Paper: arXiv 2512.07162
Authors: Kieran A. Malandain, Selim Kalici, Hakob Chakhoyan
Abstract
Real-time calibration of stochastic volatility models (SVMs) is computationally bottlenecked by the need to repeatedly solve coupled partial differential equations (PDEs). In this work, we propose DeepSVM, a physics-informed Deep Operator Network (PI-DeepONet) designed to learn the solution operator of the Heston model across its entire parameter space. Unlike standard data-driven deep learning (DL) approaches, DeepSVM requires no labelled training data. Rather, we employ a hard-constrained ansatz that enforces terminal payoffs and static no-arbitrage conditions by design. Furthermore, we use Residual-based Adaptive Refinement (RAR) to stabilize training in difficult regions subject to high gradients. Overall, DeepSVM achieves a final training loss of \(10^{"-5"}\) and predicts highly accurate option prices across a range of typical market dynamics. While pricing accuracy is high, we find that the model’s derivatives (Greeks) exhibit noise in the at-the-money (ATM) regime, highlighting the specific need for higher-order regularization in physics-informed operator learning.
Complexity vs Empirical Score
- Math Complexity: 8.5/10
- Empirical Rigor: 6.0/10
- Quadrant: Holy Grail — high math complexity, high empirical rigor
Why this score: The paper demonstrates high mathematical complexity through advanced operator learning theory, PDE transformations, and physics-informed neural network architecture. Empirical rigor is solid with detailed implementation specs (Sobol sequences, adaptive refinement) and validation results, though it lacks real-market backtesting or code availability, placing it between lab research and street-ready application.
Research Flowchart
flowchart TD A["Research Goal<br>Real-time calibration of<br>Stochastic Volatility Models"] --> B["Methodology<br>Physics-Informed Deep Operator Network<br>PI-DeepONet"] B --> C["Data/Input Sources<br>No labeled data<br>Heston PDE residuals & boundary conditions"] C --> D["Computational Processes<br>1. Hard-constrained ansatz<br>2. Residual-based Adaptive Refinement<br>3. Backpropagation on PDE loss"] D --> E["Key Findings/Outcomes<br>Accurate pricing (Loss ~10^-5)<br>Noisy Greeks in ATM regime<br>Need for higher-order regularization"]
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 DeepSVM, a physics‑informed Deep Operator Network that learns the Heston pricing operator across the full parameter space without labelled data.
Source sentence
we propose DeepSVM, a physics-informed Deep Operator Network (PI-DeepONet) designed to learn the solution operator of the Heston model across its entire parameter space.
Assumes 2
- Zero dividend yield is assumed (q = 0). model
Source sentence
We assume a zero dividend yield (q = 0) for this study, though the framework generalizes trivially.
- Feller condition is satisfied so no boundary condition at ν = 0 is required. model
Source sentence
requiring no boundary condition at ν = 0 provided the Feller condition is satisfied.
Data
Synthetic data only.
Costs
Not applicable
Finds 4
- Training converges to a total loss of order 10⁻⁵ after 5,000 L‑BFGS iterations. performance
Source sentence
reduces the total loss to O(10−5) within 5,000 iterations.
- DeepSVM predicts option prices with high fidelity across most of the domain, with the largest errors near the ATM regime. pricing-calibration
Source sentence
DeepSVM reproduces the price surface with high fidelity across the majority of the domain; the largest discrepancies occur in the narrow region near the ATM regime (x ≈0).
- After training, inference is near‑instantaneous with O(1) computational cost across the global parameter space. performance
Source sentence
Once trained, the network provides near-instantaneous inference (O(1) cost) across the global parameter space.
- Greeks (Delta and Gamma) computed from DeepSVM exhibit noise in the at‑the‑money region. negative
Source sentence
while pricing accuracy is high, we find that the model's derivatives (Greeks) exhibit noise in the at-the-money (ATM) regime
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 45 related papers). Coverage is partial, so this is neither replication nor refutation.
2 papers in the archive : PIRanNN: Random-Feature Pricing and Calibration under … ; Beyond Time Induction: Structured Conditional Neural Solvers …
Code
No repository by the authors found in the paper.
Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.
Cited References (14)
- Physics-Informed Neural Networks (PINNs) in Finance — link
- Option Pricing in the Heston Model with Physics Inspired Neural Networks — link
- Financial Machine Learning (2023) — link
- Deep Learning for Solving and Estimating Dynamic Macro-finance Models — → 2305.09783
- Towards Fast Option Pricing PDE Solvers Powered by PIELM — link
- Learning nonlinear operators via DeepONet based on the universal approximation theorem of operators (2021)
- A Closed-Form Solution for Options with Stochastic Volatility with Applications to Bond and Currency Options (1993)
- Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations (2019)
and 6 more
- Learning the solution operator of parametric partial differential equations with physics-informed DeepONets (2021) — link
- Artificial Neural Networks for Solving Ordinary and Partial Differential Equations (1998)
- DeepXDE: A Deep Learning Library for Solving Differential Equations (2021)
- Efficient backprop (2012) — link
- A comprehensive study of non-adaptive and residual-based adaptive sampling for physics-informed neural networks (2023)
- Sobolev Training for Neural Networks (2017) — link
1 of 14 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.