We provide a homotopy equivalence for the loop space of the polyhedral product associated with a simplicial complex formed via the polyhedral join operation, and give sufficient conditions under which this loop space is a finite-type product of spheres and loops on spheres.

In deep learning (DL), the instability phenomenon is widespread and well documented, and the most commonly used measure of stability is the Lipschitz constant. While a small Lipchitz constant is traditionally viewed as guarantying stability, it does not capture the instability phenomenon in DL for classification well. The reason is that a classification function – which is the target function to be approximated – is necessarily discontinuous, thus having an ‘infinite’ Lipchitz constant. As a result, the classical approach will deem every classification function unstable, yet basic classification functions a la ‘is there a cat in the image?’ will typically be locally very ‘flat’ – and thus locally stable – except at the decision boundary. The lack of an appropriate measure of stability hinders a rigorous theory for stability in DL, and consequently, there are no proper approximation theoretic results that can guarantee the existence of stable networks for classification functions. In this paper, we introduce a novel stability measure $\mathcal{S}(f)$, for any classification function $f$, appropriate to study the stability of classification functions and their approximations. We further prove two approximation theorems: First, for any $\epsilon \gt 0$ and any classification function $f$ on a compact set, there is a neural network (NN) $\psi$, such that $\psi - f \neq 0$ only on a set of measure $\lt \epsilon$; moreover, $\mathcal{S}(\psi ) \geq \mathcal{S}(f) - \epsilon$ (as accurate and stable as $f$ up to $\epsilon$). Second, for any classification function $f$ and $\epsilon \gt 0$, there exists a NN $\psi$ such that $\psi = f$ on the set of points that are at least $\epsilon$ away from the decision boundary.

We introduce a natural boundary value problem for a triholomorphic map $u$ from a compact almost hyper-Hermitian manifold $M$ with smooth boundary $\partial M$ into a closed hyperKähler manifold $N$ with free boundary $u(\partial M)\subset \Gamma$ lying on some geometrically natural closed supporting submanifold $\Gamma\subset N$, called tri-isotropic submanifold. We establish partial regularity theory and energy quantization result in this boundary setting under some additional assumption on the $W^{2,1}$ norm of the weakly converging sequences.

In deep learning, interval neural networks are used to quantify the uncertainty of a pre-trained neural network. Suppose we are given a computational problem $P$ and a pre-trained neural network $\Phi _P$ that aims to solve $P$. An interval neural network is then a pair of neural networks $(\underline {\phi }, \overline {\phi })$, with the property that $\underline {\phi }(y) \leq \Phi _P(y) \leq \overline {\phi }(y)$ for all inputs $y$, where the inequalities are meant componentwise. $(\underline {\phi }, \overline {\phi })$ are specifically trained to quantify the uncertainty of $\Phi _P$, in the sense that the size of the interval $[\underline {\phi }(y),\overline {\phi }(y)]$ quantifies the uncertainty of the prediction $\Phi _P(y)$. In this paper, we investigate the phenomenon when algorithms cannot compute interval neural networks in the setting of inverse problems. We show that in the typical setting of a linear inverse problem, the problem of constructing an optimal pair of interval neural networks is non-computable, even with the assumption that the pre-trained neural network $\Phi _P$ is an optimal solution. In other words, there exist classes of training sets $\Omega$, such that there is no algorithm, even randomised (with probability $p \geq 1/2$), that computes an optimal pair of interval neural networks for each training set ${\mathcal{T}} \in \Omega$. This phenomenon happens even when we are given a pre-trained neural network $\Phi _{{\mathcal{T}}}$ that is optimal for $\mathcal{T}$. This phenomenon is intimately linked to instability in deep learning.

In this note, we prove that minimizers of convex functionals with a convexity constraint and a general class of Lagrangians can be approximated by solutions to fourth-order Abreu-type equations. Our result generalizes that of Le (Twisted Harnack inequality and approximation of variational problems with a convexity constraint by singular Abreu equations. Adv. Math. 434 (2023)) where the case of quadratically growing Lagrangians was treated.

The descent method is one of the approaches to study the Brauer–Manin obstruction to the local–global principle and to weak approximation on varieties over number fields, by reducing the problem to ‘descent varieties’. In recent lecture notes by Wittenberg, he formulated a ‘descent conjecture’ for torsors under linear algebraic groups. The present article gives a proof of this conjecture in the case of connected groups, generalizing the toric case from the previous work of Harpaz–Wittenberg. As an application, we deduce directly from Sansuc’s work the theorem of Borovoi for homogeneous spaces of connected linear algebraic groups with connected stabilizers. We are also able to reduce the general case to the case of finite (étale) torsors. When the set of rational points is replaced by the Chow group of zero-cycles, an analogue of the above conjecture for arbitrary linear algebraic groups is proved.

