Computing methods for multiloop
Feynman integrals
Vladimir A. Smirnov
Nuclear Physics Institute of Moscow State University
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.1
Methods to evaluate Feynman integrals:
analytical, numerical, semianalytical
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.2
Methods to evaluate Feynman integrals:
analytical, numerical, semianalytical
applicable not only to Feynman integrals but also, e.g., to
Wilson loops
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.2
Methods to evaluate Feynman integrals:
analytical, numerical, semianalytical
applicable not only to Feynman integrals but also, e.g., to
Wilson loops
Evaluating Feynman integrals (STMP 211, Springer 2004)
Feynman Integrals Calculus (Springer 2006)
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.2
Methods to evaluate Feynman integrals:
analytical, numerical, semianalytical
applicable not only to Feynman integrals but also, e.g., to
Wilson loops
Evaluating Feynman integrals (STMP 211, Springer 2004)
Feynman Integrals Calculus (Springer 2006)
replaced by
Analytic Tools for Feynman Integrals (STMP, Springer 2013)
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.2
Methods to evaluate Feynman integrals:
analytical, numerical, semianalytical
applicable not only to Feynman integrals but also, e.g., to
Wilson loops
Evaluating Feynman integrals (STMP 211, Springer 2004)
Feynman Integrals Calculus (Springer 2006)
replaced by
Analytic Tools for Feynman Integrals (STMP, Springer 2013)
lists of misprints
http://theory.sinp.msu.ru/∼smirnov
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.2
Feynman integrals: basic notation, definitions and
properties. Dimensional regularization.
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.3
Feynman integrals: basic notation, definitions and
properties. Dimensional regularization.
Alpha and Feynman parameters. Sector
decompositions.
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.3
Feynman integrals: basic notation, definitions and
properties. Dimensional regularization.
Alpha and Feynman parameters. Sector
decompositions.
The method of Mellin–Barnes representation.
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.3
Feynman integrals: basic notation, definitions and
properties. Dimensional regularization.
Alpha and Feynman parameters. Sector
decompositions.
The method of Mellin–Barnes representation.
Integration by parts. Reduction to master integrals.
Laporta’s algorithm.
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.3
Feynman integrals: basic notation, definitions and
properties. Dimensional regularization.
Alpha and Feynman parameters. Sector
decompositions.
The method of Mellin–Barnes representation.
Integration by parts. Reduction to master integrals.
Laporta’s algorithm.
The method of differential equations.
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.3
Feynman integrals: basic notation, definitions and
properties. Dimensional regularization.
Alpha and Feynman parameters. Sector
decompositions.
The method of Mellin–Barnes representation.
Integration by parts. Reduction to master integrals.
Laporta’s algorithm.
The method of differential equations.
DRR
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.3
Feynman integrals: basic notation, definitions and
properties. Dimensional regularization.
Alpha and Feynman parameters. Sector
decompositions.
The method of Mellin–Barnes representation.
Integration by parts. Reduction to master integrals.
Laporta’s algorithm.
The method of differential equations.
DRR
Asymptotic expansions in various limits.
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.3
Feynman integrals: basic notation, definitions and
properties. Dimensional regularization.
Alpha and Feynman parameters. Sector
decompositions.
The method of Mellin–Barnes representation.
Integration by parts. Reduction to master integrals.
Laporta’s algorithm.
The method of differential equations.
DRR
Asymptotic expansions in various limits.
Symbols: attend Claude’s lectures!
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.3
Perturbation theory. Feynman rules. A graph Γ = {V, L, π± }
with vertices and lines (edges), where V is the set of
vertices, L is the set of lines, and π± : L → V .
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.4
Perturbation theory. Feynman rules. A graph Γ = {V, L, π± }
with vertices and lines (edges), where V is the set of
vertices, L is the set of lines, and π± : L → V .
A given Feynman graph Γ → tensor reduction → various
scalar Feynman integrals that have the same structure of
the integrand with various distributions of powers of
propagators.
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.4
Perturbation theory. Feynman rules. A graph Γ = {V, L, π± }
with vertices and lines (edges), where V is the set of
vertices, L is the set of lines, and π± : L → V .
