Research

My research connects reliable optimization algorithms, optimal transport and experimental design, and learning for intelligent systems. Across these directions, I exploit mathematical structure to make demanding computations efficient and trustworthy.

Reliable & Scalable Optimization

Topics ↑

Bridge the gap between theory and practice. I design optimization algorithms that remain reliable under inexact computation and scale with problem size. Convergence theory, Newton and proximal methods, low-rank structure, and parallel or mixed-precision computation connect mathematical guarantees with practical performance.

  • Mixed-precision GPU acceleration for large-scale minimum enclosing ball problems Preprint

    Ling Liang, Lei Yang

    2026. arXiv.

    GPU Computing Mixed Precision

    Low-precision screening and high-precision refinement reduce GPU work while retaining an accurate, fully checked enclosing ball.

    Runtime in seconds on a logarithmic scale for mp-ripALM, ripALM, CGAL, SCS, and CLARABEL, with 1,000 to 100,000 balls in dimension 50.
    Runtime at d = 50 for 1,000–100,000 balls. ripALM and mp-ripALM run on GPU; CGAL, SCS, and CLARABEL run on CPU.
  • Fast and effective computation of generalized symmetric matrix factorization Preprint

    Lei Yang, Han Wan, Min Zhang, Ling Liang

    2026. arXiv.

    Matrix Factorization Machine Learning

    Exactness results make generalized symmetric factorization amenable to the convergent A-NAUM alternating method.

    Two exactness results
    LayerWhat becomes exact
    ModelA finite quadratic penalty enforces symmetry.
    RelaxationStationary points are linked through an auxiliary-variable formulation.
    Both statements require the conditions established in the paper.
  • D-ripALM: A tuning-friendly decentralized relative-type inexact proximal augmented Lagrangian method Preprint

    Jiayi Zhu, Hong Wang, Ling Liang, Lei Yang

    2026. arXiv.

    Decentralized Optimization

    Relative-error control makes decentralized ALM easier to tune, with convergence analysis covering nonsmooth, generally convex objectives.

    A decentralized network of local objectives with inexact solves and a relative-error check.
    A simplified view of the inner/outer switching rule and neighbor communication.
  • Convergence of a relative-type inexact proximal ALM for convex nonlinear programming Preprint

    Lei Yang, Jiayi Zhu, Ling Liang, Kim-Chuan Toh

    2025. arXiv.

    Inexact Optimization

    A unified analysis explains the global behavior and asymptotic rates of relative-type inexact preconditioned proximal ALM.

    Convergence at three levels
    ObjectGuarantee
    IteratesGlobal convergence
    Local behaviorAsymptotic linear or superlinear rates
    Ergodic averagesObjective-residual and feasibility bounds
    Each guarantee is subject to its stated assumptions; local rate results require additional conditions.
  • Nesterov's accelerated Jacobi-type methods for large-scale symmetric positive semidefinite linear systems

    Ling Liang, Qiyuan Pang, Kim-Chuan Toh, and Haizhao Yang

    SIAM Journal on Scientific Computing 47, no. 6 (2025). SISC, arXiv, code.

    Linear Systems Parallel Computing

    Nesterov acceleration improves Jacobi-type convergence while preserving the simple, parallel structure of the updates.

    Parallel row-block updates feeding an accelerated Jacobi step, with an O(1/k squared) objective rate.
    The rate is for the associated convex quadratic objective, under the method’s assumptions.
  • An inexact Halpern iteration with application to distributionally robust optimization

    Ling Liang, Zusen Xu, Kim-Chuan Toh, and Jia-Jie Zhu

    Journal of Optimization Theory and Applications 260, no. 58 (2025): 1-41. JOTA, arXiv, code.

    Distributionally Robust Optimization

    Inexact Halpern updates retain O(1/k) residual rates—expected residuals in the stochastic setting—and enable practical Wasserstein DRO.

    Blue empirical samples and orange worst-case perturbed samples, connected by trajectories from inexact stochastic Halpern updates.
    Empirical samples (blue) move toward worst-case perturbations (orange) in the nonlinear DRO example.
  • A squared smoothing Newton method for semidefinite programming

    Ling Liang, Defeng Sun, and Kim-Chuan Toh

    Mathematics of Operations Research 50, no. 4 (2025): 2433-3282. MOOR, arXiv.

