PaperlayerView exact revision on arXiv

Source-linked AI summary

Koopman-von Neumann Approach to Quantum Simulation of Nonlinear Classical Dynamics

Ilon Joseph

arXiv:2003.09980v4quant-phmath-phmath.MPphysics.class-phphysics.comp-ph

TL;DR

Classical-dynamics simulation consumes substantial computational resources, raising the question of whether quantum computers can improve its efficiency. The paper uses the generalized Koopman-von Neumann formulation to represent classical phase-space evolution as unitary dynamics on Hilbert space. It reports exponential savings over Eulerian discretization under a sparsity condition and a quadratic improvement over classical probabilistic algorithms.

  • Problem

    Whether quantum computers can improve classical-dynamics simulation matters because substantial computational resources are devoted to simulating classical dynamics.

    Source: A. Motivation · p. 1

  • Method

    The generalized Koopman-von Neumann formulation recasts Liouville evolution as a Schrödinger equation on Hilbert space with a Hermitian Hamiltonian and unitary evolution.

    Source: D. Koopman-von Neumann Approach · p. 2, §VI · p. 15

  • Results

    For sparse KvN Hamiltonians, quantum simulation saves exponentially over deterministic Eulerian discretization of the Liouville equation and quadratically over classical probabilistic Monte Carlo algorithms.

    Source: F. Quantum Simulation · p. 4, §VI · p. 15

  • Takeaways & Limitations

    A finite-resource quantum computer can approximate the unitary evolution operator governing the phase-space probability distribution.

    Source: §VI · p. 15

    When the underlying system is Hamiltonian and simulating its quantized version suffices, that approach is more efficient because KvN doubles the phase-space dimension.

    Source: C. Complexity Estimates · p. 14

Abstract

from arXiv · show

Quantum computers can be used to simulate nonlinear non-Hamiltonian classical dynamics on phase space by using the generalized Koopman-von Neumann formulation of classical mechanics. The Koopman-von Neumann formulation implies that the conservation of the probability distribution function on phase space, as expressed by the Liouville equation, can be recast as an equivalent Schr\"odinger equation on Hilbert space with a Hermitian Hamiltonian operator and a unitary propagator. This Schr\"odinger equation is linear in the momenta because it derives from a constrained Hamiltonian system with twice the classical phase space dimension. A quantum computer with finite resources can be used to simulate a finite-dimensional approximation of this unitary evolution operator. Quantum simulation of classical dynamics is exponentially more efficient than a deterministic Eulerian discretization of the Liouville equation if the Koopman-von Neumann Hamiltonian is sparse. Utilizing quantum walk techniques for state preparation and amplitude estimation for the calculation of observables leads to a quadratic improvement over classical probabilistic Monte Carlo algorithms.

I. INTRODUCTION

The introduction asks whether quantum computers can improve the simulation of classical dynamics, then presents the generalized Koopman-von Neumann formulation as a unitary Hilbert-space representation for classical phase-space evolution. It outlines the method’s scope and the conditions under which quantum simulation can reduce computational work.

A. Motivation

  • The introduction asks whether quantum computers can provide efficiency gains for classical-dynamics simulations, which consume substantial computational resources.It notes that nonlinear classical systems are often non-Hamiltonian and that a prior approach has costs exponential in time steps and nonlinearity degree.Source: A. Motivation · p. 1

B. Comparison of Classical and Quantum Resource Requirements

  • For Hamiltonian systems, Eulerian discretization of the Liouville equation requires work and memory scaling as ∼L^D, where D=2Md for M particles in d dimensions.The passage contrasts this with the similar resource requirements of probabilistic classical and quantum system descriptions.Source: B. Comparison of Classical and Quantum… · p. 2

C. Quantization of Classical Dynamics

  • Direct quantization can be cheaper because its Schrödinger PDE uses half the classical phase-space dimension, but finite-ℏ dynamics differ through tunnelling, interference, and localization.The passage also notes that finding a natural quantization procedure for general non-Hamiltonian systems is not straightforward.Source: C. Quantization of Classical Dynamics · p. 2

