High Energy Physics (Theory)
Mock Modularity, Resurgence, and Dual False Theta Functions
Mrunmay Jagadale
The $\hat{Z}$ invariant of a three-
manifold $M_{3}$, a half-index of the $3d$ $\mathcal{N}=2$ theory $T[M_{3}]$, is well understood for negative-definite plumbed three-
manifolds. For orientation-reversed
manifolds, however, the known formulas fail to produce $q$-series, and making sense of the operation $q \rightarrow q^{-1}$ remains a central open problem. Seifert
manifolds, whose $\hat{Z}$ invariants are built from false theta functions, offer a natural testing ground. We study “dual false theta functions”, the objects that should replace false theta functions under orientation reversal. Two characterizations of these duals exist in the literature: a modular one, based on mock modularity with a prescribed shadow, and a resurgent one, based on Borel–Mordell integrals and transseries. We prove that the two characterizations are equivalent. We construct a new family of dual false theta functions for values of positive integer $p$ satisfying a Pell-equation condition. Its members are built from Zwegers' indefinite theta functions, have integer coefficients, and have effective central charge $c_{\text{eff}} \leq 1$. We also show that the mock theta functions of Li and Schwagenscheidt satisfy these characterizations and arise from a natural regularization of the divergent product $\eta^{3}(\tau)\psi_{p, a}(- \tau)$.
Read the full abstract ▾
Machine Learning (Statistics)
Fractional Laplace Neural Operators: Exact Architectures, an Expressivity Frontier at Criticality, and Certified Stability for Memory-Driven Network Dynamics
Mauricio Herrera-Marín
Neural operators learn maps between function spaces, while hereditary network dynamics are described by Volterra resolvents with non-rational Laplace symbols. We introduce a fractional Laplace neural operator (fLNO) that embeds this structure in the learned map. For commuting excitation–
Laplacian pairs, one block graph-spectral layer represents the full linear Volterra solution operator exactly. We establish an expressivity frontier for finite rational realizations: they approximate fractional
memory geometrically on compact frequency windows, but cannot reproduce the non-integer critical asymptotics generated by a branch point, and on the half-line the best rational rate is root-exponential. The same theory yields trainable parametrizations that enforce a prescribed stability margin by construction, and a graphon-transfer
theorem separates genuine operator consistency from parameter sharing. In a common-data
benchmark, positive rational operators can match or exceed fLNO accuracy on finite horizons, whereas in controlled near-critical experiments fLNO recovers the branching coordinate more faithfully with far fewer parameters; unconstrained rational fits can cross the stability boundary, while certified parametrizations cannot. A four-parameter spectral law transfers without retraining from graphs of size 48 to 192 with 0.51–0.62% relative error. Applications to Chilean aftershock sequences and to renewal models for Chile and 21 Italian regions illustrate structured inference with explicit uncertainty. The contribution is an operator-learning architecture in which exact
memory structure, physical coordinates and stability guarantees coexist with competitive accuracy.
Read the full abstract ▾
High Energy Physics (Phenomenology)
Unified description of longitudinal and transverse parton dynamics in cold nuclear matter
Weiyao Ke
We present a unified description of longitudinal momentum
evolution and transverse-momentum broadening in cold
nuclear matter using soft collinear effective theory with Glauber
gluon interactions (SCET$_{\rm G}$). In a finite
nucleus, Landau–Pomeranchuk–Migdal interference couples medium-induced energy loss to the transverse scale resolved by the observable. Our framework combines RG
evolution of impact-parameter-dependent energy loss with BFKL-type rapidity
evolution of the collision kernel, improved by kinematical constraints and running coupling. With a common set of cold-
nuclear-matter parameters, it accounts for the main
nuclear modification patterns in semi-inclusive DIS and Drell–Yan production and provides predictions for TMD
hadron multiplicities at the EIC. We also discuss the one-point energy correlator in DIS, whose energy weighting suppresses the leading final-state energy-loss contribution. In the EIC kinematics studied, the residual dynamical corrections are small, leaving the
nuclear modification dominated by the initial-state
nuclear parton distributions.
Read the full abstract ▾
Astrophysics of Galaxies
Star Formation Mitosis: The Effects of Particle Splitting on the Star Formation Outcome
Shivan Khullar, Michael Y. Grudić, Christopher D. Matzner et al. (+3 more)
Particle splitting can increase the dynamic range in Lagrangian astrophysical simulations by several orders of magnitude. However, such methods can introduce numerical noise that may affect the rate of convergence, particularly for collapse problems where fragmentation could be artificially suppressed or enhanced. Thus it is necessary to determine how these errors propagate to key quantities of interest, and to correct or mitigate them where possible. Using GIZMO's MFM-MHD method, we perform a set of $\approx 220$ simulations of a uniform-density turbulent gas sphere undergoing collapse, systematically varying the cloud mass and radius, feedback physics, initial resolution, the pace and cadence at which we split particles and the sink (accreting) or single-age
stellar population (SSP; non-accreting) particle formation prescription. Crucially, we compare with control runs that start with uniformly-high resolution without splitting. While we find no evidence for artificial fragmentation caused by splitting particles, particle splitting runs have a smoother density field than their non-splitting counterparts. Due to this, particle splitting runs exhibit minor differences in the density and
turbulence statistics. However, they form about the same total
stellar mass as the control non-splitting runs, albeit with a slightly different mass spectrum. As a result, particle splitting runs have a different evolutionary history, compared to runs without particle splitting, especially when
stellar feedback is present. Despite the changes seen in runs where we split particles, the numerical solution lies within the scatter inherently present in these chaotic systems, which we model by modifying the initial turbulent velocity field. Our results here argue that particle splitting generally “works" for
star formation applications, despite some systematic differences in the specifics.