    Semidefinite Optimization

    Huber smoothing preserves zero eigenvalue contributions, allowing a squared smoothing Newton method to exploit structure in SDP.

    The positive-part function and Huber approximations for epsilon 0.25 and 0.75, all zero on negative inputs.
    Scalar Huber smoothing preserves the zero region and smooths the PSD projection when applied to eigenvalues.
  • ripALM: A relative-type inexact proximal augmented Lagrangian method with applications to quadratically regularized optimal transport Preprint

    Jiayi Zhu, Ling Liang, Lei Yang, and Kim-Chuan Toh

    2024. arXiv.

    Optimal Transport

    One relative tolerance replaces delicate error-sequence tuning while retaining convergence guarantees for inexact proximal ALM.

    Practical relative-error control
    ComponentripALM
    Inner toleranceOne relative parameter in [0, 1)
    Tolerance sequenceNo prescribed summable sequence
    Correction stepNot required
    “One parameter” refers to the inner error criterion, not all algorithm settings.
  • Accelerating nuclear-norm regularized low-rank matrix optimization through Burer-Monteiro decomposition

    Ching-pei Lee, Ling Liang, Tianyun Tang, and Kim-Chuan Toh

    Journal of Machine Learning Research 25, no. 379 (2024): 1-52. JMLR, arXiv, code.

    Low-Rank Optimization Machine Learning

    Cheap factorized steps gain the safeguards of convex optimization, while the algorithm adjusts the factorization rank automatically.

    Low-rank Burer–Monteiro steps alternate with aperiodic proximal safeguards and rank adaptation.
    The schematic separates the fast factorized phase from the global-convergence safeguard.
  • QPPAL: A two-phase proximal augmented Lagrangian method for high dimensional convex quadratic programming problems

    Ling Liang, Xudong Li, Defeng Sun, and Kim-Chuan Toh

    ACM Transactions on Mathematical Software 48, no. 3 (2022): 1-27. TOMS, arXiv, code.

    Quadratic Programming

    A two-phase solver pairs inexpensive warm-start iterations with accurate proximal ALM refinement for high-dimensional convex QP.

    QPPAL phase I uses symmetric Gauss–Seidel updates and phase II refines the solution with proximal ALM.
    Phase I targets low-to-medium accuracy; Phase II efficiently refines the warm start.
  • On degenerate doubly nonnegative projection problems

    Ying Cui, Ling Liang, Defeng Sun, and Kim-Chuan Toh

    Mathematics of Operations Research 47, no. 3 (2022): 2219-2239. MOOR, arXiv.

    Conic Optimization

    Augmented Lagrangian subproblems overcome the singular Newton systems that make degenerate doubly nonnegative projection difficult.

    Handling degenerate DNN projection
    ApproachNewton equations
    Direct KKT solveThe Jacobian can be singular under degeneracy.
    Augmented LagrangianA sequence of better-conditioned nonsmooth equations.
    The DNN cone combines positive semidefiniteness with entrywise nonnegativity.
  • A new homotopy proximal variable-metric framework for composite convex minimization

    Quoc Tran-Dinh, Ling Liang, and Kim-Chuan Toh

    Mathematics of Operations Research 47, no. 1 (2022): 508-539. MOOR, arXiv.

    Statistical Estimation

    A homotopy parameterization and primal-dual-primal construction preserve useful regularizer structure in proximal Newton subproblems.

    A primal problem is dualized, treated with homotopy proximal Newton, and its subproblem is solved back in primal space.
    In the covariance-estimation specialization, the method avoids matrix inversion and Cholesky factorization.
  • An inexact augmented Lagrangian method for second-order cone programming with applications

    Ling Liang, Defeng Sun, and Kim-Chuan Toh

    SIAM Journal on Optimization 31, no. 3 (2021): 1748-1773. SIOPT, arXiv.

    Conic Optimization

    Quadratic-growth conditions connect SOCP geometry to fast ALM residual convergence and an effective solver for structured cone problems.

    A second-order cone beside the implication from quadratic growth to R-superlinear ALM residual convergence.
    The convergence implication holds under the regularity conditions established in the paper.

Optimal Transport & Experimental Design

Topics ↑

Compare distributions and choose informative experiments. I develop efficient methods for optimal transport and experimental design, from Gromov–Wasserstein matching and capacity constraints to D-optimal and Bayesian design. Exploiting geometry, sparsity, and problem structure makes these models tractable at larger scales.

