Machine Learning
Papers
Fresh from arXiv, typeset properly. Highlighted words trade on the market.
All of arXiv
Thursday 3 September
Expert pruning reduces the memory and serving cost of Mixture-of-Experts (MoE) models by removing low-importance experts identified by the router, assuming router probabilities provide a reliable importance signal. We observe that this assumption breaks down under over-dispersed routing, a regime associated with aggressive load-balancing during training, in which tokens are distributed nearly uniformly across experts and importance signals collapse. In this regime, perplexity does not predict downstream task accuracy: on gpt-oss-20B, the lowest-perplexity pruning configuration yields the worst mathematical reasoning, while the highest-perplexity configuration preserves it. This does not occur under standard routing (e.g., Mixtral-8x7B-Instruct), where perplexity and accuracy degrade together. Pruning under over-dispersed routing also exposes a capability trade-off in which no single scoring metric dominates: activation-aware scoring preserves mathematical reasoning but severely degrades knowledge-intensive science (an 18-point gap on GPQA), whereas frequency-based scoring exhibits the reverse. We propose Minimax Expert Score Allocation (MESA), a domain-aware method that iteratively boosts importance scores for experts serving whichever domain is currently worst-affected, minimizing worst-case domain degradation rather than average accuracy. At 25% expert pruning MESA achieves the smallest worst-case degradation across domains, outperforming activation-aware baselines on 7 of 11 benchmarks at a correspondingly reduced memory footprint, and it generalizes to gpt-oss-120B, Gemma-4-26B-A4B, and OLMoE-1B-7B. Our results indicate that over-dispersed routing is a qualitatively distinct pruning regime in which standard assumptions fail, and that recognizing it is a prerequisite for principled expert pruning of load-balanced MoE models.
Computation and Language
TRILOGUE: A Trilingual Spoken Dialogue Fact-Checking Benchmark with Evidence and Paired Audio
Modern misinformation is often heard before it is read, yet fact-checking systems are still evaluated mainly on clean written claims. Spoken dialogue remains different even when systems operate on transcripts: claims may be distributed across speakers and turns, depend on prior context, and become harder to verify when Automatic Speech Recognition (ASR) errors distort the available text. Prior spoken dialogue fact-checking resources are small, English-centric, or focused on annotation rather than end-to-end benchmarking, leaving no large multilingual benchmark with paired speech and turn-level labels. We introduce TRILOGUE (TRIlingual spoken diaLOGUE fact-checking), a large-scale trilingual benchmark of source-grounded spoken dialogues in English, Russian, and Kazakh. It contains nearly 12K dialogues, 187K turns, and 390 hours of paired audio with ASR transcripts and word-level timestamp alignments across all three languages, including nearly 5K human-recorded Russian and Kazakh dialogue files. TRILOGUE supports claim check-worthiness detection, source-article evidence retrieval, and claim verification with claim-only, gold-evidence, and retrieved-evidence inputs. Baselines show that ASR degradation and cross-lingual transfer remain challenging, especially for Kazakh, while retrieved source evidence substantially narrows the gap to gold-evidence verification.
Read the full abstract
Computational Physics
Reduced Precision Diffusion Synthetic Acceleration for S$_N$ Neutron Transport in LLNL's ARDRA using Hypre
Fast solution times in production discrete ordinates transport codes often require the use of diffusion synthetic acceleration (DSA). A good implementation of DSA will substantially reduce the number of iterations to converge the transport solution at the cost of solving a diffusion-like linear system at each iteration. Often fewer transport iterations will lead to smaller times to solution. Recent work has shown the promise of reduced precision preconditioning. Time and memory cost can be decreased as long as the reduced precision does not substantially degrade the convergence behavior of the higher precision linear system being solved. In this paper, we discuss the algorithmic implications of accelerating transport iterations in double precision (64 bit) while using a reduced precision (32 bit) DSA solve. We initially implement a stand alone code to investigate any theoretical limitations of reduced precision DSA in slabs. Then we implement reduced precision DSA in ARDRA, Lawrence Livermore National Laboratory's (LLNL) discrete ordinance neutral particle transport code, coupled with hypre, LLNL's scalable linear solver and multigrid methods library capable of mixed precision linear solves Then we solve various neutron transport problems of interest. We find that in most circumstances, DSA acceleration in reduced precision has little to no impact on the acceleration properties of DSA when used with source iteration in most circumstances. However, as tighter tolerances are required, reduced precision DSA may fail to successfully accelerate the transport solve and increase time to solution.
