Combinatorics
Papers
Fresh from arXiv, typeset properly. Highlighted words trade on the market.
All of arXiv
Sunday 8 November
We enumerate lattice points of preorder polytopes by support size for two-row Ferrers posets and recursively defined comb posets. For two-row posets, the order-ideal inequalities give a reflected-walk model in the nonnegative quadrant. Its transfer operators have a rank-one commutator, from which we derive an identity with five independent parameters. A matching-parameter specialization factors into the enumerators of a Motzkin excursion and a Motzkin bridge and proves the conjectured product of type~A and type~B Narayana polynomials for a two-row grid with its maximum removed. The general identity also records the two row supports separately. For combs, maximum ideal excess leads to a scalar recursion and a cubic generating function with independent spine and tooth weights. This proves a compositional-inverse conjecture of Athanasiadis and Chapoton and gives explicit $\gamma$-coefficients by Lagrange inversion. We prove $\gamma$-positivity for all two-row Ferrers cell posets and for the combs. The truncated-grid enumerators have simple negative zeros.
Computer Vision
Auto-GlaKG: Automated Biomarker Reasoning for Explainable Glaucoma Detection
Interpretable glaucoma detection requires a transparent connection between fundus images, clinically meaningful biomarkers, and diagnostic predictions. Knowledge graph (KG) reasoning provides this structure, but performance measured using annotated biomarkers does not establish performance when only images are available. We present Auto-GlaKG, an automated framework that combines image-based biomarker prediction with fixed clinical-prior rules. A ResNet50 network predicts the cup-to-disc ratio (CDR) and six morphological signs using biomarker supervision, and a KG reasoner aggregates their signed rule contributions into a glaucoma score. On 689 fundus images evaluated using stratified, group-aware five-fold cross-validation, Auto-GlaKG achieves a pooled area under the receiver operating characteristic curve (AUC) of 0.951 (95% confidence interval: 0.931–0.969), with sensitivity of 0.970 and specificity of 0.779. The matched image classifier also achieves an AUC of 0.951, while the rule stage adds 0.2~ms per image. Replacing predicted CDR with its annotation increases fold-mean AUC from 0.953 to 0.995. The framework provides an explicit rule-level account of image-based predictions and identifies CDR estimation as a priority for validation using independently measured biomarkers.
Read the full abstract
Dynamical Systems
Learning calibrated probabilistic forecasts using Koopman mode decomposition and the continuous ranked probability score
Well-calibrated forecasting for non-linear dynamical systems is essential in science and engineering applications such as fluid dynamics and weather prediction. We present a forecasting method combining probabilistic machine learning with the structured linear dynamics of Koopman mode decomposition. Within this framework, the observation, linear dynamics, and state reconstruction operators are constructed using standard machine learning architectures, producing predictive distributions via the pushforward of injected noise samples. The model is trained using a proper scoring rule, the continuous ranked probability score, to directly target calibrated, accurate forecasts. The approach is validated on two non-linear ordinary differential equation examples and two partial differential equation examples, vortex shedding and cavity flow, using both large and small, noisy datasets. Results demonstrate that known Koopman eigenfunctions and spectra can be identified and used for accurate forecasting. The model can also produce multi-modal predictive distributions, demonstrated for a trajectory initialised on the boundary between two basins of attraction. Uncertainty is captured by the sample ensemble spread, with spread–skill ratios close to one over the training horizon in several experiments, although for extended rollouts and noisy datasets, the ensemble may become under- or over-dispersed. In low-noise experiments, temporal rollout is generally accurate for at least twice the training horizon, and up to twenty times in some cases. Approximation quality is comparable to a deep ensemble approximation over the training horizon, but at a reduced computational cost and with better robustness in identifying Koopman eigenvalues.
Read the full abstract
Materials Science
Engineering ferroelectric nanocapacitors through interfacial phonon bias
As ferroelectric devices approach the nanometer scale, their behavior becomes increasingly sensitive to boundary conditions. In perovskite oxide nanocapacitors, epitaxial strain and electrostatic screening are well-established control parameters. Then, while expected to contribute to the epitaxial stabilization of ferroelectric ultrathin films, the lattice-dynamical state of the electrodes has not been isolated as an independent tuning knob. Here we call this poorly explored coupling interfacial phonon bias, whereby the electrodes' phonon energy landscape reshapes that of the adjacent ferroelectric across their shared interface. Using first-principles-accurate machine-learned interatomic potentials and molecular-dynamics simulations, we examine this mechanism in symmetric BiFeO${}_3$ nanocapacitors in contact with a metallic perovskite oxide. By independently controlling the orientatio and amplitude of the electrodes' octahedral-tilt landscape, while keeping conventional elastic and electric boundary conditions fixed, we show that the electrodes' phonon energy landscape can modify the stability of competing BiFeO${}_3$ phases. In particular, this allows us to shift the ferroelectric transition temperature and even induce antiferroelectric-like switching at ambient conditions. We thus showcase interfacial phonon bias as a complementary and powerful strategy to modify and optimize the properties of ultrathin epitaxial films.