Read the full abstract ▾
All authors Shivan Khullar, Michael Y. Grudić, Christopher D. Matzner, Philip F. Hopkins, Norman Murray, Claude-André Faucher-Giguère
Probability
Generalized Feynman Graph Expansions for Lévy-Driven Pseudo-Differential Evolution Equations
Iskander O. Hamdiken and Boubaker Smii
We study a class of Cauchy problems driven by pseudo-differential operators with Lévy-type symbols. We extend earlier results from exponential initial data to a broader set of initial conditions. We demonstrate the existence and uniqueness of the solution, which has an explicit convolution representation with the associated kernel. We then derive an asymptotic expansion of the solution, with coefficients expressed in terms of moments and, after taking logarithms, in terms of cumulants. These expansions are organized through generalized Feynman graph representations, providing a clear combinatorial description of both the solution and its logarithm. We also establish a Borel summability result under suitable conditions and recover the classical exponential case as a special instance. Finally, we illustrate the convolution kernel and solution through a one-dimensional Lévy-Gaussian example.
Read the full abstract ▾
Artificial Intelligence
Before Agents Decide: Epistemic Action in LLM-Based Systems
Yizhi Liu, Balaji Padmanabhan and Siva Viswanathan
Before a difficult decision, people often act simply to understand the situation better. We turn an object to see another side, place alternatives next to each other, or change one condition and observe what happens. These actions may not complete the task, but they improve the evidence needed for the next choice. LLM-based
agents can search and explore, yet
agent design gives less
attention to an earlier question: is the available evidence ready for the decision? Sometimes necessary evidence is missing. In other cases, the evidence is present but its form hides what matters, or the comparison needed to judge it does not yet exist. Cognitive science calls actions that improve the basis for a later choice epistemic actions. We bring this idea to LLM-based
agents and distinguish three modes: acquiring missing evidence, transforming available evidence, and probing a system to create a revealing response. We use the term epistemic scaffolding for the interfaces, tools, and environments that make these actions possible and auditable. This paper argues that
agent design must address how decision-ready evidence is produced.
Read the full abstract ▾
Signal Processing
Active Inference for Slow and Fast Interaction Loops in Holographic-Type Communication
Kamran Fotovat, Kian Fotovat, Nika Hosseini et al. (+1 more)
Holographic-type communication requires a new hologram when the viewer moves beyond the display's angular viewing zone, whereas image-to-3D reconstruction takes tens of seconds per asset. We separate these operations into a slow loop that reconstructs a persistent mesh and a fast loop that renders each viewpoint from the stored mesh. The fast loop takes 0.117 s independently of camera motion, compared with 13.8-14.8 s for a full reconstruction. An active-inference controller determines when to begin reconstruction from learned user-state transitions, whether to perform a full or regional update, and how much computation to allocate based on a
monocular-depth detail measure. Regional updates preserve geometry outside the edited area. In a modeled 100-event session, separating the loops reduces mean latency per event from 14.8 to 4.4 s (3.4-fold); predictive reconstruction reduces it further to 1.7-2.9 s, depending on user routine strength.
Read the full abstract ▾
All authors Kamran Fotovat, Kian Fotovat, Nika Hosseini, Martin Maier
Instrumentation and Methods for Astrophysics
Prototyping the Next-Generation Spectrograph for Space Exploration: Digital Micromirror Device-based Multi-Object Spectrograph (DMD-MOS) Calibration and Performance Assessment
Jiayi Emma Xu, Sarik Jeram, Shaojie Chen et al. (+7 more)
Abandoning the traditional approaches to multi-object spectrographs, the Digital Micro-mirror Device-based Multi-Object Spectrograph (DMD-MOS) leverages a robust, elegant method for creating multi-object spectroscopy slit masks through use of a DMD, which is an array of microscopic flat mirrors where each micro-mirror can be addressed to physically flip between two discrete angles. We have designed two optical channels for DMD-MOS - one for each of these two positions: 1) an imaging channel that can be used for multi-object target selection and 2) a spectrograph channel. Using this two-channel approach, DMD-MOS can simultaneously image the
telescope field of view (FOV) with the imaging channel and actuate the specific DMD micro-mirrors to send light from chosen targets into the spectrograph channel. In this work, we present the results of our laboratory testing of DMD-MOS including spectral resolution, line spread function, wavelength coverage, and the mapping between the DMD and imaging channel planes, the latter of which enables DMD-MOS and our control software to instantaneously configure the DMD to observe chosen target(s) in the acquisition field.
