Affine Structure of the Brownian Signature

Dimitri Sotnikov; Eduardo Abi Jaber; Elie Attal

The Brownian signature is affine in a path-dependent sense: its Fourier–Laplace transform expands as a signature series solving a linear equation, and the log expands locally via a tensor-algebra Riccati equation.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · grok-4.5

2026-07-12 10:50 UTC pith:TTVT2DSJ

load-bearing objection Solid existence/uniqueness theory for signature Fourier–Laplace transforms and Riccati equations; the recentering fix for locality is the real practical contribution.

arxiv 2606.29622 v2 pith:TTVT2DSJ submitted 2026-06-28 math.PR

Affine Structure of the Brownian Signature

Eduardo Abi Jaber , Elie Attal , Dimitri Sotnikov This is my paper

classification math.PR MSC 60L1060J6534G20

keywords path signaturesFourier–Laplace transformsRiccati equationtensor algebraaffine structureBrownian motionnon-Markovian

verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper shows that the time-augmented Brownian signature carries a generalized affine structure even though the setting is fully path-dependent. For a carefully chosen class of linear functionals of the signature, the conditional Fourier–Laplace transform equals an entire series in the signature whose coefficients solve an infinite-dimensional linear ODE on the extended tensor algebra. Taking the logarithm yields a local signature expansion whose coefficients solve a Riccati equation driven by the shuffle product. Unlike classical finite-dimensional affine processes, this representation cannot be global: zeros of the transform in the complex plane make the logarithm non-entire. Global formulas are recovered by recentering the expansion at the current signature, which produces a family of randomized Riccati equations with path-dependent terminal conditions. Uniqueness holds inside a growth-and-regularity class that matches the Gaussian tails of Brownian motion. The resulting transform theory supplies a practical route to conditional distributions in non-Markovian models, notably signature-volatility pricing.

Core claim

For admissible coefficients p belonging to the class B, the conditional Fourier–Laplace transform of the time-augmented Brownian signature admits an entire signature expansion whose deterministic coefficients solve the linear equation on the extended tensor algebra, while its logarithm admits a local signature expansion whose coefficients solve the associated Riccati equation; global representations are recovered by recentered Riccati equations whose terminal conditions depend on the current path.

What carries the argument

The signature Riccati equation on the extended tensor algebra (driven by the right-shift generator L and the shuffle square), obtained from the linear equation by an algebraic Cole–Hopf transform (shuffle logarithm), together with the recentering map that replaces the terminal condition p by the left-shifted coefficient X|p.

Load-bearing premise

The linear functional must belong to the admissible class B (even-degree leading terms with strictly positive real parts that dominate all other signature coordinates) and the Fourier–Laplace transform must stay non-zero on the whole time interval.

What would settle it

Exhibit a coefficient p in B for which either the linear series fails to equal the conditional expectation for some group-like element, or the associated Riccati solution produces a logarithm that disagrees with the true log-transform on a set of positive probability inside the claimed radius of convergence.

Watch this falsifier — get emailed when new claim-graph text bears on it.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Desk Editor's Note

Solid existence/uniqueness theory for signature Fourier–Laplace transforms and Riccati equations; the recentering fix for locality is the real practical contribution.

read the letter

This paper finally supplies the missing existence, uniqueness and convergence theory for the infinite-dimensional linear and Riccati equations that people have been writing down formally for signature transforms. Earlier work either stayed Markovian (polynomials of W_T) or simply assumed the series and the ODEs existed. Here they construct the admissible class B via Lyndon-word bounds, prove the entire expansion of the conditional Fourier–Laplace transform by an algebraic factorization-plus-Fubini argument, get the local log-expansion via the shuffle logarithm, derive the linear equation from Itô, pass to the Riccati by algebraic Cole–Hopf, and prove uniqueness inside an Itô-admissible sub-Gaussian class. The recentering construction that replaces the terminal condition by a path-dependent left shift is clean and restores a global representation (only the empty-word component) at the price of solving a new Riccati at each t.

