Peer Revenue Play credits. Real papers.

Papers

Fresh from arXiv, typeset properly. Highlighted words trade on the market.

All of arXiv

Tuesday 8 September

Astrophysics of Galaxies

RIGEL: Ultra-faint dwarf galaxy diversity shaped by inhomogeneous cosmic reionization

Ultra-faint dwarf galaxies (UFDs) are among the smallest and oldest galaxies in the Universe and are widely regarded as relics of cosmic reionization. To investigate how reionization quenches star formation and shapes the diversity of UFDs, we present a suite of eight cosmological zoom-in simulations of isolated UFDs with present-day halo masses of $\sim10^9\,{\rm M}_\odot$. The simulations are performed with the radiation-magnetohydrodynamic galaxy formation framework Realistic ISM modeling in Galaxy Evolution and Lifecycles (RIGEL), coupled to realistic large-scale radiation fields extracted from the THESAN reionization simulation. Despite residing in similar $z=0$ halos, the simulated galaxies span nearly two orders of magnitude in stellar mass and broadly reproduce the observed luminosities, sizes, metallicities, and stellar kinematics of Local Group UFDs. We find that reionization quenches star formation through a two-stage process. The arrival of the ionization front rapidly photoionizes the diffuse circumgalactic and intergalactic gas, suppressing further gas accretion onto the galaxy. Star formation nevertheless continues for several hundred Myr using the surviving self-shielded gas reservoir and ceases only after this gas is consumed or dispersed. Within 500 Myr after reionization, less than 40% of the initial gas mass remains in the halo, with photoevaporation constituting the dominant gas-loss channel. We further show that the halo mass at the time of reionization is a key parameter governing the subsequent evolution of UFDs. Galaxies residing in more massive halos at reionization retain gas for longer periods and undergo more extended chemical enrichment. Consequently, the halo mass at reionization strongly correlates with the final stellar mass, stellar age spread, and chemical evolution of the galaxy.
Read the full abstract

All authors Yunwei Deng, Hui Li, Ewald Puchwein, Boyuan Liu, Aaron Smith, Rahul Kannan, Federico Marinacci, Greg L. Bryan, Kung-Yi Su, Mark Vogelsberger

Astrophysics of Galaxies

SHELLQs. Black Hole Mass and Eddington Ratio Distributions of Intermediate-Luminosity Quasars at 6<z<7

We present near-infrared spectroscopy of 21 quasars at 6.07 < z < 6.90 obtained with JWST/NIRSpec and Subaru/MOIRCS. These targets have absolute ultraviolet magnitudes of -25 < M_1450 < -22 and are drawn from a sample of z>6 quasars identified in the Subaru High-z Exploration of Low-Luminosity Quasars (SHELLQs) project. These quasars occupy the intermediate-luminosity regime between luminous quasars and the faint high-redshift AGNs uncovered by JWST. We detect broad Balmer emission lines and MgII 2798 with underlying continua from the NIRSpec and MOIRCS targets, respectively. The virial black hole masses of this sample span a wide range of 7.2 < log M_BH/M_sun < 9.4, with Eddington ratios of -1.3 < log L_bol/L_Edd < 0.4. Combining these measurements with our previous mass estimates for other SHELLQs quasars, we construct a sample of 27 quasars with M_BH estimates at 6 < z < 7 and derive the distributions of M_BH and L_bol/L_Edd. When compared with luminosity-matched quasars at z ~ 1.3, we find median offsets of delta log M_BH/M_sun = -0.4 and Delta log L_bol/L_Edd = 0.2 in the high-redshift sample. In addition, 11% of the high-redshift quasars are accreting near or above the Eddington limit. As a systematic check, adopting the Eddington-ratio-dependent calibration of single-epoch M_BH estimates increases the inferred super-Eddington fraction to 22%. These results indicate that a representative population of distant supermassive black holes undergo a rapid accretion phase, although not as extreme as that seen in the most luminous quasar population, consistent with the anti-hierarchical growth scenario of supermassive black holes.
Read the full abstract

All authors Masafusa Onoue, John D. Silverman, Yoshiki Matsuoka, Xuheng Ding, Camryn L. Phillips, Michael A. Strauss, Junya Arita, Takuma Izumi, Mahoshi Sawamura, Nobunari Kashikawa, Irham Andika, Kentaro Aoki, Shunsuke Baba, Anna-Christina Eilers, Seiji Fujimoto, Tomotsugu Goto, Masatoshi Imanishi, Kohei Inayoshi, Kazushi Iwasawa, Knud Jahnke, Yuki Kaneko, Toshihiro Kawaguchi, Kotaro Kohno, Chien-Hsiu Lee, Alessandro Lupi, Tohru Nagao, Dragan Salak, Malte Schramm, Yoshiki Toba, Hideki Umehata, Marta Volonteri, Fabian Walter, Feige Wang, Jinyi Yang

Astrophysics of Galaxies

A Spectral Framework for Testing the Quasi-Star Hypothesis in Little Red Dots I: Weighing LRDs by Their Super-Eddington Luminosity Ratios—No Signs of Overmassive Black Holes