Read the full abstract ▾
All authors Jiayi Emma Xu, Sarik Jeram, Shaojie Chen, Ting S. Li, Ziang Chen, Zain Azam, Adi Khandelwal, Mark Barnet, Yixuan Cheng, Suresh Sivanandam
High Energy Astrophysical Phenomena
X-ray Through Radio Observations and Modeling of the Soft X-ray Flash GRB 250419A
Genevieve Schroeder, Anna Y. Q. Ho, Gavin P. Lamb et al. (+24 more)
Soft X-ray flashes (XRFs) are predicted from a variety of hypothesized phenomena, including "dirty fireballs" (relativistic jets with significant
baryon loading), and gamma-ray bursts viewed marginally off-axis. For a given burst, detailed multiwavelength observations are necessary for distinguishing between possible progenitor scenarios, but the number of XRFs with such data remains small. Here we present X-ray, optical, and
radio (mm to cm) data of GRB 250419A, an XRF (with peak energy $E_p<4\,$keV) discovered by the Space-based multi-band astronomical Variable Objects Monitor. We observe a clear achromatic break in the X-ray and optical bands, and likely two spectral components in the
radio emission. The afterglow is similar to the long-duration gamma-ray burst population, but slightly underluminous, and the optical afterglow has an unusually shallow decay phase at $\lesssim2\,$d. We find that the event is well explained by a high-Lorentz factor jet with low kinetic energy ($E\sim10^{49}$-$10^{50}\,$erg), with energy injection that could arise from radial stratification and/or the jet being viewed marginally off-axis. We conclude that at least some X-ray flashes are an extension of classical gamma-ray bursts down to lower energies.
Read the full abstract ▾
All authors Genevieve Schroeder, Anna Y. Q. Ho, Gavin P. Lamb, Tanmoy Laskar, Daniel A. Perley, Ore Gottlieb, Tomas Ahumada, Igor Andreoni, Aleksandra Bochenek, Michael Bremer, Michael Camilo, Jonathan Carney, S. Bradley Cenko, Alessandra Corsi, Michael W. Coughlin, James Freeburn, Erica Hammerstein, Rahul Jayaraman, Mansi M. Kasliwal, Noel Klingler, Brendan O'Connor, Cassie Sevilla, Gokul P. Srinivasaragavan, Robert Stein, Jada L. Vail, Sylvain Veilleux, Lin Yan
Instrumentation and Methods for Astrophysics
Automated multi-stressor optical fiber focal ratio degradation and throughput characterization system for future massive multiplex spectroscopic survey
Daksh Singh, Ting S. Li, Sarik Jerma et al. (+1 more)
Future spectroscopic survey facilities such as Spec-S5 and the Wide-Field Spectroscopic
Telescope (WST) will contain more than 25,000 optical fibers. Characterizing the focal ratio degradation (FRD) and throughput of these fibers is essential for spectroscopic calibration, sky subtraction, and overall instrument performance, but manual testing of such large fiber populations is prohibitively time-consuming. In this work, we present an automated multi-stressor optical fiber characterization platform capable of measuring both FRD and relative throughput while systematically varying optical and mechanical stress conditions. The system enables characterization as a function of input angle, wavelength, bend radius, twist angle, and externally applied load, while also allowing combinations of stressors to be investigated simultaneously. A custom software framework automates hardware control, data acquisition, and data analysis, enabling large characterization campaigns to be performed with minimal user intervention. Validation measurements performed on a multimode optical fiber demonstrate the ability of the platform to resolve both pronounced and subtle stress-dependent variations in FRD and relative throughput. The results show the expected dependence of FRD on input angle and bend radius, reveal comparatively weak sensitivity to externally applied compressive loading under the investigated conditions, and identify a reproducible periodic dependence on fiber twist angle. In addition, relative throughput measurements indicate that the reduction in relative throughput with increasing input angle cannot be explained solely by a simple projected-area model. The automated workflow and scalability of the system provide a versatile framework for the large-scale characterization of optical fibers required by next-generation spectroscopic survey facilities.
Read the full abstract ▾
All authors Daksh Singh, Ting S. Li, Sarik Jerma, Shaojie Chen
Quantum Physics
Noisy Quantum Query Complexity via Fractional Block Sensitivity
Mehil Agarwal, Shravas Rao and Fang Song
We study
quantum query complexity under several models of imperfect oracle access, and develop lower bounds through a common framework based on fractional block sensitivity ($\operatorname{fbs}$). For a negligent oracle that applies the correct query with probability $1-p$, we prove a lower bound in terms of $\operatorname{fbs}(f,0^n)$. We also show, perhaps surprisingly, that negligence need not destroy
quantum speedups: any function $f$ can be transformed into a partial function $f'$ whose negligent query complexity essentially preserves the
quantum query complexity of $f$. This gives partial functions with exponential
quantum speedups even under negligent queries, and in particular rules out a general lower bound in terms of $\operatorname{fbs}(f)$ for partial functions in this model. For two other models, we obtain general lower bounds in terms of $\operatorname{fbs}(f)$ for all Boolean functions. For hybrid
algorithms using $Q$ coherent and $C$ classical queries, we prove the tradeoff $C+Q^2=\Omega(\operatorname{fbs}(f))$. For an IID dephasing noisy model where each query dephases the query-index register at rate $p$ independently, we prove $\Omega\!\left(p\,\operatorname{fbs}(f)\right)$ queries are necessary. Finally, we introduce a broader family of time-varying dephasing models and identify a variational resource that is always lower bounded by $\operatorname{fbs}(f)$. Computing this resource reduces to a convex optimization problem, providing a simple way to derive lower bounds for new noise schedules. As applications, we recover the hybrid and IID dephasing bounds and determine the query complexity of unstructured search when the dephasing rate grows over time.