A given Feynman graph Γ → tensor reduction → various
scalar Feynman integrals that have the same structure of
the integrand with various distributions of powers of
propagators.
dd k1 dd k2 . . .
! !
FΓ (a1 , a2 , . . .) = ...
(p21 − m21 )a1 (p22 − m22 )a2 . . .
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.4
Perturbation theory. Feynman rules. A graph Γ = {V, L, π± }
with vertices and lines (edges), where V is the set of
vertices, L is the set of lines, and π± : L → V .
A given Feynman graph Γ → tensor reduction → various
scalar Feynman integrals that have the same structure of
the integrand with various distributions of powers of
propagators.
dd k1 dd k2 . . .
! !
FΓ (a1 , a2 , . . .) = ...
(p21 − m21 )a1 (p22 − m22 )a2 . . .
Dimensional regularization: d = 4 − 2" ; d4 k → dd k
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.4
Perturbation theory. Feynman rules. A graph Γ = {V, L, π± }
with vertices and lines (edges), where V is the set of
vertices, L is the set of lines, and π± : L → V .
A given Feynman graph Γ → tensor reduction → various
scalar Feynman integrals that have the same structure of
the integrand with various distributions of powers of
propagators.
dd k1 dd k2 . . .
! !
FΓ (a1 , a2 , . . .) = ...
(p21 − m21 )a1 (p22 − m22 )a2 . . .
Dimensional regularization: d = 4 − 2" ; d4 k → dd k
k = (k0 , #k) = (k0 , k1 , k2 , k3 )
k1 , k2 , . . . are loop momenta;
p1 , p2 , . . . are momenta of the lines; they are linear
combinations of k1 , k2 , . . . and external momenta q1 , q2 , . . .
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.4
The propagator as a building block
1 1
2 2
= lim 2 2
,
k − m + i0 δ→0 k − m + iδ
k 2 = k 2 − #k 2 = k 2 − k 2 − k 2 − k 2
0 0 1 2 3
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.5
The propagator as a building block
1 1
2 2
= lim 2 2
,
k − m + i0 δ→0 k − m + iδ
k 2 = k 2 − #k 2 = k 2 − k 2 − k 2 − k 2
0 0 1 2 3
HQET, NRQCD,. . . → other types of propagators, e.g.
1
, v = (1, #0)
v · k ± i0
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.5
For example,
q12 = q22 = 0, Q2 = −(q1 − q2 )2 = 2q1 ·q2
q1
q1 − k
q1 − k − l
k
l
q2 − k − l
q2 − l
q2
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.6
For example,
q12 = q22 = 0, Q2 = −(q1 − q2 )2 = 2q1 ·q2
q1
q1 − k
q1 − k − l
k
l
q2 − k − l
q2 − l
q2
d d
d d
! !
2 k l
FΓ (Q ; a1 , . . . , a6 , a7 , d) =
[−(k + l)2 + 2q1 ·(k + l)]a1
(2k · l)−a7
×
[−(k + l)2 + 2q2 ·(k + l)]a2 (−k 2 + 2q1 ·k)a3 (−l2 + 2q2 ·l)a4 (−k 2 )a5 (−l2 )a6
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.6
UV, IR and collinear divergences
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.7
UV, IR and collinear divergences
"
UV degree of divergence ω = 4h − 2L or ω = 4h − 2 al
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.7
UV, IR and collinear divergences
"
UV degree of divergence ω = 4h − 2L or ω = 4h − 2 al
ω < 0: there are no UV divergences.
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.7
UV, IR and collinear divergences
"
UV degree of divergence ω = 4h − 2L or ω = 4h − 2 al
ω < 0: there are no UV divergences.
ω = 0: there is a logarithmical divergence.
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.7
UV, IR and collinear divergences
"
UV degree of divergence ω = 4h − 2L or ω = 4h − 2 al
ω < 0: there are no UV divergences.
ω = 0: there is a logarithmical divergence.
IR off-shell and IR on-shell divergences.