We present a spectral test of the quasi-star hypothesis for Little Red Dots (LRDs) whereby a black hole grows inside a stellar-like envelope. We use Prospector to fit host galaxies and TLUSTY photospheres to 5 LRDs that show strong molecular or atomic absorption. We approximate each object's electron-scattered Eddington luminosity ratio, $\phi \equiv \kappa_{\rm es}\sigma T_{\rm eff}^{4}/(gc)$, and we use MESA-QUEST to simulate their envelopes. All 5 are super-Eddington at $\phi = {87}$–$299$, with the caveat that the largest sources sit at the edge of our atmosphere grid and beyond our simulations. We derive envelope masses between $700$–${15{,}000}\,M_\odot$, where quasi-star theory requires the black hole to be less than a third of that. GN-28074 falls 5 decades below its published virial mass estimate, alleviating the overmassive black hole problem. Their black holes double every $\sim0.05$–${0.2}$~Myr and can produce intermediate-mass black holes in $\lesssim30$~Myr. Our 3 water-absorbing objects have cold components that are $2$–$3$~dex denser than their hot components, which we interpret as the water forming in cold dense clouds. We then extend our measurements to 82 archival LRDs, finding that the population has super-Eddington photospheres and wind speeds that increase with $\phi$, which implies that LRDs evolve from massive sources with slow winds to having more eruptive winds as they mature and shed their outer envelopes. We thus constrain the Eddington ratios and masses of LRDs and show self-consistently that quasi-stars may be the central engines powering LRDs.
Read the full abstract

All authors Olivia Curtis, Nikko J. Cleri, Jakob M. Helton, Joel Leja, Jason T. Wright

Computation and Language

StochBench: A Domain-Specific Benchmark for Stochastic Processes in Lean

Leading benchmarks for formal theorem proving with large language models are small collections drawn from competition math, such as the IMO and Putnam, that poorly represent field-specific applications. We introduce StochBench, a Lean 4 benchmark of 450 graduate stochastic-processes problems at varying abstraction levels, each paired with its natural-language source. Addressing a field underrepresented in Mathlib, it covers finite and countable Markov chains, renewal processes, random walks, martingales, stopping times, queues, Brownian motion, stochastic calculus, weak convergence, and Poisson and continuous-time Markov processes. Our Opus 4.8-based agent achieves a 34.9% proof rate (157/450) under a 15-minute per-problem limit. StochBench better represents domain-specific applied mathematics while remaining challenging for advanced provers.
Read the full abstract

All authors Idan Davidovich, Debargha Ganguly, Vikash Singh, Vipin Chaudhary

Sound

Voice or Stereotype? Disentangling Acoustic and Content-Based Gender in Speech-to-Speech Models

Speech-to-speech (S2S) models now run inside dubbing, translation, and voice agents. Unlike text models, they hear the speaker's voice, which carries the speaker's gender. A faithful system should treat a speaker as who they sound like, not as whoever usually says what they said. Testing this is harder than it looks, since most S2S models answer in a single, fixed output voice, hard-coded so it cannot drift toward a stereotype. Checking the output voice comes back clean even when the model is biased. We therefore ask two questions. When a model re-speaks the input, does the stereotype in the words shift the perceived gender of the output voice (voice rendering)? And when the model states the speaker's gender, does it follow the voice or the content (gender attribution)? We answer both with one controlled experiment crossing male and female voices with masculine-, neutral-, and feminine-stereotyped passages, on five open- and closed-source models in English, Spanish, and Mandarin. The rendered voice shows no stereotype drift. But every model decides the speaker's gender from the content, not the voice. Making the content one step more feminine (masculine -> neutral -> feminine) multiplies the odds of a "female" judgment by 1.7-24. When the content clashes with the voice, the worst model misgenders the speaker in 90% of cases. When they agree, it misgenders in only 2%. The bias thus hides in gender attribution, where fixed-voice evaluation cannot see, and where audits must look as S2S systems increasingly speak for real people.
Read the full abstract

All authors Xiaoqun Liu, Tanu Mitra, Harshit Rajgarhia, Abhishek Mukherji

High Energy Physics (Phenomenology)

Level-Four Modular Symmetry Selected by the Harmonic Pattern of Quark and Lepton Masses

The mass ratios of the quarks and charged leptons follow a simple pattern. The logarithmic generation steps come in a two to one ratio, a note to its octave, and the ordering is mirrored between the up-down and down-lepton comparisons. The pattern, read as charge counting on a $Z_N$ clock, is realized at level four and at no level that is not a multiple of four, which motivates modular level four and the finite group $S_4$, with the generations in the triplet at $T$ charges $(3,1,0)$ and the mirror in the sign singlet. The charged-lepton mass ratios lie on a theta-dressed lattice at $0.02\%$, which sets the base $\varepsilon=0.18664$, taken as $14/75$, and the harmonic modulus $\tau=1.0685i$. Five postulates in the weight one-half theta constants, identified on the same data, reproduce the six quark masses from three lepton masses and the electroweak scale. Two more and one empirical relation for $|V_{td}|$ close the CKM matrix, giving $|V_{us}|=\sqrt{m_d/m_s}=0.2255$, $|V_{cb}|=0.04197$, $|V_{ub}|=0.003721$, and $\delta=1.138$ against the measured $1.139\pm0.023$, with no continuous parameter. The unitarity triangle follows, with apex $(0.161,0.348)$ against the measured $(0.161\pm0.010,0.347\pm0.010)$, and $J=3.11\times10^{-5}$ against $(3.09\pm0.07)\times10^{-5}$. The same structure carries the neutrinos. A single theta insertion gives $\sin\theta_{13}=\theta_2(\tau)m_2/m_3$, which holds at $0.7\sigma$ on the 207-day JUNO data, and the harmonic condition completes a normal-ordered spectrum with $m_1=0.25$~meV, $\Sigma m_\nu=0.0589$~eV, a CP-conserving phase $\delta_{CP}=\pi$, and effective Majorana masses quantized at $1.3$ to $3.9$~meV. The ultraviolet completion belongs to the magnetized torus class, with Pati–Salam the natural gauge embedding.
Read the full abstract

