Papers

We develop tools for the analysis and implementation of explicit first-order methods for minimizing functions in uniformly convex metric spaces. We formulate sufficient conditions for convergence of descent sequences in terms of directional subderivatives, function values and iterates under regularity assumptions including boundedness, geodesic smoothness and a metric Polyak-Łojasiewicz property. We show that there exists a steepest descent direction in which the assumptions for convergence are satisfied. This work provides a foundation for first-order explicit algorithms for locally smooth, nonconvex optimization.

We consider the problem of learning high-dimensional semi-global feedback controllers under hard safety constraints enforced by control barrier functions (CBFs). Incorporating CBFs into end-to-end policy training requires embedding a quadratic-program-based safety filter as an optimization layer, but computational and differentiation bottlenecks have largely restricted prior approaches to low-dimensional systems, typically with at most 16 state dimensions. We address this limitation by combining operator splitting with the recently developed Jacobian-Free Backpropagation (JFB) method to enable scalable end-to-end training while preserving hard safety guarantees through the CBF safety filter. We justify this training methodology theoretically using nonsmooth analysis techniques and demonstrate its effectiveness on high-dimensional multi-agent nonlinear control problems with state and control dimensions up to 1200 and 400, respectively.

Sudoku is a widely popular puzzle whose complete grids have been enumerated computationally: there are approximately $6.67 \times 10^{21}$ of them and, up to symmetry and renaming of digits, $5,472,730,538$ essentially different ones. The classical enumeration reduces the count to $44$ equivalence classes of the first band through a chain of ad hoc reductions, leaving the number $44$ without any apparent structural explanation. We present an alternative derivation of these $44$ classes, in which they arise as isomorphism classes of unordered triples (multisets) of column partitions under relabeling, a single invariant that replaces the original chain of reductions. This invariant makes it possible to apply Burnside's Lemma by hand: we recover $44$ through a closed derivation requiring no computational enumeration.

We find necessary and sufficient conditions for the rank of the signless incidence matrix of a weakly connected oriented graph with non-zero complex edge weights. We use this to find the combinatorial formulas for the Moore-Penrose inverse of the complex signless incidence and complex signless Laplacian matrix of a weakly connected oriented graph with non-zero complex edge weights. This resolves the open problem posed in the concluding remarks of Barik et. al. (Discrete Mathematics 349 (9), 115117, 2026).

Unique and rapid classification of knots and links is an open mathematical problem that is relevant to a range of (bio)physical systems, including polymer melts, DNA, and proteins. In this paper, we explore a data-driven approach to the classification problem of link topology. Extending the framework introduced in Ref. 1 (Sleiman et al, 2024 Soft Matter, 20(1), pp.71-78), we show that a feedforward neural network trained on the writhe density matrix classifies thermally equilibrated configurations of the first six prime links with 97% accuracy. We demonstrate that this accuracy remains high across a range of temperatures and lengths of link components, while rapidly deteriorating with the addition of topology-altering Gaussian noise; a result consistent with the writhe density matrix containing features sensitive to topology. Our results show that neural networks based on the writhe density matrix efficiently classify two-component links, establishing machine learning as a promising tool for rapid classification of more complex link topologies, e.g. Borromean rings and multi-component links, as the computational cost of exact numerical calculation of topological invariants becomes prohibitive.

This paper introduces planar web geometry into the study of characterization problems for probability laws arising from independence preservation. Let $F(x,y) = (u, v)$ be a real-analytic diffeomorphism defined on an open domain in $\mathbb{R}^2$. Let $X$ and $Y$ be independent random variables and set $(U, V) = F(X, Y)$. The map $F$ is said to preserve the independence if $U$ and $V$ are independent. This condition is highly restrictive and can therefore be used to characterize the laws for which it holds. This framework provides unified proofs of characterization theorems, including classical results such as the Kac–Bernstein and Lukacs theorems and the Matsumoto–Yor property, as well as recent results for the quadrirational Yang–Baxter maps $H_I^+$, $H_{II}^+$, and $H_{III}^A$. For measures with positive continuous densities, the map preserves independence if and only if their logarithmic densities solve the corresponding inhomogeneous Abelian functional equation. The solution space is identified with an affine space modeled on the space of Abelian relations, whose dimension is at most three according to Bol's rank bound. The same method also produces further examples, including a class generalizing transformations studied by Koudou and Vallois; every map in this class admits at least a two-parameter family of preserved product measures. The independence-preserving property of a map is preserved by coordinatewise one-variable reparametrizations, which also preserve the associated web. Thus the web associated with a map is a natural invariant under the corresponding equivalence relation of independence-preserving maps. Together with the local algebraizability of nonsingular planar $4$-webs of maximal rank, this correspondence suggests a route toward a complete classification of maps preserving the independence of a three-parameter family of measures.

