Finite Path Integral Limits

• Finite path integral limits are rigorous constructions that replace the formal sum over all paths with finite-dimensional or bounded integrals, ensuring well-defined convergence.
• They employ methods such as time slicing, piecewise geodesics, and reflected path spaces to capture quantum evolution in curved manifolds and systems with boundaries.
• These approaches provide controlled convergence mechanisms that clarify operator ordering, boundary conditions, and regularization in quantum and statistical models.

Finite path integral limits are constructions in which a formal path integral is replaced by finite-dimensional integrals, finite products of short-time kernels, bounded-domain integrals, or explicitly specified function spaces, and the physically relevant propagator, semigroup, partition function, or stationary-phase contribution is then recovered by a controlled limit. Across curved manifolds, manifolds with boundary, constrained configuration spaces, Schwinger–Keldysh formulations, strong-coupling perturbation theory, and relativistic particle models, the shared objective is to replace an ill-defined “sum over all paths” by an approximation scheme whose convergence can be proved, whose boundary data can be specified unambiguously, or whose breakdown can be identified precisely (Miyanishi, 2015, Ludewig, 2016, Fine et al., 2018, Kaya, 2012, Edery, 11 Jul 2025, Vercauteren, 13 Jan 2026).

1. Conceptual scope and principal constructions

The phrase denotes several distinct but structurally related limiting procedures. In one class, one composes short-time kernels over finer and finer partitions of a time interval. In a second, one integrates over finite-dimensional spaces of piecewise geodesic or reflected-geodesic paths and sends the mesh size to zero. In a third, one restricts the integration range in configuration space or the admissible path space itself, develops a convergent expansion on that restricted domain, and only afterward removes the restriction. In a fourth, one specifies a Hilbert space of admissible field histories so that the Gaussian functional integral becomes a self-adjoint spectral problem rather than a formal distributional object (Miyanishi, 2015, Ludewig, 2016, Kaya, 2012, Koch et al., 2014, Fine et al., 2018, Edery, 11 Jul 2025).

Setting	Finite object	Limiting object
Compact Riemannian manifolds	$\bigl(U_\chi(t/N)\bigr)^N$ built from shortest geodesics	$e^{\,it(\Delta-R/6)}$ on the stated domain
Manifolds with boundary	$P_{\mathcal T}$ over piecewise reflected geodesics	$e^{-tL}$
Generalized Laplacians on bundles	$K^{*P}(t)$ for refined partitions $P$	the heat kernel $e^{-tA/2}$
Finite integration bounds or walls	convergent perturbative series at finite $\beta$ or finite $L$	the original unbounded theory as the cutoff is removed

This typology already shows that “finite” may refer to time slicing, configuration-space bounds, finite-dimensional path spaces, or finite regularity classes. A plausible implication is that the central issue is not finiteness in a single literal sense, but the replacement of a formal continuum measure by a controlled approximation mechanism.

2. Geometric time slicing on Riemannian manifolds

On compact Riemannian manifolds, a direct transplantation of the Euclidean Feynman kernel is obstructed by the global multiplicity of paths, the local smoothness of the distance function, and the singular nature of short-time Schrödinger kernels. One response is to build the short-time approximation only from shortest geodesic segments. For $d(x,y)$ smaller than the injectivity radius, the action is taken along the unique minimizing geodesic,

$e^{\,it(\Delta-R/6)}$0

and the amplitude is chosen as the van Vleck determinant,

$e^{\,it(\Delta-R/6)}$1

The resulting oscillatory integral operator is

$e^{\,it(\Delta-R/6)}$2

with $e^{\,it(\Delta-R/6)}$3 supported inside the injectivity radius (Miyanishi, 2015).

