arXiv subjects

Wenbo Cao

Publications and source records attributed to Wenbo Cao.

At least 19 recordsLinked to original sources

arXiv2026

Single-condition neural solvers encode transferable response spaces for parametric differential equations

Operator learning for parametric partial differential equations (PDEs) typically builds global models over prescribed domains, requiring cross-condition data or costly physics-constrained training. Here we show that the output Jacobian of a neural solution model trained at one condition defines a reusable response space for cross-condition solution variations. We introduce Linearized Subspace Transfer (LST) to exploit this space and recover target solutions by minimizing the target PDE-system residual over response-space coordinates. Because any single response space has finite coverage, Active Transfer Modeling (ATM) uses post-transfer residuals as coverage indicators to selectively acquire response spaces from additional single-condition models. Across six systems, single-condition response spaces supported cross-condition transfer, with enrichment improving accuracy when added spaces expanded representation capacity. Relative to evaluated physics-informed operator baselines, ATM reduced error and offline construction cost, with orders-of-magnitude accuracy gains in representative cases and millisecond-to-second target adaptation. These results establish neural solvers as reusable local parametric models.

cs.LG↗

arXiv2026

Linearized subspace refinement framework to expose hidden accuracy in trained neural networks

Neural networks trained by gradient-based methods often exhibit optimization-induced accuracy plateaus in scientific machine learning tasks. We present Linearized Subspace Refinement (LSR), an architecture-agnostic post-training framework that exploits the local linearized model at a fixed trained state. By solving a reduced direct least-squares problem in a Jacobian-defined low-dimensional space, LSR computes a subspace-optimal linearized correction and yields a refined predictor with markedly improved accuracy. Across function approximation, data-driven operator learning, physics-informed operator fine-tuning, and noisy inverse problems, LSR shows that standard nonlinear training can remain far above this subspace-attainable error level. Similar accuracy plateaus persist even for the convex quadratic problem from local linearization when solved with standard iterative optimizers, identifying numerical ill-conditioning as a primary bottleneck. LSR frequently delivers order-of-magnitude error reductions, while the subspace rank provides an explicit capacity-control mechanism that balances correction strength, numerical stability, and noise sensitivity. Together, LSR exposes conditioning-limited attainable accuracy in trained-state linearized models and provides direct access to it.

cs.LG↗

arXiv2026

FinTrace: Holistic Trajectory-Level Evaluation of LLM Tool Calling for Long-Horizon Financial Tasks

Recent studies demonstrate that tool-calling capability enables large language models (LLMs) to interact with external environments for long-horizon financial tasks. While existing benchmarks have begun evaluating financial tool calling, they focus on limited scenarios and rely on call-level metrics that fail to capture trajectory-level reasoning quality. To address this gap, we introduce FinTrace, a benchmark comprising 800 expert-annotated trajectories spanning 34 real-world financial task categories across multiple difficulty levels. FinTrace employs a rubric-based evaluation protocol with nine metrics organized along four axes -- action correctness, execution efficiency, process quality, and output quality -- enabling fine-grained assessment of LLM tool-calling behavior. Our evaluation of 13 LLMs reveals that while frontier models achieve strong tool selection, all models struggle with information utilization and final answer quality, exposing a critical gap between invoking the right tools and reasoning effectively over their outputs. To move beyond diagnosis, we construct FinTrace-Training, the first trajectory-level preference dataset for financial tool-calling, containing 8,196 curated trajectories with tool-augmented contexts and preference pairs. We fine-tune Qwen-3-8B/32B using supervised fine-tuning followed by direct preference optimization (DPO) and show that training on FinTrace-Training consistently improves intermediate reasoning metrics, with DPO more effectively suppressing failure modes. However, end-to-end answer quality remains a bottleneck, indicating that trajectory-level improvements do not yet fully propagate to final output quality.

cs.AI↗

arXiv2026

Physics-informed neural networks for shock capturing in inviscid flows around an airfoil