Since manuscript \cite{R}, recently posted on arXiv, is closely related to our previous work, we would like to clarify some facts about it. Namely, we would like to point out that several results in \cite{R} are already known (Theorem 1 and Corollaries 2, 3 and 4) and some observations regarding our work \cite{FJ} are not correct. We also notice that Rédei's Theorem, a classical result in graph theory, gives a simple tool to show Theorems 5 and 9.

We construct the sine-Gordon measure and its renormalized effective potential in finite volume up to the seventh threshold. The construction relies on a weak stochastic control problem, its associated weak FBSDE, and an analysis of the renormalization flow equation using a multipole Mayer expansion. The analysis is fully inductive, and we include a complete order-by-order analysis in the subcritical regime $\beta^{2}<8\pi$.

We show that the self-correcting memory in 3d Wegner gauge theory is stable to arbitrary small enough perturbations of the Hamiltonian. Our proof relies on a new method we dub “gauge averaging”, which gives conditions under which explicitly broken gauge symmetries are effectively restored by fluctuations. These conditions show that the self-correcting memory phase is fluctuation stabilized, with its robustness to perturbations increasing with increasing temperature, up to some $T_c > 0$.

We show that the set $A(n, d)$ of polynomial automorphisms $F : \Bbb C^n \to \Bbb C^n$ of degree at most $d$ and with $Jac(F ) = 1$ is Zariski closed. In particular every irreducible component of the set $A(n,d)$ of polynomial mappings with Jacobian $1$ is either composed with polynomial automorphisms or (generically) with counterexamples to the Jacobian Conjecture. Moreover every such component has dimension at least $n^2-1.$ In particular if the set $X(n,d)$ is irreducible, and $n\ge 3, d\ge 7$, then a generic element of this set is a counterexample to the Jacobian Conjecture.

Let $X_{G_h}(\mathbf{x};q)=\sum_{\mu\vdash n}c_\mu(q)e_\mu(\mathbf{x})$ be the chromatic quasisymmetric function of the natural unit interval graph attached to a Hessenberg function $h$. We establish an infinite class, valid in all orders, for which every nonzero polynomial $c_\mu(q)$ has a nonnegative, log-concave coefficient sequence with interval support. Namely, this holds whenever $h$ is abelian and its complement-Ferrers partition $\lambda$ satisfies $\min\{\lambda_1,\ell(\lambda)\}\leq 3$; equivalently, the diagram has at most three rows or at most three columns. Cubic interpolation reduces the rank-three case to a uniform theorem for a difference of two products of four $q$-integers, proved by positive decomposition, interval methods, and finite-window smoothing. The argument also yields explicit formulas for every supported elementary coefficient in complement-Ferrers rank at most three. We also include a connected 13-vertex natural unit interval graph for which one elementary coefficient is positive, palindromic, and unimodal but not log-concave, thereby recording the failure of the unrestricted conjecture. Thus low complement-Ferrers rank gives a substantial positive regime even though coefficientwise $e$-log-concavity fails in general.

The Ulam–Kaplansky problem up to similarity in operator algebras asks whether $A\oplus A\sim B\oplus B$ implies $A\sim B$. We construct block-diagonal compact operators for which this implication fails. For von Neumann algebras with separable predual, we prove that this cancellation property is equivalent to subhomogeneity. Moreover, we obtain a cancellation law for finite ampliations in unital Banach algebras when the commutant of one element is finite dimensional modulo its Jacobson radical.

We study Gaussian-width complexity on statistical manifolds through a pair of functionals: the primal Fisher width $w_G(T) = w(G^{1/2}T)$, induced by the Fisher metric, and the inverse-Fisher width $w_{G^{-1}}(T) = w(G^{-1/2}T)$, induced by the inverse Fisher metric. The two widths play complementary statistical roles. On the learning side, the Fisher width measures the size of local parameter fluctuations in the geometry induced by the Fisher information. For Fisher-regular losses, we prove that the scale \(w_G(H_r)/\sqrt n\) is attained on sufficiently small Fisher balls. On the recovery side, the inverse-Fisher width captures the effect of anisotropic Gaussian measurements whose covariance is determined by the inverse Fisher information. For sparse recovery, the resulting geometry depends not only on sparsity but also on the position of the active coordinates in the Fisher spectrum. We obtain a two-sided estimate for the corresponding statistical dimension, together with support-sensitive recovery estimates and a natural ordering of supports with different curvature profiles. Finally, we establish a sharp relation between the primal and inverse-Fisher widths. On any common compact coordinate set $T$, they satisfy \[ w_G(T)w_{G^{-1}}(T)\geq w(T)^2. \] Thus, Fisher anisotropy may transfer complexity from one geometry to the other, but cannot reduce both widths relative to the Euclidean scale.

