Challenges
Datasets
Workspaces
Papers
Discussions
Leaderboard
Log inSign up
Challenges
Datasets
Workspaces
Papers
Discussions
Leaderboard
Blog
Job Board
Q3AS

© 2026 Aqora Quantum S.A.S.

TermsPrivacyLegal Notice

arXiv:2601.04685 · APS Open Science

Regularization from Superpositions of Time Evolutions

Yakir Aharonov, Eliahu Cohen, Tomer Shushi

quant-phSubmitted Jan 8, 2026• Announced Jan 9, 2026• 0 citations• Revised May 18, 2026 (v2)
v1 · Jan 8, 2026v2 · May 18, 2026
arXivHTMLPDF

Abstract

Short-time approximations and path integrals can be dominated by high-energy or large-field contributions, especially in the presence of singular interactions, motivating regulators that are suppressive yet removable. Standard regulators typically impose such suppressions by hand (e.g. cutoffs, higher-derivative terms, heat-kernel smearing, lattice discretizations), while here we show that closely related smooth filters can arise as the conditional map produced by interference in a coherently controlled, postselected superposition of evolutions. A successful postselection implements a single heralded operator that is a coherent linear combination of time-evolution operators. For a Gaussian superposition of time translations in quantum mechanics, the postselected step is $V_{σ,Δt}=e^{-iHΔt}\,e^{-\frac12σ^2Δt^2H^2}$, i.e.\ the desired unitary step multiplied by a Gaussian energy filter suppressing energies above order $1/(σΔt)$. This renders short-time kernels in time-sliced path-integral approximations well behaved for singular potentials, while the target unitary dynamics is recovered as $σ\to0$ and (for fixed $σ$) also as $Δt\to0$ at fixed $t$. In scalar QFT, a local Gaussian smearing of the quartic coupling induces a positive $(σ^2/2)φ^8$ term in the Euclidean action, providing a symmetry-compatible large-field stabilizer; it is naturally viewed as an irrelevant operator whose effects can be renormalized at fixed $σ$ (together with a conventional UV regulator) and removed by taking $σ\to0$. We give short-time error bounds and analyze multi-step success probabilities.

Journal reference
APS Open Science 1, 000049 (2026)
DOI
10.1103/4yq5-y6wf
Comments
Updated version, 13 pages

Related papers

Numerical studies on real time evolution in quantum systems - Quantum decoherence and the beginning of the universe

Naoyuki Yamamori

Real-time path integrals provide a natural formulation of quantum dynamics also in situations where a Hamiltonian description is unavailable or impractical, but their direct evaluation is obstructed by severe phase cancellation from the oscillatory weight.A guiding theme of this thesis is that, after complexifying the integration variables, real-time path integrals are naturally organized by complex saddle points.In favorable cases the original contour can be replaced by steepest-descent manifolds (Lefschetz thimbles) attached to the relevant saddles, providing a controlled saddle-based description of real-time observables.In Part I, we study quantum decoherence in a closed-system real-time path-integral formulation: the total system (subsystem plus environment) evolves unitarily, and the environment is traced out only at the final time to obtain the reduced density matrix.For the Caldeira-Leggett model with Gaussian initial conditions, the discretized effective action becomes a quadratic form, so that the reduced density matrix is characterized by a complex saddle point and the thimble integral reduces to a finite-dimensional Gaussian integral.The complex saddle is obtained by solving a large but sparse linear system, allowing an exact evaluation of the reduced density matrix without relying on the usual master-equation approximations.We then extend this methodology to a three-dimensional cubic lattice toy model by separating center-of-mass and relative coordinates, treating the center-of-mass mode as the subsystem and the relative modes as an effective environment.This enables controlled numerical studies of how the suppression of off-diagonal matrix elements depends on the coupling, inverse temperature, and system size, and allows us to extract the early-time scaling behavior of decoherence diagnostics from the complex-saddle evaluation.In Part II, we investigate spacetime dynamics in the Lorentzian IKKT matrix model.To define a Lorentzian formulation that is not continuously connected to the Euclidean model by the standard contour deformation, we include a Lorentz invariant mass term with coefficient γ, for which the classical equations admit real, smooth, expanding solutions when γ > 0. Since the Lorentz group is noncompact, the path integral can contain a divergent volume from the boost directions; we remove this volume nonperturbatively by implementing a boost "gauge-fixing" that selects the frame minimizing tr(A 0 ) 2 , leading to the condition tr(A 0 A i ) = 0 with the associated Faddeev-Popov determinant.We then apply the complex Langevin method to the resulting complex-weight matrix integral.Time and spatial observables are defined from matrix configurations, with holomorphic time ordering implemented by auxiliary variables, and spatial structure characterized through block observables such as the extent R 2 (t) and the moment-of-inertia tensor.In the bosonic model we find approximately real time and no indication of spontaneous breaking of the spatial symmetry in the i parameter range studied.Including fermions, we control the singular-drift problem induced by the Pfaffian by introducing a fermionic mass regulator m f and an anisotropic mass deformation, which enhances fermionic effects while maintaining numerical stability.In this setup we observe a real, expanding spatial extent and a dynamical dominance of three spatial directions at late times, consistent with a (3 + 1)-dimensional expanding spacetime.We finally discuss the intended limiting procedure-taking the large-N limit first and then approaching the undeformed theory-and comment on future directions, including a comparison with partially supersymmetric deformations that preserve 16 supercharges.

