We introduce a natural bilinear fractional integral type operator induced by a third order hypermetric on Ahlfors regular quasi-metric spaces. Given a quasi-metric space $(X,d)$ the function $ρ(x,y,z)$, defined as the distance, in $X^3$, of $(x,y,z)$ to the diagonal $\bigtriangleup_3=\{(x,x,x)\in X^3:x\in X\}$ is said to be a third order hypermetric in $X$. When $(X,d)$ is a Euclidean space or, more generally, when $(X,d,μ)$ is $η$-Ahlfors regular for some $η$ positive, the function $ρ(x,y,z)$ generates kernels for bilinear operators of the type $T^γ(f,g)(x)=\iint_{X\times X}ρ(x,y,z)^{-γ}f(y)g(z)dμ(y)dμ(z)$, for a given positive $γ$. In the setting of $η$-Ahlfors regular space, the power $-γ=-2η$ of $ρ(x,\cdot,\cdot)$ provides the natural singularity for this family of kernels. In this paper we consider the fractional integral rank $0<γ<2η$. We prove boundedness properties of the type $\|T^γ(f,g)\|_{p_3}\leq C\|f\|_{p_1}\|g\|_{p_2}$ for adequate values of the exponents $p_1,p_2$ and $p_3$. The proof is based on three upper bounds for $T^γ(f,g)$ in terms of the classical linear fractional Riesz operators $I_{η-\fracγ{2}}$, using the linear Hardy-Littlewood-Sobolev inequality.

We analyze greedy routing in a random graph G_n constructed on the vertex set V = {1, 2, ..., n} embedded in Z. Vertices are inserted according to a uniform random permutation pi, and each newly inserted vertex connects to its nearest already-inserted neighbors on the left and right (if they exist). This work addresses a conjecture originating from empirical studies (Ponomarenko et al., 2011; Malkov et al., 2012), which observed through simulations that greedy search in sequentially grown graphs exhibits logarithmic routing complexity across various dimensions. While the original claim was based on experiments and geometric intuition, a rigorous mathematical foundation remained open. Here, we formalize and resolve this conjecture for the one-dimensional case. For a greedy walk GW starting at vertex 1 targeting vertex n -- which at each step moves to the neighbor closest to n -- we prove that the number of steps S_n required to reach n satisfies S_n = Theta(log n) with high probability. Precisely, S_n = L_n + R_n - 2, where L_n and R_n are the numbers of left-to-right and right-to-left minima in the insertion-time permutation. Consequently, E[S_n] = 2H_n - 2 ~ 2 log n and P(S_n >= (2+c) log n) <= n^(-h(c/2) + o(1)) for any constant c > 0, with an analogous lower tail bound for 0 < c < 2, where h(u) = (1+u) ln(1+u) - u is the Bennett rate function. Furthermore, we establish that this logarithmic scaling is robust: for arbitrary or uniformly random start-target pairs, the expected routing complexity remains E[S_{s,t}] = 2 log n + O(1), closely mirroring decentralized routing scenarios in real-world networks where endpoints are chosen dynamically rather than fixed a priori.

We consider the dissipative generalized Surface Quasi-Geostrophic equation with dissipation given by any fractional power of the Laplacian. In the inviscid limit, it is proved that anomalous dissipation of the Hamiltonian is prevented by the strong compactness of the solutions in the lowest norm that makes the nonlinearity well-defined. In fact, only the dynamics at certain frequencies matters. The argument is quite robust as it applies regardless of the criticality regime and of the presence of a, possibly noncompact, external forcing. This reveals a more general mechanism behind some recent results obtained for the Navier-Stokes and the critical dissipative Surface Quasi-Geostrophic equations. Because of nonuniqueness issues, in our broader context it is important to work with Leray solutions enjoying suitable higher-order bounds. The existence of such solutions is shown and it might be of independent interest. Finally, we prove that the strong compactness is guaranteed for any initial datum with critical integrability, from which global existence of conservative, although Onsager's supercritical, weak solutions of the inviscid problem is deduced. This offers the largest class of initial data for which global existence is known so far, matching with the one considered by Delort at the endpoint.

