Papers

Embedding trained neural networks as surrogates within optimisation problems is an established practice in operations research. The prevailing approach uses feedforward neural networks (FNNs) with ReLU activations, whose piecewise-linear structure admits an exact but computationally intensive mixed-integer programming (MIP) reformulation as the networks grow. We advocate input convex neural networks (ICNNs) as structurally superior surrogates when the underlying response is approximately convex or concave. The convex architecture offers two computational advantages. First, the ICNN-MIP formulation tends to yield a tighter linear programming (LP) relaxation than its FNN-MIP counterpart, with no integrality gap in favourable instances. Second, ICNNs uniquely admit an LP-based reformulation via epigraph representations of ReLU activations, though this embedding is not always exact. When it is not, we exploit the properties of ICNNs to construct the strongest continuous relaxation over box domains, namely, the convex hull of the ICNN's graph, bounded below by the epigraph and above by the concave envelope; this construction is tractable under input convexity but hard for general ReLU networks. On this basis, we develop a branch-and-bound algorithm that builds this relaxation at each node, branches directly on input variables rather than intermediate variables as in MIP reformulations, and terminates at the root node whenever the epigraph embedding is valid. Case studies on humanitarian food aid, oil well routing, and wine blending show that ICNN surrogates match FNN accuracy and deliver gains in solve time and scalability, supporting ICNN as the default surrogate when the underlying function is convex, concave, or well-approximated as such.

We prove exact boundary controllability for a class of one-dimensional singular/degenerate wave equations of Sturm–Liouville type, \[ u_{tt}-(x^\alpha u_x)_x-\beta x^{\alpha-1}u_x-\mu x^{\alpha-2}u=0, \qquad x\in(0,1), \] with a Dirichlet condition at the regular endpoint and a boundary control acting at the singular endpoint \(x=0\). The analysis is carried out in the fractional energy space % \[ X=\mathcal H^{\nu+1/2}\times \mathcal H^{\nu-1/2}, \] associated with the corresponding singular Sturm–Liouville operator. Using the spectral decomposition induced by Bessel functions, we establish admissibility of the boundary observation operator and derive precise lower estimates for the observation coefficients. Exact observability is obtained through Ingham-type inequalities together with an abstract observability result for unitary groups. The proof treats simultaneously the subcritical, critical logarithmic, and limit-point regimes by means of singular Sturm–Liouville theory and boundary traces identified through the Lagrange bracket of the singular Sturm–Liouville expression.

We consider smooth convex minimization over the spectrahedron using Frank-Wolfe-type methods based only on extreme-eigenvector computations. In our recent work \cite{garber2026randomized} we presented the first ambient-dimension-independent linear convergence rate under quadratic growth. However, the method makes an additional strong strict complementarity assumption, it is randomized, its linear rate holds only after a burn-in phase and in expectation, and it requires the objective smoothness constant. We show that these limitations can be removed. Assuming quadratic growth and that all optimal solutions have the same rank, but without assuming strict complementarity, we give a deterministic and parameter-free Frank-Wolfe-type method with a global ambient-dimension-independent linear convergence rate.

Optimization theory is a widely used tool for intelligent decision-making. While classical optimization deals with fixed, time-invariant objective functions, many modern applications operate in dynamic environments where data arrive sequentially, and the learning objective evolves over time, often under decentralized data and communication constraints. Motivated by these trends, we study decentralized optimization from streaming data through a structured time-varying formulation in which the global objective is a temporally weighted average of losses observed across the network. We analyze multi-iteration decentralized first-order methods, including decentralized gradient descent. For strongly convex and smooth losses, we develop guarantees for the Euclidean-norm tracking error through a contraction-mapping viewpoint. The resulting bounds decompose the tracking error into a fixed-point tracking component and a bias term induced by decentralization and data heterogeneity. We specialize our analysis to uniform and exponentially discounted weights, as well as their finite-memory windowed counterparts. The bounds explicitly characterize the roles of the temporal weighting rule, per-step iteration budget, step size, and network connectivity. Uniform weighting yields a vanishing fixed-point tracking contribution of order $\mathcal O(1/t)$, whereas discounted and windowed strategies generally induce non-vanishing tracking floors governed by the discount factor and effective memory, respectively. In all cases, decentralization induces an additional non-zero bias floor under a constant step size. Numerical experiments illustrate the predicted trends.

