Optimization and Control
Papers
Fresh from arXiv, typeset properly. Highlighted words trade on the market.
Sunday 12 April
Research on multi-objective combinatorial optimization and on the Cutting Stock Problem (CSP) has been widely developed over the years. In contrast, the multi-objective Cutting Stock Problem has received limited attention and has been explored in only a small number of studies. In this paper a bi-objective study of the one-dimensional and the two-dimensional CSP is presented. It distinguishes itself from other research in the literature in two key aspects, among others. The first regards the model used to represent the problem and the second is the solution strategy based on dynamic column generation embedded into scalarization methods. Three methods adapted from the literature to analyse the trade-off between the minimization of the total number of objects and the total number of saw cycles are implemented. The computational results show that the use of dynamic column generation provides a better approximation of the Pareto front. As for the scalarization methods, none stood out. In fact, they can be viewed as complementary. The cardinality and hypervolume of an approximation of the Pareto front built from the union of the points generated by the three methods is always greater than the ones generated by each one. Dealing with the multi-objective nature of CSP, this study provides insights and computational tools to help industry obtain pragmatic solutions.
Mathematical Physics
No-Go Theorem for Quasiparticle BEC
We discuss a no-go theorem for Bose-Einstein condensation (BEC) of quasiparticles (phonons) from the viewpoint of operator algebras, using the van Hove model. The $\beta$-KMS states of the van Hove model satisfy the self-consistency condition of arXiv:1207.4621. However, the self-consistency condition is a constraint concerning the definition of the field, and is insufficient to establish the no-go theorem for BEC. In this paper, we prove the no-go theorem for BEC via two routes. First, imposing time cluster properties on the $\beta$-KMS states precludes BEC. Second, under nonlinear dispersion with $s > 2$, the treatment of infrared divergences automatically reduces the algebra of physical observables, and BEC is mathematically excluded on the reduced algebra. In particular, the latter property admits an interpretation in terms of the ideal theory of the resolvent algebra.
Read the full abstract
Logic
Iterating Generalised Perfect Set Forcing Along Well-Founded Orders
The technique of geometric forcing iteration was developed by Kanovei \cite{zbMATH01335192} and used to prove that the perfect set forcing can be iterated with countable supports along any partial order, while preserving $\aleph_1$. In \cite{Property-B} we considered a generalised perfect set forcing with respect to a filter on a cardinal $\kappa$ satisfying $\kappa^{<\kappa}=\kappa$, which we denoted ${\mathbb P} (\mathcal F)$, and we proved that its iteration with supports of size $\le\kappa$ along any ordinal preserves cardinals up to and including $\kappa^+$. We show that there is a version of the geometric iteration technique that applies to ${\mathbb P} (\mathcal F)$ and yields that for $\kappa$ satisfying $\kappa^{<\kappa}=\kappa$ and for appropriate filters $\mathcal F$, the forcing ${\mathbb P} (\FF)$ can be iterated with supports of size $\le\kappa$ along any well-founded partial order and preserve cardinals up to and including $\kappa^+$. As an application of our technique we obtain that common notions of arboreal forcings on $\omega$ can be iterated with countable supports along any well-founded partial order and that such iterations preserve $\aleph_1$.
Read the full abstract
Number Theory
Beatty solutions of almost Golomb functional equations
We study the almost Golomb equation of order $r$, $a\bigl(S(n)\bigr)=n$, where $S(n)=a(n)+a(n-1)+\cdots+a(n-r+1)$, for nondecreasing sequences of positive integers. Its greedy solution is $r$-regular in the sense of Allouche and Shallit. Beyond that one, and for every non-square order $r$, the equation has another solution, an inhomogeneous Beatty sequence $a(n)=\lfloor n/\sqrt{r}+d \rfloor$ where $d$ is a parameter. No other positive slope occurs, and the equation holds for all $n\ge r$ exactly when $d$ lies in an explicit closed interval depending on $r$. That interval is a single point when $r=2$ and has positive length for every non-square $r\ge 3$. Iterating the equation gives, for $k\ge 1$, $a^{\circ k}\bigl(S(n)\bigr)=a^{\circ(k-1)}(n)$, with $a^{\circ 0}$ the identity. Its Beatty solutions of positive slope are again of the form $\lfloor n/\sqrt{r}+d \rfloor$, and the sets of admissible $d$ form an increasing chain. The equation for $k$ holds exactly when $a^{\circ(k-1)}$ agrees at $n$ and at $a(S(n))$. We determine these solutions for $k=2$ when $r=2$ and $r=3$. At the right endpoint of the $r=2$ interval the first equation fails on a thin set of indices, which we identify as the return times of an irrational rotation to an explicit interval. The proofs combine equidistribution with an exact computation in $\mathbb{Z}[\sqrt{r}]$ over a finite range of orders.
