Searcharxiv⌕ Search

arXiv subjects

Zaiwen Wen

Publications and source records attributed to Zaiwen Wen.

At least 19 recordsLinked to original sources
arXiv2026

Learning to Restart: Differentiable Restart Policies for Optimization Dynamics

Adaptive restart mechanisms, which reset the momentum or other auxiliary state of an iterative optimizer, play a central role in the empirical performance of accelerated first-order methods, yet existing restart rules remain almost entirely hand-designed. The obstruction to learning restart policies is that, unlike the typical Learning to Optimize (L2O) setting, the discrete restart events make the resulting trajectory non-differentiable in the algorithm parameters and thus block gradient-based optimization of the restart policy. We resolve this obstruction by developing a differentiable framework for restart policies through their continuous-time limit. We introduce surrogate gradients for the restart events and optimization trajectories that are efficient to compute and provably converge to their continuous-time counterparts. We validate the effectiveness of our surrogate gradients in L2O settings, training restart policies that improve upon standard handcrafted adaptive restart criteria on quadratic and linear programs, and that transfer zero-shot from synthetic quadratics to logistic regression on real datasets.

cs.LG↗
arXiv2026

The multiplier step size in classical two-block ADMM: convergence and counterexamples

How large can the multiplier step be in the classical two-block alternating direction method of multipliers (ADMM)? We keep the penalty fixed and use the usual primal updates. We consider finite-dimensional problems with closed proper convex objectives. We assume that a Karush-Kuhn-Tucker point exists and that each block subproblem attains its minimum at every step. Under these assumptions, the constraint images and the multiplier converge for a range of factors that extends beyond the golden ratio. The constraint matrices may be rank deficient. For larger steps, we construct two fixed compact polytopes and give an explicit threshold below two. The polytopes admit an infeasible periodic orbit for every factor at or above this threshold. At that threshold, ADMM can cycle even for smooth strongly convex objectives. This gives a negative answer to Glowinski's question about the full interval from zero to two. A different pair of piecewise-quadratic objectives gives geometrically growing residuals. For the general convex class, we locate the endpoint of the uniform convergence interval near 1.88688256. A convergence certificate on the unit circle gives the lower bound, while a separate periodic orbit gives the upper bound. Exact arithmetic verifies that the two bounds differ by less than 0.000000174.

math.OC↗
arXiv2026

Riemannian EXTRA: Communication-efficient decentralized optimization over compact submanifolds with data heterogeneity

We consider decentralized optimization over a compact Riemannian submanifold in a network of $n$ nodes, where each node $i$ owns a smooth nonconvex local objective $f_i$ defined by its private data, and the goal is to minimize the global objective $F(x)=\frac{1}{n}\sum_{i=1}^n f_i(x)$ using only local communications. A naive Riemannian extension of EXTRA leaves the auxiliary normal-space error uncontrolled and does not inherit the Euclidean convergence guarantees. We develop projection-based Riemannian EXTRA (REXTRA) and a retraction-based counterpart that incorporate a normal correction into both the communicated variable and the auxiliary update, while using only one local-variable-sized message per optimization iteration. By jointly controlling consensus, tangent, and normal errors, our analysis interprets the update as a network-based Riemannian inexact-gradient step. For projection-based REXTRA, a sufficiently small constant stepsize and a local initialization yield vanishing consensus, stationary accumulation points, and an $\mathcal O(1/k)$ best-iterate stationarity--consensus bound. For the retraction-based method, we prove that its difference from the projection update is uniformly second order and establish the same convergence properties and rate under a local initialization. Numerical experiments on representative manifold-valued learning tasks demonstrate the communication efficiency of the REXTRA updates.

