Papers

We consider linear regression models fitted by minimizing Bregman losses of the form \[ \frac{1}{n}\sum_{i=1}^n \left[ \phi(y_i)-\phi(x_i^\top\theta) -\phi'(x_i^\top\theta) \bigl(y_i-x_i^\top\theta\bigr) \right], \] where \(y_i\) is the observed response and \(x_i^\top\theta\) is the linear prediction. Even when the generating potential \(\phi\) is strongly convex, the resulting regression objective may be nonconvex in \(\theta\). This creates a gap between the convexity of the generating potential and the optimization geometry of the fitted model. The Hessian can be written as a weighted Gram matrix whose weights depend on the derivatives of the potential and the current residuals. This representation gives simple conditions for local strong convexity, smoothness, and conditional linear convergence of gradient descent. For the quadratic–quartic potential, we derive an exact scalar convexity condition, identify the interval of negative curvature, and obtain local and global sufficient conditions for positive curvature. Numerical experiments indicate that these scalar conditions may fail while the full Hessian remains positive definite at the evaluated points. They also indicate that the range of tested gradient-descent step sizes leading to convergence decreases as the quartic parameter grows. These results characterize how residual-dependent curvature interacts with the generating potential and the design matrix in reverse Bregman regression.

In this paper, we propose a fast gradient algorithm for the problem of minimizing a differentiable (possibly nonconvex) function in Hilbert spaces. We first extend the dry friction property for convex functions to what we call the dry-like friction property in a nonconvex setting, and then employ a line search technique to adaptively update parameters at each iteration. Depending on the choice of parameters, the proposed algorithm exhibits subsequential convergence to a critical point or full sequential convergence to an “approximate” critical point of the objective function. We also establish the full sequential convergence to a critical point under the Kurdyka–Lojasiewicz (KL) property of a merit function. Thanks to the parameters' flexibility, our algorithm can reduce to a number of existing inertial gradient algorithms with Hessian damping and dry friction. By exploiting variational properties of the Moreau envelope, the proposed algorithm is adapted to address weakly convex nonsmooth optimization problems. In particular, we extend the result on KL exponent for the Moreau envelope of a convex KL function to a broad class of KL functions that are not necessarily convex nor continuous. Simulation results illustrate the efficiency of our algorithm and demonstrate the potential advantages of combining dry-like friction with extrapolation and line search techniques.

Understanding how experts make decisions and being able to transfer that knowledge is important, especially in complex engineering applications. It is highly valuable for training novices, improving the performance of human-machine systems, and potentially enabling fully autonomous systems that perform as well as human experts. However, an expert's decision-making strategy, developed through years of experience, is often not directly accessible, since the implicit preferences and decision rules involved can be difficult to specify explicitly. This has motivated the use of observed decisions made by the expert to learn an interpretable model that captures the expert's decision-making process. In this work, we develop an inverse optimization approach to jointly learn the decision-maker's preferences (or perceived costs) and the decision rules governing their choices. We demonstrate the general applicability of our approach using three case studies that consider a shift assignment problem, a production planning problem, and a real-world routing problem, respectively. Across these case studies, modeling both perceived costs and decision rules leads to better predictions, highlighting the value of the proposed framework and its greater flexibility in capturing and replicating expert decision making.

We study continuous-time projection neural networks for variational inequalities on closed convex sets. A positive definite matrix that depends on the state preconditions the operator, and its inverse defines the projection metric. Existing convergence analyses of this flow cover Hessian-generated inverse metrics and state-dependent scalar metrics. In the first case, a Bregman distance eliminates metric-derivative terms. We treat a matrix metric whose inverse is of neither kind. We prove joint regularity of the projection in its argument and metric. Under common spectral bounds, the Euclidean Lipschitz estimate improves from the squared bound to the bound itself. For Lipschitz strongly monotone operators and Lipschitz metrics, we prove local exponential convergence with explicit rate and radius. On compact feasible sets, an explicit metric-variation bound yields global exponential convergence. A twice continuously differentiable, uniformly positive definite metric and a strongly monotone linear operator produce an annulus of periodic orbits. The inverse metric violates Hessian integrability throughout the annulus. Deterministic computations confirm the analytic formulas and quantify slack in both convergence certificates.

