3.4.1 Wick Rotation#

Prompts

  • What is the maximum-entropy principle, and why does it force the thermal occupation probabilities to take the Boltzmann form \(p_n \propto \mathrm{e}^{-\beta E_n}\)?

  • Why is the partition function \(Z(\beta)\) the only quantity needed in equilibrium thermodynamics? How do thermal observables emerge from its derivatives?

  • Why are real-time evolution \(\mathrm{e}^{-\mathrm{i}\hat{H}t/\hbar}\) and the thermal weight \(\mathrm{e}^{-\beta\hat{H}}\) the same operator under \(t = -\mathrm{i}\hbar\beta\)? What does this identification motivate?

  • Why does the Wick rotation turn the oscillating weight \(\mathrm{e}^{\mathrm{i}S/\hbar}\) into a real positive weight \(\mathrm{e}^{-S_E/\hbar}\), and why does the potential pick up a \(+V\) in the Euclidean action?

  • Why does taking the trace force imaginary-time paths to be periodic with period \(\hbar\beta\)? In what sense is imaginary time a circle whose circumference is set by temperature?

Lecture Notes#

Overview#

Quantum mechanics and statistical mechanics look like two unrelated theories. Quantum amplitudes carry oscillating phases \(\mathrm{e}^{-\mathrm{i}\hat{H}t/\hbar}\); thermal averages carry real Boltzmann weights \(\mathrm{e}^{-\beta\hat{H}}\). Yet both feature the same operator \(\hat{H}\) inside an exponential. This lesson traces that coincidence to its source. Starting from the partition function of statistical mechanics, we recognize \(\mathrm{e}^{-\beta\hat{H}}\) as the same evolution operator we already met in §3.2, with imaginary time \(\tau\) playing the role of inverse temperature. The substitution \(t \to -\mathrm{i}\tau\) — the Wick rotation — turns the oscillating real-time path integral into a positive-measure Euclidean path integral, and the trace forces the imaginary-time direction to close into a circle of circumference \(\hbar\beta\). Time, in this language, is temperature.

Recap: Statistical Mechanics in One Page#

Take a quantum system with Hamiltonian \(\hat{H}\) and energy eigenstates \(\hat{H}\vert n\rangle = E_n\vert n\rangle\). Place it in contact with a heat bath at temperature \(T\). What is the probability \(p_n\) of finding the system in state \(\vert n\rangle\)?

The bath knows nothing about the system except its average energy \(U = \langle E\rangle\), so \(p_n\) should be the least biased distribution consistent with that single constraint. Maximizing the Shannon entropy \(S = -k_B\sum_n p_n\,\ln p_n\) subject to \(\sum_n p_n = 1\) and \(\sum_n p_n E_n = U\) yields the Boltzmann distribution

(116)#\[ p_n \;=\; \frac{\mathrm{e}^{-\beta E_n}}{Z(\beta)},\qquad \beta \;\equiv\; \frac{1}{k_B T}, \]

where the Lagrange multiplier \(\beta\) acquires its physical meaning as inverse temperature. (Detailed derivation in §6.1.2 Entropy.)

Partition Function is All You Need#

Partition function

The normalization defines the partition function:

(117)#\[ Z(\beta) \;=\; \sum_n \mathrm{e}^{-\beta E_n}. \]

\(Z(\beta)\) is the central object of equilibrium statistical mechanics: every thermodynamic observable follows from its derivatives,

(118)#\[\begin{split} \begin{aligned} \langle E\rangle &\;=\; -\frac{\partial \ln Z}{\partial \beta},\\ F &\;=\; -k_B T\,\ln Z,\\ S &\;=\; -\frac{\partial F}{\partial T},\\ C &\;=\; \frac{\partial \langle E\rangle}{\partial T}. \end{aligned} \end{split}\]

In this sense \(Z(\beta)\) is the generating function of statistical mechanics. To do equilibrium thermodynamics is to compute \(Z\).

From Spectral Sum to Trace#

Equation (117) is written in the energy eigenbasis. Rewrite it without referring to \(\{\vert n\rangle\}\) at all. The sum of diagonal matrix elements is a trace:

(119)#\[ Z(\beta) \;=\; \sum_n\langle n\vert \mathrm{e}^{-\beta\hat{H}}\vert n\rangle \;=\; \mathrm{Tr}\,\mathrm{e}^{-\beta\hat{H}}. \]

The same expression now makes sense in any basis. In the position basis,

(120)#\[ Z(\beta) \;=\; \int \mathrm{d}x\;\langle x\vert \mathrm{e}^{-\beta\hat{H}}\vert x\rangle. \]

This rewriting innocuously moves \(\beta\) into the exponent of an operator. But that operator looks oddly familiar.

