Download as pdf or txt
Download as pdf or txt
You are on page 1of 51

A Complex Path Around the

Sign Problem
(a multimessenge talk)

Paulo Bedaque,
A.Alexandru, G.Basar, H.Lamm,
S.Lawrence, G.Ridgway, N.Warrington
The “sign problem”

−S[ϕ] 𝒩
∫ Dϕ 𝒪e 1
𝒩∑
⟨𝒪⟩ = ≈ 𝒪[ϕ]
∫ Dϕ e −S[ϕ] n=1

ϕ distributed ~ e −S

What if S is complex? (finite density, real time, …)


The “sign problem”

−iSI[ϕ] −SR[ϕ]
∫ Dϕ 𝒪e e
⟨𝒪⟩ =
∫ Dϕ e −iSI[ϕ]e −SR[ϕ]
The “sign problem”

−iSI[ϕ] −SR[ϕ] −SR[ϕ]


∫ Dϕ 𝒪e e ∫ Dϕ e
⟨𝒪⟩ =
∫ Dϕ e −SR[ϕ] ∫ Dϕ e −iSI[ϕ]e −SR[ϕ]
The “sign problem”

−iSI[ϕ] −SR[ϕ] −SR[ϕ]


∫ Dϕ 𝒪e e ∫ Dϕ e
⟨𝒪⟩ =
∫ Dϕ e −SR[ϕ] ∫ Dϕ e −iSI[ϕ]e −SR[ϕ]
1 𝒩 −iSI[ϕn]
𝒩
∑n=1 e 𝒪[ϕn] ⟨𝒪e −iSI
⟩SR
≈ =
1

𝒩
e −iSI [ϕn] ⟨e −iSI⟩SR
𝒩 n=1
The “sign problem”

−iSI[ϕ] −SR[ϕ] −SR[ϕ]


∫ Dϕ 𝒪e e ∫ Dϕ e
⟨𝒪⟩ =
∫ Dϕ e −SR[ϕ] ∫ Dϕ e −iSI[ϕ]e −SR[ϕ]
1 𝒩 −iSI[ϕn]
𝒩
∑n=1 e 𝒪[ϕn] ⟨𝒪e −iSI
⟩SR
≈ =
1

𝒩
e −iSI[ϕn] ⟨e −iSI⟩SR
𝒩 n=1

exponentially small on
spacetime volume
Central idea: deform the contour into the complex
plane:
Z
(z i20)2
p
dxe = ⇡


1.0

0.8

0.6
4 × 10173

0.4
2 × 10173

0.2

-3 -2 -1 1 2 3
-3 -2 -1 1 2 3
-2 × 10173

-4 × 10173

Cristoforetti, DiRenzo, Scorzato, ‘12


How to find good contour deformations ?

• integrands have no pole except at infinity


• outside the “good” asymptotic directions the integral diverges

Z
(hz+z 2 +z 4 )
dz e

integral depends
idle
,
'
: only
good on initial and final
asymptotic
: '
\

directions:
directions homological classes
How to find good deformations ?
method 0: thimbles

Thimbles are the generalization of


steepest descent = constant phase
paths
Problems:

• finding the thimbles that contribute


• sampling isolated thimbles
How to find good deformations ?
method I: approach the thimbles

d R
i @S R
@S I
= =
tff
d i @S
dt @ Ri @ I
i
A

= ) d Ii @S R @S I
dt @ i = =
dt @ Ii @ R
i

(real) field space gradient flow hamiltonian


of SR , flow of SI,
keeps integral keeps phase fixed
well defined
How to find good deformations ?
method I: approach the thimbles
SR

relevant region
before flow
SR under the flow
How to find good deformations ?
method I: approach the thimbles
SR

relevant region
after flow
SR under the flow
How to find good deformations ?
method I: approach the thimbles
SR

if this integral exists, so does that

1) homology class preserved by the flow


How to find good deformations ?
method I: approach the thimbles
SR
SR

ϕ
<latexit sha1_base64="/CWAeh720fOm1hlGjfM1Hk2o3XQ=">AAAB63icdVDLSsNAFJ3UV62vqks3g0VwFSZpqHZXdOOygn1AG8pkOmmGziRhZiKU0F9w40IRt/6QO//GSVtBRQ9cOJxzL/feE6ScKY3Qh1VaW9/Y3CpvV3Z29/YPqodHXZVkktAOSXgi+wFWlLOYdjTTnPZTSbEIOO0F0+vC791TqVgS3+lZSn2BJzELGcG6kIZpxEbVGrJR3WsgDyLbaTpuo2GIh+pN14WOjRaogRXao+r7cJyQTNBYE46VGjgo1X6OpWaE03llmCmaYjLFEzowNMaCKj9f3DqHZ0YZwzCRpmINF+r3iRwLpWYiMJ0C60j99grxL2+Q6fDSz1mcZprGZLkozDjUCSweh2MmKdF8ZggmkplbIYmwxESbeComhK9P4f+k69qOiejWq7WuVnGUwQk4BefAARegBW5AG3QAARF4AE/g2RLWo/VivS5bS9Zq5hj8gPX2CW0Yjn0=</latexit>

relevant region
after flow
2) sign fluctuations are reduced
How to find good deformations ?
method I: approach the thimbles

.is#.Ii&
÷€÷€f
;
i

critical point:

§
@ S̄
=0
@ i thimble: SI=const.
N (real) dimensions

Small regions are mapped (close) to thimbles and


contribute significantly to the integral, SI varies little.
The algorithm J

∂ϕ̃
∫ℝN dϕ𝒪 ∂ϕ e −SR−iSI
∫ dϕ𝒪e −SR−iSI ∫ℳ dϕ̃𝒪e −SR−iSI
⟨𝒪⟩ = = =
∫ dϕe −SR−iSI ∫ℳ dϕ̃e −SR−iSI ∂ϕ̃
∫ℝN dϕ ∂ϕ
e −SR−iSI
Seff

