Real-Time Path Integrals in Quantum Dynamics

• Real-time path integrals are a formulation of quantum dynamics that represent transition amplitudes as oscillatory sums over histories in Minkowski time, posing unique numerical challenges.
• Advanced techniques such as double exponential quadrature, Picard-Lefschetz thimble decomposition, and complex semiclassical approaches mitigate severe sign problems and convergence issues.
• These methods enable practical computations in scattering, quantum field theory, gravity, and non-Markovian open systems by addressing caustics and phase oscillations.

A real-time path integral is a representation of quantum dynamics where transition amplitudes or correlation functions are expressed as oscillatory sums over histories in Minkowski time. Unlike imaginary-time (Euclidean) path integrals, which are foundational to equilibrium quantum statistical mechanics and benefit from the positive-definite exponential weight, real-time path integrals encode unitary dynamics at the cost of highly oscillatory integrals and severe numerical sign problems. The modern research agenda for real-time path integrals is characterized by the development of computational frameworks, advances in semiclassical and thimble analytics, rigorous non-perturbative definitions, and algorithmic breakthroughs for non-equilibrium quantum systems.

1. Formal Structure and Definition

In non-relativistic quantum mechanics, the propagator $U(x_b, t_b; x_a, t_a)$ for a particle of mass $m$ in a potential $V(x)$ is given by the Feynman real-time path integral:

$$U(x_b, t_b; x_a, t_a) = \int \mathcal{D}x(t)\, \exp\left\{i \int_{t_a}^{t_b} \! dt\,\left[\tfrac{1}{2} m \dot{x}^2(t) - V(x(t))\right]\right\}$$

Discretizing the time interval with $N+1$ slices of size $\Delta T = (t_b-t_a)/(N+1)$ leads to

$$U_N(x_b, t_b; x_a, t_a) = \lim_{N \to \infty} \left(\frac{m}{2\pi i \Delta T}\right)^{(N+1)/2} \int dx_1 \ldots dx_N\, \exp\left\{ i \Delta T \sum_{j=1}^{N+1} \left[ \frac{m}{2}\left(\frac{x_j-x_{j-1}}{\Delta T}\right)^2 - \frac{V(x_j)+V(x_{j-1})}{2} \right] \right\}$$

with $x_0 = x_a$ and $x_{N+1} = x_b$ (Rosenfelder, 2021). The action exponent is purely imaginary, and thus the integrand oscillates violently on the real domain.

In the field-theoretic context, real-time path integrals underlie nonequilibrium quantum field theory, quantum information, high-harmonic generation, strong-field physics, and quantum gravity, appearing in both continuum and lattice formulations.

2. Oscillatory Integrals, Numerical Quadratures, and Convergence

The principal numerical challenge of real-time path integrals is the exponential cancellation arising from the rapidly oscillating phase, the "sign problem," which precludes naive Monte Carlo sampling.

Several lines of research focus on stable and efficient quadratures for oscillatory, high-dimensional integrals:

• Double Exponential (Ooura's) Quadrature: Applied to integrals of the form $I(\omega) = \int_0^\infty f(x) e^{i\omega x}\,dx$, Ooura's formula provides a decomposition that converges rapidly due to the double-exponential decay of the integration measure after a variable transform (Rosenfelder, 2021).
• Multidimensional Gauss-Fresnel and Hyperspherical Decomposition: For free particles, the multidimensional oscillatory integral $$\int_{\mathbb{R}^N} d^N y\, e^{i\omega \sum_k y_k^2}$$ naturally separates into radial and angular components. The radial integral is highly oscillatory and handled efficiently using DE quadrature, while angular integrals are computed via (complex) Gauss-Chebyshev or Monte Carlo VEGAS algorithms with call budgets $O(10^4{-}10^6)$ for the angular sector when $N\lesssim 20$, rendering many-body real-time integrals accessible for moderate $N$ (Rosenfelder, 2021).

Algorithmic control is achieved by imposing target tolerances (e.g., $\epsilon_{\text{Ooura}}\sim 10^{-8}$), adjusting the number of quadrature nodes ($k_\max$), and ensuring regularization near singularities (see Section 5).