The associated finite path integral limit is the time-slicing product

$e^{\,it(\Delta-R/6)}$4

For compact, oriented, rank-1 locally symmetric Riemannian manifolds, the strong limit is proved on the low-energy sector: $e^{\,it(\Delta-R/6)}$5 and if $e^{\,it(\Delta-R/6)}$6 is a finite sum of Laplace eigenfunctions, the spectral projection is unnecessary. The limit is therefore not the plain Laplace–Beltrami evolution but the curvature-corrected evolution with the DeWitt term $e^{\,it(\Delta-R/6)}$7. The paper explicitly emphasizes that the convergence is strong operator convergence, not operator-norm convergence, and that uniform convergence fails in general without spectral truncation (Miyanishi, 2015).

A related but imaginary-time formulation appears for generalized Laplacians on vector bundles. There the finite object is a product

$e^{\,it(\Delta-R/6)}$8

over a partition $e^{\,it(\Delta-R/6)}$9, where $P_{\mathcal T}$0 is kernel composition. For tame data, the approximate kernel

$P_{\mathcal T}$1

defines an approximate semigroup and approximate heat kernel. The refined products converge in the $P_{\mathcal T}$2-norm,

$P_{\mathcal T}$3

and the limit is identified with the genuine heat kernel $P_{\mathcal T}$4. In the twisted $P_{\mathcal T}$5 supersymmetric quantum mechanics case, this completes the path-integral proof of the Atiyah–Singer index theorem for the twisted Dirac operator (Fine et al., 2018).

These two approaches share a common geometric structure: local normal-coordinate analysis, explicit short-time kernels, and curvature terms that are not optional corrections but part of the convergence mechanism. This suggests that on curved spaces, finite path integral limits are inseparable from the geometry of the exponential map, Jacobians, and operator ordering.

3. Boundaries, reflections, and constrained configuration spaces

When a manifold has boundary, the appropriate finite-dimensional approximation is no longer the ordinary geodesic path space. For a compact Riemannian manifold $P_{\mathcal T}$6 with boundary and a formally self-adjoint Laplace type operator

$P_{\mathcal T}$7

on a metric vector bundle, finite-dimensional time slicing is carried out over piecewise geodesic paths in the closed case and piecewise reflected geodesics in the boundary case. The reflection law is

$P_{\mathcal T}$8

and involutive boundary conditions are encoded by a boundary operator $P_{\mathcal T}$9 satisfying $e^{-tL}$0 and covariant constancy along the boundary. Dirichlet and Neumann conditions arise as $e^{-tL}$1 and $e^{-tL}$2, while Robin boundary conditions are not included except in the trivial case $e^{-tL}$3. The finite approximation operator is

$e^{-tL}$4

and the limit

$e^{-tL}$5

holds uniformly in $e^{-tL}$6 for $e^{-tL}$7 and in $e^{-tL}$8 for $e^{-tL}$9, $K^{*P}(t)$0. The proof uses Chernoff’s theorem and gives qualitative convergence rather than an explicit global error rate (Ludewig, 2016).

For one-dimensional systems on a finite interval or the half-line, a different strategy is used. The coordinate $K^{*P}(t)$1 is transformed to a canonical variable $K^{*P}(t)$2,

$K^{*P}(t)$3

The completeness of the new momentum eigenfunctions enables a Hamiltonian time-sliced path integral. The short-time kernel must then be extended to a covering space in order to capture reflected paths. For symmetric and generalized Pöschl–Teller systems, as well as the radial oscillator on the half-line, reflected sectors carry phase factors determined by near-boundary behavior, such as $K^{*P}(t)$4. The familiar minus sign for odd reflections of a particle in a box appears as the special case $K^{*P}(t)$5, where $K^{*P}(t)$6. The paper explicitly argues that the reflection phase is not purely geometric; it depends on parameters characterizing the singular boundary potential (Sakoda, 2018).

Boundary implementation by delta-function constraints inside the field-theoretic path integral leads to a further distinction between boundary types. For the Casimir effect with infinitely thin plates, Dirichlet conditions imposed by delta functionals work cleanly and reproduce the standard energy and pressure, including

$K^{*P}(t)$7

in $K^{*P}(t)$8 dimensions. For Neumann conditions, the same BRW prescription produces divergent self-terms because the effective action contains $K^{*P}(t)$9, leading to divergent $P$0-integrals in the self-couplings. The paper criticizes commonly used prescriptions for discarding these divergences and concludes that, in the regulated treatments it studies, infinitely thin Neumann-type plates yield vanishing Casimir pressure, while the standard result is recovered only for finite-thickness plates or bulk regularizations (Vercauteren, 13 Jan 2026).