•Mar 24, 2026•0 citations

Path Integrals and Singular Potentials: A Novel Regularization for Heat Kernel Asymptotics and Brownian Motion

Revista, Zen, MATH, 10

We present a novel regularization scheme for path integrals applied to quantum mechanical systems with singular potentials. Our approach focuses on the heat kernel asymptotics and their connection to Brownian motion. We introduce a short-time expansion for the heat kernel, incorporating the effects of singular potentials through a carefully constructed effective potential. This effective potential is derived using a combination of distributional calculus and stochastic calculus, ensuring that the resulting path integral is well-defined and produces physically meaningful results. The regularization method is then applied to specific examples of singular potentials, such as the inverse square potential, demonstrating its effectiveness in resolving the ambiguities associated with these potentials. We further explore the implications of our regularization scheme for the behavior of Brownian motion in the presence of singular potentials, connecting the quantum mechanical description to the classical stochastic process.

•Dec 5, 2025•0 citations

Existence of real time quantum path integrals

Job Feldbrugge, Neil Turok

Many interesting physical theories have analytic classical actions. We show how Feynman’s path integral may be defined non-perturbatively, for such theories, without a Wick rotation to imaginary time. We start by introducing a class of smooth regulators which render interference integrals absolutely convergent and thus unambiguous. The analyticity of the regulators allows us to use Cauchy’s theorem to deform the integration domain onto a set of relevant, complex “thimbles” (or generalized steepest descent contours) each associated with a classical saddle. The regulator can then be removed to obtain an exact, non-perturbative representation. We show why the usual method of gradient flow, used to identify relevant saddles and steepest descent “thimbles” for finite-dimensional oscillatory integrals, fails in the infinite-dimensional case. For the troublesome high frequency modes, we replace it with a method we call “eigenflow” which we employ to identify the infinite-dimensional, complex “eigenthimble” over which the real time path integral is absolutely convergent. We then bound the path integral over high frequency modes by the corresponding Wiener measure for a free particle. Using the dominated convergence theorem we infer that the interacting path integral defines a good measure. While the real time path integral is more intricate than its Euclidean counterpart, it is superior in several respects. It seems particularly well-suited to theories such as quantum gravity where the classical theory is well developed but the Euclidean path integral does not exist.

arXiv:2207.12798•Annals of Physics•Apr 20, 2023•21 citations

A Rigorous Path Integral Construction in any Dimension

Alexander Dynin

We propose a new rigorous time-slicing construction of the phase space Path Integrals for propagators both in Quantum Mechanics and Quantum Field Theory for a fairly general class of quantum observables (e.g. the Schroedinger hamiltonians with smooth scalar potentials of any power growth). Moreover we allow time-dependent hamiltonians and a great variety of discretizations, in particular, the standard, Weyl, and normal ones.

arXiv:math/9802058•Feb 11, 1998•0 citations

Gate-Level Quantum Simulation of Nonunitary Linear Dynamics with Hybrid Oscillator-Qubit Architecture