Read the full abstract
Earth and Planetary Astrophysics
Circumbinary planets in coplanar triple-star systems: I. Minimum eccentricity variation region
We aim to determine the conditions under which Lidov-Kozai oscillations can arise for circumbinary planets in hierarchical triple-star systems, and to identify the orbital regions where competing secular perturbations minimize eccentricity excitation. We analyzed the statistical properties of hierarchical triples from the Multiple Star Catalog to construct representative synthetic configurations. We then performed N-body simulations of circumbinary planets embedded in these systems, systematically exploring their orbital evolution and long-term stability across a range of semimajor axes and inclinations. For representative compact hierarchical triple-star systems with masses $m_0 \sim m_1 \approx 0.75\,{\rm M}_\odot$ and $m_2 \approx 0.6\,{\rm M}_\odot$ and eccentricities $e_1 \sim 0.1$ and $e_2 \sim 0.4$, dynamically significant stable circumbinary regions exist only in sufficiently hierarchical configurations, with period ratios $P_2 /P_1 \gtrsim 10^2$. In this regime, the secular competition between the tertiary perturbations and the apsidal precession induced by the compact inner binary determines both the onset of Lidov-Kozai oscillations beyond a critical distance ($a_{\rm LK}$) and the location of a region where eccentricity variations are minimized, i.e., the minimum eccentricity variation region (MER). We derived analytical estimates for these dynamical features, showing that the MER is well described by the analytical quantity $a_{\rm short}$ on short and intermediate timescales, while on secular timescales it converges toward the equilibrium prediction $a_{\rm ME}$. These analytical estimates agree well with N-body simulations. Application to the observed triple system WDS 08403+1921 confirms that $a_{\rm LK}$ and $a_{\rm ME}$ accurately identify the main dynamical features of the stability map.
Read the full abstract
Chemical Physics
Learning multistate kinetics with a variational multistate committor network
The long-time dynamics of complex molecular systems often involves rare transitions across networks of metastable states. Building on transition-path theory, which provides a rigorous framework for describing rare transitions between two metastable states, we introduce the variational multistate committor network (VMCN), a neural framework that learns the probabilities of reaching each metastable state directly from molecular simulation data. From this representation, VMCN identifies state-specific commitment, candidate transition regions and committor-consistent pathways between state pairs, and an effective kinetic network characterized by transition rates. The model is trained using finite time-lag trajectory data together with boundary conditions defined on conservative state cores. Applications to a triple-well potential, trialanine isomerization, and the $c$–ring rotation in the V$_{\rm o}$ domain of a vacuolar ATPase show that VMCN recovers metastable organization, provides committor-consistent descriptions of transition mechanisms, and estimates state-to-state kinetics. VMCN further provides diagnostics for incomplete state decompositions and enables adaptive exploration of candidate metastable states and their connecting regions. By integrating VMCN with generative committor-guided path sampling (Gen-COMPAS) for chignolin, we start from two end point structures, identify a misfolded state and a candidate folding intermediate, and we direct subsequent sampling toward the resulting multistate transition network.
Read the full abstract
General Relativity and Quantum Cosmology
From Nonmetricity Operators to the Physical Spectrum: A Branch-Complete Analysis of Local Four-Derivative Symmetric Teleparallel Gravity
We develop a branch-complete classical free-spectrum analysis of local parity-even symmetric teleparallel gravity containing all pure-gravity nonmetricity operators through four derivatives. After canonicalisation and integration by parts, the nonlinear action contains 116 independent operators distributed as $5+13+29+69$ among $Q^2$, $(\nabla Q)^2$, $Q^2\nabla Q$ and $Q^4$. Around Minkowski spacetime in the coincident gauge only the first two sectors contribute to the quadratic action, which reduces to four two-derivative and five four-derivative bilinears. We map the original coefficients to this nine-dimensional basis, construct the Hessian and momentum kernel, and classify local linear gauge symmetries. Unrestricted linearised diffeomorphism invariance imposes six independent relations and leaves a three-parameter family $(A,B,C)$. We also recover TDiff, Weyl, WTDiff and additional accidental symmetry branches. Barnes–Rivers decomposition, gauge fixing, exact inversion and conserved-source saturation reduce the physical exchange to spin-two and scalar factors $A+2Cq$ and $2A+(3B-2C)q$. The generic regular spectrum contains a massless graviton, a massive spin-two state and a massive scalar, with eight degrees of freedom. The massive spin-two residue is necessarily opposite to that of the healthy massless graviton. The regular classification leaves only GR/STEGR, $A>0$ with $B=C=0$, and the scalar extension, $A>0$, $B>0$, $C=0$, as fully healthy Minkowski loci. The latter has three degrees of freedom and mixed ultraviolet behaviour, with tensor exchange scaling as $k^{-2}$ and scalar exchange as $k^{-4}$. Thus, within this finite local metric theory, full four-derivative tensor suppression requires the finite massive spin-two pole with opposite residue.