Physics-informed neural networks (PINNs) have shown remarkable prospects in solving forward and inverse problems involving partial differential equations (PDEs). However, PINNs still face challenges in solving fluid mechanics problems involving shocks, especially in steady inviscid flows around an airfoil, where they may even fail to capture shocks. In this study, we first point out that the reason PINNs fail to capture shocks is that the steady Euler equations used to construct the loss function impose weak constraints, which are difficult to correct the continuous function approximation preference of neural networks, causing gradient descent converges to a smooth local optimum. Based on this insight, we propose to strengthen the physical constraints by reconstructing steady shock capturing as temporal evolution that gradually converges to the steady state solution. The unsteady Euler equations constructed by introducing time derivative terms into the steady equations are used to constrain PINNs. The output of PINNs is no longer required to directly approximate a flow field with shocks by minimizing the residuals of the steady Euler equations. Instead, shocks gradually form under the guidance of the temporal evolution law of the flow field. This additional temporal penalty alleviates the tendency of PINNs to converge to a smooth local optimum. Since obtaining the steady state solution requires solving the unsteady Euler equations over a long time in the time dimension, while the capability of PINNs to solve such problems is poor, we introduce a PDE loss function that embeds the concept of pseudo time-stepping to avoid this issue. In addition, to further improve the shock capturing accuracy, we develop a simplified formulation of the Euler equations. By solving four forward problems involving different flow conditions and geometries, we validate the effectiveness of the proposed method.

physics.flu-dyn↗

arXiv2026

Verified residual-specific explicit derivative kernels for physics-informed learning and discretized PDE adjoints

Derivative computation is central to scientific computing, from space-time derivatives in physics-informed neural networks (PINNs) to residual Jacobian actions and discrete-adjoint operators in computational fluid dynamics (CFD). General-purpose automatic differentiation (AD) reduces implementation effort, but can incur substantial runtime and memory overhead for high-order residuals and complex discretized operators. Explicit derivative kernels can exploit problem-specific structure and provide efficient, controllable evaluations, but their use has been limited by derivation and implementation costs. This work revisits explicit differentiation (ED) as a residual-specific and verifiable route enabled by agent-assisted implementation and stringent numerical verification. For PINNs, we propose residual-specific partial-jet propagation, which makes the derivative-state closure of the target PDE residual explicit and realizes it through specialized layerwise kernels, rather than relying only on nested AD or a generic Taylor-mode transform. Relative to nested AD, the resulting ED kernels achieve floating-point-level agreement in residual and parameter-gradient evaluations and accelerate complete PINN training, often reaching 2-4x speedups while reducing peak GPU memory in most cases. For discretized PDE adjoints, we apply the same verification-driven strategy to a finite-volume CFD residual. The generated tangent-action and transpose-action kernels pass Taylor-remainder, inner-product, and reduced-gradient consistency checks, and are embedded into a GPU-resident discrete-adjoint workflow for freestream Mach-number and angle-of-attack inversion. These results suggest that verified explicit derivative kernels, supported by agent-assisted implementation, can serve as a practical, structure-aware complement to general-purpose AD for derivative-intensive scientific computing.

physics.comp-ph↗

arXiv2026

Herculean: An Agentic Benchmark for Financial Intelligence

As AI agents improve, the central question is no longer whether they can solve isolated well-defined financial tasks, but whether they can reliably carry out financial professional work. Existing financial benchmarks offer only a partial view of this ability, as they primarily evaluate static competencies such as question answering, retrieval, summarization, and classification. We introduce Herculean, the first skilled benchmark for agentic financial intelligence spanning four representative workflows, including Trading, Hedging, Market Insights, and Auditing. Each workflow is instantiated as a standardized MCP-based skill environment with its own tools, interaction dynamics, constraints, and success criteria, enabling consistent end-to-end assessment of heterogeneous agent systems. Across frontier agents, we find agents perform relatively well on Trading and Market Insights, but struggle substantially on Hedging and Auditing, where long-horizon coordination, state consistency, and structured verification are critical. Overall, our results point to a key gap in current agents in turning financial reasoning into dependable workflow execution in high-stakes financial workflows.

