Path Integration in Statistical Field Theory:
from QM to Interacting Fermion Systems
Andreas Wipf
Theoretisch-Physikalisches Institut Friedrich-Schiller-University Jena
Methhods of Path Integration in Modern Physics 25.-31. August 2019
Andreas Wipf (TPI Jena) Path Integration in Statistical Field Theory
1 Introduction
2 Path Integral Approach to Systems in Equilibrium: Finite Number of DOF Canonical approach
Path integral formulation
3 Quantized Scalar Field at Finite Temperature
Lattice regularization of quantized scalar field theories Äquivalenz to classical spin systems
4 Fermionic Systems at Finite Temperature and Density Path Integral for Fermionic systems
Thermodynamic potentials of relativistic particles
5 Interacting Fermions
Interacting fermions in condensed matter systems Massless GN-model at Finite Density in Two Dimensions Interacting fermions at finite density ind=1+1
why do we discretize quantum (field) theories? 2 / 79
weakly coupled subsystems:perturbation theory if not:strongly coupled system
properties can only be explained by strong correlations of subsystems example of strongly coupled systems:
ultra-cold atoms in optical lattices high-temperature superconductors statistical systems near phase transitions strong interaction at low energies
Andreas Wipf (TPI Jena) Path Integration in Statistical Field Theory
theoretical approaches 3 / 79
exactly soluble models(large symmetry, QFT, TFT) approximations
mean field, strong coupling expansion, . . . restiction toeffective degrees of freedom
Born-Oppenheimer approximation, Landau-theory, . . . functional methods
Schwinger-Dyson equations
functional renormalization group equation numerical simulations
lattice field theories = particular classical spin systems
⇒powerful methods of statistical physics and stochastics
Quantum mechanical system in thermal equilibrium
HamiltonianHˆ :H 7→ H
system inthermal equilibriumwith heat bath canonical %ˆβ= 1
Zβ
Kˆ(β), Kˆ(β) = e−βHˆ, β= 1 kT normalizingpartition function
Zβ =trKˆ(β) expectation value of observableOˆin ensemble
hOiˆ β=tr %ˆβOˆ
Andreas Wipf (TPI Jena) Path Integration in Statistical Field Theory
thermodynamic potentials 5 / 79
inner and free energy
U=hHiˆ β=− ∂
∂β logZβ , Fβ=−kTlogZβ
⇒all thermodynamic potentials, entropyS=−∂TF, . . . specific heat
CV =hHˆ2iβ− hHiˆ 2β=−∂U
∂β >0 system of particles: specify Hilbert space andHˆ identical bosons: symmetric states
identical fermions: antisymmetric states traces on different Hilbert spaces
quantum statistics: canonical approach 6 / 79
Path integral for partition function in quantum mechanics
euclidean evolution operatorKˆ satisfiesdiffusiontype equation Kˆ(β) = e−βHˆ=⇒ d
dβ
Kˆ(β) =−HˆKˆ(β)
compare with time-evolution operator and Schrödinger equation U(tˆ ) = e−iH/ˆ ~=⇒ i~d
dtU(t) = ˆˆ HU(tˆ ) formally:U(tˆ =−i~β) = ˆK(β), imaginary time
~quantum fluctuations,kT thermal fluctuations
Andreas Wipf (TPI Jena) Path Integration in Statistical Field Theory
heat kernel 7 / 79
evaluate trace in position space
hq|e−βHˆ|q0i=K(β,q,q0) =⇒Zβ= Z
dq K(β,q,q)
“initial condition” for kernel: limβ→0K(β,q,q0) =δ(q,q0)
free particle in d dimensions (Brownian motion)
Hˆ0=−~2
2m∆=⇒K0(β,q,q0) = m 2π~2β
d/2
e−2~m2β(q0−q)2
HamiltonianHˆ = ˆH0+ ˆV bounded from below⇒ e−β( ˆH0+ ˆV) =s− lim
n→∞
e−βHˆ0/ne−βV/nˆ n
, Vˆ =V( ˆq)
derivation of path integral 8 / 79
insert for every identity1in e−βnHˆ0e−βnVˆ
1 e−βnHˆ0e−βnVˆ
1· · ·1 e−βnHˆ0e−βnVˆ the resolution1=R
dq|qihq| ⇒ K β,q0,q
= lim
n→∞
q0
e−βnHˆ0e−βnVˆn
q
= lim
n→∞
Z
dq1· · ·dqn−1 j=n−1
Y
j=0
qj+1
e−βnHˆ0e−βnVˆ qj
,
initial and final positionsq0=q andqn=q0 define smallε=~β/nand finally use
e−βnV( ˆq) qj
= qj
e−βnV(qj) qj+1
e−βnHˆ0 qj
= m 2π~ε
d/2
e−2~εm (qj+1−qj)2
Andreas Wipf (TPI Jena) Path Integration in Statistical Field Theory
derivation of path integral 9 / 79
discretized “path integral”
K(β,q0,q) = lim
n→∞
Z
dq1· · ·dqn−1
m 2π~ε
n/2
·exp
−ε
~
j=n−1
X
j=0
m 2
qj+1−qj
2
+V(qj)
divide interval[0,~β]intonsub-intervals of lengthε=~β/n consider pathq(τ)with sampling pointsq(τ=kε) =qk
s q
0 ε 2ε nε=~β
q=q0
q1
q2 qn=q′
interpretation as path integral 10 / 79
Riemann sum in exponent approximatesRiemann integral
SE[q] =
β
Z
0
dτm
2q˙2(τ)+V(q(τ)) SEisEuclidean action(∝action for imaginary time)
integration over all sampling pointsn→∞−→ formal path integralDq path integral with real and positive density
K(β,q0,q) =C
q(~β)=q0
Z
q(0)=q
Dq e−SE[q]/~
Andreas Wipf (TPI Jena) Path Integration in Statistical Field Theory
partition function as path integral 11 / 79
on diagonal = integration over all pathq→q
K(β,q,q) =hq|e−βHˆ|qi=C
q(~β)=q
Z
q(0)=q
Dq e−SE[q]/~
trace
tr e−βHˆ= Z
dqhq|e−βHˆ|qi=C I
q(0)=q(~β)
Dq e−SE[q]/~
partition functionZ(β) integral overall periodic pathswith period~β.