math.OC↗
arXiv2026

MM-OptBench: A Solver-Grounded Benchmark for Multimodal Optimization Modeling

Optimization modeling translates real decision-making problems into mathematical optimization models and solver-executable implementations. Although language models are increasingly used to generate optimization formulations and solver code, existing benchmarks for this capability are almost entirely text-only. This omits many optimization-modeling tasks that arise in operational practice, where requirements are described in text but essential instance information is often conveyed through visual artifacts such as tables, graphs, maps, schedules, and planning dashboards. We introduce multimodal optimization modeling, a benchmark setting where models must construct both a mathematical formulation and executable solver code from a text-and-visual problem specification. To evaluate this setting, we develop a solver-grounded framework that generates structured optimization instances, verifies each instance with an exact solver, and builds both model-facing inputs and hidden reference files from the same verified source. We instantiate it as MM-OptBench, a benchmark of 780 solver-verified instances spanning 6 optimization families, 26 subcategories, and 3 structural difficulty levels. We evaluate 9 multimodal large language models (MLLMs), including 6 frontier general-purpose and 3 math-specialized models, with aggregate, family-level, difficulty-level, and failure-mode analyses. The task remains far from solved: the best two models reach 52.1% and 51.3% pass@1, while on average across the six general-purpose MLLMs, pass@1 is 43.4% on easy and 15.9% on hard instances. All three math-specialized MLLMs solve 0/780 instances. Post-hoc extraction diagnostics and oracle-reading evaluation reveal difficulties in both multimodal grounding and downstream solver-artifact generation. MM-OptBench provides a concrete testbed for solver-grounded, decision-oriented multimodal intelligence.

cs.AI↗
arXiv2026

Accelerating LLM Pre-Training through Flat-Direction Dynamics Enhancement

Pre-training Large Language Models requires immense computational resources, making optimizer efficiency essential. The optimization landscape is highly anisotropic, with loss reduction driven predominantly by progress along flat directions. While matrix-based optimizers such as Muon and SOAP use fine-grained curvature information to outperform AdamW, their updates tend toward isotropy, which is relatively conservative along flat directions yet potentially aggressive along sharp ones. To address this limitation, we first develop a unified Riemannian Ordinary Differential Equation (ODE) framework that offers a continuous-time interpretation of adaptive algorithms: the preconditioner induces a Riemannian geometry that mitigates ill-conditioning, while momentum serves as a Riemannian damping term that promotes convergence. Guided by these insights, we propose LITE, a generalized acceleration strategy that enhances training dynamics by applying larger Hessian damping coefficients and learning rates along flat trajectories. Extensive experiments demonstrate that LITE significantly accelerates both Muon and SOAP across diverse architectures (Dense, MoE), parameter scales (130M to 1.3B), datasets (C4, Pile), and learning-rate schedules (cosine, wsd). Theoretical analysis confirms that LITE facilitates faster convergence along flat directions in anisotropic landscapes, providing a principled approach to efficient LLM pre-training. The code is available at https://github.com/SHUCHENZHU/LITE.

cs.LG↗
arXiv2026

PACE: Research Proving Agent for Adaptive Conjecture Exploration

Many mathematical research agents focus on solving problems with fixed statements. However, conjecture research, especially in applied mathematics, may require revising candidate questions or conclusions as investigation proceeds, coupling problem formulation with proof development. We present PACE, a research-proof agent that supports this process through a structured research loop that selects mathematical objectives and develops or challenges candidate mechanisms. The whole prover constructs proofs from these mechanisms, and an independent verifier checks them against the original target. Evidence-tracking memory stores assumptions and dependencies arising during research and marks evidence for revalidation when a candidate or condition changes. This memory supports task-specific context composition that keeps checked premises, exploratory material and unresolved obligations distinct. To evaluate proof construction and conjecture research, we introduce ResearchProofBench, an applied-mathematics benchmark focused on optimization, comprising 51 known-proof tasks with 331 hidden checkpoints under historical abstract access and six open-problem cases. The known-proof tasks assess reconstruction of published results, while the open-problem cases assess mathematical investigation and proof development. On the 51 known-proof tasks, PACE achieves an aggregate Proof Score of 75.16 and completes 20 proofs. In conjecture research, PACE develops conditional results, with its submissions receiving higher review scores than the compared baselines.

