Complex Semi-Classical Paths

• Complex semi-classical paths are analytically continued trajectories from classical mechanics that enable evaluation of oscillatory integrals and exploration of tunneling phenomena.
• They employ techniques such as Picard–Lefschetz theory, complex saddle point analysis, and Hamilton–Jacobi methods to navigate singularities and branch points on Riemann surfaces.
• These paths unify barrier tunneling, resonance-assisted tunneling, and dynamical transitions, offering crucial insights into semiclassical propagation and nonperturbative structures.

Searching arXiv for recent and foundational papers on complex semi-classical paths, complex saddles, tunneling, and related Picard–Lefschetz methods. Searching arXiv for recent and foundational papers on complex semi-classical paths, complex saddles, tunneling, and related Picard–Lefschetz methods. arXiv search query: "complex semiclassical paths tunneling Picard-Lefschetz complex classical trajectories" Complex semi-classical paths are complexified stationary-phase trajectories that arise when the variables of classical mechanics or field theory—time, coordinates, momenta, or fields—are analytically continued away from the real domain in order to evaluate oscillatory or Euclidean path integrals semiclassically. In this sense they include Euclidean instantons as special vertical contours in complex time, but also more general complex saddles, complex bions, and analytically continued trajectories that cross singularities or move on nontrivial Riemann sheets. Across the literature, these paths provide a unified framework for classically forbidden evolution, dynamical tunneling, resonance-assisted tunneling, coherent-state propagation, and nonperturbative vacuum effects, especially in settings where singularities are present or purely Euclidean solutions do not exist (Bramberger et al., 2016, Behtash et al., 2015, Feldbrugge et al., 2023, Feldbrugge et al., 25 Aug 2025).

1. Formal definition and analytic continuation

A standard starting point is the real-time action for a single degree of freedom,

$$S[\phi]=\int_{t_i}^{t_f} dt \,\Bigl[\tfrac12 \Bigl(\frac{d\phi}{dt}\Bigr)^2 - V(\phi)\Bigr].$$

The complex-time formulation keeps the same functional form but allows $t\in\mathbb C$ and $\phi(t)$ to become complex-valued. Introducing a real parameter $\lambda$ and a lapse $n(\lambda)$ with $dt=n(\lambda)d\lambda$, the action becomes

$$S_C[\phi]=\int d\lambda \,\Bigl[\tfrac12 \frac{(\phi'(\lambda))^2}{n(\lambda)}-n(\lambda)V(\phi)\Bigr],$$

and variation yields the complex Euler–Lagrange equation

$$\frac{d^2\phi}{dt^2}+V'(\phi)=0,\qquad t\in\mathbb C.$$

This is the basic formulation used to study tunneling from paths in complex time (Bramberger et al., 2016).

A phase-space version complexifies the canonical variables by

$x\to Z\equiv x+i\,y,\qquad p\to P\equiv p-i\,q,$

extends the classical Hamiltonian holomorphically to $H_{\rm cl}(Z,P)$, and decomposes it into real and imaginary parts,

$t\in\mathbb C$0

The corresponding real action on the enlarged four-dimensional phase space is

$t\in\mathbb C$1

supplemented by the first-class constraint $t\in\mathbb C$2. The equations of motion take Hamilton form,

$t\in\mathbb C$3

so the extra variables $t\in\mathbb C$4 permit trajectories to go around classical turning points in the complex plane rather than reflect from them (Rivers, 2012).

In semiclassical propagation and dynamical tunneling, the same analytic continuation appears in Hamilton–Jacobi or coherent-state form. One writes the time-dependent Hamilton–Jacobi equation with $t\in\mathbb C$5,

$t\in\mathbb C$6

or complexified Hamilton equations

$t\in\mathbb C$7

and then evaluates the action on the resulting complex paths (Mertig et al., 2012, Lando, 2020).

