Learning PDEs for Portfolio Optimization with Quantum Physics-Informed Neural Networks

Abdel Lisser; Letao Wang; Sreejith Sreekumar; Zeno Toffano

REVIEW 3 major objections 2 minor 1 references

Learning PDEs for Portfolio Optimization with Quantum Physics-Informed Neural Networks

T0 review · 3 major / 2 minor · reviewed 2026-07-13 · grok-4.5

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

Pith's one-line read Quantum and quantum-inspired PINNs solve the Merton portfolio HJB PDE with lower error and far fewer parameters than classical fully connected PINNs.

desk verdict Wrong full text was supplied (YOLOv11, not the QPINN paper), so the Merton HJB claims are unauditable; the abstract alone sketches a plausible, structure-exploiting idea but nothing more. read the letter →

arxiv 2604.03346 v2 pith:IO5L5OLL submitted 2026-04-03 quant-ph

Letao Wang , Abdel Lisser , Sreejith Sreekumar , Zeno Toffano This is my paper · ORCID

classification quant-ph MSC 91G1065M9968T0781P68 PACS 03.67.Ac02.60.Lj89.65.Gh

keywords quantumphysics-informedneuralnetworksportfoliooptimizationHamilton-Jacobi-BellmanequationparameterizedcircuittensorrankdecompositionMertonproblemquantum-inspiredPINN

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

Partial differential equations that set optimal portfolio weights are hard for classical solvers and ordinary neural networks. This paper builds a parameterized quantum circuit that realizes a polynomial via tensor-rank decomposition, so that when the rank stays moderate the circuit uses only polynomial rather than exponential resources. From that circuit the authors construct a Quantum Physics-Informed Neural Network and a classical quantum-inspired counterpart; both are guaranteed to contain a polynomial approximation of the PDE solution. On the Hamilton–Jacobi–Bellman equation of the Merton problem (how much to put in a risky asset versus a risk-free asset) the quantum models reach lower training losses and smaller approximation errors than a standard fully connected PINN while using substantially fewer trainable parameters, and they also beat a classical PINN given a similar structural bias. The result supplies experimental evidence that the quantum circuit’s inductive bias helps on PDEs whose solutions have exploitable low-rank structure, and it sketches a resource-efficient route for both near-term quantum hardware and classical simulators.

What carries the argument

A parameterized quantum circuit that implements a multivariate polynomial through tensor-rank decomposition, reducing quantum resource cost from exponential to polynomial whenever the rank remains moderate; this circuit is the trainable ansatz inside both the QPINN and the quantum-inspired PINN.

What would settle it

Re-run the identical Merton HJB experiment with the same classical baselines: if the QPINN and quantum-inspired models fail to produce lower loss and approximation error with fewer parameters, or if deliberately high-rank solutions erase the advantage, the central claim collapses.

Watch

Extended reading notes

Core claim

On the HJB PDE of the Merton portfolio problem, a Quantum Physics-Informed Neural Network and a quantum-inspired PINN built from a tensor-rank polynomial circuit achieve lower losses and approximation errors than a classical fully connected PINN while using substantially fewer trainable parameters, and they further outperform a classical PINN that shares a similar inductive bias, furnishing experimental evidence of quantum-induced improvement in the tested settings.

Load-bearing premise

The PDE solution must be well approximated by a polynomial whose tensor rank is moderate enough for the circuit to keep its cost polynomial and to deliver the observed accuracy and parameter savings.

Editorial extensions

If this is right

PDEs whose solutions admit moderate-rank polynomial structure become candidates for the same quantum or quantum-inspired PINN treatment.
Near-term quantum hardware can host the circuit ansatz with only polynomial gate count when rank stays moderate.
Classical simulators of the same circuit yield a practical, parameter-light PINN that already beats ordinary fully connected networks on the Merton problem.
Portfolio-optimization and other HJB-type finance PDEs can be attacked with far smaller trainable models once the low-rank structure is exploited.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

The same rank-based circuit may transfer to other low-dimensional financial HJB equations (e.g., option pricing with stochastic volatility) without redesign of the ansatz.
If rank can be estimated a priori from the PDE coefficients, one could decide automatically whether the quantum circuit is worth deploying versus a classical dense PINN.
The quantum-inspired classical model already extracts most of the reported gain, suggesting that the inductive bias, not necessarily quantum hardware, is the immediate practical payoff.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

We do not have the paper. The cache dump is YOLOv11 Demystified (2604.03349), not Learning PDEs for Portfolio Optimization with Quantum PINNs (2604.03346). Every load-bearing claim—circuit construction, existence argument, losses, parameter counts, and the “quantum-induced improvement” vs a matched-bias classical PINN—sits behind an unreadable experimental section. Treat this as abstract-only until the right PDF appears.