We study estimation of the p*p residual scatter (shape) matrix in a high-dimensional multivariate linear regression, where p and n grow proportionally. When the coefficient matrix obeys a known linear restriction of rank q < d, as in multivariate analysis of variance, growth-curve models, and reduced-rank regression, the restricted fit leaves additional residual degrees of freedom that sharpen estimation of the shape matrix. To accommodate heavy-tailed errors, we work with independent elliptically distributed rows under a mild scale condition, a finite second moment on the radii, which is far weaker than the usual sub-Gaussian assumptions and covers every multivariate-t law with more than two degrees of freedom. Shrinking the restricted residual sample covariance directly is unsound here, since its limiting spectrum depends on the radial distribution. We instead shrink a scale-invariant scatter of the restricted residuals, whose spectrum is distribution-free over the elliptical family and obeys the same limiting law as under Gaussian errors, at a smaller effective aspect ratio. The resulting estimator attains the rotation-equivariant oracle and is asymptotically optimal within that class, and a Stein-type combination with the unrestricted estimator dominates it while remaining safe under misspecification. We further correct for the case in which the restriction is itself selected from the data. Simulations, a growth-curve experiment, and two real-data analyses illustrate the results.

We show that for every fixed $\varepsilon>0$, there exist arbitrarily large families of point-line pairs $(x_1,\ell_1),\ldots,(x_n,\ell_n)$ in $[0,1]^2$, with $x_i \in \ell_i$ for all $i$, and such that $\operatorname{dist}(x_i,\ell_j)\ge n^{-2/3-\varepsilon}$ for all $i \neq j$. Combined with a previous result of Cohen, the author, and Zakharov, this solves the minimal distance problem. The same construction also comes with an unexpected finite field consequence: for every $\varepsilon>0$, there exists a set of primes $q$ of positive relative density for which $\mathbb F_q^2$ contains an induced point-line matching of size $\gtrsim q^{3/2-\varepsilon}$. This disproves a conjecture of Hunter, the author, Verstraëte and Zhang.

We study pinned nonempty-interior problems for scalar two-point configurations and for volumes of simplices. For $E\subset\mathbb{R}^d$, $d\geq 2$, compact and a smooth scalar configuration map $\Phi(x,y)$, whose corresponding localized generalized Radon transforms are nondegenerate Fourier integral operators of smoothing order $(d-1)/2$, we first note how a calculation due to Greenleaf, Iosevich and Taylor can be used to obtain positive Lebesgue measure of $\Delta_\Phi^y(E)=\{\Phi(x,y):x\in E\}$ for almost every pin $y$ when $\dim_{\mathcal H}(E)>(d+1)/2$. Our first main result is to prove that the corresponding one-frequency-loss estimate for differentiation in the level parameter yields a continuous pinned density, and hence nonempty interior, for almost every pin when $d\geq3$ and $\dim_{\mathcal H}(E)>(d+2)/2$. Concrete applications include generalized norm distances, regular variable-coefficient and Riemannian distances, and dot products or nondegenerate bilinear forms on regular patches. Our principal geometric application concerns volumes of simplices. We prove a cylinder-averaging estimate for triangle areas in $\mathbb{R}^d$ and obtain positive measure for doubly pinned area sets at a dimensional threshold $(d+1)/2$ and nonempty interior at $(d+2)/2$. A projection theorem then reduces higher simplex-volume problems to triangle areas. In particular, for $3\leq k \leq d$, if $\dim_{\mathcal H}(E)>(d+k-1)/2$, then for every prescribed base point $x_0$ and every prescribed second vertex $y\in E\setminus\{x_0\}$, the set of $k$-dimensional volumes generated by $x_0,y$ and $k-1$ further points of $E$ has nonempty interior. Thus the result is doubly strongly pinned in its first two vertices.

