General Mathematics
Papers
Fresh from arXiv, typeset properly. Highlighted words trade on the market.
Friday 10 April
Using the log-convexity of the Gamma function and Euler's reflection formula, we give a new proof of a classical weighted sine product inequality. Two different parameter choices yield two competing upper bounds for the same product. We determine precisely, via algebraic criteria, when one bound is sharper than the other. Explicit results are given for the general $n$-angle case, the $2n$-angle case, and for two and three angles. Several sharp corollaries are derived, including $\sin(\pi x)\leq \sin(2\pi x(1-x))$.
History and Overview of Mathematics
Reminiscences of S. K. Godunov. The Russian Mathematician
These personal reminiscences of the great Russian mathematician Sergey K. Godunov (1929-2023) arose from a request by his daughter, Ekaterina, to contribute a piece to a book she is writing about her father's life. I was honoured to accept this invitation and to give written form to the rewarding experience of conducting research on themes pioneered by Professor Godunov, interacting with him personally on several memorable occasions, and helping to establish research collaboration with his Novosibirsk group. Our association began at a conference in Lake Tahoe (USA) in 1995 and was followed by a number of subsequent meetings, notably in Novosibirsk, Manchester, Oxford, and Cambridge. Briefer encounters also took place in the Porquerolles Island (France), in Lyon (France), and in St. Petersburg (Russia). These notes bear witness to the global impact of Godunov's mathematical creativity across multiple branches of science, as well as to its lasting influence on the careers of generations of mathematicians in both academia and industry.
Read the full abstract
Dynamical Systems
Melnikov Analysis of Deterministic and Stochastic Manifold Splitting in the Kuramoto–Sivashinsky Equation
We develop a Melnikov framework for the Kuramoto Sivashinsky (KS) equation under weak deterministic and stochastic forcing. By treating KS as an infinite dimensional dynamical system, we derive a Melnikov functional that measures splitting of stable and unstable manifolds of a homoclinic orbit. Periodic forcing leads to phase dependent transverse intersections, while stochastic forcing produces random manifold splitting characterized by a variance determined by the adjoint solution. This provides a geometric mechanism linking invariant manifold theory to spatiotemporal chaos in dissipative partial differential equations.
Read the full abstract
Statistics Theory
Bayesian bivariate survival estimation
There is no easy extension of Kaplan-Meier and Nelson-Aalen estimators to the bivariate case, and estimating bivariate survival distributions nonparametrically is associated with various non-trivial problems. The Dabrowska estimator will for example associate negative mass to some subsets. Bayesian methods hold some promise as they will avoid the negative mass problem, butare also prone to difficulties. We simplify and extend an example by Pruitt to show that the posterior distribution from a Dirichlet process prior is inconsistent. We construct a different nonparametric prior via Beta processes and provide an updating scheme that utilizes only the most relevant parts of the likelihood, and show that this leads to a consistent estimator.
Read the full abstract
Probability · Cross-listed from q-bio.PE
Scale-dependent Temporal Signatures of Arboviral Transmission in Urban Environments
Understanding epidemic dynamics in urban environments requires models that capture interactions across space and time while incorporating biological constraints. In this work, we propose a probabilistic spatiotemporal framework based on pairwise interaction kernels to analyze arboviral transmission using large-scale georeferenced data from Recife, Brazil. The model describes interactions as a function of spatial distance and temporally delayed influence, with parameters estimated via maximum likelihood. Our results reveal a marked asymmetry between spatial and temporal components. The spatial parameter systematically collapses, indicating that spatial proximity does not provide discriminatory information between diseases at the urban scale. In contrast, temporal dynamics exhibit scale-dependent behavior: statistical differentiation between dengue, Zika, and chikungunya emerges only beyond a critical temporal window. We show that unconstrained models primarily capture short-term co-occurrence, leading to apparent but non-robust differences, while biologically constrained models reveal a common underlying transmission structure. Additionally, reconstructed transmission networks exhibit localized and structured interaction patterns consistent with plausible epidemic propagation. These findings demonstrate that epidemic differentiation is not intrinsic, but an emergent phenomenon dependent on temporal scale, highlighting the importance of biologically grounded and scale-aware modeling in spatiotemporal epidemic analysis.