From Quantum Evolution to Thermal Averaging#

Compare two operators we have met:

Object

Where it appears

Physics

\(\hat{U}(t) \;=\; \mathrm{e}^{-\mathrm{i}\hat{H}t/\hbar}\)

Real-time evolution (§3.2)

Quantum amplitudes

\(\hat{\rho}(\beta) \;\propto\; \mathrm{e}^{-\beta\hat{H}}\)

Thermal density matrix

Statistical averages

They differ only in the prefactor of \(\hat{H}\): \(-\mathrm{i}t/\hbar\) versus \(-\beta\). Setting them equal,

(121)#\[ \frac{-\mathrm{i}\,t}{\hbar} \;=\; -\beta \quad\Longleftrightarrow\quad t \;=\; -\mathrm{i}\,\hbar\beta. \]

Real time and inverse temperature are related by analytic continuation. Quantum evolution at imaginary time \(t = -\mathrm{i}\hbar\beta\) is thermal averaging at temperature \(T = 1/(k_B\beta)\). This is not a metaphor — the same operator appears on both sides. The two pillars of physics are two faces of one mathematical structure.

The Wick Rotation#

To make the identification (121) precise, introduce a real variable \(\tau\) along the imaginary axis:

(122)#\[ t \;\to\; -\mathrm{i}\,\tau,\qquad \tau\in\mathbb{R}. \]

Geometrically, this rotates the time contour by \(-\pi/2\) in the complex plane. The new variable \(\tau\) has dimensions of time and is called imaginary time. The matrix element of the evolution operator becomes

(123)#\[ \langle x'\vert \mathrm{e}^{-\mathrm{i}\hat{H}t/\hbar}\vert x\rangle \;\xrightarrow{t\,\to\,-\mathrm{i}\tau}\; \langle x'\vert \mathrm{e}^{-\hat{H}\tau/\hbar}\vert x\rangle. \]

Setting \(\tau = \hbar\beta\) recovers the thermal density matrix; setting \(\tau = \mathrm{i}t\) recovers real-time quantum mechanics. One operator, two physics.

Action and Path-Integral Weight#

Apply the Wick rotation inside the real-time path integral \(\int\mathcal{D}[x]\,\mathrm{e}^{\mathrm{i}S[x]/\hbar}\). With \(\mathrm{d}t = -\mathrm{i}\,\mathrm{d}\tau\) and \(\dot{x} = \mathrm{i}\,x'(\tau)\) (where \(x'(\tau)\equiv\mathrm{d}x/\mathrm{d}\tau\)), the kinetic term flips sign while \(\mathrm{d}t\) contributes another \(-\mathrm{i}\):

(124)#\[ S \;=\; \int\!\Bigl(\tfrac{1}{2}m\dot{x}^2 - V\Bigr)\,\mathrm{d}t \;\xrightarrow{t\,\to\,-\mathrm{i}\tau}\; \mathrm{i}\int\!\Bigl(\tfrac{1}{2}m\,x'(\tau)^2 + V(x)\Bigr)\,\mathrm{d}\tau \;\equiv\; \mathrm{i}\,S_E[x]. \]

Both terms in the Euclidean action \(S_E\) are positive:

Euclidean action

(125)#\[ S_E[x] \;=\; \int_0^{\tau_f}\!\Bigl(\tfrac{1}{2}m\,x'(\tau)^2 + V(x)\Bigr)\,\mathrm{d}\tau \;\ge\; 0. \]

The kinetic and potential terms add (no minus sign), so \(S_E\) is a genuine sum of energies along the path.

Consequently the path-integral weight rotates from oscillation to exponential damping:

(126)#\[ \mathrm{e}^{\mathrm{i}S/\hbar} \;\to\; \mathrm{e}^{-S_E/\hbar}. \]

A wildly oscillating quantum integrand becomes a real, positive, exponentially-suppressed weight — a probability measure on path space.

Partition Function as a Periodic Path Integral#

Now combine the trace (119) with the Wick-rotated path integral. The trace is the essential point: it identifies the initial and final quantum state. In a position basis this means the imaginary-time path starts at some \(x_0\) and must return to the same \(x_0\) after imaginary time \(\tau_f=\hbar\beta\).

