Searcharxiv⌕ Search

arXiv subjects

Yeoneung Kim

Publications and source records attributed to Yeoneung Kim.

At least 19 recordsLinked to original sources

Newton's Method and Policy Iterations for Viscous Hamilton--Jacobi Equations

We study Newton's method for the Cauchy problem for viscous Hamilton--Jacobi equations with Hamiltonians that need not be convex. Each Newton step solves a linear parabolic equation obtained by linearizing the equation at the current iterate. We prove that the iterates converge geometrically in \(L^\infty\) from any initial guess satisfying the initial condition, and quadratically in the parabolic Hölder space \(C^{2+α,1+α/2}\) from initial guesses close to the solution. When the Hamiltonian admits a control representation, the Newton iterates coincide with those of classical policy iteration, which therefore inherits both convergence rates. We extend the local quadratic convergence to uniformly parabolic fully nonlinear equations. For small viscosity, we introduce a Nash--Moser variant that smooths the iterate and the residual in space, and we give a convergence criterion with constants independent of the viscosity. Finally, we show how computable residuals certify the error of inexact Newton corrections computed with physics-informed neural networks.

math.NA↗

Soft-confinement approximation of no-flux McKean-Vlasov dynamics

We study the approximation of the no-flux McKean-Vlasov equation on a bounded convex domain by a whole-space equation with quadratic confinement of strength $λ^{-1}$ outside the domain. For a smooth bounded interaction kernel and a common bounded initial density with square-integrable gradient, we establish an $O(λ^{1/4})$ approximation rate. The estimate controls the $L^2$ density error uniformly on every finite time interval and the gradient error in space-time $L^2$. Without interaction, a spectral gap independent of $λ$ yields the $O(λ^{1/4})$ rate uniformly in time for the density error and uniformly over all unit time intervals for the gradient error. With interaction, we construct a counterexample showing that even weak convergence can fail to be uniform in time.

math.AP↗

Continuous-Time Wasserstein Distributionally Robust Control: Limits, Duality, and Stability

A sampled robust control problem averages future values over disturbance laws, while its continuous-time counterpart evolves with an averaged drift. We establish a fluid limit connecting these two models for Wasserstein ambiguity sets. Under a uniform disturbance moment bound and subcritical growth, sampled Bellman values converge locally uniformly to the upper and lower values of a discounted differential game. The argument accommodates unbounded disturbances without requiring square-integrable increments. We also derive an explicit bound on the change in either game value when both the radius and center of the ambiguity set vary. For Euclidean disturbance space, uniform Lipschitz continuity in the disturbance, common state and cost Lipschitz bounds, and a discount rate exceeding the drift's state Lipschitz constant, this estimate converts distributional confidence bounds into uniform value approximation bounds. Static Wasserstein duality reformulates pointwise distributional optimization; affine disturbance dependence permits an exact reduction to a ball of means. Experiments illustrate the common limit and decreasing value error with sample size. For a nonlinear disturbance model, uniform shifts overestimate worst-case recovery speed by about 36\%. In a two-product inventory model, increasing ambiguity raises production near balance while reducing the region of full production in deeper backlogs.

math.OC↗

Convergence of alternating score-system iterations for regularized Landau particle dynamics

We analyze an alternating iteration for score-based particle approximations of the homogeneous Landau equation. Each step computes a regularized score along the previous trajectory and recomputes the trajectory with that score frozen. First, at fixed positive regularization, we prove existence, uniqueness, and $W_1$-Lipschitz stability of a Tikhonov score minimizer in a vector-valued reproducing kernel Hilbert space embedded in $C_b^2$. For bounded Lipschitz collision matrices, this yields factorial convergence to the coupled regularized dynamics on every finite interval, with constants independent of particle number. Global objective gaps and system residuals control inexact solves, and structural kernel assumptions ensure momentum and energy conservation. The mean-field and alternating-iteration limits commute at fixed regularization in this regime. Second, score-difference cancellation extends the theory to far-field-truncated Landau kernels with $γ\geq -2$ and the uncut case $γ=-2$. For the uncut three-dimensional Coulomb kernel, a separate continuum construction gives local factorial convergence from bounded, compactly supported densities. Third, we characterize the obstruction to removing regularization at fixed particle number and establish a separate conditional density-level convergence criterion under uniform regularity and approximation assumptions, using relative entropy. A heat-equation counterexample shows that an unregularized density-level Picard iteration can diverge. Atomic Coulomb and simultaneous particle-zero-regularization limits remain open.