can construct well-defined Wiener-measure measure(differentable paths)=0
measure(continuous paths)=1
not only classical paths contribute 12 / 79
Andreas Wipf (TPI Jena) Path Integration in Statistical Field Theory
Mehler’s formula 13 / 79
exercise (Mehler formula)
show that the harmonic oscillator with Hamiltonian Hˆω=−~2
2m d2
dq2+mω2 2 q2 has heat kernel
Kω(β,q0,q) =
r mω
2π~sinh(~ωβ)expn
−mω 2~
(q2+q02)coth(~ωβ)− 2qq0 sinh(~ωβ)
o
equation of euclidean motionq¨=ω2qhas for givenq,q0the solution q(τ) =qcosh(ωτ) + q0−qcosh(ωβ)sinh(ωτ)
sinh(ωβ) action
S= m 2
Z β
0
( ˙q2+ω2q2) = mω 2 sinhωβ
(q2+q02)coshωβ−2qq0
second derivative
∂2S
∂q∂q0 =− mω sinh(ωβ) semiclassical formula exact for harmonic oscillator
K(β,q0,q) = s
− 1 2π
∂2S
∂q∂q0 e−S
yields above results for heat kernel diagonal elements
Kω(β,q,q) =
r mω
2π~sinh(~ωβ) exp (
−2mωq2
~
sinh2(~ωβ/2) sin(~ωβ)
)
Andreas Wipf (TPI Jena) Path Integration in Statistical Field Theory
spectrum of harmonic oscillator 15 / 79
partition function Zβ= 1
2 sinh(~ωβ/2) = e−~ωβ/2
1− e−~ωβ = e−~ωβ/2
∞
X
n=0
e−n~ωβ
evaluate trace withenergy eigenbasisofHˆ ⇒ Z(β) =tr e−βHˆ=hn|e−βHˆ|ni=X
n
e−βEn
comparison of two sums⇒ En=~ω
n+1
2
, n∈N
thermal correlation functions 16 / 79
position operator
q(t) =ˆ eitH/ˆ ~qˆe−itH/ˆ ~, q(0) = ˆˆ q
imaginary timet =−iτ⇒euclidean operator
qˆE(τ) = eτH/ˆ ~ˆqe−τH/ˆ ~, qˆE(0) = ˆq
correlations at different0≤τ1≤τ2≤ · · · ≤τn≤β in ensemble qˆE(τn)· · ·qˆE(τ1)
β≡ 1
Z(β) tr
e−βHˆqˆE(τn)· · ·qˆE(τ1) considerthermal two-point function(now we set~=1)
hqˆE(τ2) ˆqE(τ1)iβ= 1 Z(β)tr
e−(β−τ2) ˆHˆq e−(τ2−τ1) ˆHqˆ e−τ1Hˆ
Andreas Wipf (TPI Jena) Path Integration in Statistical Field Theory
thermal correlation functions 17 / 79
spectral decomposition:|niorthonormal eigenstates ofHˆ⇒ h. . .iβ= 1
Z(β) X
n
e−(β−τ2)Enhn|qˆ e−(τ2−τ1) ˆHq|niˆ e−τ1En
insert1=P
|mihm| ⇒ h. . .iβ= 1
Z(β) X
n,m
e−(β−τ2+τ1)Ene−(τ2−τ1)Emhn|ˆq|mihm|ˆq|ni
low temperatureβ→ ∞: contribution of excited states toP
n(. . .) exponentially suppressed,Z(β)→exp(−βE0)⇒
hqˆE(τ2) ˆqE(τ1)iβ β→∞−→ X
m≥0
e−(τ2−τ1)(Em−E0)|h0|ˆq|mi|2 likewise
hqˆE(τ)iβ−→ h0|ˆq|0i
connected correlation functions 18 / 79
connected two-point function
hqˆE(τ2) ˆqE(τ1)ic,β ≡ hqˆE(τ2) ˆqE(τ1)iβ− hˆqE(τ2)iβhˆqE(τ1)iβ term withm=0 inP
m(. . .)cancels⇒exponential decay withτ1−τ2:
β→∞lim hqˆE(τ2) ˆqE(τ1)ic,β =X
m≥1
e−(τ2−τ1)(Em−E0)|h0|ˆq|mi|2
