differential Equations, 13(1977), 204-211.
[7] Ismailov, M.I., Kanca, F., An inverse coefficient problem for a parabolic equation in the
case of nonlocal boundary and overdetermination conditions, Math. Meth. Appl. Sci.,
34(2011), 692-702.
[8] Kanca, F., Ismailov, M.I., The Inverse Problem of Finding the Time-dependent Diffu-
sion Coefficient of the Heat Equation from Integral Overdetermination Data, Inverse
Problems in Science and Engineering, 20(2012), 463-476.
[9] Ladyzhenskaya, D.A., Boundary Value Problems of Mathematical Physics, Springer,
Newyork, 1985.
 Numerical solution for a quasilinear parabolic equation 581

[10] Nakhushev, A.M., Equations of Mathematical Biology, Moscow, 1995 (in Russian).
[11] Sakinc (Baglan), I., Numerical Solution of a Quasilinear Parabolic Problem with Periodic
Boundary Condition, Hacettepe Journal of Mathematics and Statistics, 39(2010), 183-
189.
[12] Samarskii, A.A., The theory of difference schemes, Marcel Dekker, New York, 2001.

Fatma Kanca
Kadir Has University
Department of Management Information Systems
34083, Istanbul, Turkey
e-mail: [email protected]
Irem Baglan
Kocaeli University
Department of Mathematics
Kocaeli 41380, Turkey
e-mail: [email protected]
 Stud. Univ. Babeș-Bolyai Math. 60(2015), No. 4, 583–595

Lines in the three-dimensional

Bolyai-Lobachevskian hyperbolic geometry
Zoltán Gábos and Ágnes Mester

Abstract. The purpose of this paper is to describe the geodesics of the three-
dimensional Bolyai-Lobachevskian hyperbolic space. We also determine the equa-
tion of the orthogonal surfaces and the scalar curvature of the surfaces of revo-
lution. The metric applied is the Lobachevskian metric extended into three di-
mensions. During the analysis we use Cartesian and cylindrical coordinates. This
article is a continuation of the paper [4].
Mathematics Subject Classification (2010): 53A35.
Keywords: Hyperbolic geometry, geodesics, orthogonal surfaces, cylindrical coor-
dinates.

1. General context
In the literature exists several models for hyperbolic geometry, see [1]-[10]. The
aim of this paper is to present a three dimensional model using [8] to describe some
classical and new properties.
We consider the following metric
z y 
ds2 = cosh2 cosh2 dx2 + dy 2 + dz 2 , (1.1)
k k
where k is the parameter of the three-dimensional hyperbolic space, and x, y, z are
the Cartesian coordinates of any P (x, y, z) point. The usage of Cartesian coordinates
is justified by the existence of such hyperbolic lines which can also be considered
Euclidean lines. These lines include the coordinate axes illustrated in figure 1. Note
that the x-value can only be determined by axis Ox. Figure 1 also represents how the
coordinates of any P (x, y, z) point are determined: x = OP2 , y = P1 P2 , z = P P1 .
From metric (1.1) we can obtain two possible symmetry operations. These consist
of the reflections across the coordinate planes and the translation of the origin along
the direction of the x-axis (the values y and z are not modified).
584 Zoltán Gábos and Ágnes Mester

Based on metric (1.1), the geodesic lines verify

z y dx
cosh2 cosh2 = C1 , (1.2)
k k ds
where C1 is constant. From this we obtain
   2
d 2 z dy 1 2 z y y dx
cosh − cosh sinh cosh = 0, (1.3)
ds k ds k k k k ds
"  2  2 #
d2 z 1 z z 2 y dx dy
2
− sinh cosh cosh + = 0. (1.4)
ds k k k k ds ds

Figure 1

If we use x instead of variable s in (1.2), we can write equations (1.3) and (1.4)
in the following form:
d2 tanh ky 1 y
2
− 2 tanh = 0, (1.5)
! dx" k k #
2
d tanh kz d tanh ky

d 1 1 2 2 y z
2 y − 2
1 + k cosh tanh = 0. (1.6)
dx cosh k dx k k dx k
If we use variable x, we can apply the results obtained in the hyperbolic plane
by determining the function y = y(x). Moreover, we claim that the projections of
the geodesics in the three-dimensional space to the xOy plane are geodesics of the
two-dimensional plane.
Using (1.1) and (1.2), we get
" 2 #
d tanh ky d tanh kz 2
  
1 1 1 2 2 1
= + k + k . (1.7)
C12 cosh2 kz cosh2 ky dx cosh4 ky dx
 Lines in the three-dimensional hyperbolic geometry 585

The curvature of the geodesics equals zero,

1
= 0. (1.8)
rg
Metric (1.1) can also be obtained by using the metric
ds2 = dx21 + dx22 + dx23 − dx20 (1.9)
defined in the four-dimensional pseudo-Euclidean space, with the help of the following
equations:
x21 + x22 + x23 − x20 = −k 2 , (1.10)
x y z y z
x1 = k sinh cosh cosh , x2 = k sinh cosh , (1.11)
k k k k k
z x y z
x3 = k sinh , x0 = k cosh cosh cosh .
k k k k
If we use equations
ρ z ρ z
x1 = k cos ϕ sinh cosh , x2 = k sin ϕ sinh cosh , (1.12)
k k k k
z ρ z
x3 = k sinh , x0 = k cosh cosh ,
k k k
we obtain metric
z 2 ρ 
ds2 = cosh2 dρ + k 2 sinh2 dϕ2 + dz 2 , (1.13)
k k
where ρ, ϕ and z represent cylindrical coordinates (figure 1).
Metric (1.13) justifies that the rotation around axis Oz (the constant choices for
ρ and z) is a symmetry operation.
By choosing s as variable, the geodesic lines verify
ρ z dρ
sinh2 cosh2 = C2 , (1.14)
k k ds
where C2 is constant. We can also write
 2
d2 ρ 2 z dρ dz ρ ρ dϕ
+ tanh − k sinh cosh = 0, (1.15)
ds2 k k ds ds k k ds
"   2 #
2
d2 z 1 z z dρ 2 ρ dϕ
− sinh cosh + k 2 sinh = 0. (1.16)
ds2 k k k ds k ds
If we consider ϕ as variable, we will use the following differential equations:
d2 coth kρ ρ
2
+ coth = 0, (1.17)
dϕ k
! " 2 #
z
d coth kρ

d 1 d tanh k z 2
− tanh 1+k = 0. (1.18)
dϕ sinh2 ρ
k
dϕ k dϕ
Using (1.13) and (1.14), we obtain
( " 2 # 2 )
d coth kρ d tanh kz
 
1 2 1 1 2 1
=k +k + . (1.19)
C22 cosh2 kz sinh2 kρ dϕ sinh4 ρ
k
dϕ
586 Zoltán Gábos and Ágnes Mester

In the following sections we describe the different types of lines. Note that each
line verifies the differential equations which characterize the geodesics. Also, C1 and C2
are constant values. During the analysis our choice of coordinates may vary depending
on the form of calculations.

2. Lines crossing the origin

Let us consider the line passing through points O and P represented in figure 1,
where ϑ is the angle of intersection with axis Oz. In the OP1 P right triangle we can
write
z ρ
tanh = cot ϑ sinh . (2.1)
k k
The projection of line OP onto the xOy plane satisfies the following equation:
y x
tanh = tan ϕ sinh . (2.2)
k k
Using (1.11) and (1.12), we obtain
x y ρ
cosh cosh = cosh .
k k k
These formulas imply
z cot ϑ sinh xk
tanh = q . (2.3)
k cos ϕ 1 − tan2 ϕ sinh2 x
k

The lines verifying equations (2.2) and (2.3) also satisfy the (1.5) and (1.6) differential
equations. Using (1.7), we get
C1 = cos ϕ sin ϑ
constant. Therefore, the lines crossing the origin satisfy the conditions mentioned in
the previous section.
In the two-dimensional hyperbolic plane the orthogonal curves of lines crossing
the origin are circles. Based on the rotational symmetry operation, we claim that in the
three-dimensional case the orthogonal surfaces are spheres. By the use of cylindrical
coordinates we can write
ρ z R
cosh cosh = cosh . (2.4)
k k k
In order to determine the curvature of the sphere surface, we use the metric

ds2 = E(ρ)dρ2 + G(ρ)dϕ2

obtained from equations (1.13) and (2.4), where

sinh2 R 2 R
k cosh k R ρ
E(ρ) =
, G(ρ) = k 2 cosh2 tanh2 .
cosh2 kρ cosh k − cosh2 kρ
2 R k k
 Lines in the three-dimensional hyperbolic geometry 587

The Christoffel symbols of the second kind are as follows:

2 sinh kρ cosh kρ − cosh2 R ρ
k tanh k 1
Γ111 = , Γ212 = ,
cosh2 R cosh2 kρ k sinh cosh kρ
ρ


k k − k
 
k ρ 2 R 2 ρ
Γ122 = − tanh cosh − cosh ,
sinh2 R
k cosh2 R
k
k k k
where we used index 1 for ρ and index 2 for ϕ.
The components of the Ricci curvature tensor are
dΓ212 dΓ1
+ Γ212 Γ212 − Γ111 , R22 = − 22 + Γ122 Γ212 − Γ111 .



R11 =
dρ dρ
Using the expressions above, we obtain for the scalar curvature
1 1 2
R = R11 + R22 = − 2 . (2.5)
E G k sinh2 R
k

3. Lines crossing the x-axis

Let us consider the line passing through points P0 (a, 0, 0) and P1 (0, b, c) illus-
trated in figure 2. If we project this line onto the xOy plane, we get the line passing
through points P0 (a, 0, 0) and P2 (0, b, 0), which verifies
y b sinh a−x
k
tanh = tanh . (3.1)
k k sinh ka
The angle of intersection between the lines P0 P1 and P0 P2 is denoted by δ.

Figure 2

We can obtain the distance d between the points P0 (a, 0, 0) and P2 (0, b, 0) from
d a b
cosh = cosh cosh .
k k k
588 Zoltán Gábos and Ágnes Mester

Furthermore, distance d1 between P0 (a, 0, 0) and P3 (x, y, 0) verifies

d1 a−x y
cosh = cosh cosh .
k k k
If we consider the P0 P P3 right triangle, we can write
z d1
tanh = tan δ sinh ,
k k
while in the right triangle P0 P1 P2
c d
tanh = tan δ sinh .
k k
These formulas imply
z c B sinh a−x
k
tanh = tanh r , (3.2)
k k sinh2 a−x
A 1− tanh2 kb sinh2 ka
k

where s
tanh2 b
r
a b
B= 1+ k
, A= cosh2 cosh2 − 1.
sinh2 ka k k
Using (1.5) and (1.6) one can easily prove that equations (3.1) and (3.2) determine
geodesic lines. Also, formula (1.13) implies
! !
1 tanh2 kb tanh2 kc
= 1+ 1+ , (3.3)
C12 sinh2 ka cosh2 ka cosh2 kb − 1
thus C1 is constant.
Now we determine the orthogonal surface of the family of lines crossing point
P0 ∈ Ox. As the translation of the origin along the direction of the x-axis into point
P0 is a symmetry operation, we obtain spheres with center P0 . If we use Cartesian
coordinates, these spheres verify
x−a y z R
cosh cosh cosh = cosh . (3.4)
k k k k
The curvature of the orthogonal surface is determined by formula (2.5).
As a −→ ∞, we obtain lines being parallel to the x-axis:
y b x z tanh kc
tanh = tanh e− k , tanh = q .
k k k cosh2 b e2x − sinh2 b
k k

Thus we get
C1 = 1.
If
R = a, (3.5)
by applying equation (3.4), we obtain the equation of a parasphere containing the
origin:
y z x
cosh cosh = e k .
k k
 Lines in the three-dimensional hyperbolic geometry 589

Therefore the parasphere which contains point P (x0 , 0, 0) verifies

y z x−x0
cosh cosh = e k . (3.6)
k k
By using condition (3.5) and equation (2.5), the curvature of the parasphere becomes
R = 0.
This implies that we can use Euclidean geometry in order to study the surface of the
parasphere, fact which was also mentioned by Bolyai in his main work [1].