We present an open-source software package that implements a provably convergent Benders-type decomposition algorithm for multistage stochastic integer programs. In addition to standard cut families, such as Benders, strengthened Benders, and Lagrangian cuts, the algorithm incorporates rectified linear unit (ReLU) cuts, which provide convergence guarantees for general mixed-integer state variables. However, the dual problems used to generate these cuts often admit multiple optimal solutions. Although each solution yields a valid cut that separates the incumbent, the resulting cuts can differ in how well they approximate the subproblem cost. To strengthen these cuts, our package implements and evaluates two cut-selection strategies based on normalization and regularization of the dual problem. We also incorporate an alternating-cut criterion that uses cheaper Benders cuts when they are effective and invokes more expensive tight cuts only when necessary. Computational experiments on four classes of multistage stochastic integer programs benchmark these methods and provide insights on how problem structure affects their practical performance.

The Yau-Yau filter reformulates nonlinear state estimation as probability-density propagation governed by the Forward Kolmogorov equation (FKE). Applying it to target tracking, however, requires efficient FKE approximation on a finite computational domain. This paper proposes a Recentered-Domain Yau-Yau Filter (RD-YYF), which solves the FKE within a fixed-size local window centered at the latest state estimate. This design concentrates numerical resolution near the dominant posterior density. Offline, physics-informed neural networks (PINNs) generate FKE solution snapshots, while principal component analysis constructs a low-dimensional representation of density evolution. A lightweight residual surrogate maps the initial-condition coefficients and domain center to the terminal-solution coefficients. Online, the pretrained surrogate predicts density evolution within the recentered window, followed by observation update and state estimation. Experiments on two geometrically constrained target-tracking examples show that RD-YYF achieves lower tracking errors than the extended Kalman filter (EKF) and particle filter (PF), while retaining efficient per-timestep inference. Ablation results indicate that domain recentering improves density approximation in high-probability regions and accelerates offline PINN convergence. These results demonstrate the potential of RD-YYF for efficient nonlinear target tracking

Iterate convergence of Bregman proximal methods (BPMs) has long remained open, especially for nonconvex objectives. Recently, \citet{chen2026skl} made progress by establishing iterate convergence for a BPM via the so-called scaled Kurdyka-Łojasiewicz (SKŁ) property, but only for the Shannon entropy kernel and linearly constrained problems. In this paper, we develop a unified iterate convergence framework that applies to a broad group of kernels and composite objective functions. Our approach extends the analytical tools in \cite{chen2026skl}, in particular the SKŁ property, which plays a central role in ensuring convergence of the generated sequences. By introducing kernel-dependent parameterization functions, we show that the extended SKŁ property holds for all continuous subanalytic functions, particularly when the kernel has a closed domain. We then verify that the assumptions of the framework are satisfied by standard BPMs under mild regularity conditions, thereby establishing their iterate convergence for a wide range of objective functions. Furthermore, based on the parameterization functions, we show that the continuous-time BPM (mirror flow) converges to a stationary point for o-minimal definable objective functions, yielding the first trajectory convergence result for mirror flow without imposing convexity assumptions on the objective function or isolation assumptions on stationary points. Taken together, these discrete- and continuous-time convergence results provide a unified trajectory convergence theory for BPMs.

Quantum state tomography (QST) has attracted considerable attention due to its fundamental role in quantum information processing. In this paper, we develop a unified theoretical framework for analyzing the sample complexity of structured QST under noisy observations arising from state preparation noise, measurement noise, and finite-shot statistical noise. The proposed framework applies to a broad family of structured quantum-state classes, including general mixed states, sparse states, low-rank states, matrix product states (MPSs), matrix product operators (MPOs), projected entangled-pair states (PEPSs), and projected entangled-pair operators (PEPOs), while further introducing physically consistent structured models—including low-rank and sparse states, low-rank MPOs (LR-MPOs), and low-rank PEPOs (LR-PEPOs)—that simultaneously exploit low-dimensional structures and preserve the physical constraint. Within this framework, we derive unified non-asymptotic sample complexity guarantees for two constrained least-squares estimators under noisy observations: a noise-aware estimator that incorporates the calibrated noise model and a noise-unaware estimator based on the ideal Born measurement model. For the noise-aware estimator, we derive unified trace-norm recovery guarantees that explicitly characterize the dependence of the sample complexity on three fundamental quantities: the complexity of the underlying structured state class, the complexity of the measurement ensemble, and the state preparation and measurement noise levels. For the noise-unaware estimator, we establish a unified non-asymptotic recovery guarantee consisting of a statistical error term and an additional deterministic bias term arising from the mismatch between the assumed reconstruction model and the noisy observation process.