Partition function as a periodic Euclidean path integral

(127)#\[ Z(\beta) \;=\; \mathrm{Tr}\,\mathrm{e}^{-\beta\hat{H}} \;=\; \oint_{x(\hbar\beta)\,=\,x(0)} \mathcal{D}[x]\;\mathrm{e}^{-S_E[x]/\hbar}. \]

Equivalently, imaginary time is periodic with thermal period

(128)#\[ \tau \;\sim\; \tau + \hbar\beta \;=\; \tau + \frac{\hbar}{k_B T}. \]

The partition function is therefore not a transition amplitude between different states. It is a sum over closed Euclidean histories. The word “circle” is just this periodic identification of imaginary time.

Quantum-classical duality

Equation (127) quietly delivers a profound reinterpretation: a \(d\)-dimensional quantum system at temperature \(T\) is mathematically identical to a \((d{+}1)\)-dimensional classical statistical system, with the extra dimension being imaginary time, of finite extent \(\hbar\beta\), with periodic boundary conditions. Modern many-body simulation methods (path-integral Monte Carlo, lattice gauge theory) exploit this duality directly; quantum critical points in \(d\) dimensions can be mapped to classical critical points in \(d{+}1\). These applications are beyond the scope of this course.

Summary#

  • Boltzmann from max entropy: \(p_n = \mathrm{e}^{-\beta E_n}/Z\) is the maximum-entropy distribution at fixed mean energy; \(\beta\) is the Lagrange multiplier interpreted as inverse temperature.

  • Partition function,: \(Z(\beta) = \sum_n\mathrm{e}^{-\beta E_n} = \mathrm{Tr}\,\mathrm{e}^{-\beta\hat{H}}\). Generating function of equilibrium thermodynamics.

  • Time becomes temperature: the operators \(\mathrm{e}^{-\mathrm{i}\hat{H}t/\hbar}\) and \(\mathrm{e}^{-\beta\hat{H}}\) coincide under \(t = -\mathrm{i}\hbar\beta\).

  • Wick rotation: \(t\to -\mathrm{i}\tau\) analytically continues real-time evolution into imaginary time.

  • Euclidean action and weight,: kinetic and potential add (both positive); \(\mathrm{e}^{\mathrm{i}S/\hbar}\to\mathrm{e}^{-S_E/\hbar}\).

  • Periodic Euclidean path integral: \(Z(\beta)\) is a path integral over closed loops of circumference \(\hbar\beta\) in imaginary time. The trace forces periodicity; temperature sets the period.

See Also

Homework#

1. Microcanonical from max entropy. The microcanonical ensemble assigns equal probability to all states with energy in a thin shell \([E, E+\delta E]\). Apply max-entropy reasoning to derive it.

(a) Maximize \(S = -k_B\sum_n p_n\,\ln p_n\) subject only to \(\sum_n p_n = 1\), where the sum runs over states inside the shell. Show that \(p_n = 1/\Omega(E)\) is uniform, where \(\Omega(E)\) is the number of states in the shell.

(b) Compute \(S = k_B\ln\Omega(E)\) — Boltzmann’s entropy formula.

(c) The canonical (Boltzmann) distribution \(p_n = \mathrm{e}^{-\beta E_n}/Z\) instead arises by adding the energy constraint \(\sum_n p_n E_n = U\). Without computing the full max-entropy stationary point, explain physically why dropping the energy constraint is unphysical for a system in contact with a thermal bath — i.e., why the bath’s role is to fix the mean energy, not the total state count.

2. Partition function derivatives. A two-level system has \(\hat{H} = E_0\vert 0\rangle\langle 0\vert + E_1\vert 1\rangle\langle 1\vert\) with \(E_0 < E_1\) and gap \(\Delta = E_1 - E_0\).

(a) Compute \(Z(\beta)\) explicitly.

(b) From (118), compute \(\langle E\rangle(T)\) and the heat capacity \(C(T)\). Sketch \(C(T)\) and identify the Schottky peak temperature \(k_B T \sim \Delta\).

(c) Verify the limits \(\langle E\rangle\to E_0\) as \(T\to 0\) and \(\langle E\rangle\to (E_0+E_1)/2\) as \(T\to\infty\).