Personalized Federated Learning (PFL) aims to learn multiple task-specific models rather than a single global model across heterogeneous data distributions. Existing PFL approaches typically rely on iterative optimization-such as model update trajectories-to cluster users that need to accomplish the same tasks together. However, these learning-dynamics-based methods are inherently vulnerable to low-quality data and noisy labels, as corrupted updates distort clustering decisions and degrade personalization performance. To tackle this, we propose FB-NLL, a feature-centric framework that decouples user clustering from iterative training dynamics. By exploiting the intrinsic heterogeneity of local feature spaces, FB-NLL characterizes each user through the spectral structure of the covariances of their feature representations and leverages subspace similarity to identify task-consistent user groupings. This geometry-aware clustering is label-agnostic and is performed in a one-shot manner prior to training, significantly reducing communication overhead and computational costs compared to iterative baselines. Complementing this, we introduce a feature-consistency-based detection and correction strategy to address noisy labels within clusters. By leveraging directional alignment in the learned feature space and assigning labels based on class-specific feature subspaces, our method mitigates corrupted supervision without requiring estimation of stochastic noise transition matrices. In addition, FB-NLL is model-independent and integrates seamlessly with existing noise-robust training techniques. Extensive experiments across diverse datasets and noise regimes demonstrate that our framework consistently outperforms state-of-the-art baselines in terms of average accuracy and performance stability.

For a graph $G$, let $f_o(G)$ denote the maximum order of an induced subgraph of $G$ all of whose vertices have odd degree, and let $χ(G)$ denote the chromatic number of $G$. Scott (CPC, 1992) proved that $f_o(G) \ge |V(G)|/(2χ(G))$ for every graph without isolated vertices, and conjectured that the factor $2$ can be removed. Wang and Wu (JGT, 2024) showed that this conjecture fails for bipartite graphs, but holds for line graphs. In this article, we confirm Scott's conjecture for claw-free graphs without isolated vertices, thereby strengthening the result of Wang and Wu. We also construct $K_{1,r}$-free graphs of arbitrarily large order to show that the conjecture fails for this broader class, for every integer $r \ge 4$. Wang and Wu also asked whether $f_o(L(G)) \ge n/2$ holds for every connected regular graph $G$ of order $n \ge 3$. We show that $C_5$ is the smallest counterexample to this problem. On the positive side, we prove that if $G$ is a connected $k$-regular $C_5$-free graph on $n$ vertices with $k \ge 2$, then $f_o(L(G)) \ge n/2$.

Juniper Green is a simple combinatorial game invented by Rob Porteous and popularized by Ian Stewart. It was originally designed to familiarize school children with the concepts of multiplication and division. We analyze this elementary game through a completely different lens and show that it recovers the Gallai-Edmonds decomposition of the divisibility graph on the vertex set $V= \left\{1,2,\dots, n\right\}$. This characterizes the winning moves of the game; as a byproduct, we show that this decomposition seems to have many interesting and curious patterns that are currently unexplained.

This work proposes a bootstrapping with positivity methodology to study random $U(N)^{D}$ invariant tensors in the large $N$ limit. As has been done for $U(N)$ invariant random matrices, we combine the Dyson-Schwinger equations and positivity constraints of moments to approximate the moments of such tensor models. As examples, we bootstrap the quartic and two hexic rank three tensor models. All models studied converge quickly, and for those which have known analytic formulae, they converge to such solutions. We conjecture new explicit formulae for all moments of the rank three quartic model and support this conjecture using bootstrapped results and explicit double-series computations with 'feyntensor'.

In [97,99,100], an fl-RDT framework is introduced to characterize \emph{statistical computational gaps} (SCGs). Studying \emph{symmetric binary perceptrons} (SBPs), [100] obtained an \emph{algorithmic} threshold estimate $α_a\approx α_c^{(7)}\approx 1.6093$ at the 7th lifting level (for $κ=1$ margin), closely approaching $1.58$ local entropy (LE) prediction [18]. In this paper, we further connect parametric RDT to overlap gap properties (OGPs), another key geometric feature of the solution space. Specifically, for any positive integer $s$, we consider $s$-level ultrametric OGPs ($ult_s$-OGPs) and rigorously upper-bound the associated constraint densities $α_{ult_s}$. To achieve this, we develop an analytical union-bounding program consisting of combinatorial and probabilistic components. By casting the combinatorial part as a convex problem and the probabilistic part as a nested integration, we conduct numerical evaluations and obtain that the tightest bounds at the first two levels, $\barα_{ult_1} \approx 1.6578$ and $\barα_{ult_2} \approx 1.6219$, closely approach the 3rd and 4th lifting level parametric RDT estimates, $α_c^{(3)} \approx 1.6576$ and $α_c^{(4)} \approx 1.6218$. We also observe excellent agreement across other key parameters, including overlap values and the relative sizes of ultrametric clusters. Based on these observations, we propose several conjectures linking $ult$-OGP and parametric RDT. Specifically, we conjecture that algorithmic threshold $α_a=\lim_{s\rightarrow\infty} α_{ult_s} = \lim_{s\rightarrow\infty} \barα{ult_s} = \lim_{r\rightarrow\infty} α_{c}^{(r)}$, and $α_{ult_s} \leq α_{c}^{(s+2)}$ (with possible equality for some (maybe even all) $s$). Finally, we discuss the potential existence of a full isomorphism connecting all key parameters of $ult$-OGP and parametric RDT.

