Machine Learning
Papers
Fresh from arXiv, typeset properly. Highlighted words trade on the market.
All of arXiv
Tuesday 8 September
An input may activate few hidden units even when different inputs collectively use an entire network. We study the statistical complexity of this input-dependent sparsity in the one-hidden-layer ReLU model of Awasthi et al. (COLT 2024). For width $s$, at most $k$ active units per input, and effective weight and bias bounds $W,B$, every size-$m$ sample in the class's fixed radius-$R$ input domain satisfies $\mathcal{R}(S)\le CWR\min\{k,\sqrt{sk/m}\log^{3/2}(2m)\}+kB/\sqrt m$. A support-preserving cover and a single normalized chaining argument remove the previous explicit dimension factor, up to logarithms. Lower bounds on appropriate i.i.d. marginals match up to those logarithms, showing how changing active units across inputs retains a width dependence. The input domain matters: zero-bias networks sparse on the entire ball have at most $2k$ nonzero units and complexity $O(kWR/\sqrt m)$, whereas bias bounds comparable to $WR$ restore the worst-case rate on that same domain in only logarithmic dimension. A spherical-cap construction proves the latter claim without assuming sparsity merely on the sampling support. For a specified normalized bounded loss and biases comparable to $WR$, we also obtain agnostic minimax excess-risk bounds of order $\min\{1,\sqrt{s/(km)}\}$ up to logarithms.
Quantum Gases
Anderson orthogonality scaling in the Rabi-driven heavy Fermi polaron
The Anderson orthogonality catastrophe (AOC) is a paradigmatic many-body phenomenon in which a local perturbation induces a macroscopic response of a Fermi sea. We probe signatures of the AOC by coherently driving heavy Fermi polarons in an ultracold $^6$Li-$^{133}$Cs mixture. We observe a power-law dependence of the measured Rabi frequency on the drive strength, with exponents consistent with AOC predictions. Finite-temperature simulations quantitatively reproduce the observed scaling, indicating that AOC signatures persist beyond the idealized zero-temperature, infinite-mass limit. The damping of the Rabi oscillations provides access to polaron dephasing and reveals a nonmonotonic drive dependence, qualitatively consistent with current theories. Our results establish coherently driven impurities as a versatile probe of quantum many-body dynamics through local coherent control.
Read the full abstract
Algebraic Geometry
On the size of the fibers of a morphism of algebraic varieties
Let u be a morphism of algebraic varieties. Given m integer or infinite, we prove that the set of points whose fiber has size m is a constructible set.
Algebraic Geometry
Unramified cohomology and Brauer–Manin pairing
We show that the Tate module of the $\ell$-adic unramified cohomology $H^3_{\nr}(X, {\Q_\ell}/{\Z_\ell})$ of a smooth projective surface $X$ over a local or strictly local field $k$ of residue characteristic prime to $\ell$ vanishes under suitable conditions. We apply this to answer some questions concerning the left and right kernels of the Brauer–Manin pairing for open subvarieties of $X$. To prove the vanishing theorem, we establish Bloch's formula for the Chow group of codimension-two cycles on a semistable model of $X$, and derive applications to the restriction map for such cycles to the closed fiber of the model.
Read the full abstract
Artificial Intelligence
A Generalization of Amari's Bayesian Duality
Amari's contributions to information geometry and machine learning are well known. Here, we revisit Amari's work on Bayesian duality which has not received as much attention. We connect Amari's Bayesian duality to a convex duality of Bayes' rule. Using this connection, we present a generalization of Amari's Bayesian duality and discuss its relevance for modern artificial intelligence.