cs.AI↗
arXiv2026

Accelerating Optimization via Differentiable Stopping Time

Optimization is an important module of modern machine learning applications. Tremendous efforts have been made to accelerate optimization algorithms. A common formulation is achieving a lower loss at a given time. This enables a differentiable framework with respect to the algorithm hyperparameters. In contrast, its dual, minimizing the time to reach a target loss, is believed to be non-differentiable, as the time is not differentiable. As a result, it usually serves as a conceptual framework or is optimized using zeroth-order methods. To address this limitation, we propose a differentiable stopping time and theoretically justify it based on differential equations. An efficient algorithm is designed to backpropagate through it. As a result, the proposed differentiable stopping time enables a new differentiable formulation for accelerating algorithms. We further discuss its applications, such as online hyperparameter tuning and learning to optimize. Our proposed methods show superior performance in comprehensive experiments across various problems, which confirms their effectiveness.

cs.LG↗
arXiv2026

FaithSieve: Fine-Grained Evaluation of Math Proofs with Faithful Formal Evidence

Large language models can now generate complex, multi-step mathematical proofs, but reliably determining their correctness and localizing early logical errors remains a critical challenge. Existing evaluation approaches largely depend on model-based natural-language judgments, which often overlook local reasoning gaps. While formal theorem provers like Lean offer a path to rigorous verification, using them to evaluate informal text requires solving locality and semantic mismatches: a prover might bypass a local flaw by proving an overly broad target, or validate an auto-formalized statement that drifts from the original mathematical intent. To address this, we introduce FaithSieve, a Lean-assisted framework for fine-grained evaluation of natural-language mathematical proofs. FaithSieve decomposes coarse proof steps into local reasoning units, extracts typed proof obligations, and verifies them through a formal evaluation agent. Formal validation is gated by semantic alignment scoring, so Lean evidence is incorporated only when the formal statement faithfully preserves the context, objects, and logical form of the original claim. We construct two expert-verified datasets, ProofLoc-Olympiad and ProofLoc-University, to benchmark first-error localization. On the 350-problem Olympiad dataset, FaithSieve using a GPT-5.4 backbone achieves 81.43% exact first-error accuracy, outperforming the direct-judging baseline of 72.29%. Furthermore, on the 200-problem ProofLoc-University benchmark spanning six advanced domains, FaithSieve reaches 84.5% exact accuracy, compared to 75.0% for the direct judge. Our work demonstrates that decomposing proofs into fine-grained units and grounding them with faithful formal evidence significantly improves reliable evaluation of natural-language reasoning.

cs.AI↗
arXiv2026

Right Answers, Costly Models: The Efficiency Gap in LLM-based Optimization Modeling

Optimization modeling formulates real-world decision problems as mathematical programs that solvers can use to find optimal decisions. Large language models (LLMs) can automate this process, but the resulting correct formulations can require substantial time and memory to construct and solve, limiting practical scalability. Therefore, we systematically investigate whether LLMs can identify problem structure from natural-language descriptions and apply suitable optimization modeling techniques to generate mathematical models and solver code that solve the problems correctly and efficiently. To this end, we first curate OptTips, a knowledge base of 50 expert modeling techniques in eight families. Using this knowledge, we develop OptDachshund, a multi-agent framework that transforms problems from existing optimization benchmarks into new tasks for evaluating LLMs' use of modeling techniques. It constructs conventional and expert mathematical models with solver code for the same task and data, providing baselines for correctness and computational cost. The resulting EfficientOpt benchmark contains 561 expert-reviewed tasks with paired reference implementations. Evaluation of 11 representative LLMs reveals an efficiency gap on correctly solved tasks with comparable measurements: for every LLM, most generated programs take longer to solve than their expert counterparts. Within the comparable reference-size subset, 57\% of programs with correct objective values and fewer variables and linear constraints have longer recorded solver times. Case studies show that different modeling techniques can achieve the same optimal value at similar recorded cost. Faster solving may not reduce execution time if the code takes longer to prepare data and build the model. LLM optimization modeling should therefore be evaluated for both correctness and computational efficiency.