Elin Ranjan Das, Muqing Zheng, Rishab Dutta +3 more

We translate the one-mode unitary-dilation framework for nonunitary linear dynamics into a gate-level hybrid oscillator--qubit architecture. An ancillary oscillator encodes the kernel through state preparation and postselection, while a qubit register simulates the discretized system operator. The construction applies to time-independent dynamics $\dot u=-(L+iH)u$, including discretized partial differential equations, and removes the $\mathcal{O}(\log M_a)$ ancilla-qubit overhead of a discrete-variable (DV) $M_a$-term quadrature register. We bound the ideal state's squeezed-Fock coefficient-projection error, which decays superalgebraically in cutoff $N$ for Schwartz-class kernels and at a stretched-exponential rate under stronger joint decay and smoothness assumptions. The finite squeezed-Fock state generically has stellar rank $N-1$, making $N$ a measure of the oracle's non-Gaussian resource. For oscillator dimension $N_{\mathrm{Fock}}$, a $p$th-order product formula reaches error $ε_t$ in $\mathcal{O}(t^{1+1/p}N_{\mathrm{Fock}}^{(p+1)/(2p)}ε_t^{-1/p})$ Trotter steps, up to generator-dependent commutator factors. A perturbation bound separates total scaled-map error from postselection probability. We benchmark Law--Eberly synthesis and assess variational SNAP+$\mathcal D$ state preparation on heat and advection--diffusion equations. At the prescribed DV sizing, the hybrid CV--DV route has smaller fixed-scale error in all ten instances, while the DV route accepts with fewer repetitions in every instance. These results quantify when a continuous qumode can replace a discretized ancilla register.

quant-pharXiv:2605.10708•Quantum Science and Technology•May 12, 2026•0 citations

From Relative Couplings to Configuration-Space Histories

Qingchun Zhang

A finite amplitude dynamics, a configuration-space propagator, and a mechanical action are different mathematical objects. We give a conditional one-dimensional bridge between them while retaining the assumptions that make each identification possible. Common-additive-shift invariance selects a relative two-mode quadratic energy; a supplied first-order kinetic term and action scale turn its graph Laplacian into reciprocal norm-preserving dynamics. The resulting amplitude action vanishes on solutions and therefore cannot simply be identified with the configuration action. A convergent ordered-jump expansion retains diagonal dwell phases, hopping factors, and time-simplex measure. On a supplied spatial lattice, inverse-square scaling gives second-order fixed-band convergence and strong convergence for each fixed square-integrable state, but not all-mode real-time operator-norm convergence. The raw return-kernel density has two distinct mesh-subsequence limits. A declared Gaussian resolution instead gives an explicit uniform lattice-to-continuum bound, separate finite-volume control, and locally uniform removal of resolution at nonzero time. With independently declared mass and action calibration, the limit has the free mechanical action and exact finite-slice normalization. For smooth positive variable couplings, midpoint edges and uniform site measure select a divergence-form operator. Different measures or orderings retain the same classical principal symbol while changing lower-order terms. The Liouville coordinate and amplitude transformation fixes a local correction, endpoint factors, and the internal measure; these must be carried together. Smooth-orbit lattice and state-level time-slicing estimates give controlled comparisons without asserting a raw variable-coefficient kernel limit. The contribution is the organized conditional bridge and its obstructions, not the invention of the standard component methods or a derivation of geometry, the action-scale value, physical couplings, or empirical probability from primitive complementarity.

