Artificial Intelligence
Papers
Fresh from arXiv, typeset properly. Highlighted words trade on the market.
All of arXiv
Thursday 5 November
Scaling test-time compute has proven highly effective at improving LLM reasoning. A popular method is best-of-n sampling, where a model generates many candidate solutions and one is selected. This motivates the question: how can we select the best candidate? When humans make choices, we find it more natural to compare two options than judge each in isolation. Additionally, when the options become complex, we like to judge the options one aspect at a time, such as comparing two job candidates on experience, communication, etc. In the test-time setting, candidate solutions should be judged in the same way. Here we introduce Aspect-wise Pairwise Tournaments (APT), which verifies by comparison and compares by aspect. Each matchup of a single elimination tournament is decided by a vote over multiple off-the-shelf aspect verifiers, each expressing a pairwise preference along its own aspect. Across five domains and three generator LLMs, APT improves over binary multi-aspect approval, self-consistency and reward model verification, with the largest gains on open-ended tasks. We find that both components matter; neither binary approval over the same aspects nor a generic pairwise judge at a matched budget matches APT. Because aspects are specified in natural language, APT also lets users steer selection toward properties they care about, better aligning with user preferences without needing to regenerate solutions. No component is trained, so APT transfers to new domains by simply rewriting prompts.
Computation and Language
Forcing a Letter Suppresses Refusal: Mechanistic Evidence from WMDP
Safety benchmarks for large language models (LLMs) commonly use multiple-choice question (MCQ) format. However, prior studies documented that MCQ format reduces refusal rates, which may significantly affect the safety behavior of the model. We provide a first mechanistic explanation of this effect on the Weapons of Mass Destruction Proxy (WMDP) dataset across two models - Llama-3.1-8B and Qwen3-8B. First, we use the tuned lens technique to show that when processing MCQ the model commits to an answer-letter token very quickly, which is in contrast to the gradual refusal build-up in open-ended formats. Additionally, we observe that activations on the MCQ format lie roughly 3x further from the refusal direction, compared with their open-ended counterparts. Moreover, ablating the attention heads that carry the MCQ format answer increases the refusal, which is a direct evidence that the MCQ answer-selection competes with refusal. These findings indicate that MCQ safety scores primarily measure knowledge capabilities and do not assess the safety-mechanisms.
Read the full abstract
Operating Systems
MORTIS: Compiler-Guided KV-Cache Management for LLM Agent Programs
Existing agent serving systems divide each prompt into a reusable prefix and an opaque suffix, managing the KV-cache at this coarse granularity without distinguishing the components within each part. We observe that prompt components differ in which LLM calls share their values and how long those values remain reusable, a property we term intra-prompt heterogeneity. Our key insight is that, in frameworks that generate prompts from typed LLM function calls, these components correspond to program variables, allowing compiler analysis to recover their sharing patterns and lifetimes. We present MORTIS, a compiler-serving co-design that combines this analysis with runtime observations to improve KV-cache management. MORTIS reorders prompt components to expose additional prefix reuse and uses their predicted reuse and lifetimes to guide cache retention, eviction, and prefetching. Our evaluation shows that MORTIS achieves up to 1.99x job completion time (JCT) speedup and 1.97x throughput improvement over SGLang, with geometric-mean speedups of 1.45x and 1.49x, respectively.
Read the full abstract
Optimization and Control
Branch-Extended Automatic Differentiation
We present branch-extended automatic differentiation (BEAD), a software package that provides information about inactive branches of piecewise-differentiable functions, intended for use in nonsmooth optimization algorithms. BEAD is implemented in Python, building on the popular JAX framework. For supported nonsmooth operations such as abs and max, BEAD identifies alternative branches within tolerance from the given input, and computes gradients of the resulting branches extended to the given input. Such information has been previously used for robust optimization methods, but has required error-prone and tedious manual differentiation, unlike BEAD, which is fully automated for functions written in NumPy-like JAX code. Our experiments show that our approach is as accurate as manually written derivative code, and offers more flexibility in modifying the function and choosing between branch-discovery rules. While we were originally motivated by the manifold sampling algorithm for nonsmooth composite objectives, we also integrate BEAD branch-extension gradients into a nonsmooth BFGS-SQP solver. On two small nonsmooth test problems, the resulting path-aware search resolves failures or substantially reduces the number of evaluations required by the BFGS-SQP solver.
