Skip to content

Finite-Temperature QFT and Euclidean Time

Thermal equilibrium is encoded by the Gibbs state ρβ=eβH/Z\rho_\beta = e^{-\beta H}/Z. Wick‑rotate tiτt\to -i\tau: eiHt/eHτ/e^{-iHt/\hbar}\mapsto e^{-H\tau/\hbar}. With a trace, Z(β)=TrU(iβ)Z(\beta)=\mathrm{Tr} \, U(-i\hbar\beta), the Euclidean time direction gets glued into a circle of circumference β=/(kBT)\hbar\beta=\hbar/(k_B T). The KMS condition characterizes equilibrium and implies (anti)periodicity of Euclidean correlators; this is exactly the boundary condition for a classical Euclidean statistical field theory on Sβ1×RdS^1_{\hbar\beta}\times \mathbb{R}^d.

1. Thermal states and the partition function

Section titled “1. Thermal states and the partition function”

For a QFT with Hamiltonian HH, thermal equilibrium at temperature TT (inverse temperature β=1/(kBT)\beta=1/(k_B T)) is described by the Gibbs state

ρβ=eβHZ(β),Z(β)=TreβH.\rho_\beta=\frac{e^{-\beta H}}{Z(\beta)},\qquad Z(\beta)=\mathrm{Tr}\,e^{-\beta H}.

Thermal expectation of an observable AA is Aβ=Tr(ρβA)\langle A\rangle_\beta=\mathrm{Tr}(\rho_\beta A).

Two remarks:

  • Trace = sum over microstates. In quantum theory, “sum over all microstates” is the trace over Hilbert space. It ensures basis‑independence and correct normalization.
  • From real to imaginary time. Real‑time evolution U(t)=eiHt/U(t)=e^{-iHt/\hbar} becomes Euclidean evolution U(iτ)=eHτ/U(-i\tau)=e^{-H\tau/\hbar} after a Wick rotation t=iτt=-i\tau.

2. The KMS condition: the fingerprint of equilibrium

Section titled “2. The KMS condition: the fingerprint of equilibrium”

Let A,BA,B be Heisenberg operators, A(t)=eiHtAeiHtA(t)=e^{\frac{i}{\hbar}Ht} A e^{-\frac{i}{\hbar}Ht}. The Kubo–Martin–Schwinger (KMS) relation states (bosonic version)

  A(t)B(0)β=B(0)A(t+iβ)β.  \boxed{\;\langle A(t) B(0)\rangle_\beta = \langle B(0) A(t+i\hbar\beta)\rangle_\beta.\;}

For fermionic operators, there is a minus sign on the right (graded KMS).

Derivation (one line).

A(t)Bβ=1ZTr ⁣(eβHeiHtAeiHtB)=1ZTr ⁣(A(t+iβ)eβHB)=BA(t+iβ)β,\begin{aligned} \langle A(t)B\rangle_\beta &=\frac{1}{Z}\mathrm{Tr}\!\big(e^{-\beta H} e^{\frac{i}{\hbar}Ht} A e^{-\frac{i}{\hbar}Ht} B\big)\\ &=\frac{1}{Z}\mathrm{Tr}\!\big(A(t+i\hbar\beta)\,e^{-\beta H} B\big) =\langle B\,A(t+i\hbar\beta)\rangle_\beta, \end{aligned}

where we used eβHA(t)=A(t+iβ)eβHe^{-\beta H} A(t)=A(t+i\hbar\beta)\,e^{-\beta H} and cyclicity of the trace.

Analytic strip. More conceptually, there exists a function FA,B(z)F_{A,B}(z) holomorphic on 0<Imz<β0<\mathrm{Im}\,z<\hbar\beta whose boundary values give the two correlators above. “Analytic in a strip + iβi\hbar\beta shift” is the mathematical fingerprint of thermal equilibrium.

3. From KMS to Euclidean (anti)periodicity

Section titled “3. From KMS to Euclidean (anti)periodicity”

Define Euclidean operators by analytic continuation AE(τ):=A(t=iτ)A_E(\tau):=A(t=-i\tau). The Euclidean two‑point function is the τ\tau-ordered thermal correlator

GAB(E)(τ):=TτAE(τ)BE(0)β.G^{(E)}_{AB}(\tau) := \langle T_\tau \, A_E(\tau) B_E(0)\rangle_\beta.

Insert t=iτt=-i\tau into KMS and use the definition of TτT_\tau. For 0<τ<β0<\tau<\hbar\beta,

  GAB(E)(τ+β)=±GAB(E)(τ)  \boxed{\;G^{(E)}_{AB}(\tau+\hbar\beta)=\pm\,G^{(E)}_{AB}(\tau)\;}

with ++ for bosons and for fermions. Thus all Euclidean nn-point functions are (anti)periodic with period β\hbar\beta.

Interpretation: The correlators depend only on τmodβ\tau\bmod \hbar\beta. The most economical realization is to place the theory on a time circle Sβ1S^1_{\hbar\beta}. That circle’s circumference is β=/(kBT)\hbar\beta=\hbar/(k_B T).

4. Why Euclidean time has period β\hbar\beta: trace = closed time

Section titled “4. Why Euclidean time has period ℏβ\hbar\betaℏβ: trace = closed time”

Start with the partition function

Z(β)=TreβH.Z(\beta)=\mathrm{Tr}\,e^{-\beta H}.