A Gordian unlink is a finite number of unknots that are not topologically linked, each with prescribed length and thickness, and that cannot be disentangled into the trivial link by an isotopy preserving length and thickness throughout. In this note, we provide the first examples of Gordian unlinks. As a consequence, we identify the existence of isotopy classes of unknots that differ from those in classical knot theory. More generally, we present a one-parameter family of Gordian unlinks with thickness ranging in $[1,2)$ and absolute curvature bounded by 1, concluding that thinner normal tubes lead to different rope geometries than those previously considered. Knots or links in the one-parameter model introduced here are called thin knots or links. When the thickness is equal to 2, we obtain the standard model for geometric knots, also called thick knots.

The unprecedented success of deep learning (DL) makes it unchallenged when it comes to classification problems. However, it is well established that the current DL methodology produces universally unstable neural networks (NNs). The instability problem has caused a substantial research effort – with a vast literature on so-called adversarial attacks – yet there has been no solution to the problem. Our paper addresses why there has been no solution to the problem, as we prove the following: any training procedure based on training rectified linear unit (ReLU) neural networks for classification problems with a fixed architecture will yield neural networks that are either inaccurate or unstable (if accurate) – despite the provable existence of both accurate and stable neural networks for the same classification problems. The key is that the stable and accurate neural networks must have variable dimensions depending on the input, in particular, variable dimensions is a necessary condition for stability. Our result points towards the paradox that accurate and stable neural networks exist; however, modern algorithms do not compute them. This yields the question: if the existence of neural networks with desirable properties can be proven, can one also find algorithms that compute them? There are cases in mathematics where provable existence implies computability, but will this be the case for neural networks? The contrary is true, as we demonstrate how neural networks can provably exist as approximate minimisers to standard optimisation problems with standard cost functions; however, no randomised algorithm can compute them with probability better than $1/2$.

We characterize the subsets E of a metric space X with doubling measure whose distance function to some negative power $\operatorname{dist}(\cdot,E)^{-\alpha}$ belongs to the Muckenhoupt A1 class of weights in X. To this end, we introduce the weakly porous sets in this setting, and show that, along with certain doubling-type conditions for the sizes of the largest E-free holes, these sets characterize the mentioned A1-property. We exhibit examples showing the optimality of these conditions, and simplify them in the particular case where the underlying measure satisfies a qualitative annular decay property. In addition, we use some of these distance functions as a new and simple method to explicitly construct doubling weights in ${\mathbb R}^n$ that do not belong to $A_\infty.$

The Born approximation of a potential in the context of the Calderón inverse problem is an object that can be formally defined in terms of spectral data of the Dirichlet-to-Neumann map of the corresponding Schrödinger operator. In this article, we prove, in the case of radial potentials in the Euclidean ball and any fixed energy, that the Born approximation is well-defined as a compactly supported radial distribution, and that the Calderón problem can be reformulated as recovering a potential from its Born approximation. In addition, we show that the Born approximation depends locally on the potential and captures exactly its singularities, and that the functional that maps the Born approximation to the potential is Hölder continuous. We also prove that the Born approximation converges to the potential in the high-energy limit. Moreover, we give an explicit formula for the Fourier transform of the Born approximation at any fixed energy, and illustrate how it can be used as the basis of an accurate procedure to approximate a potential from its Dirichlet-to-Neumann map.

For associative rings with anti-involution several homology theories exist, for instance reflexive homology as studied by Graves and involutive Hochschild homology defined by Fernàndez-València and Giansiracusa. We prove that the corresponding homology groups can be identified with the homotopy groups of an equivariant Loday construction of the one-point compactification of the sign-representation evaluated at the trivial orbit, if we assume that 2 is invertible and if the underlying abelian group of the ring is flat. We also show a relative version where we consider an associative k-algebra with an anti-involution where k is an arbitrary commutative ground ring.

In this paper, we investigate the ideal structure of uniform Roe algebras for general metric spaces beyond the scope of Yu’s Property A. Inspired by the ideal of ghost operators coming from expander graphs and in contrast to the notion of geometric ideal, we introduce a notion of ghostly ideal in a uniform Roe algebra, whose elements are locally invisible in certain directions at infinity. We show that the geometric ideal and the ghostly ideal are, respectively, the smallest and the largest element in the lattice of ideals with a common invariant open subset of the unit space of the coarse groupoid by Skandalis–Tu–Yu, and hence the study of ideal structure can be reduced to classifying ideals between the geometric and the ghostly ones. We also provide a criterion to ensure that the geometric and the ghostly ideals have the same $K$-theory, which helps to recover counterexamples to the coarse Baum–Connes conjectures. Moreover, we introduce a notion of partial Property A for a metric space to characterize the situation in which the geometric ideal coincides with the ghostly ideal. As an application, we provide a concrete description for the maximal ideals in a uniform Roe algebra in terms of the minimal points in the Stone–Čech boundary of the space.