Learning & Autonomous Systems

Topics ↑

Connect optimization with learning and autonomous decision-making. I combine optimization and learning to automate mathematical modeling, tune algorithms, and uncover symbolic PDE structure. Algorithms for reinforcement learning and certifiable robot motion planning extend this work toward efficient learning and reliable decisions.

  • OptimAI: Optimization from natural language using LLM-powered AI agents

    Raghav Thind, Youran Sun, Ling Liang, Haizhao Yang

    Journal of Machine Learning, accepted, 2026. JML, arXiv.

    LLM Agents Automated Modeling

    Four cooperating LLM roles formulate, plan, implement, and debug optimization problems expressed in natural language.

    The OptimAI formulator, planner, coder, and code critic, including a feedback loop for code refinement.
    A simplified architecture: structured roles and iterative debugging turn descriptions into executable solutions.
  • Accelerating multi-block constrained optimization by learning penalty parameters

    Ling Liang, Cameron Austin, and Haizhao Yang

    Communications on Pure and Applied Analysis, accepted, 2026. CPAA, arXiv.

    Learning to Optimize Constrained Optimization

    Learned penalty parameters accelerate multi-block MPALM, reducing the need for manual tuning on Lasso and optimal transport problems.

    Log-normalized MSE versus iteration for learned-penalty LMPALM and baselines on a Lasso problem with m=10 and n=20.
    Lasso example, m = 10 and n = 20: log-normalized MSE versus iterations. LMPALM uses learned penalties.
  • Fast and certifiable trajectory optimization

    Shucheng Kang, Xiaoyang Xu, Jay Sarva, Ling Liang, and Heng Yang

    in Algorithmic Foundations of Robotics XVI, Volume 1: Proceedings of the Sixteenth Workshop on the Algorithmic Foundations of Robotics, Springer Proceedings in Advanced Robotics, vol. 37, pp. 43-65, Springer, 2026. Springer, arXiv, code, Best Paper Award Finalist.

    Robotics Semidefinite Optimization

    Sparse SDP relaxations and GPU computation produce practical robot trajectories with computable global-optimality certificates.

    Three optimized car back-in trajectories with successive car poses colored along each path.
    Car back-in trajectories. Across the paper’s five benchmarks, certified suboptimality is below 1%.
  • From equations to insights: Unraveling symbolic structures in PDEs with LLMs Preprint

    Rohan Bhatnagar, Ling Liang, Krish Patel, Haizhao Yang

    2025. arXiv.

    Scientific Discovery Symbolic PDEs

    LLMs predict operators in PDE solutions, narrowing symbolic search and improving interpretable analytical approximations.

    Symbolic PDE and boundary information feeds an LLM that predicts operators for a downstream symbolic solver.
    A simplified view of operator prediction guiding symbolic solution construction.
  • On the stochastic (variance-reduced) proximal gradient method for regularized expected reward optimization

    Ling Liang, Haizhao Yang

    Transactions on Machine Learning Research, 2024. TMLR, arXiv.

    Reinforcement Learning

    Importance-sampling-based variance reduction improves sample complexity for general regularized expected-reward optimization.

    Samples to reach ε-stationarity
    MethodSample complexity
    Stochastic proximal gradientO(ε⁻⁴)
    Variance-reduced proximal gradient with PAGEO(ε⁻³)
    Bounds hold under the respective assumptions; the improved rate uses additional conditions.
  • An inexact projected gradient method with rounding and lifting by nonlinear programming for solving rank-one semidefinite relaxation of polynomial optimization

    Heng Yang, Ling Liang, Luca Carlone, and Kim-Chuan Toh

    Mathematical Programming 201, no. 1-2 (2023): 409-472. MP, arXiv, code.

    Robotics Semidefinite Optimization

    Safeguarded low-rank local search accelerates a globally convergent SDP method, combining fast nonconvex steps with convex guarantees.

    A polynomial objective and its SDP relaxation connected by rounding and lifting, with projected descent steps.
    The paper’s univariate example explains how rounding, local search, lifting, and SDP descent work together.

Papers are grouped by their main research direction; tags highlight related methods and applications. Preprint identifies work not yet published or accepted.