4. Lines crossing the z-axis

If we consider the set of lines crossing the z-axis, we can differentiate three types
of lines. The first set contains lines crossing the xOy plane, the second set consists
of lines which do not cross the xOy plane, finally, the lines of the third family are
parallel to the xOy plane.
In each case the projections of the lines contain the origin. Note that the rotation
around the z-axis is a symmetry operation. Therefore, we can determine the relevant
lines by using surfaces of revolution which are created by rotating the curves around
the z-axis in the xOy plane (the role of x is taken by ρ). On the other hand, the lines
which cross a projected line onto the xOy plane while being parallel to the z-axis
determine an orthogonal surface perpendicular to the xOy plane. The intersection of
this orthogonal surface and the surface of revolution determines the lines in question.
For fixed ϕ we obtain from metric (1.13)
z
ds2 = cosh2 dρ2 + dz 2 , (4.1)
k
which describes the orthogonal surfaces.
If we use s as variable, we can write
 2
2 z dρ d2 z 1 z z dρ
cosh = C, − sinh cosh = 0. (4.2)
k ds ds2 k k k ds
Then by substituting s with ρ, we obtain the following differential equation:
d2 tanh kz 1 z
− 2 tanh = 0. (4.3)
dρ2 k k
From (4.1) and (4.2) we get
2
d tanh kz

1 1 2
= +k .
C 2
cosh2 z
k
dρ
Applying (2.1), we obtain the condition
C = sin ϑ
for the lines passing through the origin.
a. Lines parallel to the xOy plane and lines crossing the xOy plane
590 Zoltán Gábos and Ágnes Mester

Using the rotational symmetry operation, we obtain for the line passing through
points P1 (0, 0, z0 ) and P2 (a, b, 0)
ρ −ρ
z z0 sinh 0k
tanh = tanh . (4.4)
k k sinh ρk0
This formula satisfies equation (4.3), where
ρ0 a b
cosh = cosh cosh .
k k k
Also
1 tanh2 zk0
= 1 + . (4.5)
C2 sinh2 ρk0
If ρ0 −→ ∞, it follows that
z z0 ρ
tanh = tanh e− k (4.6)
k k
and
C = 1.
b. Lines not crossing the xOy plane
In this case the lines have a minimum point. If we apply the rotational symmetry,
we obtain
ρ −ρ
z z0 cosh mk
tanh = tanh , (4.7)
k k cosh ρkm
where ρm denotes the value of ρ determined by the minimum point. For the value of
C we have
1 tanh2 zk0
= 1 − . (4.8)
C2 cosh2 ρkm
If ρm = 0, which means that the intersection coincides with the minimum point, we
can write
z z0 ρ
tanh = tanh cosh (4.9)
k k k
and
z0
C = cosh .
k
If we consider the lines passing through P (0, 0, z0 ), we get for the orthogonal curves
circles with center P in the xOz plane. Therefore, because of the rotational symmetry,
the orthogonal surfaces of the lines containing P (0, 0, z0 ) are spheres with center P ,
which verify
z0 z ρ z0 z R
cosh cosh cosh − sinh sinh = cosh . (4.10)
k k k k k k
Now let us consider the orthogonal surface which is perpendicular to the xOy plane
and contains the projected line. Here we use variables ρ and z. Furthermore, we will
use indexes 1 and 2 for two arbitrary lines which intersect in point P (ρ, z) on this
surface. Thus we obtain
   
dz1 dz2 z
+ cosh2 = 0,
dρ1 P dρ2 P k
 Lines in the three-dimensional hyperbolic geometry 591

where the lower index P means that we need to substitute the coordinates of the
intersection.
Using equations (4.4), (4.6) and (4.7), we get for the lines crossing the z-axis
z1 z0  ρ1 ρ1 
tanh = tanh cosh − p sinh . (4.11)
k k k k
In the three different cases (lines crossing the xOy plane, lines not crossing the xOy
plane, lines parallel to the xOy plane) the required values are as follows:
ρ0 ρm
coth , tanh and 1.
k k
By deriving equation (4.11) we obtain
dz1 z0 z1  ρ1 ρ1 
= tanh cosh2 sinh − p cosh . (4.12)
dρ1 k k k k
Then, using (4.11) and (4.12), we eliminate variable p. Thus we get
dz1 tanh zk0 cosh2 zk1  z0 z1 ρ1 
= ρ1 coth tanh cosh −1 .
dρ1 sinh k k k k
After differentiating equation (4.10) we obtain
dz2 coth zk0 sinh ρk2
=− .
dρ2 coth zk0 tanh zk2 coth ρk2 − 1
In the point of intersection we have ρ1 = ρ2 = ρ and z1 = z2 = z. Thus the orthogo-
nality condition holds, which proves the validity of equation (4.10).
Equation (4.10) can be written in the following form:
z0
ρ cosh R z
k + sinh k sinh k
cos = = F (ρ). (4.13)
k cosh zk0 cosh kz
Furthermore, equation (1.13) yields metric
 2
1 − F 2 + cosh2 kz dF
dρ z
ds2 = dρ2 + k 2 cosh2 F 2 − 1 dϕ2 .


(4.14)
F2 − 1 k
Using metric (4.14) and formula (4.13), we can obtain the curvature of the orthogonal
surface. The Ricci scalar is determined by formula (2.5).

5. Family of lines not having common point

In this section we consider two sets of lines.
a. Lines parallel to the z-axis
In this case, on the orthogonal surfaces the value of z is constant, z = z0 . Indeed,
the lines verify dρ1 = 0, while on the orthogonal surface dz2 = 0. Thus we obtain the
following orthogonality condition:
z
cosh dρ1 dρ2 + dz1 dz2 = 0.
k
592 Zoltán Gábos and Ágnes Mester

The orthogonal surface called hypersphere verifies

z0  2 ρ 
ds2 = cosh2 dρ + k 2 sinh2 dϕ2 , (5.1)
k k
the scalar curvature is
2
R = 2.
k
b. Lines having minimum point on the z-axis
Here we use formula (4.9), where the parameter is tanh zk0 .
By deriving (4.9) and eliminating the parameter, we obtain
dz1 z1 z1 ρ1
= sinh cosh tanh .
dρ1 k k k
The rotational symmetry operation induces for the orthogonal surface equation
ρ z ρ0
sinh cosh = sinh , (5.2)
k k k
where ρ0 is constant. Hence we get
dz2 z2 ρ2
= − coth coth .
dρ2 k k
This and the orthogonality condition proves formula (5.2).
From equations (1.13) and (5.2) we obtain metric
2 ρ
ρ0 1 + coth ρ0 2
ds2 = sinh2 k
dρ0 + k 2 sinh2 dϕ .
k sinh2 kρ k
Hence the scalar curvature of the orthogonal surface is
R = 0.
This means that this orthogonal surface is the dual of the parasphere.

6. Surfaces with constant curvature

For lines crossing axis Oz we applied equations of type
z
tanh = Φ(ρ), (6.1)
k
which were as follows: equation (2.2), (4.4), (4.7) and (4.6).
Using formulas (5.1) and (6.1), we obtain
ds2 = E(ρ)dρ2 + G(ρ)dϕ2
for the metric, where
 2
2 dz A
E(ρ) = cosh z(ρ) + = 2,
dρ (1 − Φ2 )
2 ρ
ρ sinh k
G(ρ) = k 2 cosh2 z(ρ) sinh2 = k2 .
k 1 − Φ2
 Lines in the three-dimensional hyperbolic geometry 593

In the four different cases (lines crossing the origin, lines crossing the xOy plane, lines
not crossing the xOy plane, lines parallel to the xOy plane) the values of the constant
A are as follows:
1 tanh2 zk0 tanh2 zk0
2 , 1+ 2 z0 , 1− and 1.
sin ϑ sinh k cosh2 ρkm
The Christoffel symbols of the second kind are as follows:
1 dE Φ dΦ
dρ
Γ111 = =2 ,
2E dρ 1 − Φ2
1 dG k ρ ρ
k2 ρ dΦ
Γ122 =− = − sinh cosh 1 − Φ2 − sinh2 Φ ,
2E dρ A k k A k dρ
 
1 dG 1 1 ρ dΦ
Γ212 2


= = 1 − Φ coth + Φ ,
2G dρ 1 − Φ2 k k dρ

while the components of the Ricci curvature tensor are

"  2 #
dΓ212 2 2 1

1 2 2 dΦ
R11 = + Γ12 Γ12 − Γ11 = 2 1−Φ +k ,
dρ 2 2
k (1 − Φ ) dρ
"  2 #
dΓ122 1 2 1

sinh2 kρ 2 2 dΦ
R22 = − + Γ22 Γ12 − Γ11 = 1−Φ +k .
dρ A (1 − Φ2 ) dρ
Therefore the scalar curvature is
1 1 2
R= R11 + R22 = 2
E G k
constant for all surfaces.