math.NA↗

Moreau-Yosida Penalization of McKean-Vlasov Dynamics: Full-Flux Limits and Boundary Reaction Measures

We study McKean-Vlasov Fokker-Planck dynamics on a compact convex set $K\subset\mathbb{R}^d$ through the whole-space Moreau-Yosida penalization $U_λ=\operatorname{dist}(\cdot,K)^2/(2λ)$, keeping the external and interaction potentials fixed. At finite times, the penalized densities converge to the unique variational solution of the reflected no-flux equation, strongly in space-time $L^2$ on $K$ and in whole-space $L^1$, and narrowly, uniformly in time. We prove that the singular penalty-force measures converge weakly-$*$ to the purely normal boundary reaction $\operatorname{Tr}(ρ)n\,\mathcal{H}^{d-1}|_{\partial K}\,\mathrm{d}t$. This measure cancels exactly the surface singularity in the distributional gradient of the zero-extended density, explaining the emergent no-flux condition at the flux level. At equilibrium, we establish sharp $\sqrtλ$ rates in $L^1$ and Wasserstein distances for kernel-free Gibbs states, and in $L^1$ and quadratic Wasserstein distance for interacting fixed points under Dobrushin smallness or positive semidefiniteness, including maximum mean discrepancy interactions. The proof combines a tilted Gaussian supersolution, nonautonomous weighted energy estimates, and normal factorization with a Green-formula identification. We also identify the limit with a nonlinear reflecting diffusion, obtain path-space convergence and $λ$-uniform propagation of chaos through monotone synchronous coupling, and derive all-time Gibbs domination in the reversible case. Numerical experiments illustrate the boundary reaction and stationary scalings on a constrained maximum mean discrepancy flow.

math.AP↗

Time-optimal vaccination in state-dependent SIR models: a single switch criterion and numerical examples outside its regime

In a susceptible, infected, and recovered (SIR) model with constant rates, a time-optimal vaccination policy has at most one switch from no vaccination to maximal vaccination. We allow the transmission and recovery rates $β(S,I)$ and $γ(S,I)$ to depend on the state. If $β+I\partial_Iβ>0$ and $G:=\partial_S(βS-γ)>0$ throughout the forward invariant state space, then every optimal control is bang-bang with at most one switch. The incidence condition allows the transmission coefficient to decrease with prevalence. The Pontryagin analysis shows that $G$, the sensitivity of per capita infected growth to the susceptible fraction, and the infected costate together determine the crossing direction of the switching function. We also prove existence of an optimal control, derive explicit a priori upper bounds on the eradication time and characterize the switching time by an optimization over one variable. For parameter sets that violate $G>0$, numerical optimization gives controls with two switches and smaller eradication times than the computed minima over the single switch family. These computations provide a reproducible numerical counterexample to an unrestricted single switch claim. In a prescribed parameter sweep satisfying $G>0$, delayed vaccination reduces the eradication time relative to immediate full vaccination by more than $1\%$ in $10.9\%$ of the cases, with a maximum reduction of $26.3\%$. In examples near this maximum, the delay increases peak prevalence by more than an order of magnitude and cumulative incidence more than fourfold. Hence minimizing eradication time does not in general minimize epidemiological burden.

math.OC↗

Unbalanced Optimal Transport for Efficient Visual Document Retrieval

Vision-Language Models(VLMs)have advanced Visual Document Retrieval (VDR) by enabling OCR-free, multi-vector representations. However, these models generate excessive tokens to preserve dense visual details, which severely limits retrieval scalability. Existing compression methods mitigate this overhead but rely on isolated heuristics that fail to balance token importance, semantic redundancy, and coverage capacity. We propose a visual token compression framework formulated as an Unbalanced Optimal Transport (UOT) problem. Our approach unifies these three objectives into a single alternating optimization process. Furthermore, we theoretically show that minimizing the transport cost upper-bounds the MaxSim retrieval error, ensuring ranking stability. Extensive experiments on the ViDoRe V1 and V3 benchmarks demonstrate that our method preserves fine-grained document semantics and improves the efficiency-effectiveness trade-off over state-of-the-art base-lines. Code is available at https://github.com/shhhhhyy/Unbalanced-Optimal-Transport-for-EVDR.

cs.CV↗

Policy Iteration for Stationary Discounted Hamilton--Jacobi--Bellman Equations: A Viscosity Approach