Read the full abstract
Cosmology and Nongalactic Astrophysics
Nonstandard Dark Matter at Nonlinear Scales
We investigate how two extensions of the standard $\Lambda$CDM scenario affect the evolution of the matter density and velocity divergence fields, from the linear regime to deeply nonlinear scales, using $N$-body simulations. We consider two well-motivated scenarios in which the canonical properties of dark matter (DM) are modified. In the first scenario, DM interacts with dark energy (DM–DE interaction). In the second scenario, we allow the DM equation-of-state parameter to deviate from the canonical pressureless limit, $w_{\rm DM}=0$. For the coupled DM–DE scenario, we find significant and scale-dependent deviations from $\Lambda$CDM of order $50$–$100\%$ in the matter power spectrum, persisting throughout the nonlinear regime, with the velocity divergence spectra providing complementary and partially independent sensitivity. For the $w_{\rm DM} \neq 0$ scenario, we find systematic but remarkably small deviations confined within $5\%$ across all scales probed. Our results identify which scales carry the greatest discriminatory power between these nonstandard scenarios and $\Lambda$CDM. We show that testing these models with current and forthcoming LSS surveys, through the analysis of density and/or velocity fields, will require a deeper theoretical understanding and accurate nonlinear modelling before robust constraints can be derived from observational data.
Read the full abstract
Analysis of PDEs
Regularity of elliptic systems with Uhlenbeck structure
We study autonomous elliptic systems with Uhlenbeck structure of the form \[ -\operatorname{div}\big(\mathfrak a(|Du|)Du\big)=0 \quad \text{in }\ \Omega\subset \mathbb R^n \] Under two-sided quantitative $(p,q)$-growth conditions on $\mathfrak a$, we prove gradient Hölder continuity with constants and exponent depending only on the dimensions, $p$ and $q$. This improves the classical theory initiated by Uhlenbeck by removing the continuity requirement on $\mathfrak a'$. The proof uses nonlinear comparison systems whose regularity follows from a scalar uniformly elliptic equation. Under weaker $(p,q)$-growth conditions on the convex function $\varphi(t):=\int_0^t s\mathfrak a(s)\,ds$, we obtain local $W^{1,s}$-regularity for every $s>1$, and hence Hölder continuity of solutions with every exponent less than one. An additional lower growth condition on $\mathfrak a$ yields local Lipschitz continuity, even when $\mathfrak a$ is discontinuous.
Read the full abstract
Machine Learning
Benchmarking Machine Learning for Chaotic Three-Body Dynamics in Astrophysics
Machine learning (ML) and generative AI have recently made impressive progress in understanding physics. We explore a hierarchy of prediction problems stemming from chaotic three-body systems and benchmark existing ML approaches, including general-purpose prediction algorithms and those that build on top of foundational physics models. We first simulate the trajectories of triple systems near the critical boundary of stable/unstable initial conditions. We then design a hierarchy of learning tasks, focusing mainly on two broad problems: stability prediction and trajectory prediction. The former classifies whether a given triple system has diverged or not up to a specific horizon, while the latter predicts the exact trajectory, both given suitable input features. We benchmark a range of ML approaches, from simple regression methods to state-of-the-art foundational physics models, across these tasks, and compare them against astrophysical methods such as the Mardling-Aarseth stability criterion. Our experiments reveal a clear order of task difficulties in chaotic three-body dynamics: early dynamics enable strong stability prediction, while precise long-horizon trajectory forecasting remains difficult. Code and dataset are publicly available at https://github.com/Simonwan03/ThreeBodyMLBench.
Read the full abstract
Cosmology and Nongalactic Astrophysics
Short-term optical monitoring of quadruply imaged quasars
We present two case studies to demonstrate that short-term optical monitoring campaigns of quadruply imaged quasars (quads) reveal their astrophysical potential. Using the Liverpool Telescope and the third facility of the Two-meter Twin Telescope (TTT3), we monitored the quads QSO 2237+0305 and PS J0147+4630 in the gr bands between July 2025 and January 2026. For QSO 2237+0305, the resulting light curves revealed microlensing variability over a timescale of ~150 days. The g-band difference light curve between images A and B was then used to reliably constrain the half-light radius of the g-band source within the 3.5-16 lt-day range (1{\sigma} confidence level). For PS J0147+4630, we derived new constraints on the AB and AC time delays from the r-band light curves of the ABC images in an arc-like configuration, obtaining 2{\sigma} uncertainties of ~2 days. These constraints were used to shed light on the lensing mass distribution. Overall, our results validate the design of short-term optical monitoring campaigns of quads as valuable astrophysical tools in the era of large-scale surveys. The proof of concept relies on observations with ground-based 2mclass telescopes that lack sophisticated systems to correct for optical distortions caused by atmospheric turbulence, and we indicate the astro-photometric properties of the targets and the monitoring strategies needed to achieve success with these telescopes. Additionally, this work represents the first validation of the TTT3 facility for precision photometric monitoring of gravitationally lensed quasars.