cs.AI↗

arXiv2026

Optimization-Based Discovery of A Non-Attracting Flow State in An Oscillating-Cylinder Wake

In the flow past a stationary circular cylinder, the classical Karman vortex street arises from a Hopf bifurcation of the steady flow at the critical Reynolds number. Although this solution becomes dynamically unstable beyond this point, it remains an exact solution of the governing equations. Motivated by this observation, we investigates whether similar non-attracting flow solutions exist in the flow past a forced oscillating cylinder at supercritical Re. In the present study, while employing PINNs to investigate the flow past a forced oscillating cylinder, we identify a class of flow solutions that are inaccessible through direct time-stepping simulations. The obtained solution remains phase-locked with the cylinder oscillation frequency, despite the corresponding parameters lying outside the lock-in regime. To verify this solution, the obtained PINNs solution is used as the initial guess for an optimization based on the optimizing a discrete loss (ODIL) framework. The results show that the solution can be consistently maintained during the optimization process. This indicates that the solution is self-consistent in the optimization sense, although it does not an attracting state of the original dynamical system. To understand the reason, we compare the numerical evolution mechanisms of each solvers. The results indicate that, flow states that satisfy the governing equations but are dynamically non-attracting can be identified and maintained as minima of the optimization problem. For the flow past a forced oscillating cylinder, non-attracting periodic solutions that satisfy the governing equations exist in addition to the attracting states obtained by conventional time-stepping simulations. Optimization-based solvers can therefore reveal such flow states that are difficult to obtain through direct time integration, providing a new perspective for understanding complex wake dynamics.

physics.flu-dyn↗

arXiv2026

Solving compressible Navier-Stokes equations using the feature-enhanced neural network

Physics-informed neural networks (PINNs) have shown remarkable prospects in solving partial differential equations (PDEs) involving fluid mechanics. However, the method has so far succeeded only in inviscid flows and incompressible viscous flows, while the solution of compressible viscous flows still faces significant challenges. In previous work, we proposed a feature-enhanced neural network (FENN), which enhances the ability of PINNs to approximate flows by introducing beneficial features into the network inputs, thereby improving the performance in solving PDEs. In this study, we extend FENN to compressible viscous flows, which are governed by the compressible Navier-Stokes equations including the continuity, momentum, and energy equations. By solving four forward problems under different flow conditions and geometries together with a parametric problem involving angle of attack, we validate the effectiveness of FENN. In contrast, existing advanced methods that are well established for inviscid flows and incompressible viscous flows fail in this scenario. To the best of our knowledge, this is the first time that a PINN-like method has successfully solved forward and parametric problems involving compressible viscous flows.

physics.flu-dyn↗

arXiv2025

Probing Scientific General Intelligence of LLMs with Scientist-Aligned Workflows

Despite advances in scientific AI, a coherent framework for Scientific General Intelligence (SGI)-the ability to autonomously conceive, investigate, and reason across scientific domains-remains lacking. We present an operational SGI definition grounded in the Practical Inquiry Model (PIM: Deliberation, Conception, Action, Perception) and operationalize it via four scientist-aligned tasks: deep research, idea generation, dry/wet experiments, and experimental reasoning. SGI-Bench comprises over 1,000 expert-curated, cross-disciplinary samples inspired by Science's 125 Big Questions, enabling systematic evaluation of state-of-the-art LLMs. Results reveal gaps: low exact match (10--20%) in deep research despite step-level alignment; ideas lacking feasibility and detail; high code executability but low execution result accuracy in dry experiments; low sequence fidelity in wet protocols; and persistent multimodal comparative-reasoning challenges. We further introduce Test-Time Reinforcement Learning (TTRL), which optimizes retrieval-augmented novelty rewards at inference, enhancing hypothesis novelty without reference answer. Together, our PIM-grounded definition, workflow-centric benchmark, and empirical insights establish a foundation for AI systems that genuinely participate in scientific discovery.