We study policy iteration (PI) for deterministic infinite-horizon discounted control problems characterized by stationary Hamilton--Jacobi--Bellman equations. For general viscosity solutions, the classical gradient-based policy improvement step need not be defined pointwise. We introduce a semi-discrete formulation with centered difference quotients at scale $h$ and a separate artificial-viscosity term of order $O(h)$. The resulting stencil is monotone, and the positive discount yields a resolvent contraction. Under bounded Lipschitz data and a globally Lipschitz minimizing policy map, we prove monotone and geometric convergence of the value iterates for each fixed $h>0$, together with a local quadratic estimate whose constant is of order $h^{-2}$. Under the additional condition $λ>\Lip_x(f)$, we establish $\|V^h-V\|_\infty\le C\sqrt h$ and combine the discretization and iteration errors into a quantitative bound. A bounded Lipschitz example shows that the $\sqrt h$ exponent is sharp for this scheme. The combined estimate gives a sufficient iteration count of order $h^{-1}\log(1/h)$ to attain an error of order $\sqrt h$. In bounded-domain experiments, the smooth one-dimensional benchmark exhibits the predicted discretization plateau, while a nonlinear two-dimensional manufactured benchmark isolates convergence to the discrete solution. Exact policy evaluation gives substantially faster local convergence than the global geometric bound. A neural evaluation diagnostic illustrates the importance of controlling boundary errors as well as interior residuals.

math.OC↗

Penalty-Uniform Localization for State-Constrained Policy Iteration

Penalizing a state constraint creates a singular localization problem: whole-space values may grow like the inverse penalty parameter, while numerical diffusion allows even an inward feedback to cross the boundary. For deterministic discounted optimal control, we show how the penalty itself supplies confinement that offsets this growth. With mesh size $h$ and penalty parameter $\varepsilon$, an inward barrier bounds the additional cost of numerical leakage by $O(h/\varepsilon)$ for a monotone centered-difference scheme with vanishing viscosity. Under $h\le\varepsilon$, an occupation estimate then yields $O(h)$ localization error with a sufficient box margin logarithmic in $1/h$ and independent of $\varepsilon$. We extend the result to bounded squared-distance penalties and combine it with residual-based evaluation bounds and discounted policy-error propagation. Under coupled refinement with vanishing penalization and discretization errors, explicit conditions on evaluation errors and greedy gaps ensure convergence of the neural value approximations to the constrained value, allowing measurable, nonunique greedy selectors. Reference calculations isolate leakage and localization across mesh and penalty scales, and distinguish evaluation, iteration, and approximation errors. An explicit cylindrical state-constraint solution provides a benchmark in arbitrary dimension, tested up to dimension twenty. Paired obstacle-navigation experiments illustrate why policy-value accuracy must be assessed alongside sampled residuals when comparing raw-residual and finite-grid-assisted neural evaluation.

math.OC↗

Monotone Neural Policy Iteration for High-Dimensional First-Order Hamilton--Jacobi--Bellman Equations

We analyze a neural semi-discrete method for high-dimensional first-order Hamilton-Jacobi-Bellman (HJB) equations with known or learned dynamics. Centered differences and an artificial viscosity $Nh=O(h)$ define a monotone operator evaluated through $2d+1$ shifted network queries; policy iteration solves the resulting Bellman equation without a tensor grid. At fixed $h$, the sharp componentwise condition $\max_i|f_i|\le2N$ turns every frozen-policy operator into a nearest-neighbor Markov-chain generator with a policy-independent total jump rate. Uniformization gives whole-space well-posedness for measurable feedbacks, an explicit Poisson-tail bound on the numerical domain of dependence, and boundary-free localization. The representation also yields a posteriori policy-evaluation bounds that account for residual and learned-model errors. A greedy-gap analysis controls inexact policy iteration at fixed $h$; a separate consistency estimate connects the semi-discrete equation to the continuous HJB equation. Experiments reproduce the extremal tail, show rates consistent with $O(\sqrt h)$ and nearly $h$-independent exact-policy-iteration decay, and assess empirical estimator effectivity. A nonsmooth example shows that the continuous residual can miss a non-viscosity solution, whereas the shifted residual detects the defect. Further tests provide a structured interval-verified certificate calibration, an early-budget benefit of policy freezing for bang-bang control, and learned-dynamics diagnostics. A structured nonlinear problem with active compact-control constraints is tested against a manufactured semi-discrete reference through $d=1024$.

cs.LG↗

Self-Consistent Adjoint Policy Iteration for Constrained Dynamic Portfolio Choice

