Lap Signature: Transform and Topological Views

• Lap Signature is a dual concept where in rough-path theory it represents a Laplace-transform analogue of expected path signatures, defined by iterated integrals with convergence conditions.
• The framework leverages Fourier–Laplace transforms and associated ODEs on tensor spaces to elucidate conditional distributions and unique signature laws in stochastic settings.
• In topology, Lap Signature refers to local signature formulas for 4-manifolds, particularly in hyperelliptic Lefschetz fibrations, illustrating distinct analytic and topological methods.

“Lap Signature” most commonly denotes a Laplace- or Fourier–Laplace-transform viewpoint on path signatures: the signature of a path is the tensor series of its iterated integrals, and the expected signature is treated as an analogue of the Laplace transform for rough paths. The available literature suggests that the expression is not standardized. In current arXiv usage it is most naturally tied to expected signatures and conditional Fourier–Laplace transforms of the Brownian signature, while in a distinct topological context the phrase can point instead to a local signature formula for hyperelliptic broken Lefschetz fibrations. The common feature is the use of a “signature” invariant, but the transformed or localized quantity differs sharply between rough-path theory and 4-manifold topology (Boedihardjo et al., 2019, Jaber et al., 28 Jun 2026, Hayano et al., 2011).

1. Signature-theoretic foundations

For a continuous path of finite variation $X:[0,1]\to\mathbb{R}^d$, the $k$-th signature tensor is the tensor of iterated integrals

$$\sigma_{i_1\cdots i_k}(X) = \int_{0<t_1<\cdots<t_k<1} dX_{i_1}(t_1)\cdots dX_{i_k}(t_k),$$

and the full signature is the sequence

$$\sigma(X)=\bigl(1,\sigma^{(1)}(X),\sigma^{(2)}(X),\dots\bigr) \in T((\mathbb{R}^d)).$$

The step-$n$ signature lives in the truncated tensor algebra $T^n(\mathbb{R}^d)$, and signatures are exactly the group-like elements characterized by the shuffle relations. In particular, if $I$ and $J$ are words, then group-like elements satisfy $\sigma_{I\shuffle J}=\sigma_I\sigma_J$. The Chen–Chow theorem identifies the image of the step-$n$ signature map of smooth paths with the step-$k$0 free nilpotent Lie group $k$1 (Améndola et al., 2018).

This algebraic picture has a geometric reformulation. For loops in $k$2, Chen’s signature can be realized as holonomy of a connection $k$3 on a principal bundle with structure group a pronilpotent Banach Lie group built from the free Lie algebra on $k$4. In that formulation, the holonomy map coincides with the Chen signature map, the logarithm of the signature lies in the free Lie algebra, and the image is governed by the shuffle relations. This places signature theory simultaneously in tensor algebra, Hopf algebra, and holonomy theory (Alonso et al., 2024).

2. Expected signature and the Laplace-transform analogy

The expected signature of a stochastic process $k$5 is defined level by level as

$k$6

In the rough-path literature it is explicitly described as an analogue of the Laplace transform for random signatures. The central question is whether this tensor series determines the law of the random signature, just as a classical Laplace transform or moment generating function can determine a scalar law (Boedihardjo et al., 2019).

A sufficient condition is the Chevyrev–Lyons infinite-radius condition: $k$7 Under this hypothesis, the law of the signature is the unique probability law with that expected signature. For Brownian motion up to fixed deterministic time $k$8, the expected signature is

$k$9

so the tensor coefficients have factorial decay and the radius of convergence is infinite. In that regime, the Laplace-transform analogy is exact enough to support a uniqueness theorem for laws on the signature group (Boedihardjo et al., 2019).

The analogy is nevertheless conditional rather than automatic. For planar Brownian motion stopped when it exits the unit disk, the expected signature at the origin has finite radius of convergence. This is the first example in which the Chevyrev–Lyons condition fails, so the expected signature ceases to be “entire” in the relevant tensor-norm sense and the standard uniqueness theorem no longer applies (Boedihardjo et al., 2019).