Read the full abstract
Quantum Physics
How Many Shots Does a Quantum Reservoir Need? Measurement Cost and Its Scaling in Buffered Quantum Reservoir Computing
Quantum reservoir computing (QRC) inherits from classical reservoir computing the property that only a linear readout is trained, but unlike a classical reservoir its state is never directly observable: every feature is an expectation value estimated from a finite number of measurement shots. The shot budget is therefore a first-class hyperparameter with no classical analogue. Its effect on QRC has recently begun to be studied; what remains unsettled is how a required budget should be defined, and whether competing definitions agree. This paper argues that they do not, and that the disagreement is systematic rather than incidental. This paper distinguishes a precision budget $N_\varepsilon$, the shots needed to estimate the reservoir features to a fixed relative accuracy, from a convergence budget $N_\star$, the shots needed for the trained readout to approach its own noise-free ceiling. Proposition~\ref{prop:confound} shows that $N_\star$ depends on the noise-free error floor by construction, so that a configuration with a worse floor appears cheaper, and that configurations should therefore be compared at a fixed measurement budget rather than by the budget each needs to reach its own ceiling.
Read the full abstract
Computer Vision
BEVCast: Path-Adaptive BEV for Efficient End-to-End Autonomous Driving
Camera-based driving models build a dense bird's-eye-view (BEV) grid and spend the same computation on every cell of it, although the vehicle crosses only a few of those cells within the planning horizon. We present BEVCast, a two-stage BEV whose resolution adapts to the predicted driving path. A coarse pass encodes the whole scene on a grid of 4.096 m cells. A heat head trained on the recorded path assigns a heat to every coarse cell and the cells with the highest heat are refined into 1.024 m cells under a fixed budget. The remaining cells keep their coarse resolution, so the trajectory head reads a mixed-resolution BEV that still covers the entire scene and needs neither an HD map nor a navigation command. Against a full grid of 1.024 m cells with the same architecture and training recipe, BEVCast at a 10% refinement budget encodes 28% of the BEV queries and needs 66% of the BEV encoder time at batch size 1. On nuScenes its mean displacement error is 1.078 m against 1.149 m and its drivable area violation rate is 1.46% against 2.29%. On NAVSIM navtest its mean displacement error is 0.577 m against 0.598 m and its extended predictive driver model score is 82.51 against 81.28. Next to published planners BEVCast reaches a comparable error on both benchmarks without a navigation command. The resolution of a driving BEV can therefore follow the task instead of the grid.
Read the full abstract
Cosmology and Nongalactic Astrophysics
VENUS: Strong Lens Mass Models and Einstein Radii of 60 Galaxy Clusters
We present the strong lens mass modeling result of 60 galaxy clusters targeted by Vast Exploration for Nascent, Unexplored Sources (VENUS), the James Webb Space Telescope (JWST) multi-epoch Treasury program. We obtain refined mass models for the 60 clusters with the glafic software by fitting positions of $\sim 4500$ multiple images, $\sim 1150$ of which are newly identified multiple images with help of multi-band JWST images. Each cluster contains typically $\sim 50$ multiple images, thanks to the selection of target clusters based on their high lensing strengths. We validate our models with the Jackknife method. The median Einstein radius is $\sim 30''$ and the distribution of the Einstein radii extends out to $\sim 60''$ for the source redshift of 9, which confirms that their lensing strengths are indeed among the highest known to date. For the source redshift of 9, we identify MACS J0717.5+3745, MACS J0600.1$-$2008, and PLCK G287.0+32.9 as clusters with the largest Einstein radii as well as the largest areas of the high-magnification ($|\mu|>10$) regions. We find that the observed Einstein radii of VENUS clusters are fully compatible with the standard cold dark matter model.