∫ℝN dϕ𝒪e −iSI+iIm ln J e −(SR − Re ln J)


=

'€t*()
∫ℝN dϕe −iSI+iIm ln J e −(SR−Re ln J)

'm
⟨𝒪e −iSI+iIm ln J⟩Seff

=
A
=
⟨e −iSI+iIm ln J⟩Seff

"

%
dJij @2S
= Jjk
J = detJ(T ) dt @zi @zk
Jij (0) = I

this is the expensive part


The algorithm J

∂ϕ̃
∫ℝN dϕ𝒪 ∂ϕ e −SR−iSI
∫ dϕ𝒪e −SR−iSI ∫ℳ dϕ̃𝒪e −SR−iSI
⟨𝒪⟩ = = =
∫ dϕe −SR−iSI ∫ℳ dϕ̃e −SR−iSI ∂ϕ̃
∫ℝN dϕ ∂ϕ
e −SR−iSI
Seff

∫ℝN dϕ𝒪e −iSI+iIm ln J e −(SR − Re ln J)


=

'€t*()
∫ℝN dϕe −iSI+iIm ln J e −(SR−Re ln J)

'm
𝒪⟨e −iSI+iIm ln J⟩Seff

=
A
=
⟨e −iSI+iIm ln J⟩Seff
our algorithm

"

%
=
Metropolis in the real space,
action Seff and
reweighted phase ei Im(ln J)-i Im(S)
Case study: massive Thirring model

Z 2
2 ¯a a g ¯a a ¯a a
S = d x ( µ @µ + µ 0 + m) + µ µ
2NF
Z
NF
! 2 d2 xAµ Aµ NF log det ( µ (@µ + Aµ ) + µ 0 + m)
2g
1 X
A projection of the thimbles: = 2
L x
A0 (x)

Imϕ

0.75
⊗ ⊗

0.70
⊗ ⊗

0.65

-1.5 -1.0 -0.5 0.5 1.0 1.5


Reϕ

critical
points
1 X
A projection of the thimbles: = 2
L x
A0 (x)

Imϕ

0.75
⊗ ⊗

■ ■ ■ ■

0.70
⊗ detD=0,
Seff = ∞
thimbles 0.65

-1.5 -1.0 -0.5 0.0 0.5 1.0 1.5


Reϕ

critical
points
1 X
A projection of the thimbles: = 2
L x
A0 (x)

Imϕ

0.75
⊗ ⊗

■ ■ ■ ■
0.70
⊗ ⊗

same



●●

●●


●●

●●


●●

●●


●●

●●


●●

●●


●●

●●


●●

●●


●●

●●


●●

●●


●●

●●


●●

●●


●●
●●


●●

●●


●●

●●


●●

●●


●●

●●


●●

●●


●●

●●


●●

●●


●●

●●


●●









































●●

●●
●●

●●
●●

●●

●●


●●

●●


●●

●●


●●

●●


●●

●●


●●

●●


●●

●●


●●

●●


●●

●●


●●

●●
● ●













































●●

●●
●●

●●
●●


●●


●●

●●


●●

●●


●●

●●


●●

●●


●●

●●

0.65

●●






































●●

●●
●●

●●
●●


●●


●●

●●


●●

●●


●●

●●


●●

●●


●●

●●


●●
● ●

●●

●●


●●
●●


●●

●●


















































●●

●●

●●
●●

●●


●●

●●


●●

●●


●●


T=0.01



●●

●●


●●

●●


●●

●●


●●

●●


●●

●●


●●

●●


●●

●●


●●

●●


●●

●●


●●

●●


●●

●●


●●

●●


●●

●●


●●

●●


●●

●●


●●

●●


●●

●●


●●

●●


●●

●●


●●

●●

●●


●●

●●


●●

●●


●●

●●
homological
class

●●

●●


●●

●● ●


●●

● ●
●●

●● ●●


●● ●●


●●

● ●●

●●


●●

●●


●●

●●


●●

●●

● ●
●●


●●

● ●

●●

●● ●●


●●
● ●


●●

●●
● ●


●●

●●


●●

●●


●●

●●


●●

●●


●●

●●


●●

● ●
●●


●●

●●


●● ●

●●

●●

● ●

●●


●●
● ●
●●

●●


●●

●● ●

●●


●●

●●


●●

●●


●●





●●

●●


●●

●●


●● ●
●●

●●


●● ●●

●●
● ●●


● ●

●●

●●


●● ●●


●●

●●


●●

●●




●●


●●

●●


●●

●●


●●

●●


●●

●●


●●

●●


●●

●●


●●

●●


●●

●●


●●

●●


●●

● ●
●●

●●


●●

●●


●●

●●


●●

●●


●●

●●


●●

●●


●●

●●
● ●
●●

●●


●●

●● ●
●●


●●
●●


●●
● ●
●●


●●

●●


●●

●●


●●

●●


●●

●●


●●

●●


●●

●●


●● ●

●●

●●


●●

●●


●●

●●


●●

●●


●●

●●


●●

●●

●●


●●

●●


●●

●●


●●

●●


●●




●●

●●


●●

●●


●●

●●
● ●

●●

●●


●●

●●


●●

●●
● ●
●●


●●

●●


●●

●●


●● ●
●●


●●

●●


●●

●●


●●

●●

●●

●●


●●

●●


●●

●●


●●

●●


●●

●●


●●

●●


●●

●● ●●


●●

●●


●●

●●


●●

●●


●●

●●


●●

●●


●●

●●


●●


●●

●●


●●

●●


●●

●●


●●

●●


●●

●●


●●

●●


●●

●●


●●

●●


●●

●●


●●

-1.5 -1.0 -0.5 0.0 0.5 1.0 1.5


Reϕ
1 X
A projection of the thimbles: = 2
L x
A0 (x)