cs.AI↗

arXiv2025

A matrix preconditioning framework for physics-informed neural networks based on adjoint method

Physics-informed neural networks (PINNs) have recently emerged as a popular approach for solving forward and inverse problems involving partial differential equations (PDEs). Compared to fully connected neural networks, PINNs based on convolutional neural networks offer advantages in the hard enforcement of boundary conditions and in reducing the computational cost of partial derivatives. However, the latter still struggles with slow convergence and even failure in some scenarios. In this study, we propose a matrix preconditioning method to improve the convergence of the latter. Specifically, we combine automatic differentiation with matrix coloring to compute the Jacobian matrix of the PDE system, which is used to construct the preconditioner via incomplete LU factorization. We subsequently use the preconditioner to scale the PDE residual in the loss function in order to reduce the condition number of the Jacobian matrix, which is key to improving the convergence of PINNs. To overcome the incompatibility between automatic differentiation and triangular solves in the preconditioning, we also design a framework based on the adjoint method to compute the gradients of the loss function with respect to the network parameters. By numerical experiments, we validate that the proposed method successfully and efficiently solves the multi-scale problem and the high Reynolds number problem, in both of which PINNs fail to obtain satisfactory results.

math.NA↗

arXiv2025

Overcoming the Loss Conditioning Bottleneck in Optimization-Based PDE Solvers: A Novel Well-Conditioned Loss Function

Optimization-based PDE solvers that minimize scalar loss functions have gained increasing attention in recent years. These methods either define the loss directly over discrete variables, as in Optimizing a Discrete Loss (ODIL), or indirectly through a neural network surrogate, as in Physics-Informed Neural Networks (PINNs). However, despite their promise, such methods often converge much more slowly than classical iterative solvers and are commonly regarded as inefficient. This work provides a theoretical insight, attributing the inefficiency to the use of the mean squared error (MSE) loss, which implicitly forms the normal equations, squares the condition number, and severely impairs optimization. To address this, we propose a novel Stabilized Gradient Residual (SGR) loss. By tuning a weight parameter, it flexibly modulates the condition number between the original system and its normal equations, while reducing to the MSE loss in the limiting case. We systematically benchmark the convergence behavior and optimization stability of the SGR loss within both the ODIL framework and PINNs-employing either numerical or automatic differentiation-and compare its performance against classical iterative solvers. Numerical experiments on a range of benchmark problems demonstrate that, within the ODIL framework, the proposed SGR loss achieves orders-of-magnitude faster convergence than the MSE loss. Further validation within the PINNs framework shows that, despite the high nonlinearity of neural networks, SGR consistently outperforms the MSE loss. These theoretical and empirical findings help bridge the performance gap between classical iterative solvers and optimization-based solvers, highlighting the central role of loss conditioning, and provide key insights for the design of more efficient PDE solvers.

cs.CE↗

arXiv2025

FENN: Feature-enhanced neural network for solving partial differential equations involving fluid mechanics

Physics-informed neural networks (PINNs) have shown remarkable prospects in solving forward and inverse problems involving partial differential equations (PDEs). However, PINNs still face the challenge of high computational cost in solving strongly nonlinear PDEs involving fluid dynamics. In this study, inspired by the input design in surrogate modeling, we propose a feature-enhanced neural network. By introducing geometric features including distance and angle or physical features including the solution of the potential flow equation in the inputs of PINNs, FENN can more easily learn the flow, resulting in better performance in terms of both accuracy and efficiency. We establish the feature networks in advance to avoid the invalid PDE loss in FENN caused by neglecting the partial derivatives of the features with respect to space-time coordinates. Through five numerical experiments involving forward, inverse, and parametric problems, we verify that FENN generally reduces the computational cost of PINNs by approximately four times. In addition, the numerical experiments also demonstrate that the proposed method can reduce the number of observed data for inverse problem and successfully solve the parametric problem where PINNs fail.

physics.flu-dyn↗

arXiv2025

Solving all laminar flows around airfoils all-at-once using a parametric neural network solver