D. Koopman-von Neumann Approach

  • The generalized KvN formulation recasts Liouville evolution as a Schrödinger equation for a complex amplitude, with a Hermitian Hamiltonian linear in momentum.It applies to arbitrary classical systems and arises from a constrained Hamiltonian system with twice the classical phase-space dimension.Source: D. Koopman-von Neumann Approach · pp. 2–3

E. Semiclassical Evolution

  • The phase factor is a gauge choice for classical dynamics; its semiclassical form is tied to the classical action and satisfies the Hamilton-Jacobi equation.Developing a self-consistent framework for semiclassical dynamics of Hamiltonian ODEs is left for future work.Source: E. Semiclassical Evolution · p. 3

F. Quantum Simulation

  • Finite-resource quantum simulation approximates the KvN Hamiltonian’s unitary evolution after discretizing phase space into a finite-qubit representation.The discretization produces a quantum-mechanical coarse-graining and regularization distinct from dissipative effects such as particle collisions.Source: F. Quantum Simulation · pp. 3–4
  • A quadratic speedup over classical probabilistic algorithms is obtained by combining quantum-walk state preparation with amplitude estimation for observables.For sparse KvN Hamiltonians, the quantum representation also saves exponentially over deterministic Eulerian discretization of the Liouville equation.Source: F. Quantum Simulation · p. 4

G. Outline

The paper develops a Koopman-von Neumann representation that recasts classical phase-space evolution as a Hilbert-space equation. It then outlines its treatment of general dynamics, the KvN Hamiltonian, and the paper’s later topics.

G. Outline

  • The paper proceeds from general phase-space dynamics to the generalized KvN Hamiltonian, then treats Hamiltonian systems, conservation laws, quantum simulation, complexity, and conclusions.The Einstein summation convention is used throughout.Source: G. Outline · p. 4

A. Classical Dynamics on Phase Space

  • Classical dynamics can be viewed either by following trajectories or by evolving observables through phase-space PDEs; the Koopman operator propagates observables, while the Perron-Frobenius operator propagates the PDF.The Liouville equation describes conservation of phase-space density, and the PDF is an invariant measure of the flow.Source: A. Classical Dynamics on Phase Space · pp. 4–5

B. Koopman-von Neumann Hamiltonian

  • The generalized KvN equation represents the PDF through a complex amplitude and has a Hermitian Hamiltonian, yielding unitary evolution for arbitrary ODE systems, including non-Hamiltonian ones.Its Hamiltonian is linear in momentum, and the Heisenberg equations for the original phase-space variables match the classical equations of motion.Source: B. Koopman-von Neumann Hamiltonian · pp. 5–6

C. Koopman-von Neumann Evolution Operator

The KvN propagator evolves the amplitude using the classical flow and its Jacobian. The phase factor is immaterial to classical dynamics but must be tracked for semiclassical applications, where multiple paths can contribute distinct phases.

C. Koopman-von Neumann Evolution Operator

  • The KvN evolution operator is constructed from the classical solution, the PDF, and the phase to determine the evolving amplitude.The phase shift from the square root of the Jacobian is indexed by the number of Jacobian zeros along the trajectory.Source: C. Koopman-von Neumann Evolution… · p. 6

D. Semiclassical Dynamics and the Phase Factor

  • The phase has no effect on classical dynamics, but it matters for semiclassical interference and coupled classical-quantum systems; extending the amplitude with indices can represent Poincaré-group components.The proposed higher-spin and gauge-field extensions leave classical dynamics unchanged, while their semiclassical dynamics is identified as future work.Source: D. Semiclassical Dynamics and the… · p. 7
  • For Hamiltonian systems, choosing the phase velocity from the classical Lagrangian makes the phase factor equal to the classical action and yields the semiclassical phase.The phase also satisfies the Hamilton-Jacobi equation.Source: D. Semiclassical Dynamics and the… · p. 7
  • For longer time intervals, multiple classical paths can contribute distinct phases, so the Maslov-index phase jumps must be accounted for in the semiclassical phase factor.The phase shift is Δϕ = −νπ/2, where ν counts zeros of the Jacobian along the trajectory.Source: C. Koopman-von Neumann Evolution… · p. 6, D. Semiclassical Dynamics and the… · p. 7

E. Constrained Hamiltonian

The constrained Hamiltonian generates arbitrary classical ODEs using momenta as Lagrange multipliers. The section then illustrates the construction with general examples and notes how other constraint types could alter the quantized dynamics.