cs.LG↗
arXiv2026

A Human-AI Collaborative Workflow for Mathematical Discovery: A Case Study in Grover-Compatible Riemannian Optimization

We investigate how large language models can be used as research tools in scientific computing while preserving mathematical rigor. We propose a human-in-the-loop workflow for interactive theorem proving and discovery with LLMs. Human experts retain control over problem formulation and assumptions, while the model searches for proofs or contradictions, proposes candidate properties and theorems, and helps construct structures and parameters that satisfy explicit constraints, supported by numerical experiments and simple verification checks. Experts treat these outputs as raw material, further refine them, and organize the results into precise statements and rigorous proofs. We instantiate this workflow in a main case study on the connection between manifold optimization and Grover's quantum search algorithm, where the pipeline identifies invariant subspaces and explores Grover-compatible retractions. The main case study uses the corresponding Grover-compatible convergence analysis, including an $O(\sqrt{N} \log(1/\varepsilon))$ PL-based bound established in the companion mathematical work, to illustrate the refinement stage of the workflow. Prompt records and reusable templates for implementing the workflow are provided. We further include a multi-oracle case study, document representative failed and corrected routes arising from this setting, and provide a structured failure-mode analysis.

cs.HC↗
arXiv2026

Non-Asymptotic Global Convergence of PPO-Clip

Reinforcement learning has gained attention for modern Large Language Model post-training. The actor-only variants of Proximal Policy Optimization (PPO) are widely applied for their efficiency. These algorithms incorporate a clipping mechanism to improve stability. Besides, a regularization term, such as the reverse KL-divergence or a more general \(f\)-divergence, is introduced to control excessive deviation from a reference policy. Despite their empirical success, a rigorous theoretical understanding of the problem and the algorithm's properties is limited. This paper advances the theoretical foundations of the PPO-Clip algorithm by analyzing a deterministic actor-only PPO algorithm within the general RL setting with \(f\)-divergence regularization under the softmax policy parameterization. We derive a non-uniform Lipschitz smoothness condition and a Łojasiewicz inequality for the considered problem. Based on these properties, we establish non-asymptotic global linear convergence in value gap for the forward KL regularizer. For the reverse KL regularizer, we derive global linear convergence from any finite softmax initialization in both value gap and squared policy distance.

math.OC↗
arXiv2026

The Minimum Q-Order of BFGS with Exact Line Search Is One

Powell asked whether the smoothness assumptions underlying classical superlinear convergence force a fixed power law between adjacent iterates of exact-line-search variable-metric methods. We answer this question negatively for BFGS: within the smooth strongly convex setting, the smallest possible adjacent-iterate Q-order is one, and this boundary is attained by a single nonterminating run. In every finite dimension at least two, and for any prescribed radius and Hessian tolerance, we construct an infinitely differentiable, globally strongly convex objective that equals the standard quadratic outside the corresponding ball and whose Hessian remains within the prescribed tolerance of the identity in operator norm. The objective has its unique minimizer at the origin and identity Hessian there. Exact-line-search BFGS, initialized with the identity matrix and started inside that ball, converges Q-superlinearly, yet no fixed power greater than one controls all sufficiently late adjacent errors.

math.OC↗
arXiv2026

A counterexample to global convergence of classical DFP under the standard strong Wolfe conditions