Read the full abstract
Quantum Physics
Proof-of-principle long-distance Sagnac twin-field quantum key distribution network
Twin-field (TF) quantum key distribution (QKD) offers a promising approach to long-distance QKD networks due to its superior performance over large channel losses. Due to specialized hardware requirements, nearly all long-distance TFQKD demonstrations have only two users exchanging keys, rather than a network with three or more users. In this work, we experimentally demonstrate a proof-of-principle three-user-pair Sagnac TFQKD network spanning 127-km using single-photon avalanche detectors without any active phase stabilization or postcompensation. We implement efficient procedures for maintaining polarization stability and circumventing Rayleigh backscattering noise to achieve a stable Sagnac interference visibility of $93\pm1$ % over one hour. Using measured data for the signal basis and simulated data for the decoy basis, an asymptotic secure key rate of $1.398\times10^{-5}$ bits per pulse is achieved over an asymmetric communication channel with 102-km fiber and 45-dB overall loss. To our knowledge, this is the first TFQKD network without active phase stabilization or postcompensation achieved over long fibers. Our results represent a highly practical and cost-effective approach to long-distance QKD networks.
Read the full abstract
Probability
A Bootstrap Proof of the Abstract Wiener Space Theorem for Fréchet Spaces
In this paper, we provide an alternative proof of a result due to Gross (1960s), extending it to the Fréchet space setting. Our argument adapts the proof technique developed by Bogachev in the Banach case, as presented in the monograph Gaussian Measures, to the more general framework of separable Fréchet spaces, simplifying several steps along the way.
Read the full abstract
Machine Learning
Conformity Breaks Conformal Prediction
A conformal certificate can be valid when an LLM answers alone and invalid when the same LLM sees peers that unanimously assert a wrong answer. The question is unchanged; the model's score for the correct answer changes. We call this a score-mechanism shift: clean calibration certifies how the model scores answers alone, but not how it scores them under peer pressure. We show that this shift silently breaks conformal prediction in multi-agent LLM systems. Across open-weight models and multiple-choice QA tasks, coverage falls from a calibrated 90% to 74% under unanimous-wrong peers at the standard alpha = 0.10 operating point. The average hides a sharper failure: by targeting the low-confidence items the certificate still covers, an attacker nearly halves coverage on that subgroup, from 87% to 47%, while the monitored average remains much higher. The failure also reaches the decision layer: a system that should escalate when uncertain can instead become confident enough to act on the attacker's wrong answer. Standard conformal fixes do not solve the problem, because the question distribution has not changed; the model's scoring behavior has.
Read the full abstract
Artificial Intelligence
HarvestBench: Measuring Whether LLM Agents Will Pay to Avoid Killing Animals
HarvestBench is the first benchmark to 1) put a price on avoiding a side effect and 2) name the side effect as a living creature. Nine LLMs each drive a crew of two tractors to gather a corn harvest. The animals in their path are not part of the goal function. When an animal blocks the route the autopilot pauses and asks the agent whether to drive over it for free or swerve for a given fuel cost. All scoring is programmatic and does not involve LLM judges. Kill rates range between 0.4% and 98.8%, though the kill rate is not ordered by capability. Every model competently avoids damaging rock hits, so every animal killed is a choice, rather than an accident. Under the morality briefing the kill rate was under 6% in 5 of 6 reasoning models. Removing it (the neutral briefing) raised the kill rate to above 84% in all six models. Every model kills wild animals more often than farmed ones. Four out of six models' kill rate per answered encounter were sensitive to price changes. The moral instruction is also fragile. Four bullets of driving mechanics change Sonnet 5's kill rate from 3% to 18% and Gemini 2.5 Flash's from 4% to 39%. A moral instruction in a system prompt is overridden by a short block of operating instructions and a value that can be ignored that easily is not a good method of ensuring agents are aligned.