7. Lines not crossing the z-axis and the xOy plane

As the projection of these lines to the xOy plane verifies
y b x
tanh = tanh cosh , (7.1)
k k k
from equations (7.1) and (4.9) it follows that
z z0 cosh xk
tanh = tanh q . (7.2)
k k 1 − tanh2 b cosh2 x
k k

By deriving (7.2) we obtain

d tanh kz tanh zk0 sinh xk
=
3 . (7.3)
dx k 1 − tanh2 kb cosh2 xk 2
Using (1.7) and (7.3), we get for the value of C1
1 1 2 z0
C12
= 2 b − tanh k .
cosh k
594 Zoltán Gábos and Ágnes Mester

C1 is real if and only if cosh kb ≤ coth zk0 . Indeed, using formula (4.9) as z0 −→ ∞, we
get for the maximal value of ρ
ρm z0
cosh = coth .
k k
From !
d 1 d tanh kz cosh xk
=
dx cosh2 ky dx
3
k 2 1 − tanh y0 cosh2 x 2 cosh2 y0
k k k
and  z 2
y d tanh k 1
1 + k 2 cosh2 =
cosh2 yk0
1 − tanh yk0 cosh2 xk


k dx
it follows that the lines verifying (7.1) and (7.2) satisfy differential equations (1.5)
and (1.6).
If we use cylindrical coordinates, from
ρ b
coth = coth sin ϕ (7.4)
k k
and equation (4.9) we get
z z0 b sin ϕ
tanh = tanh coth q . (7.5)
k k k coth2 b sin2 ϕ − 1
k

These lines verify differential equations (1.17) and (1.18). Applying (1.19), we get for
the value of C2
k2
 
1 2 z0 2 b
= 1 − tanh cosh .
C22 sinh2 kb k k
Thus C2 is real if and only if cosh kb ≤ coth zk0 .
If we use equation (4.9) and formula
ρ a
coth = coth (sin ϕ + cos ϕ) , (7.6)
k k
we obtain a different line which satisfies
z z0 a sin ϕ + cos ϕ
tanh = tanh coth q . (7.7)
k k k coth2 a (sin ϕ + cos ϕ)2 − 1
k

The curves verifying (7.6) and (7.7) also satisfy the differential equations (1.17) and
(1.18). Also, from
1 cosh2 ka + cosh2 zk0
2 = k2
C2 sinh2 ka cosh2 zk0
we obtain a constant value for C2 . Thus these lines are lines of the hyperbolic space.
If we use Cartesian coordinates, instead of (7.6) and (7.7) we may write
y a x x
tanh = tanh cosh − sinh
k k k k
and
z z0 cosh xk
tanh = tanh q
2 .
k k
1 − tanh ka cosh xk − sinh xk
 Lines in the three-dimensional hyperbolic geometry 595

For each surface of revolution the scalar curvature equals the curvature of the xOy
plane and xOz is a plane of symmetry. Hence we obtain surfaces on the left and
the right side of the xOz plane. However, only equation (4.9) provides a necessary
condition. Let us consider a line crossing axis Oz, which connects two distinct surfaces.
The transitions between the line and the surfaces are smooth (the tangent vector field
is continuous) only in the case of (4.9). Therefore, new lines can only be derived by
the surface of revolution (4.9).

References
[1] Bolyai, J., Kárteszi, F., Appendix, a tér tudománya, Akadémiai Kiadó, Budapest, 1977.
[2] Dávid L., Bolyai geometria az Appendix alapján, Kolozsvár, 1944.
[3] Dubrovin, B.A., Fomenko, A.T., Novikov, S.P., Modern Geometry - Methods and Appli-
cations, Springer, New York, 1984.
[4] Gábos, Z., Mester, Á., Curves with constant geodesic curvature in the Bolyai-
Lobachevskian plane, Stud. Univ. Babeş-Bolyai Math., 60(2015), no. 3, 449-462.
[5] Liebmann, H., Nichteuklidische Geometrie, Leipzig, Berlin, 1923.
[6] Lobacsevszkij., N.I., Kárteszi F., Geometriai vizsgálatok a párhuzamosok elméletének
köréből, Akadémiai Kiadó, Budapest, 1951.
[7] Lobacsevszkij, N.I., Pangeometrie, Leibzig, 1902.
[8] Sanielevici, S., Opera matematică (Romanian), [Mathematical Works], Editura
Academiei Republicii Socialiste România, Bucharest, 1968.
[9] Szász, P., Bevezetés a Bolyai-Lobacsevszkij-féle geometriába, Akadémiai Kiadó, Bu-
dapest, 1973.
[10] Thorpe, J. A., Elementary Topics in Differential Geometry, Springer, New York, 1994.

Zoltán Gábos
Babeş-Bolyai University, Faculty of Physics
Cluj-Napoca, Romania
e-mail: [email protected]
Ágnes Mester
Babeş-Bolyai University, Faculty of Mathematics and Computer Science
Cluj-Napoca, Romania
e-mail: [email protected]
Stud. Univ. Babeş-Bolyai Math. 60(2015), No. 4, 597–601

Multisymplectic connections on supermanifolds

Masoud Aminizadeh and Mina Ghotbaldini

Abstract. In this paper we show that on any multisymplectic supermanifold there

exist a connection compatible to the multisymplectic form.
Mathematics Subject Classification (2010): 58A50, 53D05.
Keywords: Multisymplectic supermanifolds, multisymplectic connections.

1. Introduction
Multisymplectic structures in field theory play a role similar to that of sym-
plectic structures in classical mechanics. In the other hand supergeometry plays an
important role in physics. In [2] and [3], the authors studied geometry of symplectic
connections and in [1], the author studied symplectic connections on supermanifold.
In this paper we study multisymplectic connections on supermanifolds.
A supermanifold M of dimension n|m is a pair (M, OM ), where M is a Hausdorff
topological space and OM is a sheaf of commutative superalgebras with unity over
R locally isomorphic to Rm|n = (Rn , ORn ⊗ Λη1 ,...,ηm ), where ORn is the sheaf of
smooth functions on Rn and Λη1 ,...,ηm is the grassmann superalgebra of m generators
(for more details see [5]).

If M is a supermanifold of dimension n|m, we define the tangent sheaf as follows,

TM (U ) = Der(OM (U )),
the OM (U )-supermodule of derivations of OM (U ). TM is locally free of dimension
n|m. The sections of TM are called vector fields.
Definition 1.1. If ξ be a locally free sheaf of OM -supermodules on M, a connection
on ξ is a morphism ∇ : TM ⊗R ξ → ξ of sheaves of supermodules over R such that

∇f X v = f ∇X v, ∇X f v = (Xf ) + (−1)X f f ∇X v and ∇

] Xv = v
e + X,
ee e

for all homogeneous function f , vector fields X and section v of ξ. (In the case ξ = TM
we speak of a connection on M).
598 Masoud Aminizadeh and Mina Ghotbaldini

We define the torsion of a connection ∇ on TM by

T (X, Y ) = ∇X Y − (−1)X Y ∇Y X − [X, Y ].
ee

Definition 1.2. A graded Riemannian metric on supermanifold M is a graded-

symmetric non-degenerate OM -linear morphism of sheaves
g : TM ⊗ TM → OM .
A supermanifold equipped with graded Riemannian metric is called a Riemannian
supermanifold. If M is a Riemannian supermanifold with Riemannian metric g, we
call a connection ∇ metric if ∇g = 0.
On a suppermanifold M with a Riemannian metric g, there exist a unique torsion
free and metric connection ∇0 , which will be called the Levi-Civita connection of the
metric(see [4]).

2. Multisymplectic connections on supermanifolds

Let us consider a multisymplectic supermanifold of degree k (M, ω), i.e. a su-
permanifold M with a closed non-degenerate graded differential k-form ω.
Definition 2.1. A multisymplectic connection on M is a connection for which:
i) The torsion tensor vanishes, i.e.
∇X Y − (−1)X Y ∇Y X = [X, Y ].
ee

ii) It is compatible to the multisymplectic form, i.e. ∇ω = 0.

To prove the existence of such a connection, take ∇0 to be the Levi-Civita
connection associated to a metric g on M. Consider tensor N on M defined by
∇0Y0 ω(Y1 , Y2 , ..., Yk ) = (−1)ωe Y0 ω(N (Y0 , Y1 ), Y2 , ..., Yk ).
f

We shall proof some properties of N .

Lemma 2.2. We have
i) ω(N (Y0 , Y1 ), Y2 , ..., Yk ) = −(−1)Y1 Y2 ω(N (Y0 , Y2 ), Y1 , ..., Yk );
ff
P
ii) ω(N (Y0 , Y1 ), Y2 , ..., Yk ) + Σki=1 (−1)i+ p<i Yp Yi ω(N (Yi , Y0 ), Y1 , ..., Ŷi , ..., Yk ) = 0,
fe

where the hats indicate omitted arguments.

Proof. We first prove (i)
ω(N (Y0 , Y1 ), Y2 , ..., Yk ) = (−1)Y0 ωe ∇0Y0 ω(Y1 , Y2 , ..., Yk )
f

= −(−1)Y0 ωe +Y1 Y2 ∇0Y0 ω(Y2 , Y1 , ..., Yk )

f ff

= −(−1)Y1 Y2 ω(N (Y0 , Y2 ), Y1 , ..., Yk ).

ff

For proof (ii) we know dω = 0 so

k
X P
0 = dω(Y0 , Y1 , ..., Yk ) = (−1)i+Yi (w+ Y p)
Yi (ω(Y0 , ..., Ŷi , ..., Yk ))
e f
e p<i

i=0
 Multisymplectic connections on supermanifolds 599
X P
+ (−1)j+ Y j Yp
ω(Y0 , ..., Yi−1 , [Yi , Yj ], Yi+1 , ..., Yˆj , ..., Yk )
f f
i<p<j

i<j
k
X e e P
= (−1)i+Yi (w+ p<i Yp ) Y (ω(Y , ..., Ŷ , ..., Y ))
f
i 0 i k
i=0
X P
+ (−1) j+ i<p<j Y j Yp
ω(Y0 , ..., Yi−1 , ∇0Yi Yj − (−1)Yi Yj ∇0Yj Yi , Yi+1 , ..., Yˆj , ..., Yk )
f f ef

i<j
k
X e e P
= (−1)i+Yi (w+ p<i Yp ) Y (ω(Y , ..., Ŷ , ..., Y ))
f
i 0 i k
i=0
X P
+ (−1) j+ i<p<j Y j Yp
ω(Y0 , ..., Yi−1 , ∇0Yi Yj , Yi+1 , ..., Yˆj , ..., Yk )
f f

i<j
X P
− (−1)j+ i≤p<j Yj Yp ω(Y0 , ..., Yi−1 , ∇0Yj Yi , Yi+1 , ..., Yˆj , ..., Yk )
ff

i<j
k
X e e P
= (−1)i+Yi (w+ p<i Yp ) Y (ω(Y , ..., Ŷ , ..., Y ))
f
i 0 i k
i=0
X P
+ (−1) j+ i<p<j Y j Yp
ω(Y0 , ..., Yi−1 , ∇0Yi Yj , Yi+1 , ..., Yˆj , ..., Yk )
f f

i<j
X P
− (−1)i+ Yei Y p
ω(Y0 , ..., Yj−1 , ∇0Yi Yj , Yj+1 , ..., Ŷi , ..., Yk )
f
j≤p<i

j<i
k
X e e P
= (−1)i+Yi (w+ p<i Yp ) Y (ω(Y , ..., Ŷ , ..., Y ))
f
i 0 i k
i=0
X P
i+ i<p<j Yei Y
− (−1) p
ω(Y0 , ..., Yi−1 , Ŷi , ..., Yj−1 , ∇0Yi Yj , Yj+1 , ..., Yk )
f

i<j
X P
− (−1)i+ Yei Y p
ω(Y0 , ..., Yj−1 , ∇0Yi Yj , Yj+1 , ..., Ŷi , ..., Yk )
f
j≤p<i

j<i
k
X e e P
= (−1)i+Yi (w+ p<i Yp ) (Y (ω(Y , ....Ŷ , ..., Y ))
f
i 0 i k
i=0
X P
− (−1)Yi (eω+ Y p)
ω(Y0 , ..., Yj−1 , ∇0Yi Yj , ..., Ŷi , ..., Yk )
e f
p<i

j
k
X e e P
= (−1)i+Yi (w+ p<i Yp ) ∇0 ω(Y , ..., Ŷ , ..., Y )
f
Yi 0 i k
i=0

= (−1)Y0 ωe ∇0Y0 ω(Y1 , ..., Yk )

f

k
X P
+ (−1)i+Yi (eω+ p<i Yp ) ∇0Yi ω(Y0 , ..., Ŷi , ..., Yk )
e f

i=1
P
= ω(N (Y0 , Y1 ), Y2 , ..., Yk ) + Σki=1 (−1)i+ Y p Yi
ω(N (Yi , Y0 ), Y1 , ..., Ŷi , ..., Yk ). 
f e
p<i
600 Masoud Aminizadeh and Mina Ghotbaldini