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.7
UV, IR and collinear divergences
"
UV degree of divergence ω = 4h − 2L or ω = 4h − 2 al
ω < 0: there are no UV divergences.
ω = 0: there is a logarithmical divergence.
IR off-shell and IR on-shell divergences.
dd k
!
FΓ =
k 2 (k + p1 )2 (k + p2 )2
at p21 = p22 = 0
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.7
Divergences → regularization
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.8
Divergences → regularization
Analytical regularization [E. Speer’68]
1 1
2 2 a
→
(−k + m − i0) (−k 2 + m2 − i0)a+λ
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.8
Dimensional regularization
[G. ’t Hooft & M. Veltman’72]
[C.G. Bollini & J.J. Giambiagi’72; P. Breitenlohner & D. Maison’77]
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.9
Dimensional regularization
[G. ’t Hooft & M. Veltman’72]
[C.G. Bollini & J.J. Giambiagi’72; P. Breitenlohner & D. Maison’77]
Algebraic part: gµµ = d, etc.
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.9
Dimensional regularization
[G. ’t Hooft & M. Veltman’72]
[C.G. Bollini & J.J. Giambiagi’72; P. Breitenlohner & D. Maison’77]
Algebraic part: gµµ = d, etc.
Analytic part: formally,
d4 k = dk0 d#k → dd k , d = 4 − 2"
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.9
Dimensional regularization
[G. ’t Hooft & M. Veltman’72]
[C.G. Bollini & J.J. Giambiagi’72; P. Breitenlohner & D. Maison’77]
Algebraic part: gµµ = d, etc.
Analytic part: formally,
d4 k = dk0 d#k → dd k , d = 4 − 2"
Informally, use alpha parameters
iπa ! ∞
1 e a−1 i(k 2 −m2 )α
2 2 a
= α e dα
(−k + m − i0) Γ(a) 0
iπa ! ∞
1 e a−1 i(v·k)α
a
= α e dα
(−v · k − i0) Γ(a) 0
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.9
Dimensional regularization:
when deriving alpha representations, apply this rule with
d = 4 − 2"
!
4 i(αk 2 −2q·k) 2 −2 −iq 2 /α
d ke = −iπ α e
!
d i(αk 2 −2q·k) iπ(1−d/2)/2 d/2 −d/2 −iq 2 /α
d ke =e π α e
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.10
Graph Γ → dimensionally regularized Feynman integral
eiπ(a+h(1−d/2))/2 π hd/2
FΓ (a1 . . . , aL ; d) = #
l Γ(al )
! ∞ ! ∞ !
al −1 −d/2 iV/U −i m2l αl
$
× dα1 . . . dα L αl U e ,
0 0 l
"
where a = ai
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.11
Graph Γ → dimensionally regularized Feynman integral
eiπ(a+h(1−d/2))/2 π hd/2
FΓ (a1 . . . , aL ; d) = #
l Γ(al )
! ∞ ! ∞ !
al −1 −d/2 iV/U −i m2l αl
$
× dα1 . . . dα L αl U e ,
0 0 l
"
where a = ai
For a Feynman integral with 1/(m2 − k 2 − i0)al propagators,
% $
U = αl ,
trees T l$∈T
T 2
% $ & '
V = αl q .
2−trees T l$∈T
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.11
The massless box,
p2i = 0, i = 1, 2, 3, 4, s = (p1 + p2 )2 , t = (p1 + p3 )2
p1 p3
1
2 4
3
p2 p4
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.12
The massless box,
p2i = 0, i = 1, 2, 3, 4, s = (p1 + p2 )2 , t = (p1 + p3 )2
p1 p3
1
2 4
3
p2 p4
trees
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.12
The massless box,
p2i = 0, i = 1, 2, 3, 4, s = (p1 + p2 )2 , t = (p1 + p3 )2
p1 p3
1
2 4
3
p2 p4
trees
2-trees
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.12
The massless box,
p2i = 0, i = 1, 2, 3, 4, s = (p1 + p2 )2 , t = (p1 + p3 )2
p1 p3
1
2 4
3
p2 p4
trees
2-trees
U = α1 + α2 + α3 + α4 , V = sα1 α3 + tα2 α4 .