Read the full abstract
Probability
A Strict Gap Between Relaxed and Partition-Constrained Spectral Compression in a Six-State Lumpable Markov Chain
This paper studies a finite reversible lumpable Markov chain for which relaxed spectral compression yields a larger determinant than partition-constrained compression. For a symmetric six-state lumpable chain and the positive operator $T=P^2$, I compare the relaxed benchmark \begin{equation*} \mathfrak D^{\mathrm{rel}}_3(T):=\sup_{U^*U=I_3}\det(U^*TU) \end{equation*} and the partition-constrained benchmark \begin{equation*} \sup_{\mathcal A\,\mathrm{3\text{-}partition}}\det Q_{\mathcal A}(T), \qquad Q_{\mathcal A}(T)=H_{\mathcal A}^*TH_{\mathcal A}. \end{equation*} Here the partition-constrained benchmark is the compression induced by normalized indicator vectors of genuine partitions of the state space. I derive closed formulas for the two analytically central partition families, prove strict upper bounds for both in a local-mode-dominated regime, and combine these bounds with an exhaustive enumeration of all $90$ partitions into three nonempty cells in an explicit six-state model. For this model, one obtains a strict global gap: \begin{equation*} \sup_{\mathcal A}\det Q_{\mathcal A}(T)<\mathfrak D^{\mathrm{rel}}_3(T). \end{equation*} Thus, in this model, indicator-based partition frames are strictly weaker than relaxed orthonormal frames even after global partition-constrained optimization.
Read the full abstract
Number Theory
Conjectural decomposition of symmetric powers of automorphic representations for $\mathrm{GL}(n)$
Let $\pi$ be a cuspidal automorphic representation for $\mathrm{GL}(n)$ over a number field. We establish a conditional upper bound on the number of cuspidal isobaric summands in the symmetric $k$-th power lift of $\pi$, assuming that the symmetric $m$-th power lift of $\pi$ is automorphic and cuspidal for all $m \leq k-1$, along with other specified Langlands functoriality conjectures. For sufficiently large $k$, the resulting bound is independent of the specific value of $k$. We further extend our study to cases in which the cuspidality assumptions on the symmetric power lifts are relaxed.
Read the full abstract
Combinatorics
Hopf substitutions in Species
In the theory of species, the species $\mathbf{L}$ of linear orders and the substitution operation $\boldsymbol{\circ}$ combine for a compelling result: given any positive comonoid $\mathbf{p}$, $\mathbf{L}\boldsymbol{\circ}\mathbf{p}$ carries the structure of Hopf monoid, freely generated by $\mathbf{p}$. Leaving aside the universal property this implies, we ask, "for which $\mathbf{b}$ does $\mathbf{b}\boldsymbol{\circ}\mathbf{p}$ carry the structure of Hopf monoid?" After answering this question, we look at basic properties of our construction. We also extend a result of the present authors, on interpolation in species, to this new context.
