Gaussian Overlap Decoherence Correction in FSSH

• Gaussian overlap decoherence correction is defined as a method to reduce excessive electronic coherence in FSSH by modeling the overlap of bifurcated nuclear Gaussian branches.
• It employs a time-dependent overlap model based on hop-induced momentum and force differences to derive decoherence times, ensuring more realistic quantum dynamics.
• The correction is selectively applied in regions of low nonadiabaticity—using the Massey parameter threshold, as benchmarked in models like Tully's extended coupling and spin-boson systems.

Gaussian overlap decoherence correction is used most explicitly for a decoherence remedy in fewest-switches surface hopping (FSSH) in which coherence loss is estimated from the overlap of Gaussian nuclear branches. In that setting, the correction addresses the standard FSSH defect of retaining too much electronic coherence after a nonadiabatic event, but the central conclusion is that the correction should not be applied everywhere: in “On decoherence in surface hopping: the nonadiabaticity threshold” the proposed strategy is to restrict it to regions of low nonadiabaticity measured by the dimensionless Massey parameter (Runeson, 24 Jul 2025). Closely related overlap-based Gaussian constructions also appear in open-system decoherence theory, dynamical-decoupling spectroscopy, neutrino wave-packet decoherence, bosonic-code decoding, and Gaussian channel correction, although those works generally use “overlap” and “Gaussian” to denote interpolation, inference, or channel engineering rather than the same FSSH damping rule (Yan et al., 2021, Szańkowski et al., 2017, Mitani et al., 2023, Vuillot et al., 2018, Tserkis et al., 2018, Machinaga et al., 8 Apr 2026, Wang et al., 11 Apr 2026).

1. Definition in surface hopping and the overcoherence problem

Within FSSH, Gaussian overlap decoherence correction is motivated by the mismatch between a trajectory picture and true quantum branching. Standard FSSH propagates nuclear trajectories on adiabatic surfaces while electronic amplitudes remain coherent superpositions, so after a nonadiabatic event it often retains too much coherence even when a quantum wavepacket would have decohered. The Gaussian-overlap construction models this missing decoherence by comparing the overlap of two nuclear branches, typically a hopped and a non-hopped branch, approximated as Gaussians with the same width parameter $\gamma$ (Runeson, 24 Jul 2025).

For two such branches, the squared overlap is written as

$$|\langle \psi_a|\psi_b\rangle|^2 = \exp\left[ -\frac{\gamma}{2}\Delta q^2 -\frac{1}{2\gamma\hbar^2}\Delta p^2 \right],$$

where $\Delta q=q_b-q_a$ and $\Delta p=p_b-p_a$. In the FSSH application, the two branches initially coincide in position but acquire a momentum difference from the hop rescaling. This yields a time-dependent overlap model,

$$D_\text{Gauss}(t)=|\langle \psi_a|\psi_b\rangle|^2 = \exp(-k_1 t-k_2 t^2).$$

The physical interpretation is specific. The correction is not introduced as a generic phenomenological damping factor; it is tied to an overlap model for the bifurcated nuclear packet. This distinguishes it from energy-gap-based decoherence prescriptions that do not explicitly derive coherence loss from branch separation.

2. Gaussian-overlap formula and decoherence time

The Gaussian-overlap model in FSSH is parameterized by coefficients determined by the hop-induced momentum change $\Delta p_{\rm hop}$ and the force difference $\Delta F = F_b - F_a}$. The coefficients are

$$k_1 = \frac{1}{\gamma\hbar^2}\Delta p_{\rm hop}\cdot \Delta F, \qquad k_2 = \frac{\gamma}{2}\left(\frac{\Delta p_{\rm hop}}{m}\right)^2 +\frac{1}{2\gamma \hbar^2}\Delta F^2.$$

In multidimensions, the paper gives

$$k_1 = \sum_j \frac{(\Delta p_{\rm hop})_j (\Delta F)_j}{\gamma_j\hbar^2}, \qquad k_2 = \sum_j \left[ \frac{\gamma_j(\Delta p_{\rm hop})_j^2}{2m_j^2} + \frac{(\Delta F)_j^2}{2\gamma_j \hbar^2} \right].$$

Each non-active state then has its own $k_1$, $$|\langle \psi_a|\psi_b\rangle|^2 = \exp\left[ -\frac{\gamma}{2}\Delta q^2 -\frac{1}{2\gamma\hbar^2}\Delta p^2 \right],$$0, and hence its own decoherence time (Runeson, 24 Jul 2025).