Recent years have witnessed increasing research interests of physics-informed neural networks (PINNs) in solving forward, inverse, and parametric problems governed by partial differential equations (PDEs). Despite their promise, PINNs still face significant challenges in many scenarios due to ill-conditioning. Time-stepping-oriented neural network (TSONN) addresses this by reformulating the ill-conditioned optimization problem into a series of well-conditioned sub-problems, greatly improving its ability to handle complex scenarios. This paper presents a new solver for laminar flow around airfoils based on TSONN and mesh transformation, validated across various test cases. Specifically, the solver achieves mean relative errors of approximately 3.6% for lift coefficients and 1.4% for drag coefficients. Furthermore, this paper extends the solver to parametric problems involving flow conditions and airfoil shapes, covering nearly all laminar flow scenarios in engineering. The shape parameter space is defined as the union of 30% perturbations applied to each airfoil in the UIUC airfoil database, with Reynolds numbers ranging from 100 to 5000 and angles of attack spanning from -5° to 15°. The parametric solver solves all laminar flows within the parameter space in just 4.6 day, at approximately 40 times the computational cost of solving a single flow. The model training involves hundreds of millions of flow conditions and airfoil shapes, ultimately yielding a surrogate model with strong generalization capability that does not require labeled data. Specifically, the surrogate model achieves average errors of 4.6% for lift coefficients and 1.1% for drag coefficients, demonstrating its potential for high generalizability, cost-effectiveness, and efficiency in addressing high-dimensional parametric problems and surrogate modeling.

physics.flu-dyn↗

arXiv2024

Physics Informed Neural Networks (PINNs) as intelligent computing technique for solving partial differential equations: Limitation and Future prospects

In recent years, Physics-Informed Neural Networks (PINNs) have become a representative method for solving partial differential equations (PDEs) with neural networks. PINNs provide a novel approach to solving PDEs through optimization algorithms, offering a unified framework for solving both forward and inverse problems. However, some limitations in terms of solution accuracy and generality have also been revealed. This paper systematically summarizes the limitations of PINNs and identifies three root causes for their failure in solving PDEs: (1) Poor multiscale approximation ability and ill-conditioning caused by PDE losses; (2) Insufficient exploration of convergence and error analysis, resulting in weak mathematical rigor; (3) Inadequate integration of physical information, causing mismatch between residuals and iteration errors. By focusing on addressing these limitations in PINNs, we outline the future directions and prospects for the intelligent computing of PDEs: (1) Analysis of ill-conditioning in PINNs and mitigation strategies; (2) Improvements to PINNs by enforcing temporal causality; (3) Empowering PINNs with classical numerical methods.

physics.comp-ph↗

arXiv2024

An analysis and solution of ill-conditioning in physics-informed neural networks

Physics-informed neural networks (PINNs) have recently emerged as a novel and popular approach for solving forward and inverse problems involving partial differential equations (PDEs). However, achieving stable training and obtaining correct results remain a challenge in many cases, often attributed to the ill-conditioning of PINNs. Nonetheless, further analysis is still lacking, severely limiting the progress and applications of PINNs in complex engineering problems. Drawing inspiration from the ill-conditioning analysis in traditional numerical methods, we establish a connection between the ill-conditioning of PINNs and the ill-conditioning of the Jacobian matrix of the PDE system. Specifically, for any given PDE system, we construct its controlled system. This controlled system allows for adjustment of the condition number of the Jacobian matrix while retaining the same solution as the original system. Our numerical findings suggest that the ill-conditioning observed in PINNs predominantly stems from the Jacobian matrix. As the condition number of the Jacobian matrix decreases, PINNs exhibit faster convergence rates and higher accuracy. Building upon this understanding and the natural extension of controlled systems, we present a general approach to mitigate the ill-conditioning of PINNs, leading to successful simulations of the three-dimensional flow around the M6 wing at a Reynolds number of 5,000. To the best of our knowledge, this is the first time that PINNs have been successful in simulating such complex systems, offering a promising new technique for addressing industrial complexity problems. Our findings also offer valuable insights guiding the future development of PINNs.