Imϕ

0.75
⊗ ⊗

■ ●







●●


● ●


■ ●
●●
●●
●●
■ ●
●●
●●
● ●
■ ●●

● ●
● ●






●●


● ● ●
●● ●● ● ● ●

● ●
● ●●
● ● ● ●


● ● ●●
●● ● ● ● ● ●



● ● ● ● ● ● ●

0.70





● ● ●
● ● ● ● ●
● ●

●●

● ● ● ● ● ● ● ●







● ●
● ● ● ● ●









● ● ● ● ●
● ●
● ●
●● ● ● ●
● ● ● ● ●
● ● ●
● ● ● ● ● ● ●

● ●
● ● ● ● ● ●
● ●

● ● ●
● ● ● ●
● ●
● ●●

⊗ ⊗
● ●
● ● ● ● ● ● ●
● ●
● ●● ●
● ●
● ●
● ●
● ●

● ● ● ●

T=0.05
● ●
● ● ● ● ●
● ● ● ●

●● ●
● ●
● ● ● ●● ●
● ●


● ●
● ●
● ●
● ●
● ●
● ●
● ●


● ●
● ●
● ● ●
● ●
● ●

● ●


● ●

● ●
● ●
● ● ●
● ●
● ●●

● ●
● ●● ●● ●
● ●
● ●
● ●


● ●

● ●

● ●
● ●
● ●
● ●

● ●


● ●
● ●
● ●
● ●
● ●
●● ●
● ●

● ● ● ●

flowed

● ●
● ●

● ●
● ●
● ●

● ●
● ●



● ● ●
● ●
● ●
● ●
● ● ●



● ●
● ●●
● ●
● ●
● ●


● ●●


● ● ●
● ●

● ●

● ●
● ●
● ●


● ●
● ●
● ●● ●
● ●
● ● ●

●● ●
● ● ●
● ●
● ● ●
● ●●

● ● ● ●
● ●

● ●
● ●
● ●

● ●
● ● ●

● ●
● ● ● ●
●● ●
● ●● ●
● ●
● ●
● ●
● ●●

● ● ● ●

● ●

● ● ● ●


●●
● ●● ●
● ● ● ●● ●
● ●●

● ●
● ● ● ● ● ●● ●


● ●
● ●
● ●
● ● ●
● ●
● ●



● ●

● ● ● ●
● ●
● ●
● ●


● ●
● ●
● ●
● ● ●● ●
● ●


● ●
● ● ● ●
● ● ●
● ●


●● ●

● ●
● ●● ● ●
● ●

● ●●

● ●● ●● ●
● ●
● ●
● ●
● ●


●● ●
● ●
● ● ● ●
● ●● ●●


● ●
● ●
● ●
● ●
● ●
● ●
● ●


● ●● ●
● ●
● ●
● ●

● ●
● ●


●● ●
● ●● ●
● ● ●
● ●
● ●●

●●
● ●
● ●

● ●
● ●
● ●● ● ●
●●
●●
● ●

● ●
● ●
● ●● ●
● ●●
● ●
●●


● ●
● ●
● ●
● ●
● ●

● ●
● ●




● ●● ●

● ●● ● ●
● ●
● ●

●●

●●

● ●

● ●
● ●
● ●

● ●
● ●

● ●

●●

●●
● ●● ●
● ●
● ●
● ●
● ● ●
●●



●●

●● ●
● ●
● ●

● ●
● ●
● ●
● ●●

●●


●●

●●
● ●● ●
● ●
● ●

● ●
● ●

● ●
●●


●●
● ●

●●


●● ●
● ● ●

● ●
● ●
● ●
● ●
●●


●●

● ●




● ●● ● ● ●● ●

tangent
●●

●●


●●

●●

●●

●●
● ●


●●

●●

●●
● ●
● ●
● ●
● ●

● ●● ●●
● ●

●●

●●

●●

● ●●

●●

●●


●●

●●

●●


●●

●●

●●

●●


●●

●●


●●

●●


●●

●●


●●

●●

●●

●●

●●

●●

● ●
● ● ●
● ● ●
● ● ●●

●●

●●

●●


●●

●●

●●

●●


●●

●●

●●

●●


●●

●●

●●

●●


●●



● ●● ●
● ●● ●

● ●●
●● ●

● ●
● ●
● ●
● ●



●● ●●
● ●
● ●
● ●

● ●



● ●
● ●
● ●● ●
● ●
●●


● ● ●●
● ●
● ●
● ●●
● ● ● ● ●

0.65 T=0.01


● ●
● ●

● ●● ●
● ●
●●


● ●
●● ●● ●
● ●
●● ●●




● ●
● ●

● ●●
● ●
● ●
●●



●● ●

● ●● ●
● ●●
● ●

●●


●●
● ●
● ●
● ●● ●
● ●

●●
●●

● ●●
● ●●
● ●
● ●

● ●●



●●


● ●
● ●
● ●● ●

● ●
●●



●●
●●

●●
●● ●


●●
● ●

● ●●
● ●

● ●● ●●

●●
● ●

●●
●●

●●
●●


●●
● ●
●● ●●


● ●
●● ●
● ●
● ●
●● ●
●●

● ●●
● ●●



●●
● ●
●●
● ●


●●

● ●
●● ●
● ● ●

●● ●●


●● ●●

● ●





●● ●
●●
● ●●

●●

● ●
●●
● ●
● ●
● ●
●● ●


●●

● ●●
● ●





● ●●
● ●●


●●
● ●

●● ●

● ●● ●

●● ●


●●

● ●●
● ●



●●

● ●
●● ●●

●●

● ●

●● ●●

●●
●●

●●
●●
● ●
● ●●
● ●●
●●
●●

●●
●●

● ●
●●

● ●●

●●

● ●

● ●

●●


● ●
●●
● ●


●●

●● ●●


● ●

●●
● ●●


● ●
●● ●
● ●●

● ●


●● ●●

●● ●
●●


●●
● ●●

● ●

●●

●●
● ●●
● ●
●●

●●

● ●

●●