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.12
1 3
2 4
trees
2-trees
U = (α1 + α2 + α3 + α4 )α5 + (α1 + α2 )(α3 + α4 ) ,
V = [(α1 + α2 )α3 α4 + α1 α2 (α3 + α4 ) + (α1 + α3 )(α2 + α4 )α5 ]q 2
The code UF.m to evaluate U and V
http://science.sander.su
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.13
Alpha representation →
Mathematical proofs (for Feynman
" integrals at
Euclidean external momenta, ( qi )2 < 0)
Analysis of convergence.
[K. Hepp’66; P. Breitenlohner & D. Maison’77; E. Speer’68,’77]
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.14
Alpha representation →
Mathematical proofs (for Feynman
" integrals at
Euclidean external momenta, ( qi )2 < 0)
Analysis of convergence.
[K. Hepp’66; P. Breitenlohner & D. Maison’77; E. Speer’68,’77]
Hepp sectors [K. Hepp’66]
α1 ≤ α2 ≤ . . . ≤ αL
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.14
Alpha representation →
Mathematical proofs (for Feynman
" integrals at
Euclidean external momenta, ( qi )2 < 0)
Analysis of convergence.
[K. Hepp’66; P. Breitenlohner & D. Maison’77; E. Speer’68,’77]
Hepp sectors [K. Hepp’66]
α1 ≤ α2 ≤ . . . ≤ αL
Speer’s sectors [E. Speer’77, A. Smirnov& VS’2009]
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.14
A tool to evaluate Feynman integrals analytically.
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.15
A tool to evaluate Feynman integrals analytically.
d ∞! ∞
d k eiπ(a1 +a2 )
! !
a1 −1 a2 −1
2 a 2 a
= dα 1 d α 2 α 1 α2
(−k ) (−(q − k) )
1 2 Γ(a1 )Γ(a2 ) 0 0
!
d i[α1 k 2 +α2 (k 2 +2q·k+q 2 )]
× d ke
eiπ(a1 +a2 +1−d/2)/2 π d/2 ∞! ∞
α1a1 −1 α2a2 −1 iα1 α2 q2 /(α1 +α2 )
!
= dα1 dα2 d/2
e
Γ(a1 )Γ(a2 ) 0 0 (α1 + α2 )
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.15
A tool to evaluate Feynman integrals analytically.
d ∞! ∞
d k eiπ(a1 +a2 )
! !
a1 −1 a2 −1
2 a 2 a
= dα 1 d α 2 α 1 α2
(−k ) (−(q − k) )
1 2 Γ(a1 )Γ(a2 ) 0 0
!
d i[α1 k 2 +α2 (k 2 +2q·k+q 2 )]
× d ke
eiπ(a1 +a2 +1−d/2)/2 π d/2 ∞! ∞
α1a1 −1 α2a2 −1 iα1 α2 q2 /(α1 +α2 )
!
= dα1 dα2 d/2
e
Γ(a1 )Γ(a2 ) 0 0 (α1 + α2 )
α1 = ηξ, α2 = η(1 − ξ), with the Jacobian η , integrate over η
and ξ
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.15
dd k
!
(−k 2 )a1 [−(q − k)2 ]a2
d/2 Γ(2 − " − a1 )Γ(2 − " − a2 ) Γ(a1 + a2 + " − 2)
= iπ 2 a +a +%−2
.
Γ(a1 )Γ(a2 )Γ(4 − a1 − a2 − 2") (−q ) 1 2
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.16
dd k
!
(−k 2 )a1 [−(q − k)2 ]a2
d/2 Γ(2 − " − a1 )Γ(2 − " − a2 ) Γ(a1 + a2 + " − 2)
= iπ 2 a +a +%−2
.
Γ(a1 )Γ(a2 )Γ(4 − a1 − a2 − 2") (−q ) 1 2
dd k
!
(−k 2 )a1 (2v·k + ω − i0)a2
d/2 Γ(2 − a1 − ")Γ(2a1 + a2 + 2" − 4) 2 a1 +%−2
= iπ 2a +a +2%−4
(v ) .