2. Contours, thimbles, Riemann surfaces, and singularities

The geometry of complex semi-classical paths is controlled by contour deformation. In complex time, one fixes initial data on a real classical history and integrates the complex ODE along contours in the complex-$t\in\mathbb C$8 plane. By Cauchy’s theorem, any two contours with the same endpoints that can be continuously deformed into one another without crossing singularities of $t\in\mathbb C$9 are equivalent. On the resulting complex-time grids, classical Lorentzian histories appear as lines parallel to the real-$\phi(t)$0 axis on which $\phi(t)$1, while Euclidean segments are vertical lines with $\phi(t)$2 (Bramberger et al., 2016).

Singularities enter in several ways. They occur wherever $\phi(t)$3 or the ODE’s right-hand side becomes singular, where $\phi(t)$4 hits a branch point in the complex $\phi(t)$5 plane, or where the analytic continuation of the potential has poles or branch points. Explicit examples include poles at $\phi(t)$6 for $\phi(t)$7, poles at $\phi(t)$8 for the Rosen–Morse barrier, and simple poles at $\phi(t)$9 for the Woods–Saxon step (Bramberger et al., 2016, Feldbrugge et al., 2023, Feldbrugge et al., 25 Aug 2025).

A complementary geometric language is Picard–Lefschetz theory. There one complexifies the space of paths, finds all classical solutions of the complexified boundary-value problem, and associates to each saddle a Lefschetz thimble defined by downward gradient flow. The original integration cycle is then decomposed into a sum over thimbles with integer intersection numbers, so the relevant saddles are precisely those reachable by upward flow or, equivalently, those with nonzero intersection number (Feldbrugge et al., 2023, Behtash et al., 2015).

For one-dimensional polynomial Hamiltonians, the topology can be organized globally by the Riemann surface of the algebraic momentum $\lambda$0. Turning points become branch points, poles at infinity define punctures, and the fundamental group $\lambda$1 classifies topologically distinct complex trajectories. In that setting, action relations among different homotopy classes encode the nonlocal structure of tunneling quantization conditions (Harada et al., 2017).

In problems with singularity crossing, the notion of a single path is sometimes generalized to an equivalence class of trajectories $\lambda$2 with fixed endpoints in $\lambda$3 but with the complex-$\lambda$4 contour for $\lambda$5 allowed to detour around singularities. All members of the class share the same value of the action. This generalization is used explicitly for smooth and discontinuous step potentials (Feldbrugge et al., 25 Aug 2025).

3. Selection rules: negative modes, caustics, and Stokes phenomena

Complexification greatly enlarges the set of formal saddles, so a central problem is to determine which saddles actually contribute. In the complex-time tunneling framework, one expands around a background $\lambda$6 and studies the fluctuation equation for the zero-eigenvalue mode,

$\lambda$7

with boundary conditions $\lambda$8 and $\lambda$9. Although $n(\lambda)$0 is complex, its eigenvalues remain real. The criterion is that a candidate path is relevant if the real part of the zero-mode solution, $n(\lambda)$1, has no nodes between the endpoints. If one cannot avoid a line of zeros of $n(\lambda)$2 by contour deformation, the solution carries at least one negative mode and must be discarded. Only solutions with zero negative modes and finite action contribute at leading semi-classical order (Bramberger et al., 2016).

A recurrent misconception is that every complex solution of the complexified boundary-value problem contributes. In Picard–Lefschetz language this is false: as parameters vary, saddles enter or leave the decomposition at Stokes lines, defined by equality of the imaginary parts of the exponents. In the Rosen–Morse analysis this is written as

$n(\lambda)$3

and crossing such a line changes the thimble decomposition and the relevant saddle set (Feldbrugge et al., 2023).