By Godunov's theorem, linear second-order finite-difference schemes for the Fokker-Planck equation cannot preserve positivity. The Diagonal Frog (DF) framework previously bypassed this barrier using eventual positivity, but required a strict minimum time step. This paper resolves the small-step limitation using a nonlinear extension of the DF solvers. We split the second-order directional operator into a monotone M-matrix core and an antidiffusive flux correction. A Zalesak-type limiter is then applied iteratively within the implicit banded solve. The resulting Flux-Corrected DF (FCDF) schemes (variants A and B) are unconditionally positive across all time steps. Because the limiter acts on fluxes rather than point values, these schemes conserve discrete mass exactly and maintain second-order accuracy. Crucially, the limiter activates only within unresolved layers. This ensures the global $L_1$ convergence remains second-order uniformly in the cell Péclet number, avoiding the first-order degradation seen in the Chang-Cooper scheme. The method's Picard iteration is contractive under a purely convective step restriction. To support arbitrary step sizes, we develop an active-set reformulation. This solves the system using a semismooth Newton iteration, where computational cost scales only with the number of nodes where positivity binds. Finally, we introduce a defect-corrected time stepping approach that restores second-order time accuracy. Numerical experiments on Ornstein-Uhlenbeck and advection-dominated benchmarks confirm our claims.

Recently, it has been reported that one-bit massive MIMO (mMIMO) detection is equivalent to a binary classification problem that can be solved efficiently using support vector machine (SVM). Inspired by this result, we first reformulate SVM in a decentralized form consisting of multiple classifiers. This enables the use of the consensus alternating direction method of multipliers (CADMM), a technique that can improve robustness and performance through its inherent consensus making. We further update CADMM to output only valid constellation points and achieve significantly improved detection performance. In our method, by changing the size of the grouped classifiers, we balance the number of classifiers for consensus accuracy with sufficient data per group to ensure classifier robustness. Ultimately, we demonstrate that our proposed method significantly outperforms existing practically feasible methods for one-bit mMIMO detection.

We prove strong laws of large numbers for random functions in the Lipschitz pseudometric. Our results hold under either a topological or a model-theoretic condition, with the latter encompassing functions jointly definable in o-minimal structures but extending substantially beyond this class. Applications include uniform convergence of limiting and Clarke subdifferentials and finite-sample identification of solutions. Consequently, we identify broad classes of functions for which the failure phenomena revealed by our previous negative results [Tian and Royset, arXiv:2511.16568, 2025] do not occur.

In this paper, we investigate extreme values of quadratic Dirichlet $L$-functions at the central point. We provide new extreme values of $L(\frac12,\chi_d)$ as $d$ is large, which improves the recent result of Darbar and Maiti.

In this note, we show that $C^*(G)$ fails the lifting property (LP) for every discrete group $G$ containing an uncountable abelian subgroup. Consequently, for every uncountable discrete abelian group $G$, the nuclear algebra $C^*(G)$ has the local lifting property (LLP) but fails the LP. This shows that the LP and the LLP do not coincide for full discrete group $C^*$-algebras.

The conjugator length function of a finitely generated group $G$ gives the minimal upper bound on the length of a conjugator for a pair of words that represent conjugate elements in $G$, as a function of the sum of the lengths of the words. Here, we seek to promote the systematic study of conjugator length functions by explaining their significance, by surveying what is known about them and by explaining fundamental techniques and examples.

We give a derivation of the Gibbs measure for the focusing nonlinear Schrödinger equation (NLS) on the circle with rough cut-off. This extends earlier work by Sohinger and the second author, which proved analogous results for smooth cut-offs. Our proof is based on the perturbative expansion developed by Fröhlich, Knowles, Schlein, and Sohinger (2017), and provides an alternative proof of the recent derivation given in Lü, Nam, and Zhu (2026). To prove convergence of the explicit terms, we employ a Wigner measure approach and an inductive argument to overcome the lack of smoothness for the cut-off. In particular, we give a derivation of the Gibbs measure for the focusing quintic NLS with the optimal cut-off from Oh, Sosoe, and Tolomeo (2022).

We classify all automorphism groups of smooth, projective, connected curves over an algebraically closed field of characteristic $0$ with simple Jacobians. For every nontrivial group in this classification, we construct a family of curves with that automorphism group and prove that a density-one set of members has simple Jacobian. These families yield infinitely many new examples of unlikely intersections in the moduli spaces of principally polarized abelian varieties.

We investigate a Hele-Shaw type free boundary problem in general spatial dimension. We establish a comparison principle and well-posedness for the problem using a notion of viscosity flows. An important feature of our work is that our hypotheses allow for the free boundary velocity to take negative values.

We prove that the mixed volume of a convex body of fixed positive volume with its reflection about the origin is maximized by simplices. This confirms a conjecture of C. Godbersen from 1938 and refines the classical Rogers-Shephard inequality. We also prove that simplices are the only extremizers among convex polytopes. Finally, we use this inequality to prove an $L_p$-version of the Rogers-Shephard inequality for convex bodies containing the origin and show that, for any $p\in(1,\infty]$, the only extremizers are simplices with a vertex at the origin.