● ●

●● ●

●● ●
●● ●

● ●
●●
● ●
●●
● ●

●●
● ●
●●

●●

●● ●●
● ●
●●


●●
● ●
●●
● ●●

●●

●●

● ●

●●


● ●

● ●

● ●

● ●
●● ●

●● ●
●●
● ●●


●●
● ●
●●

●●

●●
● ●●
● ●●



●●

●● ●●
● ●


●●

●●

●●
● ●●

●●


● ●

●● ●

●● ●

●● ●

●● ●
●● ●●
● ●●


●●

● ●

●●

●●


●● ●
●●
● ●


●●
●●

●● ●●

●● ●●


●●

●●

●●

● ●●


●●


●●
● ●●

● ●
●● ●

● ●
●● ●

● ●●

● ●


●●


●●

● ●●

●●

●●


●●

● ●
●●
● ●
●●

●●

●●


●●
● ●

●● ●


●●

●●

●●

●●


●●
● ●
●●


●●


●●

●●

● ●

● ●
●●
● ●

●● ●

●● ●

●● ●●

● ●


●●


●●


●●

●● ●●

●●

●●


●●

●●

●●
● ●●

●● ●
●●


●●

●●

●●

●● ●
●●

●● ●●

●●

●●


●●

●●

●●

●●


●●

●●

●●

●●


●●

●●

●●

●●


●●

●●

●●


●●
● ●
●●
● ●
●●
● ●

●● ●

●● ●

● ●

●● ●

●●


●●


●●


●●

●●


●●

●●

●●

●●


●●

●●

●●

●●


●●

●●

●●

●●


●●
● ●

●●
● ●
●●

●●

●●



●●

●●

●●

●●
● ●
●●

●● ●

●●
● ●●

● ●
●●

● ●
●●

● ●
●●

● ●
●●
● ●

●●

● ●●

●●


●●

●●





●●

●●

●●

●●


●●

●●

●●
● ●

●●

●●

●●

●●


●●

●●

●●

●●

●●

● ●
●●

●●

● ●
●●

● ●

●● ●●

●● ●

●●
● ●

●● ●
●●

● ●●

●●


● ●

●●

●●

●●

●●

●●


●●

●●

●●

●●

● ●
●●
●●


●●

●●

●●

●●


●●

●●


●●


●●

●●

●●

●●


●●

●●


●●

●●

●●


●●

●●

●●

●●


●●

●●

●●

● ●●

●●

●●
● ●
●●

●● ●
●●

● ●

●●

● ●●


● ●●

●● ●

●●

● ●●

●●


●● ●●

●●

●●

●●


●●

●●

●●

●●


●●

●●

●●

●●


●●

●●

●●

●●


●●

●●
● ● ● ● ● ● ● ●

space


●●

●●

●●

●● ●

●●


● ●

●●

● ●

●●

●● ●
●●


●● ●
●●

●● ●
●●


●●
● ●
●●

●●

●●

●●

●●


●●

●●

●●

●● ●

●●

●●

● ●

●●

●● ●

●●

●●
● ●

●●


●● ●

●●

●● ●●

●●

●●
● ●

●●

●●

●●


●●

●●


●●

●●

●●

●●


●●

●●

●●

● ●●

●●

●●


●●

● ●

●●

●●

● ●
●●


●●

● ●●


●●

●● ●
●●


●●

● ●●

●●

●●


●●
● ●●

●●

●●


●●

●●

●●

●●






●●

●●

●●

●●


●●
● ●
●●

●●

●●
● ●

●●

● ●


●●
● ●

●●

● ●●

●● ●●

●●

●●

●● ●


●●

●●

●●

●●


●●


●●

●●

●●


●●

●●

●●

●●


●●

●●

●●

●●


●●

●●

●●

●●


●●

●●

●●

●●


●●

●●

●●

● ●

●●


●●

●●

●●

●●


●●

●●

●●

●●

●●


●●

●●

●●

●●

●● ●

●●


●●

●●

● ●
●●


●●


●●
● ●


●●


●●

●● ●

●●

●●

●●

●●
● ●
●●

●●

●●

●●

●●

●●

●●


●●

●●

●●

●●


●●

●●

●●

●●
● ●

●●

●●

●●

●●


●●

●●

●●

●●


●●

●●

●●

●●

●●

●●


●●

●●

●●

●●


●●

●●

●●





●●

●●

●●

●●


●●

●●

● ●

●●

●●

●●

●●

●●

●●

●● ●●


●●


●●


●●

●● ●
●●


●●

●●


●●


● ●
●●

●●

●●

●●

●●

●●

●● ●
●●

●●


●●

●●


●●

●●

●●

●●

●●

●●


●●

●●

●●

●●

●●


●●

●●

●●

●●


●●




●●





●●





●●


●●
●●

●●


●●


●● ●

●●

●●


●●


●●


●●



●●




●●





●●



●●

●●


●●

●●

●●

●●

●●


●●

●●

●●

●●


●●


●●


● ●
●●

●●

●●

●●


●●

●●

●●

●●


●●

●●

●● ●

●●

-1.5 -1.0 -0.5 0.0 0.5 1.0 1.5


Reϕ
1 X
A projection of the thimbles: = 2
L x
A0 (x)