We study constructive discrete multimarginal optimal transport (MOT) among $m$ distributions on $n$ points, where the cost tensor $C$ has $N=n^m$ entries. For additive accuracy $\varepsilon$, let $\kappa=\max\{1,(\max C-\min C)/\varepsilon\}$. Classically, we give two algorithms that return exactly feasible additive-$\varepsilon$ couplings. A deterministic box–simplex method with rounding runs in $O(m^2N\kappa\log N)=\widetilde O(m^3N\kappa)$ time, while an exact reduction to positive packing followed by rank-one completion runs in randomized time $\widetilde O(m^2N\kappa)$. At fixed $\kappa$, both match the $\Omega(N)$ cost of writing a dense coupling, up to factors in $m$ and logarithms. Quantumly, in the general entry-access model, tensor Sinkhorn followed by sparse recovery returns an exactly feasible additive-$\varepsilon$ coupling as a classical list of $\widetilde O(m^2n\kappa^2)$ atoms, using $\widetilde O(m^4\sqrt{Nn}\,\kappa^3)$ coherent cost queries without materializing the tensor. Finally, for fixed $m$ and constant normalized accuracy, we prove sparse-output lower bounds of $\widetilde\Omega(N)$ randomized classical queries, even with unrestricted output size, and $\widetilde\Omega(\sqrt{Nn})$ quantum queries for outputs with at most $n\operatorname{polylog}(n)$ atoms. Hence, for fixed $m$ and normalized accuracy between $(\log n)^{-O(1)}$ and a sufficiently small constant, explicit sparse MOT construction has query complexity $\widetilde\Theta(N)$ classically and $\widetilde\Theta(\sqrt{Nn})$ quantumly.

Let $T$ be an exponentially stable strongly continuous semigroup on a Hilbert space $X$, $A$ its generator, $X_1$ the domain of $A$ with the norm $\|v\|_1 := \|Av\|$, and $C$ an admissible observation operator taking values in a Hilbert space $Y$. Russell and Weiss conjectured that if $(A,C)$ satisfies the infinite-dimensional Hautus test, then $(A,C)$ is exactly observable. After a counterexample was found their conjecture was modified to include the additional assumption that $T$ is similar to a contraction semigroup. We disprove this modified conjecture by constructing a counterexample in which $A$ is normal, $X$ has an orthonormal basis of eigenvectors of $A$ and $C: X_1 \to Y$ is Hilbert-Schmidt. Since $A$ is normal and $T$ is exponentially stable, it follows that $T$ is a contraction semigroup. In contrast, we prove that the Hautus test implies exact observability for self-adjoint $A$. Finally, we show that if $A$ is normal and $C:X_1 \to Y$ is compact, then the Hautus test implies that $A$ has compact resolvent. Consequently, for normal $A$ and finite-dimensional $Y$ the Hautus test implies exact observability.

This note aims to serve as an entry point to the literature on learning in games, a topic with significant theoretical appeal and a wide range of applications – from machine learning and data science to economics and beyond. Our presentation is structured around two complementary viewpoints: We first consider a single agent – the learner – engaged in a sequential decision process in an unknown, non-stationary, and possibly adversarial environment. We then examine what happens when the environment is shaped by the decisions of several interacting agents, not necessarily aware of each other's actions or goals, and all seeking to improve their individual rewards. In this general context, we examine a family of regularized learning policies based on best-responding to the past history of play, up to a regularization penalty intended to encourage exploration and prevent over-commitment to suboptimal choices. In the single-agent setting, we present some basic regret bounds for regularized learning in adversarial multi-armed bandits; in the multi-agent setting, we describe an ergodic equilibrium convergence result for zero-sum games in the spirit of classical results on fictitious play, as well as a "folk theorem" linking strategic and dynamic notions of stability – Nash equilibria and attracting points of regularized learning, respectively. We pay special attention to the information available to the players and, through a unified analysis framework, we study both oracle- and payoff-based (bandit) methods. Our goal is to provide a coherent and comprehensible – albeit, by necessity, not comprehensive – account of some recent ideas in the field, and to discuss their implications for the study of rationality.