E. Constrained Hamiltonian

  • In the ℏ→0 limit, the KvN Hamiltonian becomes H(x, P, t) = P · v(x, t) + W(x, t), linear in momentum and corresponding to a constrained Hamiltonian system.Variation with respect to the multipliers enforces the coordinate equations of motion; variation with respect to coordinates gives the multiplier equations.Source: E. Constrained Hamiltonian · pp. 7–8
  • Weyl quantization of the constrained Hamiltonian yields the KvN Hamiltonian, while holonomic constraints would require Dirac brackets and could introduce nontrivial commutators and deviations from exact classical dynamics.For the constraints considered here, the quantized original coordinates commute.Source: E. Constrained Hamiltonian · p. 8

F. General Examples

  • For the linear examples, Heisenberg’s equations match the classical equations of motion, and a Hermitian block off-diagonal quadratic Hamiltonian produces symplectic block-diagonal equations.The examples include linear dynamics with exponential growth or damping and a general linear equation set.Source: F. General Examples · p. 8

III. HAMILTONIAN DYNAMICS

The canonical formulation uses Lagrange multipliers alongside the original coordinates and momenta. Its examples show how the KvN construction represents both harmonic and integrable nonlinear oscillator dynamics.

A. Canonical Hamiltonian Systems

  • The KvN formulation promotes Poisson brackets of the Lagrange multipliers to commutation relations, while the uncertainty principle applies to the original-variable and multiplier pairs, not pairs of original variables.Hamilton’s equations are divergence-free in canonical coordinates, with canonical measure ddqddp.Source: A. Canonical Hamiltonian Systems · pp. 8–9

B. Canonical Hamiltonian Examples

  • For the harmonic oscillator, the extended phase-space dynamics consists of two harmonic oscillators rotating in the same direction, with a corresponding quantum Hamiltonian.The constrained Hamiltonian exactly represents the dynamics.Source: B. Canonical Hamiltonian Examples · p. 9
  • For integrable nonlinear oscillators, the number operators are conserved like the action coordinates, while the angle operators obey Heisenberg equations of motion.Source: B. Canonical Hamiltonian Examples · p. 9

C. Generalized Hamiltonian Systems

Generalized Hamiltonian and variational systems extend the Koopman-von Neumann construction beyond canonical Hamiltonian equations. The variational formulation is equivalent to the generalized Hamiltonian form when the symplectic 2-form is non-degenerate.

C. Generalized Hamiltonian Systems

  • Generalized Hamiltonian systems are specified by H(x,t) and a Poisson bracket whose tensor is antisymmetric and satisfies the Jacobi identity.The construction gives a corresponding KvN Hamiltonian; the generalized equations also admit a KvN equation treating the PDF and wavefunction as scalar fields when the equations are non-singular.Source: C. Generalized Hamiltonian Systems · pp. 9–10

D. General Variational Systems

  • In the variational formulation, the action uses the Poincaré 1-form, whose exterior derivative defines the symplectic 2-form and Lagrange bracket.The Poisson tensor is the inverse of the Lagrange form, and the Jacobi identity follows because the symplectic form is closed.Source: D. General Variational Systems · p. 10
  • The generalized Hamiltonian form and the variational form are equivalent when the symplectic 2-form is non-degenerate.The variational equations can also be embedded in Hamilton’s equations on extended phase space by adding coordinates (τ, pτ) and using ¯H = H + pτ.Source: D. General Variational Systems · p. 10
  • The variational equations follow from a constrained Hamiltonian in which Lagrange multipliers enforce the equations of motion.The corresponding KvN Hamiltonian is constructed using a Hermitian form for the analogous quantized operators; a canonical volume form permits scalar-field PDF and wavefunction treatment.Source: D. General Variational Systems · p. 10

IV. WAVE ACTION PRINCIPLE AND CONSERVATION LAWS

The wave action principle generates the KvN wavefunction equation as well as the Liouville equation and its adjoint. For canonical Hamiltonian systems, a semiclassical phase satisfying the Hamilton-Jacobi equation makes the resulting KvN Hamiltonian equal the classical energy.