A long-standing open question in quasi-Newton optimization asks whether the classical Davidon--Fletcher--Powell (DFP) method converges globally on uniformly convex objectives when all accepted steps satisfy the standard weak Wolfe conditions. We show that the answer is no, even under the standard strong Wolfe conditions. Fix $0<c_1<2/3$ and $2/3\le c_2<1$. We construct a function $f\in C^2(\mathbb{R}^2)$ such that $\frac{1}{2}I\preceq\nabla^2 f(x)\preceq\frac{3}{2}I$ for all $x\in\mathbb{R}^2$. We also choose a fixed positive definite initial inverse Hessian approximation and a sequence of positive step lengths. The classical DFP iteration is well defined, and all accepted steps satisfy the standard strong Wolfe conditions, but $|\nabla f(x_k)|$ converges to a positive constant. The global Hessian condition number is at most three. The construction uses an alternating two-step DFP sequence near a one-dimensional invariant center manifold. Along this sequence, the smaller eigenvalue of the inverse Hessian approximation tends to zero. The changes in the gradient norm between cycle starts are summable, but the total rotation of the associated eigenvectors is unbounded. The accumulation points of the DFP sequence form a circle. A uniform separation bound allows us to interpolate the prescribed function values and gradients. We add smooth functions with pairwise disjoint supports to a quadratic and keep the global Hessian bounds. An affine change of variables gives an identity-initialized example with problem-dependent Hessian bounds. An orthogonal direct sum extends the result to every dimension $n\ge 2$.

math.OC↗
arXiv2026

A Fixed-Penalty Linearized Augmented Lagrangian Method with Classical Multiplier Updates

Augmented Lagrangian methods are effective for nonlinear equality-constrained optimization, but solving their nonlinear primal subproblems can be expensive. For smooth nonconvex problems with deterministic or stochastic objectives, we propose a nonlinear-residual linearized augmented Lagrangian method (NR-LALM) that replaces this subproblem by a regularized Gauss-Newton-type step while retaining the classical multiplier update based on the nonlinear constraint residual. The resulting step is computed from one symmetric positive-definite linear system, but the mismatch between the linearized primal model and the nonlinear-residual update produces a quadratic constraint-linearization error in the multiplier identity. We show that this error can be controlled under local regularity; multiplier boundedness and trajectory localization are derived rather than assumed. With fixed, accuracy-independent parameters, deterministic NR-LALM finds an $\varepsilon$-approximate Karush-Kuhn-Tucker (KKT) pair in $O(\varepsilon^{-2})$ iterations and first-order oracle evaluations. For stochastic objectives, a projected stochastic path-integrated differential estimator with safeguarded restarts requires, in expectation, $O(\varepsilon^{-3})$ stochastic-gradient evaluations and $O(\varepsilon^{-2})$ constraint and Jacobian evaluations. Compactness and a Kurdyka-Lojasiewicz condition further yield finite-length convergence of the deterministic primal-dual sequence. An optional minimum-norm second-order correction reduces the constraint-linearization error from second to fourth order without changing the complexity orders. All theoretical results are formalized in Lean 4. Numerical experiments confirm the predicted error orders and show favorable performance on high-dimensional deterministic and stochastic problems.

math.OC↗
arXiv2026

ReasFlow: Assisting Reasoning-Centric Scientific Discovery in Applied Mathematics via a Knowledge-Based Multi-Agent System

Recent advances in Large Language Models have fueled autonomous AI agents capable of tackling complex scientific tasks, yet existing automated research systems remain predominantly focused on empirically driven domains with quantitative benchmarks, leaving theory-driven discovery, particularly in mathematically grounded disciplines requiring rigorous proofs and synthesis of domain knowledge, largely underexplored. Key challenges include the difficulty of verifying theoretical reasoning at scale, insufficient reasoning ability for autonomous frontier exploration, and a scarcity of procedural heuristics in the literature. We introduce ReasFlow, an end-to-end autonomous agent system for reasoning-centric scientific discovery that operationalizes a collaborative paradigm where the human expert acts as Principal Investigator while the agent executes rigorous derivations as a capable graduate student. ReasFlow incorporates (i) a robust internal verification loop that audits logical coherence and corrects fundamental errors prior to human inspection, and (ii) an automated knowledge retrieval and self-improvement mechanism that proactively surfaces both declarative facts and overlooked procedural heuristics, substantially reducing expert intervention. The system unifies literature synthesis, algorithm design, theorem proving, experimentation, and manuscript preparation in a single system. Deployed to autonomously generate five complete research papers with rigorous theoretical and empirical content from minimal prompts, ReasFlow consistently achieves the highest evaluation scores among state-of-the-art open-access baselines under a curated LLM-based review rubric. ReasFlow is publicly accessible via the ReasLab platform, providing a collaborative workspace for AI-assisted theoretical research. Github repo: https://github.com/reaslab/ReasFlow.git.