Read the full abstract
Physics and Society
Publish or Perish: A population-dynamics model of the corruption of peer review
Peer review is commonly treated as a mechanism of quality control. This article examines what happens when participants are split between two groups. Explicators assess whether submissions present logic-based arguments (whether deductive, inductive or abductive), independently of whether they agree with the reasons or the conclusion. Interpreters assess submissions according to their outlook. Their employment of logic is opportunistic, as a means of presenting their own views, but not as a means of assessing others' views. Using a deterministic population-dynamics model of iterative games of "Publish or Perish," this article shows how this asymmetry permits interpreters to increase their population share even when they begin as a small minority. It compares fixed and survivor-drawn reviewer panels, alternative replenishment rules, hostile explicator responses, generational differences in ratios, and different panel sizes. Under survivor-drawn review, explicator hostility to interpreters produces an unstable equilibrium at an explicator share of $q^* = 2 - \sqrt{3} \approx 0.2679$: below this threshold, interpreters can survive without explicator cooperation and continue toward dominance. The threshold is not universal, however. Exogenous replenishment in favour of explicators can eliminate the point of no return, and sufficiently lagged, generational replenishment can permit explicators to recover after it has apparently been crossed. The model therefore identifies not only a mechanism by which peer review becomes corrupted, but also the institutional conditions determining whether that corruption becomes self-sustaining or reversible.
Read the full abstract
Computation and Language
GRACE: Graph-Grounded Reflective Agent Copilot Engine for Expert-in-the-Loop Knowledge Expansion
Large language models deployed in high-stakes settings frequently generate plausible but ungrounded claims. Standard retrieval-augmented generation (RAG) pipelines offer limited remedy, since they retrieve isolated passages without tracking cross-document evidence relationships or quantifying uncertainty. We introduce GRACE (Graph-grounded Reflective Agent Copilot Engine), a framework that deconstructs LLM responses into atomic claims and grounds them against trusted knowledge priors within a weighted bipartite graph. Edge weights encode the closeness of each claim to the priors, enabling weighted centrality analysis that classifies claims as Grounded, Refuted, or Boundary. Such classification identifies not just hallucinations but also novel or contested claims at the frontier of the model's knowledge. To efficiently allocate human or agent resources, we formulate a Return on Attention (RoA) objective that defers a claim to expert review only when its priority-weighted uncertainty exceeds the cost of verification. Claims verified by experts are promoted to new evidence anchors, closing a validator-LLM evolutionary loop that expands the knowledge base across iterations. We evaluate GRACE across multiple language models and on datasets spanning both general and domain-specific knowledge. Our results show that our knowledge base serves as a reliable foundation for retrieval that outperforms RAG baselines, and that the RoA framework efficiently selects valuable boundary knowledge for expert verification. These findings demonstrate that graph-structured representations combined with expert-in-the-loop verification can mitigate hallucination at the system level rather than at the generation level. Code available at https://github.com/johnsk95/grace_code
Read the full abstract
High Energy Physics (Phenomenology)
Stability of Collective Neutrino Oscillations – A Distributional Approach
We study the stability of collective neutrino oscillations using a distributional approach motivated by the statistical mechanics of Kuramoto synchronization. Treating the ensemble of neutrino flavor polarization vectors in the thermodynamic limit $N\to\infty$, we derive an exact nonlinear Fokker–Planck (continuity) equation for the one-body distribution $F(\vec{\mathbf{S}},\omega,t)$ on the flavor sphere. This equation admits a two-parameter family of azimuthally symmetric stationary solutions, whose stability we analyze by linearizing around them. The resulting eigenvalue condition determines the growth or decay rate of small perturbations from any initial distribution – not merely from a state close to full flavor coherence – thereby going significantly beyond the conventional linear stability analysis of collective modes. In special limits the condition reproduces known synchronization thresholds in the two-beam model, providing a non-trivial check of the framework. We present analytical results for the eigenvalue equation and explore stability phase diagrams for physically relevant frequency distributions.
Read the full abstract
Geometric Topology
A topological version of the Cartan-Hadamard Theorem and asphericity complexity
We investigate a topological version of the Cartan-Hadamard theorem that allows one to study asphericity of spaces via distinguished families of paths. A topological space has the distinguished path property (dpp) if there exists a continuous map from the space $\Pi(X)$ of homotopy classes of paths (relative endpoints) to the path space $X^I$ that is a right inverse to the canonical quotient map. For complete metric spaces endowed with a locally convex metric, the distinguished paths are precisely the local geodesics. We show that if $X$ has the dpp, then its universal cover is contractible and, in particular, $X$ is aspherical. The space $\Pi(X)$ is a fiber bundle over $X$ whose fiber is the universal cover of $X$, and in the case of Riemannian manifolds of non-positive curvature it is naturally isomorphic to the tangent bundle. We prove that every aspherical CW-complex that is locally finite or countable has the dpp, and this allows us to reinterpret asphericity of CW-complexes in terms of the existence of continuous sections. We also define and study the notion of asphericity complexity of spaces by means of local sections from $\Pi(X)$ to $X^I$ and relate it to the concept of equivariant topological complexity introduced by Colman and Grant.