Read the full abstract ▾
Quantum Physics
Walshness: an intrinsic neural-network representability metric for quantum states
Nisarga Paul, Kehuang Chen, Jessica K. Jiang et al. (+2 more)
Neural
quantum states (NQS) have emerged as powerful representations of
quantum states with rapidly expanding applications across
quantum many-body physics. Yet our understanding of when neural networks can efficiently represent physical
quantum states remains limited, in part due to the nonlinear parameterization of NQS, the intricate sign structure of
quantum states, and the sensitive dependence on basis choice. We introduce Walshness, a complexity metric of a
quantum state which quantifies whether it admits a compact neural-network representation. We rigorously prove that a general
quantum many-body state admits an efficient NQS representation if it has bounded Walshness, and vice versa, for various NQS architectures. We provide an
algorithm for finding the optimal basis by minimizing the energy of a corresponding classical
spin model. This allows us to empirically improve the learnability of various physical
quantum states, including ground states of the transverse-field Ising model and mixed-field toric code, often reducing infidelity by orders of magnitude, and generalizes the basis choice given by the well-known Marshall sign rule in frustrated antiferromagnets. Our results provide an intrinsic complexity metric for
quantum states with provable connections to their neural-network representability.
Read the full abstract ▾
All authors Nisarga Paul, Kehuang Chen, Jessica K. Jiang, Haimeng Zhao, Di Luo
Quantum Physics
Graph Neural Post-selection for Quantum Error Correction
Conor Carty, Tamas Noszko, Joschka Roffe et al. (+1 more)
Post-selection improves the logical reliability of
quantum computation by rejecting shots which are likely to result in logical failure. We introduce graph neural networks that predict decoder failure without additional decoder executions, using only syndromes, and for high-rate
quantum low-density parity-check (qLDPC) codes, existing belief-propagation posteriors. We evaluate rotated surface and bivariate bicycle (BB) code
memories under uniform depolarising
circuit-level noise. Retaining 90% of shots gives approximately 3700x and 740x reductions in logical error rates for [[72,12,6]] and [[144,12,12]] BB codes at realistic physical error rates. For the distance 5 surface code at physical noise p=0.003, our neural post-selector achieves logical error suppression comparable to the leading complementary gap method without running the decoder.
Read the full abstract ▾
All authors Conor Carty, Tamas Noszko, Joschka Roffe, Roberto Bondesan
General Relativity and Quantum Cosmology
Black hole sensitivity in Einstein–aether theory
Morgan Kidnew, Andrew Kovachik and Sergey Sibiryakov
We numerically find the solutions for slowly moving Schwarzschild black holes in Einstein–aether theory. We use these solutions to compute the sensitivity parameter which describes the coupling of the black hole to the aether field within point-particle effective theory. A simple analytic approximation for the dependence of the sensitivity on the Einstein–aether parameters is derived.
Read the full abstract ▾
Quantum Physics
Quantum Algorithms for OPI Variants Beyond Locality and Classical Decodability
Seyoon Ragavan and Noah Shutty
Regev's reduction quantumly finds codewords satisfying nonlinear constraints by decoding the dual code. To date, applications that have not been dequantized have relied on efficient classical decoders and coordinate-wise constraints. We overcome these restrictions separately. Our first contribution uses a
quantum decoder to find solutions $\mathbf{y}\in(\mathbb{F}_q\setminus\{0\})^m$ to $\mathbf{B}\mathbf{y}=0$, where $\mathbf{B}\in\mathbb{F}_q^{n\times m}$. For fixed
prime $q>2$, Chen, Liu, and Zhandry solve this problem for random
matrices with $m=\Omega(n^2)$, a regime now covered by classical
algorithms. We adapt their template to codes (spanned by the rows of $\mathbf{B}$) that satisfy a "two-fold multiplication property": the coordinate-wise products of pairs of codewords span a space of dimension smaller than $m$. Under suitable distance conditions, this allows us to solve instances with $m\leq n^{2-\Omega(1)}$, beyond the established guarantees of classical
algorithms. We also give an efficient classical
algorithm under a stronger three-fold multiplication property, leaving intermediate regimes as candidates for
quantum advantage. For Reed-Muller codes punctured at random points, our
algorithm finds a word in the unpunctured dual code supported exactly on those points. This works beyond known efficient classical decoding regimes. Our second contribution retains classical decoding but allows global constraints on symbol frequencies. We study "histogram-local" constraints, which specify the allowed numbers of occurrences of each symbol. For broad families, stability under resampling one coordinate yields efficient
quantum algorithms for variants of optimal
polynomial intersection (OPI) combining coordinate-wise and histogram-local constraints. Adapting Yamakawa-Zhandry, we prove a
quantum-classical separation for these problems relative to a classical random oracle.