Read the full abstract
Combinatorics
Ratio of the number of $\mathbf{1}$-nearly independent vertex subsets and the Merrifield-Simmons index
The number $\sigma_k(G)$ of induced subgraphs with size $k$ of a graph $G$ was introduced recently as the number of $k$-nearly independent vertex subsets of $G$. Results highlighting similarity and difference in the behaviours of $\sigma_1$ and $\sigma_0$, have been reported. In this paper, we provide more comparison tools, by studying the ratio $\frac{\sigma_1(G)}{\sigma_{0}(G)}$. We establish sharp lower and upper bounds for this ratio over various classes of graphs, including connected graphs, trees, and forests.
Read the full abstract
Statistics Theory
Confidence envelopes for the false discoveries with heterogeneous data
In the context of selective inference, confidence envelopes for the false discoveries allow the user to select any subset of null hypotheses while having a statistical guarantee on the number of false discoveries in the selected set. Many constructions of such envelopes have been proposed recently, using local test families (Genovese and Wasserman, 2006; Goeman and Solari, 2011), paths (Katsevich and Ramdas, 2020) or interpolation (Blanchard et al., 2020a). All those methods have in common that they have been well-studied for the homogeneous case where all p-values under the null have a uniform distribution over [0, 1]. However, in many applications the data are heterogeneous and discrete, hence the p-values have heterogeneous, discrete distributions, and the previous constructions may incur a loss of power, in the sense that they over-estimate the number of false discoveries. In this paper, we bridge the previous constructions under the homogeneous case with new tools. We also apply these tools to propose several confidence envelopes based on tools tailored for heterogeneous data, like the Bretagnolle inequality, or a new variant of the Simes inequality. We compare these new envelopes to their homogeneous counterparts on simulated data.
Read the full abstract
K-Theory and Homology
The $K$-theory of finite Tambara fields: away from $p$
In previous work, the author and David Chan computed the algebraic $K$-theory of the constant $C_2$-Tambara field with value the field with two elements, using a method which fails at odd primes. Herein we make progress towards the corresponding odd primary computations using a completely new idea. Particularly, we show that the $K$-theory groups of any constant $C_{p^n}$-Tambara field with value a characteristic $p$ finite field are torsion, and we completely determine these groups after inverting $p$. The away-from-$p$-torsion satisfies a simple pattern predicted by previous work, and a computer-aided computation shows that the $p$-power torsion is nontrivial in general.
Read the full abstract
Statistics Theory · Cross-listed from stat.ME
Partial correlation networks of Gaussian processes
In Gaussian graphical models, conditional independence and partial correlations are natural inferential targets for understanding direct relationships in multivariate data. No comparable framework exists for spatial processes, where multivariate analysis defaults to modeling unconditional cross-covariance structure, even when direct relationships remain of scientific interest. We address this gap by establishing a novel characterization of process-level partial correlation for multivariate Gaussian processes that recovers a direct link with Gaussian graphical models. Our analysis proceeds through a class of stationary multivariate processes, termed spectrally inside-out, in which a precision matrix modulates the strength of conditional dependence and yields necessary and sufficient conditions for conditional independence. Within this class, partial cross-correlation functions factorize into a process-level partial correlation coefficient and an attenuation term independent of cross-process parameters. The spectrally inside-out class includes the separable coregionalization model, a process convolution construction, and the parsimonious multivariate Matérn, for which such a characterization was previously thought unavailable. We further show that a nonstationary inside-out model satisfies the same factorization and admits the same necessary and sufficient conditions. Our results clarify the limitations of existing approaches: linear coregionalization models encode conditional independence through the zero pattern of the inverse factor loading matrix and do not result in interpretable partial cross-correlation functions. Low-rank spatial factor models lack a meaningful graphical characterization. Methods that enforce network structure through auxiliary graphical layers only characterize presence or absence of graph edges. We illustrate our results through synthetic and real data.
Read the full abstract
Statistics Theory
On a copula product characterizing independence and perfect functional dependence
Recently studied dependence measures take values in [0,1], where 0 characterizes independence of X and Y, and 1 characterizes perfect functional (not necessarily monotone) dependence of Y on X. The most prominent example is Chatterjee's rank correlation, which is based on the concept of conditional independence. In contrast, we show that dependence measures such as Wasserstein correlations and rearranged dependence measures are naturally linked to the concept of conditional comonotonicity. This connection is captured by a new copula product T(C)=C\vee\Pi which models conditional comonotonicity and characterizes independence and perfect functional dependence. As a main contribution, we prove that the mapping T acts as a reflection on the class of stochastically increasing copulas. Further, T^2=T \circ T projects a copula onto its increasing rearranged copula. To better understand the behavior of such dependence measures, we also study fixed points, ordering results, and continuity properties of T. Our results demonstrate that conditional comonotonicity is a rather intrinsic feature of dependence measures, whereas conditional independence underlying Chatterjee's rank correlation is a quite exceptional property.