Read the full abstract
Quantum Gases
Few-body bound states in the anyon-Hubbard model
Quantum statistics in low-dimensional systems predicts anyonic particles with fractional exchange statistics which are neither that of bosons nor fermions. While anyons are typically found in two dimensions as excitations of topologically-ordered states of matter, anyon-like exchange statistics has also been discussed in one dimension, for instance, in the context of the anyon-Hubbard model (AHM), the physics of which has recently been observed in experiment [Kwan et al., arXiv:2306.01737; Dhar et al., arXiv:2412.21131; and Bakkali-Hassani et al., arXiv:2602.20421]. The AHM can be formulated in terms of bosons featuring density-dependent Peierls phases, described by a statistical phase angle $\theta$, which controls asymmetric transport and the formation of dynamically bound pairs at finite momentum. Here, we show theoretically that the AHM also hosts exact two-body bound states in the continuum (BICs) for arbitrary $\theta\neq 0$, and genuine three- and four-body bound states. Unlike conventional bound states stabilized by attractive (or repulsive) interactions, which are energetically localized with a large effective mass, these clusters here are bound by a purely kinematic mechanism endowing them with fast chiral transport properties. We provide a simple variational approximation to the three-body bound states and explain their binding mechanism. Moreover, we show that the signatures of three-body bound states in the AHM can be directly probed experimentally from the expansion dynamics starting from three localized particles.
Read the full abstract
Computer Vision
Canonical Color as a Lens into Concept Decodability in Vision Encoders and VLMs
Visual encoders construct a representation of the image input for Vision-Language models. How much conceptual, as opposed to immediately visible, information does this representation contain? We use canonical color as a controlled test case to ask whether vision encoders make canonical-color information linearly accessible, even when color is removed from the input image. We construct a dataset of objects with canonical colors, and probe vision encoders for both color and object identity using color and grayscale images. We find that canonical color remains decodable from grayscale images, and is tied to predicted object identity, indicating a conceptual link. Extending this analysis to full VLMs, we find that VLM post-training can have a surprisingly large effect on color decodability in the vision encoder. Overall, canonical color provides a usefully controllable lens for tracing object-level conceptual semantic information in vision encoders and VLMs.
Read the full abstract
Computer Vision
Mask Forcing: Improving Autoregressive Video Diffusion Distillation via Dual-Noise Masking Rollout
Autoregressive (AR) video diffusion models have shown great potential in real-time video generation. Recent methods distill pretrained bidirectional video diffusion models into causal AR students through Distribution Matching Distillation (DMD), but the generated videos often suffer from over-saturation and over-smoothing issues, resulting in limited visual quality and realism. The key contributing factor is the mode-seeking behavior of the reverse KL objective in DMD, which can cause the student distribution to collapse onto only a few modes of the teacher distribution. To address this, we propose Mask Forcing, a Dual-Noise Masking Rollout strategy that perturbs the AR student self-rollout to mitigate mode collapse induced by reverse-KL mode seeking. The core idea is to inject cleaner signals into noisy rollout inputs via random masks along spatial and temporal axes during the self-rollout process of AR diffusion distillation. Such perturbations encourage the student rollouts to explore more regions of the teacher distribution, allowing DMD to provide learning signals beyond the modes already covered by the student. Moreover, the cleaner tokens act as denoising guidance for other noisier tokens, improving the intermediate rollout predictions and reducing error accumulation. Extensive experiments demonstrate that our method improves multiple AR video diffusion distillation methods with higher visual quality efficiently, without incorporating real video data or additional post-training stages.
Read the full abstract
Astrophysics of Galaxies
The fading hierarchy of Galactic open clusters
Galactic open clusters provide a record of both hierarchical star formation and the subsequent dynamical evolution of the Milky Way disk. We use the two-point correlation function to characterize the spatial and kinematic clustering of open clusters in the solar neighborhood, considering their three-dimensional (3D) distributions and their projections onto the Galactic plane, as well as subsamples divided by age, mass, and survey volume. We quantify the clustering through fractal dimensions and characteristic correlation scales. We find that the youngest clusters have fractal dimensions comparable to those measured in the interstellar medium, suggesting that they retain part of the hierarchical structure of their natal molecular clouds. The clustering strength then decreases systematically with age and becomes weak on a characteristic timescale of order 100 Myr. This evolution is scale dependent: the excess correlation at tens of parsecs fades more rapidly than the weaker correlation at hundreds of parsecs. By contrast, we find only a weak dependence of the clustering properties on cluster mass and volume. Using the same cluster sample, we derive an empirical mapping between projected and 3D fractal dimensions, providing an observational calibration for comparing the Milky Way with studies of external galaxies, where only projected measurements are available. The open-cluster population exhibits characteristic spatial and kinematic clustering scales of approximately 370 pc and 11 km/s, respectively. These results support a picture in which young open clusters inherit a spatial hierarchy from star formation and progressively lose their spatial correlations through subsequent evolution.
