Thawed Gaussian Coherence Thermofield Dynamics

• Thawed Gaussian Coherence Thermofield Dynamics is a single-trajectory finite-temperature method that reformulates vibronic spectra as a zero-temperature wavepacket autocorrelation using an augmented configuration space.
• It employs a local-harmonic approximation on the excited-state surface to propagate a thawed Gaussian wavepacket, simplifying the evaluation of dipole autocorrelation functions.
• The approach offers computational efficiency and qualitative accuracy at low anharmonicity, though it may produce unphysical negative intensities and miss hot-band features under strong anharmonic conditions.

Searching arXiv for the specified paper and closely related thermofield/coherence/Herman–Kluk work. arXiv search query: (Kröninger et al., 29 Jul 2025) Herman-Kluk coherence thermofield dynamics vibronic spectra Thawed Gaussian coherence thermofield dynamics is a single-trajectory finite-temperature method for vibronic spectroscopy in which coherence thermofield dynamics is combined with a thawed-Gaussian propagation of the thermofield wavepacket. In the formulation associated with "Vibronic spectra at nonzero temperatures from Herman-Kluk coherence thermofield dynamics" (Kröninger et al., 29 Jul 2025), the finite-temperature dipole trace is rewritten exactly as a zero-temperature wavepacket autocorrelation on an augmented configuration space of doubled dimension, and, in the special case emphasized here, the thermofield or primed degrees of freedom are kept fixed at the minimum of the ground-state potential, so that only one classical trajectory and one time-dependent Gaussian must be propagated.

1. Thermofield reformulation of the finite-temperature coherence

In the Condon approximation, the absorption cross-section is proportional to the Fourier transform of the dipole autocorrelation,

$$\sigma(\omega)=\frac{2\pi \omega}{3\hbar c}\int_{-\infty}^{\infty} C_{\mu\mu}(t)e^{i\omega t}dt,$$

with

$$C_{\mu\mu}(t)=\mathrm{Tr}\!\bigl[\hat\mu^\dagger e^{-i\hat H_e t/\hbar}\hat\mu\,\hat\rho\,e^{i\hat H_g t/\hbar}\bigr]=\mu^2\,C(t),$$

and

$$C(t)=\mathrm{Tr}\bigl[e^{-i\hat H_e t/\hbar}\,\hat\rho\,e^{i\hat H_g t/\hbar}\bigr].$$

Coherence thermofield dynamics replaces this finite-temperature trace by a pure-state overlap on a doubled coordinate space (Kröninger et al., 29 Jul 2025).

The thermofield pure state is introduced as

$$\bar\psi_0(\bar q)=\langle q|\hat\rho^{1/2}|q'\rangle,$$

with doubled coordinate

$\bar q=(q,q').$

One then obtains the exact identity

$$C(t)=\langle \bar\psi_0|\,e^{-i\hat{\bar H}t/\hbar}\,|\bar\psi_0\rangle,$$

where the augmented Hamiltonian is

$$\hat{\bar H}=\tfrac12\,\hat{\bar p}^{T}\bar m^{-1}\hat{\bar p}+\bar V(\hat{\bar q}),$$

with

$$\bar m=\mathrm{diag}(m,-m),\qquad \bar V(q,q')=V_e(q)-V_g(q'),$$

and

$\bar p=(p,p').$

The key structural consequence is that a finite-temperature trace becomes a zero-temperature wavepacket autocorrelation on an augmented Hamiltonian system.