Use eβH=U(iβ)e^{-\beta H}=U(-i\hbar\beta) to write

Z(β)=TrU(iβ).Z(\beta)=\mathrm{Tr}\,U(-i\hbar\beta).

In the path‑integral representation the trace glues the initial Euclidean time slice to the final one; the Euclidean time coordinate τ\tau is therefore compact with circumference β\hbar\beta.

  • Bosons: periodic around the circle.
  • Fermions: anti‑periodic (due to the graded nature of the trace/KMS).

Expanding along S1S^1 yields Matsubara frequencies

ωnbos=2πnβ,ωnferm=(2n+1)πβ.\omega_n^{\rm bos}=\frac{2\pi n}{\hbar\beta},\qquad \omega_n^{\rm ferm}=\frac{(2n+1)\pi}{\hbar\beta}.

5. From finite‑TT QFT to a classical statistical field theory

Section titled “5. From finite‑TTT QFT to a classical statistical field theory”

The Euclidean functional integral of the finite‑TT QFT is

Z(β)=(anti)periodic ⁣ ⁣DΦ  exp ⁣[1SE[Φ]],τ[0,β),  \boxed{\quad Z(\beta) = \int_{\text{(anti)periodic}}\!\!\mathcal{D}\Phi\; \exp\!\left[-\frac{1}{\hbar}\,S_E[\Phi]\right], \qquad \tau\in[0,\hbar\beta),\;}

which is precisely the partition function of a classical Euclidean statistical field theory on Sβ1×RdS^1_{\hbar\beta}\times\mathbb{R}^d with weight eSE/e^{-S_E/\hbar}.

Discretize Euclidean time with spacing aτa_\tau and define the transfer matrix

T:=eaτH/.T := e^{-a_\tau H/\hbar}.

Then

Z(β)=TrTNτ,Nτ=βaτ,Z(\beta)=\mathrm{Tr}\,T^{N_\tau},\qquad N_\tau=\frac{\hbar\beta}{a_\tau},

which is literally a classical partition sum over configurations on a (d+1)(d+1)-dimensional lattice with local Boltzmann weights. This is the foundation of finite‑temperature lattice QFT.

Because non‑zero Matsubara modes have gaps 2πn/β\sim 2\pi n/\hbar\beta, at high TT one can integrate them out. For static observables the effective theory becomes a dd-dimensional classical field theory for the Matsubara zero modes.

6. Vacuum wave functional from Euclidean projection

Section titled “6. Vacuum wave functional from Euclidean projection”

The ground‑state wave functional Ψ0[ϕ]ϕ0\Psi_0[\phi]\equiv \langle \phi|0\rangle admits a Euclidean representation:

Ψ0[ϕ(x)]    τ<0ϕ(τ=0,x)=ϕ(x) ⁣ ⁣Dϕ  eSE[ϕ]/.\Psi_0[\phi(\mathbf x)] \;\propto\; \int_{\substack{\tau<0\\ \phi(\tau=0,\mathbf x)=\phi(\mathbf x)}}\!\!\mathcal D\phi\;e^{-S_E[\phi]/\hbar}.

Reason: the Euclidean kernel KE=ϕ2eHTϕ1K_E=\langle \phi_2|e^{-HT}|\phi_1\rangle admits a spectral sum neEnTΨn[ϕ2]Ψn[ϕ1]\sum_n e^{-E_nT}\Psi_n[\phi_2]\Psi_n^*[\phi_1]. As TT\to\infty, the factor eEnTe^{-E_nT} projects onto the ground state, leaving the path integral on a half‑space with boundary value ϕ(τ=0)=ϕ\phi(\tau=0)=\phi.

  • Fermions: anti‑periodic b.c. on Sβ1S^1_{\hbar\beta} (graded KMS / Grassmann trace).
  • Chemical potential: in the grand canonical ensemble, HHμQH\to H-\mu Q. KMS twists the boundary condition by a phase: Φ(τ+β)=eβμqΦ(τ)\Phi(\tau+\hbar\beta)=\mp\,e^{\beta\mu q}\Phi(\tau) (minus for fermions). Geometrically, a flat U(1)U(1) Wilson line along the time circle.
  • Gauge theories: include gauge fixing and ghosts; reflection positivity holds at the level of gauge‑invariant observables.
  • Sign problems: if SES_E is not real and bounded below (e.g., finite density or θ\theta-terms), the probabilistic interpretation as a classical ensemble may fail (Monte Carlo obstruction).
  • Geometry link (black holes): near a static horizon, the Euclidean metric looks like a cone; removing the conical singularity fixes the time circle’s period 2π/κ2\pi/\kappa, giving Hawking’s T=κ/(2πkB)T=\hbar\kappa/(2\pi k_B).
  • KMS condition \Rightarrow equilibrium correlators are analytic in a strip and related by an iβi\hbar\beta shift.
  • Wick rotation + trace \Rightarrow Euclidean time is a circle of circumference β\hbar\beta; bosons periodic, fermions anti‑periodic; Matsubara modes follow.
  • Functional integral \Rightarrow a finite‑TT QFT is equivalent to a classical Euclidean statistical field theory on Sβ1×RdS^1_{\hbar\beta}\times\mathbb{R}^d.
  • Transfer matrix / OS reconstruction \Rightarrow this is not a coincidence but a structural, (semi‑)rigorous equivalence under standard assumptions.

We keep \hbar and kBk_B explicit. In “natural units”, set =kB=1\hbar=k_B=1, so the thermal circle has circumference β=1/T\beta=1/T.