A target moves between two sites according to a discrete-time Markov chain with a $2\times2$ transition matrix $M$. At each epoch one site is searched at positive cost, and a search may overlook a target that is present. Ross conjectured that an optimal policy is threshold in the posterior probability that the target is at site~1. MacPhee and Jordan proved the conjecture throughout the nonpositive-determinant ($\det M\le0$) regime and for part of the positive-determinant ($\det M>0$) regime, leaving the remaining cases open. We prove threshold optimality throughout the positive-determinant regime, completing Ross's conjecture for all parameter values.

Optimal transport and Gromov–Wasserstein distances are useful tools for comparing probability measures and metric measure spaces, but their balanced formulations force all mass to be matched. This constraint is often too strong for data with outliers, missing parts, or only partial overlap. In this paper, we develop entropic partial optimal transport for Gaussian mixture models and define a partial mixture Gromov–Wasserstein distance. For the finite entropic partial optimal transport problem, we prove the existence and uniqueness of the minimizer and establish quantitative large-penalty estimates. Moreover, the resulting entropic partial component couplings induce continuous partial transport plans through Gaussian optimal maps. We analyze their large-penalty and subsequent zero-entropy limits and construct the associated displacement interpolations and barycentric projection maps. In addition, by identifying each Gaussian mixture with a finite metric measure space of Gaussian components, we establish the metric property and large-penalty limit of the partial mixture Gromov–Wasserstein distance. Finally, numerical experiments on synthetic Gaussian mixtures and point clouds illustrate the effects of the penalty and entropic regularization and the robustness of partial matching to outliers.

We prove that no randomized integral or fractional algorithm for online vertex cover under general vertex arrivals achieves a competitive ratio strictly below $1+\sqrt{e}/2\approx1.824360635$, even on bipartite graphs and against an oblivious adversary. This improves the previous lower bound of approximately $1.753$. Our proof extends the complete-bipartite alternating construction of Wang and Wong to an arbitrary number of alternations. The resulting adversary is described by a monotone integral recurrence. If the recurrence never violates the competitive budget, its iterates converge to an integrable fixed point; classifying all such fixed points forces the excess ratio to be at least $\sqrt{e}/2$. A truncated discrete recurrence and a Riemann-sum argument convert every strict continuous violation into a finite, algorithm-dependent but realization-oblivious input. We also exhibit a critical fixed point showing that $1+\sqrt{e}/2$ is the exact limit of this homogeneous complete-bipartite recurrence, rather than a numerical artifact.

We develop a primal–dual interior-point method for nonsymmetric conic optimization based on a conjugate-free scaling matrix. The scaling is obtained from a single-secant BFGS update of the primal barrier Hessian. In contrast to multi-secant BFGS scalings, it does not require conjugate-barrier derivatives. This feature is important for high-dimensional nonsymmetric cones, where conjugate-barrier derivatives may be unavailable in closed form or expensive to compute. We embed the conjugate-free scaling in a homogeneous self-dual predictor–corrector framework. Using a split central-path neighborhood that separately controls the conic variables and the scalar homogeneous variables, we prove that the scaling matrix remains uniformly comparable to the primal barrier Hessian. This comparison bound is used to prove neighborhood preservation and to show that the complementarity measure and the linear residual decrease at a uniform rate. Consequently, the method attains an iteration bound of $\mathcal{O}(\sqrt{\nu}\log(1/\varepsilon))$, improving the $\mathcal{O}(\nu\log(1/\varepsilon))$ bound of Badenbroek and Dahl [Optim. Methods Softw., 37 (2022), pp. 1027–1064] and matching the best-known complexity order for interior-point methods. Numerical experiments on instances involving the operator perspective epigraph cone and the quantum relative entropy cone show that the method is competitive with QICS, a specialized solver for conic models arising in quantum information.

Large-scale renewable power-to-hydrogen (ReP2H) plants in cold regions suffer from prolonged startup and repeated thermal stress during frequent startup-shutdown operation. The situation becomes worse due to the lack of coordinated heat management among the alkaline electrolysis stacks, balance of plant (BoP), plant thermal utility system (PTUS), and plant building. This paper presents a plant-wide thermal topology and a hierarchical electricity-heat management framework to address the issues. Bidirectional thermal coupling between the stack cluster and PTUS enables preheating, thermal standby, and waste heat recovery, while minute-scale production scheduling is coordinated with second-scale thermal regulation. Case studies based on an 80 MW plant in Northern China show that the proposed framework eliminates cold startups in year round, increases hydrogen yield by 1.50, improves energy and exergy efficiencies by 0.99 and 4.33 percentage points, respectively, and reduces the levelized cost of hydrogen by 3.22%. It also reduces thermal fatigue damage and startup-shutdown-induced voltage degradation.