Read the full abstract
Analysis of PDEs
Differentiability of Lipschitz mappings into metric spaces: area and co-area formulas
We give a self-contained exposition of metric differentiability for Lipschitz mappings from sets in Euclidean spaces into arbitrary metric spaces, together with the corresponding area and co-area formulas. The novelty is a new metric implicit function theorem for mappings into metric spaces, which is then used to give a direct and geometric proof of the co-area formula for Lipschitz mappings into metric spaces.
Read the full abstract
High Energy Physics (Phenomenology)
Complex Scalar Dark Matter with a Vector-Like Quark and Lepton: Precision, Flavor, and HL-LHC
We investigate a minimal extension of the SM consisting of a $\mathbb{Z}_3$-stabilized complex scalar dark matter (CSDM) candidate, a down-type vector-like quark (VLQ), and a charged vector-like lepton (VLL). The additional vector-like fermions not only enable the CSDM to reproduce the observed relic abundance beyond the Higgs-resonance region through semi-annihilation and co-annihilation processes, but also induce correlated signatures across flavor, electroweak precision, dark matter, and collider observables. We perform a comprehensive one-loop analysis of neutral meson mixing, rare meson decays, charged lepton flavor violation, anomalous magnetic moments of charged leptons, and $Z$-pole observables. We find that neutral meson mixing and rare meson decays provide the dominant constraints on the VLQ sector, while charged lepton flavor-violating processes strongly restrict the VLL Yukawa couplings. Current direct-detection limits require Higgs-DM coupling $\lambda_{\Phi H} \lesssim5\times10^{-3}$ and the VLQ Yukawa coupling $\mathtt{y}_d \lesssim 0.05$ for TeV-scale VLQ masses, whereas present indirect-detection searches impose no additional constraints. Combining all flavor, electroweak precision, dark matter, and collider constraints, we identify viable parameter regions with CSDM masses above approximately $1.0~\rm TeV$ and VLQ masses above about $1.5~\rm TeV$. We also find that the LHC can exclude VLQs (VLLs) with masses up to approximately $1.6~(0.38)$ TeV, depending on the CSDM mass, at the $2\sigma$ confidence level. The HL-LHC can further probe an extended region of the parameter space, with discovery prospects at the $3\sigma$ level. Finally, we demonstrate the complementarity of flavor, dark matter, and collider searches in probing this framework.
Read the full abstract
Robotics
DeCAL: Towards Physically-Grounded Dexterous Vision-Language-Action Models via Contact-Aware Latent Co-Imagination
Dexterous manipulation involves contact-rich and fine-grained interactions with the physical world, posing significant challenges for existing vision-language-action (VLA) models due to severe visual occlusions and complex contact dynamics. While recent works have incorporated tactile sensing into robotic manipulation, most approaches still rely on homogeneous multimodal fusion, lacking adaptive tactile integration and explicit modeling of physical dynamics. In this work, we present DeCAL, a physically-grounded dexterous vision-language-action model that unifies understanding, imagination and action generation for contact-rich dexterous manipulation. Built upon a Mixture-of-Transformers (MoT) architecture, DeCAL leverages specialized experts for each capability while enabling efficient information flow among them. To effectively leverage tactile information, we introduce Adaptive Visuo-Tactile Fusion that dynamically regulates tactile interactions via a contact-aware gating strategy. Furthermore, we propose Visuo-Tactile Latent Co-Imagination to jointly model visual and tactile dynamics, equipping the policy with implicit physical world knowledge. Experimental results show that DeCAL consistently achieves state-of-the-art performance across all tasks, attaining a 71% average success rate and an 83.4% progress success rate, while also demonstrating strong generalization to unseen scenarios. The website is available at https://aureleopku.github.io/DeCAL.