Read the full abstract ▾
Statistics Theory
Weighted Power Fréchet Means in Metric Spaces with Curvature Bounded Above
Christof Schötz
We establish non-asymptotic risk bounds for power Fréchet means in geodesic metric spaces with
curvature bounded above. The observations form weighted, possibly infinite sequences of independent random variables whose laws and means may differ. We treat three settings. For $2$-Fréchet means in Hadamard spaces, we obtain a sharp mean squared error bound that becomes an identity in
Hilbert spaces. For $2$-Fréchet means in CAT($\kappa$) spaces with $\kappa>0$, we establish variance and Wasserstein contraction inequalities with optimal constants depending on the circumradius $\varrho$ of the closed convex data domain, and obtain mean squared error bounds throughout the maximal range $\varrho<\pi/(2\sqrt{\kappa})$. For $\alpha$-Fréchet means in Hadamard spaces, $1<\alpha<2$, we derive finite $L^\alpha$ risk bounds under weighted $\alpha$-moment conditions, allowing even an infinite $\alpha$-moment of the population mixture. The proofs combine variance, quadruple, and contraction inequalities with a leave-one-out stability technique. Applications give prior and posterior bounds for Dirichlet-process Fréchet means and finite-sample guarantees for local constant Fréchet regression. The regression results require no response-space
entropy condition and replace density smoothness assumptions common in earlier work with transport smoothness; for $\alpha<2$, the bounds remain finite even when the responses have infinite variance.
Read the full abstract ▾
Combinatorics
The Algebraic Connectivity and Laplacian Spectral Radius of Token Graphs
Xiaodi Song, Cristina Dalfó, Miquel Àngel Fiol et al. (+1 more)
For a graph $G=(V,E)$ of order $n$ and an integer $k$ between $1$ and $\lfloor\frac{n}{2}\rfloor$, its
token graph $F_k(G)$ is the graph whose vertices consist of the $\binom{n}{k}$ $k$-subsets of $V$, and two vertices of $F_k(G)$ are adjacent whenever their symmetric difference is an edge in $E$. It was found that the algebraic connectivity of a graph is greater than or equal to that of its
token graph, while the
Laplacian spectral radius of a graph is less than or equal to that of its
token graph. Moreover, a
conjecture that the algebraic connectivity of a graph coincides with that of its
token graph has been proved by using the theory of continuous Markov chains of random walks. In this paper, we derive some results about the algebraic connectivities of a graph and the same graph after adding new edges and their
token graphs to obtain a combinatorial/algebraic proof. Besides, we provide some conditions under which the
Laplacian spectral radius of a graph is less than that of its
token graph. Finally, we characterize the graphs that have the same
Laplacian spectral radius as their
token graphs, including trees.
Read the full abstract ▾
All authors Xiaodi Song, Cristina Dalfó, Miquel Àngel Fiol, Shenggui Zhang
Machine Learning
Denoising Surface: Modeling and Predicting Inference Cost for Diffusion LLM Serving
Haoyu Zheng, Fangcheng Fu, Binhang Yuan et al. (+6 more)
As
diffusion large
language models (dLLMs) become more capable, they are moving from research settings to real-world serving, where request management (such as scheduling and resource allocation) relies on accurate estimation of per-request inference cost. However, common cost proxies fall short for dLLMs: output length ignores that one forward pass can unmask multiple
tokens, and denoising-step count ignores the heterogeneous per-step costs. We observe that the block-autoregressive generation mechanism induces a two-dimensional execution structure over output blocks and within-block denoising steps, whereas these proxies collapse it into a scalar, discarding information essential for characterizing the cost. Motivated by this insight, we propose the Denoising Workload Surface (DWS), which preserves this two-dimensional block-step structure as a probability surface to weight the heterogeneous per-step costs. We then design a coarse-to-fine training scheme that enables a lightweight
prompt-only predictor to accurately predict the complex DWS. This predictor runs efficiently even on a single CPU core, avoiding GPU contention with the serving model. Since DWS decouples request-dependent execution behavior from deployment-specific cost factors, the predictor transfers across hardware configurations without retraining. In real-world serving experiments, DWS reduces cost-prediction error by up to $2.50\times$ over scalar-based predictors, while the DWS-guided shortest-job-first scheduler reduces end-to-end latency by up to $1.92\times$ for online chatbots.
Read the full abstract ▾
All authors Haoyu Zheng, Fangcheng Fu, Binhang Yuan, Yongqiang Zhang, Liang Deng, Hao Wang, Yuanyuan Zhu, Xiao Yan, Jiawei Jiang
Statistical Methodology
Heteroskedastic Canonical Polyadic Tensor Decomposition
Kyle Ritscher and Carlos Llosa-Vite
When minimizing the squared-error loss, the popular CP decomposition can be interpreted as parameter inference in a Gaussian model with a low-rank mean tensor and constant variance across the tensor entries. We introduce heteroskedastic-CP (HCP), which models entrywise variability with a non-constant, low-rank precision tensor, and develop an alternating block-coordinate ascent method to recover both the low-rank mean and precision tensors from noisy observations. Our procedure is computationally competitive, with the same leading-order factor-update complexity as CP-ALS. We demonstrate HCP on synthetic experiments and an
EEG application.