energy gapE1−E0and matrix element|h0|q|1i|2from
hqˆE(τ2) ˆqE(τ1)ic,β→∞−→e−(E1−E0)(τ2−τ1)|h0|q|1i|ˆ 2, τ2−τ1→ ∞
Andreas Wipf (TPI Jena) Path Integration in Statistical Field Theory
path integral for thermal correlation function 19 / 79
for path-integral representation consider matrix elements Dq0|e−βHˆeτ2Hˆqˆe−τ2Hˆeτ1Hˆqˆe−τ1Hˆ|qE resolution of the identity andq|uiˆ =u|ui:
h. . .i= Z
dvduhq0|e−(β−τ2) ˆH|vivhv|e−(τ2−τ1) ˆH|uiuhu|e−τ1Hˆ|qi path integral representations each propagator (β > τ2> τ1):
sum over paths withq(0) =qandq(τ1) =u sum over paths withq(τ1) =uandq(τ2) =v sum over paths withq(τ2) =v andq(β) =q0 multiply with intermediate positionsq(τ1)andq(τ2)
Rdudv: path integral over all paths withq(0) =q andq(β) =q0
thermal correlation functions 20 / 79
insertion ofq(τ2)q(τ1in path integral hqˆE(τ2) ˆqE(τ1)iβ= 1
Z(β) I
Dqe−SE[q]q(τ2)q(τ1) similarly: thermaln−point correlation functions
hqˆE(τn)· · ·qˆE(τ1)iβ= 1 Z(β)
I
Dqe−SE[q]q(τn)· · ·q(τ1)
conclusion
there exist a path integral representation for all equilibrium quantities, e.g.
thermodynamic potentials, equation of state, correlation functions
Andreas Wipf (TPI Jena) Path Integration in Statistical Field Theory
real and imaginary time 21 / 79
real time: quantum mechanics action from mechanics
S= Z
dtm
2q˙2−V(q) real time path integral
hq0|e−itH/ˆ ~|qi
=C Z q(t)=q0
q(0)=q
Dq eiS[q]/~]
correlation functions h0|Tqˆ(t1) ˆq(t2)|0i
=C Z
Dq eiS[q]/~]q(t1)q(t2) oscillatory integrals
imaginary time: quantum statistics euclidean action
SE = Z
dτm
2q˙2+V(q) imaginary time path integral
hq0|e−βH/ˆ ~|qi
=C
Z q(~β)=q0
q(0)=q
Dq e−SE[q]/~]
correlation functions hqˆE(τ1) ˆqE(τ2)iβ
=C Z
Dq e−SE[q]/~]q(τ1)q(τ2) exponentially damped integrals
stochastic methods are required 22 / 79 numerical simulations: discrete (euclidean) time
system on time lattice =classical spin system Zβ = lim
n→∞
Z
dq1· · ·dqn
m 2π~ε
n/2
e−SE(q1,...,qn)/~ expectation values of observables
Z
dq1. . .dqnF(q1, . . . ,qn) high-dimensional integral (sometimesn=106required)
curse of dimension: analytical and numerical approaches do not work stochastic methods, e.g. Monte-Carloimportant sampling
what can be determined?
energies, transitions amplitudes and wave functions in QM potentials, phase transitions, condensates and critical exponents bound states, masses and structure functions in particle physics . . .