General Relativity and Quantum Cosmology

Reconstructing sign-switching dark energy histories: Scalar-field regularity, conditional potential comparison, and representative dynamics

Phenomenologically similar sign-switching dark-energy histories need not have comparable scalar-field realizations. We reconstruct minimally coupled scalars with fixed kinetic sign for three prescribed histories: the error-function model (ECDM), the compact smooth-step model (SSCDM), and the ladder-like model (L$\Lambda$CDM). Continuous negative-to-positive density crossings select the phantom branch. ECDM yields a smooth on-shell potential at every finite redshift; its equation-of-state pole at the density zero is only a ratio singularity. Exact SSCDM has a regular trajectory but a $C^1$, non-$C^2$ endpoint potential, $V-V_e\propto|\phi-\phi_e|^{4/3}$. Its non-Lipschitz force permits delayed departures from frozen plateaus, so the reconstructed potential and plateau data do not uniquely generate the prescribed history. The exact Ladder requires distributional kinetic stress and has no ordinary classical realization in the adopted one-field action. In conditional synthetic comparisons, the sigmoid–Gaussian family ranks highest for ECDM and the generalized axion-like family for SSCDM, although the fitted axion exponents $n<1/2$ imply a divergent endpoint force. Representative regular forward solutions exhibit negative-to-positive scalar-density crossings, with the potential zero preceding the density zero. In a closure test, unretuned evolution of the top-ranked ECDM template tracks the target approximately, $\max|\Delta\widetilde\Omega_\phi|\simeq0.16$. Closure fails structurally for compact SSCDM: no potential with a locally Lipschitz force can reproduce its exact finite-duration frozen plateaus from exactly frozen initial data. Field-map existence, invertibility, and endpoint regularity must be assessed before interpreting a phenomenological history as scalar dynamics. The phantom action is used only as a homogeneous effective proxy.
Read the full abstract

All authors Shahnawaz A. Adil, Özgür Akarsu, Mariam Bouhmadi-López, Beñat Ibarra-Uriondo, Nihan Kat{\i}rc{\i}, J. Alberto Vázquez

Number Theory

Effective Euler Sums

Euler sums, also called alternating multiple zeta values (MZVs), have been shown to play important roles in many research areas in mathematics and theoretical physics. Motivated by a similar conjecture for MZVs by Kaneko and Zagier, the author proposed a conjectural isomorphism between the space of finite Euler sums and the space of classical ones modulo $\zeta(2)$-products. In this paper, we explicitly construct those finite elements (for which we call effective Euler sums) that correspond to the classical Euler sums under this conjecture in depth one and two. In the appendix, we offer another group of plausible candidates of effective Euler sums when depth is two and weight is even, by extending the heuristic argument of Kaneko and Zagier for the double zeta case. Kina recently obtained independently some the same results in the MZV setting. In particular, using one of his results we prove in the appendix that Kaneko and Zagier's heuristically defined effective double zeta values coincides with ours.
Read the full abstract

General Mathematics

Complex numbers: uncommon knowledge

In this note we unveil some subfields of the field of complex numbers which are well known to specialists, but seem to remain hidden for a wider mathematical audience. The existence and construc- tions of these fields follow directly from famous results of the 20-th century, however standard graduate textbooks avoid these topics.
Read the full abstract

Machine Learning (Statistics)

Tensor Network Moral Graph Recovery of Discrete Probability Distributions

We present a method for recovering the moral graph of a causal DAG from a probability distribution over discrete variables, using fully connected tensor networks (FCTNs) with nuclear-norm-regularized bond corrections. Each bond matrix is parameterized as a baseline all-ones matrix plus a low-rank correction $C_{ij} = U_{ij}V_{ij}^\top$, and the nuclear norm of the correction implemented via the variational Frobenius norm penalty on the factors drives unnecessary bonds to zero. We prove that under faithfulness, positivity, and a no-implicit-rerouting assumption on the local tensor architecture, every optimal FCTN with zero reconstruction error $\varepsilon = 0$ has effective graph exactly equal to the moral graph. For the approximate regime ($\varepsilon > 0$), we provide explicit recovery bounds using the Fannes-Audenaert continuity of conditional mutual information, and derive a sufficient condition on the regularization parameter $\beta$. The effective graph is read directly from the optimized bond matrices.
Read the full abstract

All authors Á. Troyano Olivas, Chi-Hang Fred Fung, Hans H. Brunner, Momtchil Peev, Vicente Martin

Machine Learning

Accountable and uncertainty-aware evaluation of sensor-based AI under distribution shift: devices, subjects, and nearly three years underground