Read the full abstract
Chaotic Dynamics
Chaos Growth in a Generic 2D Hamiltonian System
We examine the geometric mechanism controlling the transition from local to global chaos in a generic 2D Hamiltonian system, the generalized Hénon-Heiles system. Using Poincaré sections and invariant manifolds, we show that the chaotic fraction grows through abrupt transitions associated with the destruction of invariant barriers. We introduce an index that gives the time required for the first penetration of a local chaotic domain into the global chaos. This index increases sharply and tends to infinity as a critical energy is approached. The set of lobes generates heteroclinic connections, while multiple unstable periodic orbits exhibit a global manifold structure characterised by repeated loops and lobes. However, in special chaotic cases close to the integrable Hénon-Heiles potential, where two asymptotic curves extend to infinity, the manifold topology differs qualitatively from the generic case, as the asymptotic curves no longer form closed structures surrounding the periodic orbits.
Read the full abstract
High Energy Physics (Phenomenology)
Probing the Supersymmetric Grand Unified Theories with Anomaly Mediation at the Future Proton-Proton Colliders and Hyper-Kamiokande Experiment
With the grand desert hypothesis, we investigate the Supersymmetric (SUSY) Grand Unified Theories (GUTs) with the minimal anomaly mediated supersymmetry breaking (AMSB) as a continuation of our program of studying SUSY GUTs at the future 100 TeV proton-proton Collider such as $\rm FCC_{hh}$ and SppC as well as proton-decay experiment such as Hyper-Kamiokande experiment. We show that for the GUT scale $M_{GUT} \ge 0.84 \times 10^{16}$ GeV, the upper mass bounds on the gluino and first two generation squarks are 8 TeV and 6 TeV, respectively. Therefore, the gluino and first two generation squarks can be probed at the $\rm FCC_{hh}$ and SppC. In addition, we discuss the mass spectra for the colored sparticles, sleptons, pseudo-scalar Higgs and electroweakinos. We also shed light on the possible search strategies of gluino and winos at the future 100 TeV colliders. Moreover, for $\rm M_{\rm GUT} \le 0.84 \times 10^{16}$ GeV, the proton lifetime for dimension-six proton decay via heavy gauge boson exchange is less than or around $5 \times 10^{34}$ years, and thus we can probe it at the Hyper-Kamiokande experiment. To be concrete, we present a couple of benchmark points representing the above two mentioned GUT conditions on $\rm M_{\rm GUT}$.
Read the full abstract
Materials Science
Effect of Correlation on the Pressure-Induced Insulator to Metal Transition in Complex Spin-Orbit Coupled Double Perovskite La$_2$CoIr$_{1-x}$Ti$_x$O$_6$
We study the effect of electron correlation ($U$) and hydrostatic pressure on the electronic properties of double perovskites La$_2$CoIr$_{1-x}$Ti$_x$O$_6$ for $x$ = 0, 0.5, and 1, using first-principles electronic structure calculations within density functional theory (DFT). At ambient pressure, we systematically vary the electron correlation, i.e., on-site Coulomb interaction to study its effect on band gap and structural parameters. Increasing the value of $U$ enhances the band gap of all three compositions while producing only a slight variation in structural parameters. The effect of hydrostatic pressure is also studied for the same compositions. The estimated band gap values at ambient pressure closes at approximately 39, 42, and 53 GPa for $x$ = 0 (for $U_\mathrm{Co}$ = 5 eV and $U_\mathrm{Ir}$ = 2 eV), 0.5 (for $U_\mathrm{Co}$ = 5 eV and $U_\mathrm{Ir}$ = 2 eV), and 1 (for $U_\mathrm{Co}$ = 4 eV), respectively, indicating pressure-induced insulator-to-metal transitions. These results show that electron correlation stabilises the insulating state, whereas hydrostatic pressure favours the metallic state by closing the band gap. These findings highlight the tunability of correlated oxides through electron correlations and pressure, offering insights for materials design.