We present PDHCG-CQP, a GPU-accelerated first-order solver for large-scale conic convex quadratic programming. PDHCG-CQP supports affine constraints and Cartesian products of nonnegative, second-order, rotated second-order, exponential, and three-dimensional power cones. At its core is a restarted averaged primal-dual hybrid gradient (PDHG) method, whose primal update is computed inexactly by solving a conic quadratic proximal subproblem with projected gradient iterations. We establish local linear convergence of the restarted averaged scheme with both exact and inexact primal proximal evaluations under a uniform local quadratic-growth condition on the smoothed primal-dual gap. We further show that this condition holds under strict complementarity by exploiting a rotated second-order-cone lifting together with local primal and dual regularity conditions. Our C/CUDA implementation combines matrix-free linear algebra, batched cone projections, adaptive inner solves, reflected-Halpern acceleration, and fully device-resident KKT residual computations. It also supports multi-GPU execution through a two-dimensional partitioning of the problem data. Extensive experiments on standard and large-scale quadratic programming (QP), convex quadratically constrained quadratic programming (QCQP), second-order cone programming (SOCP), and quasilinear Fisher equilibrium benchmarks demonstrate that PDHCG-CQP achieves state-of-the-art robustness among first-order solvers while scaling efficiently to 8 GPUs and instances with up to $4.4\times10^8$ stored primal coordinates. PDHCG-CQP is open source and available at https://github.com/Lhongpei/PDHCG.

We study decision-focused learning (DFL) in shortest-path network interdiction (SPNI) games, a Stackelberg game where an interdictor (leader) strengthens the networks' arcs against attacks, while an evader (follower) who is uncertain about costs of attacking network arcs relies on a machine-learned predictor to identify the shortest path. While DFL is highly effective as an end-to-end optimization framework, we show that it faces a fundamental structural failure when employed in this game setting: its training objective admits a broad decision-equivalence class of cost estimators that achieve zero nominal loss yet fail under interdiction, reversing DFL's usual advantage over a naive prediction-focused learning (PFL) approach. To address this, we propose Adversarial DFL (A-DFL), which replaces nominal training samples with interdicted scenarios to collapse the harmful equivalence class. Experiments on synthetic and real-world networks confirm that A-DFL restores DFL's advantage in this game setting, enabling effective end-to-end optimization.

What should a machine learning model learn when data is missing during training? We look at the learning process from a dynamical systems perspective, cast data missingness as a structured loss of actuation that limits controllability of the parameter error dynamics, and ultimately derive adaptation mechanisms with Lyapunov stability characteristics that throttle model updates in ways that preserve learning coherence under partial, intermittent observability. Under recurrent excitation, our analysis provides ISS-type residual-to-state bounds with respect to a bounded closed-loop mismatch between the loss residual and the preconditioned update geometry. We evaluate the efficacy of our directional observability-aware adaptive learning approach on multimodal contexts, reinforcing its premise in promoting learning coherence and stability even in pathologically sparse domains and problems.

In this paper, we develop provable parameter-free fixed-point algorithms for contractive mappings, with an emphasis on automatically exploiting hidden contractivity without requiring prior knowledge of the contraction factor. Our first method is a parameter-free variant of the Halpern fixed-point iteration with no line-search, bisection, or prior estimate of the contraction factor, while retaining essentially the same per-iteration computational cost as classical fixed-point schemes. We establish explicit linear convergence rates for both the fixed-point residual and the distance to the unique fixed point. The second algorithm is also a parameter-free Halpern method, but it requires a line-search procedure to find a local estimate for the contraction factor, which may need a few additional operator evaluations. This method also enjoys explicit linear convergence guarantees. We further extend these ideas in two directions. First, using the relation between Halpern iterations and Nesterov-type accelerated fixed-point schemes, we derive parameter-free Nesterov-type (three-term) variants that inherit linear convergence in the contractive setting. Second, by combining the proposed fixed-point schemes with Tikhonov regularization, we obtain new methods for solving cocoercive equations and establish an iteration complexity of $\mathcal{O}({\epsilon^{-1}\log(\epsilon^{-1})})$ for computing an $\epsilon$-solution, provided that a valid lower bound on the cocoercivity constant is available. Numerical experiments on several examples demonstrate that the proposed algorithms are competitive with, and often outperform, existing adaptive fixed-point methods. In particular, the methods successfully exploit contractive behavior when it is present on the selected nonexpansive test problems.