The corresponding damping timescale is

$$|\langle \psi_a|\psi_b\rangle|^2 = \exp\left[ -\frac{\gamma}{2}\Delta q^2 -\frac{1}{2\gamma\hbar^2}\Delta p^2 \right],$$1

Operationally, non-active amplitudes are damped, and the active-state amplitude is renormalized to conserve total population. The paper’s central methodological point is that this timescale is useful only when evaluated in regions where the surfaces are effectively uncoupled; in strongly nonadiabatic and anharmonic regions, the assumptions behind the Gaussian overlap estimate are least reliable (Runeson, 24 Jul 2025).

3. Nonadiabaticity threshold, GONT, and instantaneous collapse

The decisive modification introduced in the threshold paper is to gate the correction by the dimensionless Massey parameter

$$|\langle \psi_a|\psi_b\rangle|^2 = \exp\left[ -\frac{\gamma}{2}\Delta q^2 -\frac{1}{2\gamma\hbar^2}\Delta p^2 \right],$$2

where $$|\langle \psi_a|\psi_b\rangle|^2 = \exp\left[ -\frac{\gamma}{2}\Delta q^2 -\frac{1}{2\gamma\hbar^2}\Delta p^2 \right],$$3 is the time-derivative coupling and $$|\langle \psi_a|\psi_b\rangle|^2 = \exp\left[ -\frac{\gamma}{2}\Delta q^2 -\frac{1}{2\gamma\hbar^2}\Delta p^2 \right],$$4 is the adiabatic energy gap. In the paper’s interpretation, $$|\langle \psi_a|\psi_b\rangle|^2 = \exp\left[ -\frac{\gamma}{2}\Delta q^2 -\frac{1}{2\gamma\hbar^2}\Delta p^2 \right],$$5 controls the rate of population transfer, while $$|\langle \psi_a|\psi_b\rangle|^2 = \exp\left[ -\frac{\gamma}{2}\Delta q^2 -\frac{1}{2\gamma\hbar^2}\Delta p^2 \right],$$6 controls phase rotation around the coherence axis. Thus $$|\langle \psi_a|\psi_b\rangle|^2 = \exp\left[ -\frac{\gamma}{2}\Delta q^2 -\frac{1}{2\gamma\hbar^2}\Delta p^2 \right],$$7 corresponds to effectively uncoupled states, whereas $$|\langle \psi_a|\psi_b\rangle|^2 = \exp\left[ -\frac{\gamma}{2}\Delta q^2 -\frac{1}{2\gamma\hbar^2}\Delta p^2 \right],$$8 indicates strong nonadiabatic coupling (Runeson, 24 Jul 2025).

The threshold criterion is

$$|\langle \psi_a|\psi_b\rangle|^2 = \exp\left[ -\frac{\gamma}{2}\Delta q^2 -\frac{1}{2\gamma\hbar^2}\Delta p^2 \right],$$9

If this condition is satisfied, decoherence is applied; if not, decoherence is suppressed. The resulting thresholded Gaussian scheme is called Gaussian Overlaps with Nonadiabaticity Threshold (GONT). A cheaper variant, Instantaneous Nonadiabaticity Threshold (INT), uses the same criterion but replaces $\Delta q=q_b-q_a$0 by immediate collapse.