Read the full abstract
Machine Learning
ALETHEIA: A Multi-Frequency Eddy Current Pulsed Thermography Dataset for Neural Operator Learning in Nondestructive Testing
Learning neural solvers for spatiotemporal partial differential equations (PDEs) under real-world constraints remains a key challenge in scientific machine learning, especially for inverse tasks with sparse and noisy boundary observations. We present the Aletheia dataset, the first 3D benchmark for learning data-driven solvers in the context of nondestructive testing (NDT). The dataset simulates eddy-current-induced heating in conductive solids and models the resulting transient heat propagation governed by the heat equation. Aletheia contains over 4,700 high-resolution samples across 10 excitation frequencies (1-100,kHz), each providing volumetric heat source and temperature fields over time. It supports both forward prediction of temperature evolution and inverse reconstruction of internal heat sources or defects from surface infrared measurements. Real infrared thermography data from cracked rail specimens are included for calibration and comparison. We define four canonical tasks on both regular and irregular grids and benchmark them using various neural operators. Aletheia establishes a unified platform for evaluating neural PDE solvers under realistic NDT conditions, enabling progress in reliable, data-driven inverse modeling. The dataset and code are publicly available at https://github.com/sherlock-0327/ALETHEIA.
Read the full abstract
Soft Condensed Matter
The Outer Eshelby Problem in Amorphous Solids with Odd Dipole Screening
The so-called "2-dimensional Eshelby problem" was defined by JD Eshelby, where a circle is cut out of an elastic sheet, deformed into an ellipse, and forced back into the material. This creates a quadrupolar response which turned out to be very relevant for describing plastic deformations in amorphous solids. Moreover, the resulting strain field has been used to describe how elasto-plastic models respond to external strains. It is thus important to examine carefully the Eshelby problem in amorphous solids, especially when anomalous elasticity is at play. Here we solve the Eshelby problem in amorphous solids with both even and odd dipole screening and a complete analytical result is obtained for the displacement and strain fields. The analytical results are crucial for understanding the anomalous responses in displacement fields measured directly from experiments and numerical simulations. The computation of strain fields clearly demonstrates the deviations from their classical counterparts for perfectly elastic solids. The decomposition of the displacement field shows the excitation of higher-order Fourier modes. The presence of higher excitations other than the imposed Eshelby geometry therefore suggests the possible nonlinear mode coupling that can influence and modify the anomalous responses.
Read the full abstract
Spectral Theory
Rayleigh's conjecture in dimensions four and five
We prove Rayleigh's conjecture for the clamped plate in dimensions four and five. In fact, we obtain a sharp comparison with the first Dirichlet eigenvalue: on every open set of finite positive measure, the first plate eigenvalue is at least the square of the Dirichlet eigenvalue multiplied by the ratio attained on a ball. The proof uses a homogeneous transformation of a trial function and its gradient. Two divergence identities reduce the comparison to positivity of a quadratic form in the Hessian. We establish the required Bessel inequalities by rational estimates for entire profiles and their remainders. An exact defect identity accounts for regular nodal sets, so the argument allows trial functions of either sign and imposes no boundary regularity. The domains may be disconnected or unbounded. We characterize equality up to sets of zero Sobolev capacity. We also obtain quantitative stability with the optimal quadratic power of the Fraenkel asymmetry.
Read the full abstract
Computation and Language
Understanding When and Where Self-Supervised Speech Models Learn Phonemes and Prosodic Cues through Layer-Wise and Time-Resolved Probing
Self-supervised speech models (S3Ms) encode a rich hierarchy of acoustic, prosodic, and phonetic information, yet when and where it emerges during pretraining remains unclear. We train frozen-encoder probes on every Transformer layer across pretraining checkpoints of two matched models: HuBERT-Base, whose stage 2 is a separately initialized model trained on new pseudo-label targets, and single-objective wav2vec 2.0-Base. Our five tasks span an acoustic-to-abstract gradient: absolute pitch (F0), speaker-relative pitch, lexical stress, phoneme boundaries, and 38-way phoneme identity. Acoustic-prosodic cues and phone boundaries are decodable from the earliest checkpoints and barely shift across layers, whereas phoneme identity is constructed during pretraining, re-localizing to upper-middle layers under HuBERT's stage-2 targets at a matched update count. Acquisition order therefore tracks abstraction from the acoustic surface, not the segmental/suprasegmental distinction, extending to prosody an account established for phonetic, lexical and syntactic structure; the no-switch wav2vec 2.0 shows no comparable re-localization.