From the abstract alone, the actual move is clear enough. They build a parameterized circuit that realizes a multivariate polynomial via tensor-rank decomposition, so resource cost drops from exponential to polynomial when the rank stays moderate. That circuit becomes both a QPINN and a classical quantum-inspired PINN, with an existence guarantee that the PDE solution can be approximated by such a rank-structured polynomial. They then target the Merton HJB (optimal mix of risky vs risk-free asset) and report lower residual/loss and approximation error than a fully connected PINN, with far fewer trainable parameters, and still better than a classical PINN given a similar inductive bias. If those numbers hold, that is a useful, resource-aware result for finance PDEs that have exploitable low-rank structure—not a field reorganizer, but a concrete near-term pathway.

The soft spot is exactly the one the abstract flags: everything hinges on the solution admitting a moderate-rank tensor polynomial. Without the manuscript we cannot check whether Merton HJB actually has that structure, whether the classical baselines are fair, or whether the parameter win is real rather than an under-powered FC net. Circularity risk looks ordinary PINN residual minimization, not tautological, but again unverified.

Who it is for: people already working quantum ML / PINNs for HJB-type finance PDEs who care about parameter efficiency and structured ansätze. A serious editor should send a real version of this to referees; the abstract is coherent and the comparison design (including a matched-bias classical control) is the right kind of claim. I would not cite or bring it to reading group on the current dump. Get the correct full text, then re-read the circuit, the rank hypothesis against the actual HJB solution, and the tables. Until then, no verdict on the experiments.

Referee Report

3 major / 2 minor

Summary. The abstract claims a parameterized quantum circuit that realizes multivariate polynomials via tensor-rank decomposition, reducing resource scaling from exponential to polynomial when the rank is moderate; from this circuit the authors define a Quantum Physics-Informed Neural Network (QPINN) and a quantum-inspired PINN that are guaranteed to represent a tensor-rank polynomial approximation of a PDE solution. Numerical experiments on the Hamilton–Jacobi–Bellman PDE of the Merton portfolio problem are said to show lower residual losses and approximation errors than a classical fully connected PINN, with substantially fewer trainable parameters, and further gains over a classical PINN engineered to share a similar inductive bias, presented as evidence of quantum-induced improvement. The body supplied under this arXiv identifier is, however, an unrelated computer-vision manuscript on YOLOv11 object detection; none of the claimed circuits, existence arguments, baselines, or numerical tables appear in the provided full text.

Significance. If the abstract’s claims were substantiated—moderate-rank polynomial structure for the Merton HJB solution, a concrete PQC realizing that structure, existence guarantees for the QPINN/quantum-inspired approximants, and controlled experiments showing parameter-efficient gains over both a generic and an inductive-bias-matched classical PINN—the work would be a useful contribution to near-term quantum and quantum-inspired solvers for structured PDEs in quantitative finance. Those strengths cannot be credited on the present submission because the load-bearing constructions and results are absent from the manuscript body.

major comments (3)

The full text under review is the YOLOv11 object-detection paper (backbone/neck/head, C3K2, SPPF, C2PSA, COCO mAP/FPS), not a manuscript on quantum or quantum-inspired PINNs. Title, abstract, and body are therefore inconsistent; every central claim of the abstract (PQC for tensor-rank polynomials, existence of a tensor-rank approximation, Merton HJB experiments, parameter counts, loss/error comparisons, and “quantum-induced improvement”) is unsupported by any equation, algorithm, theorem, table, or figure in the supplied text.
Abstract claim of resource reduction “from exponential to polynomial when the corresponding tensor rank is moderate” and of an existence guarantee for a tensor-rank polynomial approximation cannot be audited: no definition of the PQC, no statement of the rank hypothesis for the Merton HJB solution, and no proof or constructive argument appear in the body.
Abstract experimental claims—lower losses and approximation errors than a fully connected PINN, fewer trainable parameters, and outperformance of a classical PINN with “similar inductive bias”—are uncheckable. There are no residual definitions, training protocols, baseline architectures, parameter counts, error metrics, or statistical controls in the provided manuscript, so the attribution of gains to quantum structure cannot be assessed.

minor comments (2)

Even at the abstract level, the phrase “quantum-induced improvement” should be reserved for comparisons that isolate circuit structure from inductive bias and optimization protocol; the abstract asserts such a control but the body does not implement it.
The abstract should name the precise form of the Merton HJB PDE (state variables, utility, market parameters) so that the moderate-rank hypothesis is falsifiable once a correct manuscript is supplied.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation identifiable; abstract describes ordinary residual-minimization PINN training plus empirical comparison, and the supplied full text is the wrong manuscript.

full rationale

The only inspectable content for arXiv:2604.03346 is the abstract. It states that a PQC implements a polynomial via tensor-rank decomposition, that QPINN / quantum-inspired PINN therefore admit a polynomial approximation of the PDE solution, and that these models are trained and compared on the known Merton HJB residual against classical PINNs. Residual minimization against a fixed PDE is ordinary PINN training; the reported lower losses and parameter counts are experimental outcomes, not quantities forced by construction from fitted constants. The existence claim is a representability statement, not a self-definitional loop. The CACHEABLE full-text block is the unrelated YOLOv11 paper (2604.03349), so no equation-level derivation chain of the quantum work can be walked or quoted. Under the hard rule that circularity may be asserted only when a specific reduction can be exhibited from the paper’s own text, no circular step is present. Score 0 with empty steps is therefore the correct finding.