Read the full abstract
Systems and Control
Multi-objective Electric Vehicle Charging Strategies at Energy Hubs with Justice-informed Profit Distribution among Users
This paper studies potential energy cost savings in electric vehicle (EV) charging hubs when users allow cooperative vehicle-to-grid (V2G), and how such savings should be redistributed. To this end, we present multi-objective operational strategies and justice-informed profit sharing methods for a charging energy hub equipped with a battery storage system and solar panels. Specifically, we first frame the dynamic energy hub operational problem aiming to minimize a combination of grid costs and emissions, accounting for time-varying prices and energy mix as well as storage system and EV battery degradation, which is especially relevant when considering V2G operations. We parse the problem as a second-order conic program that can be efficiently solved to global optimality. Our real-world case-study shows the importance of flexibility for hub operation: Equipping a charging station with a stationary battery and leveraging dynamic V2G can significantly reduce the total costs and emissions compared to conventional settings without a battery or dynamic charging by more than 10%. To redistribute these cost benefits to users, we compare standard Nash bargaining with a justice-informed sufficientarian version which caps charging prices to safeguard adequate access to affordable charging. Our results show that the sufficientarian approach better accounts for inequalities in flexibility capital without significantly affecting aggregate costs, highlighting how justice-informed pricing only marginally affects total revenues.
Read the full abstract
High Energy Physics (Theory)
NLO corrections to charged-current correlators near extremality: non-Abelian hydrodynamics meets AdS$_2$
We analytically compute the longitudinal charged-current correlator in a near-extremal holographic plasma with quark and isospin chemical potentials, $\mu_q$ and $\mu_3$ respectively. To this end, we study fluctuations of gauge fields around a low-temperature AdS$_5$ Reissner-Nordström black hole charged both under the quark number and the isospin. The result is obtained at next-to-leading order in a hydrodynamic expansion where apart from the temperature and isospin chemical potential, the frequency and momenta are small compared to the quark chemical potential, $\omega,k,T,\mu_3\ll\mu_q$. This approach captures the interplay between the hydrodynamic-like pole and the tower of infrared modes produced by the AdS$_2$ throat controlling the geometry of the black hole near extremality. At zero isospin chemical potential, the resulting quasinormal-mode formula reproduces the collision of the diffusive pole with the infrared spectrum, while the finite isospin chemical potential lifts this collision. The analytic spectrum agrees precisely with the full numerical results. The next-to-leading-order correlator also substantially improves the spectral function beyond the standard hydrodynamic regime $\omega,k,\mu_3\ll T$, extending the agreement with numerics to larger frequencies and momenta.
Read the full abstract
Combinatorics
Evasive subspaces, minimal rank-metric codes and subspace designs via auxiliary fields
We extend the auxiliary-field method for scattered subspaces to evasive subspaces and subspace designs. For $1\le h\le k<hn$ and coprime $h,k+1$ with $n(k-h+1)\ge k+1$, we construct spanning $(h,k)$-evasive $\mathbb{F}_q$-subspaces of $\mathbb{F}_{q^n}^r$ of dimension $rn(k-h+1)/(k+1)$ in every sufficiently large admissible ambient dimension, with a threshold independent of $q$. When $k<n$, this dimension is maximum. The proof combines an intersection lemma, a rank-support density inequality, scalar orbit counting, and a profile preserved by direct sums. The coding consequences include, for every fixed odd extension degree $m\ge3$ and every sufficiently large $K$, minimal rank-metric codes of optimal length $2K$ with first two generalized weights $(m+1)/2$ and $m+1$. For these codes the stable profile is equivalent to minimality with the largest possible minimum distance; for even $m$ the analogous class consists of minimal MRD codes. More generally, for $1\le\tau\le m-2$ and $\gcd(m,\tau+1)=1$, we obtain nondegenerate $\tau$-minimal codes of length $(\tau+1)K$ for every sufficiently large $K$, uniformly in $q$. We also construct maximum $h$-designs whose number of members grows exponentially with the ambient dimension at fixed $q,n,h$, and the associated maximum sum-rank distance (MSRD) codes with a growing number of equal long blocks. In extension degree five, the ordinary and $2$-minimal constructions hold for every $q$ and every $K\ge32$ and $K\ge59$, respectively.
Read the full abstract
Artificial Intelligence
LLM-Based Feature Extraction for Multi-Criteria Inventory Classification
E-commerce supply chains generate large volumes of customer feedback and operational data that remain underutilized in inventory management. Traditional inventory classification techniques such as ABC analysis rely primarily on annual dollar usage, while multi-criteria inventory classification (MCIC) methods extend this but still depend on structured quantitative data and do not utilize unstructured information available in modern supply chains. This paper proposes an LLM-based feature extraction approach that transforms unstructured sources such as customer reviews into quantitative features using vector embeddings and LLM-generated evaluation criteria. The enriched feature set is incorporated into an MCIC framework based on the Hadi-Vencheh model. Experimental results show measurable reclassification effects when customer sentiment signals are included, demonstrating that integrating unstructured data into MCIC can enhance inventory prioritization.