In this paper, we develop a systematic framework for deriving explicit error bounds for semidefinite feasibility problems without assuming strict feasibility (Slater's condition), a setting in which existing results are limited. Our main technical contribution is the introduction of radial-type Hölder error bounds, where the error bound constant depends explicitly on the norm of the reference matrix through radial modulus functions. By combining facial reduction with recently developed facial residual functions, we obtain explicit descriptions of these modulus functions, yielding qualitative radial-type Hölder error bounds without imposing any constraint qualifications. Our results complement the classical work of Sturm by providing explicit estimates for the constants involved in Sturm's local Hölder error bounds over bounded sets with a given size. We further analyze the asymptotic behavior of these bounds as the dimension of the underlying matrix space grows, identifying regimes in which they can be asymptotically tight up to a dimension-free constant. As an application, we establish explicit error bounds for the optimality system of semidefinite programs by reformulating them as feasibility problems, a setting where Slater's condition typically fails. Under the generically satisfied strict complementarity condition, we derive radial-type error bounds without assuming the usual solution uniqueness requirement, and demonstrate their asymptotic tightness through an explicit example.

We prove a sharp lower bound for smooth nonconvex stochastic optimization with uniformly bounded gradient noise. In the \(K=1\) fresh-sample model, every randomized adaptive algorithm requires $$\Omega\left( \frac{\Delta L}{\epsilon^2} + \frac{\Delta L\sigma^2}{\epsilon^4} \right)$$ queries to find a point with expected gradient norm at most \(\epsilon\). This matches the standard upper bound and, to the best of our knowledge, resolves the question raised by [Arjevani et al. 2023] of whether almost-surely bounded oracle error permits a better rate than bounded variance. The proof was independently generated with GPT-5.6 Sol in Codex's Ultra mode during a two-hour session. The human author supplied the prompt and was responsible only forchecking the proof and revising and polishing the manuscript.

Optimization under zeroth-order (i.e., bandit) feedback is central to many engineering problems where the analytic forms of objectives and/or constraints are unavailable. In modern applications, such as online control and online learning, optimization problems often evolve with time, requiring adaptive optimization methodologies. Yet, existing methods in this seting are largely confined to adaptations of methodologies developed for time-invariant or first-order optimization, and thus often rely on gradient surrogates that fail to fully exploit the zeroth-order structure of the available information. In this paper, we propose a randomized two-point direct-search algorithm for nonconvex time-varying optimization and derive iteration-complexity bounds under both constant and diminishing probing ratios. The resulting analysis yields explicit stationarity bounds in terms of the temporal variability of the problem and possible oracle errors. Our complexity bounds recover the complexity of existing zeroth-order methods in the time-invariant setting, while extending direct- search methods beyond static settings. As an illustrative application, we show that the methodology is naturally suited to solve optimal (equilibrium-selection) control problems for dynamical systems. In this setting, the analysis yields explicit stationarity bounds in terms of the temporal variability of the problem, measured through the effects of plant dynamics and exogenous disturbance variations.

We characterize the subsets of a finite distributive lattice that arise as the minimizer sets of modular functions. It is immediate that the minimizer set of any modular function is a decomposition-closed sublattice. Our main theorem establishes the converse: every decomposition-closed sublattice is the minimizer set of a modular function. Our proof is constructive. Using Birkhoff's representation, we decompose the problem along the intervals determined by an arbitrary maximal chain of the given sublattice. The key ingredient is a weight assignment on connected difference posets that yields local modular functions vanishing precisely at the endpoints of each interval. We also provide an example showing that this characterization fails for nondistributive lattices.