Read the full abstract
Statistics Theory · Cross-listed from cs.LG
Online Covariance Estimation in Averaged SGD: Improved Batch-Mean Rates and Minimax Optimality via Trajectory Regression
We study online covariance matrix estimation for Polyak–Ruppert averaged stochastic gradient descent (SGD). The online batch-means estimator of Zhu, Chen and Wu (2023) achieves an operator-norm convergence rate of $O(n^{-(1-\alpha)/4})$, which yields $O(n^{-1/8})$ at the optimal learning-rate exponent $\alpha \rightarrow 1/2^+$. A rigorous per-block bias analysis reveals that re-tuning the block-growth parameter improves the batch-means rate to $O(n^{-(1-\alpha)/3})$, achieving $O(n^{-1/6})$. The modified estimator requires no Hessian access and preserves $O(d^2)$ memory. We provide a complete error decomposition into variance, stationarity bias, and nonlinearity bias components. A weighted-averaging variant that avoids hard truncation is also discussed. We establish the minimax rate $\Theta(n^{-(1-\alpha)/2})$ for Hessian-free covariance estimation from the SGD trajectory: a Le Cam lower bound gives $\Omega(n^{-(1-\alpha)/2})$, and a trajectory-regression estimator–which estimates the Hessian by regressing SGD increments on iterates–achieves $O(n^{-(1-\alpha)/2})$, matching the lower bound. The construction reveals that the bottleneck is the sublinear accumulation of information about the Hessian from the SGD drift.
Read the full abstract
Functional Analysis
CPD $n$th roots of subnormal operators are subnormal
We investigate the $n$th root problem for bounded operators on a Hilbert space within the class of conditionally positive definite (CPD) operators determined by the Lévy–Khintchine formula. The class contains subnormal operators, complete hypercontractions of order $2$, and $3$-isometries. Our main result shows that if $T$ is a CPD operator such that $T^n$ is subnormal (resp., quasinormal, normal, or a $3$-isometry), then $T$ belongs to the corresponding class. This establishes the invariance of these classes under taking $n$th roots within the CPD class and extends several earlier results in operator theory. Furthermore, we provide characterizations of quasinormal and normal operators in terms of their CPD property and the structure of the representing triplet. Finally, we show that the classes of CPD and normaloid operators are distinct by means of both theoretical arguments and explicit examples.
Read the full abstract
Analysis of PDEs
Determining wavenumbers for Hall and electron magnetohydrodynamics turbulence
In turbulent flows, the Kolmogorov wavenumber characterizes the smallest scales at which viscous effects dominate. A mathematical analogue of this notion first introduced by Foias and Prodi [8] – a determining wavenumber – quantifies the minimal set of modes that uniquely determine the long-time behavior of solutions. Extending this framework from the Navier-Stokes equations to magnetized plasma models, we focus on the Hall-MHD and Electron-MHD turbulence in sub-ion and dissipation ranges. We prove existence of time-dependent determining wavenumbers for weak solutions of the Hall- and electron-MHD, improving upon previous results that were not optimal and lacked any comparison with phenomenological dissipation scales. Under explicit scale-localized intermittency assumptions, we show that their time averages are bounded above by Kolmogorov-like dissipation wavenumbers predicted by phenomenological studies of plasma turbulence. For strong electron-MHD solutions, we also establish a uniform bound on the magnetic determining wavenumber from Besov regularity.
Read the full abstract
Number Theory
On the Chevalley-Bass number of a field
We give upper and lower bounds on the Chevalley-Bass number of a field of characteristic zero, whenever this quantity is well-defined. We also describe an algorithm which computes the Chevalley-Bass number of a field, provided its maximal abelian subextension is known. As a primary application, we improve the value of a constant related to exponential diophantine equations.
Read the full abstract
Numerical Analysis
A DPG method for the circular arch problem
We consider an elastic model for a circular arch that incorporates membrane, transverse shear, and bending effects. The central line of the arch is partitioned into elements, and an ultra-weak variational formulation is developed alongside a discontinuous Petrov-Galerkin (DPG) approximation procedure based on so-called optimal test functions. The formulation uses discontinuous stress and displacement interpolations on the element mesh, with corresponding interface variables defined at the nodes. Theoretical analysis predicts optimal convergence rates for all quantities of interest, while also revealing potential error amplification influenced by the curvature of the arch and the imposed boundary conditions. The method is tested on examples with different support configurations. The numerical experiments confirm the theoretical predictions and further demonstrate that the accuracy of the DPG method can be improved by employing a suitably scaled test space norm.
