Matsubara dynamics starts from a problem that looks almost classical: how can we compute real-time equilibrium correlation functions while retaining the quantum Boltzmann distribution? The first useful stop is LSC-IVR, where the exact Kubo expression is mapped into phase space and then propagated with a classical Liouvillian.

Background

The Kubo-transformed correlation function for two operators \(A\) and \(B\) can be written as

\[ C_{AB}(t)= \operatorname{Tr}\left[ K_\beta(A)e^{iHt/\hbar}B e^{-iHt/\hbar} \right], \qquad K_\beta(A)=\frac{1}{\beta}\int_0^\beta e^{-\lambda H}A e^{-(\beta-\lambda)H}\,d\lambda . \]

The imaginary-time smearing in \(K_\beta(A)\) is what makes this object compatible with path-integral statistics. A direct classical correlation has the right phase-space intuition, but it does not in general conserve the quantum Boltzmann distribution.

Weyl representation

The Weyl transform maps an operator to a phase-space function:

\[ A_W(q,p)= \int d\Delta\, \left\langle q-\frac{\Delta}{2}\right|A \left|q+\frac{\Delta}{2}\right\rangle e^{ip\Delta/\hbar}. \]

After inserting the corresponding Fourier identity into the exact correlation function, one obtains a formally exact phase-space expression involving \([K_\beta(A)]_W\) and \([B(t)]_W\).

LSC-IVR approximation

The exact quantum Liouvillian contains the Moyal series,

\[ \hat L= \frac{p}{m}\frac{\partial}{\partial q} - \sum_{\lambda=1,\mathrm{odd}}^\infty \frac{1}{\lambda!} \left(\frac{i\hbar}{2}\right)^{\lambda-1} \frac{\partial^\lambda V}{\partial q^\lambda} \frac{\partial^\lambda}{\partial p^\lambda}. \]

LSC-IVR truncates this to the classical Liouvillian,

\[ \mathcal L= \frac{p}{m}\frac{\partial}{\partial q} - \frac{\partial V}{\partial q}\frac{\partial}{\partial p}, \]

and estimates the Kubo correlation with classical phase-space propagation. This is exact for a harmonic oscillator but no longer exact for nonlinear potentials.

Workflow

  1. Use an exact analytic Kubo expression for the harmonic oscillator.
  2. Use sinc-DVR diagonalization as the quantum Kubo reference for \(V(q)=q^4/4\).
  3. Sample the classical Boltzmann phase-space density on a deterministic grid.
  4. Propagate each phase-space point with velocity-Verlet under the classical force.
  5. Compare \(C_{qq}(t)=\langle q(0)q(t)\rangle\) against the quantum Kubo reference.

Result

Harmonic and quartic benchmark for LSC-IVR-style phase-space propagation
LSC-IVR-style benchmark. Left: harmonic oscillator with horizontal axis time \(t\) and vertical axis \(C_{qq}(t)\). The black solid Kubo curve and red dashed classical-Liouvillian curve overlap; the maximum absolute error is \(4.3\times10^{-6}\). Right: quartic oscillator \(V(q)=q^4/4\), with the black solid curve computed by sinc-DVR Kubo summation and the red dashed curve computed by the same phase-space approximation. The RMS difference is \(1.55\times10^{-1}\), showing that the classical Liouvillian truncation is not a controlled replacement for nonlinear quantum dynamics.

Why this sets up Matsubara dynamics

The central weakness is not that phase space is useless; the harmonic test shows that it can be exactly right. The weakness is the way the quantum Liouvillian is truncated. Matsubara dynamics changes the truncation: instead of cutting the Moyal series directly in powers of \(\hbar^2\), it keeps the smooth low-frequency imaginary-time normal modes. Part II develops that mode-space picture.

Code used in this note

References

  • Hele, Willatt, Muolo, and Althorpe, J. Chem. Phys. 142, 134103 (2015).
  • Prada, Pos, and Althorpe, J. Chem. Phys. 158, 114106 (2023).

Matsubara series map

  1. Part I: From Kubo to LSC-IVR - phase-space representation and the harmonic/quartic benchmark.
  2. Part II: Low Modes and the Matsubara Phase - smooth imaginary-time paths, Matsubara truncation, and phase conservation.
  3. Part III: Quartic Benchmark and Phase Reweighting - practical estimator, phase problem, and noisy higher-mode convergence.
  4. Part IV: 绝热与非绝热 Matsubara 实时动力学推导 - parallel derivations with a cyclic electronic trace and explicit nuclear Gaussian integrals.