This paper considers a networked tracking architecture in 6G integrated sensing and communication (ISAC) systems, where multiple base stations (BSs) cooperatively transmit radio signals and process received echo signals to track multiple moving targets. Compared to the single-BS counterpart, networked tracking allows the moving targets to be associated with different BSs over time such that the wireless resources can be dynamically allocated among BSs based on target locations. However, networked tracking imposes new challenges for algorithm design and resource allocation. In this paper, we first design the networked Kalman Filter (NKF) that is suitable for multi-BS based tracking, then characterize the posterior Cramer-Rao bound (PCRB) under this NKF, and last design the beamforming vectors of all the BSs to minimize the tracking PCRB. Numerical results show that our dynamic beamforming design can properly associate the targets to the suitable BSs at various sensing blocks and reduce the tracking mean-squared error (MSE).

We study the local Lipschitz one subsets of a finite dimensional space, that is, sets for which there exists a continuous function whose local Lipschitz derivative is the characteristic function of said set. We give a characterization of a local Lipschitz one set on the real line in terms of a certain measure-theoretic density condition, which we call quasi-density. We show that any local Lipschitz one set needs to be quasi-dense, but the converse does not hold. Finally, we show that any regular closed subset of a normed space is a local Lipschitz one set, but there exist local Lipschitz one sets that are not regular closed.

The lowest eigenstates of the hopping matrix on the line graph of a cubic lattice with periodic boundary conditions are highly degenerate, they form a lowest flat band. Further, these states are localized. If one considers a repulsive bosonic Hubbard model on this lattice it is possible to construct exact multi-particle ground states simply by putting particles in the localized single particle ground states such that they avoid each other. This can be done up to a certain critical particle number $N_c$. We prove that at this particle number the ground state entropy is subextensive $\propto N_c^{2/3}$. For lower densities the entropy is extensive. We further show that the problem is related to the number of 4-cycle decompositions of the cubic lattice with periodic boundary conditions.

We study null controllability for the parabolic equation on $\mathbb{S}^{2}$ endowed with its canonical almost-Riemannian structure. For a spherical crown $ω=\{α<x_3<β\}$, where $0\le α<β\le1$, we prove the sharp minimal time formula $T_{\min}(ω)=\ln(1/\sqrt{1-α^{2}})$ for null controllability in $ω$. We also prove that, whenever the control region contains the equator, null controllability holds in every positive time. The proof combines two complementary tools. First, after Fourier decomposition with respect to the periodic variable, we establish observability estimates for a family of one-dimensional singular parabolic equations, with constants uniform with respect to the Fourier mode; the singularities at the poles are handled via a Hardy-Poincaré inequality. Second, for crowns away from the equator, we use the moment method to construct controls on the pole-touching crown $α<x_3< 1$ from sharp weighted lower bounds on associated Legendre functions, and then pass to a general crown $α<x_3<β$ by a cut-off argument on the full domain combined with the arbitrary-time controllability of crowns containing the equator. The result closes the large-time gap left in earlier work and gives the exact null-controllability threshold for the canonical almost-Riemannian heat equation on $\mathbb S^2$.

The standard Monte Carlo estimator $\widehat{I}_N^{\mathrm{MC}}$ of $\int fdω$ relies on independent samples from $ω$ and has variance of order $1/N$. Replacing the samples with a determinantal point process (DPP), a repulsive distribution, makes the estimator consistent, with variance rates that depend on how the DPP is adapted to $f$ and $ω$. We examine two existing DPP-based estimators: one by Bardenet & Hardy (2020) with a rate of $\mathcal{O}(N^{-(1+1/d)})$ for smooth $f$, but relying on a fixed DPP. The other, by Ermakov & Zolotukhin (1960), is unbiased with rate of order $1/N$, like Monte Carlo, but its DPP is tailored to $f$. We revisit these estimators, generalize them to continuous settings, and provide sampling algorithms.