Imϕ ●

● ●

● ●

● ●

0.75
⊗ ⊗

■ ●●
● ●
● ● ●● ■ ●
■ ●
■ ●
●●
● ●● ●
T=0.5

● ● ●● ●● ●● ●
● ● ● ● ●● ● ●●

● ●
● ●● ●● ●
● ●
● ● ●
● ●●● ●
● ● ●●

● ● ●
● ● ●● ●

● ● ● ●
● ●


● ● ●

●● ● ● ●● ●● ● ●


● ● ● ●

● ●
● ●
●● ●●
● ●
●● ● ● ●


● ● ●
●●
● ●
●●
● ● ● ●


● ● ● ●


● ● ●●
● ● ●
●● ● ●
● ● ● ●

0.70

● ● ●
●● ● ● ● ● ● ●

●● ● ●

●●
● ● ●
●●
● ● ●● ●● ● ●
●● ● ●●
● ● ●
● ● ● ● ●● ● ●









●● ● ●
● ● ● ● ●
● ● ●

● ●







● ●● ● ● ● ●
● ● ●
● ●
●● ● ●● ● ● ●
● ● ● ● ● ● ●● ● ●
● ● ●●
● ●
● ● ● ● ● ●● ● ●

● ●
● ●
● ● ● ● ● ● ● ●
● ●
● ●

● ● ●
● ●
● ● ● ●
● ●
● ●
● ●●

⊗ ⊗
● ●
● ●
● ● ● ● ● ●● ● ● ● ●●
● ● ●
● ●
● ●
●● ● ●
● ●● ● ●
● ● ●● ●
● ●

● ● ● ● ●

T=0.05
● ●
● ●●
● ● ● ● ● ●
● ● ● ●● ● ●

●● ●
● ●

● ● ●● ● ● ●● ● ●● ●● ●
●● ●
● ●


● ●
● ●
●● ● ●
● ●
● ● ●
● ●●
● ●
● ●

● ●




●●


●●
●●
●● ●●● ●





● ● ● ●







●●
● ●●


●● ●





●●

● ●
● ●
●● ●
● ●
● ●
● ● ●
● ●● ●
●● ●
● ●


● ●

● ●●
●● ●● ●
● ●
● ● ● ●
● ●
● ●● ●●

● ●●
● ●


● ●
● ●●
● ●
● ●●
● ●
● ●
● ●
●● ●●●● ●
●● ●
● ●

● ● ● ● ● ●

flowed

● ●
● ●
●●
● ●●● ●

● ●● ● ● ● ●
● ●

● ●●● ●●
●● ●
● ●



● ● ●
●●
● ●●● ●
● ●
● ● ●
● ● ●

●● ● ●



● ●
● ●●
●●
● ●
●● ●

● ●
● ● ● ● ●
● ●
● ●●
● ●
●●
● ● ●●


● ● ●
●●
● ●
●●
● ●
● ●

● ●
● ●

● ●
●●
● ●●
●●
● ●
● ●


● ●
● ●
●●

●● ●●
●●
● ●
● ●
● ● ● ● ●
● ●
● ●
●●
●● ●●

●● ● ●

●● ● ●●
●●
●● ●●

●●
●● ● ●
● ● ● ●
●●
●●
●● ●●
●●
●● ●
● ●●
● ●● ●●
●●
●●
● ●
●●
●●

●● ● ●
● ● ● ●
● ●
● ●
●●
●●

●● ●
●●
●●
●● ●
● ●

● ●
● ●
●●
●●
●●

●●


●●

●● ●
●●

●●

●●

●●

●●
●● ● ●●
● ●

● ● ●●

●●

●●

●●

●●

●● ●●


●●

●●


●●
●●
●●
● ● ●

●● ● ●

●●

●●


●●

●●

●●

●●


●●

●● ●
● ●
● ● ● ● ●● ●
● ●●

●●


●●

●●

●●


●●

●●


●●
● ● ●●

● ●
● ● ●

● ● ● ●
● ● ●
● ●


● ●● ●● ● ● ●● ●
● ●
●●


● ●
● ● ● ● ● ● ● ● ●
● ●


● ●
● ●
● ●
● ● ● ● ●
● ●
● ●



● ●

● ● ● ● ●
● ●
● ●


● ●
● ●
● ●
● ● ● ● ●● ●
● ●


● ●
● ● ● ● ● ● ● ● ●
● ●


●● ●

● ●
● ●
● ● ● ● ●
● ●

● ●●

● ●● ●
● ●
● ● ● ● ●
● ●
● ●
● ●


●● ●
● ●
● ● ● ● ● ●
● ●
● ●●


● ●
● ●● ●
● ● ● ● ●
● ●
● ●


● ●● ●
● ●
● ● ● ●
● ●

● ●
● ●


●● ●
● ●
● ●
● ● ● ● ●
● ●
● ●●

●●
● ●
● ●●
● ●
● ● ● ●
● ●● ● ●
●●
●●
● ●

● ●
● ●
● ● ● ●● ●
● ●

● ●
●●


● ●
● ●
● ●
● ● ● ●
● ●

● ●
● ●




● ●● ●

● ●
● ● ● ● ●
● ●
● ●

●●

●●

● ●

● ●
● ●
● ● ● ●

● ●
● ●

● ●

●●

●●
● ●● ●
● ●
● ●
● ●
● ●● ●
● ● ●
●●



●●

●● ●
● ●
● ●

● ● ● ●
● ●
● ●
● ●●

●●


●●

●●
● ●● ●
● ●
● ● ● ●

● ●
● ●
●● ●
●●


●●
● ●

●●


●● ●
● ● ●●
● ●● ●
● ●
● ●
● ●
● ●
●●


●●

● ●




● ●● ● ● ●● ●

tangent
●●

●●


●●

●●

●●

●●
● ●


●●

●●

●●
● ●
● ●● ●
● ● ● ●

● ●● ●

● ●

●●

●●

●●

● ●●

●●

●●


●●

●●

●●


●●

●●

●●

●●


●●

●●


●●

●●


●●

●●


●●

●●

●●

●●

●●

●●

● ●
● ● ●
● ●● ●
● ● ●
● ● ●●

●●

