arXiv:2610.06797
From Quantum Tunneling to Classical Optimization through Instantons
Jiayu Shen, Dylan Herman, Guneykan Ozgul, Akshay Seshadri, Junhyung Lyle Kim, Shree Hari Sureshbabu, Shouvanik Chakrabarti, Niraj Kumar
Abstract
We use an instanton-based analysis to show that Path Integral Stochastic Gradient Descent (PISGD), a classical optimization algorithm instantiated from path integral Monte Carlo (PIMC), can inherit the quantum tunneling mechanism. Quantum tunneling is the phenomenon in which a quantum particle passes through a potential barrier that a classical trajectory cannot cross in the same way. This mechanism can contribute to the performance of quantum optimization algorithms, since a common bottleneck in optimization is that the solution becomes trapped in a local minimum, separated from better solutions by a potential barrier. Classical noise-driven optimizers, such as stochastic gradient descent (SGD), have escape times that scale exponentially with the potential barrier height. For quantum tunneling, the escape time is instead exponential in the bounce action, which equals twice the Agmon distance between the two basins. When the barrier is tall but thin, quantum tunneling can allow escaping local minima faster than such classical methods. Remarkably, this scaling is not exclusive to algorithms with genuine quantum dynamics. For PISGD applied to classical nonconvex global optimization, we adapt Kramers' escape-rate theory to path space and obtain the same exponential scaling as quantum tunneling, in terms of the bounce action rather than the barrier height. We confirm this scaling numerically by implementing PISGD for both separable and nonseparable potentials. That said, PISGD retains only incoherent tunneling, since coherence among multiple tunneling paths is absent. Our analysis therefore does not rule out a potential quantum advantage for quantum-tunneling-based algorithms over classical ones in general.
AI Summary
The paper asks whether a classical gradient-based optimizer can reproduce tunneling-assisted escape in continuous nonconvex optimization. Path Integral Stochastic Gradient Descent (PISGD) applies noisy gradient descent to a discretized Euclidean action: each iterate is a periodic worldline, with nearest-neighbor coupling between its time slices. An instanton analysis identifies the bounce as the escape saddle in path space. Adapting Kramers–Langer theory to its translational zero mode gives the leading escape-time scaling $\mathcal T_{\mathrm{PISGD}}\sim e^{\tilde\lambda B_V}$, matching semiclassical incoherent tunneling, rather than SGD’s $e^{\tilde\lambda H_V}$. Here $B_V$ is twice the relevant Agmon distance and $H_V$ is the barrier height. The stationary single-slice marginal approaches the quantum ground-state density in the continuum, long-imaginary-time limits. For identical separable components with global-minimum curvature $(\omega^\star)^2$, imposing concentration within radius $\varepsilon$ yields exponents $dH_V/[\varepsilon^2(\omega^\star)^2]$ for SGD and $dB_V/(2\varepsilon^2\omega^\star)$ for PISGD. Thus PISGD has the smaller exponent when $B_V<2H_V/\omega^\star$. Numerical tests use modified quartics, Gaussian-spike barriers, and a nonseparable squared-quartic potential; the spike example has an SGD exponent about $8.5$ times larger. Functional-determinant calculations also supply escape-rate prefactors. The analysis is partly physics-style rather than fully rigorous. Explicit concentration-adjusted results primarily concern separable landscapes; the nonseparable scaling is hypothesized and tested numerically. Comparisons use fixed noise, not optimized annealing schedules, and reported runtimes omit the per-iteration cost of evaluating all slices. PISGD lacks coherent interference and can encounter QMC topological obstructions, leaving general tunneling-based quantum advantages unresolved.
Large language model tools were used interactively at all stages of the preparation of this manuscript: from the earliest exploratory research stages to the final exposition and writing steps.
Central claim may not hold: The claimed equality between PISGD and quantum escape-time exponents is not established for the stated closed, confining quantum systems, because the assumed irreversible exponential quantum decay does not generally follow from their unitary dynamics. The abstract overstates the nonseparable numerical confirmation: the proposed maximal-bounce-action scaling is hypothetical, and the reported data do not distinguish its exponent from the alternative predictions.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.