We provide an explicit mathematical characterization of regional characteristics and the resulting epidemiology to evaluate whether elimination or suppression is feasible using targeted measures, such as case isolation and contact tracing. We derive an epidemic model where community members can be infected by either travellers or other community members, and with constraints on the public health resources available to support isolation of arriving travellers and infected community members. We prove that the optimal controls are to immediately implement public health measures at their maximum levels. We find that elimination or suppression strategies are feasible with targeted measures if public health capacity is high, the transmission rate is low, or if the arrival rate of infected travellers is low. Our results illustrate that a particular country that implemented a mitigation strategy during the pandemic may not necessarily have achieved better outcomes if it had instead implemented elimination because elimination may not have been feasible in countries with low public health capacity to support control measures, high transmission rates, or high traveller arrival rates. Combinations of regional characteristics determine whether an elimination, suppression or mitigation strategy without resurgence is feasible. We find that regions with the low traveller arrival rates and the capacity to implement and enforce isolation or quarantine of arriving travellers are less likely to exceed their capacity to contact trace and isolate community members. Travel measures prevent only a small number of community infections via their direct effect, but indirectly, by protecting contact tracing capacity, travel measures can prevent epidemic resurgence and substantially reduce the number of people infected during a pandemic before a vaccine or therapy is developed.

Spacecraft rendezvous on elliptical orbits must hold a safety corridor under navigation noise, unmodelled perturbations, and thrust errors driven by propellant mass uncertainty. This paper develops a tube-based model predictive controller for the Yamanaka-Ankersen linear time-varying dynamics, carrying constraint-tightening tube methods from circular to elliptical orbits. A horizon-dependent, element-wise error bound propagates the actual closed-loop matrices along the prediction window, so early prediction steps keep nearly the full corridor that a constant-width tube would surrender. A multiplicative-to-additive conversion folds mass and thrust uncertainty into the same tightening recursion, and a Perron-Frobenius spectral-radius condition supplies a computable certificate that the tightening converges, with an explicit input-to-state stability gain. In a paired Monte Carlo campaign on a Mars sample-return orbit, the tube controller cuts mean corridor violations by more than an order of magnitude against a nominal predictive baseline at comparable fuel cost. A nonlinear truth-model test with oblateness perturbations well beyond the assumed disturbance budget leaves the corridor unviolated, and when the design envelope is exceeded the tightened problem becomes infeasible and the controller reverts to a saturated linear fallback, making the loss of guarantee explicit rather than silent. The construction requires only element-wise arithmetic and an open-source quadratic-programming solver.

Many robust planning problems are solved by checking a family of ordinary optimization problems, one for each uncertainty threshold. Repeatedly building and solving those relaxations can dominate runtime. We show that, when a decision chooses exactly one option from each group, the entire threshold family of multiple-choice knapsack relaxations can instead be bounded together. A cancellation removes threshold-specific baselines and reduces each group's contribution to two simple envelopes. After sorting, one multiplier is evaluated across all thresholds in time linear in their number and nearly linear in the number of options. The resulting interval bound is valid, matches the ordinary relaxation on a single feasible threshold, and is at least as strong as the matched group-clique comparator. A certified one-dimensional search controls multiplier error, while exact checks resolve numerically ambiguous feasibility and comparison cases. In an otherwise identical adaptive search, the method wins all 60 instance-level median comparisons and achieves a 2.37-fold geometric-mean speedup. Additional ablations, independent linear-program checks, and nine instances transformed from a published archive test the mechanism beyond the internal generator. A separate integer audit shows the boundary: faster relaxation bounds need not help when integer search dominates.

Many operational decisions require solving a linear program whose cost vector is unknown at decision time and must be predicted from contextual information. Because prediction and decision are only weakly aligned, the emerging integrated learning and optimization (ILO) paradigm trains the predictor through the downstream problem, judging a prediction by the decision it induces. However, predictions are inevitably imprecise, so robustness often enters the decision stage. To address this issue, we propose an integrated learning and robust optimization (ILRO) framework, where a robust decision problem is used both to define the training problem (termed the RSPO loss problem), and to produce the deployed decision. Thus, this framework simultaneously achieves both robustness and learning-decision alignment. To tackle its computational challenges, we develop a convex surrogate, RSPO+, and characterize when it is Fisher consistent. Moreover, the RSPO loss possesses informative gradients, allowing us to develop first-order computational methods. We also derive finite-sample excess risk bounds for both RSPO and RSPO+ predictors. Numerical experiments on transportation and portfolio problems, in comparison with multiple benchmarks, show the advantage in decision quality of the proposed framework. The gain is more pronounced for scenarios with limited samples, high-dimensional decisions, and model misspecification.