●●

●●


●●

●●

●●

●●


●●

●●

●●

●●


●●

●●

●●

●●


●●



● ●
● ●
● ●
● ●
● ●
● ●

● ●

●● ●
●● ●
● ● ● ●
● ●
● ●
●●

●● ●
● ●
● ●
● ●
● ●
● ●

● ●●

● ●●
● ●
● ●● ●
● ●● ●
● ●

●●
● ● ●
● ●
● ●
● ●
● ●
● ●

●● ● ● ● ● ● ● ●●

0.65 T=0.01


● ●
● ●

● ●
● ●
● ●● ●
● ●●



● ●●
● ●
● ●● ●
● ●
● ●●
● ●





● ●
● ●●
● ●
● ●
● ●●
● ●
● ●●




●● ●●
● ●
● ●● ●● ●
● ●●
● ●●




●● ●
● ●● ●
●● ●

● ● ●● ●
●●


●●

● ●
● ●
● ●
● ●
● ●

● ●
● ●
●●

●●

● ●●
● ●●
● ●●
● ●●
● ●● ●●
● ●

●●


●●
●●

●●
● ●●

●●
● ●
● ●
● ● ● ●
● ●●
● ●
●●


● ●●

●●

●●
●●
●●
● ●

● ●
●●

● ●

● ●

● ● ● ●● ●
●● ●


●● ●
●● ●
●●


●●

● ●

●● ●●

●●
● ●

●● ●● ●● ● ●
● ●
●● ●

●●

● ●

●● ●●



●● ●

●● ●
●●


●● ●

● ●
● ● ●
● ●
● ●

● ●
●●


●● ●

●● ●●




●● ●
●● ●
●●


●● ●

●● ●● ●
● ● ●● ●

●● ●
●●


●● ●
● ●




●● ●

●● ●
●●

●●
● ●


● ●
●●
●●
●●

●●
● ●
●● ●
● ●
● ●
● ●
●●
●●

●●
●●
● ●

●●
● ●

●●

●● ●
●●
● ●●





● ●
●● ●●


●●

● ●
●●

● ●●

●● ●


●● ●
●● ● ● ●

● ●●

●● ●●


● ●●


● ●

●●


● ●●
● ●



●●
● ●
●● ●

●●

●●
● ●●
●●
● ●●

● ●
●●
● ●
● ●● ●● ●

● ●●
● ●

●● ●

●●
● ●

●●

●●

● ●
●● ●

●●
●●

● ●

●● ●
●●

●●

●●
● ●●

●●

● ●
●●
● ●
●●
● ●●

● ●● ●● ●
●● ●
●●
● ●

●● ●
●●
●●
●● ●

●●

●●

●● ●
●●
● ●
●●



●●
● ●
●●
● ●●

●●

●●


● ●
●●


●● ●●
● ●●
● ●

● ●● ●
● ●
●● ●●
● ●
●● ●

●●

●● ●


●●

●●

● ●
●● ●●





●●

● ●●

● ●

●●

●●

●●


●● ●●


●●


●● ●
●●
● ●●

● ●●
● ●● ●
● ●
●●
● ●
●● ●
●●
● ●

●●

●●


● ●
●●

●●

●●


●●
● ●

●● ●

●●





●●

● ●


● ●
●●

●●

●●


●●

●●
● ●


●●


●●


●●
● ●
●●
● ●

● ●●

● ●●
● ●●
● ●
●● ●
●●
● ●
●●
● ●

●●


●●


●●


● ●

●●

●●

●●

●●


●● ●●


● ●

●●





●●

●●

● ●

●●

● ●●

●●

●●

●●

●●

●●


●●

●●

●●

●●

●●

●●


●●

●●

●●

●●

●●


●●

●●

●●


●●

●●


● ●
●● ●●

● ●●

● ●
● ● ●
●●
● ●

●● ●●

● ●

●●


●●


●●


●●


●●

●●

●●

●●

●●

●●

●●


●●

●●

●●

●●

●●

●●

●●

●●


●●

●●

● ●●
●●
● ●

●●

●●



●●

●●

●●


● ●

●●

● ●●

●● ●

●● ●


●● ●●
● ●●
● ●

●● ●●

● ●

●● ●


●●
● ●

●●

●●


●●




●●

● ●●


●●

●●

●●

●●

●●

●●

●● ●

●●

●● ●
●●
● ●
●●
● ●
●●

● ●
● ●
●● ●●


● ●●

● ●

●●
● ●●

●●

● ●●

●●

●●

●●

●●

●●

●●


●● ●

●●

●●




●●

●●


●●

●●

●●

●●


●●

●●

●●

●●


●●

●●


●●

●●

●●


●●

●●

●●

●●


●●

●●

●●

●●

●●

●●

●●
● ●

●●

●●
● ●
●●

●● ●●

●● ●●

●●
● ●
●● ●● ●


●●
● ●

●●
● ●●

●●
● ●

●●

●●
● ●
●●

●●

●●


●●

●●

●●

●●


●●

●●

●●

●●


●●

●●

●●

●●


●●

●●

●●

●●


●●

●●

●●
●●


●●

●●

●●

●●
●●

● ●●
● ●
● ●
● ●
● ●
● ●
● ●
● ●● ●●

space


●●

●●

●●

●● ●
●●

●●
● ●●

●●
● ●

●●


● ●
●● ●
●● ●
●●


●● ●
●●

●● ●
●●

●●

● ●●

●●

●●

●●


●●

●●

●●


●●

●● ●●

●●


●● ●●

●●

● ●

●●


●● ●●

● ●
●●
● ●

●●


●● ●

●●

●● ●