We develop simulation-based policy iteration for continuous-time portfolio choice with predictable returns and convex constraints. Each outer step re-evaluates a fixed-latent open-loop backpropagation-through-time (OL-BPTT) adjoint after deployment and solves the constrained update. Shifted-adjoint cancellation controls the adjoint--HJB Hamiltonian-gradient discrepancy by the policy-improvement residual. For CRRA portfolios, exact HJB policy iteration identifies the optimal reduced value factor, while population OL-BPTT iteration converges globally when the adjoint update is directionally improving and approximate stationarity is asymptotically HJB-compatible. A theorem-matched occupation audit yields maximal $95\%$ upper endpoints of $0.066$ for the primitive directional ratio and $0.074$ for a stronger norm-relative ratio, both against the half-step threshold $0.75$. In the high-precision $50$--$50$ occupancy/broad-anchor design of a three-factor, fifty-asset benchmark, current-policy re-evaluation outperforms matched pooled refinement under the on-policy and broad evaluation laws.

math.OC↗

Solving nonconvex Hamilton--Jacobi--Isaacs equations with PINN-based policy iteration

We propose a mesh-free policy iteration framework that combines classical dynamic programming with physics-informed neural networks (PINNs) to solve high-dimensional, nonconvex Hamilton--Jacobi--Isaacs (HJI) equations arising in stochastic differential games and robust control. The method alternates between solving linear second-order PDEs under fixed feedback policies and updating the controls via pointwise minimax optimization using automatic differentiation. Under standard Lipschitz and uniform ellipticity assumptions, we prove that the value function iterates converge locally uniformly to the unique viscosity solution of the HJI equation. The analysis establishes equi-Lipschitz regularity of the iterates, enabling provable stability and convergence without requiring convexity of the Hamiltonian. Numerical experiments demonstrate the accuracy and scalability of the method. In a two-dimensional stochastic path-planning game with a moving obstacle, our method matches finite-difference benchmarks with relative $L^2$-errors below %10^{-2}%. In five- and ten-dimensional publisher-subscriber differential games with anisotropic noise, the proposed approach consistently outperforms direct PINN solvers, yielding smoother value functions and lower residuals. Our results suggest that integrating PINNs with policy iteration is a practical and theoretically grounded method for solving high-dimensional, nonconvex HJI equations, with potential applications in robotics, finance, and multi-agent reinforcement learning.

math.NA↗

Physics-Informed Policy Iteration for High-Dimensional Hamilton--Jacobi--Bellman Equations: Interior Error Bounds without Boundary Data

We develop a physics-informed policy-iteration method for stationary second-order Hamilton--Jacobi--Bellman equations arising in continuous-time stochastic control. Each policy-evaluation step is a linear elliptic PDE and is approximated by a mesh-free neural residual solver; policy improvement is then performed pointwise from the surrogate gradient. The analysis addresses bounded-domain training without prescribed boundary data. We prove well-posedness of the PDE-defined evaluation for every Borel Markov policy, establish Lipschitz stability of the greedy map on bounded gradient ranges, and derive an exponential attenuation estimate for unresolved boundary information. These ingredients yield a closed finite-step interior error bound whose floors are determined by a continuous $L^p$ residual, with a finite exponent $p>d$, and an attenuated amplitude term. The experiments measure the quantities in the estimate. On a linear--quadratic testbed with an exact reference, the gradient-error floor scales nearly linearly with a fresh-sample $L^p$ residual estimate, and finite-difference probes identify when a training buffer is beneficial. At matched architecture and budget, linear fixed-policy training becomes markedly more reliable than direct minimization of the nonlinear HJB residual as the tested problems become more difficult. The method also produces effective feedback on an inverted pendulum, a planar quadrotor, and a 100-dimensional posterior-seeking problem. These nonlinear tests expose two limitations not visible from sampled residuals alone: the collocation distribution may miss the region visited by the learned closed loop, and evaluation without boundary data may leave undetermined the gradient component on which the greedy update depends. On-policy collocation, rollout-anchored evaluation, and rollout-based stopping provide effective model-only safeguards.

cs.LG↗

A Physics-Informed, Global-in-Time Neural Particle Method for the Spatially Homogeneous Landau Equation