The single-selection prophet inequality is a canonical Bayesian online selection problem in which independent nonnegative values arrive sequentially and the decision-maker must irrevocably select at most one. Classical single-threshold guarantees are tight in the worst case, but the hard instances that prove tightness are highly irregular: the prophet's advantage is driven by rare, very large realizations of the maximum. We refine this worst-case picture by imposing a bound on the relative variance of the prophet's value, $\mathrm{Var}(\max_{i\in[n]}X_i)/\mathbb E[\max_{i\in[n]}X_i]^2$. This yields a nonparametric complexity measure that interpolates between deterministic instances, where the full prophet value can be recovered, and the unrestricted worst-case regime. Our main technical contribution is a general kernel method for single-threshold prophet inequalities. The method represents an instance by the quantile function of the maximum and rewrites the payoff of a threshold as a linear kernel functional of this quantile. This turns the worst-case analysis into an infinite-dimensional convex program, restores strong minimax duality in quantile space, and reduces the bounded-variance adversary's problem to a one-parameter variational family. Applying this framework, we obtain an exact characterization of the IID bounded-variance curve and asymptotically optimal finite-horizon thresholds, a closed-form expression for the fixed-order non-identical model, and a prophet-secretary lower-bound program together with a strict separation from the IID benchmark at every positive finite variance constraint. As a further application of the same kernel viewpoint, we derive an exact formula for IID random horizons under a convexity condition on the horizon pgf, which includes monotone-hazard-rate horizons, highlighting the broad applicability of this new technique for single threshold settings.

We study exact Kullback–Leibler (KL) projection for low-rank factorizations whose two nonnegative factors have prescribed row marginals and a shared, learned column marginal. For arbitrary positive row marginals of equal total mass, the joint KL projection reduces exactly to a strictly convex gauge-fixed dual with only $r-1$ effective variables; its Hessian is a sum of categorical covariance terms and admits $O((n+m)r)$ matrix-free Hessian–vector products. The projection theorem is objective-independent. We then specialize this geometry to doubly stochastic (DS) graph learning through $W=U\operatorname{Diag}(g)^{-1}V^\top$, where row-simplex factors with a common column mass induce an exactly DS graph without materializing an $n\times n$ optimization variable. Combined with observed-edge sparse fitting, a stochastic anchor-reduced manifold regularizer, and Bregman backtracking, the resulting mirror-descent method preserves exact feasibility at every accepted step. Under a nonvanishing latent-mass condition, it satisfies sufficient decrease and an $O(1/N)$ mirror-stationarity bound, while strictly positive accumulation points are KKT stationary. Matched clustering experiments show competitive accuracy, feasibility residuals near numerical precision, and favorable anytime behavior without a dense learned graph.

Data Envelopment Analysis (DEA) is a widely used non-parametric method for evaluating the relative efficiency of decision-making units (DMUs). However, the classic Charnes-Cooper-Rhodes (CCR) model often suffers from a lack of discriminatory power, particularly when the number of input and output variables is large relative to the number of DMUs. Under such conditions, many DMUs are evaluated as "CCR-efficient" with a score of unity, making it difficult to rank or distinguish between top performers. To address this limitation, we propose a novel Maximum-Margin DEA (MM-DEA) model that integrates the concept of "margin maximization" – inspired by structural risk minimization in machine learning – directly into the efficiency measurement framework. Like the traditional super-efficiency model, the proposed MM-DEA model excludes the target DMU from the reference set used to evaluate it; unlike the super-efficiency model, however, MM-DEA imposes an explicit upper bound on the target DMU's own score, ensuring that every score remains within the standard $[0,1]$ range rather than potentially exceeding one. We theoretically demonstrate that the MM-DEA model is a generalized extension of the CCR model, reducing to the latter when the trade-off parameter $\alpha$ is zero. Furthermore, we show that the fractional programming formulation of MM-DEA can be rigorously transformed into a linear programming (LP) problem, ensuring computational efficiency. In our numerical experiments, MM-DEA produced a unique ranking for a high-dimensional synthetic dataset on which the CCR model failed to discriminate among any of the DMUs. Our results suggest that the "margin" serves as a critical metric for managerial resilience and competitive advantage.