Γ(a1 )Γ(a2 )ω 1 2
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.16
λ2
d/2
λ1 + λ2 − d/2
= iπ G(λ1 , λ2 )×
λ1
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.17
λ2
d/2
λ1 + λ2 − d/2
= iπ G(λ1 , λ2 )×
λ1
λ2
d/2 2 λ1 −d/2
2λ1 + λ2 − d
= iπ Ḡ(λ1 , λ2 )(v ) ×
λ1
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.17
"L
α= ηαl& ,
l = 1, 2, . . . , L − 1, η = l=1 αl , integrate over η ,
&
"L−1 &
introduce αL = 1 − l=1 αl by inserting an integration over
(" )
& with δ L & by α :
αL α
l=1 l − 1 , replace α l l
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.18
"L
α= ηαl& ,
l = 1, 2, . . . , L − 1, η = l=1 αl , integrate over η ,
&
"L−1 &
introduce αL = 1 − l=1 αl by inserting an integration over
(" )
& with δ L & by α :
αL α
l=1 l − 1 , replace α l l
& d/2 'h
iπ Γ(a − hd/2)
FΓ (q1 , . . . , qn ; d; a1 . . . , aL ) = #
l Γ(al )
! ∞ ! ∞ *% L
+ #
al −1 a−(h+1)d/2
l lα U
× ... δ αl − 1 & " 2 'a−hd/2 dα1 . . . dαL
0 0 l=1 −V + U ml α l
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.18
"L
α= ηαl& ,
l = 1, 2, . . . , L − 1, η = l=1 αl , integrate over η ,
&
"L−1 &
introduce αL = 1 − l=1 αl by inserting an integration over
(" )
& with δ L & by α :
αL α
l=1 l − 1 , replace α l l
& d/2 'h
iπ Γ(a − hd/2)
FΓ (q1 , . . . , qn ; d; a1 . . . , aL ) = #
l Γ(al )
! ∞ ! ∞ *% L
+ #
al −1 a−(h+1)d/2
l lα U
× ... δ αl − 1 & " 2 'a−hd/2 dα1 . . . dαL
0 0 l=1 −V + U ml α l
Cheng–Wu
(" theorem:
)
L &" '
δ l=1 αl − 1 → δ l∈ν αl − 1 → δ (αl − 1)
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.18
"L
α= ηαl& ,
l = 1, 2, . . . , L − 1, η = l=1 αl , integrate over η ,
&
"L−1 &
introduce αL = 1 − l=1 αl by inserting an integration over
(" )
& with δ L & by α :
αL α
l=1 l − 1 , replace α l l
& d/2 'h
iπ Γ(a − hd/2)
FΓ (q1 , . . . , qn ; d; a1 . . . , aL ) = #
l Γ(al )
! ∞ ! ∞ *% L
+ #
al −1 a−(h+1)d/2
l lα U
× ... δ αl − 1 & " 2 'a−hd/2 dα1 . . . dαL
0 0 l=1 −V + U ml α l
Cheng–Wu
(" theorem:
)
L &" '
δ l=1 αl − 1 → δ l∈ν αl − 1 → δ (αl − 1)
" "L
Proof. Use η = l∈ν αl instead of η = l=1 αl
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.18
dd k dd l
! !
(−k 2 + m2 )λ1 [−(k + l)2 ]λ2 (−l2 + m2 )λ3
( )2 Γ(λ + λ + " − 2)Γ(λ + λ + " − 2)Γ(2 − " − λ )
1 2 2 3 2
= iπ d/2
Γ(λ1 )Γ(λ3 )
Γ(λ1 + λ2 + λ3 + 2" − 4)
×
Γ(λ1 + 2λ2 + λ3 + 2" − 4)Γ(2 − ")(m2 )λ1 +λ2 +λ3 +2%−4
choose δ (α1 + α3 − 1)
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.19
Feynman parameters:
1
(m21 − p21 )a1 (m22 − p22 )a2
! 1
Γ(a1 + a2 ) dξ ξ a1 −1 (1 − ξ)a2 −1
=
Γ(a1 )Γ(a2 ) 0 [(m21 − p21 )ξ + (m22 − p22 )(1 − ξ)]a1 +a2
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.20
Feynman parameters:
1
(m21 − p21 )a1 (m22 − p22 )a2
! 1
Γ(a1 + a2 ) dξ ξ a1 −1 (1 − ξ)a2 −1
=
Γ(a1 )Γ(a2 ) 0 [(m21 − p21 )ξ + (m22 − p22 )(1 − ξ)]a1 +a2
" ! 1 ! 1 "
1 Γ( al ) $
a l −1
δ ( ξl − 1)