What they do well is keep the hypotheses explicit and use them tightly: B is stable under left shifts, matches known martingale conditions for signature volatility, and the non-vanishing assumption is stated up front. The counter-example showing that the plain log-expansion cannot be global (zeros of the quartic transform on the imaginary axis) is honest rather than papered over. The application to the joint transform of the signature martingale and its bracket, and thence to the signature-vol model, is the natural payoff and closes a gap left open in their earlier formal work.

Soft spots are real but proportionate. Everything lives inside B and non-vanishing; outside that the estimates fail. The plain Riccati expansion is only local, so the practical route is the randomized family. There is no shipped code, but this is pure theory. Citation pattern is normal self-citation for motivation; the proofs stand alone.

This is for people working on signature models, non-Markovian affine transforms, or Fourier pricing in path-dependent volatility. It deserves a serious referee at a probability journal. I would engage with it and expect to cite the existence/uniqueness and recentering results.

Referee Report

0 major / 6 minor

Summary. The paper develops an infinite-dimensional affine transform theory for the time-augmented Brownian signature. For coefficients p in an admissible class B, the conditional Fourier–Laplace transform admits an entire signature expansion whose coefficients solve a linear ODE on the extended tensor algebra (Theorems 4.6, 5.2). Its logarithm admits a local signature expansion whose coefficients solve a Riccati equation (Theorems 4.7, 5.7); the locality is shown to be structural via zeros of the transform (Theorem 4.8). Global representations are recovered by recentering, yielding randomized Riccati equations with path-dependent terminal conditions (Theorem 5.11). Uniqueness is established within Itô-admissible classes under sub-Gaussian growth (Theorem 5.5, Corollary 5.6, Proposition 5.10). Applications include joint transforms for signature volatility models (Section 6).

Significance. If correct, the results supply the first rigorous existence, uniqueness, and convergence theory for signature Fourier–Laplace and log-Fourier–Laplace expansions in a genuinely path-dependent (non-Markovian) setting, where Gaussian-density regularization is unavailable. The algebraic strategy—Lyndon-word bounds (Lemma 4.1), factorization via shuffle exponentials, and Cole–Hopf at the tensor-algebra level—extends prior Markovian or formal results and cleanly explains both the entire expansion of the transform and the intrinsic locality of its logarithm. The recentering construction and the stability of B under left shifts are new and practically useful. The framework justifies transform methods previously used under unverified assumptions in signature volatility models and opens a route to non-Markovian control. Proofs are self-contained; the counterexample establishing non-entireness of the log-transform is a genuine contribution rather than a caveat.

minor comments (6)

Definition 4.2 and Remark 4.1: a short explicit checklist of how the three classical examples (polynomial, integrated, generic) embed into B would help readers verify membership without re-deriving the leading-term conditions each time.
Section 5.3 / Definition 5.4: the Itô-admissibility condition is clear, but a one-sentence pointer that the probabilistic solution of Theorem 5.2 satisfies it via (5.4) and Lemma 4.1 would make Corollary 5.6 easier to check on a first reading.
Figure 1 caption and surrounding text: specify the truncation order M_max = 140 and the numerical scheme (reference to Abi Jaber–Li–Lin) more prominently so that the radius comparison with the first zero is reproducible without hunting the text.
Notation: the dual use of | for word length, right-shift, and absolute value is standard but dense; a brief notational table early in Section 3 would reduce cognitive load.
References to the companion control paper (Abi Jaber–Attal–Sotnikov 2026a) and the martingale paper (2026b) are appropriate; ensuring arXiv identifiers or DOIs are final before publication would aid readers.
Appendix B: the saddle-point contour argument is correct; a short remark that the same method applies to other even degrees with negative real leading coefficient would clarify the scope of Theorem 4.8 beyond the quartic case.