Sensor-based AI systems are rarely operated under the conditions under which they were trained: devices, personnel and recording epochs change, and each change degrades performance in ways a random train-test split cannot reveal. We propose a staged, accountable evaluation protocol that treats the evaluation of a deployed model as a measurement with declared reference levels and a quantified uncertainty. Four cumulative generalisation stages hold out devices, subjects and time. Each stage is judged on quantiles of repeated trainings against chance references with the correct class count, an out-of-present-scope rate exposes silent misdirection towards classes that are no longer present in deployment relative to training, and an explicit decision rule ties roll-out decisions not to means but to 5% quantiles. We demonstrate the protocol on infrastructure-free geomagnetic localisation with smartphone-based recurrent classifiers in two real underground mines, including a replication of the scheme's training stages at the second site. Unchanged models are re-evaluated on data recorded 34 months after the training campaign, on a device generation unknown at training time and with a held-out surveyor. The 5% quantile of their present-conditioned precision there is 0.39 over 299 repeated trainings, 16.5 times the chance level; across the composition of the 42 reachable location classes the figure varies by +/-0.08, several times the spread between repeated runs. Repeated trainings of a single configuration show why means mislead: a bimodal configuration passes a mean-based test decisively while its 5% quantile lies more than an order of magnitude below chance.
Read the full abstract

All authors Benny Platte, Rico Thomanek, Christian Roschke, Marc Ritter

Algebraic Geometry

From Closed to Relative Higher-Genus Gromov-Witten Invariants via Resurgent Functions

Higher genus Gromov-Witten invariants of Calabi-Yau threefolds are encoded in a generating function which is an asymptotic series in a formal parameter $\lambda$. Using resurgence, analytic functions in this formal parameter were uncovered. In this paper we focus on the resolved conifold and study the enumerative meaning of the strong-coupling asymptotic expansion, in powers of $ 1/\lambda $, of the resurgent analytic functions. We show that this expansion contains both a closed curve-counting contribution in dual variables and a contribution governed by relative Gromov-Witten invariants, naturally interpreted in logarithmic geometry.
Read the full abstract

Machine Learning (Statistics)

CAST: Canonical Approximate Schur Tree for Approximate Cholesky on Graphs

Graph-data workloads such as diffusion estimation, ranking, semi-supervised learning, and network optimization often solve many Laplacian or symmetric diagonally dominant M-matrix (SDDM) systems with the same coefficient matrix. Approximate Cholesky preconditioners eliminate vertices one at a time and store the resulting sparse approximate factorization, the factor, whose construction cost is amortized across these solves. But eliminating a vertex, the pivot, creates a dense Schur-complement clique among its $d$ active neighbors. We introduce CAST (Canonical Approximate Schur Tree), which replaces this clique with a weighted random spanning tree sampled directly from it. Every realization is connected and contains exactly d-1 edges, while reweighting each selected edge by the reciprocal of its tree-inclusion probability makes the update unbiased. The distribution is independent of the ordering of the pivot neighbors, and we prove that its leverage-score marginals minimize the largest normalized reweighted-edge contribution among unbiased inverse-marginal one-tree estimators. We also introduce CAST-$\rho$, which replaces each pivot neighbor with $\rho$ copies, each carrying a $1/\rho$ share of that neighbor's incident weight, samples a weighted random spanning tree on the expanded clique, and contracts the copies back to the original neighborhood. The resulting update remains unbiased and connected, can be sampled exactly in $O(\rho d)$ time, and satisfies a $1/\rho$ bound on the second moment of the normalized local Schur error. Increasing $\rho$ therefore reduces certified local sampling variability, but may increase construction cost and downstream fill. Empirically, we observe that CAST-1 is the faster default, whereas CAST-2 is preferable when its additional edge contributions remain inexpensive.
Read the full abstract

Machine Learning

DiffLUT-Net: Differentiable Training of FPGA LUT Networks with Learnable Connectivity

Field-programmable gate arrays (FPGAs) enable efficient neural-network inference, but most deployment flows either accelerate multiply-accumulate operations or convert pretrained quantized models into lookup tables (LUTs). We present DiffLUT-Net, an FPGA-native network connected by six-input LUTs that are trained from scratch. We jointly learn the 64 truth-table entries of a LUT and the source to each of its six input ports using a differentiable LUT function relaxation and hardware source selection. After training, the truth tables and connections are discretized, unused logic can be pruned, and the network is exported directly as synthesizable Verilog. Across five benchmarks, DiffLUT-Net achieves favorable accuracy-resource trade-offs. These results demonstrate the effectiveness of jointly learning LUT functions and sparse connectivity for compact FPGA-native inference. The code is available at https://github.com/TUDa-HWAI/DiffLUT-Network.
Read the full abstract

All authors Jiaqi Ye, Xinrui Gong, Jingcun Wang, Olga Kondrateva, Bing Li, Grace Li Zhang