Read the full abstract
Probability
Variable-Length Markov Chains on Finite Quivers: Boundary-Window Identifiability, Exact Depth, and Local Rank Comparison
Variable-length Markov chains on finite quivers provide a natural framework for context-dependent stochastic growth under incidence constraints. I study quiver-valued variable-length Markov chains observed through finite boundary windows and develop a first-order theory of visible-depth identifiability via stationary visible one-step transition laws and their restricted differentials on prescribed tangent blocks. For visible depth $m$, the main object is the stationary one-step informative map $q_{\mathcal{Q}}^{(m)}$. In the edge-homogeneous regime, once the local visible support is fixed and the representation hypothesis holds, all admissible visible depths encode the same edge-level extension law and hence have the same first-order rank. In the exact-depth regime of context length $r$, the depth-$r$ boundary process is the canonical finite-state Markov chain, smaller visible windows are deterministic truncations, and every coarser informative map factors $C^1$-smoothly through the depth-$r$ informative map on the relevant affine transition-array neighborhood. Hence rank cannot increase beyond depth $r$. After quotienting a tangent block by directions already invisible at depth $r$, I characterize strict coarse-depth loss exactly by coarse rank deficiency, equivalently by strict rank drop from depth $r$ to depth $m$ on the original block. I also give subspace-based and global selected-coordinate criteria, a global one-coordinate branching criterion, and an explicit depth-two example. Under full fine-depth rank and strict coordinate-rank loss at every smaller depth, a global coordinate-rank theorem yields $m_*(T,\theta_0)=r$. Reduced local coordinates remove stochastic redundancies, first-order criteria are invariant under $C^1$ reparameterization, and the statistical and LAN consequences remain conditional on additional estimation and likelihood-level hypotheses.
Read the full abstract
Combinatorics
Extremal chromatic bounds for distance Laplacian eigenvalues
For a connected simple graph $G$ on $n$ vertices with chromatic number $\chi$, the distance Laplacian matrix is $\DL(G)=\operatorname{diag}(\Tr_G(v_1),\dots,\Tr_G(v_n))-D(G)$, where $D(G)$ is the distance matrix and $\Tr_G(v)=\sum_{u\in V(G)} d_G(u,v)$ is the transmission. The eigenvalues of $\DL(G)$ are ordered as $\partial^{L}_1(G)\ge \partial^{L}_2(G)\ge \cdots \ge \partial^{L}_n(G)=0$. Building on the chromatic lower bound $\partial^{L}_1(G)\ge n+\ceil{n/\chi}$ and subsequent developments, we prove a color-class majorization principle: if $(\ell_1,\dots,\ell_\chi)$ are the color-class sizes in an optimal $\chi$-coloring with $\ell_1\ge\cdots\ge\ell_\chi$, then the first $\ell_1-1$ distance Laplacian eigenvalues satisfy $\partial^{L}_i(G)\ge n+\ell_1$, for $1\le i\le \ell_1-1$. This gives sharp lower bounds on the number of eigenvalues above the chromatic threshold $b_\chi=n+\ceil{n/\chi}$, thereby refining distribution theorems of [Aouchiche and Hansen, Filomat, 2017] and [Pirzada and Khan LAA, 2021]. We further refine clique/independent-set based multiplicity results by deriving explicit chromatic criteria in terms of neighborhood compression, and we generalize the extremal problem for minimum $\partial^{L}_1$ at fixed chromatic number by characterizing the balanced complete multipartite minimizers. Finally, we present a Ky Fan type result, and complement-component consequences of the majorization principle.
