arXiv:2609.39861
Reverse quantum state diffusion from differential geometry
Dinh-Long Vu, Patrick Rebentrost
Abstract
Quantum state diffusion unravels a Lindblad master equation into stochastic trajectories of pure states. The ensemble of trajectories is described by a probability density on the manifold of pure states, which contains much more information than the density matrix, its first moment. We derive the time reversal of this diffusion. Writing the reversal of a general diffusion on a manifold in Stratonovich form, and passing through the manifold of pure states, we obtain a backward stochastic differential equation for the state vector, valid for any Hilbert space dimension and any Lindbladian. The backward stochastic equation provides a physical, nonlinear unravelling of the inverse Lindbladian that holds only for the ensemble from which it was constructed. Following score-based generative models, we then use the backward equation to generate a new ensemble close to the original. For the depolarizing channel the unravelling is a Brownian motion on complex projective space, so the uniform distribution of pure states is its stationary law in any dimension. Starting the backward equation from that prior, we bound the error of the resulting distribution compared to the original one in different settings: exact and learned scores, continuous and discrete time.
AI Summary
The paper asks how to time-reverse quantum state diffusion and reconcile physical pure-state trajectories with inverse Lindblad evolution, which is not generally CPTP. It treats diffusion intrinsically on $\mathbb{CP}^{d-1}$, distinguishing the ensemble distribution from its density-matrix first moment. An earlier derivation of reverse quantum state diffusion is acknowledged; the focus here is a geometric formulation and quantitative sampling-error bounds. Using Stratonovich time reversal and an explicit Fubini–Study divergence formula, it derives a backward state-vector SDE with unchanged noise operators and drift corrections involving geometry and directional scores. Under smoothness and positivity assumptions, the construction applies in arbitrary finite dimension. Its ensemble average follows $-\mathcal L$ when initialized from the corresponding forward ensemble, not for arbitrary inputs. For depolarizing dynamics with $L_k=\sqrt\gamma\,\sigma_k$, the trajectory generator is $\gamma\Delta_g$, with uniform Fubini–Study stationary law and spectral gap $\lambda_1=2\gamma d$. With exact scores, reversal initialized from this prior has asymptotic errors $O(e^{-2\lambda_1T})$ in total variation and $O(e^{-4\lambda_1T})$ in relative entropy, with explicit leading coefficients. Girsanov arguments bound learned-score error by prior mismatch plus integrated squared score error. A rotation–diffusion scheme has discretization bounds derived from score martingales and Li–Yau Fisher-information control, requiring $O((d-1)\log^2(T/\delta)/\varepsilon)$ score evaluations for error tolerance $\varepsilon$. Numerical tests cover $d=2,3,4,6,8$, using symmetry-reduced distributions and prescribed score perturbations rather than trained models. Singular initial distributions require stopping at $\delta>0$ for the error guarantees. Non-depolarizing priors and jump unravellings remain open directions.
Central claim may not hold: The claimed general backward unravelling requires a smooth, positive ensemble density on the pure-state manifold, but valid Lindblad evolutions can preserve lower-dimensional supports where the theorem's full-manifold score is undefined, and the paper supplies no extension covering these ensembles.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.
- Comments
- 31 pages, 5 figures