Read the full abstract
Analysis of PDEs
Vorticity-Response Blowup in Axisymmetric Models
We study how the orientation and elliptic depth of swirl-generated vorticity response affect singularity formation in axisymmetric model equations. For a sign-reversed Hou–Li model on the whole line, the dynamics factorize into two real Riccati–diffusion channels. For every viscosity $\nu>0$, every nontrivial smooth datum in an infinite-dimensional cone of even, nonnegative, monotone channel data develops finite-time amplitude blowup; in the inviscid case, characteristics give the exact blowup time. For the same smooth swirling inviscid datum, the standard response sign is global whereas the reversed sign blows up. We then introduce a three-parameter 5D response family separating transport geometry, response orientation, and elliptic depth. Exact divergence, energy, and scaling identities show that physical three-dimensional incompressibility selects one geometry, weighted 5D incompressibility selects another, and Navier–Stokes scaling makes two elliptic inversions critical. The frozen physical response is critical and dissipative, while a natural one-inversion 5D-solenoidal model is supercritical and energy-pumping. These results identify response orientation and causal depth as load-bearing structure in axisymmetric model dynamics.
Read the full abstract
Optimization and Control · Cross-listed from cs.AI
New Hybrid Fine-Tuning Paradigm for LLMs: Algorithm Design and Convergence Analysis Framework
Fine-tuning Large Language Models (LLMs) typically involves either full fine-tuning, which updates all model parameters, or Parameter-Efficient Fine-Tuning (PEFT), which adjusts a small subset of parameters. However, both approaches have inherent limitations: full fine-tuning is computationally expensive, while PEFT often struggles to learn new knowledge and exhibits suboptimal performance. To overcome these issues, we propose a novel hybrid fine-tuning approach that jointly updates both LLMs and PEFT modules using a combination of zeroth-order and first-order optimization methods. To analyze our new algorithm, we develop a theoretical framework centered on the concept of hybrid smoothness condition, which accounts for the heterogeneous nature of the optimization landscape in joint LLM and PEFT training. We derive a rigorous convergence analysis for the convergence of reshuffling-type SGD algorithm under multiple learning rates and demonstrate its effectiveness through extensive empirical studies across various downstream tasks and model architectures. On the practical side, our results demonstrate consistent performance improvement, making the approach a viable solution for large-scale language model fine-tuning.
Read the full abstract
Functional Analysis
Normed lattices majorizing in their norm completions
This note is a follow-up to \cite{bt}. We focus on conditions under which a normed lattice $X$ is majorizing in its norm completion. We show that \cite[Question 8.17]{bt} – namely, whether this holds whenever every norm-null sequence in $X$ has an order-bounded subsequence – is equivalent to the question whether every P-ideal on $\N$ is meager. This is a longstanding open problem in Set Theory, and it has a negative answer under various set-theoretical assumptions, in particular under the Continuum Hypothesis. We also present several equivalent conditions to both of the two aforementioned properties, and give a simple proof of a well-known Riesz-Fischer-style characterization of completeness of a normed lattice.
Read the full abstract
Analysis of PDEs
Resolvent estimates for the Schrödinger operator with $L^\infty$ electric and magnetic potentials and applications to the local energy decay
We establish resolvent estimates that extend earlier results to a larger class of electric potentials $V\in L^\infty(\mathbb{R}^d;\mathbb{R})$, $d\ge 3$, and magnetic potentials $b\in L^\infty(\mathbb{R}^d;\mathbb{R}^d)$ such that $V(x), b(x)=O_k\left(|x|^{-k}\right)$, $|x|\gg 1$, for every integer $k$. More precisely, we prove estimates for the derivatives of the weighted resolvent of the corresponding magnetic Schrödinger operator, which are uniform with respect to both the spectral parameter and the order of derivation. We also show that these resolvent estimates still hold for the Dirichlet self-adjoint realization of the Schrödinger operator in the exterior of a non-trapping obstacle in $\mathbb{R}^d$, $d\ge 2$, provided the magnetic potential is supposed identically zero. As an application of these resolvent estimates, we obtain the rate of decay of the local energy of solutions to the corresponding wave equation. In particular, we show that for potentials satisfying $|V(x)|+|b(x)|\le Ce^{-c|x|^s}$, $c,C>0$, $0<s<1$, the rate of decay of the local energy is $e^{-c_0t^s}$ with some constant $c_0>0$, where $t\gg 1$ is the time variable.
