Paper: arXiv 2403.11738

Authors: Alexandre Pannier, Cristopher Salvi

Abstract

We develop a provably convergent kernel-based solver for path-dependent PDEs (PPDEs). Our numerical scheme leverages signature kernels, a recently introduced class of kernels on path-space. Specifically, we solve an optimal recovery problem by approximating the solution of a PPDE with an element of minimal norm in the signature reproducing kernel Hilbert space (RKHS) constrained to satisfy the PPDE at a finite collection of collocation paths. In the linear case, we show that the optimisation has a unique closed-form solution expressed in terms of signature kernel evaluations at the collocation paths. We prove consistency of the proposed scheme, guaranteeing convergence to the PPDE solution as the number of collocation points increases. Finally, several numerical examples are presented, in particular in the context of option pricing under rough volatility. Our numerical scheme constitutes a valid alternative to the ubiquitous Monte Carlo methods.

Complexity vs Empirical Score

  • Math Complexity: 8.5/10
  • Empirical Rigor: 4.0/10
  • Quadrant: Lab Rats — high math complexity, limited empirical testing

Why this score: The paper introduces a novel kernel-based solver for path-dependent PDEs, involving advanced functional analysis, signature kernels, and reproducible closed-form solutions, which is highly mathematically dense. However, the empirical evidence is limited to theoretical consistency proofs and a few numerical examples without comprehensive backtesting or real-world data validation.

Research Flowchart

  flowchart TD
  A["Research Goal:<br>Solve Path-Dependent PDEs<br>without Monte Carlo"] --> B["Methodology:<br>Formulate Optimal Recovery<br>in Signature RKHS"]
  B --> C["Input:<br>Collocation Paths &<br>Boundary Conditions"]
  C --> D["Compute:<br>Signature Kernel<br>Evaluations"]
  D --> E["Solve Linear System:<br>Kernel Matrix Inversion"]
  E --> F["Outcome:<br>Closed-Form Solution<br>for PPDE"]
  F --> G["Validation:<br>Consistency Proof &<br>Option Pricing Examples"]

Evidence Card

What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card

Proposes

The paper proposes a kernel-based solver for path‑dependent PDEs using signature kernels, with a convergence theory and closed‑form solution in the linear case.

Source sentence
We develop a kernel-based solver for path-dependent PDEs (PPDEs) along with a convergence theory.

Assumes 5

  • Strict assumptions guaranteeing consistency and convergence of the scheme as the number of collocation points increases. other
    Source sentence
    Under strict assumptions, we prove consistency of the proposed scheme, guaranteeing convergence to the PPDE solution as the number of collocation points increases.
  • Injectivity of the signature (point‑separating property) ensured by paths having one strictly monotone coordinate. model
    Source sentence
    This is a technical assumption needed to ensure injectivity of the signature and consequently the point separating property assumption needed to establish universality of the signature kernel.
  • The static kernel g is four times differentiable with bounded derivatives up to order 4. statistical
    Source sentence
    Assumption 3.10. g : V × V →R is four times differentiable and, for all 0 ≤n, m ≤4, we have ... < ∞.
  • State‑dependent payoffs are used in numerical experiments (though the method allows path‑dependent payoffs). data
    Source sentence
    For clarity of exposition and because well‑posedness results are scarcer in the path‑dependent case, we will however stick to state‑dependent payoffs in our numerical experiments.
Show 1 more
  • The kernel‑based solver does not require additional training costs; the Gram matrix is estimated once offline. execution
    Source sentence
    Our kernel‑based solver does not require additional “training” costs: the estimation of the Gram matrix is performed once, offline, and can then be reused…

Data

  • Synthetic rough‑volatility option pricing data synthetic unknown
    Source sentence
    We consider two values of H = 0.1 and H = 0.3 and two strike values K = 0.1 and K = 1.0. We take the same number of sample paths and collocation points for Monte Carlo and the signature kernel solver respectively...

Costs

Not applicable The paper reports no trading or execution result.

Finds 2

  • The numerical performance (MSE and MAE) of the signature‑kernel PPDE solver is comparable to Monte Carlo across different numbers of paths and collocation points. performance
    Source sentence
    As it can be observed from the tables, the numerical performance in terms of MSE and MAE of the two algorithms are comparable across the different settings.
  • The solver’s computational complexity scales as O(h⁻² ĤN d²) and can be reduced to O(h⁻¹ ĤN d²) on a GPU. theoretical
    Source sentence
    The complexity of the resulting solver scales as O(h−2 ˆNd2) where this time ˆN is the number of collocation paths. This cost can be further reduced to O(h−1 ˆNd2) if the operations are carried out on a GPU.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 20 related papers). Coverage is partial, so this is neither replication nor refutation.

Cited by 4 papers in the archive (3 by independent authors): Deep learning interpretability for rough volatility (uses its method or data)

Code

https://github.com/crispitagorico/sigppde — the authors' own repository

Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.