# al = # dξ1 . . . dξL ξl !
Al Γ(al ) 0
" a
0 l ( Al ξl ) l
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.20
Suppose, we are dealing with a finite Feynman integral.
Then one can expand the integrand in parametric
representations in Taylor series in ".
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.21
Suppose, we are dealing with a finite Feynman integral.
Then one can expand the integrand in parametric
representations in Taylor series in ".
A straightforward strategy [F. Brown’08]
Perform integrations in a given alpha-parametric integral
one by one, in some order.
Presumably, results are expressed in terms of multiple
polylogarithms.
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.21
Suppose, we are dealing with a finite Feynman integral.
Then one can expand the integrand in parametric
representations in Taylor series in ".
A straightforward strategy [F. Brown’08]
Perform integrations in a given alpha-parametric integral
one by one, in some order.
Presumably, results are expressed in terms of multiple
polylogarithms.
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.21
∞ ∞
(iπ 2 )2
! ! "
δ ( αl − 1)
F (q 2 ; 1, 1, 1, 1, 1; 4) = dα1 . . . dα5 .
q2 0 0 UV
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.22
∞ ∞
(iπ 2 )2
! ! "
δ ( αl − 1)
F (q 2 ; 1, 1, 1, 1, 1; 4) = dα1 . . . dα5 .
q2 0 0 UV
Apply the Cheng–Wu theorem by choosing the delta
function δ (α5 − 1), with the integration over the rest of the
four variables from zero to infinity and integrate in
Mathematica
& 2 '2
2 iπ
F (q ; 1, 1, 1, 1, 1; 4) = 2
6ζ(3)
q
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.22
A tool to evaluate Feynman integrals numerically.
Modern sector decompositions
[T. Binoth & G. Heinrich’00; C. Bogner & S. Weinzierl’07; A.V. Smirnov & M.N. Tentyukov’08;
A.V. Smirnov, VS & M.N. Tentyukov’08; J. Carter & G. Heinrich’10, S. Borowka &
G. Heinrich’11–13]
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.23
A tool to evaluate Feynman integrals numerically.
Modern sector decompositions
[T. Binoth & G. Heinrich’00; C. Bogner & S. Weinzierl’07; A.V. Smirnov & M.N. Tentyukov’08;
A.V. Smirnov, VS & M.N. Tentyukov’08; J. Carter & G. Heinrich’10, S. Borowka &
G. Heinrich’11–13]
Public computer codes:
SecDec, sector decomposition, FIESTA
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.23
The factorization can be always achieved in Hepp sectors:
h(γl ) ,
$ -
U= tl 1+ Pf
l
where γl = {1, . . . , l} and h is the number of loops.
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.24
The factorization can be always achieved in Hepp sectors:
h(γl ) ,
$ -
U= tl 1+ Pf
l
where γl = {1, . . . , l} and h is the number of loops.
For example, U = α1 + α2 .
Two sectors: α1 ≤ α2 and α2 ≤ α1 .
In the first one, α1 = t1 t2 , α2 = t2 so that U = t2 (t1 + 1)
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.24
The factorization can be always achieved in Hepp sectors:
h(γl ) ,
$ -
U= tl 1+ Pf
l
where γl = {1, . . . , l} and h is the number of loops.
For example, U = α1 + α2 .
Two sectors: α1 ≤ α2 and α2 ≤ α1 .
In the first one, α1 = t1 t2 , α2 = t2 so that U = t2 (t1 + 1)
Hepp and Speer sectors are applicable only at external
Euclidean momenta, i.e. when ( i∈ν pi )2 < 0 for any
"
nonempty subset ν . The second function of alpha
parameters is not generally proper factorized.