Scheme	Criterion	Action
GONT	$\Delta q=q_b-q_a$1	Apply Gaussian-overlap damping using $\Delta q=q_b-q_a$2
INT	$\Delta q=q_b-q_a$3	Collapse non-active amplitude instantaneously
EDC	No nonadiabaticity threshold	Use $\Delta q=q_b-q_a$4

The numerical guidance reported in the paper is unusually concrete. In the Tully extended coupling model, a safe threshold window of roughly

$\Delta q=q_b-q_a$5

is identified; $\Delta q=q_b-q_a$6 is presented as the best general compromise, while $\Delta q=q_b-q_a$7 is safer in delicate coherent cases. The practical recommendation is to run two GONT simulations, one with $\Delta q=q_b-q_a$8 and one with $\Delta q=q_b-q_a$9, use the averaged $\Delta p=p_b-p_a$0 as a robust population estimate, and use the difference between the two runs as an uncertainty indicator (Runeson, 24 Jul 2025).

4. Benchmark behavior and comparison with unrestricted decoherence

The thresholded scheme is benchmarked on Tully models, spin-boson models, the FMO complex, conical intersection models, and spin-orbit coupling models. Across nearly all systems, the reported pattern is that uncorrected FSSH often gives fairly good $\Delta p=p_b-p_a$1-type population measures but inconsistent $\Delta p=p_b-p_a$2-type wavefunction populations, unrestricted Gaussian-overlap decoherence or EDC can over-suppress coherence and worsen results, GONT with a low threshold usually gives the best compromise, and INT often works for fast-decoherence systems but is generally less accurate than GONT for slower-decoherence or coherent systems (Runeson, 24 Jul 2025).

The Tully extended coupling model is the key demonstration. Bare FSSH gets the first crossing right but fails after the second crossing. Unrestricted Gaussian overlap decoherence is too strong. The paper reports that $\Delta p=p_b-p_a$3 gives only slight improvement, $\Delta p=p_b-p_a$4 is adequate, and $\Delta p=p_b-p_a$5 gives excellent accuracy, with $\Delta p=p_b-p_a$6 similarly good but with slight inconsistency. In scattering probabilities for the same model, GONT with $\Delta p=p_b-p_a$7 or $\Delta p=p_b-p_a$8 gives accurate results and removes oscillations that appear in bare FSSH; EDC fails to remove some oscillations and can be inaccurate (Runeson, 24 Jul 2025).

The comparison with energy-based decoherence correction is central because it frames the controversy over “more decoherence” versus “better decoherence.” EDC uses

$\Delta p=p_b-p_a$9

with

$$D_\text{Gauss}(t)=|\langle \psi_a|\psi_b\rangle|^2 = \exp(-k_1 t-k_2 t^2).$$0

where $$D_\text{Gauss}(t)=|\langle \psi_a|\psi_b\rangle|^2 = \exp(-k_1 t-k_2 t^2).$$1 a.u. The criticism given is that EDC depends on the total kinetic energy, which is not physically relevant to local nonadiabaticity, depends on an arbitrary energy scale $$D_\text{Gauss}(t)=|\langle \psi_a|\psi_b\rangle|^2 = \exp(-k_1 t-k_2 t^2).$$2, and can over-decohere coherent dynamics. The benchmarks in coherent spin-boson and FMO-like systems are reported to confirm this tendency (Runeson, 24 Jul 2025).

5. Related overlap-based Gaussian decoherence formalisms

Outside surface hopping, several works use a related Gaussian-overlap logic, but not the same correction protocol. In “Decoherence factor as a convolution: an interplay between a Gaussian and an exponential coherence loss,” the decoherence factor is not given by a hand-added Gaussian correction; rather, it is written as the convolution of an exponential overlap contribution and a Gaussian spectral contribution,

$$D_\text{Gauss}(t)=|\langle \psi_a|\psi_b\rangle|^2 = \exp(-k_1 t-k_2 t^2).$$3

equivalently,

$$D_\text{Gauss}(t)=|\langle \psi_a|\psi_b\rangle|^2 = \exp(-k_1 t-k_2 t^2).$$4

Weak coupling gives predominantly exponential decay, strong coupling gives predominantly Gaussian decay, and the crossover is controlled by the system-environment coupling. The authors are explicit that this is not universal; low-temperature quantum Brownian motion can develop long-time power-law tails, so the Gaussian/exponential picture is mainly an early-time or intermediate-time description (Yan et al., 2021).