Read the full abstract
Dynamical Systems
Singularities and Collision Criteria in the Gravitational Straight-Segment Problem with Arbitrary Positive Density
We study singularities and collision criteria for a massless particle around a fixed finite straight segment with an arbitrary positive continuous density. We first show that every singularity forces the trajectory to approach the material segment. Using the axial symmetry of the potential, we prove that any trajectory approaching the segment is confined to an invariant plane containing it. Moreover, we show that every singularity is either a collision singularity or an ejection singularity. Finally, in the invariant plane, we introduce an angular-cone condition for positions and momenta directed toward the segment. We prove that if the initial position-momentum pair belongs to the angular cone, then the trajectory collides with the segment in finite time, and we derive an upper bound for the collision time. We further prove that the final arc of any collision satisfies the angular-cone condition.
Read the full abstract
Analysis of PDEs
Nonzero solutions for Dirichlet problems with the competing (p,q)-Laplacian
The following equation is considered: $$-\Delta_p u(\boldsymbol{x}) + \Delta_q u(\boldsymbol{x}) = f\big(\boldsymbol{x}, u(\boldsymbol{x})\big)\text{ for } \boldsymbol{x} \in \Omega, \qquad u(\boldsymbol{x}) = 0 \text{ for } \boldsymbol{x}\in \partial \Omega.$$ We start by showing that the energy functional associated with the considered problem no longer has the properties typical of its non-competing counterpart, which prevents a direct application of standard variational and monotonicity methods. Using Ekeland's variational principle combined with properties imposed on the nonlinear term~$f$, we prove the existence of a nonzero generalized solution to the problem. It is also shown that in the one-dimensional case $\Omega = (-1,1)$, assuming that $f$ satisfies the symmetry conditions $$f(x,u) = f(-x,u) = -f(x,-u),$$ together with the sign condition $f(x,u) \geq 0$ for $u \geq 0$ and an appropriate growth bound on $f$, a convex relaxation of the action functional yields the existence of a positive weak solution. Furthermore, we prove that in this setting, and under a slightly stronger sign condition $$f(x,u) > 0\text{ for }u > 0,$$ all global minimizers of the action functional do not vanish in $(-1,1)$ but fail to be of class $C^1$.
Read the full abstract
Probability
A tractable transition function for the Wright-Fisher diffusion with selection via gamma duality
The transition function of the Wright-Fisher diffusion with selection is a fundamental object for non-neutral evolutionary modelling and likelihood-based inference, but, unlike in the neutral case, existing representations are impractical or inefficient. Spectral approaches rely on truncation levels that grow with selection strength and the number of alleles, thus leading to rapidly increasing computational costs, while existing duality-based expansions lead to multi-type dual processes whose transition probabilities involve intractable normalising constants. In this paper we obtain a new transition-function representation for the multi-allelic Wright-Fisher diffusion with parent-independent mutation and genic selection that makes transition-law evaluation and simulation practically feasible in regimes where existing methods are costly or unavailable. The key tool is a new gamma duality, obtained by augmenting classical polynomial duality by an exponential tilting and yields a multivariate birth-death dual with fully explicit infinitesimal rates and an entrance-boundary construction. This is used to express the Wright-Fisher transition law as an explicit mixture of standard Dirichlet components, with mixing weights obtained from the dual started from its entrance boundary. We characterise the dual stationary distribution and its small-time behaviour, and develop practical and efficient sampling procedures for both transition and stationary laws, including a fast approximate stationary sampler with explicit error control. Numerical experiments illustrate substantial runtime gains over specialised methods in regimes where they apply, while our approach remains applicable for arbitrary numbers of alleles and selection coefficients.