Read the full abstract
Computer Vision
Segmentation of the aorta in 4D flow MRI using 4D convolutional kernels and learning from sparse annotations
Automated aortic segmentation in 4D flow MRI is essential for reproducible hemodynamic assessment but is limited by scarce dense annotations and high computational demands. We developed a fully automated 4D (3D+time) U-Net for segmenting the ascending aorta, arch, and proximal descending aorta, using a parameter-efficient hybrid 4D kernel to capture temporal context and sparse 4D labels derived from existing 2D expert contours and centerlines, thereby avoiding the need for dense 4D annotations. Training comprised 268 scans from 8 centers and 2 vendors, with evaluation on an internal test set (32 scans) and an external post-contrast set (30 scans; different site, protocol, and annotator), compared against frame-wise 3D networks and two semi-automatic references. Against time-resolved annotations, the 4D U-Net achieved Dice scores of 0.927 (internal) and 0.911 (external), versus 0.919/0.847 for the 3D U-Net, 0.893 for static PC-MRA, and 0.808 for registration-based propagation; differences were small in systole but pronounced in diastole. Agreement with expert contours for peak velocity, net flow, axial and circumferential wall shear stress, and diameters was excellent (ICC >=0.954 internal, >=0.980 external), while semi-automatic references performed worse. The method thus provides reproducible, time-resolved aortic segmentation for automated hemodynamic analysis and generalizes across multicenter, multivendor, and independent post-contrast data. The model is publicly available.
Read the full abstract
Computer Vision
ICM-Bench: Person-Level Identity Reasoning in Multimodal Agents with Long-Term Memory
Long-horizon multimodal agents should remember not only what happened but also who participated. This capability depends on linking recurring faces, voices, names, person-associated objects, events, and social relations to consistent identities over time. Existing long-video and multimodal-agent benchmarks measure broad memory question answering, but they do not isolate the ability to maintain recurring person identities and reason over their cross-time relations. We introduce ICM-Bench (Identity-Centric Memory Benchmark), which, to the best of our knowledge, is the first benchmark specifically designed to evaluate identity-centric reasoning over long video memories in multimodal agents. The benchmark contains 839 synthetic clips spanning 141 minutes and 1,217 open-ended questions about six recurring adults in a one-year life album. A theme-configurable pipeline generates the video collection and associates each question with its target identities and traceable supporting evidence. We compare direct caption-memory baselines, memory-augmented agents, and graph-retrieval systems. Gemini 3.1 Pro achieves the highest overall accuracy of 74.0%, yet its score falls to 60.3% on questions that require long-term identity profiles. The results show that current systems recover many event-level memories but remain less reliable when evidence must be accumulated around a stable person.
Read the full abstract
Astrophysics of Galaxies
KKR 18: A Late Arrival Galaxy?
KKR 18 is a dwarf irregular galaxy in the Local Void at a distance of 9.46 Mpc. It is gas rich and was experiencing vigorous star formation 20 - 100 Myr ago. Its metallicity is 4% of Solar from an emission spectrum, qualifying is as extremely metal poor (XMP). Its most remarkable characteristic is an extremely deficient Red Giant Branch. This feature is shared with a small number of other dwarf galaxies and it is inferred that they lack ancient stellar components. It is suggested that these are late arrival galaxies (LAG).
Read the full abstract
Signal Processing
Opportunistic Data Offloading for Robotic Operations in Dynamically Varying Channel Environments
This paper studies energy-efficient operation of autonomous vehicles (AVs) in dynamic environments with moving obstacles and while communicating over mmWave channels. The obstacles induce severe attenuation of the mmWave channel resulting in a highly dynamic communication environment. In this setting, we consider the problem of jointly optimizing motion and communication energy for an AV that safely navigates among dynamic obstacles toward a designated destination while ensuring timely transmission of onboard sensing or telemetry data over mmWave channels. We then seek a real-time methodology to compute energy-efficient trajectories in a setting where dynamic obstacles induce both safety constraints and time-varying mmWave blockage, leading to tightly coupled motion-communication trade-offs. We propose a nonlinear model predictive control (NMPC) framework that enables anticipative communication and motion decision-making and energy co-optimization, augmented with a control barrier function (CBF) to ensure safety. Extensive simulation results demonstrate the effectiveness of our approach, reducing total energy consumption by up to 37.3% compared to baseline strategies. Overall, our results demonstrate that the proposed NMPC-based framework significantly enhances energy efficiency and performance of AVs under dynamic, blockage-sensitive mmWave communication constraints.