This paper explores a version of the classical Ces`aro integral operator for the Lebesgue space L2(0, 1) where we discuss its norm, adjoint, spectral properties, and invariant subspaces. An important tool will be semigroups of weighted composition operators on L2(0, 1).

We implement methods from the geometry of numbers to give explicit estimates for the number of integral ideals in a number field. We pay particular attention to minimising the effect of the degree $n$ of the number field on the error term and avoid terms on the order of $n^{n^2}$. We do this by studying fundamental domains for the action of multiplying with units of infinite order in Minkowski space. With some lattice theory we show that one can make different choices for such a fundamental domain, which yield a smaller error, especially when the degree of the field extension is large. We also adapt Schmidt's partition trick to this generalised setting.

This paper develops a conceptual extension of the Kinetic Theory of Active Particles, building upon the framework introduced in [2]. Living systems cannot be adequately described within classical single-scale paradigms, even when refined. To overcome this limitation, we introduce a Multiscale Kinetic Theory of Active Particles (MS-KTAP), in which a sub-microscopic scale of interacting entities is incorporated into the description of collective dynamics. In this framework, the activity variable is interpreted as an emergent quantity arising from lower-scale regulatory mechanisms and influenced by interactions across higher scales. The proposed framework captures key features of living systems, including heterogeneity, adaptive decision-making, nonlinear and non-conservative interactions, spatial dynamics, and cross-scale feedback, within a unified mathematical structure. Competition and cooperation are thus described across multiple levels of organization. The first part of the paper derives the mathematical framework, while the second presents how specific models can be obtained. The paper concludes with perspectives on further developments, including possible integrations with scientific machine learning.

Starting from pseudometrics and preorders on sets of integers, we extend the focus to sets of finite sequences of integers, in particular sequences of consecutive integers. We outline existing concepts for deriving centred pseudometrics and preorders in a given pseudometric space and their application to $\mathbb{Z}$ and develop approaches to generalize the ideas to $\mathbb{Z}^m$. Sequences of consecutive integers represent a special case here and are examined in more detail. Another main topic is the use of arithmetic functions in this context. The types of pseudometrics and preorders examined in this paper can be induced by suitable arithmetic functions. We derive fundamental conclusions about relationships between functions and preorders, as well as about equivalent and potentially distinct types of preorders.

Air Traffic Flow Management (ATFM) traffic regulations are being increasingly used as rising demand meets persistent workforce shortages. This operational strain has amplified a critical phenomenon that we call \emph{regulation cascading}: the compounding, non-linear interactions that occur when multiple regulations influence one another in unpredictable ways. As the number and complexity of regulations grow, cascading effects become more pronounced, undermining the network operator's ability to protect sectors reliably. To address this challenge, we introduce Regulation Zero 2, an updated sequential planning framework that natively operates in the regulation space, optimizing over ordered sequences of flow-level regulations that remain compatible as much as possible with existing slot-allocation systems such as CASA and RBS++. We equipped Regulation Zero 2 with new heuristics to render flow finding more efficient. At its core, the method employs a hierarchical Monte Carlo Tree Search (MCTS) that first samples congestion hotspots and then selects candidate regulations synthesized by a local proposal engine. Each proposal is evaluated by a fast First-Planned-First-Served (FPFS) allocator to estimate its reward, with these feedbacks guiding the subsequent MCTS exploration. Experiments on many pan-European summer-peak traffic days that Regulation Zero delivers promising and consistent performance. Compared to a flight-centric simulated-annealing and NSGA-II baselines, it achieves markedly higher objective improvements, while maintaining a tighter scope of impact on the network. Ablation studies also found that Regulation Cascading could reduce up to 50\% of potential effectiveness. RZ also preserves FPFS fairness and supports expert knowledge injection, offering a pragmatic and low-disruption pathway toward automation in operations.

Graphs of solutions to the minimal surface equation over simply connected domains with boundary values 0 can have at most exponential growth.