3. Fourier–Laplace transforms of the Brownian signature

A more recent development studies conditional Fourier–Laplace transforms of the time-augmented Brownian signature. For $$\sigma_{i_1\cdots i_k}(X) = \int_{0<t_1<\cdots<t_k<1} dX_{i_1}(t_1)\cdots dX_{i_k}(t_k),$$0, the time-augmented signature $$\sigma_{i_1\cdots i_k}(X) = \int_{0<t_1<\cdots<t_k<1} dX_{i_1}(t_1)\cdots dX_{i_k}(t_k),$$1 records all iterated Stratonovich integrals in the letters $$\sigma_{i_1\cdots i_k}(X) = \int_{0<t_1<\cdots<t_k<1} dX_{i_1}(t_1)\cdots dX_{i_k}(t_k),$$2 and $$\sigma_{i_1\cdots i_k}(X) = \int_{0<t_1<\cdots<t_k<1} dX_{i_1}(t_1)\cdots dX_{i_k}(t_k),$$3, where $$\sigma_{i_1\cdots i_k}(X) = \int_{0<t_1<\cdots<t_k<1} dX_{i_1}(t_1)\cdots dX_{i_k}(t_k),$$4 corresponds to $$\sigma_{i_1\cdots i_k}(X) = \int_{0<t_1<\cdots<t_k<1} dX_{i_1}(t_1)\cdots dX_{i_k}(t_k),$$5 and $$\sigma_{i_1\cdots i_k}(X) = \int_{0<t_1<\cdots<t_k<1} dX_{i_1}(t_1)\cdots dX_{i_k}(t_k),$$6 to $$\sigma_{i_1\cdots i_k}(X) = \int_{0<t_1<\cdots<t_k<1} dX_{i_1}(t_1)\cdots dX_{i_k}(t_k),$$7. For a suitable class $$\sigma_{i_1\cdots i_k}(X) = \int_{0<t_1<\cdots<t_k<1} dX_{i_1}(t_1)\cdots dX_{i_k}(t_k),$$8 of linear functionals $$\sigma_{i_1\cdots i_k}(X) = \int_{0<t_1<\cdots<t_k<1} dX_{i_1}(t_1)\cdots dX_{i_k}(t_k),$$9 on the extended tensor algebra, the conditional transform

$$\sigma(X)=\bigl(1,\sigma^{(1)}(X),\sigma^{(2)}(X),\dots\bigr) \in T((\mathbb{R}^d)).$$0

admits an entire signature expansion

$$\sigma(X)=\bigl(1,\sigma^{(1)}(X),\sigma^{(2)}(X),\dots\bigr) \in T((\mathbb{R}^d)).$$1

where $$\sigma(X)=\bigl(1,\sigma^{(1)}(X),\sigma^{(2)}(X),\dots\bigr) \in T((\mathbb{R}^d)).$$2 is the group of group-like elements (Jaber et al., 28 Jun 2026).

The coefficient family $$\sigma(X)=\bigl(1,\sigma^{(1)}(X),\sigma^{(2)}(X),\dots\bigr) \in T((\mathbb{R}^d)).$$3 solves a linear ODE on tensor space,

$$\sigma(X)=\bigl(1,\sigma^{(1)}(X),\sigma^{(2)}(X),\dots\bigr) \in T((\mathbb{R}^d)).$$4

with

$$\sigma(X)=\bigl(1,\sigma^{(1)}(X),\sigma^{(2)}(X),\dots\bigr) \in T((\mathbb{R}^d)).$$5

This is an infinite-dimensional affine transform formula: the conditional transform is linear in the signature variable, while the coefficients evolve by a tensor-algebra analogue of the heat equation. The construction is genuinely path-dependent and is formulated directly on the tensor algebra rather than on a finite-dimensional Markov state space (Jaber et al., 28 Jun 2026).

The logarithm of the transform has a local signature expansion. If $$\sigma(X)=\bigl(1,\sigma^{(1)}(X),\sigma^{(2)}(X),\dots\bigr) \in T((\mathbb{R}^d)).$$6, one defines $$\sigma(X)=\bigl(1,\sigma^{(1)}(X),\sigma^{(2)}(X),\dots\bigr) \in T((\mathbb{R}^d)).$$7, and then, locally in the signature variable,

$$\sigma(X)=\bigl(1,\sigma^{(1)}(X),\sigma^{(2)}(X),\dots\bigr) \in T((\mathbb{R}^d)).$$8

The coefficient $$\sigma(X)=\bigl(1,\sigma^{(1)}(X),\sigma^{(2)}(X),\dots\bigr) \in T((\mathbb{R}^d)).$$9 satisfies a Riccati equation on the tensor algebra,

$n$0

where

$n$1

This produces an infinite-dimensional affine–Riccati duality for Brownian signatures and gives a transform-theoretic framework for conditional distributions in non-Markovian models (Jaber et al., 28 Jun 2026).