Read the full abstract
Atmospheric and Oceanic Physics
FESOM2.8 in single precision: climate fidelity over six decades and performance on CPUs and GPUs
Halving the floating-point precision of an ocean model can make it substantially faster, but whether the rounding errors stay negligible over climate time scales has rarely been tested. We implement single precision in the global ocean-sea-ice model FESOM2 and compare 60-year hindcasts with double-precision controls. A first implementation lost salt steadily, accumulated a heat-content deficit of 129 ZJ, and shifted major ocean fronts. Controlled experiments trace the drift to the tracer update: small increments added to a large salinity value are rounded away, and the increments themselves are biased. Accumulating tracer tendencies and layer thicknesses in 64 bits, and carrying the rounding error of each tracer update into the next, removes the drift for at most 8% additional runtime. Single- and double-precision simulations then differ in heat content and mean state about as much as two double-precision runs with slightly perturbed initial conditions, sea ice differs by at most 0.2%, and he salinity difference is four orders of magnitude smaller than before. In the coupled model AWI-ESM3, single precision adds no detectable heat source or sink over 60 years. Single precision speeds up the 60-year hindcasts by a factor of 1.35, and the model by up to 1.66 on meshes of up to 5 km resolution; on GPUs, a C++/Kokkos version runs up to 1.27 times faster and needs 43% less memory. Whether reduced precision is safe for climate integrations is decided by long-term budgets, not by short tests.
Read the full abstract
Geometric Topology
On the monotonicity of amenable category for geometric 4-manifolds and their connected sums
High Energy Physics (Phenomenology)
Direct Light-Cone Reconstruction of the $B$-Meson Shape Function in HQET
The $B$-meson shape function encodes the nonperturbative residual-momentum dynamics of the heavy quark within heavy quark effective theory (HQET). From a Wilsonian perspective, HQET provides a natural framework for lattice calculations when the heavy quark mass is comparable to or exceeds the inverse lattice spacing. Motivated by this observation, we introduce a $B$-meson quasi-shape function directly in HQET and calculate its one-loop matching to the light-cone shape function, avoiding an intermediate matching step involving a full-QCD light-cone distribution. We determine the one-loop renormalization kernels and matching coefficients in both the $\overline{\mathrm{MS}}$ and regularization-independent momentum-subtraction (RI/MOM) schemes for the quasi-shape function, while keeping the light-cone shape function in the $\overline{\mathrm{MS}}$ scheme. The matching coefficients are obtained at leading power in the large-$v^z$ expansion and are independent of the infrared regulator. Our results provide the perturbative ingredients for a direct lattice HQET determination of the $B$-meson shape function.
Read the full abstract
Quantum Physics
Hardware-Efficient Continuous-Variable Quantum Computation: Finite-$N$ Non-Gaussian Gates in a Floquet-Engineered Cavity-BEC
Continuous-variable (CV) quantum computation in bosonic modes offers a hardware-efficient route toward fault tolerance, yet synthesizing deterministic, macroscopic non-Gaussian gates with Wigner negativity remains challenging. Here, we establish two physical breakthroughs and two tensor-network innovations using a single-cavity 87Rb Bose-Einstein condensate (BEC) as a macroscopic bosonic logical mode. Physically, building on our orthogonal optical dual-driving architecture, we propose deterministic, unitary Floquet-engineered Two-Axis Twisting (TAT) non-Gaussian gates without post-selection, and introduce a paradigm shift from quantum metrology to CV quantum computation: via a normal-ordered Holstein-Primakoff expansion and non-perturbative spin simulations, we show that the over-squeezed regime ($t>t_{opt}$, where Wineland squeezing $\xi_R^2$ degrades) harnesses finite-$N$ Bloch-sphere curvature - manifesting at $O(1/S)$ as cubic shearing and quartic Kerr $(a^\dagger a)^2$ interactions - to deterministically generate Wigner negativity. Methodologically, we formulate an exact automaton Matrix Product Operator in the compact SU(2) spin basis with an $N$-independent Liouvillian bond dimension ($D_L=6$), bypassing Fock cutoff errors ($N_{max}$) and $O(N^3)$ permutational-invariant scaling. Finally, we decouple $4^N$-dimensional Lindblad evolution from collective observation via an Exact Bilateral Dynamic Programming Dicke Projector with log-gamma normalization, extracting the $(N+1)\times(N+1)$ symmetric projection $\tilde{\rho}_{n_L,n_R}$ without moment truncation while tracking symmetry retention $p_{sym}(t)$. Our proof-of-principle $N=50$ benchmark resolves deep Wigner negativity ($W_{min} \approx -0.214$ at $t=2.00\mu$s, after $\xi_R^2 \approx 0.124$ at $t_{opt} \approx 0.64\mu$s), establishing a scalable architecture for $N \sim 10^3-10^5$ macroscopic BEC processors.