Read the full abstract
Analysis of PDEs
A Carleman contraction method for inverse initial data recovery in the Navier-Stokes equations with unknown body force
We solve an inverse initial data problem for the incompressible Navier-Stokes system. The objective is to recover the initial velocity and pressure from lateral boundary observations, without assuming that the time-independent body force is known. To eliminate this unknown force, we differentiate the momentum equation with respect to time and then apply a Legendre polynomial-exponential time-dimensional reduction. This procedure yields a coupled system of elliptic equations for the expansion coefficients. We then construct a contractive map for this reduced system on a suitable admissible set equipped with a Carleman-weighted norm. Its fixed point yields an approximate solution of the time-dimensional reduction model, and the contraction property gives rise to a globally convergent Picard iteration. Finally, we present a numerical algorithm based on this framework and numerical experiments showing accurate reconstructions of the initial velocity and pressure from synthetic boundary data.
Read the full abstract
Probability
Explicit cutoff profiles for colored top-$m$-to-random shuffles
We study $p$-colored top-$m$-to-random on the wreath product $G_{n,p}=C_p\wr S_n$, with $m$ fixed. Using the Nakano-Sadahiro-Sakurai basis elements $B_m$, we obtain exact nested-set occupancy mixtures and reduce the likelihood ratio to the single statistic $L_p$. This yields exact formulas for separation and $L^\infty(U)$, and exact one-dimensional formulas for total variation, $L^q(U)$ ($1\le q<\infty$), $\chi^2$, and relative entropy. At $k=\Bigl\lfloor \frac{n}{m}(\log n+c)\Bigr\rfloor$, the number of never-chosen labels in the associated $m$-subset occupancy model converges in law to $\mathrm{Poisson}(e^{-c})$, giving the total-variation profile $f_p(c)$, the separation profile, and the corresponding $L^q(U)$, $L^\infty(U)$, $\chi^2$, and relative-entropy profiles. For $m=1$ we recover colored top-to-random; for $p=1$, the total-variation profile reduces to the Diaconis-Fill-Pitman profile. For the reversed chain, we also identify optimal strong stationary times whose tail probabilities are exactly the separation distances.
Read the full abstract
Optimization and Control · Cross-listed from eess.SY
Online Electricity Pricing from Frequency Measurements
Frequency dynamics in power systems reflect active power imbalance in real time, thereby providing an instantaneous signal to inform electricity pricing. However, existing real-time markets operate on much slower timescales and fail to exploit this signal. In this letter, we develop integrated market–frequency dynamics that enable online pricing directly from frequency measurements. Representing the real-time market as a dynamic price-discovery process, and integrating this process with the grid frequency dynamics, we derive an explicit price formation mechanism from frequency measurements. This mechanism manifests as a distributed PID-like controller for each generator, where frequency response is driven and remunerated by electricity prices derived solely from local frequency measurements.
Read the full abstract
Analysis of PDEs
On Kirchhoff-type p(.)-Laplacian problems with sandwich-type and arbitrary growth
We establish the existence of a positive bounded weak solution for a class of Kirchhoff-type $p(\cdot)$-Laplacian problems involving an arbitrary growth and a sandwich-type growth $s(\cdot)\in (\inf p,\sup p)$. This setting leads to substantial analytical difficulties in the variational analysis of the associated energy functional. By combining truncation arguments with a priori estimates, we prove the existence result under suitable assumptions on the data.