Caustics provide another organizing principle. They occur when two real saddles coalesce and continue beyond the coalescence point as a complex-conjugate pair. In configuration-space semiclassics this is encoded by the vanishing of a Jacobian or the divergence of the van Vleck determinant; in coherent-state semiclassics, a phase-space caustic occurs when $n(\lambda)$4 or equivalently $n(\lambda)$5. Near such points, the quadratic approximation fails and must be replaced by a uniform Airy-type approximation. In the Nelson-Hamiltonian example, the resulting uniform propagator remains finite at the caustic and shows less than $n(\lambda)$6 relative error everywhere when compared with the full-quantum result (Ribeiro, 2010).

The selection problem is not exhausted by simple decay criteria. For one-dimensional polynomial Hamiltonians, it is possible to enumerate homotopy classes on the Riemann surface, but the proper treatment of Stokes phenomena remains essential: in normal-form Hamiltonians, some apparently decaying saddles must still be discarded, indicating that the correct criterion requires the full machinery of exact WKB or resurgent analysis (Harada et al., 2017).

4. Barrier tunneling, reflection, and singularity crossing

The most direct application is ordinary one-dimensional barrier tunneling. In the quartic double well,

$n(\lambda)$7

the standard Euclidean instanton is recovered from the special vertical contour $n(\lambda)$8, and its real-time complex deformations yield the same saddle action as long as no singularity is crossed. Along more general contours in the complex-time plane, the single-instanton path has no zeros of $n(\lambda)$9, while multi-winding paths acquire nodes and therefore carry negative modes (Bramberger et al., 2016).

Potentials with complex singularities sharpen the distinction between admissible and inadmissible paths. For

$dt=n(\lambda)d\lambda$0

the relevant tunneling paths are those that pass just beyond the nearest pole but avoid crossing any zero-lines of $dt=n(\lambda)d\lambda$1; paths looping around the farther pole encounter an unavoidable node and are discarded. This example illustrates that singularities do not merely obstruct contour deformation; they also structure the nodal pattern that determines saddle relevance (Bramberger et al., 2016).

In the symmetric Rosen–Morse barrier,

$dt=n(\lambda)d\lambda$2

real saddles coalesce at a caustic and continue into the forbidden region as a complex-conjugate pair. The relevant complex path may then hit a pole at $dt=n(\lambda)d\lambda$3, at which point the naive boundary-value solution ceases to exist. The resolution is analytic continuation to a new Riemann sheet. The on-shell action acquires the universal shift

$dt=n(\lambda)d\lambda$4

and this shift generates the exponential suppression factor characteristic of tunneling (Feldbrugge et al., 2023, Feldbrugge et al., 2023).

Step and smooth-step potentials exhibit related but more intricate behavior. For the Heaviside step $dt=n(\lambda)d\lambda$5 and its Woods–Saxon regularization

$dt=n(\lambda)d\lambda$6

complex saddles are connected to caustics, may cease to exist as naive solutions when they encounter singularities of the potential, and can be continued beyond the crossing by passing to equivalence classes of trajectories. In this framework one identifies the instanton responsible for quantum reflection. Most complex contributions are small, but in the sharp-step limit or certain geometries one finds $dt=n(\lambda)d\lambda$7, so the contribution is unsuppressed and persists into the semi-classical limit (Feldbrugge et al., 25 Aug 2025).

5. Mixed systems, coherent-state propagation, and complexified phase space

Complex semi-classical paths are not restricted to barrier penetration in one-dimensional configuration space. In mixed regular-chaotic systems, there is generally no real classical path connecting a regular torus to the chaotic sea, so dynamical tunneling is described by complex trajectories in complexified phase space. The tunneling amplitude is written as a sum over complex paths,

$dt=n(\lambda)d\lambda$8

and the dominant contribution is controlled by the path of smallest positive $dt=n(\lambda)d\lambda$9. For the standard map, the semiclassical prediction

$$S_C[\phi]=\int d\lambda \,\Bigl[\tfrac12 \frac{(\phi'(\lambda))^2}{n(\lambda)}-n(\lambda)V(\phi)\Bigr],$$0

shows excellent agreement with numerically exact tunneling rates over several orders of magnitude (Mertig et al., 2012).