We propose a physics-informed neural particle method (PINN--PM) for the spatially homogeneous Landau equation. The method adopts a Lagrangian interacting-particle formulation and jointly parameterizes the time-dependent score and the characteristic flow map with neural networks. Instead of advancing particles through explicit time stepping, the Landau dynamics is enforced via a continuous-time residual defined along particle trajectories. This design removes time-discretization error and yields a mesh-free solver that can be queried at arbitrary times without sequential integration. We establish a rigorous stability analysis in an $L^2_v$ framework. The deviation between learned and exact characteristics is controlled by three interpretable sources: (i) score approximation error, (ii) empirical particle approximation error, and (iii) the physics residual of the neural flow. This trajectory estimate propagates to density reconstruction, where we derive an $L^2_v$ error bound for kernel density estimators combining classical bias--variance terms with a trajectory-induced contribution. Using Hyvarinen's identity, we further relate the oracle score-matching gap to the $L^2_v$ score error and show that the empirical loss concentrates at the Monte Carlo rate, yielding computable a posteriori accuracy certificates. Numerical experiments on analytical benchmarks, including the two- and three-dimensional BKW solutions, as well as reference-free configurations, demonstrate stable transport, preservation of macroscopic invariants, and competitive or improved accuracy compared with time-stepping score-based particle and blob methods while using significantly fewer particles.

math.NA↗

Physics-informed approach for exploratory Hamilton--Jacobi--Bellman equations via policy iterations

We propose a mesh-free policy iteration framework based on physics-informed neural networks (PINNs) for solving entropy-regularized stochastic control problems. The method iteratively alternates between soft policy evaluation and improvement using automatic differentiation and neural approximation, without relying on spatial discretization. We present a detailed $L^2$ error analysis that decomposes the total approximation error into three sources: iteration error, policy network error, and PDE residual error. The proposed algorithm is validated with a range of challenging control tasks, including high-dimensional linear-quadratic regulation in 5D and 10D, as well as nonlinear systems such as pendulum and cartpole problems. Numerical results confirm the scalability, accuracy, and robustness of our approach across both linear and nonlinear benchmarks.

math.NA↗

Censored Sampling for Topology Design: Guiding Diffusion with Human Preferences

Recent advances in denoising diffusion models have enabled rapid generation of optimized structures for topology optimization. However, these models often rely on surrogate predictors to enforce physical constraints, which may fail to capture subtle yet critical design flaws such as floating components or boundary discontinuities that are obvious to human experts. In this work, we propose a novel human-in-the-loop diffusion framework that steers the generative process using a lightweight reward model trained on minimal human feedback. Inspired by preference alignment techniques in generative modeling, our method learns to suppress unrealistic outputs by modulating the reverse diffusion trajectory using gradients of human-aligned rewards. Specifically, we collect binary human evaluations of generated topologies and train classifiers to detect floating material and boundary violations. These reward models are then integrated into the sampling loop of a pre-trained diffusion generator, guiding it to produce designs that are not only structurally performant but also physically plausible and manufacturable. Our approach is modular and requires no retraining of the diffusion model. Preliminary results show substantial reductions in failure modes and improved design realism across diverse test conditions. This work bridges the gap between automated design generation and expert judgment, offering a scalable solution to trustworthy generative design.

cs.LG↗

A diffusion-based generative model for financial time series via geometric Brownian motion

We propose a novel diffusion-based generative framework for financial time series that incorporates geometric Brownian motion (GBM), the foundation of the Black--Scholes theory, into the forward noising process. Unlike standard score-based models that treat price trajectories as generic numerical sequences, our method injects noise proportionally to asset prices at each time step, reflecting the heteroskedasticity observed in financial time series. By accurately balancing the drift and diffusion terms, we show that the resulting log-price process reduces to a variance-exploding stochastic differential equation, aligning with the formulation in score-based generative models. The reverse-time generative process is trained via denoising score matching using a Transformer-based architecture adapted from the Conditional Score-based Diffusion Imputation (CSDI) framework. Empirical evaluations on historical stock data demonstrate that our model reproduces key stylized facts heavy-tailed return distributions, volatility clustering, and the leverage effect more realistically than conventional diffusion models.

cs.LG↗

Approximate Thompson Sampling for Learning Linear Quadratic Regulators with $O(\sqrt{T})$ Regret

We propose a novel Thompson sampling algorithm that learns linear quadratic regulators (LQR) with a Bayesian regret bound of $O(\sqrt{T})$. Our method leverages Langevin dynamics with a carefully designed preconditioner and incorporates a simple excitation mechanism. We show that the excitation signal drives the minimum eigenvalue of the preconditioner to grow over time, thereby accelerating the approximate posterior sampling process. Furthermore, we establish nontrivial concentration properties of the approximate posteriors generated by our algorithm. These properties enable us to bound the moments of the system state and attain an $O(\sqrt{T})$ regret bound without relying on the restrictive assumptions that are often used in the literature.

stat.ML↗