Read the full abstract
Emerging Technologies
STEMPix: A Phase-Transition-Material-Based Pixel Sensor for Resolving Edge-Movement Direction
This paper proposes a spatio-temporal edge-movement direction pixel (STEMPix) for generating compact direction-aware edge movement information inside a CMOS-compatible image sensor array. The proposed design targets specialized sensing applications where local boundary movement is more important than full-frame intensity reconstruction. Instead of transferring full multi-bit frames for external processing, STEMPix generates a 3-bit local edge direction code (LEDC) by combining pixel-level temporal change information with neighboring-pixel spatial edge information. We design the architecture using a two-tier organization, where the photodiode layer is separated from the computation layer to preserve light-collection area while accommodating the additional in-array processing circuitry. The proposed circuit is evaluated through HSPICE transient simulations. The estimated implementation achieves a horizontal pitch of 1.73 {\mu}m, a vertical pitch of 2.36 {\mu}m, and a geometric fill factor of 95.47%. The average active switching energy is 0.465 fJ per LEDC operation across representative edge-movement cases. The proposed STEMPix operation also supports global-shutter capture and dynamic thresholding. These results indicate that STEMPix can provide a compact and scalable front-end representation for edge-movement-aware sensing systems.
Read the full abstract
Computation and Language
What Attention Recalls and Recurrence Controls in Hybrid Language Models
Hybrid language models combine attention with a fixed-size recurrent state, but the role of each channel remains unclear. We introduce two cache-level interventions. Split-prefill keeps only the KV cache or only the recurrent state from a prefilled context, then generates an answer. State-swap pairs the KV cache from one context with the recurrent state from another in a single forward pass. On Qwen3.5 and Falcon-H1, the two channels split sharply by function. Exact retrieval survives only through attention (64-98% of full accuracy) and collapses to zero through recurrence. Output language and persona reverse the pattern: both survive recurrence (70-80% and 3-5x) while KV-only drops to ~1% language accuracy. State-swap confirms this causally: the answer takes its value from the KV side and its language from the recurrent side. Recurrent-only generation also accepts words that were never in the context but share meaning or parts with seen items. Attention provides a lookup over what was said; the recurrent state shapes how the model says it next.
Read the full abstract
Numerical Analysis
Improving the efficiency of a Hyperpower method for approximating inverse matrices
In this manuscript, scalar norm-type accelerators are used for first time in order to increase the order of convergence of known iterative methods for solving the matrix equation $X^{-1}-A=0$, while decreasing the number of matrix-matrix products. Specifically, our proposed scheme upgrades the order of convergence in a $50\%$, achieving the sixth-order of convergence with the computational cost of fourth-order Hyperpower method, in terms of matrix-matrix multiplications. In the main results, not only the convergence and its order is proven, but also its efficiency and stability. This new technique provides good numerical results, even compared with existing fourth-, sixth- and eighth-order schemes in large matrices of up to $10^6$ entries, but also promises the opening of a new kind of procedures with good performance and scalability. Finally, an application to digital image restoration is made to show the behavior of the method, compared with existing ones.
Read the full abstract
Chemical Physics
Recovering molecules from coarse-grained beads: free-energy-conditioned generative backmapping across chemical space
Transferable coarse-grained (CG) force fields compress chemical space: by aggregating atoms into a reduced set of interaction beads, models such as MARTINI reduce the number of distinguishable compounds by roughly three orders of magnitude, making high-throughput screening of thermodynamic properties tractable across soft matter, with drug–membrane permeability as a well-developed example. The compression is lossy and, so far, one-way: a screen returns a combination of beads, with no established route back to the compounds it stands for. Recovering those compounds–compositional backmapping–is a one-to-many inverse map, distinct from the better-studied conformational problem of rebuilding atomic coordinates from a known mapping. Here we formulate compositional backmapping as conditional graph generation by introducing juniper, a discrete denoising diffusion model over molecular graphs conditioned on the octanol–water partition free energy $\Delta G_{\mathrm{W} \mapsto \mathrm{O}}$, the principal driver of MARTINI bead type assignment and hence a proxy for bead identity. Trained on molecules of up to 9 heavy atoms mapped onto one or two beads, juniper generates molecules that are 93% valid and 92% unique for two-bead targets, and whose $\Delta G_{\mathrm{W} \mapsto \mathrm{O}}$ distributions track the target $\Delta G^{\mathrm{CG}}_{\mathrm{W} \mapsto \mathrm{O}}$ linearly ($r^{2} \geq 0.96$), departing only in the hydrophobic and hydrophilic tails. Although the model receives no chemical information beyond a single scalar, the functional groups shift systematically with the imposed free energy, from branched hydrocarbons at the apolar end to amides, imides, and isocyanates at the polar end. A bead combination flagged by a CG screen can therefore be turned into candidate molecules for atomistic study or synthesis.