When a nonlinear resonance chain is present, the same logic yields a complex-path description of resonance-assisted tunneling. In the effective pendulum Hamiltonian with one $$S_C[\phi]=\int d\lambda \,\Bigl[\tfrac12 \frac{(\phi'(\lambda))^2}{n(\lambda)}-n(\lambda)V(\phi)\Bigr],$$1 resonance, one identifies direct paths, resonance-bridge paths, and partner-torus paths, leading to

$$S_C[\phi]=\int d\lambda \,\Bigl[\tfrac12 \frac{(\phi'(\lambda))^2}{n(\lambda)}-n(\lambda)V(\phi)\Bigr],$$2

For the standard map with kicking strength $$S_C[\phi]=\int d\lambda \,\Bigl[\tfrac12 \frac{(\phi'(\lambda))^2}{n(\lambda)}-n(\lambda)V(\phi)\Bigr],$$3, the closed-form prediction reproduces numerical tunneling rates to within a few percent over many orders of magnitude, including the resonance-assisted peaks (Fritzsch et al., 2016).

In semiclassical propagation, complexification also resolves representation-dependent singularities. The van Vleck–Gutzwiller propagator requires a root-search for all real trajectories connecting fixed endpoints and diverges at caustics. In the Bargmann or coherent-state representation, one instead uses complex classical paths and a complex generating function $$S_C[\phi]=\int d\lambda \,\Bigl[\tfrac12 \frac{(\phi'(\lambda))^2}{n(\lambda)}-n(\lambda)V(\phi)\Bigr],$$4, obtaining a semiclassical propagator whose amplitude factor never vanishes for real Hamiltonians. Through inverse Segal–Bargmann transforms, this construction yields the Herman–Kluk initial-value representation, which is free of caustic singularities and identical to the H-K propagator. Numerical studies in the Kerr system further reveal a pronounced breakdown at half the Ehrenfest time and an accumulation of trajectories around caustics as a function of increasing time, termed “caustic stickiness” (Lando, 2020).

A related geometric use of complex arcs appears in the Weyl representation of nonequilibrium work identities. There, imaginary-time stationary-phase trajectories define a pseudo-Hamiltonian $$S_C[\phi]=\int d\lambda \,\Bigl[\tfrac12 \frac{(\phi'(\lambda))^2}{n(\lambda)}-n(\lambda)V(\phi)\Bigr],$$5 and a “pseudo-classical” trajectory given by the midpoint of the chord of a frozen-time imaginary arc. Integrating $$S_C[\phi]=\int d\lambda \,\Bigl[\tfrac12 \frac{(\phi'(\lambda))^2}{n(\lambda)}-n(\lambda)V(\phi)\Bigr],$$6 along that trajectory yields a semi-classical work variable entering the leading-order Jarzynski identity (Brodier et al., 2019).

6. Complex bions, effective paths, and nonperturbative structure

The nonperturbative significance of complex semi-classical paths is particularly explicit in resurgence theory and in models with vanishing or degenerate perturbative sectors. In Euclidean path integrals, a complete semiclassical analysis requires complexification of the action and measure together with the inclusion of complex saddles satisfying the holomorphic Newton equation

$$S_C[\phi]=\int d\lambda \,\Bigl[\tfrac12 \frac{(\phi'(\lambda))^2}{n(\lambda)}-n(\lambda)V(\phi)\Bigr],$$7

In tilted double-well and quantum-deformed sine-Gordon systems, one finds exact real bions, exact complex bions, and even singular or multi-valued complex saddles. Their actions differ by phases $$S_C[\phi]=\int d\lambda \,\Bigl[\tfrac12 \frac{(\phi'(\lambda))^2}{n(\lambda)}-n(\lambda)V(\phi)\Bigr],$$8, described as hidden topological angles, and these phases are necessary for the cancellation of perturbative ambiguities and for consistency with supersymmetry in the supersymmetric limit (Behtash et al., 2015).