Now we show that on any multisymplectic supermanifold there exist a connection

compatible to the multisymplectic form.
Theorem 2.3. Let (M, ω) be a multisymplectic supermanifold. Then on M there is at
least a multisymplectic connection.
Proof. We define now a new connection ∇ as follows
1 (−1)X Y
ee
∇X Y = ∇0X Y + N (X, Y ) + N (Y, X).
k+1 k+1
It is easy to show that ∇ is a torsion free connection. We show that the connection
is compatible with the multisymplectic form ω, i.e. ∇ω = 0. We have
∇Y0 ω(Y1 , ..., Yk ) = Y0 (ω(Y1 , ..., Yk ))
k
X P
− (−1)Y0 (eω+ Y p)
ω(Y1 , ..., Yi−1 , ∇Y0 Yi , Yi+1 , ..., Yk )
f f
p<i

i=1
k
X P
= Y0 (ω(Y1 , ..., Yk )) − (−1)Y0 (eω+ p<i Yp ) ω(Y1 , ..., Yi−1 , ∇0Y0 Yi
f f

i=1

1 (−1)Y0 Yi
fe
+ N (Y0 , Yi ) + N (Yi , Y0 ), Yi+1 , ..., Yk )
k+1 k+1
k
X P
= Y0 (ω(Y1 , ..., Yk )) − (−1)Y0 (eω+ 1≤p<i Yp ) ω(Y1 , ..., Yi−1 , ∇0Y0 Yi , Yi+1 , ..., Yk )
f f

i=1
k
1 X P
− (−1)Y0 (eω+ Y p)
ω(Y1 , ..., Yi−1 , N (Y0 , Yi ), Yi+1 , ..., Yk )
f f
1≤p<i
k+1 i=1
k
1 X P
− (−1)Y0 (eω+ 1≤p≤i Yp ) ω(Y1 , ..., Yi−1 , N (Yi , Y0 ), Yi+1 , ..., Yk )
f f
k + 1 i=1
= ∇0Y0 ω(Y1 , ..., Yk )
k
1 X e P
− (−1)i−1 (−1)Y0 ωe +Yi 1≤p<i Yp ω(N (Y0 , Yi ), Y1 , ..., Ŷi , .., Yk )
f f
k + 1 i=1
k
1 X e P
− (−1)i−1 (−1)Y0 ωe +Yi 0≤p<i Yp ω(N (Yi , Y0 ), Y1 , ..., Ŷi , ..., Yk )
f f
k + 1 i=1
k
= (−1)Y0 ωe ω(N (Y0 , Y1 ), Y2 , ..., Yk ) − (−1)Y0 ωe ω(N (Y0 , Y1 ), Y2 , ..., Yk )
f f
k+1
k
1 X e P
+ (−1)i+Y0 ωe +Yi 0≤p<i Yp ω(N (Yi , Y0 ), Y1 , ..., Ŷi , ..., Yk )
f f
k + 1 i=1
1
= (−1)Y0 ωe (ω(N (Y0 , Y1 ), Y2 , ..., Yk )
f
k+1
 Multisymplectic connections on supermanifolds 601

k
X e P
+ (−1)i+Yi p<i Yp ω(N (Yi , Y0 ), Y1 , ..., Ŷi , ..., Yk )) = 0. 
f

i=1
0
Let now ∇ be a multisymplectic connection and ∇X Y = ∇X Y + S(X, Y ), where
S is a tensor field on M. We have
Theorem 2.4. ∇0 is a multisymplelectic connection if and only if S is supersymmetric
and X P
(−1) p<i Y0 Yp ω(Y1 , ..., Yi−1 , S(Y0 , Yi ), Yi+1 , ..., Yk ) = 0.
ff

Proof. If we want ∇0 to be torsion free then

∇Y X + S(X, Y ) − (−1)X Y ∇Y X − (−1)X Y S(Y, X) = [X, Y ].
ee ee

So S(X, Y ) = −(−1)X Y S(Y, X). If ∇0 be compatible to the multisymplectic form ω.

ee

We have 0
0 = ∇Y0 ω(Y1 , ..., Yk ) = Y0 (ω(Y1 , ..., Yk ))
X P 0
− (−1)Y0 (eω+ p<i Yp ) ω(Y1 , ..., Yi−1 , ∇Y0 Yi , Yi+1 , .., Yk )
f f

i
P
= ∇Y0 ω(Y1 , ..., Yk ) − (−1)Y0 ωe (Σi (−1) p<i Y0 Yp ω(Y1 , ..., Yi−1 , S(Y0 , Yi ), Yi+1 , ..., Yk )).
f ff

So X P
(−1) p<i Y0 Yp ω(Y1 , ..., Yi−1 , S(Y0 , Yi ), Yi+1 , ..., Yk ) = 0. 
ff

References
[1] Blaga, P.A., Symplectic connections on supermanifolds: Existence and non-uniqueness,
Stud. Univ. Babeş-Bolyai Math., 58(2013), no. 4, 477-483.
[2] Bieliavsky, P., Cahen, M., Gutt, S., Rawnsley, J., Schwachhofer, L., Symplectic connec-
tions, math/0511194.
[3] Gelfand, I., Retakh, V., Shubin, M., Fedosov manifolds, Advan. Math., 136(1998), 104-
140.
[4] Goertsches, O., Riemannian supergeometry, Math. Z., 260, (2008), 557-593.
[5] Leites, D.A., Introduction to the theory of supermanifolds, Russian Mathematical Sur-
veys, 35(1980), 1-64.

Masoud Aminizadeh
“Vali-e-Asr” University of Rafsanjan
Department of Mathematics
Rafsanjan, Iran
e-mail: [email protected]
Mina Ghotbaldini
“Vali-e-Asr” University of Rafsanjan
Department of Mathematics
Rafsanjan, Iran
Stud. Univ. Babeş-Bolyai Math. 60(2015), No. 4, 603–609

On applications of Andrica-Badea and Nagy

inequalities in spectral graph theory
Igor Milovanović, Emina Milovanović and Edin Glogić

Abstract. Applications of Andrica-Badea and Nagy inequalities for determining

bounds of graph invariants of undirected, connected graphs are investigated. We
consider bounds of the following invariants: the first Zagreb index, general Randić
index, Laplacian linear spread and normalized Laplacian spread of graphs.
Mathematics Subject Classification (2010): 60E15, 05C50.
Keywords: Inequalities, first Zagreb index, spread of graph.

1. Introduction
Andrica and Badea (see [1]) have proved the following result.
Let p1 , p2 , . . . , pn be non-negative real numbers and a1 , a2 , . . . , an and b1 , b2 , . . . , bn
real numbers with the properties
0 < r1 ≤ ai ≤ R1 < +∞ and 0 < r2 ≤ bi ≤ R2 < +∞
for each i = 1, 2, . . . , n. Further, let S be a subset of In = {1, 2, . . . , n} which minimizes
the expression
n
X 1X
pi − pi . (1.1)
2 i=1
i∈S

Then
n
X n
X n
X n
X
pi pi ai bi − pi ai pi bi
i=1 i=1 i=1 i=1

n
!
X X X
≤ (R1 − r1 )(R2 − r2 ) pi pi − pi (1.2)
i∈S i=1 i∈S

In [17] Nagy has proved the following result:

604 Igor Milovanović, Emina Milovanović and Edin Glogić

Let a1 , a2 , . . . , an are real numbers with the property r ≤ ai ≤ R, for each i =

1, 2, . . . , n. Then
n n
!2
X
2
X n
n ai − ai ≥ (R − r)2 . (1.3)
i=1 i=1
2

In this paper we consider bounds of some graph invariants and prove that they
are direct corollaries of inequalities (1.2) and (1.3). Some of the obtained bounds are
better than those obtained in the literature so far.
In the next section we recall some results from spectral graph theory needed for
our work.

2. Laplacian and normalized laplacian spectrum of graph

Let G = (V, E), V = {1, 2, . . . , n}, be undirected connected graph with n vertices
and m edges, with sequence of vertex degrees d1 ≥ d2 ≥ · · · ≥ dn > 0, di = d(i),
i = 1, 2, . . . , n. Denote with A adjacency matrix of G. Its eigenvalues λ1 ≥ λ2 ≥
· · · ≥ λn represent ordinary eigenvalues of graph G. If D = diag(d1 , d2 , . . . , dn ) is
diagonal matrix of vertex degrees, then L = D − A is the Laplacian matrix of the G.
Eigenvalues of L, µ1 ≥ µ2 ≥ · · · ≥ µn−1 > µn = 0 represent Laplacian eigenvalues of
graph G. The main properties of these eigenvalues are (see [3, 7, 8, 15])
n−1
X n
X n−1
X n
X n
X
µi = di = 2m and µ2i = d2i + di = M1 + 2m, (2.1)
i=1 i=1 i=1 i=1 i=1
Pn
where M1 = i=1 d2i is the first Zagreb index (see [13]).
Because the graph G is assumed to be connected, it has no isolated vertices
and therefore the matrix D−1/2 is well–defined. Then L∗ = D−1/2 LD−1/2 is the
normalized Laplacian matrix of the graph G. Its eigenvalues are ρ1 ≥ ρ2 ≥ · · · ≥
ρn−1 > ρn = 0. Main properties of these eigenvalues are given by (see [19])
n−1
X n−1
X
ρi = n and ρ2i = n + 2R−1 , (2.2)
i=1 i=1
−1
P
where R−1 = {i,j}∈E (di dj ) is the general Randić index (see [6, 18]).

3. Main result
In the following theorem we prove the inequality that establishes lower and upper
bounds for invariant M1 in terms of parameters n, m, d1 and dn .
Theorem 3.1. Let G = (V, E) be undirected connected graph with n, n ≥ 2, vertices
and m edges. Then
4m2 1 4m2
+ (d1 − dn )2 ≤ M1 ≤ + nα(n)(d1 − dn )2 , (3.1)
n 2 n
 Andrica-Badea and Nagy inequalities in spectral graph theory 605

where
 (1
(−1)n+1 + 1
  
1 jnk 1 jnk 1 , if n is even
α(n) = 1− = 1− = n4 2 −1
n 2 n 2 4 2n2 4n2 , if n is odd
Equality holds if and only if G is regular graph.
Proof. For ai = di , i = 1, 2, . . . , n, R = d1 and r = dn , the inequality (1.3) transforms
into
n n
!2
X
2
X n
n di − di ≥ (d1 − dn )2 ,
i=1 i=1
2
i.e. according to (2.1), into
n
nM1 − 4m2 ≥ (d1 − dn )2 , (3.2)
2
wherefrom the left part of inequality (3.1) is obtained.
For pi = 1, i = 1, 2, . . ., n and S = {1, 2, . . . , k} ⊂ In , the expression (1.1)
reaches the minimum for k = n2 . Now for S = {1, 2, . . . , b n2 c}, pi = 1, ai = bi = di ,
i = 1, 2, . . . , n, r1 = r2 = dn and R1 = R2 = d1 , the inequality (1.2) becomes
n n
!2
X X jnk  j n k
2
n di − di ≤ (d1 − dn )2 n−
i=1 i=1
2 2
i.e.
nM1 − 4m2 ≤ n2 (d1 − dn )2 α(n) (3.3)
where
 (1
(−1)n+1 + 1
  
1 jnk 1 jnk 1 , if n is even
α(n) = 1− = 1− 2
= n4 2 −1 .
n 2 n 2 4 2n 4n2 , if n is odd

From (3.3) right side of inequality (3.1) immediately follows.