In this paper, we investigate the dimension theory of the one-parameter family of Okamoto’s function. We compute the Hausdorff, box-counting, and Assouad dimensions of the graph for a typical choice of parameter. Furthermore, we study the dimension of the level sets. We give an upper bound on the dimension of every level set, and we show that for a typical choice of parameter, this value is attained for Lebesgue almost every level set.

We consider the existence, nonexistence and multiplicity of normalized solutions to the $p$-Laplacian equation on a bounded domain(0.1)

\begin{equation}\left\{\begin{array}{ll}-\Delta_p u+\lambda |u|^{p-2}u=g(u),&\text{in }\Omega,\\\int_\Omega |u|^p=\rho, u=0&\text{on }\partial \Omega,\end{array}\right.\end{equation}

where $\Omega$ is a bounded domain, $p\geq 2$. Firstly, under suitable assumptions on $\rho$, if $g$ is at most mass-critical at infinity, we prove the existence of infinitely many solutions. Secondly, for $\rho$ large, if $g$ is mass-supercritical, we perform a blow-up analysis to show the nonexistence of finite Morse index solutions. At last, for $\rho$ suitably small, combining with the monotonicity argument, we obtain a multiplicity result. In particular, when $p=2$, we obtain the existence of infinitely many normalized solutions.

In this paper, we study the validity of the sub-supersolution method for the equation

\begin{equation*}\begin{cases}-\mbox{div}(K(x)\nabla u)=K(x)|x|^{\alpha-2}f(x,u) \,\mbox{in } {\mathbb{R}}^{N},\\u \gt 0 \,\mbox{in } {\mathbb{R}}^{N},\end{cases}\end{equation*}

$N \geq 3$

$K(x)=exp(|x|^{\alpha}/4)$

$\alpha\geq 2$

$f$

$f$

$f$

We study the existence and multiplicity of positive bounded solutions for a class of nonlocal, non-variational elliptic problems governed by a nonhomogeneous operator with unbalanced growth, specifically the double phase operator. To tackle these challenges, we employ a combination of analytical techniques, including the sub-super solution method, variational and truncation approaches, and set-valued analysis. Furthermore, we examine a one-dimensional fixed-point problem.To the best of our knowledge, this is the first workaddressing nonlocal double phase problems using these methods.

We prove that a partially hyperbolic attracting set for a $C^2$ vector field, having slow recurrence to equilibria, supports an ergodic physical/SRB measure if, and only if, the trapping region admits non-uniform sectional expansion on a positive Lebesgue measure subset. Moreover, in this case, the attracting set supports at most finitely many ergodic physical/SRB measures, which are also Gibbs states along the central-unstable direction. This extends to continuous time systems a similar well-known result obtained for diffeomorphisms, encompassing the presence of equilibria accumulated by regular orbits within the attracting set. In codimension two the same result holds, assuming only the trajectories on the trapping region admit a sequence of times with asymptotical sectional expansion, on a positive volume subset. We present several examples of application, including the existence of physical measures for asymptotically sectional hyperbolic attracting sets, and obtain physical measures in an alternative unified way for many known examples: Lorenz-like and Rovella attractors, and sectional-hyperbolic attracting sets (including the multidimensional Lorenz attractor).

We study necessary and sufficient conditions for a 4-dimensional Lefschetz fibration over the 2-disk to admit a ${\text{Pin}}^{\pm}$-structure, extending the work of A. Stipsicz in the orientable setting. As a corollary, we get existence results of ${\text{Pin}}^{+}$ and ${\text{Pin}}^-$-structures on closed non-orientable 4-manifolds and on Lefschetz fibrations over the 2-sphere. In particular, we show via three explicit examples how to read-off ${\text{Pin}}^{\pm}$-structures from the Kirby diagram of a 4-manifold. We also provide a proof of the well-known fact that any closed 3-manifold M admits a ${\text{Pin}}^-$-structure and we find a criterion to check whether or not it admits a ${\text{Pin}}^+$-structure in terms of a handlebody decomposition. We conclude the paper with a characterization of ${\text{Pin}}^+$-structures on vector bundles.

A meta-conjecture of Coulson, Keevash, Perarnau, and Yepremyan [12] states that above the extremal threshold for a given spanning structure in a (hyper-)graph, one can find a rainbow version of that spanning structure in any suitably bounded colouring of the host (hyper-)graph. We solve one of the most pertinent outstanding cases of this conjecture by showing that for any $1\leq j\leq k-1$, if $G$ is a $k$-uniform hypergraph above the $j$-degree threshold for a loose Hamilton cycle, then any globally bounded colouring of $G$ contains a rainbow loose Hamilton cycle.