In this paper, we consider a class of multiblock nonconvex nonsmooth optimization problems, which covers many applications such as the analysis of pre-earthquake anomalies and machine learning. To solve this class of problems, we propose the inertial block proximal linearized method with two-phase adaptive momentum (IBPL$^+$-TP). Compared to the current methods, our method possesses three main advantages: (1) it introduces a two-phase adaptive momentum strategy to effectively update the extrapolation parameters, (2) it allows using two different extrapolation points to accelerate the convergence, (3) it allows the extrapolation parameters of these two extrapolation points to be independent of and unconstrained by all other parameters. While maintaining the above advantages, we prove that our method ensures the monotonic convergence of the objective function of this class of problems, and we also prove that the sequence generated by our method globally converges to a critical point, as well as establish the convergence rate of our method. To demonstrate the effectiveness of our method, we apply it to solve two nonconvex and nonsmooth machine learning problems, namely sparse nonnegative matrix factorization with $\ell_0$-constraints and sparse nonnegative CP decomposition with $\ell_0$-constraints. The numerical experimental results on solving these problems show that our method outperforms several state-of-the-art methods.

This paper presents a discrete-time passivity-based analysis of first-order momentum-based methods for a class of functions whose gradient has lower and upper sector bounds of $0$ and $L$, respectively. Through a loop transformation, it is shown that momentum-based methods can be represented as a passive controller in negative feedback with an output strictly passive (OSP) system. The weak passivity theorem is then used to derive explicit hyperparameter conditions under which the shifted gradient asymptotically vanishes. Under an additional assumption that requires the existence of a unique stationary point and excludes arbitrarily small gradients far from that point, convergence of the iterates to the global minimizer is established.

In this paper, We prove that: (A) If $m\ge 4,$ there exists a Platonic geodesic tree having $m$ edges (geodesic segments), which form $m$ platonic angles with measure $120^{o}$ and $(\frac{m(m-1)}{2}-m)$ platonic angles with measure $90^{o}.$ (Existence of a Fermat Coordinate system) (B) If $m=2,$ there exists a Platonic geodesic tree with three vertices and 2 edges, which meet at $120^{\circ}$ or $150^{\circ}$ or $180^{\circ}$ (Minimal spanning geodesic tree). (C) If $m=3,$ there exists a Platonic tree and the Steiner point is identical with the Fermat point, having three edges, which meet at $120^{\circ}$ (Hexagonal coordinate system). (D) If $m$ is even and $m \ge 4,$ there exists a Platonic geodesic tree having $m$ edges , which form platonic angles with only two distinct measures $90^{o}$ and $180^{\circ}.$ (Existence of a Cartesian coordinate system). (E) If $m\ge 4$, there does not exist a Platonic geodesic tree having $m$ edges , which form only platonic angles with measure $120^{o}$ (Non existence of a Fermat coordinate system). (F) If $m\ge 3,$ there does not exist a Platonic geodesic tree having $m$ edges, which form platonic angles with only one measure $150^{o},$ or two distinct measures $120^{\circ}$ and $150^{\circ}$ or two distinct measures $90^{\circ}$ and $150^{\circ}$ (Non existence of a Fermat coordinate system). A classification of platonic geodesic trees with boundary vertex set $V$ and added vertex set $Q$ in the $K-$Space is given by conditions (A), (B), (C), (D), (E), (F), which characterize each vertex in $V\cup Q.$

This work introduces a proximal stochastic subgradient method for minimizing the sum of an expected cost, whose integrand is potentially nonsmooth and nonconvex, and a lower semicontinuous, prox-bounded function. We target a broad class of integrands obeying a nonsmooth, localized variant of the descent lemma in the decision variable, a structural assumption that simultaneously covers smooth losses with Lipschitz gradient and differences of such losses with convex functions. At each iteration the expected cost is replaced by a sample average that is progressively refined, and the proximal-subgradient stepsize is selected by an Armijo-type line search enforcing a sufficient-decrease property up to stochastic errors induced by the sample-based approximation. This framework accommodates substantially more general problem formulations than existing methods, in particular, it requires neither (weak) convexity of the regularizer nor a uniform bound on the variance of the stochastic oracle, and our analysis yields convergence guarantees that are new even in the smooth setting. Specifically, we establish almost sure convergence of the sequence of function values and stationarity of every accumulation point of the trajectories under the relaxed requirement that the sample-size sequence be merely nondecreasing and unbounded, with no prescribed growth rate. Leveraging the Kurdyka-Lojasiewicz (KL) property, we further upgrade this subsequential guarantee to convergence of the whole trajectory to a single stationary point. Finally, for exponential-type KL desingularizing functions and polynomially growing sample sizes, we derive explicit polynomial convergence rates, up to a logarithmic factor, for both the function values and the iterates.