Equalities in (3.2) and (3.3) hold if and only if d1 = d2 = · · · = dn , so the
equalities in (3.1) hold if and only if G is a regular graph. 

Remark 3.2. Since (d1 − dn )2 ≥ 0, left inequality in (3.1) is stronger than

4m2
M1 ≥ (3.4)
n
which was proved in [9].
q
nM1
Remark 3.3. In [2] the invariant irrEB (G) = 4m2 − 1 as the irregularity measure
of graph was introduced. In [11] another irregularity measure irrg (G) = ddn1 − 1 was
defined. According to (3.1) we can establish the following relationship between these
two measures
r r
ndn n2 dn α(n)
2
irrg (G) ≤ irrEB (G) ≤ irrg (G).
8m 4m2
606 Igor Milovanović, Emina Milovanović and Edin Glogić

The linear Laplacian spread of graph G is defined as µ1 − µn−1 . The following

theorem establishes lower and upper bounds for this invariant in terms of parameters
n, m and M1 .
Theorem 3.4. Let G = (V, E) be undirected connected graph with n, n ≥ 2, vertices
and m edges. Then
s r
(n − 1)(M1 + 2m) − 4m2 2((n − 1)(M1 + 2m) − 4m2 )
2
≤ µ1 − µn−1 ≤ . (3.5)
(n − 1) α(n − 1) n−1
where
(−1)n + 1
     
1 n−1 1 n−1 1
α(n − 1) = 1− = 1− .
n−1 2 n−1 2 4 2(n − 1)2
∼ Kn .
Equalities hold if and only if G is a complete graph, G =
Proof. For n := n−1, ai = µi , i = 1, 2, . . . , n−1, R = µ1 and r = µn−1 , the inequality
(1.3) becomes
n−1 n−1
!2
X
2
X (n − 1)
(n − 1) µi − µi ≥ (µ1 − µn−1 )2
i=1 i=1
2
i.e. according to (2.1) we have
(n − 1)
(n − 1)(M1 + 2m) − 4m2 ≥ (µ1 − µn−1 )2 (3.6)
2
wherefrom right side of (3.5) is obtained.
For n := n − 1, pi = 1, ai = bi = µi , i = 1, 2, . . . , n − 1, r1 = r2 = µn−1 and
R1 = R2 = µ1 , the inequality (1.2) transforms into
n−1 n−1
!2    
X
2
X n−1 n−1
(n − 1) µi − µi ≤ (µ1 − µn−1 )2 n−1−
i=1 i=1
2 2
i.e.
(n − 1)(M1 + 2m) − 4m2 ≤ (µ1 − µn−1 )2 (n − 1)2 α(n), (3.7)
where
    ( 1
1 n−1 1 n−1 if n is even
α(n − 1) = 1− = 4n(n−2) .
n−1 2 n−1 2 4(n−1)2 , if n is odd
From (3.7) left part of inequality (3.5) is directly obtained.
Equalities (3.6) and (3.7) hold if and only if µ1 = µ2 = · · · = µn−1 , hence
equalities in (3.5) hold if and only if G is a complete graph, G ∼
= Kn . 
Remark 3.5. Right side of inequality (3.5) was proved in [16]. Since α(n − 1) ≤ 14 , for
each n, left side of inequality (3.5) is stronger than
2 p
µ1 − µn−1 ≥ (n − 1)(M1 + 2m) − 4m2 ,
n−1
for even n. The above inequality was proved in [10] and [20].
 Andrica-Badea and Nagy inequalities in spectral graph theory 607
 
2m
From inequality (3.4) and M1 ≤ m n−1 + (n − 2) , proved in [5], and inequality
4m2
M1 ≥ n , proved in [9], the following corollary of Theorem 3.4 holds.
Corollary 3.6. Let G = (V, E) be undirected connected graph with n, n ≥ 2, vertices
and m edges. Then
s r
1 2m(n(n − 1) − 2m) 2m(n(n − 1) − 2m)
≤ µ1 − µn−1 ≤ .
n−1 nα(n − 1) n−1
∼ Kn .
Equalities hold if and only if G is a complete graph, G =
In the following theorem we determine lower and upper bounds for graph invari-
ant R−1 in terms of parameters n, ρ1 and ρn−1 .
Theorem 3.7. Let G = (V, E) be undirected connected graph with n, n ≥ 2, vertices
and m edges. Then
n 1 n (n − 1)α(n − 1)
+ (ρ1 − ρn−1 )2 ≤ R−1 ≤ + (ρ1 − ρn−1 )2 . (3.8)
2(n − 1) 4 2(n − 1) 2
Equalities hold if and only if G is a complete graph, G ∼
= Kn .
Proof. for n := n − 1, ai = ρi , i = 1, 2, . . . , n − 1, r = ρn−1 and R = ρ1 the inequality
(1.3) becomes
n−1 n−1
!2
X
2
X n−1
(n − 1) ρi − ρi ≥ (ρ1 − ρn−1 )2
i=1 i=1
2
i.e. according to (2.2)
n−1
(n − 1)(n + 2R−1 ) − n2 ≥ (ρ1 − ρn−1 )2 , (3.9)
2
wherefrom left side of inequality (3.8) is obtained.
For n := n − 1, pi = 1, ai = bi = ρi , i = 1, 2, . . . , n − 1, r1 = r2 = ρn−1 and
R1 = R2 = ρ1 , inequality (1.3) transforms into
n−1 n−1
!2    
X
2
X n−1 n−1
(n − 1) ρi − ρi ≤ (ρ1 − ρn−1 )2 n−1− ,
i=1 i=1
2 2
i.e.
(n − 1)(n + 2R−1 ) − n2 ≤ (n − 1)2 α(n − 1)(ρ1 − ρn−1 )2 , (3.10)
wherefrom right part of inequality (3.8) is obtained.
Equalities in (3.9) and (3.10) hold if and only if ρ1 = ρ2 = · · · = ρn−1 , therefore
equalities in (3.8) hold if and only if G is complete graph, G ∼
= Kn . 
Remark 3.8. Since (ρ1 − ρn−1 )2 ≥ 0, it follows that left side of inequality (3.8) is
stronger than inequality
n
R−1 ≥
2(n − 1)
which was proved in [14].
608 Igor Milovanović, Emina Milovanović and Edin Glogić

Remark 3.9. Inequalities in (3.8) can be presented in an equivalent form as

s r
2(n − 1)R−1 − n 2(2(n − 1)R−1 − n)
≤ ρ1 − ρn−1 ≤ . (3.11)
(n − 1)2 α(n − 1) n−1
For even n, left side of inequality (3.11) is stronger than inequality
2 p
ρ1 − ρn−1 ≥ 2(n − 1)R−1 − n,
n−1
which was proved in [4].
Acknowledgement. This work was supported by the Serbian Ministry of Education
and Science, Project No TR32012 and TR32009.

References
[1] Andrica, D., Badea, C., Grüs inequality for positive linear functionals, Period. Math.
Hungar, 19(1988), no. 2, 155–167.
[2] Bell, F.K., A note on the irregularity of graphs, Lin. Algebra Appl., 161(1992), 45–54.
[3] Biggs, N., Algebraic graph theory, Cambridge Univ. Press, Cambridge, UK, 1993.
[4] Butler, S., Eigenvalues and structures of graph, Ph. Dissertation, University of Califor-
nia, San Diego, 2008.
[5] de Caen, D., An upper bound on the sum of squares of degrees in a graph, Discr. Math.,
185(1998), no. 1-3, 245–248.
[6] Cavers, M., Fallat, S., Kirkland, S., On the normalized Laplacian energy and general
Randic index of graphs, Lin. Algebra Appl., (2010), no. 433, 172–190.
[7] Chung, F.R.K., Spectral graph theory, Am. Math. Soc., Providence, 1997.
[8] Cvetković, D., Doob, M., Sachs, H., Spectra of graph theory and application, New York,
Academic Press, 1980.
[9] Edwards, C.S., The largest vertex degre sum for a triangle of a graph, Bull. London
Math. Soc., 9(1977), 203–208.
[10] Fath-Tabar, G.H., Ashrafi, A.R., Some remark on Laplacian eigenvalues of graphs, Math.
Commun., 15(2010), no. 2, 443-451.
[11] Goldberg, F., New results on eigenvalues on degree deviation http://orXiv
org.1403.269V1, 2014.
[12] Gutman, I., Furtula, B., Elphick, C., Tree new/old vertex-degree based toplogoical indices,
MATCH Commun. Math. Comput. Chem., 72(2014), 617–632.
[13] Gutman, I., Trinajstić, N., Graph theory and molecular orbitas. Total π-electron energy
of alternant hydrocarbons, Chem. Phys. Letters, 17(1972), 535–538.
[14] Li, X., Yang, Y., Sharp bounds for the general Randić index, MATCH Commun. Math.
Comput. Chem., 51(2004), 155–166.
[15] Merris, R., Laplacian matrices of graphs: A survey, Lin. Algebra Appl., 197-198(1994),
143–176.
[16] Milovanović, I., Milovanović, E., Remark on inequalities for the Laplacian spread of
graphs, Czeh. Math. J., 64 (2014), no. 139, 285–287.
 Andrica-Badea and Nagy inequalities in spectral graph theory 609

[17] Nagy, J.V.S., Uber algebraische Gleichungen mit lauter reellen Wurzeln, Jahresbericht
der deutschen mathematiker-Vereingung, 27(1918), 37-43.
[18] Randić, M., On characterization of molecular branching, J. Amer. Chem. Soc., 97(1975),
6609–6615.
[19] Zumstein, P., Comparison of spectral methods through the adjacency matrix and the
Laplacian of a graph, Ph. Dissertation, ETH Zürich, 2005.
[20] You, Z., Liu, B., The Laplacian spread of graphs, Czech. Math. J., 62(2012), 159–168.