4. Locality, zeros, and the limits of Laplace-type signature methods

Two limitations recur in the literature. The first is radius of convergence. The expected signature of Brownian motion stopped on the boundary of the unit disk has finite radius of convergence, so the expected-signature analogue of the Laplace transform is not globally entire even for a canonical stopped diffusion (Boedihardjo et al., 2019).

The second is locality of the logarithm. For Fourier–Laplace transforms of the Brownian signature, zeros of the transform in the complex plane prevent any global log-expansion. In the scalar model $n$2, the analytic continuation of the relevant transform has infinitely many zeros on the imaginary axis. Consequently, the representation

$n$3

is intrinsically local and only valid up to a positive stopping time or within a neighborhood in signature space. To recover global representations, the theory introduces recentering and randomized Riccati equations with path-dependent terminal conditions (Jaber et al., 28 Jun 2026).

These two phenomena rule out a naive identification of “Lap Signature” with an everywhere-entire transform calculus on signatures. A plausible implication is that signature-based Laplace methods behave more like local analytic charts on path space than like a single global moment-generating function.

5. Broader signature geometry and adjacent transforms

The Laplace-transform viewpoint is embedded in a broader signature ecosystem. In algebraic geometry, fixed-level signature tensors define projective varieties. The universal signature variety $n$4 is the Zariski closure of all level-$n$5 signature tensors, and expected Brownian signature varieties $n$6 are studied as non-commutative lifts of Gaussian moment varieties. This situates Laplace-type constructions among algebraic constraints on deterministic and random signatures, including piecewise linear, polynomial, Brownian, and mixture models (Améndola et al., 2018).

Higher-dimensional generalizations reinforce that the word “signature” is not confined to one-dimensional paths. For piecewise linear surfaces, the surface signature is defined as surface holonomy of a universal translation-invariant $n$7-connection. It decomposes into a boundary component, which is the ordinary Chen signature of the boundary loop, and an abelian component given by integrals of polynomial $n$8-forms over a canonically closed surface. The main injectivity statement is that the signature uniquely characterizes a surface up to translation and thin homotopy in the PL category (Bischoff et al., 20 Jun 2025).

Other papers introduce transformed signatures for approximation theory rather than Laplace theory. The flip signature reverses tensor order, and the sawtooth signature mixes continuous signature information with discrete increments in a generalized Euler–Maclaurin formula. These constructions are not called “Lap signatures,” but they show that signature theory routinely produces new transforms adapted to specific analytic tasks (Bellingeri et al., 2024).

6. Distinct topological usage: local signatures of fibrations

In a different branch of the literature, “signature” refers not to iterated integrals but to the Hirzebruch signature of a $n$9-manifold and its localization over singular fibrations. For a hyperelliptic directed broken Lefschetz fibration $T^n(\mathbb{R}^d)$0, the global signature decomposes as

$T^n(\mathbb{R}^d)$1

where $T^n(\mathbb{R}^d)$2 is Endo’s local signature for hyperelliptic Lefschetz singular fibers and $T^n(\mathbb{R}^d)$3 is a rational-valued homomorphism on a subgroup of the hyperelliptic mapping class group preserving the fold vanishing cycle (Hayano et al., 2011).

This topological local signature is constructed from Meyer’s signature cocycle, its rational cobounding function on the hyperelliptic mapping class group, and the signature contribution of round $T^n(\mathbb{R}^d)$4-handle cobordisms. For Lefschetz fibers of type $T^n(\mathbb{R}^d)$5 and $T^n(\mathbb{R}^d)$6, the local signatures are explicitly

$T^n(\mathbb{R}^d)$7

The paper explicitly states that it does not concern Laplace-type signatures: all signature notions there are topological signatures of $T^n(\mathbb{R}^d)$8-manifolds or $T^n(\mathbb{R}^d)$9-dimensional cobordisms, and the central analytic object is Meyer’s signature $I$0-cocycle (Hayano et al., 2011).

This is the principal terminological ambiguity surrounding “Lap Signature.” In rough-path theory, the phrase points toward Laplace-transform analogues built from Chen signatures. In the topology of broken Lefschetz fibrations, the nearby notion is instead a local signature formula for the ordinary signature of a $I$1-manifold.