For a connected graph $G$ and a set $S\subset V(G)$, the Steiner distance $d_G(S)$ is the minimum number of edges in a connected subgraph of $G$ containing $S$. The Steiner-Wiener $k$ index is defined by $\mathrm{SW}_k(G) = \sum_{S\subset V(G), |S|=k} d_G(S)$. We study the inverse problem for this invariant restricted to trees: for fixed $k$, which positive integers occur as $\mathrm{SW}_k(T)$ for a finite tree $T$? We prove that all sufficiently large positive integers occur as $\mathrm{SW}_k(T)$ for some finite tree $T$ if and only if $k$ is even. For odd $k$, we further show that the set of attainable values has asymptotic density of order $k^{-\delta}(\log k)^{-3/2}$, where $\delta$ is the Erdős-Tenenbaum-Ford constant.

Physics-informed learning of partial differential equations (PDEs) has been dominated by multilayer perceptrons (MLPs), whose spectral bias and dense parameterization limit both accuracy and interpretability. Kolmogorov Arnold Networks (KANs) mitigate these limitations because their learnable spline activations are structurally aligned with the piecewise-polynomial bases of classical discretizations. However, the way a PDE is cast into a loss functional is as decisive as the choice of approximator: strong-form residual minimization requires high-order derivatives and heavily weighted losses, the energy (Bubnov-Galerkin) form is restricted to self-adjoint operators and, as we show, collapses to a trivial solution for parameter-identification problems, and boundary integral forms require a known fundamental solution. We propose PG-KINN, a physics-informed KAN built on a Petrov-Galerkin formulation in which the trial space is a KAN and the test space is an independent, compactly supported, piecewise-polynomial space evaluated with Gauss-Legendre quadrature. Integration by parts lowers the differentiation order while retaining applicability to general non-self-adjoint, nonlinear, and inverse problems; the localized test functions turn the global residual into a set of element-wise weak residuals with favorable conditioning. On a suite of benchmarks spanning crack singularities, stress concentration, Neo-Hookean hyperelasticity, inverse parameter identification in heterogeneous media, and complex geometries, PG-KINN consistently outperforms legacy MLP baselines and state-of-the-art KAN-based strong/energy/inverse formulations (PIKAN). These results position the Petrov-Galerkin coupling of KAN trial spaces and polynomial test spaces as a robust and accurate route for AI-based computational mechanics.

Let $s, t\ge2 $ be integers. A graph $G$ is $(s,t)$-splittable if $V(G)$ can be partitioned into two sets $S$ and $T$ such that $\chi(G[S ]) \ge s$ and $\chi(G[T ]) \ge t$. The Erdős-Lovász Tihany Conjecture from 1968 asserts that every graph $G$ satisfying $\omega(G)<\chi(G)=s+t-1$ is $(s,t)$-splittable. A vertex of a graph is bisimplicial if the set of its neighbors can be expressed as the union of two cliques. Let $G$ be a graph with $\omega(G)<\chi(G)=s+t-1$. We prove that if $G$ does not contain $C_4$ as an induced subgraph and every induced subgraph of $G$ has a bisimplicial vertex, then $G$ is $(s,t)$-splittable. Combining our result with a recent result of Chudnovsky and Seymour, which states that every non-empty even-hole-free graph has a bisimplicial vertex, we obtain that the Erdős-Lovász Tihany Conjecture holds for all even-hole-free graphs.

Posterior probabilities are bounded martingales, but this fact alone does not identify the experiment that generates them. We show that every time-homogeneous martingale diffusion on $[0,1]$ with strictly positive $C^1$ volatility on the interior and absorbing endpoints, which we call an autonomous win-martingale, can be realized as the Bayesian posterior of an explicit binary sequential experiment observed through a scalar diffusion. Conversely, in a general binary diffusion experiment, the current time and observation are jointly sufficient for the hidden state precisely when the difference between the signal drifts under the two hypotheses satisfies a Riccati equation. Equivalently, this holds when the likelihood ratio is generated by a Doob $h$-transform associated with a positive space-time harmonic function for the null dynamics. Within this class, when that function is spatially harmonic, the posterior is an autonomous martingale. Outside this harmonic case, autonomy forces the drift difference to be constant, recovering the classical two-state Shiryaev-Wonham filter from binary testing. These results give concrete sequential inference interpretations to a broad class of bounded martingale diffusions, including absorbed Brownian motion, the Wright-Fisher diffusion, and the Aldous martingale of the "most exciting game."