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.24
Example: the massless on-shell box in the sector
α2 ≤ α1 ≤ α3 ≤ α4 = 1, with
U = 1 + α1 + α2 + α3 , V = sα1 α3 + tα2 .
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.25
Example: the massless on-shell box in the sector
α2 ≤ α1 ≤ α3 ≤ α4 = 1, with
U = 1 + α1 + α2 + α3 , V = sα1 α3 + tα2 .
In the sector variables α2 = t1 t2 t3 , α1 = t2 t3 , α3 = t3 ,
we have
V = t1 t3 (st3 + t t1 )
so that a further sector decomposition is desirable.
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.25
Recursively sector decompositions
[T. Binoth & G. Heinrich’00]
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.26
Recursively sector decompositions
[T. Binoth & G. Heinrich’00]
The goal is" to obtain a factorization of U and
W = −V + m2l αl in final sector variables, i.e. to represent
them as products of sector variables in some powers times
a positive function.
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.26
Recursively sector decompositions
[T. Binoth & G. Heinrich’00]
The goal is" to obtain a factorization of U and
W = −V + m2l αl in final sector variables, i.e. to represent
them as products of sector variables in some powers times
a positive function.
Primary sectors ∆l
αi ≤ αl , l &= i = 1, 2, . . . , L ,
with new variables
.
αi /αl if i &= l
αi& =
αl if i = l
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.26
The contribution of a primary sector for ∆L :
& d/2
'h 1 1 L−1
iπ Γ(a − hd/2)
! ! $
F (L) = (−1)L # ... αlal −1
l Γ(al ) 0 0 l
a−(h+1)d/2
× ÛΓ W hd/2−a dα1 . . . dαL−1 ,
where
*L−1 L−1
+
% $
WΓ = −V̂Γ + ÛΓ m2l αl! + m2L ,
l=1 l=l!
ÛΓ = U (α1 , . . . , αL−1 , 1) ,
V̂Γ = VΓ (α1 , . . . , αL−1 , 1)
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.27
Let us choose a subset I = {i1 , . . . ik } of {1, . . . , n}, with
n ≡ L − 1.
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.28
Let us choose a subset I = {i1 , . . . ik } of {1, . . . , n}, with
n ≡ L − 1.
The unit hypercube {(t1 , . . . , tn )|0 ≤ ti ≤ 1∀i ∈ (1, . . . , n)} is
then decomposed into k sectors
Sl = {(t1 , . . . , tn )|ti ≤ til ∀i ∈ I} ,
for l = 1, . . . , k , and the new (sector) variables are
introduced as follows:
ti = t&i ∀i &∈ I
til = t&il
tir = t&il t&ir ∀ir ∈ I, r &= l
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.28
The integration region in the new variables t&i is again a unit
hypercube.
Then for each of the k resulting sectors subsets of the
indices are chosen and new sectors are introduced in a
similar way.
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.29
The integration region in the new variables t&i is again a unit
hypercube.
Then for each of the k resulting sectors subsets of the
indices are chosen and new sectors are introduced in a
similar way.
The rules according to which these subsets are chosen
form a sector decomposition strategy.
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.29
The integration region in the new variables t&i is again a unit
hypercube.
Then for each of the k resulting sectors subsets of the
indices are chosen and new sectors are introduced in a
similar way.
The rules according to which these subsets are chosen
form a sector decomposition strategy.
This process is terminated when the contribution of each of
the final sectors takes a proper factorized form
! 1 ! 1 n
$
... f (t1 , . . . , tn ; ") tai i +bi % dti
0 0 i=1
with a function f which is regular near the origin.
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.29
To make the poles in " manifest, the integrations over the
final sector variables ti are analyzed one by one.
! 1
G(") = ta+b% g(t, ")dt
0
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.30
To make the poles in " manifest, the integrations over the
final sector variables ti are analyzed one by one.
! 1
G(") = ta+b% g(t, ")dt
0
a ≥ 0 → the integrand can be expanded in ".