This study addresses the critical yet under researched area of supported evacuation for vulnerable populations during wildfires, such as hospital patients and long term care residents, by developing a two stage stochastic optimization model that optimizes facility location, fleet sizing, and vehicle routing under strict time windows. To overcome the problem NP hard complexity, the authors propose an innovative solution methodology leveraging Logic Based Benders Decomposition, featuring Combinatorial Benders Cuts and logic based inequalities. Extensive numerical experiments and real world data from a community wildfire drill in Roxborough Park, Colorado, demonstrate that the proposed approach yields high quality solutions, significantly improving shelter placement, vehicle utilization, and overall evacuation efficiency compared to alternative policies.

A Hider distributes two identical balls among three boxes whose search costs satisfy $a\ge b\ge c>0$. A Searcher opens boxes adaptively until both balls are found; every opening incurs the corresponding box cost and recovers at most one ball. We give an exact solution of this heterogeneous multi-look search-cost game. The value is the maximum of three explicit rational functions. In the three parameter regimes, an optimal Hider strategy is the product-form distribution restricted respectively to three, five, or all six placements. Every non-wasteful deterministic Searcher policy is payoff-equivalent to one of $72$ elementary decision trees, which induce only $42$ distinct opening-count profiles. An analytic argument settles the first regime. Two exact Searcher mixtures settle the second, and four settle the third. Feasibility of the parameterized mixtures over the full cost region is established by exact rational Bernstein-basis certificates. Independent implementations reproduce the policy set, the zero-sum linear-program values, and all $154$ dyadic nodes examined by the Bernstein verifier.

This paper introduces a contour integral method (CIM) for efficiently computing outputs of parametric linear systems in control form over specified time intervals and to a user-prescribed accuracy. The CIM approximates the inverse Laplace transform via a quadrature rule applied along a modified integration contour. For parametric systems, we show how CIM integrates effectively with projection-based model order reduction (MOR) where a greedy algorithm builds the projection spaces following an error estimate we derive for this setting. We additionally demonstrate that the developed projection framework naturally enforces Hermite interpolation conditions. This combination substantially lowers the cost of evaluating the input-output relations across the parameter domain, for a wide range of input functions, and for initial conditions well captured by a low-dimensional subspace.. We demonstrate the accuracy and efficiency of the approach on benchmark non-parametric and parametric control systems, comparing against state-of-the-art projection-based MOR methods.

Gradient descent on a factored model $W = UV^\top$ is implicitly biased toward low-rank solutions, while Adam, starting from the same small initialization, is not. We trace the difference to the gauge symmetry of the loss, its invariance under $(U, V) \mapsto (UQ, VQ)$. Gradient flow's low-rank mechanism is available to an optimizer only if that optimizer is gauge-equivariant, a condition necessary for the transfer but not sufficient for low-rank recovery. Gradient descent, momentum, "shared-scalar" Adam, Muon, and Shampoo satisfy it. Adam, RMSProp, and the other coordinate-wise methods do not. A structure theorem characterizes the memoryless equivariant rules as exactly the Gram-determined left preconditioners, and a transfer theorem carries gradient flow's pathwise properties to common-scalar flows. We then sort nine update rules on underdetermined matrix sensing by recovery error against the planted ground truth. A one-parameter family from coordinate-wise to shared-scalar preconditioning restores the bias monotonically, isolating anisotropy as the cause. A "spectral schedule" reconciles two opposing reports about Muon: equal-rate updates recover exactly low-rank targets but lose their edge as the spectral tail grows. In transformers, Adam separates two gauge-equivalent initializations at the first step, where the equivariant optimizers stay at float precision, and ends with the per-head invariants $W_Q^\top W_K$ 56% apart in relative Frobenius distance, a gap no per-head rotation can close. On two hyperspectral datasets at matched training loss, gradient descent cuts held-out error by 43-44% at the lowest sampling density, and at lower effective rank. Basis choice is therefore not a tuning detail but a decision about which interpolant the optimizer selects.