Read the full abstract ▾
Machine Learning
Gumbel Straight Flow: Distilling Autoregressive Models into One-step Flow Maps
Yeongmin Kim, Arnaud Doucet, Andrew Campbell et al. (+3 more)
We present Gumbel Straight Flow (GSF), a continuous flow map
language model that leverages the noise-data coupling of a pretrained autoregressive
language (AR) model. We theoretically demonstrate that the coupling between Gumbel noise and one-hot
token sequences induced by an autoregressive model yields non-intersecting linear paths connecting the noise to the sequence representations. To further enhance high-quality few-step path sampling, we use a flow map
semigroup objective where the tangent (velocity) condition is guided directly by the AR teacher. Across various
benchmarks, including pretraining and downstream tasks, GSF can outperform current few-step
language generation baselines.
Read the full abstract ▾
All authors Yeongmin Kim, Arnaud Doucet, Andrew Campbell, Valentin De Bortoli, Thomas Mensink, David Ruhe
Materials Science
Orbital angular momentum in crystalline solids: non-equilibrium theory
Sangeeta Sharma, Peter Elliott and Samuel Shallcross
We develop a general operator theory of orbital angular momentum (OAM) in periodic solids that applies equally to
equilibrium and coherently driven states. Employing the
crystal momentum representation of Adams and Blount, we construct a periodic-solid OAM operator exactly corresponding to the symmetrized $\hat{\mathbf r}\times\hat{\mathbf v}$ observable, and show that it separates into local circulation (LC) and itinerant circulation (IC) sectors that are Hermitian and $U(1)$, and more generally $U(N)$, gauge covariant. In
equilibrium, the LC and IC sectors reduce to the established modern theory of orbital magnetism; out of
equilibrium, the same operator structure exposes new contributions generated by
laser-written momentum-space textures and interband coherence. Applied to gapped
graphene, we uncover a rich phenomenology of ultrafast non-
equilibrium orbital angular momentum, including a quasi-stationary light-induced OAM. This latter finding, underpinned by a close connection between band-
manifold phase texture and nonequilibrium OAM, points towards new possibilities for light control over OAM in solids as well as a potentially rich role for topological band structures in designed OAM materials.
Read the full abstract ▾
High Energy Astrophysical Phenomena
A Multi-Wavelength Picture of the Surprisingly Short 2024/25 Outburst from EXO 0748-676
Amy H. Knight, Lauren Rhodes, Noel Castro Segura et al. (+11 more)
After sixteen years in quiescence, EXO 0748-676 unexpectedly returned to outburst in March 2024. However, in stark contrast to its previous outburst (1985 - 2008), EXO 0748-676 returned to quiescence in January 2025, after ~ 10 months in outburst. Here, we collate data from nine observing facilities, covering wavelengths from
radio to X-ray, to provide the first multi-wavelength study of the 2024/25 outburst. Unlike the 1985 outburst, both the rise and decay are fully captured at X-ray and optical wavelengths. The light curves do not exhibit the hallmarks of a canonical transient low-mass X-ray binary outburst. Instead, the X-ray and optical fluxes remain reasonably stable during the 2024/25 outburst. The wealth of optical data of EXO 0748-676 reveals a skewed optical phase curve and an irradiation-dominated optical light curve which lasts until the onset of hydrogen recombination in the outer disc at ~11000K, triggering a rapid fade at the end of the outburst. X-ray observations by XMM-Newton and NICER revealed that EXO 0748-676 was heavily obscured early on, with spectra exhibiting a dampened soft X-ray contribution. We find that a spectral model including a clumpy, ionised absorber describes the obscuration, although the exact cause of the early-time obscuration remains uncertain. X-ray colour analysis indicates that EXO 0748-676 spent most of the ~10 month outburst in a faint, hard state. EXO 0748-676 increased marginally in brightness towards the end of the outburst, but never entered a soft spectral state. This behaviour is characteristic of a failed transition outburst.
Read the full abstract ▾
All authors Amy H. Knight, Lauren Rhodes, Noel Castro Segura, Douglas J. K. Buisson, Aru Beri, David M. Russell, Kevin Alabarta, Maria C. Baglio, Jakob van den Eijnden, Matthew J. Middleton, Payaswini Saikia, Chris Done, Rob Fender, Fraser Lewis
Quantum Physics
Causal Trade-Offs in Superluminal Extensions of Relativity. A no-go theorem
Amrapali Sen, Matthias Salzger and V. Vilasini
Extensions of relativity that include superluminal reference frames have recently been proposed by Dragan and Ekert as a way to reconcile relativistic
symmetry with
quantum uncertainty. In this work, we develop a causal modelling framework to formally study this proposal and prove a no-go
theorem demonstrating that a theory permitting superluminal reference frames can avoid superluminal signalling only if it severely constrains which spacetime regions are accessible to operational protocols, if the choice of reference frame has observable consequences or if free variables do not exist or their existence depends on the reference frame. Crucially, our result goes beyond the familiar intuition that superluminal transformations may lead to superluminal signalling; our no-go result shows that Dragan and Ekert's proposal of introducing objective uncertainty does not, by itself, eliminate this conflict. In fact, because superluminal signalling might instead be avoided by relaxing other fundamental assumptions in the presence of superluminal reference frames, the proposed objective uncertainty may not be necessary at all. Our results clarify the limitations of uncertainty-based resolutions and the causal trade-offs faced by superluminal extensions of relativity.