Read the full abstract
Mesoscale and Nanoscale Physics
Topological Impurity Bands
The effects of disorder on topological phases of matter are typically either preservation of topology or its destruction. For example, the quantum Hall effect's quantization and chiral edge mode remain robust until a strong-disorder percolation transition exits the topological phase. Disorder can also enlarge a preexisting topological phase, yielding a topological Anderson insulator (TAI). However, the reliance of TAI and other theoretical treatments on perturbatively averaging out long-wavelength randomness suggests that additional effects might appear when the disorder involves localized defects. Here we show that a finite density of randomly distributed defects can generate various new topological phase transitions. This route to disorder-induced topology can be viewed as the formation of topological impurity bands (TIB) and is applicable even if the initial clean model is a trivial band insulator. The theory relies on an orbital-dependent vorticity of impurity bound states, which we derive analytically near a generic Dirac cone phase transition as well as using the Chern insulator lattice models of Qi-Wu-Zhang and Haldane. Their numerically computed phase diagrams at 3% defect density show robust topological transitions in distinct TIB and TAI regimes, suggesting a local route to topological phases.
Read the full abstract
Number Theory
The Archimedean place is a blurred interval at infinity
Classically, the places of $\mathbb{Q}$ are often regarded as a one-point compactification of $\mathrm{Spec}(\mathbb{Z})$, with the real place corresponding to a formal “prime” added at infinity. Re-examining this picture from a topos-theoretic perspective reveals a subtler geometry: while the non-Archimedean places are identified with singletons indexed by the non-zero prime ideals of $\mathbb{Z}$, the Archimedean place is represented by the space of upper reals $\overleftarrow{[0,1]}$, which may be informally thought of as the unit interval equipped with a non-Hausdorff topology. On a technical level, our analysis brings together geometric logic and descent techniques from topos theory, distinguishing standard descent from lax descent toposes both at the level of sheaves and of the geometric theories they classify. More broadly, this paper brings into conversation two parallel distinctions: on the number-theoretic side, between Archimedean and non-Archimedean phenomena, and on the topos-theoretic side, between standard and lax descent. Looked at from a high level, these perspectives begin to converge on a common theme: how should the connected and the disconnected interact?
Read the full abstract
Machine Learning
When Does Scale-Invariant Optimization Become Unstable? An Exact Schedule Law with Weight Decay
Normalization renders large parts of neural networks effectively scale invariant, inducing a hidden feedback loop in which learning-rate schedules and weight decay interact through the parameter norm to control the effective step taken by the optimizer. We show that this interaction is governed by an exact discrete-time law: a single scalar quantity captures all schedule and decay forcing, while norm growth induces an opposing geometric self-quenching effect. This yields a sharp boundary that cleanly separates contraction- and expansion-dominated effective learning rate regimes. To understand the underlying mechanism, we provide exact analysis of a fully solved normalized regression model where the dynamics reduce to two dimensions and show that the balance point is intrinsically unstable, implying that constant learning rate with weight decay cannot stably maintain an interior equilibrium and instead produces recurrent behavior driven by discrete-time Jacobian structure. We further extend this perspective across optimizers through unified homogeneous-optimizer framework that reveals a structural dichotomy in self-quenching strength, providing a first-principles explanation for why adaptive methods exhibit systematically weaker stabilization under normalization. Across dynamical systems and neural networks (MLP, CNN, GPT2 / MNIST, CIFAR, wikiText, OpenWebText), the predicted law holds with high precision and enables direct control of training via the identified scalar, with performance peaking sharply at the predicted boundary. Together, these results isolate a single governing quantity for scale-invariant optimization, providing a precise and actionable lens on training dynamics, optimizer behavior, and schedule design in modern deep learning. Code is available in https://github.com/shasanamin/normalized-optimization-dynamics.