Read the full abstract
Number Theory
SIC dimension towers via cyclotomic polynomials
We prove a structure theorem for the dimension towers~$\{d_k(D)\}_{k\geq0}$ which arise in the number-theoretic formulation of Zauner's SIC-POVM conjecture over a real quadratic field~$K=\Q(\qD)$. If~$\eps$ denotes the first totally positive power of a fundamental unit of~$K$ and $t_k = \eps^k + \eps^{-k}$ the trace of its $k$-th power, then the SIC dimension tower~$\{d_k=1+t_k\}_{k\geq0}$ is the level~$m=3$ row of an infinite two-dimensional cyclotomic array~$\{\Psi_m(t_k)\}_{m\geq1,k\geq0}$ attached to~$K$, while the auxiliary factors~$(d_k+1)$ and $(d_k-3)$ are its ramified levels $m=2$ and $m=1$. Here~$\Psi_m$ denotes the minimal polynomial of~ $\zeta_m+\zeta_m^{-1} = 2\cos{2\pi/m}$. This construction arose initially from an attempt to formulate relations among SIC dimensions in $q$-algebraic terms. The central object is a single closed composite norm relation for the two-parameter family $c_{m,k}=1-\zeta_m\eps^k$ over the cyclotomic field tower $\{K(\mu_m)\}_{m\geq1}$. This framework sheds new light on the mod-$p$ analogue of Leopoldt's conjecture, by placing the central 3-symmetry of Zauner's conjecture within a broader arithmetic context. Away from the primes dividing~$2mD$, the valuations $v_p(\Psi_m(t_k))$ at every fixed level~$m$ are described exactly in terms of a single local unit valuation, which is then related, through the~$p$-adic class number formula, to the corresponding $p$-adic $L$-value.
Read the full abstract
Category Theory
Algebra of spectral duality structures
This paper establishes the algebraic and categorical foundations of spectral duality structures (SDS). We show that the class of all SDS admits the structure of a graded symmetric monoidal category, where the degree $K$ is the fundamental invariant indexing the strata. We define operations of Cartesian product and disjoint union, morphisms preserving involution and weights, and prove that $C^\ast=1/(1+\sqrt K)$ is a functor constant on each stratum. We classify objects up to isomorphism by the combinatorial type $(k,f)$, the degree $K$, and a multiset of inversion classes $[r]=\{r,r^{-1}\}$. In each degree, skeletons are exactly the finite $C_2$-sets over the additive group of the reals with $C_2$ acting by $t\mapsto -t$, which explains the monoidal structure and identifies the classification with the decomposition into orbits. Furthermore, we prove that the response rank is the rank of an explicit matrix built from the amplitudes and the weights, and hence exactly the number of non-trivial pairs $k$ under the full-rank condition, and we connect the categorical structure with the Fisher–Rao geometry of Paper V. The SDS, originally discovered in stochastic resetting problems, is thus an autonomous mathematical object with a rich algebraic structure and a natural geometric realisation.
Read the full abstract
Networking
Waves on the Walls: Empirical Characterization of mmWave Lateral Waves for Enhanced Indoor Coverage
High-frequency millimeter-wave (mmWave) communication systems are constrained by the surrounding environment, where walls are traditionally treated as obstacles that block or reflect signals indoors. Consequently, current beamforming strategies are tailored to circumvent these obstructions. In this paper, a paradigm shift is introduced that leverages lateral wave propagation along building interfaces to extend mmWave coverage. Unlike traditional reflections, lateral waves travel along the boundary between two media of different refractive indices and decay algebraically with distance, offering a potential alternative path for mmWave connectivity. While well-established at low frequencies in natural media, the existence of lateral waves at mmWave frequencies along engineered building materials has not been demonstrated before. To this end, the first experimental characterization of mmWave lateral waves along a wall is reported. Extensive controlled measurements are employed to characterize the signal-grazing geometry and to establish a frequency and distance-dependent path-loss model for this phenomenon. The results provide the first empirical foundation for a new class of interface-guided mmWave links.