Andreas Wipf (TPI Jena) Path Integration in Statistical Field Theory
harmonic and anharmonic oscillators 23 / 79
m=1, µ=1, λ=0 m=1, µ=−3, λ=1
q
|ψ0(q)|2
Hˆ= pˆ2
2m +µˆq2+λqˆ4 Monte-Carlo simulation (Metropolis algorithm) square of the ground state wave function
parameters in units of lattice constantε A. Wipf, Lecture Notes Physics 864 (2013)
path integral for linear chain 24 / 79
exercise: harmonic chain
find free energy for periodic chain of coupled harmonic oscillators H= 1
2m
N
X
i=1
p2i +mω2 2
X
i
qi+1−qi2
, qi =qi+N
periodicq(τ)⇒may integrate by parts in LE = m
2 Z
dτ q˙2+ω2 qi+1−qi2 matrix notation
LE = m 2
Z
dτqT
− d2 dτ2+A
q, A=ω2 2δij−δi,j+1−δi,j−1 hint: non-negative eigenvalues and orthonormal eigenvectors ofA:
ωk =2ωsinπk
N and ek
Andreas Wipf (TPI Jena) Path Integration in Statistical Field Theory
free energy of linear chain 25 / 79 expandq(τ) =P
ck(τ)ek
LE =X
k
m 2
Z
dτ( ˙c2k+ω2kck2
Ndecoupled oscillatorswith frequenciesωk ⇒ hq|e−βHˆ|qi=Y
k
Kωk(β,qk,qk)
results for one-dimensional oscillator⇒ Zβ=Y
k
eβωk/2 eβωk−1 =Y
k
e−βωk/2
1−e−βωk, ωk =2ωsinπk N
free energy contains zero-point energy
Fβ= 1 2
X
k
~ωk+kTX
k
log 1− e−~ωk/kT
field theories 26 / 79
spin 0:scalar field(Higgs particle, inflaton,. . . ) spin 12:spinor field(electron, neutrinos, quarks, . . . )
spin 1:vector field(photon, W-bosons, Z-boson, gluons, . . . ) a quick way from quantum mechanics to quantized scalar field theory:
scalar fieldφ(t,x)satisfiesKlein-Gordon type equation(~=c =1) 2φ+V0(φ) =0
Lagrangian = integral of Lagrangian density over space L[φ] =
Z
space
dxL(φ, ∂µφ), L=1
2∂µφ∂µφ−V(φ)
Andreas Wipf (TPI Jena) Path Integration in Statistical Field Theory
field theories 27 / 79
momentum field, Legendre transform⇒Hamiltonian(fixed timet) π(x) = δL
δφ(˙ x) = ∂L
∂φ(˙ x)= ˙φ(x) H=
Z
dx πφ˙−L
= Z
dxH, H=1 2π2+1
2(∇φ)2+V(φ) free particle:V ∝φ2⇒Klein-Gordan2φ+m2φ=0
infinitely many dof: one at each space point
one of many possible regularizations:discretize space field theory onspace lattice:x =εnwithn∈Zd−1
φ(t,x)−→φx=εn(t) , Z
dx −→εd−1X
n
from field theory to point mechanics 28 / 79
N1
N2
ε
N1
N2
ε
N1
N2
ε
N1
N2
ε
N1
N2
ε
N1
N2
ε
N1
N2
ε
N1
N2
ε
N1
N2
ε
N1
N2
ε
N1
N2
ε
N1
N2
ε
N1
N2
ε
N1
N2
ε
N1
N2
ε
N1
N2
ε
N1
N2
ε
N1
N2
ε
N1
N2
ε
N1
N2
ε
N1
N2
ε
N1
N2
ε
N1
N2
ε
N1
N2
ε
N1
N2
ε
N1
N2
ε
N1
N2
ε
N1
N2
ε
N1
N2
ε
N1
N2
ε
N1
N2
ε
N1
N2
ε
N1
N2
ε
N1
N2
ε
N1
N2
ε
N1
N2
ε
e.g. periodic bc lattice constantε
# of lattice sitesN=Q Ni
linear extendsLi =εNi physical volumeV =εd−1N
finitehypercubic lattice in space x =εn with ni ∈ {1,2, . . . ,Ni} continuum fieldφ(x)→lattice fieldφx integral→Riemann sum
Z
dx −→εd−1X
n
derivative→difference quotient
∂φ(x)
∂xi
−→(∂iφ)x
example: symmetric “lattice derivative”
(∂iφ)x = φx+εei−φx−εei 2ε
Andreas Wipf (TPI Jena) Path Integration in Statistical Field Theory
from field theory to point mechanics 29 / 79
finite lattice→mechanical system with finite number of dof H=εd−1 X
x∈lattice
1 2πx2+1
2(∂φ)2x +V(φx) path integral quantizationknown
{φ0x} e−iH/ˆ ~
{φx}
=C Z Y
x
Dφx eiS[{φx}]/~ (formal) path integral over paths{φx(t)}in configuration space
φx(0) =φx and φx(t) =φ0x, ∀x =εn high-dimensionalquantum mechanical system with action
S[{φx}] = Z
dtεd−1X
x
1 2φ˙2x −1
2(∂φ)2x −V(φx)
quantum field theory at finite temperature 30 / 79
canonical partition function
Zβ=C I Y
x
Dφx e−SE[{φx}]/~, φx(τ) =φx(τ+~β) real euclidean action
SE[{φx}] = Z
dτ εd−1X
x
1 2
φ˙2x +1
2(∂φ)2x +V(φx) path-integral well-defined afterdiscretization of “time”