A particularly sharp example is a one-dimensional potential

$$S_C[\phi]=\int d\lambda \,\Bigl[\tfrac12 \frac{(\phi'(\lambda))^2}{n(\lambda)}-n(\lambda)V(\phi)\Bigr],$$9

for which all ordinary perturbative coefficients vanish identically to all orders. The nonperturbative vacuum energy is then reproduced by finite-action complex solutions of the holomorphic Newton equation. At $$\frac{d^2\phi}{dt^2}+V'(\phi)=0,\qquad t\in\mathbb C.$$0, the numerical action is

$$\frac{d^2\phi}{dt^2}+V'(\phi)=0,\qquad t\in\mathbb C.$$1

so that

$$\frac{d^2\phi}{dt^2}+V'(\phi)=0,\qquad t\in\mathbb C.$$2

The exact imaginary part $$\frac{d^2\phi}{dt^2}+V'(\phi)=0,\qquad t\in\mathbb C.$$3 ensures that $$\frac{d^2\phi}{dt^2}+V'(\phi)=0,\qquad t\in\mathbb C.$$4 is real up to a minus sign, and fitting $$\frac{d^2\phi}{dt^2}+V'(\phi)=0,\qquad t\in\mathbb C.$$5 reproduces the full nonperturbative vacuum energy over a range of $$\frac{d^2\phi}{dt^2}+V'(\phi)=0,\qquad t\in\mathbb C.$$6 (Shuryak, 15 Mar 2026).

A distinct but related notion is the complex effective path obtained by extremizing the quantum effective action $$\frac{d^2\phi}{dt^2}+V'(\phi)=0,\qquad t\in\mathbb C.$$7. Because $$\frac{d^2\phi}{dt^2}+V'(\phi)=0,\qquad t\in\mathbb C.$$8 can be complex, the effective path is generally complex as well, and its imaginary part encodes vacuum decay or particle creation. In the forced harmonic oscillator, the imaginary part of the effective action is tied to the vacuum-persistence probability and the effective path is obtained with Feynman rather than retarded boundary conditions. In the inverse-square near-horizon problem, the growth rate of the effective path reproduces a Planck distribution and the Hawking temperature (Singh et al., 2011).

The interpretive status of such paths is not uniform across the literature. One line of work emphasizes that attempts to identify quantum-to-classical correspondence inside forbidden regions often lead to non-physical paths involving complex time or spatial coordinates, and proposes instead a real operational notion of quasi-classical path defined by post-selected time-of-arrival probabilities. In that approach, the path is extracted from maxima of $$\frac{d^2\phi}{dt^2}+V'(\phi)=0,\qquad t\in\mathbb C.$$9 with respect to $x\to Z\equiv x+i\,y,\qquad p\to P\equiv p-i\,q,$0 at fixed $x\to Z\equiv x+i\,y,\qquad p\to P\equiv p-i\,q,$1, and no analytic continuation of $x\to Z\equiv x+i\,y,\qquad p\to P\equiv p-i\,q,$2 or $x\to Z\equiv x+i\,y,\qquad p\to P\equiv p-i\,q,$3 is introduced (Anastopoulos et al., 2016). This does not negate the semiclassical role of complex saddles; rather, it indicates that “path” can denote different mathematical objects in different formalisms: a contributing saddle of a path integral, an effective trajectory of a quantum action, or an operationally reconstructed time-of-arrival locus.

Complex semi-classical paths thus occupy a broad but coherent domain. They unify Euclidean instantons with real-time complex saddles, provide contour-based criteria for dominance through negative modes or thimble intersection numbers, resolve classically forbidden transport in mixed systems, and furnish the nonperturbative structures required by resurgence and by models with no perturbative series. The common theme is that semiclassical analysis in quantum mechanics and allied field-theoretic settings is often incomplete unless the classical equations are allowed to explore their complex domain (Bramberger et al., 2016, Behtash et al., 2015).