A. Action Principle for General Systems

B. Action Principle for Canonical Hamiltonian Systems

  • Other action principles can yield alternative KvN phase-factor dynamics for further exploration.The discussion also considers a semiclassical phase chosen equal to the classical action and examines the difference between the resulting Hamiltonian expressions.Source: A. Action Principle for General Systems · p. 11, B. Action Principle for Canonical… · p. 11
  • The phase-density relation, once true initially, is preserved by the dynamics through the Jacobi identity.When both f and W depend only on adiabatic invariants, the Hamiltonian density is (H + W)f.Source: B. Action Principle for Canonical… · p. 11
  • With a semiclassical phase satisfying the Hamilton-Jacobi equation, the canonical-system KvN Hamiltonian expressions yield the classical energy.In action-angle coordinates with H0(J), the action is given by the resulting classical-energy expression; the condition also simplifies results of Ref..Source: B. Action Principle for Canonical… · p. 11

C. Symmetries and Conservation Laws

The section connects KvN symmetries to conserved densities, then discusses uncertainty, numerical discretization, and complexity. Its complexity estimates describe speedups over Eulerian discretization and Monte Carlo under stated conditions, while noting boundaries on those gains.

C. Symmetries and Conservation Laws

  • KvN number density is conserved under constant phase shifts; time-independent Hamiltonians and coordinate-independent Hamiltonians conserve the corresponding Hamiltonian and momentum densities.For a time-independent classical Hamiltonian, conservation of probability density also conserves classical energy density; coordinate independence similarly conserves momentum density.Source: C. Symmetries and Conservation Laws · p. 12

A. Heisenberg Uncertainty

  • The extended phase space allows simultaneous measurement of all original phase-space variables, or all Lagrange multipliers.The uncertainty principle applies to each original variable and its conjugate Lagrange multiplier, not to pairs of original variables.Source: A. Heisenberg Uncertainty · p. 12

B. Numerical Discretization

  • Finite-qubit simulation requires a finite-dimensional approximation of the KvN Hamiltonian, with periodic coordinates discretized into levels.The discussion also describes squeezed initial and final measurement states as a way to saturate the uncertainty limit set by numerical discretization.Source: B. Numerical Discretization · p. 12
  • Squeezed states reduce coordinate uncertainty to the numerical grid spacing, σxj = ∆xj, while increasing uncertainty in the conjugate multipliers.The phase-space area imposed by discretization is ∆xj∆Pj = h/Lj, below the Heisenberg limit; squeezed states can reach the discretization limit.Source: B. Numerical Discretization · p. 12, Figure 3 · p. 12

C. Complexity Estimates

  • Against Eulerian discretization, the approach offers exponential speedup in D and polynomial speedup in L; quantum-walk sampling reduces work from ε^-2 to ε^-1.The Eulerian comparison assumes a sufficiently sparse quantum Hamiltonian, and the dimension-dependent gains can fall to polynomial at best when many time steps are required.Source: C. Complexity Estimates · p. 13
  • Quantum sampling gives O(sDT/ε) complexity, while estimating observables by ordinary repeated averaging would retain the classical O(1/ε^2) sampling law.Amplitude amplification and estimation provide the quadratic sampling improvement; observable estimation requires repeated simulation, with four KvN simulations per step in the described procedure.Source: C. Complexity Estimates · pp. 13–14
  • The quadratic gain over classical time-dependent Monte Carlo requires relatively sparse initialization and output, rather than setting every initial condition or measuring the full PDF.The proposed pattern uses readily constructed initial states and a small number of measurements, such as low-order PDF moments; long simulation times can also reduce the Eulerian-comparison savings.Source: C. Complexity Estimates · pp. 13–14

VI. CONCLUSION

The generalized Koopman-von Neumann formulation recasts nonlinear, non-Hamiltonian classical dynamics as unitary evolution on Hilbert space, enabling finite-dimensional quantum simulation. Under sparsity, this approach improves on deterministic Eulerian discretization exponentially and on classical probabilistic algorithms quadratically, while the authors identify further algorithmic and example-specific work as future directions.

