Numerical analytic continuationor
a nice interpolation and extrapolation method
Ralf-Arno Tripolt
Palaver
Frankfurt, October 29, 2018
Analytic continuation - the idea
[commons.wikimedia.org]
The interpolation method
The interpolation method
The interpolation method
For details...
[courtesy L. Holicki]
arXiv: 1610.03252
arXiv: 1801.10348
Simple Example: f(x) = 1/(x+ 100)
I we use 2 input points for theSPM
I can we reconstruct thefunction?
-300 -200 -100 0 100 200 300
-0.2
-0.1
0.0
0.1
0.2
x
f(x)
f(x)Input
f(x)=1
x + 100
Simple Example: f(x) = 1/(x+ 100)
I Only 2 input points areneeded to reconstructf(x) = 1
x+100
I it is the “first guess” of themethod
-300 -200 -100 0 100 200 300
-0.2
-0.1
0.0
0.1
0.2
x
f(x)
f(x)InputSPM
f(x)=1
x + 100
Another simple example: f(x) = x
I we use 3 input points for theSPM
I can we reconstruct thefunction?
-10 -5 0 5 10
-10
-5
0
5
10
x
f(x)
f(x)=xInput
Another simple example: f(x) = x
I 3 input points are needed toreconstruct f(x) = x
I with 15 digits precision oneobtains for example
f(x) =22 + 1.8 · 1015x
1.8 · 1015 − x ≈ x
-10 -5 0 5 10
-10
-5
0
5
10
x
f(x)
f(x)=xInputSPM
Another example: f(x) = ex
I we use 11 input points
I can we reconstruct thefunction?
-10 -5 0 5 10
0
5000
10000
15000
20000
x
f(x)
f(x)=exp(x)Input
Another example: f(x) = ex
-10 -5 0 5 10
0
5000
10000
15000
20000
x
f(x)
f(x)=exp(x)InputSPM
I for 11 input points we obtain
f(x) =263504 + 170536x+ 46451x2 + 10389x3 + 756x4 + 148x5
265568− 98809x+ 15473x2 − 1274x3 + 55x4 − x5
Another example: f(x) = ex
-10 -5 0 5 10 15 20-1×106
-500000
0
500000
1×106
x
f(x)
f(x)=exp(x)InputSPM
I for 11 input points we obtain
CN (x) =263504 + 170536x+ 46451x2 + 10389x3 + 756x4 + 148x5
265568− 98809x+ 15473x2 − 1274x3 + 55x4 − x5
What is a spectral function?
Ω
ΡHΩL
What is a spectral function?
Ω
ΡHΩLΠ
Ω = mΠ
What is a spectral function?
Ω
ΡHΩLΠ
Ω = mΠ
2 Γ
What is a spectral function?
Ω
ΡHΩLΠ
Ω = mΠ
2 Γ
Π Ψ Ψ
Ω ³ 2mΨ
Model for a spectral function
We use
ρ(ω2) = − 1
πIm
(1
ω2 −M2 −Π(ω2)
)with the self energy
Π(ω2) = S1 log(T 21−ω2)+S2 log(T 2
2−ω2)
and the parametersM = 50 MeV,S1 = 2000 MeV2,T1 = 0 MeV,S2 = 3000 MeV2,T2 = 300 MeV,
ω → ω + iε with ε→ 0
0 100 200 300 400 5001.×10-7
5.×10-71.×10-6
5.×10-61.×10-5
5.×10-5
ω [MeV]ρ[MeV
-2]
ρ(ω)
Model for a spectral function
0 100 200 300 400 5001.×10-7
5.×10-71.×10-6
5.×10-61.×10-5
5.×10-5
ω [MeV]
ρ[MeV
-2]
ρ(ω)Input
Model for a spectral function
0 100 200 300 400 5001.×10-7
5.×10-71.×10-6
5.×10-61.×10-5
5.×10-5
ω [MeV]
ρ[MeV
-2]
ρ(ω)InputSPM
Model for a spectral function
0 100 200 300 400 5001.×10-7
5.×10-71.×10-6
5.×10-61.×10-5
5.×10-5
ω [MeV]
ρ[MeV
-2]
ρ(ω)Input
Model for a spectral function
0 100 200 300 400 5001.×10-7
5.×10-71.×10-6
5.×10-61.×10-5
5.×10-5
ω [MeV]
ρ[MeV
-2]
ρ(ω)InputSPM
Model for a spectral function
0 100 200 300 400 5001.×10-7
5.×10-71.×10-6
5.×10-61.×10-5
5.×10-5
ω [MeV]
ρ[MeV
-2]
ρ(ω)Input
Model for a spectral function
0 100 200 300 400 5001.×10-7
5.×10-71.×10-6
5.×10-61.×10-5
5.×10-5
ω [MeV]
ρ[MeV
-2]
ρ(ω)InputSPM
Plot f(x) in the complex plane
We find poles in the complexplane at:ωP ≈ (236.43± i12.64) MeV
This pole characterizes the
particle which is associated to
the peak in the spectral
function
Model for overlapping resonances - 3 particles
The form factor could be measuredby experiment:
F (s) =
3∑i=1
M2i
M2i − s− iΓi
M2i√s
(k(s)
k(M2i )
)3with
k(s) =
√s
2
√1− 4m2
π/s
and the parametersM1 = 0.5 GeV,Γ1 = 0.2 GeV,M2 = 0.6 GeV,Γ2 = 0.1 GeV,M3 = 0.7 GeV,Γ3 = 0.15 GeV,mπ = 0.137 GeV
0.1 0.2 0.3 0.4 0.5 0.6
1.5
2.0
2.5
3.0
s @GeV2D
»FHsL»
Model for overlapping resonances - 3 particles
0.1 0.2 0.3 0.4 0.5 0.6
1.5
2.0
2.5
3.0
s [GeV2]
|F(s)|
|F(s)|
Input
SPM
Plot F(s) in the complex plane
exact result, F (s): reconstruction, SPM:
Complex pole of the charged ρ(770) meson
[ALEPH collaboration, Phys. Rept. 421, 191-284, 2005, arXiv:hep-ex/0506072]
Complex pole of the charged ρ(770) meson
[R.-A. T., I. Haritan, J. Wambach and N. Moiseyev, arXiv:1610.03252]
Complex pole of the charged ρ(770) meson
We find the complex pole ofthe charged ρ(770) meson tobe at
√sρ = Mρ − iΓρ/2 with
Mρ = 761.8± 1.9 MeV,
Γρ = 139.8± 3.6 MeV.