Read the full abstract
Machine Learning
A Repeated-Measurement Study for Cultural Analytics of English Song Lyrics Using Five Large Language Models
Large language models (LLMs) are increasingly used to annotate cultural texts at scales that are impractical for human coders. However, before their outputs are treated as measurements of latent social constructs, it is necessary to establish whether those measurements are reliable. This study evaluates five LLMs as zero-shot annotators of four social constructs expressed in English song lyrics: self-esteem, self-control, seeking belonging, and seeking recognition. Using repeated annotations of a large lyric corpus, we examine three properties of LLM-based measurement: consistency across repeated runs, convergence across models, and transferability of consensus labels to supervised classification. The findings show that LLM-based measurement is not uniformly reliable across constructs. Self-esteem exhibits the strongest repeated-measurement reliability across models, while seeking recognition is generally less stable; self-control and seeking belonging show intermediate but model-dependent reliability. Downstream classification further indicates that consensus LLM labels contain learnable signal, although transferability does not itself establish construct validity. Repeated-measurement stability and cross-model convergence should therefore be reported before LLM annotations are treated as scalable measurements in cultural analytics.
Read the full abstract
Atmospheric and Oceanic Physics
Exploring the role of input data on hail nowcast skill using spatiotemporal neural networks
Hail can cause large financial losses and poses risks to public safety, making reliable nowcasts essential for timely warnings. Deep-learning approaches have emerged as a strong alternative to conventional methods, but how input data choices affect performance has not been deeply explored. We investigate how the skill of a deep-learning hail nowcasting model can be improved without changing the model architecture. Sensitivity experiments assess the impact of training-data volume, random data augmentation (mirroring and rotation) and the number of input timesteps. Increasing the years of training data substantially improves forecast skill, by up to 25 minutes at later lead times. Augmentation improved performance for larger datasets but interestingly degraded performance for smaller ones. Sensitivity to input timesteps was weaker than sensitivity to training years. These findings show that improvements in data selection and preprocessing can yield substantial gains even with a fixed architecture, offering guidance for future deep-learning nowcasting development.
Read the full abstract
Optimization and Control
Multi-Principal Competition for an Exclusive Agent: characterization and approximation of Nash Equilibria
This paper introduces and studies a Principal-Agent competition model in which two risk-neutral Principals compete to hire a single Agent. The Agent chooses to work exclusively for one Principal, making the Agent's reservation utility endogenous, as it is determined by the competing offers. We characterize the Nash equilibria of the resulting game by relating them to the individual Principal-Agent contracting problems and analyzing the Agent's selection bias. We further introduce two numerical algorithms to approximate Nash equilibria by locating sign changes of the value functions of the Principals' individual problems. The first is a Newton-type method that exploits differentiability of the value functions, while the second is a bisection method that does not require differentiability. The proposed procedures are illustrated through several examples involving exponential, logarithmic, and constant relative risk aversion (CRRA) utility functions within the polynomial framework of Renner and Schmedders (2015). We also provide technical results regarding the regularity of the value function of a classical Principal-Agent problem, including differentiability under suitable conditions.
Read the full abstract
Machine Learning
Beyond a Universal Forecasting Selector: Demand-Conditioned Model Selection across Demand Patterns and Horizons
Forecasting-model selection remains difficult in heterogeneous demand because the most suitable decision rule may vary with demand structure, data availability, and forecasting horizon. This study examines whether the selector itself should be treated as a context-dependent component of the forecasting process. Five selection mechanisms - RMSSE, ERA, OWA, CCG-AHSC, and CCG-AHSCD - are compared across 24 optimized forecasting models, nine datasets, three training-testing partitions, and horizons from 1 to 12 cycles. Selector performance is evaluated ex post using Global Relative Accuracy (GRA), statistical tests, and a best-attainable-model reference. No selector dominates across all conditions. CCG-AHSC and CCG-AHSCD are more competitive for Smooth demand and several Erratic configurations, whereas OWA and ERA perform better in Intermittent and Lumpy settings. Selector suitability also changes with historical data availability and horizon, supporting a context-dependent rather than universal approach to forecasting-model selection.
Read the full abstract
Number Theory
Multiple zeta values in $\lambda$-rings
We propose a framework for multiple zeta values in $\lambda$-rings that unifies both classical multiple zeta values and multiple $q$-zeta values. As in the classical case, we show that the multiple zeta function defines an algebra homomorphism from the quasi-shuffle algebra to the $\lambda$-ring. We also prove a formula for multiple polylogarithms in $\lambda$-rings, similar to the iterated integral expression in the classical case. As an application, we explicitly describe the algebra of regular multiple $q$-zeta values, compute the corresponding Poincaré series, and exhibit a small spanning set.