A recurrent misconception is that boundary effects can always be imposed by a simple image sum or by delta-function constraints with no change in analytic structure. The boundary results show instead that reflections, phase factors, and even the viability of the limiting procedure depend strongly on the operator class and on the precise boundary condition.

4. Functional measure, self-adjointness, and admissible histories

Finite path integral limits are not only limits of discretizations; they can also arise from specifying the function space on which the path integral is defined. In the in-in formalism for a real scalar field, the generating functional involves doubled fields $P$1 and $P$2, coupled at a return time $P$3. The relevant configuration space is a Hilbert space of doublets

$P$4

with inner product

$P$5

The admissible histories are not merely those satisfying $P$6; they must also satisfy

$P$7

The second condition is derived from the discretized path integral and removes the boundary term that otherwise prevents symmetry of the free kinetic operator (Kaya, 2012).

After integration by parts, the Gaussian sector is governed by the doubled operator

$P$8

With both return-time matching conditions imposed, $P$9 becomes symmetric and, in the explicit flat-space treatment and in de Sitter by a deficiency-index argument, essentially self-adjoint. The spectral representation of $e^{-tA/2}$0 then reproduces the standard Schwinger–Keldysh propagators. The paper therefore treats the subtle boundary integration at the return time as part of the definition of the functional measure rather than an external prescription (Kaya, 2012).

This framework also gives a rigorous interpretation of certain nonperturbative stationary-phase configurations. Instanton-like cosmological solutions are shown to belong to the path-integral function space by square-integrability arguments, and more strongly, they can be identified as limits of Cauchy sequences of admissible classical configurations in the Hilbert space. That limiting-sequence interpretation converts what might otherwise appear as a formal saddle into a bona fide element of the functional domain (Kaya, 2012).

The same general principle appears in a different form in differentiable-path integrals. There the fluctuation around a classical path is expanded as

$e^{-tA/2}$1

and the admissible set is restricted by

$e^{-tA/2}$2

The usual unbounded Fourier-mode integration is replaced by bounded mode-by-mode integration, so the path integral is performed over a subspace of $e^{-tA/2}$3-type regularity rather than over arbitrary continuous Feynman paths. The emergent time scale $e^{-tA/2}$4 separates an ordinary quantum regime from a UV-modified regime in which quantities such as $e^{-tA/2}$5 remain finite (Koch et al., 2014).

5. Finite domains, convergent perturbative series, and Euclidean compactification

A particularly direct use of finite path integral limits appears in strong-coupling perturbation theory. For the quartic integral

$e^{-tA/2}$6

the usual perturbative procedure expands $e^{-tA/2}$7 and then integrates term by term over the entire real line, producing the asymptotic series

$e^{-tA/2}$8

for $e^{-tA/2}$9. Replacing the infinite domain by $\beta$0 yields instead

$\beta$1

which is absolutely convergent for finite $\beta$2. The same mechanism applies when $\beta$3, a case the paper describes as not even Borel summable in the standard treatment. Its central claim is that divergence comes from expanding the interaction and then integrating to infinity, whereas finite integration limits make the ratio test trivial: $\beta$4 For the $\beta$5 dimensional anharmonic oscillator, finite limits are implemented as infinite walls at $\beta$6, encoded by a parameter $\beta$7 with $\beta$8 corresponding to $\beta$9. The ground-state energy is expanded as

$L$0

and for any finite $L$1 the series is absolutely convergent. The paper reports agreement with the exact strong-coupling energy to within less than $L$2 at $L$3 for $L$4, while stressing that the limit $L$5 must be taken after summing the convergent series (Edery, 11 Jul 2025).