[R.-A. T., I. Haritan, J. Wambach and N. Moiseyev, arXiv:1610.03252]
The analytic continuation problem
Calculations at finite temperature are often performed using imaginary energies:
Ω
ip0
Ω
ip0
→
The analytic continuation problem
Analytic continuation problem: How to get back to real energies?
Ω
ip0
?
?
Analytic Continuation - Free Particle
We want to reconstruct thespectral function of a freeparticle
ρ(ω) = sgn(ω)δ(ω2 −m2)
starting from the freepropagator
DE(p0) =1
p20 +m2
with p0 = 2nπT ,m = 200 MeV, andT = 1/β = 10 MeV.
-1500 -1000 -500 0 500 1000 15000
5
10
15
20
25
p0 @MeVD
DE
Hp 0L@G
eV
-2
D
Ω
ip0
?
?
Analytic Continuation - Free Particle
We choose 51 input pointsfrom the Euclidean propagatorDE(p0) and apply the SPM toobtain the real-timepropagator DR(ω = ip0) andthe spectral function:
ρ(ω) = − 1
πImDR(ω)
-2000 -1000 0 1000 2000
0.5
1.0
2.0
5.0
10.0
20.0
p0 @MeVD
DE
Hp 0L@G
eV
-2
D
0 200 400 600 800 100010-20
10-15
10-10
10-5
1
105
1010
w @MeVD
rHwL
@Me
V-
2D
Summary
The Schlessinger point method can be used to obtain the analytic continuationof a function that is given in the form of numerical data.
I one can reconstruct the underlying function not only along the real axisbut also in the complex plane
I can be used to identify resonance poles and to “predict” decay thresholds(branch cuts)
I can be used to perform an analytic continuation based on Euclidean data
Schlessinger Point Method (SPM)
Given a finite set of N data points (xi, fi) we construct the rational interpolantp(x)/q(x) with polynomials p(x) and q(x) that is given by the continued fraction
p(x)/q(x) = CN (x) =f1
1 +a1(x− x1)
1 +a2(x− x2)
... aN−1(x− xN−1)
,
where the coefficients ai are given recursively by a1 = f1/f2−1x2−x1
and
ai =1
xi − xi+1
(1 +
ai−1(xi+1 − xi−1)
1+
ai−2(xi+1 − xi−2)
1+· · · a1(xi+1 − x1)
1− f1/fi+1
)The polynomials are of order (N/2− 1, N/2) for an even number of input points and
((N − 1)/2, (N − 1)/2) for an odd number of input points
[L. Schlessinger, Physical Review, Volume 167, Number 5 (1968)]
[R.W. Haymaker and L. Schlesinger, Mathematics in Science and Engineering, Volume 71, Chapter 11 (1970)]
[H.J. Vidberg and J.W. Serene, Journal of Low Temperature Physics, Vol. 29, Nos. 3/4 (1977)]
[A. Pilaftsis and D. Teresi, Nucl. Phys. B 874 (2013) 594-619]
[G. Marko, U. Reinosa and Z. Szep, arXiv: 1706.08726]
Analytic Continuation and Radius of Convergence
I an analytic continuation into thecomplex plane can be performed bychoosing x in CN (x) to be complex,i.e. x = αeiθ
I rational interpolants can exactlyreproduce polar singularities, thusextending the ‘radius of convergence’to the first non-polar singularity, e.g. abranch point
I even non-polar singularities may bewell approximated by poles and zerosof the rational fraction
I a rational fraction can have only onesheet in the complex plane - amany-sheeted function can only bereconstructed on a single sheet
BP