convenient:same lattice constantεin time and spatial directions replaceτ∈[0,~β]−→τ∈ {ε,2ε, . . . ,N0ε}withN0ε=~β lattice sites(xµ) = (τ,x) = (εnµ)withnµ ∈ {1,2, . . . ,Nµ}
⇒d-dimensionalhypercubic space-time lattice
Andreas Wipf (TPI Jena) Path Integration in Statistical Field Theory
space-time lattice 31 / 79
lattice fieldφxdefined on sites ofspace-time latticeΛ
lattice regularization 32 / 79
d-dimensional Euclid’sche space-time→latticeΛ, sitesx ∈Λ continuous fieldφ(x)→lattice fieldφx,x ∈Λ
finite lattice: extend in directionµ:Lµ=εNµ
finite temperature:L1=· · ·=Ld−1L0≡β=1/(kT) scalar fieldperiodic in imaginary time direction
φx=(x0+εN0,x)=φx=(x0,x)=⇒temperature-dependence typically: also periodic in spatial directions
⇒identificationxµ∼xµ+Lµ(torus) space-time volumeV=εdN1N2· · ·Nd
some freedom inchoice of lattice derivative(use symmetries)
Andreas Wipf (TPI Jena) Path Integration in Statistical Field Theory
space-time lattice 33 / 79
dimensionless fields and couplings (~=c=1)
natural units~=c=1⇒all units in powers of lengthL dimensionless action (unitL0)
SE = Z
ddx1
2(∂φ)2+X
a
λpha φa
Rddx(∂φ)2dimensionless⇒[∂φ] =L−d/2⇒[φ] =L1−d/2 λpha R
ddxφadimensionless⇒[λpha] =L−d−a+ad/2 in particularλph2 ∝m2⇒[m] =L−1
4 space-time dimensions⇒λph4 dimensionless
dimensionless lattice field and lattice constants (x =εn) φx =ε1−d/2φn, λpha =ε−d−a+ad/2λa
dimensions 34 / 79
lattice action with dimensionful quantities SLph=εdX
x
1 2
φx+εeµ−φx−εeµ 2ε
2
+X
a
λphaφax
!
⇒lattice action with dimensionless quantities SL=X
n
1
2 φn+eµ−φx−eµ2
+X
a
λaφan
partition function
Zβ =C
Z N0N1···
Y
n=1
dφne−SL[{φn}]
finite-dimensional well-defined integral(lattice regularization) lattice formulation without any dimensionful quantity
processor knows numbers, not units!
Andreas Wipf (TPI Jena) Path Integration in Statistical Field Theory
renormalization 35 / 79
merely lettingε→0: no meaningful continuum limit λamust be changed asε→0
condition:dimensionful observables approach well-defined finite limits existence of suchcontinuum limitnot guaranteed
example: consider correlation length in hφ(n)φ(m)ic ∝ e−|n−m|/ξ, 1
ξ =m=dimnsionless mass ξdepends on dimensionless couplingsξ=ξ(λa)
relates to (given) dimensionful massmph =1/(εξ)⇒ε mph from experiment,ξ(λa)measured on lattice
renormalization:keepmph (and further observables) fixed⇒λa
renormalizable field theories 36 / 79
extend of physical objectsseparation of lattice points extend of physical objectsbox size
conditions(scaling window) small discretization effectsξ1 small finite size effectsξNµ
strict continuum limit:ξ→ ∞
2’nd order phase transition required in system withNspatial → ∞ theory renormalizable:only a small number ofλamust be tuned relevantrenormalizable field theories
non-Abelian gauge theories ind ≤4 scalar field theories ind<4 four-Fermi theories ind≤3 non-linear sigma-models ind≤3 Einstein-gravity ind≤4 (???)
Andreas Wipf (TPI Jena) Path Integration in Statistical Field Theory
simulation 37 / 79
input in simulations:only a few observables (masses) simulate with stochastic algorithms in scaling window repeat simulations with same observables but decreasingε output:many (dimensionful) observables
extrapolate toε→0
if theoryrenormalizable: converge to a continuum limit asε→0 finite temperature:N0given,εfrom matching to observable⇒β=εN0.
⇒temperature dependence of free energy
condensates pressure, densities
free energy of two static charges (confinement) phase diagram
screening effects
correlations in heat bath, . . .