Circularity Check

0 steps flagged

No significant circularity: expansions and linear/Riccati equations are derived from the probabilistic transform, not assumed or fitted.

full rationale

The derivation chain is constructive and self-contained. Class B (Def. 4.2) is defined via leading even-degree terms with positive real parts; integrability follows from Lyndon/Radford bounds (Lemma 4.1) and shuffle-compatible norms (Lemma 4.5). Theorem 4.6 builds u_t := E[ξ_t] from the conditional expectation and justifies Fubini, yielding the entire expansion. The linear equation (Thm 5.2) is then verified by Itô on that probabilistic object; uniqueness holds inside an Itô-admissible sub-Gaussian class proved here (Thm 5.5 / Cor 5.6). The Riccati equation (Thm 5.7) is the algebraic Cole–Hopf transform of that linear equation via the shuffle logarithm; locality is proved by an independent saddle-point counterexample (Thm 4.8), and global recovery uses left-shift stability of B (Lemma 4.4) plus Chen (Thm 5.11). Prior self-citations (signature vol, control, formal Riccati) are explicitly framed as works that assumed existence/convergence; this paper supplies those missing proofs rather than importing them as load-bearing uniqueness or ansatz. No equation reduces to its input by definition, and there is no data fitting. Score 1 only for routine reuse of the authors’ prior signature Itô formula as a technical tool, which does not encode the target affine claims.

Axiom & Free-Parameter Ledger

0 free parameters · 6 axioms · 2 invented entities

The paper rests on standard stochastic calculus, the algebraic theory of signatures and shuffle products, and a newly delineated class B of admissible coefficients whose real parts are controlled by Lyndon-word bounds. No free parameters are fitted; the only modelling choices are the definition of B and the growth class for uniqueness.

axioms (6)

standard math Chen’s identity and the group-like property of path signatures
Used throughout for factorization and recentering (Sections 3–5).
standard math Itô formula for linear functionals of signatures (Theorem 3.5)
Taken from prior work of the authors and applied to derive the linear equation.
standard math Radford’s theorem: the shuffle algebra is generated by Lyndon words
Invoked to prove the signature bounds of Lemma 4.1 (Appendix A).
ad hoc to paper Class B of admissible coefficients (Definition 4.2) with Re(leading coefficients) > 0
Constructed so that Re⟨p, signature⟩ is bounded above; required for all integrability and interchange arguments.
domain assumption Non-vanishing of the Fourier-Laplace transform on [0,T]
Needed to define the continuous branch of the shuffle logarithm and the Riccati equation.
ad hoc to paper Sub-Gaussian growth condition for uniqueness of the linear equation
Calibrated to Brownian tails; excludes Tychonoff-type counter-examples (Theorem 5.5).

invented entities (2)

Class B of admissible signature coefficients no independent evidence
purpose: Guarantees that the real part of the linear functional is bounded above, enabling entire expansions and integrability.
Defined ad hoc in Definition 4.2 by combining even-degree leading terms with Lyndon-controlled remainders; closed under left shifts (Lemma 4.4).
Randomized Riccati equation with path-dependent terminal condition no independent evidence
purpose: Restores a global representation of the log-transform after the local expansion fails.
Obtained by recentering via Chen’s identity and left shifts (Theorem 5.11); terminal condition becomes X|p for the current signature X.

pith-pipeline@v1.1.0-grok45 · 50327 in / 2780 out tokens · 28297 ms · 2026-07-12T10:50:47.251496+00:00 · methodology

0 comments

read the original abstract

We establish an infinite-dimensional affine transform theory for the time-augmented Brownian signature. Our first main result shows that, for a suitable class of linear functions of the signature, the conditional Fourier-Laplace transform admits an entire signature expansion. We prove that the associated coefficients solve an infinite-dimensional linear differential equation on the extended tensor algebra. Our second main result shows that the logarithm admits a local signature expansion whose coefficients satisfy a Riccati equation on the extended tensor algebra, revealing a generalized affine structure of the Brownian signature in a genuinely path-dependent setting. In contrast to conventional affine processes, we show that this representation is intrinsically local: zeros of the Fourier-Laplace transform in the complex plane prevent any global expansion. To recover global representations, we introduce a new class of randomized Riccati equations with path-dependent terminal conditions through a recentering argument. Furthermore, we establish uniqueness of solutions to the linear and Riccati equations within a suitable class of solutions. Our results provide a theoretical framework for transform methods in non-Markovian settings, with applications to the computation of conditional distributions.

Figures

Figures reproduced from arXiv: 2606.29622 by Dimitri Sotnikov, Eduardo Abi Jaber, Elie Attal.