•Sep 28, 2026•0 citations
Short-time approximations and path integrals can be dominated by high-energy or large-field contributions, especially in the presence of singular interactions, motivating regulators that are suppressive yet removable. Standard regulators typically impose such suppressions by hand (e.g. cutoffs, higher-derivative terms, heat-kernel smearing, lattice discretizations), while here we show that closely related smooth filters can arise as the conditional map produced by interference in a coherently controlled, postselected superposition of evolutions. A successful postselection implements a single heralded operator that is a coherent linear combination of time-evolution operators. For a Gaussian superposition of time translations in quantum mechanics, the postselected step is Vσ,Δt=e−iHΔt e−12σ2Δt2H2V_{σ,Δt}=e^{-iHΔt}\,e^{-\frac12σ^2Δt^2H^2}Vσ,Δt​=e−iHΔte−21​σ2Δt2H2, i.e.\ the desired unitary step multiplied by a Gaussian energy filter suppressing energies above order 1/(σΔt)1/(σΔt)1/(σΔt). This renders short-time kernels in time-sliced path-integral approximations well behaved for singular potentials, while the target unitary dynamics is recovered as σ→0σ\to0σ→0 and (for fixed σσσ) also as Δt→0Δt\to0Δt→0 at fixed ttt. In scalar QFT, a local Gaussian smearing of the quartic coupling induces a positive (σ2/2)φ8(σ^2/2)φ^8(σ2/2)φ8 term in the Euclidean action, providing a symmetry-compatible large-field stabilizer; it is naturally viewed as an irrelevant operator whose effects can be renormalized at fixed σσσ (together with a conventional UV regulator) and removed by taking σ→0σ\to0σ→0. We give short-time error bounds and analyze multi-step success probabilities.
Real-time path integrals provide a natural formulation of quantum dynamics also in situations where a Hamiltonian description is unavailable or impractical, but their direct evaluation is obstructed by severe phase cancellation from the oscillatory weight.A guiding theme of this thesis is that, after complexifying the integration variables, real-time path integrals are naturally organized by complex saddle points.In favorable cases the original contour can be replaced by steepest-descent manifolds (Lefschetz thimbles) attached to the relevant saddles, providing a controlled saddle-based description of real-time observables.In Part I, we study quantum decoherence in a closed-system real-time path-integral formulation: the total system (subsystem plus environment) evolves unitarily, and the environment is traced out only at the final time to obtain the reduced density matrix.For the Caldeira-Leggett model with Gaussian initial conditions, the discretized effective action becomes a quadratic form, so that the reduced density matrix is characterized by a complex saddle point and the thimble integral reduces to a finite-dimensional Gaussian integral.The complex saddle is obtained by solving a large but sparse linear system, allowing an exact evaluation of the reduced density matrix without relying on the usual master-equation approximations.We then extend this methodology to a three-dimensional cubic lattice toy model by separating center-of-mass and relative coordinates, treating the center-of-mass mode as the subsystem and the relative modes as an effective environment.This enables controlled numerical studies of how the suppression of off-diagonal matrix elements depends on the coupling, inverse temperature, and system size, and allows us to extract the early-time scaling behavior of decoherence diagnostics from the complex-saddle evaluation.In Part II, we investigate spacetime dynamics in the Lorentzian IKKT matrix model.To define a Lorentzian formulation that is not continuously connected to the Euclidean model by the standard contour deformation, we include a Lorentz invariant mass term with coefficient γ, for which the classical equations admit real, smooth, expanding solutions when γ > 0. Since the Lorentz group is noncompact, the path integral can contain a divergent volume from the boost directions; we remove this volume nonperturbatively by implementing a boost "gauge-fixing" that selects the frame minimizing tr(A 0 ) 2 , leading to the condition tr(A 0 A i ) = 0 with the associated Faddeev-Popov determinant.We then apply the complex Langevin method to the resulting complex-weight matrix integral.Time and spatial observables are defined from matrix configurations, with holomorphic time ordering implemented by auxiliary variables, and spatial structure characterized through block observables such as the extent R 2 (t) and the moment-of-inertia tensor.In the bosonic model we find approximately real time and no indication of spontaneous breaking of the spatial symmetry in the i parameter range studied.Including fermions, we control the singular-drift problem induced by the Pfaffian by introducing a fermionic mass regulator m f and an anisotropic mass deformation, which enhances fermionic effects while maintaining numerical stability.In this setup we observe a real, expanding spatial extent and a dynamical dominance of three spatial directions at late times, consistent with a (3 + 1)-dimensional expanding spacetime.We finally discuss the intended limiting procedure-taking the large-N limit first and then approaching the undeformed theory-and comment on future directions, including a comparison with partially supersymmetric deformations that preserve 16 supercharges.