lattice field theory as spin model 38 / 79
path integral for finite temperature QFT= classical spin model
no non-commutative operators, instead: path or functional integration over fields scalar field:
assignφn∈Rto each lattice site sigma models:
φn∈Sphere
discretespin models:
φn∈ discrete group example: Potts-model:
φn∈Zq
figures: 3−state Potts-type model
Andreas Wipf (TPI Jena) Path Integration in Statistical Field Theory
relativistic fermions 39 / 79
electron, muon, quarks, . . . are described by 4-component spinor fieldψα(x) metric tensor in Minkowski space-time
(ηµν) =diag(1,−1,−1,−1) 4×4 gamma-matrices
γ0, . . . , γ3, {γµ, γν}=2ηµν1 covariantDirac equation for free massive fermions
i∂/−m
ψ(x) =0, ∂/=γµ∂µ
Euler-Lagrange equation of invariant action S=
Z
d4xψ(i¯ ∂/−m)ψ, ψ¯=ψ†γ0=⇒πψ=−iψ†
quantization of Dirac field 40 / 79
quantization:ψ(x)→ψ(xˆ )
satisfiesanti-commutation relation
{ψˆα(t,x),ψˆβ†(t,y)}=δαβδ(x −y) Hamilton operator:β=γ0, α=γ0γ:
Hˆ = Z
dx ψˆ†(x)(ˆhψ)(ˆ x), ˆh=iα· ∇+mβ derive path integral representation ofpartition function
Zβ=tr e−βHˆ leads to imaginary time path integral replacet→ −iτ and
γ0E=γ0 and γiE=iγi
Andreas Wipf (TPI Jena) Path Integration in Statistical Field Theory
path integral for partition function 41 / 79 ACR with euclidean metric
{γµE, γEν}=2δµν1, γEµhermitean lattice regularization(drop index E)
space-timeR4→finite (hypercubic) latticeΛ continuum fieldψ(x)onR4→lattice fieldψx expected path integral
Zβ=trrege−βHˆ=C I Y
α,x∈Λ
dψα,x† dψα,x e−SL[ψ,ψ†]
integration overanti-periodic fields(ACR forψ, see below) ψx(τ+β) =−ψx(τ), also on time lattice SLsome lattice regularization of
SE= Z
ddxψ†(i∂/+im)ψ
Grassmann variables 42 / 79
quantized scalar field obey equal-time CR φ(t,ˆ x),φ(tˆ ,y)
=0, x 6=y
⇒commuting fields in path integral
[φ(x), φ(y)] =0, ∀x,y quantized fermion field obey equal-time ACR
ψˆα(t,x),ψˆ†β(t,y) =0, x 6=y ,
⇒anti-commuting fields in path integral
ψα(x), ψβ†(y) =0, ∀x,y
variables{ψα,n, ψ†α,n}in fermion path integral:Grassmann variables
Andreas Wipf (TPI Jena) Path Integration in Statistical Field Theory
Graussian integrals 43 / 79
free theories have quadratic action
Gaussian integralswithA=AT positive matrix;exercise⇒ Z N
Y
n=1
dφn exp
−1 2
XφnAnmφm
= (2π)n/2
√detA what do we get for fermions?
simplify notation:ψα,n≡ηi andψα,n† ≡η¯i withi =1, . . . ,m objects{ηi,η¯i}form complex Grassmann algebra:
{ηi, ηj}={η¯i,η¯j}={ηi,η¯j}=0=⇒η2i = ¯ηi2=0
Grassmann integration defined by (a, b ∈ C )
Z
linear, Z
dηi(a+bηi) =b, Z
dη¯i(a+bη¯i) =b
Gaussian integrals with Grassmann variables 44 / 79
Grassmann integrals with
D¯ηDη ≡
m
Y
i=1
dη¯idηi
free fermions⇒Gaussian Grassmann integral Z =
Z
D¯ηDη e−¯ηAη, ηAη¯ =X
i,j
¯ ηiAijηj
expand exponential function:R
D¯ηDη ηAη¯ k
=0 fork 6=m remaining contribution (useη¯i2=0)
1 n!
Z
DηDη¯ (¯ηAη)m = Z
D¯ηDη X
i1,...,im
(¯η1A1i1ηi1)· · ·(¯ηmAmimηim)
= Z
D¯ηDηY
i
(¯ηiηi) X
i1,...,im
εi1...imA1i1· · ·Amim
= (−1)m Z
Y
i
(d¯ηiη¯idηiηi)detA= (−1)mdetA
Andreas Wipf (TPI Jena) Path Integration in Statistical Field Theory
generating function 45 / 79
simple formula
Z
D¯ηDη e−¯ηAη=detA
generalization:generating function Z( ¯α, α) =
Z
D¯ηDη e−¯ηAη+ ¯αη+ ¯ηα= e−αA¯ −1α detA expand in powers ofα,¯ α⇒
h¯ηiηji ≡ 1 Z
Z
D¯ηDη e−¯ηAηη¯iηj = (A−1)ij application to Dirac fields: abovepartition function
Zβ= I
DψDψ¯ e−SL, DψDψ¯ =Y
α,n
dψ†α,ndψα,n
Graussian integrals for fermions and bosons 46 / 79
dimensionless field and couplings SL=X
n∈Λ
ψn†(i∂/nm+imδnm)ψn=X
n
ψ¯nDnmψm
lattice partition function
Zβ=CdetD expectation value in canonical ensemble
hAiˆ β= 1 Zβ
I
DψDψ¯ A( ¯ψ, ψ)e−SL(ψ,ψ)¯ formula forcomplex scalar field
Zβ= I
DφDφ¯exp
−Xφ¯mCmnφn
∝ 1
detC boson fields: periodic in imaginary time
fermion fields: anti-periodic in imaginary time
Andreas Wipf (TPI Jena) Path Integration in Statistical Field Theory
Thermodynamic potentials for Gas or relativistic particles 47 / 79
neutral scalars (+: periodic bc) SE = 1
2 Z
φ(−∆+m2)φ=⇒Fβ=kT
2 log det+(−∆+m2) +. . . Dirac fermions (−: anti-periodic bc)
SE = Z
ψ†(i∂/+im)ψ=⇒Fβ=−2kTlog det−(−∆+m2) +. . .