2. Single-trajectory thawed-Gaussian ansatz

The thawed-Gaussian approximation assumes that the thermofield wavepacket remains a single Gaussian at all times (Kröninger et al., 29 Jul 2025):

$$\bar\psi(\bar q,t)=N(t)\,\exp\Bigl\{-\tfrac{1}{2\hbar}(\bar q-\bar q_t)^T A_t(\bar q-\bar q_t) +\tfrac{i}{\hbar}\bar p_t^T(\bar q-\bar q_t) +\tfrac{i}{\hbar}\gamma(t)\Bigr\},$$

where

$$C_{\mu\mu}(t)=\mathrm{Tr}\!\bigl[\hat\mu^\dagger e^{-i\hat H_e t/\hbar}\hat\mu\,\hat\rho\,e^{i\hat H_g t/\hbar}\bigr]=\mu^2\,C(t),$$0

$$C_{\mu\mu}(t)=\mathrm{Tr}\!\bigl[\hat\mu^\dagger e^{-i\hat H_e t/\hbar}\hat\mu\,\hat\rho\,e^{i\hat H_g t/\hbar}\bigr]=\mu^2\,C(t),$$1 is a $$C_{\mu\mu}(t)=\mathrm{Tr}\!\bigl[\hat\mu^\dagger e^{-i\hat H_e t/\hbar}\hat\mu\,\hat\rho\,e^{i\hat H_g t/\hbar}\bigr]=\mu^2\,C(t),$$2 complex symmetric width matrix, and $$C_{\mu\mu}(t)=\mathrm{Tr}\!\bigl[\hat\mu^\dagger e^{-i\hat H_e t/\hbar}\hat\mu\,\hat\rho\,e^{i\hat H_g t/\hbar}\bigr]=\mu^2\,C(t),$$3 is a real phase.

In the single-trajectory version, the primed packet remains fixed at the ground-state minimum,

$$C_{\mu\mu}(t)=\mathrm{Tr}\!\bigl[\hat\mu^\dagger e^{-i\hat H_e t/\hbar}\hat\mu\,\hat\rho\,e^{i\hat H_g t/\hbar}\bigr]=\mu^2\,C(t),$$4

Accordingly, only the unprimed trajectory $$C_{\mu\mu}(t)=\mathrm{Tr}\!\bigl[\hat\mu^\dagger e^{-i\hat H_e t/\hbar}\hat\mu\,\hat\rho\,e^{i\hat H_g t/\hbar}\bigr]=\mu^2\,C(t),$$5 and the corresponding $$C_{\mu\mu}(t)=\mathrm{Tr}\!\bigl[\hat\mu^\dagger e^{-i\hat H_e t/\hbar}\hat\mu\,\hat\rho\,e^{i\hat H_g t/\hbar}\bigr]=\mu^2\,C(t),$$6 block of the width matrix are propagated. This is the defining simplification of the thawed-Gaussian coherence thermofield dynamics variant summarized in the source material.

The initial thermofield state is obtained from the square-root density for a harmonic expansion of $$C_{\mu\mu}(t)=\mathrm{Tr}\!\bigl[\hat\mu^\dagger e^{-i\hat H_e t/\hbar}\hat\mu\,\hat\rho\,e^{i\hat H_g t/\hbar}\bigr]=\mu^2\,C(t),$$7 around its minimum. The initial width is

$$C_{\mu\mu}(t)=\mathrm{Tr}\!\bigl[\hat\mu^\dagger e^{-i\hat H_e t/\hbar}\hat\mu\,\hat\rho\,e^{i\hat H_g t/\hbar}\bigr]=\mu^2\,C(t),$$8

with

$$C_{\mu\mu}(t)=\mathrm{Tr}\!\bigl[\hat\mu^\dagger e^{-i\hat H_e t/\hbar}\hat\mu\,\hat\rho\,e^{i\hat H_g t/\hbar}\bigr]=\mu^2\,C(t),$$9

$$C(t)=\mathrm{Tr}\bigl[e^{-i\hat H_e t/\hbar}\,\hat\rho\,e^{i\hat H_g t/\hbar}\bigr].$$0

$$C(t)=\mathrm{Tr}\bigl[e^{-i\hat H_e t/\hbar}\,\hat\rho\,e^{i\hat H_g t/\hbar}\bigr].$$1