We study the Max-$k$-Cut problem from a node-feature perspective, where each vertex is associated with a feature vector and edge weights are given by pairwise inner products. We first examine the semidefinite relaxation of Max-$k$-Cut from this perspective. Using a normal-cone argument, we derive a general sufficient condition for exactness of the Frieze–Jerrum relaxation and show that it is satisfied in two feature-structural regimes: perfect feature balance, where the aggregate feature vectors of the parts are equal, and feature dominance, where a small set of large nonnegative feature vectors determines the structure of an optimal partition. We then show that the Max-$k$-Cut objective is equivalent to minimizing the sum of squared norms of the aggregate feature vectors assigned to the $k$ parts, thereby connecting the problem to vector balancing. Motivated by this observation, we show that a greedy feature-balancing algorithm retains the classical $1-1/k$ worst-case approximation guarantee and recovers an optimal partition under feature dominance. For rank-$1$ feature graphs with nonnegative features, classical bounds of Chandra and Wong for greedy load balancing yield a computable a posteriori optimality-gap certificate that depends only on the returned partition and requires no knowledge of the optimum.

We study second- and higher-order methods for solving smooth monotone variational inequalities (MVI). Monteiro and Svaiter (SIAM J. Optim., 2012) showed that a second-order method, NPE, converges at the rate of $\mathcal{O}(T^{-1.5})$. For convex-concave minimax optimization, a subset of MVI problems, Chen, Liu, Luo, and Zhang (COLT 2025) recently improved the complexity to $\tilde{\mathcal{O}}( T^{-1.75})$ . However, it is open whether the conjectured complexity for MVI can be improved. In this paper, by using a large-step inexact Halpern iteration, we propose a novel Halpern-NPE method that achieves an even faster rate of $\tilde{\mathcal{O}}(T^{-2})$ for solving MVIs. We also provide the $p$th-order generalization of our method. We first introduce an Anchored Tensor Method (ATM) that achieves the rate of $\mathcal{O}(T^{-(p-1)})$, and then combine it with the Halpern iteration to achieve a faster convergence rate of $\tilde{\mathcal{O}}(T^{-p})$. This improves all prior results for $p \ge 2$ and matches the classical extragradient method for $p=1$.

We study constrained statistical learning over infinite-dimensional hypothesis classes in the fully nonconvex setting, and establish universal PACC learnability of the solutions of dual algorithms: Probably Approximately Correct on Constraints, guaranteeing optimality and constraint satisfaction at once. This strengthens near-PACC results, whose feasibility residual no amount of data can remove. Optimality is caught between generalization, governed by Rademacher complexity and favoring small classes, and strong Lagrangian duality, which rests on Lyapunov convexity for vector measures and needs decomposability, a demand pulling the other way. We reconcile the two by posing the population problem over a universal RKHS $\mathcal{H}_K$, dense in a decomposable envelope, and learning over norm balls of growing radius. This yields the Tikhonov complexity $\mathfrak{T}^{\varepsilon}_{n}$, the least RKHS norm reaching an $\varepsilon$-optimal Lagrangian level set; we prove it finite, obtain exact learnability of the optimal value, and make the sample threshold explicit and polynomial in $1/\varepsilon$ under a source condition. Feasibility is harder: absent convexity the Lagrangian may not attain its infimum, and dual information pins down only an averaged constraint-risk vector, not the risks of any returned predictor. We introduce the closure-realization gap $\varepsilon^\star_\infty$, an index of how well $\mathcal{H}_K$ retrieves feasible solutions from dualization; it is a property of the problem, not of a modeling choice. Learnability is exact when $\varepsilon^\star_\infty=0$, in particular under dual differentiability, and near-PACC with residual exactly $\varepsilon^\star_\infty$ otherwise. Finally, no distribution-free threshold exists already in the unconstrained specialization, so universality is the canonical frame for dual algorithms over large hypothesis classes.