This paper develops a principled foundation for goal-oriented semantic communication for logical decision-making. Consider a setting where autonomous agents engage in collaborative perception. In such settings, the volume of sensory data and limited bandwidth often make transmission of raw observations infeasible, requiring intelligent selection of task-relevant information. Because these scenarios are safety-critical, the selection and decision processes must also be transparent and verifiable. To address this, we propose an explainable semantic communication framework grounded in a First-Order Logic (FOL) hierarchical representation of the world. We define semantic information, entropy, conditional entropy, and mutual information by assigning an inductive logical probability measure over semantic structures in the language. Based on these definitions, we formulate a goal-oriented semantic communication objective through semantic rate-distortion theory and, equivalently, through the semantic information bottleneck principle. In this framework, task rules are represented as goal-oriented states, defined as a layer over the world states to capture decision-relevant abstractions. The resulting principle selects evidence that is most informative about these states, aiming to transmit only those FOL clauses most critical for decision-making while preserving logical verifiability. We demonstrate the effectiveness of the approach in a deduction-based safe path-following task within an FOL-based urban environment simulator with multiple dynamic agents.

Classical work of Thurston and Gabai shows that finitely many taut sutured manifold hierarchies determine the Thurston norm of a compact oriented irreducible $3$-manifold with toroidal boundary. We give an explicit procedure to extract this information from such hierarchies. This is achieved via the maw dual graph construction, which can be incorporated into a general method for computing the Thurston norm of a manifold. As an application, we compute the Thurston norm of the exterior of all alternating and some nonalternating pretzel links with three components. Using these computations, we give a negative answer to a question of Baker--Taylor. Moreover, we show that if a nonseparating surface $S$ in a Haken manifold $M$ with toroidal boundary is disjoint from a boundary torus, then the class $[S] \in H_2(M,\partial M)$ does not lie in the interior of a top-dimensional cone of the Thurston norm. In particular, if two components $\ell_i$ and $\ell_j$ of a nonsplit link have zero linking number, then neither represents a class in an open top-dimensional cone of the Thurston norm ball of the link exterior.

We introduce and analyse a mathematical model describing the dynamics of particles generated by charge-exchange interactions. The model extends the well-established exchange-driven growth model, previously studied in several works, by allowing for particle densities defined on the entire integer lattice. Despite the many similarities between the two models, substantial differences arise both in their qualitative behaviour and in their mathematical analysis. Under suitable assumptions on the kernel in the collision operator, we establish global well-posedness in the class of nonnegative densities with finite first moment. Moreover, under a detailed balance condition, we investigate the structure of equilibria and analyse their stability by means of entropy methods.

Fractional PDEs involving the fractional Laplacian on bounded domains are challenging because of hypersingular nonlocal kernels, exterior Dirichlet constraints, reduced boundary regularity, and the high computational cost in high dimensions. To address these issues, we first adopt a spatially varying radius with directional distance-to-boundary information, which yields a geometry-adaptive three-part decomposition of the fractional Laplacian: singular near-field, regular interior far-field, and analytical exterior far-field contributions. Then we employ Gauss-Jacobi quadrature for the singular radial integral, Gauss quadrature for the regular interior radial integral, and Monte Carlo sampling for the angular variables. A feature-enhanced physics-informed neural network trial space is finally used to tackle the low-regularity behavior near the boundary. Through the above steps, we obtain a quadrature-enhanced Monte Carlo fractional physics-informed neural network (QE-MC-fPINN) method. Numerical experiments on fractional Poisson equations and time-dependent fractional PDEs show that, on the tested benchmarks, the proposed method outperforms two representative MC-fPINN discretizations in accuracy and convergence, especially for solutions with strong boundary singularities.

It is a long-standing open question to determine whether the Sierpiński carpet attains its conformal dimension or not. While this problem remains unresolved, we prove that Cartesian products $\mathbb{S}^k$, where $\mathbb{S}$ is the Sierpiński carpet and $k \geq 2$, do not attain their conformal dimension. Our approach is based on the Sobolev spaces and energy measures on $\mathbb{S}$ -- constructed by Shimizu, Kigami, and Murugan and Shimizu -- together with a certain singularity result of energy measures from the theory of analysis on fractals. This work formulates a general non-attainment result of conformal dimension for product metric spaces $X^k$ for $k \geq 2$ in terms of self-similarity and energy measures of the factor $X$. It applies, in particular, to the cases where $X$ is the Sierpiński carpet, the Sierpiński gasket, the Menger sponge, and the Laakso diamond.