Read the full abstract
Optics
Modulation Induced Memory Resonances in Dynamically Driven Optical Cavities
Temporal modulation of optical cavities provides a powerful mechanism for controlling photonic states beyond the static cavity paradigm. Here, we investigate the dynamics of periodically driven optical cavities using a time-dependent cavity-field model with modulated resonance frequencies. We demonstrate modulation-induced tuning of cavity resonances and analyze the resulting intracavity field evolution in both amplitude and phase. The interplay between periodic modulation and cavity dissipation gives rise to pronounced hysteretic behavior, revealing memory effects associated with the finite photon lifetime of the cavity. By quantifying the hysteresis area as a function of modulation frequency, we identify distinct memory resonances characterized by enhanced field-history dependence at specific driving frequencies. Furthermore, we examine the time-resolved local density of optical states (LDOS) and show that temporal modulation enables dynamic spectral redistribution of photonic states, with the degree of confinement governed by the cavity loss rate. Our results establish a simple framework for understanding and engineering memory effects in dynamically driven optical resonators and provide new opportunities for tunable photonic environments, dynamic LDOS control, and reconfigurable light-matter interactions.
Read the full abstract
High Energy Physics (Phenomenology)
Opening up derivative baryon-number-violating operators
Baryon number violation is our most sensitive probe of physics beyond the Standard Model, typically encoded in higher-dimensional effective operators. Operators involving derivatives are often ignored in phenomenological analyses since they are generically suppressed compared to non-derivative operators. We systematically construct tree-level UV completions of derivative baryon-number-violating operators up to mass dimension 12 in the Standard Model effective field theory (extended by right-handed neutrinos) and show that derivative operators need not be suppressed. Our public code can be used to UV-complete other derivative operators.
Read the full abstract
Quantum Physics
What are local complementations, really?
A local complementation at a vertex $v$ of a graph $G$ replaces the induced neighborhood of $v$ with its complement. Local complementations are, at face value, peculiar operations. Yet they famously show up when describing the local Clifford equivalence among graph states, and, because of this, have become a fundamental notion in the study of graph states. Following Bouchet, we show here how local complementations on graphs are more naturally thought of in terms of switchings of (matroid) bases of stabilizer states. Bases and switchings reveal structural properties of stabilizer states that are harder to glean from the graph-theoretic description alone, such as vertex-minors and statistics of local Pauli measurements. Bases also simplify prior proofs about graphs up to local complementations, of which we give several examples. Bases are foundational in the study of (multi)matroids; as such this paper serves as a gentle introduction to the applications of multimatroids to quantum information theory.
Read the full abstract
Quantum Physics
Quantum-enabled measurement of gravitational forces at the milligram scale
Gravity is by far the weakest among all known interactions. While it dominates the large-scale structure of the universe, gravitational interactions between small masses in the microscopic range remain underexplored because they are too weak to be experimentally accessible. The interface between gravity, described by general relativity, and quantum mechanics, which describes the microscopic world, is of particular interest. From a theory standpoint, these two major descriptions of nature have not been combined in a satisfactory manner. Technology is gradually becoming mature enough such that probing gravity at the microscopic scale, where quantum mechanics and gravity potentially meet, is becoming accessible. Explorations towards genuine quantum gravity have to be carried out under conditions where the gravitating masses possess non-classical properties, which necessitates ultimate protection against disturbances, such as ultralow temperatures and a small range of motion. Here, we present an experiment at millikelvin temperatures, where the gravity due to a micron-range coherent displacement of an 87 mg source mass imprints on the motion of an oscillating membrane loaded with a 1.3 mg test mass. Vibrations of the membrane are measured in the microwave optomechanical scheme, which is a key setup for investigating non-classical motional states of mechanical oscillators. When considering the combination of total source and test masses, our setup presents two orders of magnitude progress in sensing gravity at the small scale, as well as a 35 dB improvement of gravitational sensitivity at kHz frequencies or above. Our setup offers a plausible approach for studying gravitational interactions between masses prepared in non-classical motional states.
Read the full abstract
High Energy Physics (Theory)
Dressed AdS$_4$ black holes in M-theory from the 3d superconformal index
The superconformal index of 4d $\mathcal N\,{=}\,4$ SYM has been shown to exhibit a rich phase structure, whose holographic interpretation in AdS$_5$ involves (dual-)dressed black holes, gray galaxies, and graviton gas contributions. Here we extend the analysis to 3d CFTs, focusing on ABJM theory and the $\mathcal N=4$ ADHM theory. We study the microcanonical superconformal index with two independent charges at finite $N$ and obtain exact signed state degeneracies up to large charges. The results exhibit the black hole physics expected to emerge at large-$N$, including isolated black holes and black holes dressed by additional matter.