We present a novel regularization scheme for path integrals applied to quantum mechanical systems with singular potentials. Our approach focuses on the heat kernel asymptotics and their connection to Brownian motion. We introduce a short-time expansion for the heat kernel, incorporating the effects of singular potentials through a carefully constructed effective potential. This effective potential is derived using a combination of distributional calculus and stochastic calculus, ensuring that the resulting path integral is well-defined and produces physically meaningful results. The regularization method is then applied to specific examples of singular potentials, such as the inverse square potential, demonstrating its effectiveness in resolving the ambiguities associated with these potentials. We further explore the implications of our regularization scheme for the behavior of Brownian motion in the presence of singular potentials, connecting the quantum mechanical description to the classical stochastic process.
Many interesting physical theories have analytic classical actions. We show how Feynman’s path integral may be defined non-perturbatively, for such theories, without a Wick rotation to imaginary time. We start by introducing a class of smooth regulators which render interference integrals absolutely convergent and thus unambiguous. The analyticity of the regulators allows us to use Cauchy’s theorem to deform the integration domain onto a set of relevant, complex “thimbles” (or generalized steepest descent contours) each associated with a classical saddle. The regulator can then be removed to obtain an exact, non-perturbative representation. We show why the usual method of gradient flow, used to identify relevant saddles and steepest descent “thimbles” for finite-dimensional oscillatory integrals, fails in the infinite-dimensional case. For the troublesome high frequency modes, we replace it with a method we call “eigenflow” which we employ to identify the infinite-dimensional, complex “eigenthimble” over which the real time path integral is absolutely convergent. We then bound the path integral over high frequency modes by the corresponding Wiener measure for a free particle. Using the dominated convergence theorem we infer that the interacting path integral defines a good measure. While the real time path integral is more intricate than its Euclidean counterpart, it is superior in several respects. It seems particularly well-suited to theories such as quantum gravity where the classical theory is well developed but the Euclidean path integral does not exist.
We translate the one-mode unitary-dilation framework for nonunitary linear dynamics into a gate-level hybrid oscillator--qubit architecture. An ancillary oscillator encodes the kernel through state preparation and postselection, while a qubit register simulates the discretized system operator. The construction applies to time-independent dynamics u˙=−(L+iH)u\dot u=-(L+iH)uu˙=−(L+iH)u, including discretized partial differential equations, and removes the O(log⁡Ma)\mathcal{O}(\log M_a)O(logMa​) ancilla-qubit overhead of a discrete-variable (DV) MaM_aMa​-term quadrature register. We bound the ideal state's squeezed-Fock coefficient-projection error, which decays superalgebraically in cutoff NNN for Schwartz-class kernels and at a stretched-exponential rate under stronger joint decay and smoothness assumptions. The finite squeezed-Fock state generically has stellar rank N−1N-1N−1, making NNN a measure of the oracle's non-Gaussian resource. For oscillator dimension NFockN_{\mathrm{Fock}}NFock​, a pppth-order product formula reaches error εtε_tεt​ in O(t1+1/pNFock(p+1)/(2p)εt−1/p)\mathcal{O}(t^{1+1/p}N_{\mathrm{Fock}}^{(p+1)/(2p)}ε_t^{-1/p})O(t1+1/pNFock(p+1)/(2p)​εt−1/p​) Trotter steps, up to generator-dependent commutator factors. A perturbation bound separates total scaled-map error from postselection probability. We benchmark Law--Eberly synthesis and assess variational SNAP+D\mathcal DD state preparation on heat and advection--diffusion equations. At the prescribed DV sizing, the hybrid CV--DV route has smaller fixed-scale error in all ten instances, while the DV route accepts with fewer repetitions in every instance. These results quantify when a continuous qumode can replace a discretized ancilla register.
A finite amplitude dynamics, a configuration-space propagator, and a mechanical action are different mathematical objects. We give a conditional one-dimensional bridge between them while retaining the assumptions that make each identification possible. Common-additive-shift invariance selects a relative two-mode quadratic energy; a supplied first-order kinetic term and action scale turn its graph Laplacian into reciprocal norm-preserving dynamics. The resulting amplitude action vanishes on solutions and therefore cannot simply be identified with the configuration action. A convergent ordered-jump expansion retains diagonal dwell phases, hopping factors, and time-simplex measure. On a supplied spatial lattice, inverse-square scaling gives second-order fixed-band convergence and strong convergence for each fixed square-integrable state, but not all-mode real-time operator-norm convergence. The raw return-kernel density has two distinct mesh-subsequence limits. A declared Gaussian resolution instead gives an explicit uniform lattice-to-continuum bound, separate finite-volume control, and locally uniform removal of resolution at nonzero time. With independently declared mass and action calibration, the limit has the free mechanical action and exact finite-slice normalization. For smooth positive variable couplings, midpoint edges and uniform site measure select a divergence-form operator. Different measures or orderings retain the same classical principal symbol while changing lower-order terms. The Liouville coordinate and amplitude transformation fixes a local correction, endpoint factors, and the internal measure; these must be carried together. Smooth-orbit lattice and state-level time-slicing estimates give controlled comparisons without asserting a raw variable-coefficient kernel limit. The contribution is the organized conditional bridge and its obstructions, not the invention of the standard component methods or a derivation of geometry, the action-scale value, physical couplings, or empirical probability from primitive complementarity.