VI. CONCLUSION

  • The generalized Koopman-von Neumann formulation recasts phase-space Liouville dynamics as an equivalent Hilbert-space Schrödinger equation with a Hermitian Hamiltonian and unitary evolution.Finite-resource quantum computers can simulate a finite-dimensional approximation of this unitary evolution.Source: §VI · pp. 14–15
  • The equivalent Schrödinger equation is linear in momenta because it quantizes a constrained Hamiltonian system with twice the original phase-space dimension.Lagrange multipliers enforce the original equations of motion; squeezing uncertainty in the phase-space variables allows uncertainty to reach the numerical-discretization limit.Source: §VI · p. 15
  • A sparse Koopman-von Neumann Hamiltonian yields exponentially greater efficiency than deterministic Eulerian discretization and, up to polylogarithmic factors, an overall quadratic complexity improvement over classical probabilistic algorithms.Quantum walks can prepare many useful initial states quadratically faster than classical Monte Carlo, while quantum algorithms for sums and integrals give a quadratic speedup for calculating observables at fixed accuracy.Source: §VI · p. 15
  • Exploring the representation on specific classical systems and developing algorithms that accurately approximate the semiclassical evolution operator remain future directions.The authors identify both advantages and disadvantages on specific examples as an open topic.Source: §VI · p. 15

Appendix A: Probability Distribution Function as a Volume Form

Appendix A treats the probability distribution as a volume form and states how it transforms and is conserved under phase-space flow. Appendix B develops a scalar-field convention and the corresponding Hermitian Koopman-von Neumann operator, with a restriction for time-dependent coordinate transformations.

Appendix A: Probability Distribution Function as a Volume Form

  • The probability distribution is a volume form on the phase-space manifold, and its integral over the manifold is normalized to unit probability.The probability density form is defined within a region of the manifold.Source: Appendix A · p. 15
  • Under a coordinate change, the transformed probability distribution is related to the original through the Jacobian determinant.The volume density also transforms with the Jacobian.Source: Appendix A · p. 15
  • The Liouville equation expresses the probability distribution as an invariant volume form for the space-time velocity.In coordinates Xµ = (t, x), the space-time velocity is Vµ = (1, v), and the equation is proportional to its space-time divergence.Source: Appendix A · p. 15
  • Appendix A defines the gradient and divergence in its coordinate conventions and notes that the usual divergence convention applies when the PDF is treated as a scalar field.The latter convention is discussed in Appendix B.Source: Appendix A · p. 16

Appendix B: Probability Distribution Function as a Scalar Field

  • With a natural phase-space volume form, the PDF can be treated as a scalar field, and the Hilbert-space inner product is defined using that volume form.The corresponding scalar-field PDF and Liouville equation are introduced in this convention.Source: Appendix B · p. 16
  • For a time-independent coordinate-transformation Jacobian, dividing the Liouville equation by the Jacobian gives the scalar-field convention used in the main text.In this case, the main-body equations apply with f → F and ψ → Ψ.Source: Appendix B · p. 16
  • For time-dependent coordinate transformations, the Jacobian is generically time-dependent, so the Liouville equation cannot be divided through by it.In the scalar-field convention, Hermitian momentum and Koopman-von Neumann Hamiltonian operators are defined with respect to the relevant measures.Source: Appendix B · p. 16

Appendix C: Probability Distribution Function on Space-Time

Appendix C formulates the probability distribution and Koopman-von Neumann equation under space-time coordinate transformations, establishing Hermiticity for the corresponding operators. Appendix D derives conservation of the symplectic form and its implications for the Poincaré form and Poisson tensor.

Appendix C: Probability Distribution Function on Space-Time

  • Under a general space-time coordinate transformation, the space-time velocity and Jacobian determine the transformed Liouville equation and its divergence form.The PDF is a space-time volume form, while the conserved phase-space probability distribution is its spatial component.Source: Appendix C · p. 16
  • The appendix gives two wavefunction definitions for the transformed distribution and derives the associated Koopman-von Neumann equations.For the scalar-wavefunction definition, the resulting equation is Hermitian with respect to the specified Hilbert-space inner product.Source: Appendix C · pp. 16–17
  • The space-time-scalar wavefunction definition leads to the Koopman-von Neumann equations and Hamiltonian operators that are Hermitian with respect to their corresponding inner products.The appendix states this correspondence for the equations and operators in the main text and Appendix B.Source: Appendix C · pp. 16–17