Read the full abstract
Combinatorics
Polynomial Time Enumeration of t-Stack-Sortable Permutations Ending in Their Least Entry
We study the behavior of West's stack-sorting map $s$ on permutations whose last entry is also their least. Let $S_{n}':=\{\pi0\mid \pi\in S_n\}$ where $\pi0$ denotes the concatenation of $\pi$ and $0$. For each permutation $\pi\in S_n'$, we introduce a new combinatorial object known as the stack-sorting tableau $T_{\pi}$, which ultimately serves as the key ingredient in the first polynomial time algorithm for counting the number of $t$-stack-sortable permutations in $S_n'$. We then establish a precise relationship between the behavior of $s$ on $S_{n}'$ and on $S_{n}$.
Read the full abstract
Optimization and Control
Pseudoconvex Problems in Operational Decision Systems: Algorithms for Joint Learning and Optimization
We consider joint optimization and learning problems arising in real-time decision systems. While most existing work focuses primarily on convex, revenue-based objectives, we extend this line of research to multi-objective formulations. In energy systems, for instance, we incorporate metrics such as renewable penetration and generation costs. Our key focus, however, is on a class of problems with a pseudoconvex structure - a natural relaxation of convexity. Representative examples include fractional objectives in energy management and logit-based revenue models in retail. The outer-level problem optimizes these pseudoconvex objectives, while the inner-level problem involves training a machine learning model using historical data. Our contributions are twofold. First, we propose a simultaneous learning-and-optimization framework that iteratively updates both inner- and outer-level variables. Second, we develop convergent algorithms for these problem classes under realistic mathematical assumptions. Using real-world datasets, we evaluate the computational performance of our methods and highlight an important observation: there exist clear trade-offs between inexact learning and computational time when assessing final solution quality.
Read the full abstract
Optimization and Control
Convergence and Stability of a Catching-Up Algorithm for Differential Inclusions with Maximally Monotone Operators
We study a catching-up algorithm with approximate projections for finite-dimensional differential inclusions governed by a maximally monotone operator and a continuous perturbation. Through the canonical decomposition of the operator into a regular part and the normal cone to the closure of its domain, the problem becomes a sweeping process, discretized by an explicit predictor step and a feasible approximate projection. For variable time steps, the approximations converge, up to a subsequence, to a solution on every finite horizon when the mesh size and the accumulated projection-error budget vanish and the nodes stay in a common compact convex set. This set comes from linear growth on finite horizons and, without any growth condition, by dissipativity of the projected field at a feasible center, which confines the nodes to a fixed ball uniformly in time for small steps and small projection errors. Under either condition, locally Lipschitz perturbations give global well-posedness, convergence of the whole family, and, for a common initial state, a trajectory-error estimate of order $O(\sqrt{h+E})$, where $h$ is the maximal step and $E$ the accumulated budget; sparse projection errors are allowed. An elastic network with unilateral stops illustrates the results, and numerical experiments compare the algorithm with the full-resolvent discretization.
Read the full abstract
Functional Analysis
Finite-codimensional subspaces of Daugavet spaces: projection constants and minimal projections
Over the real or complex field, we establish a duality formula for projection constants of finite-codimensional subspaces of Banach spaces with the Daugavet property. If \[ Y=\bigcap_{j=1}^n \ker f_j \subset X, \qquad W=\operatorname{span}\{f_1,\dots,f_n\} \subset X^*, \] then \[ \lambda(Y,X)=1+\lambda(W,X^*), \] and minimal projections onto $Y$ correspond exactly to weak$^*$-continuous minimal projections onto $W$. This yields, in particular, a complete description of the hyperplane case: every hyperplane has projection constant $2$, and $\ker f$ admits a minimal projection if and only if $f$ attains its norm. We then specialise to the real space $X=C[0,1]$. Our second ingredient is a transfer principle from duplication-stable finite-dimensional subspaces of $\ell_1^N$ to piecewise-constant subspaces of $L_1[0,1]\subset M[0,1]=C[0,1]^*$. For the regular symmetric spaces constructed by Chalmers and the second-named author and the second named author and Prophet, respectively, the transferred subspaces retain their projection constants but admit no weak$^*$-continuous minimal projections. Passing to annihilators yields finite-codimensional subspaces of the real space $C[0,1]$ for which the infimum defining the projection constant is not attained. As a consequence, for every $\Lambda\in[2,\infty)$ there exists a finite-codimensional subspace $Y$ of the real space $C[0,1]$ such that \[ \lambda(Y,C[0,1])=\Lambda, \] and the infimum defining $\lambda(Y,C[0,1])$ is not attained. For each even codimension $n$ we moreover realise every value in the interval $(2,1+\beta_n]$, where \[ \beta_n = \mathsf E_{{\mathsf P}_n}\Bigl|\sum_{j=1}^n \varepsilon_j\Bigr| = n2^{-n}\binom{n}{n/2} \sim \sqrt{\frac{2n}{\pi}}, \] $(\varepsilon_j)$ is a Rademacher family on $\Omega_n=\{-1,1\}^n$, and $\mathsf{P}_n$ is the uniform probability measure.