3. Spectral sum equals trace. For the qubit of HW 3.4.1.2, verify \(Z(\beta) = \mathrm{Tr}\,\mathrm{e}^{-\beta\hat{H}}\) in two ways: (a) directly from the definition \(\sum_n\mathrm{e}^{-\beta E_n}\), and (b) by computing the matrix exponential \(\mathrm{e}^{-\beta\hat{H}}\) in the \(\{\vert 0\rangle,\vert 1\rangle\}\) basis and tracing. Confirm the two results agree.

4. Free-particle Wick rotation. For a free particle (\(V = 0\)) with real-time action \(S = \int_0^T\tfrac{1}{2}m\dot{x}^2\,\mathrm{d}t\):

(a) Apply \(t \to -\mathrm{i}\tau\), \(\dot{x}\to\mathrm{i}\,x'(\tau)\), \(\mathrm{d}t\to -\mathrm{i}\,\mathrm{d}\tau\) and show \(S \to +\mathrm{i}\,S_E\) with \(S_E = \int_0^{T_E}\tfrac{1}{2}m\,x'(\tau)^2\,\mathrm{d}\tau \ge 0\).

(b) Verify \(\mathrm{e}^{\mathrm{i}S/\hbar}\to\mathrm{e}^{-S_E/\hbar}\), and explain why this exponentially suppresses paths with large kinetic energy.

5. Harmonic oscillator Euclidean action. For \(V(x) = \tfrac{1}{2}m\omega^2 x^2\):

(a) Apply the Wick rotation to derive \(S_E = \int_0^{T_E}\bigl(\tfrac{1}{2}m\,x'(\tau)^2 + \tfrac{1}{2}m\omega^2 x^2\bigr)\,\mathrm{d}\tau\).

(b) Both terms in \(S_E\) are positive. Explain in one sentence why this guarantees convergence of the Euclidean path integral.

(c) For the trial path \(x(\tau) = A\cos(\omega\tau)\) with small \(A\), compute \(S_E\) over one period \(T_E = 2\pi/\omega\) and verify it is positive.

6. Time becomes temperature: numerical estimate. Estimate \(\hbar\beta\) for an electronic system at room temperature (\(T = 300\,\mathrm{K}\)).

(a) Compute \(\hbar\beta\) in seconds. How does this compare to the inverse Bohr-orbit frequency \(\hbar/E_{\mathrm{Ry}}\) where \(E_{\mathrm{Ry}} \approx 13.6\,\mathrm{eV}\)?

(b) On what physical timescale does Wick rotation matter for thermal physics? In one sentence, why is room-temperature electronic structure essentially “frozen” on the thermal-circle scale?

7. Free-particle thermal trace and equipartition. The Euclidean propagator of a free particle is \(K_E(x,\tau;x_0,0) = \sqrt{m/(2\pi\hbar\tau)}\,\exp[-m(x-x_0)^2/(2\hbar\tau)]\).

(a) Set \(\tau = \hbar\beta\), \(x = x_0\), and integrate over \(x_0\) in a box of length \(L\) to obtain \(Z_{\mathrm{free}}(\beta) = L/\lambda_{\mathrm{th}}\) with \(\lambda_{\mathrm{th}} = \sqrt{2\pi\hbar^2\beta/m}\).

(b) From \(\langle E\rangle = -\partial\ln Z/\partial\beta\), recover the 1D classical equipartition result \(\langle E\rangle = \tfrac{1}{2}k_B T\).

8. Thermal circle, geometrically. Sketch the imaginary-time circle of circumference \(\hbar\beta\) for two regimes: \(\hbar\beta \ll \lambda_{\mathrm{th}}/v\) (high \(T\)) and \(\hbar\beta \gg \lambda_{\mathrm{th}}/v\) (low \(T\)), where \(v = \sqrt{k_B T/m}\) is a typical thermal velocity.

(a) Argue that in the high-\(T\) limit a typical path barely moves during one trip around the circle, so the thermal trace reduces to a classical Boltzmann integral over a single position \(x_0\).

(b) Argue that in the low-\(T\) limit only the lowest-energy eigenstate contributes to \(Z(\beta)\). Show \(Z(\beta)\to \mathrm{e}^{-\beta E_0}\) as \(\beta\to\infty\) and conclude \(E_0 = -\lim_{\beta\to\infty}\beta^{-1}\ln Z(\beta)\) — the long-\(\beta\) projection onto the ground state.