Assumptions & free parameters 0 free parameters · 4 assumptions · 2 invented entities

Abstract-only review. Load-bearing ingredients are the moderate-rank tensor decomposition of the solution polynomial, the claim that the PQC implements that polynomial with polynomial resources, the PINN residual training setup, and the Merton HJB model. No free parameters or invented physical entities can be enumerated from numbers in the abstract; the main modeling axioms are domain-standard finance/PDE assumptions plus the paper-specific moderate-rank hypothesis.

assumptions (4)

ad hoc to paper The target PDE solution can be approximated by a polynomial that admits a moderate tensor-rank decomposition, making quantum resource scaling polynomial rather than exponential.
Stated as the condition under which the PQC reduces complexity; central to both resource claim and existence guarantee.
domain assumption The Merton portfolio problem is correctly reduced to an HJB PDE whose solution yields the optimal investment fraction between risky and risk-free assets.
Standard continuous-time finance modeling assumption used as the numerical testbed.
domain assumption Physics-informed residual minimization (PDE residual plus boundary/initial conditions) is a valid training objective for approximating the PDE solution.
Inherited from the classical PINN literature; both quantum and classical models rely on it.
ad hoc to paper A classical PINN can be constructed to share a similar inductive bias to the quantum models so that residual performance gaps can be attributed to quantum structure.
Required for the abstract’s claim of “quantum-induced improvement” rather than mere architectural bias.

invented entities (2)

Parameterized quantum circuit implementing a multivariate polynomial via tensor rank decomposition
purpose: Serve as the function approximator inside QPINN / quantum-inspired PINN with reduced resource scaling when rank is moderate.
Core architectural object introduced for this work; independent hardware validation is not provided in the abstract.
Quantum Physics-Informed Neural Network (QPINN) and Quantum-inspired PINN with existence guarantee for a tensor-rank polynomial approximation
purpose: Guarantee that a PDE solution approximation of the stated polynomial form exists and can be trained via residual loss.
Named models built on the PQC; existence claim is paper-internal without inspectable proof in the available text.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Learning PDEs for Portfolio Optimization with Quantum Physics-Informed Neural Networks." pith.science (2026). https://pith.science/paper/IO5L5OLL

@misc{pith2026260403346,
  author       = {Pith},
  title        = {Pith review of: Learning PDEs for Portfolio Optimization with Quantum Physics-Informed Neural Networks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/IO5L5OLL}},
  note         = {Machine review of arXiv:2604.03346}
}

read the original abstract

Partial differential equations (PDEs) play a crucial role in financial mathematics, particularly in portfolio optimization, and solving them using classical numerical or neural network methods has always posed significant challenges. Here, we investigate the potential role of quantum circuits for solving PDEs. We design a parameterized quantum circuit (PQC) for implementing a polynomial based on tensor rank decomposition, reducing the quantum resource complexity from exponential to polynomial when the corresponding tensor rank is moderate. Building on this circuit, we develop a Quantum Physics-Informed Neural Network (QPINN) and a Quantum-inspired PINN, both of which guarantee the existence of an approximation of the PDE solution, and this approximation can be represented as a polynomial that incorporates tensor rank decomposition. Numerical experiments are conducted on the Hamilton--Jacobi--Bellman (HJB) PDE arising from the Merton portfolio optimization problem, which determines the optimal investment fraction between a risky and a risk-free asset. The results show that our quantum models achieve lower losses and approximation errors than a classical fully connected PINN while using substantially fewer trainable parameters. Our quantum models further outperform a classical PINN constructed to share a similar inductive bias, providing experimental evidence of quantum-induced improvement in the tested settings and highlighting a resource-efficient pathway toward classical and near-term quantum solvers for PDEs with exploitable solution structure.

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

1 extracted references · 1 linked inside Pith

[1]

1.YOLOv11 architecture diagram, showing the backbone (left), neck with SPPF, and multi-scale detection head

YOLOv11 Demystified: A Practical Guide to High-Performance Object Detection Nikhileswara Rao Sulake1 Rajiv Gandhi University of Knowledge Technologies, Nuzvid, India, nikhil01446@gmail.com, WWW:https://nikhil-rao20.github.io/ Fig. 1.YOLOv11 architecture diagram, showing the backbone (left), neck with SPPF, and multi-scale detection head. Abstract. YOLOv11...

arXiv 2026

Pith tools

Reviewed July 13, 2026 · model on record in the stance chip above.

[1] [1]

1.YOLOv11 architecture diagram, showing the backbone (left), neck with SPPF, and multi-scale detection head

YOLOv11 Demystified: A Practical Guide to High-Performance Object Detection Nikhileswara Rao Sulake1 Rajiv Gandhi University of Knowledge Technologies, Nuzvid, India, nikhil01446@gmail.com, WWW:https://nikhil-rao20.github.io/ Fig. 1.YOLOv11 architecture diagram, showing the backbone (left), neck with SPPF, and multi-scale detection head. Abstract. YOLOv11...

arXiv 2026