3. Semiclassics, Complex Trajectories, and Maslov Indices

The $\hbar\to0$ limit of the real-time path integral is governed by stationary phase/saddle point contributions, corresponding to solutions of the complexified Euler–Lagrange equations. For quadratic actions, Gaussian integration yields exact propagators; for anharmonic or discontinuous potentials, complex semiclassical paths and caustic structures emerge.

• Classically Allowed and Forbidden Paths: Real solutions dominate classically allowed transitions; complex solutions (instantons) encode classically forbidden (tunneling or reflection) processes. These trajectories can pass through caustics (where real solutions coalesce and split into complex conjugate pairs), and their contributions are quantified by topology (intersection numbers) (Feldbrugge et al., 25 Aug 2025).
• Maslov Correction: For the harmonic oscillator, discretized analysis shows the propagator acquires a phase $-\pi/2$ each time the path passes through a focal point (caustic), as extracted from diagonalization of the quadratic form in discretized (path integral) variables using Chebyshev polynomials (Rosenfelder, 2021).
• Unsuppressed Complex Saddles: In discontinuous potentials (e.g., step potentials), complex semiclassical phase-space contributions can avoid exponential suppression and control phenomena such as quantum reflection in the high-energy limit (Feldbrugge et al., 25 Aug 2025).

Fourier and Laplace diagnostics of the numerically computed propagator allow detection of relevant semiclassical actions through peaks in transformed amplitudes, permitting direct identification of both real and complex contributing paths (Feldbrugge et al., 25 Aug 2025).

4. Picard-Lefschetz Theory, Thimble Decomposition, and Intersection Numbers

To render real-time path integrals numerically tractable and conceptually rigorous, one exploits the analytic properties of the action and employs complex contour deformations:

• Lefschetz Thimbles: The original oscillatory integral over $\mathbb{R}^L$ is deformed into a sum over steepest-descent (Lefschetz) thimbles $\mathcal{J}_\sigma$ attached to complex saddle points $z_\sigma$. On each thimble, $\operatorname{Im} S$ is constant, and the real part decays, taming the oscillations and converting the sign problem into an exponential decay (Shoji et al., 7 Oct 2025, Feldbrugge et al., 2022, Tanizaki et al., 2014, Cherman et al., 2014).
• Intersection Numbers: Only those thimbles with non-zero intersection number $n_\sigma$ with the original cycle contribute. Recent algorithms implement robust high-dimensional calculations of these $n_\sigma$ via multiple shooting methods, determining both magnitude and sign, even for $20$-dimensional discretized path integrals (Shoji et al., 7 Oct 2025).
• Sewed-Around-Saddle Methods: Hybrid techniques, such as the "sewed almost-Lefschetz thimble" approach, formally join analytically tractable Gaussian near-thimble regions to numerically constructed flow-based thimbles further away, yielding improved numerical stability and reduced phase fluctuations (Mou et al., 2024).

The correctness of these constructions is rigorously supported in polynomial theory settings, with absolute convergence ensured by analytic regulators and mode-by-mode "eigenflow" deformations (Feldbrugge et al., 2022).

5. Regularization, Damping, and Caustics

Integration near caustics (focal points) is hampered by singularities in the fluctuation determinants or spiking of the phase. To stabilize numerics:

• Analytic Regularization: An exponential damping factor $D_\eta(y)=\exp(-\eta y)$, with $\eta>0$ small, is introduced in the integrand, corresponding to shifting the integration variable into the complex (damped) plane. This is formally equivalent to regularizing denominators $1/(ξ-ξ_n^{(N)}+i0^+)\to 1/(ξ-ξ_n^{(N)}+i\eta)$, smoothing singularities and permitting integration (Rosenfelder, 2021).
• Convergence Control: The overall modulus of the prefactor is correspondingly suppressed by a factor $[(ξ-ξ_n)^2+\eta^2]^{-1/4}$, transforming the sharp caustic into a continuous, numerically tractable feature. One can then extrapolate $\eta\to 0^+$ for physical results (Rosenfelder, 2021).

These measures are essential for correct calculation of phase shifts, scattering amplitudes, and for the accurate evaluation of path-integral determinants in the vicinity of Stokes lines and caustics.