Read the full abstract
Optimization and Control
Convex Synthesis of First-Order Methods for Time-Varying Smooth Strongly Convex Optimization
Time-varying optimization is fundamental to decision-making in dynamic environments, where objectives evolve over time due to exogenous signals or data streams. However, algorithms designed for static problems yield suboptimal decisions in dynamic scenarios, even asymptotically. In this paper, we develop a robust control synthesis framework to systematically design first-order methods for smooth strongly convex problems that vary in time. Our approach leverages both convex robust control synthesis in the static setting and the internal model principle by directly embedding a model of the underlying variability into the designed algorithm.
Read the full abstract
Functional Analysis
Quantitative Stability and Numerical Resolution of the Moment Measure Problem
The moment measure problem consists in finding a convex function $\psi$ whose moment measure, i.e., the pushforward by $\nabla \psi$ of the measure with density $e^{-\psi(\,\cdot\,)}$, is prescribed. It is highly non-linear and less understood than the related optimal transport problem. We establish a quantitative stability estimate for this problem. This estimate validates, as well as leads us to introduce, an approach to the numerical resolution of the moment measure problem inspired by semi-discrete optimal transport, consisting in approximating the prescribed measure by a finitely supported one. We describe a Newton method for solving the discrete problem thus obtained, and perform numerical experiments, studying the experimental rates of convergence of the approximation beyond the predictions of the stability estimate.
Read the full abstract
Numerical Analysis · Cross-listed from cs.LG
Last-Iterate Convergence of Randomized Kaczmarz and SGD with Greedy Step Size
We study last-iterate convergence of SGD with greedy step size over smooth quadratics in the interpolation regime, a setting which captures the classical Randomized Kaczmarz algorithm as well as other popular iterative linear system solvers. For these methods, we show that the $t$-th iterate attains an $O(1/t^{3/4})$ convergence rate, addressing a question posed by Attia, Schliserman, Sherman, and Koren, who gave an $O(1/t^{1/2})$ guarantee for this setting. In the proof, we introduce the family of stochastic contraction processes, whose behavior can be described by the evolution of a certain deterministic eigenvalue equation, which we analyze via a careful discrete-to-continuous reduction.
Read the full abstract
Mathematical Physics
Strictly correlated electrons in a quantum ring: from Kohn-Sham to Kantorovich potentials
Our goal in this paper is twofold. First, we characterize the class of pairwise interactions for which the Seidl conjecture on the structure of optimal plans for the symmetric multimarginal optimal transport problem with one-dimensional marginal holds. This extends previous results by Colombo, De Pascale, and Di Marino [Can. Jou. Math., 67 (2015), https://doi.org/10.4153/CJM-2014-011-x], which treated the case of translation-invariant, convex and decreasing interactions. In particular, our results apply to physically relevant interactions for electrons living on a quantum ring. The second main goal of the paper is to rigorously derive the leading order asymptotics of the adiabatic connection potential for strongly interacting systems. More precisely, we show that for electrons in a quantum ring (or one-dimensional interval), not only the Lieb density functional converges to the optimal transport (or strictly correlated) functional in the semiclassical limit, but also the representing potential converges to a regular Kantorovich potential. As an intermediate step, we also extend previous results on the strongly interacting limit of the Lieb functional to periodic systems in arbitrary dimensions.
Read the full abstract
Information Theory
Diffusion Denoiser Achievable Analysis for Finite Blocklength Unsourced Random Access
Polyanskiy proposed a framework for the unsourced multiple access channel (MAC) problem where users employ a common codebook in the finite blocklength regime. However, existing approaches handle channel noise before the joint decoder. In this work, we introduce a decoder compatible diffusion denoiser as a lightweight analysis within joint decoding. The score network is trained on samples drawn from the channel output distribution, making the method easy to integrate with existing code designs. In our theoretical analysis, we derive a diffusion-denoiser random-coding achievable bound that is strictly tighter. Simulations on existing decoders, including FASURA, MSUG-MRA and pilot-based method, show consistent performance gains with at least a $0.5$ $\mathrm{dB}$ improvement in required $\mathrm{E_b/N_0}$ at a fixed error target.
Read the full abstract
Classical Analysis and ODEs
The attainable almost sure large dimensions
In this paper we study the range of possible almost sure dimensions of random measures arising from a natural model of random Moran measures. Specifically, we consider the Assouad-like “large” $\Phi$-dimensions of these measures. These dimensions can be tuned to consider a specific range of depths in scale and so provide refined local geometric information. The quasi-Assouad dimension is a well-known and important example of a “large” $\Phi$-dimension. We determine the range of possible almost sure $\Phi$-dimensions for random measures generated by the model and supported on any given random Moran set. We do this for both the case when the probability weights depend on the scaling factors and the case when they do not. In the later situation, we show that usually there is a “gap” between the dimension of the set and that of the smallest attainable upper dimension and largest attainable lower dimension. As a consequence of our results, we also determine the a.s. dimensions of the underlying random Moran set.