Astrophysics of Galaxies

Virgo Filaments VII: MeerKAT HI-imaging of the VirgoIII filament

We present MeerKAT HI observations of galaxies in the nearby Virgo III filament. At a median distance of $\sim30$ Mpc, the filament extends over $\approx16$ Mpc across $\sim56$ deg$^2$. Our survey comprises 15 targeted MeerKAT pointings centred on HI deficient galaxies. Multi-resolution HI imaging at 8-90 arcsec with an rms of $\sim0.3$ mJy beam$^{-1}$ reaches column densities of $\sim4\times10^{20}$ to $\sim4\times10^{18}$ atoms cm$^{-2}$. This probes HI in galaxy discs at $\sim1.2$ kpc resolution and their outskirts at $\sim13.5$ kpc. We detect HI emission from 80 sources within the 500-3000 km/s range; most are optically detected galaxies, while 10 per cent lack confirmed counterparts. We also identify three HI clouds without optical counterparts near an interacting system, possibly displaced during the interaction. Galaxies in the filament lie below the field HI gas-fraction scaling relations at fixed stellar mass. Crucially, $\sim40$ per cent of galaxies not selected for HI deficiency fall more than $1\sigma$ below these relations, with an even greater fraction lying under the mean relation itself, indicating that the low HI gas fractions are widespread in the filament rather than confined to the pre-selected deficient population. Quantifying HI morphologies using an asymmetry parameter, we find a wide range of disturbances without a clear trend with projected distance to the filament spine or local density, suggesting a combination of local interactions and the filament itself. This paper presents the survey description, catalogue, and HI atlas enabling investigations of gas removal mechanisms in Virgo III galaxies.
Read the full abstract

All authors M. Ramatsoku, G. Nagaraj, P. Jablonka, R. A. Finn, B. Vulcani, G. Rudnick, F. Combes, G. De Lucia, G. Castignani, D. Zaritsky, Y. Bahè, R. A. Koopmann, D. Zakharova, K. Conger

Rings and Algebras

Linear Preservers of Infinitely Divisible Matrices

An infinitely divisible nonnegative matrix is an entry-wise nonnegative matrix that admits an entry-wise nonnegative $m$th root, with respect to usual matrix multiplication, for every positive integer $m$; it is called strongly infinitely divisible when it is, in addition, invertible. The study of such matrices has its origins in the theory of infinitely divisible probability distributions and the embedding problem for Markov matrices and is closely connected with continuous one-parameter semigroups; in the invertible case, this connection admits a natural description in terms of the exponential map. In this paper, we characterize the bijective linear maps on $M_n(\mathbb{R})$ that preserve strongly infinitely divisible matrices and infinitely divisible nonnegative matrices. In the former case, we combine the Inverse Function Theorem with Zariski-density techniques, using the Zariski-density approach recently developed in linear preserver theory by Fallat and Mondal [Proc. Amer. Math. Soc., 2026]. In the latter case, we first show that preserving infinite divisibility forces preservation of the cone of entry-wise nonnegative matrices and then exploit the additional structure afforded by infinite divisibility to complete the characterization.
Read the full abstract

Materials Science

The energetics of force errors in machine-learned molecular dynamics

The energetic effect of a force error depends on atomic motion. We establish a directional residual-work coefficient combining directional curvature mismatch with the spatial distribution of the residual response. For conservative potentials force-matched at an anchor, it determines the leading signed work at the first crossing of a small force-error budget. At a 474-atom lithium-electrolyte interface, predictions fixed before future reference evaluations differ from measurements by less than 4.6% of predicted work across 24 prescribed endpoints. Changing only the initial velocity direction at fixed structure and initial total kinetic energy reverses the force-work ranking. At the same admitted time of 0.25 fs, one direction gives an 11.6% larger maximum force residual but 36.1% less work. The reversal recurs at a second structure. The framework connects force tolerances to reference-energy transfer, providing a physical basis for potential assessment and adaptive reference allocation.
Read the full abstract

All authors Peng Kang, Da Wan, Shulin Bai, Vincent Michaud-Rioux, Zhen Li, Yu Liu, Lei Zheng, Li-Dong Zhao

Robotics

No Free Checker: A Survey of Verifiers for Robot Policies

A verifier for robot policies reads a candidate behavior and returns a score for how well it did, used both to evaluate vision-language-action policies and to train them. Verifiers range from success detectors and reward models to runtime monitors, safety filters, and temporal-logic specifications. We survey roughly 150 verifiers and compare them along two properties. Availability is how much a verdict costs, how early in a rollout the verdict arrives, and how often a verdict can be asked for. Availability rises as verdicts get cheaper, earlier, and denser. Credibility is how much a high score tells us about the task. Credibility falls as the judgment becomes gameable and self-serving. We group the verifiers by who supplies the judgment: human verifiers, rule-based and formal verifiers, learned and pretrained verifiers, and model-intrinsic verifiers. Across the four families, we find that credibility falls as availability rises. Regardless of who supplies the judgment, there is no free checker. We then examine what validates a verifier itself, and how much a high score tells us. Three measures appear in the literature: agreement with human labels, the performance of the policy it trains, and behavior under reward hacking. We close with nine metrics that make a verifier claim checkable, and coordinates for the verifiers still to be built.
Read the full abstract

All authors Yang Wan, Xihang Yue, Zhirui Liu, Ziyuan Chu, Shuxun Wang, Yuhan Chen, Xiaonan Jiang, Xukun Zhu, Yubo Dong, Linchao Zhu