6. Applications: Scattering, Quantum Dynamics, and Beyond

The direct evaluation of real-time path integrals, now practical via the above methods, enables a variety of quantum applications:

• Scattering Amplitudes: For short-range potentials, scattering matrices (T-matrix elements) can be computed directly via real-time path-integral methods, using transfer-matrix factorizations and cylinder-set complex probability measures. This procedure bypasses the need for analytic continuation and allows the extraction of on-shell quantities from large time-limits and grid discretizations (Polyzou et al., 2018, Polyzou et al., 2017).
• Quantum Field Theory and Lattice Gauge Theory: Real-time lattice gauge theory has been constructed using manifestly unitary actions (HFK, heat-kernel and its variants), with oscillatory sums rendered absolutely convergent by contour deformations and auxiliary-integer reparametrizations. Monte Carlo algorithms have demonstrated proof-of-principle non-perturbative real-time gauge calculations in $1+1$D U(1) and SU(3), overcoming the exponential cost barriers of prior imaginary-time methods (Kanwar et al., 2021).
• Quantum Gravity and Black Hole Thermodynamics: Real-time path integrals are directly applicable to gravitational thermodynamics, offering a route that obviates the conformal factor instability of Euclidean gravity, allowing a definition in terms of absolutely oscillatory (unit modulus) integrals over Lorentzian metrics, with black hole saddle-points emerging as dominant contributions after deformation to appropriate subcontours (Marolf, 2022).
• Non-Markovian Open Systems: Real-time path integrals, combined with augmented density tensor propagation and improved memory cutoff schemes, provide numerically exact access to reduced density matrix dynamics in strongly non-Markovian regimes, including physical features such as population revivals and information backflow (Strathearn et al., 2017).

In all these cases, recent advances in convergence algorithms, thimble decompositions, and quadrature methods have greatly extended the practical range of real-time path integral techniques.

7. Limitations, Open Problems, and Outlook

Despite progress, several mathematically and computationally challenging issues remain:

• Infinite Number of Saddles and Intersection Numbers: For non-quadratic potentials, the set of complex contributing saddle points is generally infinite; the identification of the relevant subset and rigorous computation of intersection numbers in infinite dimensional spaces is only partially solved, despite high-dimensional advances (Shoji et al., 7 Oct 2025, Tanizaki et al., 2014).
• Scaling with System Size: While Picard–Lefschetz decomposition and recent multidimensional shooting methods have extended reach to tens of degrees of freedom, the cost still grows rapidly with system size in general, and full field-theoretic applications require further developments.
• Numerical Instabilities: Integration along complex flows can exhibit stiffness and instability for long flows or when sampling over directions with significant phase fluctuations; tempering, adaptive sampling, or higher-order integration is required for large systems (Mou et al., 2024).
• Convergence near Caustics: In the vicinity of focal points (caustics), analytic regularization is required, and controlling the extrapolation $\eta\to 0^+$ demands additional care.

Nonetheless, the field continues to progress rapidly, driven by algorithmic innovation, analytic understanding of complex field structures, new numerical architectures, and insights from resurgent analysis and quantum computing.

---

References:

(Rosenfelder, 2021): On the numerical evaluation of real-time path integrals: Double exponential integration and the Maslov correction (Feldbrugge et al., 25 Aug 2025): The real-time Feynman path integral for step potentials (Polyzou et al., 2018): Scattering with real-time path integrals (Feldbrugge et al., 2022): Existence of real time quantum path integrals (Shoji et al., 7 Oct 2025): Stable Evaluation of Lefschetz Thimble Intersection Numbers: Towards Real-Time Path Integrals (Mou et al., 2024): Computing real-time quantum path integrals on Sewed, almost-Lefschetz thimbles (Kanwar et al., 2021): Real-time lattice gauge theory actions: unitarity, convergence, and path integral contour deformations (Marolf, 2022): Gravitational thermodynamics without the conformal factor problem: Partition functions and Euclidean saddles from Lorentzian Path Integrals (Strathearn et al., 2017): Efficient Real-Time Path Integrals for Non-Markovian Spin-Boson Models (Cherman et al., 2014): Real-Time Feynman Path Integral Realization of Instantons (Tanizaki et al., 2014): Real-time Feynman path integral with Picard--Lefschetz theory and its applications to quantum tunneling