Read the full abstract
Artificial Intelligence
MeClear: Cooperative Game-Theoretic Attribution and Risk-Aware Memory Clearance for Long-Horizon LLM Agents
Long horizon Large Language Model (LLM) agents rely on external memory systems to preserve user preferences and task knowledge across extended interactions. Conventional retrieval mechanisms optimize semantic compatibility rather than downstream utility, frequently introducing outdated, misleading, or conflicting evidence into the active context. We present MeClear, a task conditioned memory clearance framework that identifies memories featuring negative downstream utility through cooperative attribution and selectively suppresses them from agent execution. MeClear combines Leave One Out screening with sampled cooperative Shapley attribution to distribute utility across interacting evidence, effectively resolving redundant conflict masking where single removal evaluations fail. Utilizing attribution rankings, MeClear executes a query scoped minimal clearance strategy over a nested filtration, verifying task recovery on the cleared context without permanently altering the persistent memory bank. Comprehensive experimental evaluations across ten long dialogue memory pools demonstrate that MeClear achieves a target recall of 85.9% and an overall task recovery rate of 82.3%, representing a 25.5 percentage point improvement over Leave One Out (LOO) baselines.
Read the full abstract
Nuclear Theory
Hydrodynamic attractors
Attractors are effective low-dimensional structures, defined in a chosen set of observables, that trajectories from different initial states approach; hydrodynamic attractors are those on which the late-time evolution is governed by hydrodynamic constitutive relations. Motivated by nuclear collisions and ultracold atomic gases, this chapter distinguishes attractorization (the loss of sensitivity to some directions in the space of initial states) from hydrodynamization (the onset of validity of hydrodynamic constitutive relations). Using mainly conformal, boost-invariant Bjorken flow, we compare Müller-Israel-Stewart-type theories, kinetic theory, holography, and classical Yang-Mills fields. Forward attraction, produced by the decay of non-hydrodynamic modes, is distinguished from pullback attraction, which can select a unique regular solution, and from expansion-driven attraction, which can suppress initial-state sensitivity before microscopic relaxation, with or without a subsequent fluid regime. Divergent gradient expansions and their transseries completions, the state-space picture, and adiabatic hydrodynamization provide complementary descriptions. Applications to nuclear collisions include particle production, transverse energy and flow, the initialization of and linear response around attracting backgrounds, jet quenching, and attractor-informed modifications of hydrodynamic models. For ultracold Fermi gases, we review theoretical proposals for bulk-channel attractors under scattering-length drives and distinguish them from recent measurements of short-time contact and momentum-distribution dynamics.
Read the full abstract
Artificial Intelligence
SAEScientist-Bench: Can AI Agents Conduct Autonomous SAE Interpretability Research?
While research on recursive self-improvement (RSI) has predominantly automated model training pipelines, reliable autonomous development demands a missing pillar: post-hoc monitoring and auditing to understand what models learn and ensure safe alignment. Mechanistic interpretability tools are essential to bridge this gap, among which Sparse Autoencoders (SAEs) serve as a cornerstone by isolating interpretable features for model inspection and steering. In this paper, we introduce SAEScientist-Bench to evaluate whether AI agents can act as scientists utilizing SAE tools for autonomous mechanistic discovery. Given a target concept, an agent designs contrastive probes and navigates a Gemma Scope dictionary of 131K+ features in Gemma-2-9B-IT to discover the optimal feature, evaluated against curated expert reference features anchored on Neuronpedia across activation rank, concept selectivity on contrastive texts, and causal steering. Across 10 agent configurations and 20 tasks, frontier agents demonstrate genuine discovery capabilities and lead different evaluation dimensions, but remain well behind the expert baseline, approaching expert levels on separating target concepts from contrastive controls while lagging substantially in causal generation steering. Further analysis reveals that although agents can design contrasts to rule out spurious candidates, they frequently misinterpret experimental measurements. These results establish experimental model understanding as a measurable capability for closed-loop autonomous AI R&D. Our code is available at https://github.com/Trae1ounG/SAEScientist.
Read the full abstract
Human-Computer Interaction
Travel Package Booking Application with API Bot
These days we are witnessing many mobile applications based on the recommended systems, which have become a great technology which is been used by the various mobile applications according to the situation. Recommendation provided by the mobile application is a key element for the person who is traveling to several places. For any tourist information application contextual information is much needed to guide the user on his interests this can be achieved by the Context-aware computing. Which provides the user most interactive system with the suggestions provided by it based on the input from the user in a certain location, here context includes the user's mental, social, physical environments. To achieve this contextual information, we will design and implement the context-aware user interface based on the user for which we have to study the user and design a rich user interface. The final outcome for which users have the satisfaction when using context-aware functionality will be much better than non-context-aware application.