●●

●●

● ●
●●

●●


●●

●●

●●


●●

●●

●●

●●


●●

●●

●●

●● ●

●●

●●

●●

●●
● ●

●●

●●

● ●

●●


●●
● ●
●●
● ●●

● ●●


●●

●● ●
●●

●●


● ●●

●●

●●

●●

● ●●

●●

●●

●●


●●

●●

●●

●●


●●

●●

●●

●●


●●

● ●●
●●

●●

●● ●


●●
● ●●


●● ●

● ●
●● ●

●●

● ●●

●● ●

●●


●●

●● ●

●●


●●

●●

●●

●●




●●

●●

●●

●●


●●

●●

●●

●●


●●

●●

●●

●●

●●

●●

●●

●●


●●

●●

●●

●●

●●

●●

● ●
●●

●●


●●

●●

●●

●●

●●


●●

●●

●●

●●

●●

●●
●●


●●
● ●●

●●

●●

●●

● ●

●●

●●


●●
● ●
●●
●●
● ●


● ●
● ●●


●●

●●

●● ●

●●

●●

●●

●●
● ●
●●●
●●


●●
●●


●●

●●

●●

●●

●●

●●

●●


●●

●●

●●

●●
● ●●

●●

●●

●●


●●

●●

●●

●●

●●


●●

●●

●●

●●

●●

●●

●●


●●

●●

●●

●●

●●

●●


●●

●●

●●

●●


●●

●●

●●

●● ●

●●

●●

●●

●●

●●

●●


● ●
●●

●●


●●


●●

●● ●●
●●

●● ●●
●●

●●
● ●

●●

●●

●●


●●

● ●

●●

●●

●●

●●

●●

●●

●● ●●


●●

●●

●●

●●


●●

●●

●●



●●

●●

●●

●●

●●


●●

●●

●●

●●

●●

●●


●●




●●





●●




●●



●●

●●


●●





●●



●●

●● ●●●●


●●




●●




●●

●●


●●





●●





●●





●●


●●

●●

●●

●●

●●


●●

●●

●●

●●

●●

●●




●●

●● ●

●●

●●


●●

●●


●●

●●


●●
●●

●●

●●
● ●●

-1.5 -1.0 -0.5 0.0 0.5 1.0 1.5


Reϕ
Case study: massive Thirring model
Wilson, 10 x10 lattice, NF=2, amf=0.3

〈n〉/mf
4 Re〈e-iSI 〉

1.0
3
RN 0.8

0.6
2
RN
0.4
1
0.2

μ/mf
μ/mf 1 2 3 4
0 1 2 3 4

sign problem
Real field calculation
Case study: massive Thirring model
Wilson, 10 x10 lattice, NF=2, amf=0.3

〈n〉/mf
4 Re〈e-iSI 〉

1.0
N
3 R
0.8
flow=0.0
2
0.6 RN
0.4 flow=0.0
1
0.2

μ/mf
μ/mf 1 2 3 4
0 1 2 3 4

sign problem improved


tangent plane calculation
Case study: massive Thirring model
Wilson, 10 x10 lattice, NF=2, amf=0.3

〈n〉/mf
4 Re〈e-iSI 〉

RN 1.0
3
flow=0.0 0.8
flow=0.4 RN
0.6
2
flow=0.0
0.4
flow=0.4
1
0.2

μ/mf
μ/mf 1 2 3 4
0 1 2 3 4

flow done with estimators for the jacobian


(difference reweighted)
flowed manifold calculation
Case study: massive Thirring model
Staggered, NF=2, T/mf=0.302

〈n〉/mf
1.5
Nt ×Nx =10×10
Nt ×Nx =10×20
1.0

0.5

μ/mf
1 2 3 4

thermodynamic limit
Case study: massive Thirring model
Staggered, NF=2, amf=0.265

a〈n〉
0.25
aT=1/10
0.20 aT=1/20

0.15 aT=1/30 step is missed


0.10
aT=1/40
in a one thimble
0.05 calculation

0.1 0.2 0.3 0.4 0.5 0.6

cold limit
Case study: massive Thirring model
Staggered, NF=2, T/mf=0.302, mfL=3.31
〈n〉/mf
2.0
Nt ×Nx =12×12
1.5
Nt ×Nx =16×16

Nt ×Nx =20×20
1.0

0.5 integer
μ/mf
particle number
0.5 1.0 1.5 2.0 2.5 3.0 3.5

continuum limit
How to find good deformations ?
method I: approach the thimbles

Same method applied to 1+1D QED


where a thimble decomposition does not even exist
Case study: Real Time Dynamics

Viscosities, conductivities, … require:


Z
0 1 H 0 1 iSc [ ] 0
h (t) (t )i = Tr(e (t) (t )) = D e (t) (t )
Z Z

Tmax
-iβ
Schwinger-Keldish
contour
Test case: The Sign Problem from Hell

Tmax
-iβ

field at a point in the real


axis does not contribute to
the damping factor in eiSc
iIm(iSc )
he i=0
Problems
•large flow needed (from RN)
•jacobian expensive
•anisotropic proposals
•tangent space in wrong homology class
“Pseudo-fermions” for the jacobian
)*÷ii±÷÷€€

J0 ⌘ = ⌘˜ ⌘˜k = J0 Re(⌘)
¥

•isotropic proposal
•no need to compute det(J0)

1+1D φ4: nt=10, nx=10, nβ=2, λ=0.1
weak coupling
Cp=0 (t)
real Cp=1 (t)

4
0.5
2

t t
0.5 1.0 1.5 2.0 0.5 1.0 1.5 2.0

-2
-0.5
-4
imaginary

p=0 p=2π/L
1+1D φ4: nt=10, nx=10, nβ=2, λ=0.1
weak coupling
Cp=0 (t) Cp=1 (t)

4
0.5
2

t t
0.5 1.0 1.5 2.0 0.5 1.0 1.5 2.0

-2
-0.5
-4

p=0 p=2π/L
1+1D φ4: nt=10, nx=10, nβ=2, λ=0.1
weak coupling
Cp=0 (t) Cp=1 (t)

4
0.5
2

t t
0.5 1.0 1.5 2.0 0.5 1.0 1.5 2.0

-2
-0.5
-4

p=0 p=2π/L
1+1D φ4: nt=10, nx=10, nβ=2, λ=1.0
strong coupling
Cp=0 (t) Cp=0 (t)

4
real 0.3

0.2
2
0.1

t t
0.5 1.0 1.5 0.5 1.0 1.5
-0.1
-2
-0.2
-4
-0.3

imaginary
p=0 p=2π/L
1+1D φ4: nt=10, nx=10, nβ=2, λ=1.0
strong coupling
Cp=0 (t) Cp=0 (t)

0.3
4
0.2
2
0.1

t t
0.5 1.0 1.5 0.5 1.0 1.5
-0.1
-2
-0.2
-4
-0.3

p=0 p=2π/L
1+1D φ4: nt=10, nx=10, nβ=2, λ=1.0
strong coupling
Cp=0 (t) Cp=0 (t)

0.3
4
0.2
2
0.1

t t
0.5 1.0 1.5 0.5 1.0 1.5
-0.1
-2
-0.2
-4
-0.3

p=0 p=2π/L
How to find good deformations ?
method II: learnifolds

.is#.Ii&
÷€÷€f
;
i

§

LT = (rough?) interpolation of points of MT


Interpolation = learning

€H
x€f)
%

.

no trapping
continuous map
between manifolds:
ϕ = ϕ̃R + it ϕI
Interpolation = learning
(xi wi + b) = y

sin(Re(A0 ))

cos(Re(A0 )) Im(A0 )

sin(Re(A1 )) Im(A1 )

cos(Re(A1 ))
outputs:
inputs: Im Aµ(0)
Re Aµ(x)
How to find good deformations ?
method III: sign optimized manifolds

−S−RelnJ
maximize i(SI+ImJ)
∫ Dϕ̃e
the average ⟨e ⟩=
phase
∫ Dϕ̃e −SR −RelnJ
σ

gradient of eiα computed by a short,


sign-problem free MC run
Mori, Kashiwa&Ohnishi, `17
Alexandru,Bedaque,Lamm&Lawrence, `17
Bursa&Kroyter, `18
How to find good deformations ?
method III: sign optimized manifolds

−SR −1
∫ℳ Dϕ̃ e (∂λSR − ReTrJ ∂λJ)
∂λ log | ⟨σ⟩ | =
∫ℳ Dϕ̃ e −SR

no sign problem
How to find good deformations ?
method II: sign optimized manifolds

€H
x€f)
%

.

no trapping

mapping of manifold: ϕ̃ = ϕ + if(ϕ)


How to find good deformations ?
method III: sign optimized manifolds
<ψψ>
0.20

2+1D 0.15

Thirring
6x6x6
0.10

model
0.05

μ
0.0 0.2 0.4 0.6 0.8 1.0

factorized ansatz: ϕ̃i = ϕi + if(ϕi)


How to find good deformations ?
method II: sign optimized manifolds
<ψψ>
0.20

2+1D 0.15

Thirring 0.10

model 8x6x6
0.05

0.00 μ
0.0 0.2 0.4 0.6 0.8 1.0

factorized ansatz: ϕ̃i = ϕi + if(ϕi)


How to find good deformations ?
method II: sign optimized manifolds
<ψψ>
0.20

2+1D 0.15

Thirring 0.10

model 12x6x6
0.05

0.00 μ
0.0 0.2 0.4 0.6 0.8 1.0

factorized ansatz: ϕ̃i = ϕi + if(ϕi)


How to find good deformations ?
method II: sign optimized manifolds
To take home:

• Deforming the integration on complex space is a good


thing

• Thimbles are just one possibility

• The beauty is on the (algorithmic) details

• Jacobians are expensive: estimators, “pseudo-fermions”,


(machine learned) ansatze, alternative flows, …

• Better proposals: solved at weak coupling, “pseudo-


fermions”
Previous attempts concentrated on using thimbles. Issues:
•Monte Carlo chain to stay on the thimble (Langevin
Cristoforetti et al. `12, HMC Fuji et al. `13)

•solution of classical e.o.m.=critical point: many thimbles


•which thimbles contribute? maybe only one thimble
matters in the i) thermodynamic limit and/or ii)
continuum limit
We’ll take a different route:

•forget thimbles
•too little flow=sign problem not ameliorated enough
•too much flow=Monte Carlo gets stuck into one region
in field space

You might also like