Appendix D: Conservation of the Symplectic Form, Poincar´e Form, and Poisson Tensor

  • The symplectic form is conserved because the symplectic form is closed and the first two terms in its Lie-derivative expression cancel by the equations of motion.The appendix also derives conservation through the time derivative of the Poincaré form, which is the differential of the Lagrangian.Source: Appendix D · p. 17
  • Conservation of the symplectic 2-form implies conservation of the Poisson 2-vector because the latter is defined as its inverse.Source: Appendix D · p. 17

Appendix E: Symplectic Volume Form for Generalized Hamiltonian and Variational Systems

Appendix E defines a canonical symplectic volume form for generalized Hamiltonian and variational systems and relates its conservation to the flow and Koopman-von Neumann construction. Appendix F derives a conservation law from a symmetry of the equations of motion.

Appendix E: Symplectic Volume Form for Generalized Hamiltonian and Variational Systems

  • In generalized Hamiltonian and variational systems, the velocity is divergence-free with respect to a preferred canonical volume form.The appendix contrasts this with Hamilton’s equations, whose velocity is divergence-free in canonical coordinates.Source: Appendix E · p. 17
  • The symplectic form defines a canonical volume form, and conservation of the symplectic form implies conservation of that volume density by the flow.The appendix expresses this as dJ/dt = ∂tJ + ∇J · v = 0 and concludes that the space-time divergence vanishes with respect to the canonical volume form.Source: Appendix E · p. 17
  • The scalar-wavefunction Koopman-von Neumann Hamiltonian for generalized Hamiltonian mechanics is Hermitian with respect to the canonical measure used to define the inner product.Its Hermitian momentum operator is defined as the volume-weighted analogue of the operator in Appendix B.Source: Appendix E · p. 17
  • For general variational systems, the appendix also gives a Koopman-von Neumann Hamiltonian and relates it to the earlier Hamiltonian expression.Source: Appendix E · pp. 17–18
  • The generalized Hamiltonian formulation may be less suitable when the symplectic 2-form is degenerate or its closed-form expression is unknown or difficult to construct.The authors say the main-text conventions may be preferable in such cases.Source: Appendix E · p. 18

Appendix F: Symmetries and Noether’s Theorem

  • A symmetry of the equations of motion is defined by invariance of dA, and the appendix derives a conservation law by taking its inner product with the solution V.The symmetry is generated by a space-time vector U.Source: Appendix F · p. 18

Appendix G: Hamilton-Jacobi Equation

Appendix G constructs solutions of the Hamilton–Jacobi equation using action-angle coordinates and canonical transformations. With ϕ₀ = 0, the resulting Hamiltonian expressions yield the classical energy density, and the constraint f_H − f, if initially zero, remains zero.

Appendix G: Hamilton-Jacobi Equation

  • Along the trajectory, the action derivatives ∂ϕ/∂Jj vanish, and the Hamilton–Jacobi solution satisfies the associated partial differential equations.The complete-solution construction begins from the equations of motion in action-angle coordinates.Source: Appendix G · p. 18
  • In action-angle coordinates, H₀(J) depends only on the action variables, allowing an explicit action integral that satisfies the Hamilton–Jacobi equation and yields a canonical transformation.The action variables are constants of motion, and choosing ϕ₀ = 0 also makes the action integral satisfy the Hamilton–Jacobi equation.Source: Appendix G · p. 18
  • For another canonical coordinate set, a mixed-variable generating function S(q, J, t) defines the transformation and yields an action integral that can be integrated because J is constant.The resulting action integral satisfies the Hamilton–Jacobi equation when ϕ₀ = 0.Source: Appendix G · p. 18
  • Choosing ϕ₀ = 0 makes the final Hamiltonian expressions yield the classical energy density Hf, simplifying results in Ref..The passage identifies this equality for the expressions in Eqs. 95 and 96.Source: Appendix G · p. 18
  • The constraint f_H − f, if zero initially, remains zero because its time derivative vanishes when the semiclassical phase rate is identified with the classical Lagrangian.The derivation combines LV f = 0 with the stated evolution relations for J and α.Source: Appendix G · p. 18

Related papers

Loading 2003.09980v4…