Read the full abstract
Representation Theory
Category $\mcal O$ for polynomial toroidal algebras and its subalgebras
In this paper we study Category $\mcal O$ for the polynomial toroidal Lie algebras and its $S,H$ type subalgebras. We classify irreducible objects of category $\mcal O$ as unique irreducble quotient of standard modules. Surprisingly, costandard objects of category $\mcal O$ arrises from Shen-Larsson type modules. We determine necessary sufficient conditions for irreducibility of Shen-Larsson modules. Finally appeling structure of Shen-Larsson modules and Soergel Tilting module theory of \cite{Soe}, we compute charcter formulas for irreducible modules and indecomposable Tilting modules of category $\mcal O$.
Read the full abstract
Optimization and Control
Stabilizability of first-order dynamics in second-order systems
We study whether second-order systems can be made to behave like prescribed first-order dynamical systems through feedback control. More precisely, we study whether prescribed vector fields on compact smooth manifolds, viewed geometrically as sections of the tangent bundle, can be asymptotically stabilized in a strong sense by second-order control systems on the base manifold. Our class of second-order systems includes most Lagrangian systems, and we obtain both positive and negative results. The positive result asserts that, for fully actuated systems, the section corresponding to any smooth vector field can be made globally exponentially stable, normally hyperbolic, and more. In particular, not only does each closed-loop solution asymptotically have the prescribed velocities, but it also converges to a trajectory of the first-order dynamics generated by the prescribed vector field at an exponential rate. Thus, the closed-loop second-order system asymptotically reproduces the prescribed first-order dynamics. In contrast, the negative result asserts that, for underactuated systems on manifolds with nonzero Euler characteristic, sections corresponding to "almost all" smooth vector fields cannot even be locally asymptotically stabilized. This includes, in particular, all vector fields with only isolated zeros. An example shows that the Euler characteristic assumption is necessary for the negative result.
Read the full abstract
Quantum Algebra
Finite Pre-Tensor Categories that are Morita Equivalent to Finite Tensor Categories
A finite pre-tensor category is a finite abelian category equipped with a right exact tensor product for which every projective object has duals. Finite tensor categories, for which every object has duals, are notable examples. More generally, the category of bimodules over an algebra in a finite tensor category is a finite pre-tensor category. In particular, it is natural to extend the notion of Morita equivalence between finite tensor categories to finite pre-tensor categories. We characterize completely those finite pre-tensor categories that are Morita equivalent to finite tensor categories. More precisely, we show that a finite pre-tensor category $\mathcal{C}$ is Morita equivalent to a finite tensor category if and only if the Drinfeld center of $\mathcal{C}$ is a finite tensor category. We also discuss higher algebraic consequences of our characterization.