The iterate convergence of Bregman projected gradient method (BPGM) has remained a long-standing open problem, especially for the widely adopted Shannon entropy kernel. Existing convergence results are often limited, relying on Lipschitz continuity of the kernel's gradient or restrictive conditions on the objective function. In this paper, we develop a novel convergence analysis framework for the BPGM with the Shannon entropy kernel, yielding strong convergence results for a broad class of objective functions under linear constraints. The cornerstone of our framework is a new concept called scaled Kurdyka-Łojasiewicz (SKŁ) property, which captures the local growth behavior of a function under the Bregman geometry. We show that the SKŁ property ensures the iterate convergence of BPGM and holds for all continuous subanalytical functions. Furthermore, we prove that the BPGM sequence exhibits linear convergence if the problem possesses an SKŁ\ exponent of $1/2$. We then furnish the examples of functions with the SKŁ\ exponent $1/2$ by proving that the SKŁ\ exponent $1/2$ is implied by the KŁ exponent $1/2$ under strict complementarity and local Lipschitz continuity of the objective's gradient. Building on these novel results, our work takes a first step towards resolving the open problem of BPGM iterate convergence.

Operating-room allocation is a major challenge for public hospitals with limited surgical capacity and large elective waiting lists. This work proposes a two-stage operating-room allocation framework for specialty-block scheduling environments. The methodology combines a mixed-integer linear programming model for medical specialty block allocation with a priority-based patient allocation procedure. The framework was evaluated through one-week and multi-week simulation scenarios using parameters estimated from historical surgical and waiting-list data. The proposed methodology was compared against an integrated mixed-integer linear programming baseline adapted from the literature under realistic operational disruptions, including failed patient confirmations and surgery suspensions. Results show that the proposed framework achieves a more balanced distribution between offered and demanded surgical time across specialties while maintaining high operating-room utilization and substantially lower computational times than the integrated baseline. In the 20-week scenario, the proposed methodology achieved lower waiting times and operated a larger number of patients while preserving operational flexibility. These results suggest that decomposition-based operating-room allocation strategies provide a practical and computationally efficient alternative for reducing surgical waiting lists in public hospitals.

Bergemann and Morris (2016) show that one information structure is more informative than another exactly when it induces a smaller set of Bayes correlated equilibrium outcomes in every game. We build the quantum analogue. An information structure becomes a family of density operators indexed by the payoff state, which the mediator observes. We show that obedience is equivalent to a Loewner domination between operators on one player's subsystem. The equilibrium set is then a nonempty compact spectrahedron computable by semidefinite programming, classical structures embed exactly, and under quantum individual sufficiency more information shrinks the equilibrium set in every game.

We develop accelerated algorithms for convex $(H_0,H_1)$-smooth optimization, where $\|\nabla^2 f(x)\|\le H_0+H_1(f(x)-f^*)$. This class generalizes standard smoothness and contains the $(L_0,L_1)$-smooth class. Combining a Nesterov-type accelerated gradient scheme with small-dimensional relaxation and phase restarts, we obtain a full-gradient method with iteration complexity $\widetilde O(\sqrt{H_0\widetilde R^2/\varepsilon}+\sqrt{H_1\widetilde R^2}\log(F_0/\varepsilon))$. We extend the same approach to randomized coordinate optimization, obtaining a coordinate method with uniform sampling whose iteration complexity carries the standard factor $d$, and a coordinate method with non-uniform sampling whose iteration complexity is governed by $S_{1/2}^{(j)}=\sum_i\sqrt{H_{j,i}}$. These results provide, to our knowledge, the first accelerated full-gradient and coordinate guarantees for this convex class. We also provide practical implementation recommendations. Experiments confirm the predicted acceleration, gains from non-uniform sampling, and the viability of inexact relaxation.

In optimization, one often encounters problems of the form $\min_x F(x)+E(x)+G(Kx)$. In this work, we combine primal-dual algorithms with multigrid techniques for their solution. To link the the fine-grid and coarse-grid problems problems, we introduce a nonsmooth primal-dual coherence condition, and an efficient partially linearized line search procedure. Our work is motivated by total variation regularized inverse imaging problems, on which we demonstrate the efficacy of the method, being able to solve problems not previously possible with forward-backward multigrid methods.