Read the full abstract
Representation Theory
Symplectic Dirac operators on homogeneous spaces
We define symplectic Dirac operators on homogeneous spaces and study their representation-theoretic role. For an invariant polarization, the symplectic Dirac operator decomposes into two symplectic Dolbeault operators. We compute their commutator as the natural symplectic analogue of the square of the classical Dirac operator. Our first main result gives a necessary and sufficient condition for this commutator to satisfy a Parthasarathy-type formula. We further prove that, whenever this condition fails, no cubic perturbation of the symplectic Dolbeault operators can yield such a formula, in contrast with Kostant's cubic Dirac operator in the orthogonal setting. As applications, we establish an ${\mathfrak s}{\mathfrak l}_2$-structure generated by the symplectic Dolbeault operators and derive Dirac-type inequalities for unitary representations of Hermitian symmetric spaces labelled by the levels of the symmetric powers of the antiholomorphic tangent space at the identity. The level-zero inequality recovers the standard Parthasarathy-Dirac inequality, while the higher levels inequalities yield new constraints. For $SU(1,n)$, we show that, for representations with a specific Kraljević corner, the level-one inequality strengthens all basic Parthasarathy inequalities of the first kind for particular $K$-types, precisely those satisfying an explicit highest-weight condition.
Read the full abstract
Metric Geometry
On universal elements for doubling geodesic trees
For $n\ge 3$ and $c\in(0,1)$, let $\mathcal{GT}(n,c)$ denote the class of geodesic metric trees of valence at most $n$ whose branch points are uniformly relatively separated with constant $c$. We prove that $\mathcal{GT}(n,c)$ has no bi-Lipschitz universal element. More precisely, we construct a family $(T_a)_{a\in[1/4,1/3]}\subset\mathcal{GT}(n,c)$ such that, for every $n_M\geq 3, c_M\in(0,1)$ and every $M\in\mathcal{GT}(n_M, c_M)$, there are at most countably many parameters $a$ for which $T_a$ admits a bi-Lipschitz embedding into $M$, whereas each $T_a$ admits a bi-Lipschitz embedding into $\mathbb R^2$. Thus the obstruction is neither dimensional nor caused by a failure of planar embeddability. This gives a negative answer to a question of Chrontsios-Garitsis, Ioannidis, and Vellis~\cite[Question~1.11]{CGIV2024}. Furthermore, we show a complementary positive result for ultrametric spaces: every bounded ultrametric space $X$ admits a bi-Lipschitz embedding into every complete metric space $Y$ satisfying $\dim_A X<\dim_{LA} Y$ where $\dim_A X$ and $\dim_{LA} X$ are Assouad and lower Assouad dimensions, respectively.
Read the full abstract
Combinatorics
Catalan-many tropical morphisms to trees; Part II: A space and a count
In their work on Brill-Noether theory, Eisenbud and Harris established the geometry of the universal parameter space of linear series over curves, proving that for even genus $g$ and degree $d = g/2 + 1$, the projection to the moduli space of curves is a finite cover of degree equal to the Catalan number $C_{g/2} = \frac{1}{g/2+1}\binom{g}{g/2}$. In this paper, we construct the tropical counterpart of this universal family: a polyhedral cone complex $\mathcal{G}_{g \to 0, d}^{\mathrm{trop}}$ parametrizing degree-$d$ tropical morphisms from genus-$g$ metric graphs to metric trees. For even $g$ and $d = g/2 + 1$, we prove that the forgetful projection $\Pi \colon \mathcal{G}_{g \to 0, d}^{\mathrm{trop}} \to \mathcal{M}_{g}^{\mathrm{trop}}$ is a branched cover of degree $C_{g/2}$ equipped with natural determinantal multiplicities. We compute this degree by showing that on caterpillars of loops the morphisms are in bijection with ballot sequences, and we establish its global invariance across $\mathcal{M}_{g}^{\mathrm{trop}}$ via a tropical balancing condition across codimension-$1$ walls. Via deformation and path lifting, this yields an effective method to construct Catalan-many gonality-witnessing maps for any generic metric graph, establishing that the tree gonality of any genus-$g$ metric graph is at most $\lceil g/2 \rceil + 1$.