Because the ground-state packet does not evolve in the single-trajectory approximation, one sets the primed block constant and propagates only the unprimed block, denoted $$C(t)=\mathrm{Tr}\bigl[e^{-i\hat H_e t/\hbar}\,\hat\rho\,e^{i\hat H_g t/\hbar}\bigr].$$2. The initial data are therefore

$$C(t)=\mathrm{Tr}\bigl[e^{-i\hat H_e t/\hbar}\,\hat\rho\,e^{i\hat H_g t/\hbar}\bigr].$$3

3. Local-harmonic dynamics on the excited-state surface

The time evolution is generated by a local-harmonic approximation to the thermofield potential around $$C(t)=\mathrm{Tr}\bigl[e^{-i\hat H_e t/\hbar}\,\hat\rho\,e^{i\hat H_g t/\hbar}\bigr].$$4 (Kröninger et al., 29 Jul 2025):

$$C(t)=\mathrm{Tr}\bigl[e^{-i\hat H_e t/\hbar}\,\hat\rho\,e^{i\hat H_g t/\hbar}\bigr].$$5

Inserting the Gaussian ansatz into the Schrödinger equation,

$$C(t)=\mathrm{Tr}\bigl[e^{-i\hat H_e t/\hbar}\,\hat\rho\,e^{i\hat H_g t/\hbar}\bigr].$$6

and matching coefficients of powers of $$C(t)=\mathrm{Tr}\bigl[e^{-i\hat H_e t/\hbar}\,\hat\rho\,e^{i\hat H_g t/\hbar}\bigr].$$7 yields the thawed-Gaussian ordinary differential equations:

$$C(t)=\mathrm{Tr}\bigl[e^{-i\hat H_e t/\hbar}\,\hat\rho\,e^{i\hat H_g t/\hbar}\bigr].$$8

$$C(t)=\mathrm{Tr}\bigl[e^{-i\hat H_e t/\hbar}\,\hat\rho\,e^{i\hat H_g t/\hbar}\bigr].$$9

$$\bar\psi_0(\bar q)=\langle q|\hat\rho^{1/2}|q'\rangle,$$0

$$\bar\psi_0(\bar q)=\langle q|\hat\rho^{1/2}|q'\rangle,$$1

Here

$$\bar\psi_0(\bar q)=\langle q|\hat\rho^{1/2}|q'\rangle,$$2

is the Hessian of the excited-state surface.

The source material states that the minus sign in front of $$\bar\psi_0(\bar q)=\langle q|\hat\rho^{1/2}|q'\rangle,$$3 drops out because $$\bar\psi_0(\bar q)=\langle q|\hat\rho^{1/2}|q'\rangle,$$4 and $$\bar\psi_0(\bar q)=\langle q|\hat\rho^{1/2}|q'\rangle,$$5 are constant, so no driving arises from $$\bar\psi_0(\bar q)=\langle q|\hat\rho^{1/2}|q'\rangle,$$6. This isolates the active dynamics onto a single classical trajectory on $$\bar\psi_0(\bar q)=\langle q|\hat\rho^{1/2}|q'\rangle,$$7, while the thermal information remains encoded in the initial thermofield width.

4. Closed-form autocorrelation and spectral reconstruction

Within the thawed-Gaussian approximation, the thermofield autocorrelation is an overlap of two Gaussians and can therefore be evaluated in closed form (Kröninger et al., 29 Jul 2025). With $$\bar\psi_0(\bar q)=\langle q|\hat\rho^{1/2}|q'\rangle,$$8, $$\bar\psi_0(\bar q)=\langle q|\hat\rho^{1/2}|q'\rangle,$$9, and $\bar q=(q,q').$0,

$\bar q=(q,q').$1

The absorption spectrum then follows by Fourier transformation:

$\bar q=(q,q').$2

or, alternatively, from the full Fourier transform

$\bar q=(q,q').$3

A notable feature of this construction is that once $\bar q=(q,q').$4 are known, no further costly overlap integrations are required. The thermal trace, the quantum time propagation, and the final spectral reconstruction are all expressed through Gaussian data generated along a single trajectory.