Reconstructing unknown electromagnetic sources from far-field radiation patterns is a fundamental inverse problem with broad applications in biomedical imaging, non-destructive testing, and telecommunications. In practical settings, however, collecting dense multi-frequency far-field measurements at the Nyquist sampling rate is often infeasible. Under-sampled or sparse data introduce non-radiating source components that sever the uniqueness of the solution, creating severe artifacts when standard inversion techniques are applied. To overcome this limitation, we present a two-stage reconstruction strategy exploiting the physical property that compactly supported, geometrically sparse sources exhibit a finite rate of innovations (FRI). In the first stage, we construct an associated wrap-around structured Hankel matrix. By leveraging the low-rank property of the matrix due to FRI of the unknown sources, we enrich the sub-sampled data. To that end, we convert missing multi-frequency far-field data recovery into a constrained matrix completion task solved via Annihilating Filter-based Low-rank Hankel Matrix Completion Approach (ALOHA). In the second stage, a Fourier inversion scheme reconstructs the current source density from the enriched dataset. Extensive numerical evaluations on electromagnetic source models show that our enrichment framework effectively eliminates under-sampling artifacts and resolves non-uniqueness challenges. The method delivers accurate and stable reconstructions under high sub-sampling rates (e.g., with $30$% to $50$% available samples) and strong noise conditions ($10$ dB SNR), outperforming standard $\ell_1$-compressed sensing baselines.

This paper develops a prescribed performance leader-following consensus protocol for directed networks under event-based broadcasting. Each node schedules its transmissions using only its current state and its last broadcast value, while followers use held neighbor and leader samples between broadcasts. To accommodate jumps in sampled neighborhood errors caused by asynchronous receptions, a decaying receiver-side performance correction term temporarily modifies the prescribed performance function. For a directed network containing a spanning tree rooted at a time-varying leader with bounded trajectory and derivative, we prove existence and uniqueness of complete non-Zeno solutions, boundedness of all closed-loop signals, and explicit prescribed bounds for the sampled and continuous neighborhood disagreements and the leader-following error. These bounds quantify the effects of the communication topology and broadcast thresholds. Simulations verify the prescribed performance bounds and illustrate the event-based communication mechanism.

Stochastic gradient descent (SGD) optimization methods are the standard instruments for the training of deep neural networks (DNNs). In many relevant artificial intelligence (AI) systems - such as popular large language models (LLMs)-not the standard SGD scheme is used as the optimization method but instead suitable accelerated variants of SGD are employed. One of the most popular methods of such accelerated SGD variants is the momentum orthogonalized by Newton-Schulz (MUON) optimizer proposed by Jordan et al. in 2024. The MUON optimizer exploits the special matrix structure of the weight parameters in the training of the DNNs and, in its original form, employs five Newton-Schultz (NS) matrix steps in each MUON iteration. In this work we propose and study a generalized variant of the MUON optimizer involving an arbitrary number of generalized NS steps with polynomials of possibly arbitrary high degree. The considered optimizer covers MUON with the original NS polynomial as well as MUON combined with the recently proposed Polar Express method as special cases. For a simple class of stochastic optimization problems (SOPs) we show for almost every mini-batch size that MUON fails to converge to the solution of the SOP as the number of gradient steps converges to infinity. We also establish an error analysis for MUON with the generalized NS steps that provides convergence rates in terms of the number of gradient steps and in terms of the size of the mini-batch. We illustrate our general error analysis for MUON in the case of several concrete examples including quadratic stochastic optimization problems (SOPs) as well as $\ell_2$ regularized logistic regression for binary classification.

We propose MELMO (Moreau Envelope Smoothing with Linear Minimization Oracles), an algorithm for composite optimization problems of the form min x f (x) + g(T x), where f is smooth and g may be non-smooth. The method leverages the Moreau envelope to smooth the non-smooth component while adapting to problem geometry through linear minimization oracles. Assuming g is $\rho$-weakly convex, we establish a family of convergence bounds parameterized by the step-size and smoothing schedules, thereby making explicit the trade-off between optimizing the smoothed objective and recovering stationarity for the original composite problem. In particular, one regime yields O(k -1/4 ) rates for both the smoothed-gradient norm and a composite stationarity proxy, while another yields O(k -1/3 ) for the smoothed-gradient norm together with O(k -1/4 ) for the composite proxy. We also establish a K-horizon-dependent convergence rate that yields O(K -1/3 ) for the composite proxy. Empirically, MELMO is competitive with variable smoothing and subgradient baselines on sparse low-rank matrix factorization and image denoising.