A different Gaussian-specific correction issue appears in dynamical-decoupling noise spectroscopy. For a finite-ranged Gaussian spectrum,

$$D_\text{Gauss}(t)=|\langle \psi_a|\psi_b\rangle|^2 = \exp(-k_1 t-k_2 t^2).$$5

direct use of the standard spectroscopic formula can create a spurious long-tail attribution because the sinc tails of the filter dominate the exponentially small true spectral wings. The proposed cure is a modified Alvarez–Suter reconstruction in which one measures $$D_\text{Gauss}(t)=|\langle \psi_a|\psi_b\rangle|^2 = \exp(-k_1 t-k_2 t^2).$$6 for increasing durations, fits

$$D_\text{Gauss}(t)=|\langle \psi_a|\psi_b\rangle|^2 = \exp(-k_1 t-k_2 t^2).$$7

and reconstructs the spectrum from the slope $$D_\text{Gauss}(t)=|\langle \psi_a|\psi_b\rangle|^2 = \exp(-k_1 t-k_2 t^2).$$8 rather than raw attenuation values. Here “correction” refers to removing a Gaussian-specific reconstruction artifact, not to damping electronic amplitudes (Szańkowski et al., 2017).

Wave-packet neutrino oscillation provides another overlap-centered formulation. In the two-packet 3D Gaussian treatment, the observed probability depends jointly on source and detector localization. Propagation, localization, and coherence loss are governed by

$$D_\text{Gauss}(t)=|\langle \psi_a|\psi_b\rangle|^2 = \exp(-k_1 t-k_2 t^2).$$9

whereas momentum matching is governed by

$\Delta p_{\rm hop}$0

The resulting coherence length is

$\Delta p_{\rm hop}$1

This is not a correction protocol, but it sharpens the general meaning of Gaussian overlap as a joint source-detector overlap problem rather than a pure propagation problem (Mitani et al., 2023).

A further extension appears in a Gaussian open-system theory with a common structured environment under relative motion. There the overlap-induced contribution is the off-diagonal noise kernel

$\Delta p_{\rm hop}$2

which turns on only when Doppler-shifted spectral supports overlap above the kinematic threshold

$\Delta p_{\rm hop}$3

Below threshold, the common environment acts mainly as a coherent mediator at leading resonant order; above threshold, it supports finite correlated decoherence (Wang et al., 11 Apr 2026).

6. Continuous-variable, bosonic-code, and Gaussian-channel analogues

In bosonic quantum information, Gaussian overlap appears most prominently in GKP decoding under Gaussian displacement noise. For the independent Gaussian displacement channel,

$\Delta p_{\rm hop}$4

a continuous Gaussian shift must be interpreted modulo the stabilizer lattice, so the likelihood of a given syndrome is a sum over translated peaks. This is the “Gaussian overlap” issue: shifted Gaussians associated with different lattice translates overlap, and decoding amounts to deciding which logical sector has the largest total posterior weight. Repeated noisy GKP correction maps to a 1D Euclidean path integral with Villain potential, and in the concatenated toric-GKP architecture the use of GKP analog information improves the toric-code threshold from about $\Delta p_{\rm hop}$5 to $\Delta p_{\rm hop}$6 when both GKP and toric-code measurements are perfect; when only the GKP error correction measurements are perfect, a threshold at $\Delta p_{\rm hop}$7 is observed; in the fully noisy setting, a new decoder finds a threshold at $\Delta p_{\rm hop}$8, corresponding roughly to GKP states with about $\Delta p_{\rm hop}$9 photons or more if imperfections arise only from finite-energy GKP-state preparation (Vuillot et al., 2018).