5. Computational structure and numerical characteristics

The numerical workflow is compact because the method is explicitly single-trajectory (Kröninger et al., 29 Jul 2025). The required operations are summarized below.

Component	Requirement	Consequence
Classical propagation	Only one $\bar q=(q,q').$5-dimensional classical trajectory $\bar q=(q,q').$6 on $\bar q=(q,q').$7 is needed	The dynamical core is low-dimensional relative to doubled-space propagation
Local electronic information	At each time step one evaluates $\bar q=(q,q').$8 and the Hessian $\bar q=(q,q').$9	The method requires curvature information along the trajectory
Correlation evaluation	$$C(t)=\langle \bar\psi_0|\,e^{-i\hat{\bar H}t/\hbar}\,|\bar\psi_0\rangle,$$0 is obtained from Eq. (9) once $$C(t)=\langle \bar\psi_0|\,e^{-i\hat{\bar H}t/\hbar}\,|\bar\psi_0\rangle,$$1 are known	No further costly overlap integrations

The ordinary differential equations may be integrated by standard symplectic or Runge–Kutta–Nyström schemes. The source material also specifies a simple split approach: update $$C(t)=\langle \bar\psi_0|\,e^{-i\hat{\bar H}t/\hbar}\,|\bar\psi_0\rangle,$$2 by Verlet and then $$C(t)=\langle \bar\psi_0|\,e^{-i\hat{\bar H}t/\hbar}\,|\bar\psi_0\rangle,$$3 by a small-step integration of the width and phase equations. No Monte Carlo sampling is required, and the cost is described as essentially one classical Hessian evaluation per time step plus Gaussian parameter updates.

This computational profile suggests why the thawed-Gaussian coherence thermofield dynamics method is attractive when a finite-temperature spectrum is needed but a multi-trajectory semiclassical treatment is too expensive.

6. Regime of validity and comparison with Herman–Kluk coherence thermofield dynamics

The method’s principal approximation is the retention of only the local harmonic curvature of $$C(t)=\langle \bar\psi_0|\,e^{-i\hat{\bar H}t/\hbar}\,|\bar\psi_0\rangle,$$4 along a single path (Kröninger et al., 29 Jul 2025). As stated in the source material, it therefore cannot capture wavepacket splitting, interference, or higher-order anharmonic effects. This limitation becomes especially consequential at high temperatures, where the thermal density populates many excited vibrational hot levels and the associated dynamics explore regions far from the reference trajectory. Those contributions generate weak hot-band peaks in the spectrum.

The reported benchmark is a comparison for Morse potentials of increasing anharmonicity, evaluated at various temperatures, against a numerically exact approach, Herman–Kluk coherence thermofield dynamics, and the single-trajectory thawed-Gaussian coherence thermofield dynamics (Kröninger et al., 29 Jul 2025). At low anharmonicity, both approximate methods yield accurate spectra. In a Morse potential with higher anharmonicity, however, the thawed-Gaussian thermofield dynamics, because it is based on the local harmonic approximation, fails to capture emerging hot bands, whereas the Herman–Kluk thermofield approach successfully reproduces them.

A further limitation stated explicitly in the source material is that the single-trajectory thawed-Gaussian coherence thermofield dynamics can produce unphysical negative intensities, since the thawed-Gaussian time evolution solves effectively a nonlinear Schrödinger equation. By contrast, the Herman–Kluk coherence thermofield approach superposes phases from many trajectories and includes the full local stability matrix, which allows it to describe interference and the broader phase-space exploration needed to recover hot-band structure.

Taken together, these comparisons place thawed Gaussian coherence thermofield dynamics in a specific methodological niche: it remains extremely efficient and often qualitatively correct at low anharmonicity, but it loses accuracy at high temperature and strong anharmonicity, precisely where the multi-trajectory Herman–Kluk treatment succeeds.