Read the full abstract ▾
Machine Learning
Score the Update, Not the Token: Descent-Aligned Routing for Combinatorial LoRA Experts
Priya Nair, Lukas Brenner, Maya Lindqvist et al. (+5 more)
Mixture-of-LoRA-experts methods raise the capacity of low-rank adaptation by routing each
token to a few low-rank experts. Nearly all of them tie one input-side factor to one output-side factor per expert, and nearly all of them route by scoring the
token: the router picks experts without seeing what any of them would write. We argue that the router should score the update. To first order, adding an expert's update to a layer output lowers the loss by the inner product between that update and the negative loss gradient at the output. This usefulness is quadratic in the
token, so a router that is linear in the
token sees only the part of it that runs through the
token mean, and routers that rank experts by the norm of their own activations never see the output factor. If each expert is split into a reader (down-projection) and a writer (up-projection), the usefulness of every reader–writer pair becomes an inner product in the shared rank-$r$ space, and all $N_AN_B$ pairs can be scored from $N_A+N_B$ vectors. We build VANE on this identity. A low-rank compass predicts the descent direction of each
token. VANE scores every pair by the
alignment between its update and the compass without forming any update, activates the top-$k$ pairs with additive gates, and gives every pair its exact first-order router gradient. On single-domain commonsense
reasoning and a four-domain multi-task mixture with Llama-3.2-3B and Llama-3.1-8B, VANE attains the best average among twelve PEFT and MoE-LoRA baselines, by 0.9–1.1 and 1.3–1.5 points respectively, with less than half the trainable parameters of an 8-expert MoE-LoRA. Its router scores also track the measured usefulness of experts far more closely than
token routers do.
Read the full abstract ▾
All authors Priya Nair, Lukas Brenner, Maya Lindqvist, Daniel Whitmore, Wen-Hsuan Liu, Tom Saliencro, Amara Okonkwo, Rohan Desai
Computation and Language
EurekaBench: Measuring Agentic Ability to Discover New Scientific Insights
Jiayi Geng, Zhengxuan Wu, Kevin S. Chen et al. (+12 more)
When Isaac Newton discovered the law of gravitation, he did so through an iterative process of analyzing observed data such as planetary patterns, finding the underlying mechanisms by describing patterns in mathematical equations, and refining his theory against the Moon's orbit, revealing the startling insight that the same force governs both falling apples and orbiting
planets. Would it be possible for AI
agents to make similar discoveries? To measure this ability, we introduce EurekaBench, a cross-domain
benchmark that tests AI
agents' ability to conduct long-horizon experiments and discover mechanisms that explain observations. We evaluate these mechanisms by the scientific insights that can be derived from them. EurekaBench contains an expert-verified set of 26 long-horizon tasks across neuroscience, computer science, chemistry, astrophysics, geophysics, and
plasma physics, with a total of 306 scientific insights that the discovered mechanisms are expected to support. Our evaluation framework tests three axes of scientific discovery:
agents' ability to follow known scientific constraints, the predictive accuracy of the discovered mechanisms, and whether these mechanisms yield scientific insights or inform future research. Our results show that current AI
agents often overly fixate on predictive accuracy optimization, surpassing human scientists, while falling substantially short in deriving scientific insights.
Read the full abstract ▾
All authors Jiayi Geng, Zhengxuan Wu, Kevin S. Chen, Seungone Kim, Joseph Janssen, Zora Zhiruo Wang, Bhupalee Kalita, Runtian Gao, Aaron Ho, Andrew Oakleigh Nelson, Olexandr Isayev, Francisco Villaescusa-Navarro, Ching-Yao Lai, Howard Chen, Graham Neubig
Data Structures and Algorithms
Dynamic Connectivity, Minimum Spanning Tree, and 2-Edge Connectivity with Polylogarithmic Worst-Case Update Time
Simon Meierhans, Maximilian Probst Gutenberg and Yu-Cheng Yeh
We give fully dynamic
algorithms for maintaining connectivity, minimum spanning tree, and $2$-edge connectivity of a graph with worst-case polylogarithmic update time. Our
algorithms are randomized and succeed with high probability against an adaptive adversary. For the minimum spanning tree and $2$-edge connectivity problems, this improves over the subpolynomial update time bounds obtained by Nanongkai, Saranurak, and Wulff-Nilsen [FOCS'17], Jin and Sun [FOCS'21], and Jin, Sun, and Thorup [SODA'24], respectively. The only randomized component of our
algorithms is the computation of static expander decompositions, and a deterministic
algorithm for said problem would directly imply deterministic
algorithms for all three problems. This reduction is novel even for the connectivity problem.
Read the full abstract ▾
Algebraic Geometry
Supermoduli spaces are non-projected in every genus $g\geq 3$
Mauricio Corrêa, Ron Donagi and Simone Noja
We prove that, for every genus $g\geq 3$, the primary obstruction of each parity component of the
moduli superstack of smooth unpunctured super Riemann surfaces is nonzero, both algebraically and holomorphically. Consequently, both components are non-projected and, in particular, non-split: neither admits an algebraic or holomorphic retraction onto its reduced
moduli stack of
spin curves.