exercise
Try to prove the results for fermions (including sign and overall factor)
zeta-function for second order operator A > 0
ζA(s) =X
n
λ−sn , eigenvaluesλn
ζ−function regularization 48 / 79
absolute convergent series in half-plane<(s)>d/2
meromorphic analytic continuation, analytic in neighborhood ofs=0 definesζ−function regularized determinant Dowker, Hawking
log detA=trlogA=X
logλn=−dζA(s) ds
s=0
correct for matrices Mellin transformations
Z ∞
0
dt ts−1e−tλ= Γ(s)λ−s
⇒relation to heat kernel ζA(s) =X
n
1 Γ(s)
Z ∞
0
dt ts−1e−tλn = 1 Γ(s)
Z ∞
0
dt ts−1tr e−tA
Andreas Wipf (TPI Jena) Path Integration in Statistical Field Theory
heat kernel 49 / 79
coordinate representation ζA(s) = 1
Γ(s) Z ∞
0
dt ts−1Z
dx K(t;x,x), K(t) = e−tA
heat kernel ofA=−∆+m2oncylinder[0, β]×Rd−1 K±(t;x,x0) = e−m2t
(4πt)d/2 X
n∈Z
(±1)ne−{(τ−τ0+nβ)2+(x−x0)2}/4t
integrate over diagonal elements ζA±(s) = βV
(4π)d/2Γ(s) Z
dt ts−1−d/2e−m2t
∞
X
n=−∞
(±)ne−n2β2/4t
Jacobitheta function
integral representation ofKelvin functions Z ∞
0
dt tae−bt−c/t =2c b
(a+1)/2
Ka+1
2√
bc
⇒series representation; ind=4 ζA±(s) = βV
16π2 m4−2s
Γ(s) Γ(s−2) +4
∞
X
1
(±)n nmβ
2 s−2
K2−s(nmβ)
!
identities
Γ(s−2)
Γ(s) = 1
(s−1)(s−2) and 1
Γ(s) =s+O(s2)
Andreas Wipf (TPI Jena) Path Integration in Statistical Field Theory
derivative ats=0⇒ Fβ±=−m4VC±
128π2
3−2 logm2
µ2 +64 X
n=1,2...
(±)nK2(nmβ) (nmβ)2
real scalarsC+=1, complex fermionsC−=−4 well-known results for massless particlesK2(x)∼2/x2
m→0lim f+(β) =−π2
90T4 , lim
m→0f−(β) =− 2 45π2T4
questions
Why is there a relative factor of 4? What is the free energies of complex scalars, Majorana fermions and photons. What is free energy of complex fermions ind dimensions?
Interacting relativistic fermions 52 / 79
condensed matter systems ind =2+1 tight binding approximation for small excitation energies
honeycomb lattice for graphen (GN):
2 atoms in every cell, 2 Dirac points
⇒4-component spinor field
interaction-driven transition metal↔insolator long rang order: AF, CDW, . . .
interacting fermions(symmetries!) condensed matter systems ind =1+1
conducting polymers (Trans- and
Cis-polyacetylen) Su, Schrieffer, Heeger
quasi-one-dimensional inhomogeneous superconductor Mertsching, Fischbeck
relativistic dispersion-relationsfor electronic excitations on
honeycomb lattice
from Castro Neto et al.
Andreas Wipf (TPI Jena) Path Integration in Statistical Field Theory
interacting fermions in 1+1 and 2+1 dimensions 53 / 79
irreducible spinor in two and three dimensions has 2 components Nf species (flavours) of spinors,Ψ = (ψ1, . . . , ψNf)
relativistic fermions
LGN= ¯Ψi∂Ψ +/ imΨΨ +¯ LInt(Ψ,Ψ),¯ e.g.ΨΨ =¯ Xψ¯iψi
parity invariant models LInt=gGN2
2Nf( ¯ΨΨ)( ¯ΨΨ) scalar-scalar,Gross-Neveu LInt=−g2Th
2Nf( ¯ΨγµΨ)( ¯ΨγµΨ) vector-vector,Thirring LInt= gPS2
2Nf( ¯Ψγ∗Ψ)( ¯Ψγ∗Ψ) pseudoscalar-pseudoscalar in even dimensionsγ∗∝Q
γµ
Hubbard-Stratonovich trickwith scalar, vector and pseudscalar field
Relativistic four-Fermi Theories 54 / 79
combinations thereof ind =4
non-renormalizableFermi theoryof weak interaction
effective models for chiral phase transition in QCD (Jona Lasino) 2 spacetime dimensions:[g] =L0
massless ThM:soluble Thirring
massless ThM incurved space withµ:soluble Sachs+AW, . . .