Igor Milovanović
Faculty of Electronic Engineering
Beogradska 14, 18000 Niš, Serbia
e-mail: [email protected]
Emina Milovanović
Faculty of Electronic Engineering
Beogradska 14, 18000 Niš, Serbia
e-mail: [email protected]
Edin Glogić
State University of Novi Pazar
Novi Pazar, Serbia
e-mail: edin [email protected]
Stud. Univ. Babeş-Bolyai Math. 60(2015), No. 4, 611–621

Conjugate free convection in a vertical channel

filled with nanofluid
Flavius Pătrulescu and Teodor Groşan

Abstract. The steady natural conjugate convection in a long vertical channel

filled with a nanofluid and including internal heat generation is presented in this
paper. A new mathematical model is proposed for the momentum, energy and
nanoparticles’ concentration equations. The system of partial differential equa-
tions is written in terms of dimensionless velocity, temperature and concentra-
tion of the nanoparticles and is solved analytically. The effects of the governing
parameters, such as the ratio between the thermophoresis parameter and the
Brownian motion parameter, R, and the buoyancy ratio parameter, N r, on the
velocity, temperature and nanoparticles’ concentration are studied. It is found
that the addition of the nanoparticles into the fluid reduces the temperature and
enhances the heat transfer. A limit case when the thermal conductivity of the
nanoparticles is much larger than the thermal conductivity of the base fluid has
been also studied.
Mathematics Subject Classification (2010): 76R10, 82D15, 82D80.
Keywords: Free convection, heat transfer, nanofluid, Brownian motion, ther-
mophoresis, heat generation, thermal energy.

1. Introduction
General fluids used in industrial processes involving heat transfer (energy gen-
eration, insulation, cooling of microelectronic components) are water, mineral oil,
ethylene glycol, etc. (see [3], [11]). Low values of the physical properties of these flu-
ids (thermal conductivity, density, expansion coefficient, etc.) limit the efficiency of
heat transport and it is necessary to obtain new type of fluids, having improved heat
transfer characteristics (see [8]). In order to enhance the thermal characteristics of
the fluids, one can form mixtures by adding ultra-fine solid particles (metallic, non-
metallic or polymeric) to the fluid. Choi [7] was probably, the first one who called the
fluids with particles of nanometer dimensions nanofluids.
Over the last twenty years, many industrial processes, biology, medicine, cat-
alytic chemistry and environmental applications started to use nanotechnologies (see
612 Flavius Pătrulescu and Teodor Groşan

[13] and [17]). Different mathematical models were used by several authors to describe
heat transfer in nanofluids. Among these models the most used are those where the
concentration of nanoparticles is constant and the addition of nanoparticles into the
base fluid improved their physical properties (see [18]). Moreover, other models based
on physical properties variation include thermal dispersion (see [12]) or Brownian mo-
tion (see [14]). A more complex mathematical model (see [4]) considers that nanopar-
ticles’ concentration is variable and incorporates the effects of Brownian motion and
thermophoresis. Recently, Celli [6] had the idea to combine the model proposed by
Buongiorno [4] and the model based on improved physical properties considering for
the last one an average concentration of nanoparticles.
Many times theoretical problems as well as industrial processes and natural
phenomena are modelled using simple geometries such as infinite channels (see [1],
[2]). However, in real simulation it is necessary to take into account the interaction
between the convective heat transfer in nanofluid and conductive heat transfer in
the thick solid walls. Such situations (i.e. conjugate heat transfer) appears in cellular
structures, cavities or channels with solid walls, etc.
Several authors such as Pătrulescu and Groşan [16], Groşan [9], Groşan and Pop
[10] and Li [15] have studied the fully developed flow in a vertical channel filled by a
nanofluid using different mathematical models for nanofluid and different boundary
conditions. In the present paper, the fully developed conjugate heat transfer in a
vertical channel filled with a nanofluid when heat generation in the solid wall is
considered has been studied analytically.

2. Notations and preliminaries

Consider the fully developed steady conjugate free convection flow of an in-
compressible nanofluid in vertical channel differentially heated. The left wall of the
channel is kept at a constant temperature TH , while the right wall has a constant
temperature TC . We consider a two-dimensional coordinate frame in which x-axis is
aligned vertically upwards, see Figure 1. The left wall is at y = 0 and has thickness
b, b > 0. The right wall is at y = L, L > b.
The field variables are the temperature in solid domain Ts , the temperature in
nanofluid domain Tf , the velocity v = (u, v) and the nanoparticle volume fraction C.
As in [15] it is assumed that the nanoparticles’ flux

qc = DB ∇C + (DT /Tf ) ∇Tf (2.1)

is zero on the solid vertical walls. In (2.1) DB and DT are the Brownian and ther-
mophoretic diffusion coefficients.
To define the effective viscosity we use a model proposed in [5], namely
µf
µnf = , (2.2)
(1 − C0 )2.5
 Conjugate free convection in a vertical channel 613

where µf represents the dynamic viscosity of base fluid and C0 is the reference
nanoparticles volume fraction concentration. Moreover, the effective thermal conduc-
tivity is approximated by a model introduced in [19], namely
kp + 2kf − 2C0 (kf − kp )
knf = kf , (2.3)
kp + 2kf + C0 (kf − kp )
where kf and kp are the thermal conductivity of the base fluid and thermal conduc-
tivity of the nanoparticles, respectively. The behavior of µnf and knf as functions of
C0 was studied in [6] for a side heated square cavity. There, the nanofluid is composed
of water as base fluid and Alumina as nanoparticles dispersed inside the base fluid.

x
q0’’’

T H +T C
T0 =
2
TH TC
g

y=0 y=b y=L

Figure 1. Geometry of the problem and the co-ordinate system

The limit case of knf is

lim1 + 2C0
knf = kf (2.4)
1 − C0
and it is obtained when the ratio kf /kp is very small. Physically this is possible when
high thermal conductivity nanopraticles such as cooper, gold, carbon nanotubes are
used. Thus, it is possible to predict the maximum achievable temperature in this
mathematical model for different kind of fluids and concentrations of nanoparticles.
Some examples concerning nanofluids’ thermo-physical properties are shown in the
following table.

Physical properties Fluid phase (water) Cu Al2 O3 T iO3

Cp (J/kgK) 4179 385 765 686.2
ρ (kg/m3 ) 997.1 8933 3970 4250
k (W/mK) 0.613 400 40 8.9538
β × 10−5 (1/K) 21 1.67 0.85 0.9
614 Flavius Pătrulescu and Teodor Groşan

3. Basic equations
In this section we provide the governing equations for the flow and heat transfer.
Thus, as in [16], we consider the following equation for temperature in solid domain
q0′′′
αs ∇2 Ts + = 0, (3.1)
(ρc)s
where q0′′′ represents the heat generation, αs is the thermal diffusivity coefficient and
(ρc)s represents the heat capacity. As in [10], in the fluid domain we consider the
following four field equations in the vectorial form. More exactly, the equations em-
body the conservation of total mass, momentum, thermal energy and nanoparticles’
concentration. Thus, we have
∇ · v = 0, (3.2)
ρf (v · ∇v) = µnf ∇2 v + {ρp C + (1 − C)[ρf (1 − β(Tf − T0 ))]}g, (3.3)
(ρc)f (v · ∇Tf ) = knf ∇2 Tf + (ρc)p [DB ∇Tf · ∇C + (DT /Tf )∇Tf · ∇Tf ], (3.4)
v · ∇C = ∇(DB ∇C + (DT /Tf )∇Tf ). (3.5)
Here ρf is the fluid density, ρp is the nanoparticle mass density, β represents the ther-
mal expansion coefficient, g is the gravitational acceleration. Finally, (ρc)f and (ρc)p
are the heat capacity of the base fluid and of the nanoparticle material, respectively.
In the rest of the paper we use the following linearized version of the momentum
equation (see [10])
ρf (v · ∇v) = µnf ∇2 v + [(ρp − ρf0 )(C − C0 ) − (1 − C0 )ρf0 β(Tf − T0 )]g, (3.6)
where ρf0 represents the reference fluid density.
Based on the fact that the flow is fully developed we introduce the following
assumptions
∂Tf ∂Ts ∂C
v = 0, = 0, = 0, = 0. (3.7)
∂x ∂x ∂x
Taking into account (2.1) and (3.7) the governing equations for the flow and heat
transfer (3.1)–(3.6) become
d2 Ts q0′′′
αs + = 0, (3.8)
dy 2 (ρc)s
d2 Tf
= 0, (3.9)
dy 2
dC DT dTf
DB + = 0, (3.10)
dy Tf dy
d2 u
µnf 2 + (1 − C0 )ρf0 β(Tf − T0 )g − (ρp − ρf0 )(C − C0 )g = 0, (3.11)
dy
subject to the boundary conditions
Ts |y=0 = TH , Tf |y=L = TC , (3.12)
Tf |y=b = Ts |y=b , (3.13)
dTs dTf
ks |y=b = knf |y=b , (3.14)
dy dy
 Conjugate free convection in a vertical channel 615

u(b) = u(L) = 0. (3.15)

In (3.14) ks denotes the thermal conductivity of solid domain. To complete the set of
equations and boundary conditions we add the following nanoparticles’ conservation
relation Z L
C(y) dy = Q0 , (3.16)
b
where Q0 is defined in the next section.

4. Dimensionless equations
In this section we solve equations (3.8)–(3.11) subject to (3.12)– (3.16). To this
end, we consider the following dimensionless variables used in [16].
y ks (Ts − T0 ) ks (Tf − T0 ) C − C0 u
Y = , Θs = , Θf = ,φ= ,U= , (4.1)
L q0′′′ L2 q0′′′ L2 C0 Uc
where Uc is the characteristic velocity given by
q0′′′ L2 2
gβ( ks )L
Uc = . (4.2)
νf
In (4.2) νf represents the kinematic viscosity of base fluid.
We substitute dimensionless variables (4.1) into equations (3.8)–(3.11) and we obtain
the following ordinary differential equations
d2 Θs
+ 1 = 0, (4.3)
dY 2
d2 Θf
= 0, (4.4)
dY 2
dφ R dΘf
+ = 0, (4.5)
dY wΘf + 1 dY
d2 U
= Nr (1 − C0 )2.5 φ − (1 − C0 )3.5 Θf , (4.6)
dY 2
where w is a dimensionless constant given by
q0′′′ L2 2q0′′′ L2
w= = (4.7)
ks T0 ks (TH + TC )
and, as in [10], N r is the buoyancy ratio parameter defined by
g(ρp − ρf0 )C0 L2
Nr = . (4.8)
µf U c
Moreover, R is given by
Nt
R=, (4.9)
Nb
and it represents the ration between the thermophoresis parameter and the Brownian
motion parameter (see [10]).
616 Flavius Pătrulescu and Teodor Groşan

The boundary conditions (3.12)–(3.15) become

Θs |Y =0 = q, (4.10)
Θs |Y =r = Θf |Y =r , (4.11)
dΘs dΘf
|Y =r = K |Y =r , (4.12)
dY dY
Θf |Y =1 = −q, (4.13)
U (r) = U (1) = 0, (4.14)
where r, q, K are given by
b ks (TH − TC ) knf
r= , q= , K= . (4.15)
L 2q0′′′ L2 ks
Finally, we choose Q0 = C0 L(1 − r) such that (3.16) takes the form
Z 1
φ(Y )dY = 0. (4.16)
r