Combinatorics

$B_h$-sets and perturbations in normed vector spaces

The subset $A = \{a_i:i \in I\}$ of a normed vector space is a $B_h$-set if every element of the sumset $hA$ has a unique representation as a sum of $h$ elements of $A$. Let $\varepsilon = \{\varepsilon_i:i \in I\}$ be a set of positive real numbers. An $\varepsilon$-perturbation of $A$ is a set $A' = \{a'_i:i\in I\}$ such that $|a_i'-a_i|<\varepsilon_i$ for all $i \in I$. Let $\Delta_{hA} = \inf\{|x'-x| : x,x' \in hA \text{ and } x\neq x'\}$. It is proved that if $A$ is finite or countably infinite set with $\Delta_{hA}>0$, then there is a $B_h$-set $A'$ that is an $\varepsilon$-perturbation of $A$.
Read the full abstract

High Energy Astrophysical Phenomena

Decoding the Imprints of Energy-Momentum Squared Gravity in Neutron Stars with Machine Learning Analysis

Neutron stars (NSs) provide a unique laboratory for testing gravity in the strong-field regime and for searching for deviations from General Relativity (GR). In this work, we investigate the effects of Energy-Momentum Squared Gravity (EMSG) on NS structure and examine whether its signatures can be identified from observable stellar properties using supervised machine learning (ML). We solve the modified Tolman-Oppenheimer-Volkoff equations for approximately $10^{4}$ nuclear equations of state (EOSs) for EMSG coupling parameters $\alpha=\{-5.01,-2.50,0,+2.50,+5.01\}\times10^{-38}\,\mathrm{erg}^{-1}\mathrm{cm}^{3}$, and calculate the gravitational mass $M$, radius $R$, dimensionless tidal deformability $\Lambda$, and fundamental $f$-mode oscillation frequency for each stellar configuration. Imposing observational constraints on $M$, $R$, and $\Lambda$, we split our datasets into train and test sets, and we employ Random Forest (RF), K-Nearest Neighbors (KNN), Support Vector Machine (SVM), Logistic Regression (LR), and Gaussian Naive Bayes (GNB) to classify the representative sectors $\alpha=\{-5.01,0,+5.01\}\times10^{-38}\,\mathrm{erg}^{-1}\mathrm{cm}^{3}$ using $(M, R,\Lambda,f)$. The RF classifier performs best, achieving an accuracy of approximately $99.85\%$ with precision, recall, and F1-scores exceeding $99.8\%$, while KNN also achieves accuracy above $99\%$. The nearly diagonal confusion matrices demonstrate that the observationally viable NS configurations associated with different EMSG sectors remain highly separable in the multidimensional observable space. Our results show that NS observables retain robust signatures of EMSG even after observational filtering, establishing ML-assisted NS observations as a promising complementary approach for probing modified gravity with current and future multi-messenger observations.
Read the full abstract

Atmospheric and Oceanic Physics

Steering Diffusion Priors with Sparse Observations for High-Resolution Temperature Downscaling

Local heatwave hazard depends on fine-scale air temperature, but ground stations are sparse and reanalysis products such as ERA5 cannot resolve the terrain and land-surface contrasts that shape real heat exposure. We present a conditional diffusion emulator for high-resolution 2-m temperature downscaling, conditioned on static geography, a training climatology, exact-time ERA5 temperature, and solar and temporal features, guided at inference by score-based data assimilation (SDA): a differentiable Gaussian observation likelihood steers the diffusion score toward sparse revealed temperature observations without any retraining. On a controlled 32-case synthetic-grid protocol over AORC, guidance improves hidden-cell reconstruction over both ERA5 and a strong observation-proximal nearest-neighbor baseline once observation density reaches 1% (RMSE 0.318 vs.\ 0.431~K, winning all 32 cases), while sparser regimes still favor direct interpolation. We further map the full guidance-strength landscape across three observation densities, showing that the optimal strength shifts systematically with density and that over-guiding causes sharp, predictable degradation – giving a concrete operating recipe rather than a single untuned setting. The resulting fields are intended as a temperature layer for downstream heatwave-hazard products such as threshold exceedance and cumulative heat-burden. The present evidence is a controlled synthetic-grid validation; station-network and held-out-year evaluations are the next steps toward deployment.
Read the full abstract

All authors Anirudh Avireddy, Manmeet Singh, Shivanshi Singh, Ayush Raj, Saptarishi Dhanuka, Parthasarathi Mukhopadhyay, Sandeep Juneja

Geophysics

Bayesian deep learning integration of geophysical and drilling data for 3D prediction of copper mineralization and drill targeting: a case study from the Kogodai prospect, Rudny Altai

Exploration drill targeting in structurally complex terranes is hindered by sparse sampling, heterogeneous datasets, and the ambiguity of geophysical inversions. Here, we present an uncertainty-aware 3D workflow for the acceleration of time-to-discovery in brownfield explorations and apply it to the Kogodai prospect in the Rudny Altai metallogenic province. We jointly analyse existing drilling and geophysical data in a comprehensive approach, revealing hidden patterns in already available data. Drillholes and trenches were desurveyed to a common 3D reference frame, and assays were composited to a consistent spatial support to facilitate joint modelling with geophysical inputs. We develop Bayesian deep-learning models to predict 3D fields of Cu grade together with chargeability and apparent resistivity while quantifying epistemic uncertainty via Monte Carlo sampling. The original contribution of this work is to treat the problem not as pointwise regression between co-located observations, but as joint learning of spatially continuous 3D fields from sparse, heterogeneous exploration evidence. The resulting 3D predictions delineate a principal mineralized trend and several localized candidate zones that coincide with elevated induced polarization (IP) responses, while uncertainty mapping highlights where predictions are robust versus where additional drilling would be most informative. The continuous Cu-grade field can also be thresholded to produce binary prospectivity maps, allowing the sensitivity of target delineation to the chosen cutoff to be evaluated. The outputs are intended for qualitative interpretation and risk-aware drill targeting rather than resource estimation, and we discuss key limitations arising from incomplete provenance metadata for geophysical products and heterogeneity of historical sampling.
Read the full abstract