cs.AI↗
arXiv2026

MECA: A Mechanism-Centered Agent for Constructing Well-Specified and Valuable Mathematical Conjectures

Automatically constructing well-specified and valuable mathematical conjectures remains a central challenge in AI-assisted mathematical discovery. Many existing open problems and conjectures are often too broad, underspecified, or difficult to connect to plausible proof or refutation strategies. We view a mathematical mechanism as a structure or reasoning principle that connects the assumptions of a candidate problem to its target conclusion, such as an inequality, invariant, decomposition, or reduction to an intermediate claim. We present MECA (MEchanism-centered Conjecture Agent), a multi-agent framework that constructs conjectures by jointly developing candidate statements and their supporting mechanisms. Explorer agents propose mechanisms, test how they apply, and revise the candidate conjecture accordingly, while critic agents assess their mathematical validity and research value. Their feedback guides changes to the assumptions, scope, and conclusion. Through this process, MECA transforms broad research directions into precise conjectures with substantive mathematical support while retaining a clearly identified unresolved core. We evaluate MECA in two complementary settings. First, we compare it with a generate-and-revise baseline on reconstructing preselected target-paper conclusions from target-conditioned but article-blind source materials. Second, we construct 100 semi-open problems from literature-derived seeds and existing open problems and evaluate them through independent proof and refutation attempts by automated provers. Our results indicate that mechanism-centered refinement produces well-specified and research-worthy conjectures that remain challenging for current automated provers.

cs.AI↗
arXiv2026

Retraction-Free Optimization over the Stiefel Manifold for the LoRA Fine-Tuning

Optimization over the Stiefel manifold plays a significant role in various machine learning tasks. Existing methods either use the retraction operators, requiring costly orthonormalization for large-scale matrices, or employ landing methods that rely on careful step size selection and penalty parameter tuning. To address these challenges, we propose a retraction-free and penalty parameter-free algorithm that directly lands on the manifold. By leveraging the strongly-convex-like property of the quadratic penalty function and the proximal smoothness of the Stiefel manifold, we establish global convergence guarantees with the best-known iteration complexities under both constant and diminishing step sizes. Then, we reformulate the low-rank adaptation (LoRA) fine-tuning problem for large language models as a manifold optimization problem, introducing Manifold-LoRA for geometry-accelerated adaptation. This approach employs the proposed landing technique and a carefully designed step size strategy to accelerate the training process. Numerical experiments on benchmark datasets demonstrate the efficiency and strong downstream performance of the proposed method.

cs.LG↗
arXiv2026

A Unified Framework for Formalizing Matrix Decomposition Proofs

Existence proofs for many matrix decompositions share a recursive routine: a local transformation prepares the matrix, a slice is selected, a recursive solution is obtained, and the result is lifted and transported back. Formalizing this routine uniformly in dependent type theory is difficult because recursive subproblems may change index types, and reconstruction must preserve structural predicates across block embeddings and reindexings. We develop a Lean~4 framework that separates decomposition schemas, transformations, reduction strategies, measures, lifting, transport, and subtype induction. The framework uses general index types, packages square and rectangular matrices in universe types, and provides a decomposition driver that assembles strategy data into subtype-induction instances. It has been instantiated across PLU, LU, LDL/Cholesky, QR variants, Gauss rank normal form, Hessenberg reductions, Schur variants, normal spectral decomposition, SVD, bidiagonalization, tridiagonalization, UTV, Smith normal form, rational canonical form, and Jordan-type forms at varying levels of statement strength. Across these instances, repeated decomposition proofs are best treated not as separate tasks but as instances of a more general inductive statement whose interface records a certified proof path compatible with the chosen decomposition statement.

math.NA↗