Cited References (84)

  1. Neural SDE s as Infinite-Dimensional GANs (2021) — link
  2. Generative adversarial nets (2014)
  3. Some properties of regularized kernel methods (2004)
  4. Machine learning of linear differential equations using Gaussian processes (2017)
  5. Numerical Gaussian processes for time-dependent and nonlinear partial differential equations (2018)
  6. Functional expansions — → 2212.13628
  7. On the convergence of monotone schemes for path-dependent PDE s (2017)
  8. Sig GPDE : Scaling Sparse G aussian Processes on Sequential Data (2021)
and 76 more
  1. Wide feedforward or recurrent neural networks of any architecture are G aussian processes (2019)
  2. Probabilistic numerical methods for PDE -constrained B ayesian inverse problems (2017)
  3. DGM : A deep learning algorithm for solving partial differential equations (2018)
  4. Signature Kernel Conditional Independence Tests in Causal Discovery for Stochastic Processes (2024) — link
  5. A Neural RDE approach for continuous-time non- M arkovian stochastic control problems (2023) — link
  6. Optimal stopping via distribution regression: a higher rank signature approach (2023) — link
  7. New directions in the applications of rough path theory (2023)
  8. Deep signature transforms (2019)
  9. Solving high-dimensional partial differential equations using deep learning (2018)
  10. Weighted signature kernels (2024)
  11. Differential equations driven by rough paths (2007)
  12. Linear generalizations of Gronwall’s inequality (1976)
  13. Derivative reproducing properties for kernel methods in learning theory (2008)
  14. Kernels for sequentially ordered data (2019)
  15. Neural signature kernels as infinite-width-depth-limits of controlled R es N ets (2023)
  16. Non-adversarial training of Neural SDE s with signature kernel scores (2023)
  17. Sk-tree: a systematic malware detection algorithm on streaming trees via the signature kernel (2021)
  18. Higher order kernel mean embeddings to capture filtrations of stochastic processes (2021)
  19. Distribution regression for sequential data (2021)
  20. Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations (2019)
  21. A Neural RDE -based model for solving path-dependent PDE s (2023) — link
  22. Deep curve-dependent PDE s for affine rough volatility (2023)
  23. Solving path dependent PDE s with LSTM networks and path signatures (2020) — link
  24. Path-dependent deep G alerkin method: A neural network approach to solve path-dependent partial differential equations (2021)
  25. Stochastic control and differential games with path-dependent influence of controls on dynamics and running cost (2019)
  26. Operator-Adapted Wavelets, Fast Solvers, and Numerical Homogenization: From a Game Theoretic Approach to Numerical Approximation and Algorithm Design (2019)
  27. Pricing under rough volatility (2016)
  28. Hybrid scheme for B rownian semistationary processes (2017)
  29. Deep curve-dependent PDE s for affine rough volatility (2019) — link
  30. Mmd gan: Towards deeper understanding of moment matching network (2017) — link
  31. Hilbert space embeddings and metrics on probability measures (2010)
  32. Theory of reproducing kernels (1950)
  33. Generalizing from several related classification tasks to a new unlabeled sample (2011)
  34. Path dependent F eynman– K ac formula for forward backward stochastic V olterra integral equations (2022)
  35. Rough volatility (2023)
  36. Neural stochastic PDE s: Resolution-invariant learning of continuous spatiotemporal dynamics (2022)
  37. Global universal approximation of functional input maps on weighted spaces — → 2306.03303
  38. Volatility is rough (2022)
  39. Neural rough differential equations for long time series (2021)
  40. Probabilistic numerical methods for partial differential equations and B ayesian inverse problems (2016) — link
  41. Sparse C holesky factorization for solving nonlinear PDE s via G aussian processes (2023) — link
  42. Solving and learning nonlinear PDE s with G aussian processes (2021) — link
  43. Scalable gradients and variational inference for stochastic differential equations (2020)
  44. The Signature Kernel is the solution of a G oursat PDE (2021)
  45. Multidimensional stochastic processes as rough paths: theory and applications (2010)
  46. Topologies on unparameterised path space (2022) — link
  47. Signature moments to characterize laws of stochastic processes (2018) — link
  48. A kernel two-sample test (2012)
  49. A martingale approach for fractional B rownian motions and related path dependent PDE s (2019)
  50. Affine V olterra processes (2019)
  51. The characteristic function of rough Heston models (2019)
  52. Stochastic differential equations (2005)
  53. On the improved rates of convergence for Mat 'e rn-type kernel ridge regression with application to calibration of computer models (2020)
  54. Rough volatility, path-dependent PDE s and weak rates of convergence — → 2304.03042
  55. A Kernel Approach for PDE Discovery and Operator Learning (2022) — link
  56. Learning “best” kernels from data in G aussian process regression. With application to aerodynamics (2022)
  57. Optimal rates for the regularized least-squares algorithm (2007)
  58. Minimax Optimal Kernel Operator Learning via Multilevel Training (2022) — link
  59. Learning partial differential equations in reproducing kernel hilbert spaces (2023)
  60. Error Analysis of Kernel/ GP Methods for Nonlinear and Parametric PDE s (2023) — link
  61. Scattered data approximation (2004)
  62. Path-dependent PDE s for volatility derivatives — → 2311.08289
  63. Brownian semistationary processes and volatility/intermittency (2009)
  64. G aussian V olterra processes as models of electricity markets (2023) — link
  65. A rough multi-factor model of electricity spot prices (2022)
  66. Asymptotic theory for B rownian semi-stationary processes with application to turbulence (2013)
  67. Lecture notes on rough paths and applications to machine learning (2024) — link
  68. Assessing relative volatility/intermittency/energy dissipation (2014)
  69. Gaussian Agency problems with memory and Linear Contracts (2022)
  70. Sample paths estimates for stochastic fast-slow systems driven by fractional B rownian motion (2020)
  71. Climate dynamics as a nonlinear B rownian motion (1998)
  72. Slow-fast systems with fractional environment and dynamics (2022)
  73. Functional limit theorems for the fractional Or nstein– U hlenbeck process (2022)
  74. Ergodicity of stochastic differential equations driven by fractional B rownian motion (2005)
  75. Consistency and robustness of kernel-based regression in convex risk minimization (2007)
  76. Fractional B rownian motions, fractional noises and applications (1968)

4 of 84 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.