We consider a reaction-diffusion equation in a one-dimensional space, where the diffusion coefficient changes sign from positive to negative and back to positive. The reaction term is bistable, with its interior zero located in the region where the diffusivity is negative. The model does not admit continuous wavefronts, i.e., continuous traveling waves that connect the steady states $0$ and $1$. We prove the existence of a family of shock wavefronts, that is, wavefronts with profiles exhibiting a jump discontinuity. We investigate the properties of these profiles and their propagation speeds. Finally, we apply the results to a recently proposed model describing the movement of a population composed of both isolated and grouped individuals.

In this paper, we theoretically and numerically study the minimizers for the Cahn-Hilliard energy with the Flory-Huggins potential under the strong anchoring condition, i.e., the Dirichlet boundary condition. We reveal bifurcation phenomena mediated by the boundary condition, the transition layer thickness, and the temperature of the system. Numerical simulations are also presented to approximate the minimizers of this energy by solving a gradient-flow equation, namely the Allen-Cahn equation constrained with strong anchoring conditions and random initial data. The effects of varying the transition layer thickness and temperature are presented to confirm the theoretical analysis.

In this paper, we study the right invariant metric $d_{H^{\infty}}$ on the analytic automorphism group $\rm{Aut}(\mathbb{D})$ of the unit open disk $\mathbb{D}$ induced by maximal modulus, that is, $d_{H^{\infty}}(φ, ψ)=\sup_{z\in\mathbb{D}}|φ(z)-ψ(z)|$ for any $φ, ψ\in \rm{Aut}(\mathbb{D})$. We give the explicit formula of the right invariant metric $d_{H^{\infty}}$ and characterize the almost regular Finsler geometric structure of $(\rm{Aut}(\mathbb{D}), d_{H^{\infty}})$.

We show that the class of twisted fractionally Calabi-Yau algebras of finite global dimension coincides with the stable endomorphism algebras of $d$-cluster tilting modules over $d$-representation-finite algebras. This is an application of our main result stating that an algebra $A$ of finite global dimension is twisted fractionally Calabi-Yau if and only if there exists $i$ such that the replicated algebra $A^{(i)}$ is a higher Auslander algebra if and only if there exist infinitely many $i$ such that $A^{(i)}$ is a higher Auslander algebra. This gives a new connection between the study of higher Auslander-Reiten theory and twisted fractionally Calabi-Yau algebras, and provides a new construction of large classes of higher Auslander algebras and higher representation-finite algebras. We give several applications such as an explicit characterisation of twisted $\frac{n}{2}$-Calabi-Yau algebras, and a triangle equivalence between the bounded derived category of a twisted fractionally Calabi-Yau algebra of finite global dimension and the $\mathbb{Z}$-graded stable module category of an associated higher preprojective algebra.

In this paper, we propose a novel machine learning method based on an adaptive tensor neural network subspace for solving quasiperiodic elliptic problems. To this end, we first provide a theoretical analysis of the associated quasiperiodic and periodic function spaces and establish regularity estimates for the quasiperiodic elliptic problems. In particular, under the Diophantine condition, we derive a suitable condition on the source term to guarantee the regularity of the solution, which provides a theoretical basis for the design of numerical schemes. An efficient numerical method is then designed by combining the projection method with tensor neural networks. Leveraging the special structure of tensor neural networks, high-dimensional integration can be performed directly and with high accuracy, without relying on Monte Carlo methods. Finally, several numerical experiments are presented to demonstrate the accuracy and efficiency of the proposed method.

The goal of machine learning is to find models that minimize prediction error on data that has not yet been seen. Its operational paradigm assumes access to a dataset $S$ and articulates a scheme for evaluating how well a given model performs on an arbitrary sample. The sample can be $S$ (in which case we speak of ``in-sample'' performance) or some entirely new $S'$ (in which case we speak of ``out-of-sample'' performance). Traditional analysis of generalization assumes that both in- and out-of-sample data are i.i.d.\ draws from an infinite population. However, these probabilistic assumptions cannot be verified even in principle. This paper presents an alternative view of generalization through the lens of sensitivity analysis of solutions of optimization problems to perturbations in the problem data. Under this framework, generalization bounds are obtained by purely deterministic means and take the form of variational principles that relate in-sample and out-of-sample evaluations through an error term that quantifies how close out-of-sample data are to in-sample data. Statistical assumptions can then be used \textit{ex post} to characterize the situations when this error term is small (either on average or with high probability).