physics.flu-dyn↗

arXiv2024

New Interpretation for error propagation of data-driven Reynolds stress closures via global stability analysis

In light of the challenges surrounding convergence and error propagation encountered in Reynolds-averaged Navier-Stokes (RANS) equations with data-driven Reynolds stress closures, researchers commonly attribute these issues to ill-conditioning through conditional number analysis. This paper delves into an additional factor, numerical instability, contributing to these challenges. We conduct global stability analysis for the RANS equations, closed by the Reynolds stress of direct numerical simulation (DNS), with the time-averaged solution of DNS as the base flow. Our findings reveal that, for turbulent channel flow at high Reynolds numbers, significant ill-conditioning exists, yet the system remains stable. Conversely, for separated flow over periodic hills, notable ill-conditioning is absent, but unstable eigenvalues are present, indicating that error propagation arises from the mechanism of numerical instability. Furthermore, the effectiveness of the decomposition method employing eddy viscosity is compared, results show that the spatial distribution and amplitude of eddy viscosity influences the numerical stability.

physics.flu-dyn↗

arXiv2024

A solver for subsonic flow around airfoils based on physics-informed neural networks and mesh transformation

Physics-informed neural networks (PINNs) have recently become a new popular method for solving forward and inverse problems governed by partial differential equations (PDEs). However, in the flow around airfoils, the fluid is greatly accelerated near the leading edge, resulting in a local sharper transition, which is difficult to capture by PINNs. Therefore, PINNs are still rarely used to solve the flow around airfoils. In this study, we combine physical-informed neural networks with mesh transformation, using neural network to learn the flow in the uniform computational space instead of physical space. Mesh transformation avoids the network from capturing the local sharper transition and learning flow with internal boundary (wall boundary). We successfully solve inviscid flow and provide an open-source subsonic flow solver for arbitrary airfoils. Our results show that the solver exhibits higher-order attributes, achieving nearly an order of magnitude error reduction over second-order finite volume methods (FVM) on very sparse meshes. Limited by the learning ability and optimization difficulties of neural network, the accuracy of this solver will not improve significantly with mesh refinement. Nevertheless, it achieves comparable accuracy and efficiency to second-order FVM on fine meshes. Finally, we highlight the significant advantage of the solver in solving parametric problems, as it can efficiently obtain solutions in the continuous parameter space about the angle of attack.

physics.flu-dyn↗

arXiv2024

A complete state-space solution model for inviscid flow around airfoils based on physics-informed neural networks

Engineering problems often involve solving partial differential equations (PDEs) over a range of similar problem setups with various state parameters. In traditional numerical methods, each problem is solved independently, resulting in many repetitive tasks and expensive computational costs. Data-driven modeling has alleviated these issues, enabling fast solution prediction. Nevertheless, it still requires expensive labeled data and faces limitations in modeling accuracy, generalization, and uncertainty. The recently developed methods for solving PDEs through neural network optimization, such as physics-informed neural networks (PINN), enable the simultaneous solution of a series of similar problems. However, these methods still face challenges in achieving stable training and obtaining correct results in many engineering problems. In prior research, we combined PINN with mesh transformation, using neural network to learn the solution of PDEs in the computational space instead of physical space. This approach proved successful in solving inviscid flow around airfoils. In this study, we expand the input dimensions of the model to include shape parameters and flow conditions, forming an input encompassing the complete state-space (i.e., all parameters determining the solution are included in the input). Our results show that the model has significant advantages in solving high-dimensional parametric problems, achieving continuous solutions in a broad state-space in only about 18.8 hours. This is a task that traditional numerical methods struggle to accomplish. Once established, the model can efficiently complete airfoil flow simulation and shape inverse design tasks in approximately 1 second. Furthermore, we introduce a pretraining-finetuning method, enabling the fine-tuning of the model for the task of interest and quickly achieving accuracy comparable to the finite volume method.

physics.flu-dyn↗