Read the full abstract ▾
Superconductivity
Electronic structure and two-orbital model of the quadlayer La$_5$Ni$_4$O$_{13}$
Jian-Xiang Sun, Haokan Xiao, Cui-Qun Chen et al. (+1 more)
The discovery of pressure-induced superconductivity in Ruddlesden–Popper (RP) nickelates has stimulated extensive interest in high-T$_c$
superconductors. Here, we systematically study the electronic properties of the quadlayer RP nickelate La$_5$Ni$_4$O$_{13}$ under ambient pressure, 5% isotropic compressive strain, and 4% $c$-axis uniaxial strain using density functional theory (DFT) and random phase approximation (RPA) calculations. DFT calculations show that isotropic strain broadens the Ni-$e_g$ bands and induces charge transfer from O-$p$ to Ni-$d$ orbitals, whereas $c$-axis uniaxial strain selectively shifts the $d_{z^2}$-derived bonding1 band upward while leaving the $d_{x^2-y^2}$ dispersion nearly unchanged. From Wannier downfolding, we construct a quadlayer two-orbital model that reproduces the low-energy Ni-$e_g$ bands. Our model reveals that under ambient pressure and 5% isotropic strain, the
Fermi surface consists of two electron pockets ($\alpha$ and $\delta$) and three hole pockets ($\beta$, $\beta^{\prime}$, and $\beta^{\prime \prime}$), while under uniaxial strain, a $\gamma$ hole pocket with $d_{z^2}$ orbital character emerges. RPA calculations reveal that the leading
spin response shifts from $\mathbf{q}\approx(2\pi/3,2\pi/3)$ at ambient pressure to $\mathbf{q}\approx(\pi,\pi)$ under both strain conditions and is enhanced under $c$-axis compression. These results suggest that $c$-axis compression may provide a favorable route to superconductivity in the quadlayer nickelate analogous to that in
bilayer and trilayer nickelates.
Read the full abstract ▾
All authors Jian-Xiang Sun, Haokan Xiao, Cui-Qun Chen, Dao-Xin Yao
High Energy Physics (Phenomenology)
A massive right-handed sterile neutrino in three-body ${\bar B}\to D^*\ell^-{\bar N}_R$ decays - a Monte Carlo simulation study
Bhubanjyoti Bhattacharya, Thomas E. Browder, Lucien M. Cremaldi et al. (+6 more)
We investigate the differential decay distributions of $\bar{B}\to D^*\ell^-{\bar N}_R $, where $N_R$ is a heavy right-handed
neutrino (RHN) and $\ell = e, \mu$. We employ a newly developed Monte Carlo event generator based on the EvtGen framework that can simulate beyond-the-Standard-Model processes. We validate the implementation of the model by comparing the Monte-Carlo results with analytical results for several kinematic variables such as $q^2$ and missing mass, as well as signatures in the angular distributions such as $A_{FB}$, $S_{3}$, $S_{5}$, etc. This MC Generator will enable new experimental searches for RHNs. We discuss the sensitivity of various observables to the presence of an RHN and the feasibility of an experimental analysis at Belle II.
Read the full abstract ▾
All authors Bhubanjyoti Bhattacharya, Thomas E. Browder, Lucien M. Cremaldi, Horacio Crotte Ledesma, Nilakshi Das, Alakabha Datta, Tejhas Kapoor, Kumar Pandey, Alexei Sibidanov
Robotics
ScaffoldM3C: A Multimodal Sequential Monte Carlo Framework for Generative Stable Construction Planning
Gadiel Sznaier Camps, Chengyang He, Guillaume Sartoretti et al. (+2 more)
Autonomously constructing physically realizable 3D structures remains a significant challenge due to combinatorial action spaces, interchangeable components, equifinal assembly sequences, and strict stability requirements during construction. State-of-the-art methods fine-tune large
language models for text-based generative construction. However, these approaches do not allow for
Multimodal (text, image, sketch) conditioning, overlook the practical role of scaffolding for stabilizing intermediate structures, and suffer from slow inference speeds. Therefore, we formulate construction as a probabilistic next-block generation task with multiple potential assembly actions and multiple potential task conditioning modalities. Concurrently, we explicitly consider the utility of scaffolding by introducing an auxiliary scaffold block
token. We present Scaffold
Multimodal Monte Carlo (ScaffoldM3C), a
multimodal, lightweight, auto-regressive model for stable block-based construction, that proposes a set of next-step candidate blocks. Leveraging these candidates, we utilize Sequential Monte Carlo (SMC) to maintain a population of possible assembly sequences, allowing us to consider multiple, potentially different, assembly directions simultaneously. We train our
multimodal architecture by extending the StableText2Brick
dataset to contain image conditioning
prompts and scaffold-stabilized build sequences. ScaffoldM3C is 4x smaller than competing baselines, yielding a 5x to 20x speedup during inference, while achieving comparable construction quality to state-of-the-art methods and higher overall stability. We demonstrate the effectiveness of our approach through simulations and real-world
robot assembly demonstrations.
Read the full abstract ▾
All authors Gadiel Sznaier Camps, Chengyang He, Guillaume Sartoretti, Eduardo Montijano, Mac Schwager