5. Results and discussions

In this section we determine the exact solutions of equations (4.3)–(4.6) and
discuss the results. Integrating equations (4.3), (4.4) and taking into account boundary
conditions (4.10)–(4.13) we obtain
Y2
Θs (Y ) = − + A1 Y + A2 , (5.1)
2
and
Θf (Y ) = A3 Y + A4 . (5.2)
The constants A1 , A2 , A3 , A4 are given by
1 2
2r + 2q
A1 = r + K , A2 = q, (5.3)
r(1 − K) − 1
and
1 2
2r
+ 2q
A3 = , A4 = −q − A3 . (5.4)
r(1 − K) − 1
Moreover, integrating (4.5) and taking into account (5.2), (5.4) and (4.16) we obtain
R Ca 
φ(Y ) = − ln(wA3 Y + wA4 + 1) − . (5.5)
w 1−r
The constant Ca is given by
Ca = (1 + z) ln(wA3 + wA4 + 1) − (r + z) ln(wA3 r + wA4 + 1) − (1 − r),
where
1 + wA4
z= .
wA3
Finally, integrating (4.6) and taking into account (5.2), (5.4), (5.5) and boundary
conditions (4.14) we obtain
U (Y ) = G(Y ) + C1 Y + C2 . (5.6)
 Conjugate free convection in a vertical channel 617

The constants C1 , C2 are given by

G(r) − G(1) rG(1) − G(r)
C1 = , C2 = , (5.7)
1−r 1−r
and function G is defined in the following way
h Y 2 z2 
G(Y ) = (1 − C0 )2.5 − Cb + zY + ln(wA3 Y + wA4 + 1)
2 2
Y 3 3 Ca Cb Y 2 Yi
−(1 − C0 )A3 + Cb + − (1 − C0 )A4 + zCb ,
6 2 1−r 2 2
where
RNr
Cb = .
w
Next we determine the Nusselt number. For the conjugate wall is defined as:
hL
Nu = |y=b , (5.8)
kf
where the convective heat transfer coefficient, h, is obtained from relation (see [16]):
dTf
−knf |y=b = h(Tf |y=b − T0 ). (5.9)
dy
Substituting (5.9) in (5.8) the dimensionless form of Nusselt number becomes:
knf 1 dΘf
Nu = − |Y =r . (5.10)
kf Θf |Y =r dY
Tacking into account (5.2) we deduce that
A3
N u = −K . (5.11)
A3 r + A4
Finally, we present the effects of the governing parameters on the velocity, temperature
and nanoparticles’ concentration. In this study we consider the following fixed values
r = 0.3, TH = 300, TC = 15, ks = 1.2, kf = 0.613, C0 = 0.08, q = 1. Figure 2 presents
the variation of the solid temperature profiles for different kind of nanoparticles. It is
obvious that the addition of the nanoparticles leads to a decrease of the temperature in
solid, specially on the solid-fluid interface and the results for copper nanoparticles are
the most close to the limit case. The influence of the parameter R on the nanoparticles’
concentration profiles in the limit case is depicted in Figure 3. The concentration
profile is almost flat for small values of the parameter R, while for large values large
differences between the concentration values on the left and on the right walls appear.
The variation of the velocity profiles with R and N r is given in Figures 4 and 5,
respectively. In both cases there is a reversed flow near the right wall. The maximum
velocity increases with the increase of R and N r, respectively.
Acknowledgement. The work of the first author has been partially supported
by project POSDRU/159/1.5/S/132400: Young successful researchers-professional
development in an international and interdisciplinary environment at Babeş-Bolyai
University.
618 Flavius Pătrulescu and Teodor Groşan

0.95

0.9

0.85
Θs (Y )

0.8

0.75

0.7

Cu
0.65 Al2 O3
Ti O2
limit case
base fluid
0.6
0 0.05 0.1 0.15 0.2 0.25 0.3
Y

Figure 2. Solid temperature

2
φ(Y )

1
R=0.5,0.75,1,1.5,2.5

0
R=0.5,0.75,1,1.5,2.5

−1

−2

−3
0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Y

Figure 3. Nanoparticles’ concentration

Conjugate free convection in a vertical channel 619

2.5

1.5

0.5
U (Y )

R=0.5,0.75,1,1.5,2.5

R=0.5,0.75,1,1.5,2.5
−0.5

−1

−1.5

−2

−2.5
0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Y

Figure 4. Velocity profile, Nr = 100

2
U (Y )

1
N =50,100,250,500
r

−1 N =50,100,250,500
r

−2

−3

−4
0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Y

Figure 5. Velocity profile, R = 1

620 Flavius Pătrulescu and Teodor Groşan

References
[1] Aung, W., Worku, G., Theory of fully developed, combined convection including flow
reversal, J. Heat Trans.-T. ASME, 108(1986), 485–488.
[2] Barletta, A., Analysis of combined forced and free flow in a vertical channel with vis-
cous dissipation and isothermal-isoflux boundary conditions, ASME J. Heat Transfer,
121(1999), 349–356.
[3] Bejan, A., Convection Heat Transfer (2nd ed.), Wiley, New York, 1995.
[4] Buongiorno, J., Convective transport in nanofluids, J. Heat Trans.-T. ASME, 128(2005),
no. 3, 240–250.
[5] Brinkman, H.C., The viscosity of concentrated suspensions and solutions, J. Chem.
Phys., 20 (1952), 571–581.
[6] Celli, M., Non-homogeneous model for a side heated square cavity filled with a nanofluid,
Int. J. Heat Fluid Flow, 44(2013), 327–335.
[7] Choi, S.U.S., Enhancing thermal conductivity of fluids with nanoparticles, ASME FED,
231(1995), 99–103.
[8] Daungthong, W., Wongwises, S., A critical review of convective heat transfer of nanoflu-
ids, Renew. Sust. Energ. Rev., 11(2007), 797–817.
[9] Groşan, T., Thermal dispersion effect on fully developed free convection of nanofluids in
a vertical channel, Sains Malays., 40(2011), 1429–1435.
[10] Groşan, T., Pop, I., Fully developed mixed convection in a vertical channel filled by a
nanofluid, J. Heat Trans.-T. ASME, 134(2012), no. 8, 082501-082501-5.
[11] Ingham, D.B., Pop, I., Transport in Porous Media, vol. 1, 2 and 3, Elsevier, 1998, 2002,
2005.
[12] Khanafer, K., Vafai, K., Lightstone, M., Buoyancy-driven heat transfer enhancement in
a two-dimensional enclosure utilizing nanofluids, Int. J. Heat Mass Transfer, 46(2003),
3639–3653.
[13] Kleinstrauer, C., Microfluidics and Nanofluidics. Theory and Selected Applications, Wi-
ley, New Jersey, 2013.
[14] Koo, J., Kleinstreuer, C., Laminar nanofluid flow in microheat-sinks, Int. J. Heat Mass
Transfer, 48(2005), 2652–2661.
[15] Li, W., Exact solutions for free convective heat transfer in nanofluids in a vertical chan-
nel, Open J. Heat Mass Momentum Transfer (HMMT), 1(2013), 19–34.
[16] Pătrulescu, F., Groşan, T., Conjugate heat transfer in a vertical channel filled with a
nanofluid adjacent to a heat generating solid domain, Rev. Anal. Numér. Théor. Approx.,
36(2010), no. 1, 141–149.
[17] Schafer, H.-E., Nanoscience. The Science of the Small in Physics, Engineering, Chem-
istry, Biology and Medicine, Springer, Heidelberg, 2010.
[18] Tiwari, R.K., Das, M.K., Heat transfer augmentation in a two sided lid-driven differ-
entially heated square cavity utilizing nanofluids, Int. J. Heat Mass Transfer, 50(2007),
2002–2018.
[19] Wang, X.Q., Mujumdar, A.S., A review on nanofluids-part I: Theoretical and numerical
investigations, Braz. J. Chem. Eng., 25(2008), 613–630.
 Conjugate free convection in a vertical channel 621

Flavius Pătrulescu
Babeş-Bolyai University
Faculty of Mathematics and Computer Sciences
1, Kogălniceanu Street, 400084 Cluj-Napoca, Romania
Tiberiu Popoviciu Institute of Numerical Analysis, Romanian Academy
P.O. Box 68-1, 400110 Cluj-Napoca, Romania
e-mail: [email protected]
Teodor Groşan
Babeş-Bolyai University
Faculty of Mathematics and Computer Sciences
1, Kogălniceanu Street, 400084 Cluj-Napoca, Romania
e-mail: [email protected]
Stud. Univ. Babeș-Bolyai Math. 60(2015), No. 4, 623–628

Book reviews

Thomas Bedürftig und Roman Murawski, Philosophie der Mathematik, 3rd

expanded and revised edition, xvii+465 pp, Walter de Gruyter, Berlin, 2015, ISBN:
978-3-11-033117-2/pbk; 978-3-11-033118-9/ebook.

Any person, well acquainted with the history of mathematics, could grasp the
fact that the development of mathematics is not a linear one. Indisputably, the most
notable conceptual changing happened in the 19 century, a moment in which the
analysis of the real numbers, the set theory and the axiomatic approach impose a
new paradigm, in which much of the old problems get a mathematical solution and in
which new topics do generate new problems. This is the conceptual ground on which,
in a very inspired and extremely flexible mode, the mathematics and the philosophy
encounter each other in an excellent book on the philosophy of mathematics.
A “philosophy of mathematics” in a time in which the mathematics is separated
from philosophy? What is the good of it? The answer to this question is also the
leading idea of the whole analysis carried out in the chapters of the book and that
can be outlined in the following terms: the mathematical solutions to some problems
do not necessarily dissolve the philosophical solutions; “we come to decisions but the
philosophical questions are not answered. They remained and will remain” (372). This
is the key subject organizing in a perfectly coherent totality the six chapters of the
book.
Although the starting point, Chapter 1, On the road to the real numbers, is a
usual one in the present mathematical practice, that of the real number, the analysis
reveals a whole range of fundamental philosophical problems connected to this notion:
irrationality of a number, incommensurability of some quantities, the geometrical
continuum and the real line, the infinite. Surely, since in the new paradigm the set
R of the real numbers gives the solutions, these are the problems mathematically
decided, but philosophically some of them are still important problems. According
to Cantor and Dedekind, for example, the linear continuum is R, but can the set
R be the linear continuum? Mathematically, the small infinity isn’t a problem any
more, but if R doesn’t express the intuitive, geometrical continuum, then does not the
intuitive meaning be captured, at least partially, by resorting to the classical notion of
infinitesimals? The actual infinity, stipulated by the axiom of infinity in the set theory,
claims to work with infinite sets as given mathematical objects, but can be understood
an infinite series like an usual given object? Moreover, why such abstract, theoretical
624 Book reviews