Read the full abstract
Analysis of PDEs
Stochastic homogenization of fractional obstacle problems
We prove a stochastic homogenization result for a class of nonlinear and nonlocal variational problems in domains with many small randomly distributed (bilateral) obstacles. Our model case is a Dirichlet problem for the fractional $p$-Laplacian, $p>1$, where a pinning condition $u=0$ is imposed on the solution in a random collection of small balls whose centers and radii are generated by a stationary marked point process. Such a general obstacle distribution allows for clustering effects to appear with positive probability. Under suitable moment conditions on the obstacle radii, we identify a critical scaling regime in which the fractional $p$-capacity density of the obstacles is asymptotically additive almost surely. In turn, this key property allows us to derive an effective homogenized problem which is formally analogous to the one obtained in the periodic setting or under the assumption of well-separation for the obstacles. The analysis also extends to the case of randomly shaped obstacles and to a broad class of nonlocal interaction kernels. At the methodological level, the paper develops a streamlined proof strategy with several new ingredients, among them the use of Palm measures.
Read the full abstract
Logic
The Henson graphs: colorings and codings
By recent work of \citet{DobrinenICM} and \citet{Balko7} we know that every finite $G$ in the Henson graph $\mathbb{H}_{n+1}$ (the universal ultrahomogeneous $(n+1)$-clique free graph) has exact finite big Ramsey degree $k({G,n})$. That is, there is a positive integer $k({G,n})$ such that for each finite coloring $C$ of the copies of $G$ in $\mathbb{H}_{n+1}$, there is $\tilde{\mathbb{H}}$, a substructure of $\mathbb{H}_{n+1}$ and isomorphic to $\mathbb{H}_{n+1}$, such that in $\tilde{\mathbb{H}}$ at most $k({G,n})$ colors are used on the copies of $G$ in $\tilde{\mathbb{H}}$. Moreover, for exactness, for some coloring and all corresponding $\tilde{\mathbb{H}}$, all $k({G,n})$ colors are needed. The ultimate result here is that if $|G|\geq 2$, then there is a finite computable coloring $C$ such that, for all such $\tilde{\mathbb{H}}$, we have that $\tilde{\mathbb{H}}$ computes $\emptyset^{(|G|-1)}$ (and hence the halting set).
Read the full abstract
Differential Geometry
Robustness of CSC Sasaki existence under the join operation
Spectral Theory
Banded Hermitian Matrices, Matrix Orthogonal Polynomials, and the Toda Lattice
We study the direct and inverse spectral theory for a class of finite Hermitian banded matrices. Using the theory of matrix orthogonal polynomials, we provide an explicit procedure for reconstructing a banded matrix from a matrix-valued measure that encodes its spectral data. We establish necessary and sufficient conditions for a measure to be the spectral measure of a matrix in the examined class. We further analyze the connections between this spectral analysis, block tridiagonalization algorithms, and the Toda lattice evolution on banded matrices.
Read the full abstract
Optimization and Control
Some properties of sets, functions, and multi-objective optimization problems using p-convexity
In this paper, we investigate the concept of p-convexity for sets and functions in n-dimensional Euclidean space. We establish novel algebraic and topological results within this generalized convexity framework. Furthermore, we analyze multi-objective optimization problems, with a particular emphasis on weakly efficient minima, under the assumption of $p$-convexity of the component functions. Several characterizations and properties of the corresponding solution sets are derived.
Read the full abstract
Dynamical Systems
Discontinuity example for the Lyapunov exponents on the boundary of the uniformly hyperbolic set
We present an example of a discontinuity point for the Lyapunov exponents when viewed as a function of the cocycle in a topology finer than the $C^0$-topology. The linear cocycle taking values in SL(2,R) is locally constant, defined over a Bernoulli shift, and lies on the boundary of the uniformly hyperbolic set. In particular, we show that it can be approximated, in the $C_{\delta-\log}$-topology, by cocycles whose Lyapunov exponents vanish.