**Figure 1.** Figure 1: Numerical illustrations of the solution u and the function ψ. Left: zeros of u(0, ·) along the imaginary axis. Right: power series approximations of ψb(0, x) for truncation orders ranging from 5 to 25. The dark blue line is the Monte Carlo estimate using 106 samples; the dotted green line corresponds to the expansions obtained via the recentered Riccati equation (5.18). The horizontal bars correspond to th… view at source ↗

Review history (2 revisions) →

discussion (0)

Reference graph

Works this paper leans on

5 extracted references · 3 linked inside Pith

[1]

Path-dependent processes from signatures

Eduardo Abi Jaber, Louis-Amand G´ erard, and Yuxing Huang. Path-dependent processes from signatures. arXiv preprint arXiv:2407.04956, to appear in Annals of Applied Probability, 2024a. Eduardo Abi Jaber, Shaun Li, and Xuyang Lin. Fourier-Laplace transforms in polynomial Ornstein-Uhlenbeck volatility models. arXiv preprint arXiv:2405.02170, to appear in Fi...

arXiv
[2]

Stochastic control with signatures via Riccati equations on the tensor algebra

Eduardo Abi Jaber, Elie Attal, and Dimitri Sotnikov. Stochastic control with signatures via Riccati equations on the tensor algebra. Working paper, 2026a. Eduardo Abi Jaber, Paul Gassiat, and Dimitri Sotnikov. Martingale property and moment explosions in signature volatility models. arXiv preprint arXiv:2503.17103, to appear in Finance and Stochastics, 20...

arXiv
[3]

Signature SDEs from an affine and polynomial perspective

Christa Cuchiero, Sara Svaluto-Ferro, and Josef Teichmann. Signature SDEs from an affine and polynomial perspective. arXiv preprint arXiv:2302.01362v2,

Pith/arXiv arXiv
[4]

Functional expansions

Bruno Dupire and Valentin Tissot-Daguette. Functional expansions. arXiv preprint arXiv:2212.13628,

Pith/arXiv arXiv
[5]

A PDE approach for solving the characteristic function of the generalised signature process

Terry Lyons, Hao Ni, and Jiajie Tao. A PDE approach for solving the characteristic function of the generalised signature process. arXiv preprint arXiv:2401.02393,

Pith/arXiv arXiv

[1] [1]

Path-dependent processes from signatures

Eduardo Abi Jaber, Louis-Amand G´ erard, and Yuxing Huang. Path-dependent processes from signatures. arXiv preprint arXiv:2407.04956, to appear in Annals of Applied Probability, 2024a. Eduardo Abi Jaber, Shaun Li, and Xuyang Lin. Fourier-Laplace transforms in polynomial Ornstein-Uhlenbeck volatility models. arXiv preprint arXiv:2405.02170, to appear in Fi...

arXiv

[2] [2]

Stochastic control with signatures via Riccati equations on the tensor algebra

Eduardo Abi Jaber, Elie Attal, and Dimitri Sotnikov. Stochastic control with signatures via Riccati equations on the tensor algebra. Working paper, 2026a. Eduardo Abi Jaber, Paul Gassiat, and Dimitri Sotnikov. Martingale property and moment explosions in signature volatility models. arXiv preprint arXiv:2503.17103, to appear in Finance and Stochastics, 20...

arXiv

[3] [3]

Signature SDEs from an affine and polynomial perspective

Christa Cuchiero, Sara Svaluto-Ferro, and Josef Teichmann. Signature SDEs from an affine and polynomial perspective. arXiv preprint arXiv:2302.01362v2,

Pith/arXiv arXiv

[4] [4]

Functional expansions

Bruno Dupire and Valentin Tissot-Daguette. Functional expansions. arXiv preprint arXiv:2212.13628,

Pith/arXiv arXiv

[5] [5]

A PDE approach for solving the characteristic function of the generalised signature process

Terry Lyons, Hao Ni, and Jiajie Tao. A PDE approach for solving the characteristic function of the generalised signature process. arXiv preprint arXiv:2401.02393,

Pith/arXiv arXiv