All authors Margarita Veshchezerova, Egor Barashov, Evgenii Gusev, Michael R. Perelshtein, Arlan Kasymzhan, Bolat M. Kabaziev, Nurlan Y. Askarov

Machine Learning (Statistics)

What Fixed-Rollout pass@k Evaluations Can Identify

Repeated-sampling evaluations increasingly extrapolate pass@k far beyond the number n of samples collected per problem. We show that, in the pooled/random-task conditional-Binomial model, fixed-n success counts identify only the n free moments of the latent per-task success distribution. Consequently, direct pass@k is identified for k <= n, but generic extrapolated pass@k, tail exponents, and tail constants are not identified for k > n, even with arbitrarily many exchangeable tasks at the same rollout budget. This is stronger than the observation that the usual estimator is undefined beyond n: it characterizes the information missing from the fixed-depth count-law experiment. We give exact count-law-preserving constructions with incompatible extrapolations, state the exceptional unique-extension case, and compute sharp population identified intervals through Hausdorff principal representations. On the public 10,000-rollout-per-problem release of Brown et al., counterfactual n = 16 evaluations leave failure at k = 1000 ambiguous by factors from 1.5 to over 2,600 across four MATH/GSM8K/CodeContests configurations. The calibration shows that intermediate-scale failure share alone does not determine width. Our result does not reject parametric inference-time scaling laws; it supplies the nonparametric baseline against which their assumptions can be evaluated. We give an exact, conservative one-coordinate finite-task confidence certificate and a reporting standard separating direct estimates, identified sets, and model-conditioned forecasts.
Read the full abstract

Machine Learning (Statistics)

Critical initialization destabilizes higher input derivatives in wide scalar-input networks

The edge-of-chaos condition preserves first-order input perturbations in wide randomly initialized networks, but physics-informed losses, score matching and derivative regularization depend on higher input derivatives. For smooth scalar-input fully connected networks, using a joint Gaussianity of the finite derivative jet that holds in the infinite-width limit at each fixed depth, we derive mean-field recursions through third order that are exact at the variance fixed point, with finite-depth corrections that decay geometrically. At criticality, the first-derivative variance is depth-invariant, whereas the second-derivative variance grows linearly whenever the activation has nonzero curvature. The resulting third-order system closes on mean-field susceptibilities. For residual networks with branch scale L^{-1/2}, we prove that every fixed finite derivative order has uniformly bounded variance under explicit regularity assumptions. Simulations verify the critical growth laws, the residual bound, and the closed recursion. The results concern initialization, not trained-network performance.
Read the full abstract

Cryptography and Security

In RAG We Trust? Measuring Robustness of Retrieval-Augmented Generation Under Post-Retrieval Context Tampering

Retrieval-augmented generation (RAG) grounds a language model in retrieved documents, which reduces hallucination but creates a new attack surface: if retrieved text is tampered with, the model may repeat the falsehood. We study how much a small quantized model, Llama 3.1 8B, degrades when a fraction of its retrieved context is poisoned. Three corruption strategies are tested, entity swap, number swap, and negation, each applied to zero, one, two, or three of the three retrieved passages, over a factorial sweep of 588 runs on a fact-checking task built from FEVER. Accuracy falls from 77.9% on clean context to 43.5% when all three passages are corrupted. Entity swap flips the largest share of answers that were correct on clean context. Number-based corruption stays flat while poisoned passages are a minority and jumps once they form a majority, a pattern we re-check with query-level bootstrap intervals. The model rarely invents new falsehoods; its dominant reaction is to abstain, and a lexical overlap proxy of unsupported generation falls under attack rather than rising. The study is a small-scale measurement with coarse automated labels; we treat the strategy contrasts as suggestive until decoding is controlled and stronger adjudication is in place.
Read the full abstract

Software Engineering

Talking to Itself While Coding: What Makes Comments Help Code Generation?