Read the full abstract
Information Theory
Entropy-Rate Selection for Partially Observed Processes
I formulate an entropy-rate maximization problem at the observable level for stochastic processes observed through an information-reducing observation map. For a visible stationary law, the map determines an observational fiber of hidden stationary laws generating that law. In the finite-state finite-memory setting, retained visible constraints determine a feasible class of stationary $(r+1)$-block laws, and the entropy maximizer is defined as the entropy-rate maximizer on this class. The paper formulates entropy-rate maximization on feasible classes induced by partial observability and develops a structural theory for the resulting maximizer. I prove existence and uniqueness of the maximizer, with uniqueness under a fixed-context-marginal hypothesis and, more generally, via a strict-concavity characterization by row proportionality. Two global characterization regimes are central: a fixed one-point marginal yields the i.i.d. maximizer, and a fixed $r$-block law yields the $(r-1)$-step Markov extension. The gap functional equals a conditional mutual information and vanishes exactly at the maximizing completion. I also derive optimality conditions, local geometry of the maximizer, a latent random-mapping realization that leaves the visible law unchanged, and a local empirical consistency theorem, and illustrate the framework by an aliased hidden-state example.
Read the full abstract
Analysis of PDEs
On the transfer of stability from the local to the fractional anisotropic Calderón problem with exterior measurements
We study the quantitative transfer of uniqueness from the classical to the fractional Calderón problem with exterior data. This allows us to deduce the first stability estimates for the principal part of the isotropic fractional Calderón problem with exterior data in the absence of Liouville transforms. Our argument relies on careful quantitative unique continuation and Runge approximation estimates. Due to the unbounded geometry and the mismatch of the dimensionalities of the measurement domains (exterior data on an open set vs boundary data on a co-dimension one manifold) novel challenges arise compared to the setting of source-to-solution measurements on closed manifolds.
Read the full abstract
Dynamical Systems · Cross-listed from physics.soc-ph
Epidemic Transmission Modelling on the Birth-death Evolving Network with Indirect Contacts
Epidemic modelling on complex networks has been studied intensively all the time. The majority of relative research assumes that the time scale of the underlying network evolution is much larger compared to the propagation dynamics on it, while the co-evolution of epidemics and networks needs exploring further. In this paper, we investigate how our recently proposed birth-death evolving network impacts the Susceptible-Infected-Recovered-Susceptible (SIRS) epidemic process. Our evolving network considers the increase and the heritable deletion of nodes, which enables to depicting individual behaviors during an epidemic, e.g., population migration and indirect contacts. To model the above processes, we construct a Markovian queueing network and perform analyses for the variation of population size of different epidemic states. In simulations, we reveal how the population migration and indirect contacts caused by our network dynamic properties influence the population sizes of each epidemic state, and find that newly-created indirect contacts facilitate epidemic transmission.
Read the full abstract
Analysis of PDEs
Rate of convergence of a nonlinear heat equation with a constraint of codimension one
We consider a nonlinear constrained heat flow evolving on the manifold $\mathcal{M}=\{v\in L^{2}:\|v\|_{L^{2}}=1\}$ over bounded smooth domains. It is known that the solution corresponding to any nonnegative initial datum remains on $\mathcal{M}$ and converges to the unique positive ground state of the associated stationary problem. In this work, we first establish certain time-regularity estimates and then use these to derive explicit exponential rates of convergence for the energy, the solution in the $L^2, H^1$ and $H^2-$norms, and the associated nonlinear eigenvalue, thereby proving a sharp exponential stability of the ground state. Moreover, using the Łojasiewicz-Simon inequality, we obtain decay rates for locally stabilized solutions toward a stationary state in the $L^2$ and $H^1-$norms, where the rate depends on the corresponding Łojasiewicz-Simon exponent. Our results are new, and the approach relies on spectral analysis of the linearized operator, uniform higher-order estimates, and the compactness of solution trajectories.
Read the full abstract
Optimization and Control
Last Iterate Convergence of AdaGrad-Norm for Convex Non-Smooth Optimization
We study the convergence of the last iterate (i.e., the $(N+1)$-th iterate) of the AdaGrad method. Although AdaGrad – an adaptive subgradient method – underpins a wide class of algorithms, most existing convergence analyses focus on averaged (or best) iterates. We derive worst-case upper bounds on the suboptimality of the final point and show that, with an optimally tuned stepsize parameter, the last iterate converges at the rate $O(1/N^{1/4})$. We complement this guarantee with matching lower-bound constructions, proving that this rate is tight and that AdaGrad's last-iterate rate is strictly worse than the classical $O(1/N^{1/2})$ rate for its averaged iterate. Technically, our analysis introduces an exponent parameter that captures the growth of the cumulative squared subgradients; combined with the last-iterate inequality of Zamani and Glineur (2025), this reduces the problem to bounding a particular series.