Read the full abstract
Optimization and Control
Distributed Linear Programming on GPU Clusters at Extreme Scale
Large linear programs can exceed the memory of a single compute node. Although first-order methods replace sparse factorizations with GPU-suited matrix-vector products, other solver phases can reintroduce a single-node memory limit. We present SHARDLP, a distributed GPU LP solver that keeps the matrix and primal-dual state partitioned from sharded input through solution output. On the Google PDLP benchmark, SHARDLP solves ten of eleven instances, compared with eight in the published CPU PDLP study. On the largest benchmark, eight H200 GPUs solve a 1.185-billion-variable, 6.338-billion-nonzero LP in 9.9 minutes; the published CPU experiment reports 21.06 hours on different hardware. Beyond this benchmark, separately checked multi-node solves reach up to 13.604 billion variables and 40.807 billion nonzeros, while validated executions span up to 76 GPUs across 29 compute nodes. For column-partitioned solves, support-aware communication skips GPUs that store no coefficients for a row; on an LP with 2.76 billion nonzeros, it cuts modelled communication by 92.97% and improves solver time by 1.27x-1.52x. The code for SHARDLP is available at https://github.com/ArnaudDeza/ShardLP.
Read the full abstract
High Energy Physics (Phenomenology)
For Whom the Xenon Recoils: Magnetic Inelastic Dark Baryons
We develop a theory of a dark baryon dark matter candidate that interacts with nuclei dominantly through inelastic scattering mediated by a transition magnetic dipole operator. Models are presented that can elegantly explain both the scattering rate and the size of the inelastic splitting to be consistent with the one event observed at LZ\@. Elastic scattering is suppressed due to accidental symmetries within the strongly-coupled sector. The nuclear response is dominated by a spin-dependent structure factor, thus we find a large range of masses, $1 \; {\rm TeV} \lesssim m_\chi \lesssim 50 \; {\rm TeV}$, and mass splitting of $100 \; {\rm keV} \lesssim \delta \lesssim 500 \; {\rm keV}$ can fit the data for a variety of possible dark matter velocity distributions. The models predict that some nuclear scattering events will be accompanied by a simultaneous photon signal with energy $\delta$, which not only distinguishes this model from other inelastic dark matter explanations but also enables directional detection in conventional direct detection experiments. We briefly comment on the dark matter abundance, indirect detection signals (including the models' significantly weaker constraints from dark matter annihilation in the Sun), and the associated collider signals that arise from the meson sector of the theory.
Read the full abstract
Cosmology and Nongalactic Astrophysics
A Two-Dimensional Test of Cosmological Parity Violation with the 3-Point Correlation Function
Searches for parity violation (PV) in the Universe's large-scale structure have recently been carried out using parity-odd modes of the galaxy 4-Point Correlation Function (4PCF). Here, we introduce a new estimator for a 3-Point Correlation Function (3PCF) that is sensitive to PV. The construction exploits the handedness of triangles formed on 2D spherical shells, where mirror-image configurations cannot be related by proper rotations within the surface. By dotting the oriented area vector with an externally-defined normal vector to the shell, we obtain sensitivity to PV. We present a discrete estimator for this statistic and also show how to correct for the survey geometry. This parity-odd 3PCF provides a complementary test of PV in galaxy surveys and a potential cross-check of PV 4PCF measurements.
Read the full abstract
Combinatorics
A hypercontractive proof of the sharp bound for cancellative pairs
Fang and Huang proved that every cancellative pair $(\mathcal{A},\mathcal{B})$ of families of subsets of $[n]$ satisfies $|\mathcal{A}|\cdot|\mathcal{B}|\leq(\frac{9}{4})^n$. Their proof uses entropy. We give a Fourier-analytic proof based on Lifshitz's one-sided noise operator $\mathrm{T}^{1/4\to1/2}$. The main tool is a near-$L^1$ hypercontractive estimate for this operator.