GNM:asymptotically free, integrable Gross-Neveu, Coleman, . . .
3 spacetime dimensions:[g] =Ls not renormalizable in PT
renormalizable in large-Nexpansion Gawedzki, Kupiainen; Park, Rosenstein, Warr
interacting UV fixed point→asymptotically safe de Veiga; da Calen; Gies, Janssen
can exhibit parity breaking at lowT lattice theories:
generically: sign problem even forµ=0
partial solution of sign problem Schmidt, Wellegehausen, Lenz, AW
Andreas Wipf (TPI Jena) Path Integration in Statistical Field Theory
Masselss GN-model at finite density in two dimensions 55 / 79
with J. Lenz, L. Panullo, M. Wagner and B. Wellegehausen
GN shows breaking of discrete chiral symmetry order parameteriΣ =hΨΨi¯
ψa→iγ∗ψa, ψ¯a→iψ¯aγ∗=⇒iΣ =hΨΨi → −h¯ ΨΨi¯ equivalent formulation with auxiliary scalar fieldHubbard-Stratonovich transformation
LGN=Lσ= ¯Ψ iD⊗1Nf
Ψ +Nf 2g( ¯ΨΨ)2 Lσ= ¯Ψ iD⊗1Nf
Ψ +λNfσ2, D=∂/−σ6=D†
chemical potential for fermion number charge 56 / 79
conservedfermion charge Q= Z
spacedxj0= Z
spacedxψ†ψ partition function ofgrand canonical ensemble
Zβ,µ=tr e−β( ˆH−µQ)ˆ, functional integral with aboveLσwherein
D=∂/+σ+µγ0
expectation values
hOi= 1 Zβ,µ
Z
DψDψDσ¯ e−SσO
Andreas Wipf (TPI Jena) Path Integration in Statistical Field Theory
chemical potential for fermion charge 57 / 79
fermion integral in Zβ,µ=
Z
DψDψDσ¯ e−Sσ[σ,ψ,ψ]¯ = Z
Dσe−NfSeff[σ]
Nf fermion species couple identically to auxiliary field⇒ det iD⊗1
= (detiD)Nf ψanti-periodic in imaginary time,σperiodic effective action after fermion integral
Seff=λ Z
d2xσ2−log(detiD)
Ward identity(lattice regularization) 1
Zβ,µ
Z
DψDψ¯Dσ d dσ(x)
e−S[σ,ψ,ψ†]
=−D dS dσ(x)
E
=0
homogeneous phases 58 / 79
exact relation
Σ≡ −ihψ(x¯ )ψ(x)i= Nf
g2hσ(x)i forNf→ ∞saddle point (steepest descend) approximation
Zβ,µ= Z
Dσe−NfSeff][σ]N−→f→∞ e−NfminSeff[σ]
translation invariance⇒minimizingσconstant:Seff= (NfβL)Ueff Ueff= σ2
4π
logσ2 σ0 −1
−1 π
Z ∞
0
dp p2 εp
1
1+ eβ(εp+µ) + 1 1+eβ(εp−µ)
one-particle energiesp=p p2+σ2 IR-scaleσ0=hσiT=µ=0
Andreas Wipf (TPI Jena) Path Integration in Statistical Field Theory
condensate in the(µ,T)-plane 59 / 79
'data.dat' u 1:2:3
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
µ/σ0 0
0.1 0.2 0.3 0.4 0.5 0.6 0.7
T/σ0
0 0.2 0.4 0.6 0.8 1
symmetric phase for largeT, µ
homogeneously broken phase for smallT, µ Wolff, Barducci
special points:(Tc, µ) = (eγ/π,0),(T,µc) = (0,1/√ 2) Lifschitz-Punkt bei(T, µ0)≈(0.608,0.318)
is homogeneity assumption really justified? 60 / 79
possible QCD-phase diagram
crystalline LOFF phase (color superconductive phase)?
problem:µ6=0⇒complex fermion determinant/ largeµbeyond reach in simulations
are there inhomogeneouscrystallic phasesin model systems?
Andreas Wipf (TPI Jena) Path Integration in Statistical Field Theory
space-dependent condensate in GN model 61 / 79
discreteεnenergies of Dirac Hamiltonian on[0,L]
hσ=γ0γ1∂x +γ0σ(x)
hidden supersymmetry h2σ=−d2
dx2 +σ2(x)−γ1σ0(x) = AA† 0 0 A†A
!
, A=−d dx +σ renormalization: fix (constant) condensateσ0atµ=T =0
introduce constantcompanion field
¯ σ2= 1
L Z
dxσ2(x) constantσ⇒¯σ=σ