Read the full abstract
Statistical Mechanics
Interacting particles crossing barriers: spatial statistics, choice of route taken and role of interactions
We consider the statistics of pairs of particles moving in a potential energy landscape exhibiting a choice of different pathways from a local minimum start point to the global energy minimum destination. We examine the impact of both attractive and repulsive interactions between the particles, treating these via either a soft Gaussian or a Lennard-Jones potential. We investigate the statistics of which particle arrives at the destination first, arrival-time distributions and where in the system the second particle is located when the first arrives. We find that the typical arrival time of the first particle can be significantly different to that of a single particle, sometimes being much longer and in other cases being shorter, depending on the nature of the particle interactions. In the case of soft Gaussian attractive interactions, which effectively bind the pair of particles together to form a single larger particle, the arrival time can be much longer, since bigger particles cross barriers more slowly. In contrast, the arrival time is shorter when there is a repulsive core interaction between the particles, due to excluded-area and particle-ordering effects. We investigate the probability distribution of particles' positions through the Jensen–Shannon divergence which unveils the dependence of the spatial distribution of the particles within the confining potential on the parameters of the particle interaction potentials.
Read the full abstract
Complex Variables
How the $p$-Bergman metric approaches the Carathéodory metric
We study the rate at which the $p$-Bergman metric approaches the Carathéodory metric on convex domains with measures comparable to $\delta^\alpha dV$, $\alpha>-1$. On smooth convex domains of finite type the logarithmic coefficient is bounded above by the harmonic-measure average of the tangential multitype dimension plus $1+\alpha$, with equality when almost every endpoint of the extremal disc is coercive. For a class of convex domains in $\mathbb{C}^2$ with $C^4$ boundary and rank-one real tangential contact along an axis disc, a winding number distinguishes the coefficient $2+\alpha$ from an upper bound $7/4+\alpha$; quartic contact makes the latter bound exact. The exact value holds along the entire axis of explicit real-algebraic strongly pseudoconvex domains, including perturbations without the weighted circle action, although their pointwise boundary models agree up to unitary equivalence with those of a domain having coefficient $2+\alpha$. A $C^2$ example has negative winding but coefficient $2+\alpha$, with a $-\frac12\log\log p$ correction, so boundary regularity is essential to the gap theorem.
Read the full abstract
Multiagent Systems
s$^3$: Simple Scaling of Swarms via Cooperative Multi-agent RL
Groups can achieve more than individuals in the same amount of time, but only if their members coordinate: they must decide who works on what and what to share. However, this organization in multi-agent systems today usually comes from arbitrary, static harnesses designed a priori without adapting to agent behaviors or task specifications. We ask whether agents can instead learn to organize themselves. We introduce s$^3$, the first multi-agent reinforcement learning method in which agents decide for themselves how long to work alone and when to share findings or read teammates' messages, with no predefined organization or roles. Our s$^3$ agents learn to take different approaches, read each other's work, exhibit emergent roles that complement each other in the swarm, and generalize to bigger teams. We study both coordination, where agents must combine private information, and parallel exploration, where sharing discoveries can make search more efficient. With matched training and inference budgets, s$^3$, swarms evaluated with twice as many agents as in training solve over 25$\times$ more coordination problems than the strongest baseline, and in exploration, communicating agents reduce repeated work by nearly 50% and outperform best-of-N by up to 9.4 percentage points. Rather than being designed in advance, s$^3$ enables organization to emerge from a shared goal. This opens the door to capable systems built from many models that learn to work together.
Read the full abstract
Systems and Control
EMT Modeling and Coordinated Mode-Transition Ride-Through Control of Modern Data Centers
Large data centers are emerging as power-electronic loads whose response to grid voltage disturbances can affect bulk-system stability. This paper presents the composition and control characteristics of a proprietary electromagnetic transient (EMT) model of a modern data center comprising an uninterruptible power supply (UPS) with battery energy storage, induction-motor cooling loads, a computer room air handler (CRAH) driven by a variable-frequency drive (VFD), and an aggregated server load. A voltage-based state machine coordinates three aspects of the facility response: the UPS operating mode (eco, double conversion, or battery), the supply path of the CRAH, and a ramped reconnection of the grid-side UPS converter after the disturbance clears. A 100 MW facility is implemented in PSCAD/EMTDC and benchmarked against the PSSE PERC1 aggregate data center model under a symmetrical three-phase fault and low- and high-voltage ride-through (LVRT/HVRT) events. The two models agree closely in their voltage and active-power trajectories, while the EMT model also resolves the converter and motor-reacceleration transients that the phasor-domain model cannot.