The differentiable-path construction provides a different boundedness mechanism. The finite mode bounds induce a characteristic scale $L$6 through

$L$7

and the mean square velocity behaves as

$L$8

For $L$9, the standard Feynman behavior is recovered to leading order; for $d(x,y)$0, the short-time divergence saturates. The kernel factorizes as

$d(x,y)$1

so unitarity requires $d(x,y)$2 with real $d(x,y)$3 (Koch et al., 2014).

Finite Euclidean time enters a further class of constructions. For the two-dimensional noncommutative harmonic oscillator, Wick rotation

$d(x,y)$4

produces a Euclidean kernel, and the partition function is obtained from the trace

$d(x,y)$5

Periodic thermal boundary conditions are encoded by $d(x,y)$6. The resulting partition function,

$d(x,y)$7

agrees with the Hamiltonian derivation. Here the finite Euclidean interval is not removed by a limit; instead it defines the thermal compactification itself (Jahan, 2012).

Taken together, these examples show that “finite” may refer either to an auxiliary cutoff to be sent away later or to a physically meaningful Euclidean interval such as $d(x,y)$8. This suggests that finite path integral limits can function both as regulators and as exact formulations of thermodynamic or boundary-value problems.

6. Symmetry reduction, stabilization, and limits of applicability

The relativistic point particle illustrates a case in which the finite-step construction does not converge by accumulating ever finer path classes. The Euclidean geometric action

$d(x,y)$9

has local Lorentz invariance, global Poincaré invariance, and Weyl invariance. A naive discretization overcounts physically equivalent intermediate configurations and yields the wrong Fourier-space propagator,

$e^{\,it(\Delta-R/6)}$00

rather than the Klein–Gordon form $e^{\,it(\Delta-R/6)}$01. The standard Chapman–Kolmogorov composition fails for the same reason. The remedy is a no-overcounting prescription $e^{\,it(\Delta-R/6)}$02 together with symmetry-compensating factors $e^{\,it(\Delta-R/6)}$03. After quotienting by the redundancy, the one-step kernel is already exact: in one dimension $e^{\,it(\Delta-R/6)}$04, and in higher dimensions the paper states

$e^{\,it(\Delta-R/6)}$05

In this model the continuum limit stabilizes immediately after symmetry reduction rather than emerging from infinitely many inequivalent refinements (Koch et al., 2017).

Several of the constructions discussed above come with explicit restrictions. The geodesic Schrödinger time-slicing theorem is proved only for compact rank-1 locally symmetric manifolds, uses shortest paths only, excludes long geodesic contributions by cutoff, and yields strong rather than operator-norm convergence (Miyanishi, 2015). The reflected-geodesic heat-kernel formula assumes compactness, smooth boundary, a formally self-adjoint Laplace type operator, involutive boundary conditions, and provides no quantitative global rate estimate of the form $e^{\,it(\Delta-R/6)}$06 (Ludewig, 2016). The thin-plate Casimir analysis concludes that infinitely thin Dirichlet plates are reliable in the BRW approach, whereas Neumann and Robin-type conditions are subtle or fail without additional structure such as finite thickness or bulk interactions (Vercauteren, 13 Jan 2026).

A common misunderstanding is that a path integral limit becomes more faithful simply by adding more slices. The relativistic example shows that additional slices may contribute only gauge redundancy; the manifold examples show that more slices must still respect local geometric validity; and the finite-wall perturbative example shows that the order of limits matters, because the convergent series must be summed before the walls are removed. Finite path integral limits therefore organize not just approximation, but also the admissible order of limiting operations (Koch et al., 2017, Edery, 11 Jul 2025).

The subject is consequently best understood not as a single theorem but as a family of replacement principles for formal path integration. Depending on the setting, the finite object may be a geodesic kernel, a reflected billiard path space, a bounded configuration-space integral, a restricted Fourier-mode measure, or a Hilbert space of doubled histories. What unifies these constructions is the insistence that the path integral be defined through a controlled finite surrogate whose convergence, stabilization, or failure can be established in the geometry, operator theory, or asymptotics of the underlying problem.