constructions, as actual infinity and eo ipso real numbers, find their applications in
the solving of a number of concrete problems?
All these matters are the subject of a historical analysis in Chapter 2, From
the history of philosophy and mathematics, from the antiquity to the recent realism-
antirealism controversy in the philosophy of mathematics. This analysis paves the way
to some possible answers to the above questions.
The effectiveness of the concept “philosophy of mathematics”, proposed by the
authors, is masterly illustrated in the next chapters. No doubt, the standard is given
by the analysis of the real number, based on the idea of continuum and on the re-
lated notion of infinity, the topic of Chapter 3, About fundamental questions of the
philosophy of mathematics. Sketched, the idea is this. The actual understanding of
the continuum (Cantor/Dedekind) is a form of the atomism, the lines and their parts
being uncountable sets of points, different from the infinitesimal understanding (Leib-
niz), in which the lines are made up of an infinite number of infinitely small segments.
Is this identification Continuum = R the answer to the question about the very nature
of the continuum? Mathematically, yes! Philosophically, no, if by “philosophical” we
understand the capture of the intuitive meaning of this notion, in which the continuum
(e.g. the line) has parts and the parts obtained by division are again the continua.
Since the points have no parts, they represent the idea of discontinuity and can form
only something discontinuous. Therefore, R is not the intuitive continuum, a notion
that, given the above identification, did not survive mathematically, but that is still
living philosophically. Can this intuitive meaning be recovered also in a formal math-
ematical way? On the authors view, the solution is given by the vindication of the
infinitesimals, in a non-standard analysis (D.Laugwitz, C.Schmieden, A.Robinson),
in which the notion of real number is extended to a non-Archimedean field of the
“hyperreal numbers” with infinitely large and infinitely small numbers. In this field
besides the real “standard” numbers, at an infinite distance there are “nonstandard”
numbers; and this new understanding occurs by reintroducing Leibniz’s infinitesimal
quantities.
Undoubtedly, one key problem of the philosophy of mathematics is that con-
cerning the nature of mathematical objects. Since the mathematical concepts can be
defined in set-theoretic terms, they are reducible to the notion of set. And, in this
way, the question referring to the nature of mathematical objects (e.g. real number)
reduces to the question of what kind of entities the sets are. Chapter 4, The sets and
the theories of sets, integrates therefore in the structure of the book, in a perfectly
coherent way, the ontological problems of what actually the sets are (hard to say!)
in the terms of the way they are defined, i.e. in the terms of the axioms describ-
ing the relations between them. Two axiomatic theories of sets (Zermelo-Frenkel and
Neuman-Bernays-Gödel), in an elegant comparative analysis, characterize ontological
this central notion of mathematical thinking that, by integration of the notion of set,
becomes set-theoretic thinking.
Certainly, the new methodological paradigm of the present mathematics is rep-
resented by the set theory and logic, axiomatically organized. Then, in a natural way,
 Book reviews 625

the analysis continues with a distinct Chapter 5, The axiomatic and the logic, fo-
cused on the new mathematical and philosophical problems, as the problem of the
consistency of mathematics, that of completeness, the problem of the nature of math-
ematical truth and the re-thinking of the continuum. Directly or indirectly, all these
problems are originated in the axiomatic setting of the present mathematical theories.
They are those which reveals the “distance” between the provability and the truth
and, at last, the limits of this kind of mathematical thinking.
A Retrospection, in Chapter 6, about the topics developed in the book and some
technical considerations on To think and to calculate infinitesimally, concludes this
excellent book on the philosophy of mathematics.
The historical, mathematical and philosophical perspectives from which the key
notions are investigated are very different. But the authors avoid to give “definitive”
solutions to these problems. Often we find the expressions of the following form:
”We don’t know what is the point” (203), ”R is not the continuum, R is a model”
(264), “What exactly are the sets, nobody knows” (310), “We do not find an effective
answer to the question about numbers” (377), “For the set theory, for the theory of
infinite, the infinite remains enigmatical” (314), “Even mathematically, the continuum
remains transcendent” (227). Therefore, the fundamental pieces of the philosophical
and mathematical investigation “remain transcendent” (386). In a perfect consonance
with the project of the construction of the book, the source of the transcendence
is to be viewed in the essential difference between a philosophical notion and its
mathematical description, and from here the necessity of the philosophical analysis of
the fundamental mathematical meanings.
The text of the book has an exemplary clearness, all the key notions are indicated
in the distinct paragraphs and the bibliographical list of the papers to which the
authors are referred is impressively vast and relevant.
The conceptual strictness, the quality of argumentation and the originality of
the analysis make of this book one of the profoundest and more elegant conceptual
construction in the philosophy of mathematics.
Virgil Drăghici

Gerard Walschap, Multivariable Calculs and Differential Geometry, Walter de

Gruyter (De Gruyter Graduate), 2015, ix+355 pages, Paperback, ISBN 978–3-11-
036949-6.

The book under review is a rigorous introduction to differential geometry (mainly

of hypersurfaces in a Euclidean space) based on a solid foundation of calculus and
linear algebra.
I give below a short description of the contents. The first chapter is a short intro-
duction to the topology of a Euclidean space. The second chapter (called, generically,
Differentiation) is concerned with a study of the mappings between two Euclidean
spaces, including, among the classical stuff (Taylor series, implicit functions), a thor-
ough introduction to vector field and Lie brackets, as well as the partition of unity
626 Book reviews

on an open set of a Euclidean space. What these two chapter have in common is the
consistent use of the tools of linear algebra (for instance, à la manière de Calculus on
Manifolds, Addison Wesley, 1965, by Michael Spivak). This language is not, usually,
the favorite of the analysts, although we can hardly call it “modern” (the book of
Spivak has been published fifty years ago). The third chapter is the first one with a
geometric flavor. Although the name of the chapter is, simply, Manifolds, in reality, it
is a short introduction to the geometry of the manifolds (well, actually submanifolds
of a Euclidean space). It includes, beside the standard notions of differential topol-
ogy (manifolds, Lie groups, maps, vector fields, the tangent bundle), also differential
geometric concepts as: covariant derivative, geodesics, the second fundamental form,
curvatures, isometries (with respect to the Euclidean scalar product on the ambient
Euclidean spaces). The following chapter is devoted to the theory of integration on
a Euclidean space. It includes the classical subjects (the definition of the integral
and of the integrability, the Fubini theorem, the classical integral theorems), but also
some applications to physics and the Sard theorem for Euclidean spaces, a quite un-
expected (but not inappropriate) presence here. The chapter five, Differential forms,
has, actually, as subject the theory of integration on manifolds (tensors and forms,
differential forms on manifolds, integration on manifolds and manifolds with bound-
ary, Stokes’ theorem). The author takes time to discuss the connection between the
Stokes’ theorem and the classical theorems of integral calculus. The chapter ends with
a short look at the de Rham cohomology. For the last two chapters, the author returns
to geometry. Thus, chapter six treats standard subjects from Riemannian geometry
(extremal properties of geodesics, Jacobi fields, the variation of the length functional,
Hopf-Rinow theorem, comparison theorems a.o.), while the final chapter focuses on
the geometry of hypersurfaces in a Euclidean space, with the induced Riemannian
metric (orientation, Gauss map, curvature a.o.). In this context, the general concepts
from manifold theory become more intuitive. In particular, there are also discussed
some classical examples of surfaces (ruled surface, surfaces of revolution).
Multivariable calculus and geometry always meet (well, at least in textbooks).
The question is where do they meet. I could mention several such intersection points:
• analytical geometry that goes together with calculus (this is common, especially,
in the American universities);
• the calculus as a prerequisite to a textbook of manifolds;
• the other way around: (sub)manifolds used as a foundation for multiple integra-
tion (for instance).
What I haven’t seen elsewhere (or, if I have, I don’t remember) is a marriage between
calculus and differential geometry, where the two subjects to be, more or less, on equal
footing. From this point of view, the author did an excellent job. The book can be
used as a textbook for a course in differential geometry for advanced undergraduate
or beginning graduate student in mathematics or physics, or for self-study.
The book includes a large number of worked examples and exercises, as well as
a number of excellent drawings that improve the presentation.
Paul A. Blaga
 Book reviews 627

The Princeton Companion to Applied Mathematics, Edited by Nicholas J. Higham,

Associate Editors: Mark R. Dennis, Paul Glendinning, Paul A. Martin, Fadil Santosa
and Jared Tanner, Princeton University Press, Princeton, NJ, 2015, xvii + 994 pp,
ISBN: 978-0-691-15039-0/hbk; 978-1-4008-7447-7/ebook.

As the editor writes in the Preface “The Princeton Companion to Applied Math-
ematics describes what applied mathematics is about, why it is important, its con-
nections with other disciplines, and some of the main areas of current research.” We
also reproduce here the nice words of Paul Halmos quoted in Editor’s (NJH) article
What is applied mathematics?
Pure mathematics can be practically useful and applied mathematics
can be artistically elegant... Just as pure mathematics can be useful,
applied mathematics can be more beautifully useless than is some-
times recognized...
On the other side, it is not easy to give a precise definition of applied mathemat-
ics and, in some cases, it is difficult to say whether a specific domain (or topic) belongs
to pure or to applied mathematics. Also, over time, a domain of pure mathematics
finds its applications, as, e.g., the applications of number theory to cryptography. Pure
mathematics was presented in another Princeton volume The Princeton companion
to mathematics (Editors: Timothy Gowers, June Barrow-Green, Imre Leader; Prince-
ton University Press, 2008), which contains some topics (Mathematics and Chemistry,
Mathematical Biology, Mathematical Statistics, Optimization and Lagrange Multipli-
ers) which can be considered to belong to applied mathematics and could be included
in the present volume as well. In order to avoid overlapping the topics presented in
the previous Companion are excluded from the present one, and in the case of some
crucial concepts (e.g. algebraic geometry, fast Fourier transform), the approach here
is different, with emphasis on applications and computational aspects. In some cases,
particular aspects of topics treated in the previous companion are included here.
The book is divided into eight parts.
Part I, Introduction to Applied Mathematics, contains some general results about
applied mathematics as: what is it, the language, algorithms, goals and history.
Part II, Concepts, contains short articles explaining specific concepts as convex-
ity, chaos, floating-point arithmetic, Markov chains, etc. This part is not a compre-
hensive, other concepts being presented in other articles.
Part III, Equations, Laws, and Functions of Applied Mathematics, contains short
presentations of some functions and equations encountered in applied mathematics, as,
e.g., Bessel functions, Mathieu functions, Euler functions, Black-Scholes law, Hooke’s
law, the equations of Cauchy-Riemann, Laplace, Korteweg-de Vries, Dirac, etc.
Part IV, Areas of Applied Mathematics, contains longer articles giving an
overview of some domains of applied mathematics as complex analysis, ordinary and
partial differential equations, data mining, random matrices, control theory, informa-
tion theory, etc.
628 Book reviews

Part V, Modeling, presents some mathematical models in chemistry, biology,

financial mathematics, wheather prediction, etc.
Part VI, Example Problems, contains short articles on various interesting prob-
lems as bubbles, foams, the inverted pendulum, robotics, random number generation,
etc.
Part VII, Application Areas contains articles on connections of mathematics with
other domains as aircraft noise, social networks, chip design, digital imaging, medical
imaging, radar imaging, etc.
The final part, VIII, Final Perspectives, contains some longer articles of gen-
eral interest on mathematical writing, the reading of mathematical papers, teaching
applied mathematics, mathematics in the media, mathematics and policy (how math-
ematicians can influence political decisions).
In the last years, partly due to high performance computers, the area of applied
mathematics enlarged considerably, so that the present volume is a welcome addi-
tion to the existing publications and a guide for the researchers interested to apply
mathematics in their domains (an unavoidable option, as it can be seen).
Together with its elder brother, The Princeton companion to mathematics, which
turned to be a very successful enterprise, the present volume covers a lot of topics
in applied mathematics, and surely it will also have great success and impact. It is
dedicated to students (starting with the undergraduate level), teachers, researchers in
various areas and specialists in various domains desiring to know what mathematics
can offer and, as it is proved in this volume, it has a lot to offer.
S. Cobzaş