Read the full abstract
High Energy Physics (Phenomenology)
Two-Loop Anomalous Dimension for Non-Global and Clustering Logarithms
We derive the two-loop anomalous dimension relevant for the resummation of subleading non-global and, for the first time, also clustering logarithms within the framework of effective field theory. Special emphasis is put on the choice of the renormalization scheme. We show that the modified minimal subtraction scheme is problematic since it requires extra dimensions to set up the parton shower to perform the resummation. We discuss a set of modified dipole subtraction schemes which are free from this complication and can be used to obtain a frame-independent result for the renormalization-group evolution. We implement the leading-color limit of our result in the Marzili framework and numerically verify the scheme independence for gap-between-jets cross sections.
Read the full abstract
Probability
Quantitative Parisi formulas and fluctuations in the Sherrington-Kirkpatrick model
We establish quantitative Parisi formulas and fluctuation bounds for the Sherrington-Kirkpatrick model at zero external field. In particular, using that the Parisi measure is supported on an interval, we show that for every fixed inverse temperature greater than one, the variance of the logarithmic partition function lies between orders $N^{4/15}$ and $N^{7/15}$. The same exponents hold for the ground-state variance. We also obtain upper and lower bounds on the finite-size bias of the mean free energy and ground-state energy, together with two-sided upper-tail estimates with exponent $6/5$ at both positive and zero temperature.
Read the full abstract
Cosmology and Nongalactic Astrophysics
TILING I: Field-level Bayesian reconstruction of cosmological initial conditions during the epoch of reionization
Reconstructing the initial conditions (ICs) of the matter field for an observed volume gives a complete picture of that region's temporal evolution. IC reconstruction is common at low redshifts but largely unexplored during the epoch of reionization (EoR; $z\gtrsim 5$), partly due to difficulties modeling inhomogeneous cosmic radiation fields. Yet the EoR spans half the observable Universe, offering unmatched potential for astrophysics and cosmology. Here we quantify how well upcoming galaxy and 21cm observations can constrain ICs during the EoR. We develop $\tt TILING$ ($\bf T$omographic $\bf I$nference of $\bf L$inear $\bf I$Cs via $\bf N$etwork $\bf G$rafting): a hybrid machine learning pipeline that first produces a point estimate of the ICs, which then improves a score-based diffusion network for generating posterior samples. We train $\tt TILING$ on mock galaxy maps at varying UV magnitude limits, plus corresponding 21cm maps at varying noise levels and foreground "wedge" contamination. For fiducial survey choices, we achieve accurate IC reconstruction (posterior mean cross-correlation coefficients >0.8 and power spectrum errors under a few percent) down to $\text{k}\lesssim 0.2 \text{cMpc}^{-1}$. While 21cm interferometry aids recovery of IC power spectra, most constraining power, especially for IC Fourier phases, comes from galaxy maps, underscoring the need for complementary observations when interpreting the 21cm signal. We show how $\tt TILING$ can recover 21cm power spectrum modes excised by foreground contamination. Our framework can also: (i) guide follow-up observations of sub-volumes of interest; (ii) reconstruct galaxy evolution and reionization morphology for specific volumes; and (iii) isolate the contribution to reionization from the vast majority of galaxies unobservable by optical/IR telescopes like $\it JWST$. Our code is publicly available.
Read the full abstract
Analysis of PDEs
The Trudinger inequality is true for $M^{1,s}$ spaces
We prove the Trudinger inequality with exponent $\frac{s}{s-1}$ for $M^{1,s}$ Sobolev spaces on metric measure spaces under the sole measure growth assumption $\mu(B(x,r))\ge br^s$. This improves the previously known exponential integrability with power one. Our argument also yields sharp asymptotic bounds for the Sobolev constants as $p\uparrow s$. On doubling spaces, we further remove the connectedness assumption from the Trudinger inequality associated with a $(1,p)$-Poincaré inequality when $p<s$. An Ahlfors regular counterexample shows that this extension fails at the critical exponent $p=s$.