A different continuous-variable correction paradigm is teleportation-based simulation of Gaussian channels. For a resource state in symmetric standard form, Braunstein–Kimble teleportation implements a Gaussian channel with

$\Delta F = F_b - F_a}$0

The paper derives the full family of finite-energy physical resource states that simulate a chosen phase-insensitive Gaussian channel and finds that the optimal states are pure, satisfy $\Delta F = F_b - F_a}$1, minimize mean energy and entanglement cost, and are equally entangled to the Choi-state in the sense of entanglement of formation. This finite-resource viewpoint is then used to generalize an earlier pure-loss error-correction protocol to thermal-loss channels (Tserkis et al., 2018).

Environment-assisted suppression of optical loss provides a more direct Gaussian-only channel correction. In that protocol, a $\Delta F = F_b - F_a}$2-squeezed vacuum is injected into the environment port of a loss channel, the leaked quadrature is measured, and feedforward with gain

$\Delta F = F_b - F_a}$3

is used to cancel added noise. The corrected quadratures are

$\Delta F = F_b - F_a}$4

and

$\Delta F = F_b - F_a}$5

The experiment tests up to five steps under $\Delta F = F_b - F_a}$6 and $\Delta F = F_b - F_a}$7 loss per step for single-photon, $\Delta F = F_b - F_a}$8-squeezed single-photon, and $\Delta F = F_b - F_a}$9-squeezed single-photon inputs, and reports systematically higher fidelity and more persistent Wigner negativity than the unsuppressed case. The measured ancilla squeezing is about $$k_1 = \frac{1}{\gamma\hbar^2}\Delta p_{\rm hop}\cdot \Delta F, \qquad k_2 = \frac{\gamma}{2}\left(\frac{\Delta p_{\rm hop}}{m}\right)^2 +\frac{1}{2\gamma \hbar^2}\Delta F^2.$$0 dB pure squeezing equivalent, and the loop has about $$k_1 = \frac{1}{\gamma\hbar^2}\Delta p_{\rm hop}\cdot \Delta F, \qquad k_2 = \frac{\gamma}{2}\left(\frac{\Delta p_{\rm hop}}{m}\right)^2 +\frac{1}{2\gamma \hbar^2}\Delta F^2.$$1 internal propagation loss (Machinaga et al., 8 Apr 2026).

7. Interpretive status, misconceptions, and limitations

Taken together, these results indicate that “Gaussian overlap decoherence correction” does not denote a single universal formal object. In FSSH it is a specific overlap-derived damping rule, and the main methodological lesson is that unrestricted application is often harmful: the correction is reported to be most reliable only in regions of low nonadiabaticity, which is why thresholding by the Massey parameter is the paper’s principal innovation (Runeson, 24 Jul 2025).

In open-system theory, the same phrase should not be read as licensing a universal Gaussian ansatz. The convolution work explicitly states that the Gaussian/exponential interpolation is not universal and can fail at long times because of spectral-edge effects (Yan et al., 2021). In dynamical-decoupling spectroscopy, Gaussian spectra are precisely the case in which naïve reconstruction is most deceptive, because filter-tail overlap can mimic a physical long tail unless the asymptotic slope is extracted (Szańkowski et al., 2017). In bosonic coding, Gaussian overlap is a decoding asset for GKP and toric-GKP codes, but linear oscillator codes still do not provide scalable protection against Gaussian displacement noise: the no-go result states that such codes merely squeeze Gaussian shift errors and do not yield a threshold improvement with increasing code size (Vuillot et al., 2018).

A plausible implication is that the operative meaning of the term depends on the layer at which “correction” is implemented. In trajectory methods it is branchwise amplitude damping; in spectroscopy it is artifact removal; in bosonic coding it is analog-information decoding over overlapping Gaussian sectors; in Gaussian channels it is environment or resource engineering. What remains common across these usages is that decoherence, inference, or mitigation is controlled by quantitatively tracking the overlap structure of Gaussian or Gaussian-derived objects rather than by using a purely hard-threshold or bit-valued description.