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.30
To make the poles in " manifest, the integrations over the
final sector variables ti are analyzed one by one.
! 1
G(") = ta+b% g(t, ")dt
0
a ≥ 0 → the integrand can be expanded in ".
If a < 0, subtract first terms of the Taylor series of g(t) in t at
the origin up to order −1 − a and obtain, after explicitly
integrating, the subtracted terms,
−1−a −1−a
/ 0
1
g (k) (0, ") % g (k) (0, ")
% !
G= + ta+b% g(t) − tk dt
k!(a + k + b" + 1) 0 k!
k=0 k=0
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.30
A first implementation [T. Binoth & G. Heinrich’00]
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.31
A first implementation [T. Binoth & G. Heinrich’00]
Strategies that are guaranteed to terminate
[C. Bogner & S. Weinzierl’07]
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.31
A first implementation [T. Binoth & G. Heinrich’00]
Strategies that are guaranteed to terminate
[C. Bogner & S. Weinzierl’07]
(It is assumed that all the terms in the function V have the
same sign.)
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.31
A first implementation [T. Binoth & G. Heinrich’00]
Strategies that are guaranteed to terminate
[C. Bogner & S. Weinzierl’07]
(It is assumed that all the terms in the function V have the
same sign.)
A, B, C, X
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.31
A first implementation [T. Binoth & G. Heinrich’00]
Strategies that are guaranteed to terminate
[C. Bogner & S. Weinzierl’07]
(It is assumed that all the terms in the function V have the
same sign.)
A, B, C, X
Strategy S [A.V. Smirnov & M.N. Tentyukov’08]
FIESTA
(Feynman Integral Evaluation by a Sector decomposiTion
Approach)
http://science.sander.su
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.31
FIESTA 2 [A.V. Smirnov, VS & M.N. Tentyukov’09]
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.32
FIESTA 2 [A.V. Smirnov, VS & M.N. Tentyukov’09]
data bases (Kyoto cabinet)
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.32
FIESTA 2 [A.V. Smirnov, VS & M.N. Tentyukov’09]
data bases (Kyoto cabinet)
multiprecision arithmetics
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.32
FIESTA 2 [A.V. Smirnov, VS & M.N. Tentyukov’09]
data bases (Kyoto cabinet)
multiprecision arithmetics
integration in c++ (cuba library
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.32
FIESTA 2 [A.V. Smirnov, VS & M.N. Tentyukov’09]
data bases (Kyoto cabinet)
multiprecision arithmetics
integration in c++ (cuba library
Speer sectors are implemented.
They are reproduced within Strategy S.
[A.V. Smirnov & VS’09]
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.32
SecDec
[J. Carter & G. Heinrich’10, S. Borowka & G. Heinrich’11–13 ]
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.33
SecDec
[J. Carter & G. Heinrich’10, S. Borowka & G. Heinrich’11–13 ]
An important new feature of SecDec 2.1 is
the possibility to apply it at physical values of kinematic
invariants, i.e. where the second function of alpha
parameters has terms of different sign.
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.33
SecDec
[J. Carter & G. Heinrich’10, S. Borowka & G. Heinrich’11–13 ]
An important new feature of SecDec 2.1 is
the possibility to apply it at physical values of kinematic
invariants, i.e. where the second function of alpha
parameters has terms of different sign.
Based on a contour deformation in parametric integrals
[D. Soper’99, C. Anastasiou, S. Beerli & A. Daleo’07]
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.33
FIESTA 3 [A.V. Smirnov’13]
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.34
FIESTA 3 [A.V. Smirnov’13]
Can be applied at physical values of kinematic invariants.
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.34
FIESTA 3 [A.V. Smirnov’13]
Can be applied at physical values of kinematic invariants.
mpi parallelization
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.34
FIESTA 3 [A.V. Smirnov’13]
Can be applied at physical values of kinematic invariants.
mpi parallelization
A geometrical strategy based on computational geometry
[ T. Kaneko & T. Ueda’10]
is implemented.
Vladmir A. Smirnov Atrani, September 30 – October 05, 2013 – p.34