Read the full abstract
Probability · Cross-listed from stat.ML
Tail-Aware Information-Theoretic Bounds for LLM Alignment under Heavy-Tailed Rewards
Classical information-theoretic learning bounds typically rely on KL mutual information and moment-generating-function (MGF) arguments, which are well matched to bounded or sub-Gaussian losses but can be ineffective when losses or rewards are heavy-tailed. We develop a tail-aware information-theoretic framework for sub-Weibull data, where the tail parameter $\theta$ controls the tail heaviness: $\theta=2$ corresponds to sub-Gaussian, $\theta=1$ to sub-exponential, and $0<\theta<1$ to genuinely heavy tails. Our key technical ingredient is a decorrelation lemma that bounds change-of-measure expectations using a shifted-log $f_\theta$-divergence, which admits explicit comparisons to Rényi divergence without MGF arguments. On the empirical-process side, we establish sharp maximal inequalities and a Dudley-type chaining bound for sub-Weibull processes, with logarithmic and entropy terms raised to the power $1/\theta$. These tools yield tail-adaptive selection bounds and a multiscale information-theoretic Dudley inequality based on shifted-log and Rényi mutual information. We apply our theory to large language models (LLMs) in the context of reward hacking within reinforcement learning from human feedback (RLHF). We show that Rényi-regularized alignment provides finite reward guarantees and ensures that best-of-N policies remain well-controlled, thereby mitigating the catastrophic Goodhart effects where standard KL-regularization fails. We illustrate Rényi-regularized RLHF by experiments, including controlled heavy-tailed rewards and token-space reward attacks.
Read the full abstract
Combinatorics
Black-white polynomials of graphs and generating functions
Let G be a graph. The black-white polynomial W_G(t) enumerates colorings of the vertices of G with two colors (black and white), where the power of t keeps track of how many white vertices have an even number of black neighbors. Such polynomials appear in quantum information theory, where they are used to capture properties of the entanglement in certain quantum states described by graphs. In this paper we describe how to use generating functions to compute these polynomials for various families X of graphs. Our main results are the following: (i) we describe some constructions under which X leads to a rational generating function; (ii) we use a matrix model to construct the exponential generating function of the black-white polynomials of all graphs; and (iii) we generalize a construction of Wright to build exponential generating functions of black-white polynomials for graphs of a given loop number.
Read the full abstract
Optimization and Control
A Hierarchical Robust Control Strategy for Stochastic Kuramoto–Sivashinsky–Korteweg–de Vries Equations
We investigate the robust Stackelberg null controllability of a one-dimensional forward linear stochastic Kuramoto–Sivashinsky–Korteweg–de Vries (KS–KdV) equation. The control framework is formulated as a hierarchical Stackelberg game involving two leaders, one follower, and worst-case disturbances acting in both the drift and diffusion terms. The first leader acts to drive the system to rest, while the second leader is introduced to overcome analytical difficulties arising from the stochastic setting. The follower, by reducing the effect of the disturbances, addresses a tracking-type control problem aimed at keeping the system state and its first and second spatial derivatives close to prescribed target trajectories. First, the robust control problem is characterized by the existence of a saddle point. Then, the analysis is reduced to the null controllability of a strongly coupled forward–backward stochastic KS–KdV system. The problem is addressed by combining a duality technique with new Carleman estimates for forward and backward stochastic fourth-order parabolic equations.
Read the full abstract
Logic
A note on iterating strongly $(<\lambda)$-closed stationary $\lambda^+$-cc forcing
We give an exposition of an iteration theorem for iterating $(<\lambda)$-closed stationary $\lambda^+$-cc forcing with supports of size $<\lambda$ and preserving these two properties. We discuss the relation of this theorem with other iteration theorems and forcing axioms that have appeared in the literature, notably the one from \cite{Sh80}.