Given finitely many compact convex bodies in $\R^n$, one seeks translates maximizing the volume of their common intersection. A lazy solver merely translates each body so as to place its centroid at the origin. We prove that the lazy strategy always captures strictly more than $\left(\tfrac{2}{n+1}\right)^n$ of the optimal volume, and that this constant is sharp: families of cones over tangent disks, indexed by finite nets on the sphere, approach it. The infimum is not attained. For two convex bodies in the plane the resulting sharp constant $4/9$ closes a gap open since 1996, when de Berg, Cheong, Devillers, van Kreveld and Teillaud proved that centroid alignment of two convex polygons captures at least $9/25$ of the maximum overlap and exhibited examples capturing only $4/9$. The proof rests on the following identity: for a convex body $K$ with centroid at the origin, the intersection of all centroid-recentered compact convex supersets of $K$ equals $\tfrac{1}{n+1}(K-K)$. We close with a promise-problem variant in which the lazy strategy captures at least $\left(\tfrac{n}{n+1}\right)^n > \tfrac1e$ of the optimum, uniformly in the dimension. All results below have been checked in Lean~4; the one classical input quoted rather than proved is the equality case of the Brunn–Minkowski inequality, which enters only for $n\ge2$.

Data-driven decision pipelines combining predictive machine learning models with downstream optimization software are increasingly used to make high-stakes operational decisions. Certifying the safety, fairness, and reliability of these decisions is essential, yet traditional scenario generation methods rely on repeated random testing, which becomes computationally prohibitive when failure events are rare and offers little insight into why failures occur. We present a framework for local violation certification designed specifically for linear decision pipelines under input uncertainty. We mathematically demonstrate that standard sampling methods fail efficiently for rare violations, motivating a direct structural approach. By analyzing the fixed decision boundary of a deployed pipeline, we show that the local risk of failure can be calculated directly in closed form using a single optimization solve. Furthermore, we introduce an exact sampling procedure and closed-form risk statistics that provide feature-level attributions (identifying which input characteristics contribute most to potential non-compliance) without requiring repetitive random trials or complex sampling algorithms. We demonstrate our approach on an economic power dispatch system subject to emissions regulations, delivering precise, auditable risk assessments at a fraction of the traditional computational cost.

Blind deconvolution has been widely applied in system identification and signal processing. While joint sparsity commonly arises in practical scenarios, effectively exploiting this structure to enhance recovery performance remains a challenging and largely open problem. In this paper, we propose a joint-sparsity-promoting optimization problem and develop a Riemannian optimization algorithm for its accurate and efficient solution. We further establish theoretical guarantees that characterize the non-asymptotic relationship between the estimation error and the sample complexity, showing that exploiting joint sparsity can significantly reduce the sample complexity required for successful recovery. Numerical experiments are provided that validate the theoretical results and demonstrate the effectiveness of the proposed approach.

We give a negative solution to MAIS-O60. We first construct an example in which an initially active ReLU neuron becomes completely inactive in finite time and thereafter remains frozen at a limit whose Fourier energy is equally distributed among all nonzero real frequency classes. The counterexample holds on an open set of initial conditions and therefore occurs with positive probability under Gaussian initialization. An appendix prepared by GPT-5.6 Sol strengthens the counterexample by showing that the same failure can occur for every Clarke trajectory from an open set of initial conditions, under the convention $\mathrm{ReLU}'(0)=0$, for smooth dead-zone approximations of ReLU, and for fixed-step full-batch gradient descent. Thus, single-frequency alignment is not a general consequence of training a single neuron on modular addition.

We study asynchronous optimization for finite-sum eigenspace computation in heterogeneous distributed systems. The theoretical foundations for asynchronous eigenspace computation remain scarce, with existing approaches offering limited coverage of dynamics directly on the Grassmannian under stale information. In this paper, we propose a Grassmannian incremental aggregation method that refreshes only arriving components and reuses cached gradients, retaining low per-update cost without global synchronization. The method employs an extrinsic polar update that preserves the intrinsic subspace geometry without requiring parallel transport of stale tangent vectors. Our analysis establishes a tight angle-dependent gradient-dominance characterization of the objective and a basin-invariance property for stale aggregated updates. These yield two-phase linear convergence, comprising an explicit broad-basin regime and a sharper local regime, with constants controlled by component spectral spreads. Experiments on serial and distributed PCA demonstrate improved sample efficiency and wall-clock convergence over representative baselines.