Large Language Models (LLMs) often generate natural-language comments while writing code, and these comments become part of the context used to generate the code that follows. However, it remains unclear which properties of comments affect code-generation performance. We study this question through observational analyses and controlled interventions. On LiveCodeBench, neither comment frequency nor broad comment intent reliably predicts pass@1. We then prefill weaker recipient models with comment blocks written by stronger source models, allowing us to separate comment surface form from the solution content they convey. Comments from source solutions that pass the tests raise recipient pass@1 by 17.2% on average. In contrast, comments describing failed solutions provide no reliable gain, while comments written for a different problem reduce pass@1 by 20.8%. Finally, across a wide range of models and prompt variants, most recipient models show no significant recovery of the external-comment gain, and the best case recovers only 24%. These results show that comments help code generation not merely because they are comments, but because they can provide correct solution content that prompting cannot reliably elicit.
Read the full abstract

All authors Dangfeng Pan, Zhensu Sun, Cenyuan Zhang, David Lo, Xiaoning Du

Machine Learning

Distribution-Consistent Inference for Dynamic Sparse Mixture-of-Experts

Mixture-of-Experts (MoE) architectures have emerged as a powerful paradigm for scaling model capacity while preserving efficient inference in large foundation models. However, most MoE models use a fixed top-$k$ expert selection policy, assigning the same expert budget to every token even when fewer experts may be sufficient. Inference-time dynamic top-$k$ routing can reduce computation without retraining, but existing methods often overlook the distributional shift caused by deviating from the training-time routing configuration. We show that reducing the number of activated experts consistently increases the RMS scale and variance of SMoE outputs, inducing a representation mismatch that contributes to downstream performance degradation in addition to the loss of expert capacity. To address this correctable component, we propose Layer-wise Distribution Alignment (LDA), a lightweight inference-time correction that uses layer-wise calibration statistics to align reduced-routing representations with the default configuration. Across multiple SMoE LLMs, benchmarks, and routing strategies, LDA recovers much of the performance lost induced by the distributional shift under reduced routing while preserving sparse-inference efficiency with negligible overhead.
Read the full abstract

Machine Learning

Scaling Post-Training Ternarisation to Qwen3-8B Capability Retention, Reproduction, Lossless Packing, and Packed Execution

Ultra-low-bit language models promise reductions in storage and memory traffic, but a nominal "1.58-bit" label does not specify the deployed representation or its execution cost. We study a scale-up of an aggressive post-training conversion pipeline from Qwen3-4B to Qwen3-8B. The conversion uses KOTMS rotation, E2M-ATQ adaptive ternarisation, and GPTQ-style error compensation in a weight-only A16 configuration. We do not claim these algorithms as new. Our contribution is the end-to-end scale-up characterisation: an external reproduction gate, matched 4B/8B capability analysis, cross-corpus perplexity, effective-bit accounting, lossless lattice-aware packing, and direct packed execution. The 8B model reaches a three-corpus perplexity ratio of 1.361x, with WikiText-2, C4, and PTB ratios of 1.318x, 1.393x, and 1.371x. On eight zero-shot tasks at n = 500, mean accuracy is 64.6% versus 72.4% for FP16, corresponding to 78.5% chance-corrected retention and a 7.8-point absolute cost. The matched 4B run retains 69.6%, yielding an 8.9-point 8B advantage. The packed checkpoint is 8.24 GiB and preserves the recorded perplexity to measurement precision. Direct packed execution reaches 15.52 tokens/s in 7.35 GiB, while a preliminary packed GEMV remains slower than FP16 cuBLAS. The result is a validated scale-up baseline: model size improves robustness to aggressive post-training discretisation, actual serialisation is solved for the measured artefact, and direct execution is feasible, while broader seeds, calibration distributions, and kernel optimisation remain open.
Read the full abstract

Information Theory

Optimality of Kløve Arrays within the Symmetric Kløve-Mossige Class

Rajam$\unicode{228}$ki and Koivunen asked whether minimum-redundancy symmetric Kl$\unicode{248}$ve-Mossige arrays with contiguous sum co-arrays are always Kl$\unicode{248}$ve arrays. We give a computer-assisted proof that, at every fixed sensor count, every maximizing sensor set belongs to the Kl$\unicode{248}$ve class. The argument classifies the overlaps between a generator and its shifted reflection, then bounds the aperture of a hypothetical optimizer outside the Kl$\unicode{248}$ve class. Comparing these bounds with a classical Kl$\unicode{248}$ve construction settles all sensor counts at least 330. An exact integer certificate covers the remaining counts from 2 through 329 and matches every equality case to a Kl$\unicode{248}$ve array as a complete set. Consequently, optimization within the full symmetric Kl$\unicode{248}$ve-Mossige class reduces to the previously known search over Kl$\unicode{248}$ve parameters. The theorem concerns this specified class, rather than unrestricted sparse arrays or all restricted additive bases.
Read the full abstract

General Relativity and Quantum Cosmology

On The Regularized McVittie Black Bounce and a Family of Traversable Wormholes

The McVittie metric describes a self-gravitating compact object embedded in an expanding universe. It inherits two singularities: a curvature singularity at $r \to 0$ and a cosmological singularity at $a(t) \to 0$. Motivated by the construction of black-bounce geometries, we regularize both of these singularities and propose a regularized McVittie metric. The central singularity is regularized by replacing $r \to \sqrt{r^{2}+b^{2}}$, leading to a geometry that interpolates between a cosmological black hole, a black bounce, and a traversable wormhole. Similarly, the cosmological singularity is regularized by introducing a non-vanishing scale factor $a \rightarrow \sqrt{a^{2} + a_b^{2}}$. The resulting spacetime is supported by an effective imperfect fluid with finite anisotropic stress. We investigate the formation of the apparent horizon, analyze the circular geodesic structure, and argue that the geometry can also be interpreted as a conformally evolving Morris-Thorne wormhole embedded in a regular cosmological background.
Read the full abstract