Read the full abstract
Artificial Intelligence
NormGym: Evaluating Normative Competence through Social Interaction
Normative competence, the ability to infer socially enforced expectations from interaction and adjust behavior accordingly, is essential for agents operating in unfamiliar human communities. Yet existing evaluations largely test whether agents can recognize or apply norms they may already know, rather than whether they can acquire previously unknown norms from social interaction. We introduce NormGym, an interactive benchmark for normative competence in which a newcomer must infer a hidden, arbitrary norm while pursuing its own task in a multi-agent community. Success requires distinguishing what is possible from what is permitted, inferring the norm from observed behavior and sanctions, and applying it in new situations. NormGym contains diverse norms and distinct environments, allowing systematic evaluation of current agents, including large language model (LLM) agents. We find that normative competence remains challenging: the strongest evaluated LLM achieves a joint task-and-compliance success rate of 55.3%. Our analysis reveals that models can sometimes revise their behavior following sanctions for their own actions, but struggle to use other agents' behavior and the sanctions they receive to infer the underlying norm. We further find that fine-tuning models may be insufficient to improve normative competence. These results suggest that current models lack a robust, transferable capability for acquiring normative knowledge through social interaction.
Read the full abstract
Astrophysics of Galaxies
Jeans Dynamical Constraints on the IMF in Massive Quiescent Galaxies at $z\sim0.7$ from JWST/NIRSpec
We present dynamical masses for nine massive quiescent galaxies at $z\sim0.7$ from the JWST-IMFERNO program, measured from ultra-deep ($\sim32$hr), spatially resolved NIRSpec/MSA spectroscopy combined with NIRCam imaging. The most massive of these galaxies have previously been found to have a bottom-heavy initial mass function (IMF), based on measurements of dwarf-sensitive absorption features. Here, we examine whether the stellar masses inferred for these IMFs are consistent with the dynamical masses of these galaxies. We derive the dynamical masses by including Jeans Anisotropic Modelling within a forward-modelling framework that accounts for instrumental optics, source morphology, and geometric effects. We compare the resulting dynamical masses to the stellar masses for different IMF assumptions. A Milky Way IMF produces physically plausible dynamical-to-stellar mass ratios but implies central dark matter fractions ($f_{\rm DM,1R_e}=0.54\pm0.06$) that are a factor $\sim2$ higher compared to local ETGs and simulations. When uniformly applying the IMF that was measured from absorption features, the mean dark matter fraction becomes low ($f_{\rm DM,1R_e}=0.02\pm0.18$), and the stellar masses exceed the dynamical mass at $2\sigma$ significance for two out of seven galaxies in the sample. When we adopt the measured IMF only in the central $R_e$ and use a MW IMF in the outskirts, we find realistic dark matter fractions ($f_{\rm DM,1R_e}=0.28\pm0.12$) and consistent dynamical masses. These results support a scenario where the cores of the most massive quiescent galaxies at $z\sim0.7$ had a bottom-heavy IMF, similar to those observed in the centres of local early-type galaxies. Our results are consistent with an inside-out growth scenario, in which size and mass growth of massive galaxies proceeds after quenching through minor mergers, leaving the bottom-heavy cores mostly intact.
Read the full abstract
Probability
Interacting Urn Schemes on Infinite Graphs: A Representation Approach
Interacting urn schemes have emerged as an important framework for studying reinforcement phenomena in interacting stochastic systems (see, e.g., \cite{launay2011interacting,launay2012generalized,WARM2014,aletti2017interacting,Synchronisation-2,kaur2023interacting,kaur2023interacting2}). Most existing results, however, are restricted to finite systems. In this article, we investigate interacting urn schemes involving infinitely many urns, where the interactions are naturally described by a directed graph. The reinforcement or drawing mechanism at each urn depends on the compositions of its in-neighboring urns. To analyze the asymptotic behavior of these systems, we develop a representation of the urn dynamics in terms of branching Markov processes evolving on a class of random forests introduced in this article. This extends the representation approach of \cite{bandyopadhyay2022new}, originally developed for single Pólya urn schemes, to interacting urn schemes on locally finite directed graphs. Using this representation, we establish almost sure convergence of the color proportions under standard conditions on the reinforcement matrix or kernel. In particular, we show that the convergence holds independently of the topology of the underlying graph, and hence our results also apply to undirected graphs, where the interactions are bidirectional. We further investigate measure-valued generalizations and other related reinforced stochastic processes, illustrating the broad applicability of the representation approach. Finally, we discuss connections with self-organized criticality.