Read the full abstract
Probability
Harmonic and Monogenic Representations of White Noise and its weighted harmonic Bergman estimates
Continuing Feichtinger's generalized stochastic process theory, we analyze harmonic and monogenic representations of white noise in Euclidean spaces. We define its convolution with square-integrable functions. Using the weighted Sobolev regularity of the white noise, we establish a high-dimensional Sokhotski-Plemelj formula for the white noise density. Furthermore, we embed these representations into new weighted Bergman spaces, proving that the weighted Bergman norms of the harmonic representations are bounded by the white noise's weighted Sobolev norms.
Read the full abstract
Computation and Language
Saidzoda-Tajik-EmbeddingGemma-300M: Knowing Every Letter Is Not Reading Tajik
EmbeddingGemma-300M is a small open embedder whose tokenizer represents every Tajik letter, and as released it is weak at Tajik retrieval: on our Tajik law test the right passage is in its top ten results for 49.3% of questions, on tax documents for 45.9%, and on the human-written Belebele Tajik set for 69.8%, against 98.6% on the Russian version of the same set. We present Saidzoda-Tajik-EmbeddingGemma-300M, the same model fully fine-tuned for one epoch on 235,140 question and passage pairs written by our own Tajik language model for laws, tax documents, Wikipedia and web text, with the same mined hard negatives, English and Russian replay, loss, global batch, learning rate and epoch as our earlier Saidzoda-Tajik-E5-Large (arXiv:2611.00592) and BGE-M3 models. After training, law reaches 92.4% and tax 91.7%; on Belebele, the one test written by people, recall@10 rises from 69.8% to 92.6% and the first result is right for 76.6% of questions instead of 51.7%. A checkpoint at step 100 of 466 already holds 88% of the law gain at recall@10. At recall@10 the result is no more than 2.8 points below the two 560M models trained on the same pairs on law, tax and Wikipedia, within the intervals, with 768-dimensional vectors and about half their parameters; at recall@1, for which we have no intervals, it stays 7.0 points behind the E5 model on Belebele. The cost is a drop of 3.0 and 2.8 points in Russian and English recall@1, larger than for the two earlier models. We also report that, according to our run notes, a bfloat16 test run produced a non-finite loss at step 3 which the training framework logged as zero, so training was done in fp32. Our generated test questions come from the same model as the training questions, so we treat those scores as an upper limit until questions written by Tajik speakers exist.
Read the full abstract
Robotics
DeepSoMo: Deformable Object Motion Estimation under Proprioceptive Partial Observability via Predictive Latent Dynamics
Estimating the motion of deformable bodies in soft robotics from proprioceptive sensing is challenging due to partial observability, as internal observations provide only indirect access to global motion, contact, and shape evolution. In such settings, instantaneous observations are insufficient, and accurate estimation requires temporally consistent latent representations. This paper studies deformable motion estimation under proprioceptive partial observability from a latent dynamics perspective. We argue that a key requirement for long-horizon estimation is latent temporal consistency, where latent states remain predictive of future observations. To this end, we present DeepSoMo, a learning-based framework that enforces predictive structure in latent dynamics through a self-supervised latent prediction constraint. We evaluate the approach in a canonical deformable dynamics setting using high-fidelity finite-element simulation as a controlled source of ground truth. Results demonstrate consistent improvements in trajectory estimation, orientation recovery, shape reconstruction, and contact inference over architecture-matched recurrent baselines.
Read the full abstract
High Energy Physics (Phenomenology)
An $SU(5)$ GUT with non-holomorphic modular $S'_4$ symmetry: fermion masses, mixing and leptogenesis
A non-supersymmetric $SU(5)$ grand unified model is constructed based on the non-holomorphic modular $S'_4$ flavor symmetry, supplemented by a global Peccei–Quinn $U(1)_{PQ}$ symmetry that constrains the Yukawa couplings. Constructed from polyharmonic Maaß forms, these couplings allow the model to reproduce the observed charged-fermion masses and mixing as well as the neutrino oscillation parameters for both normal ordering (NO) and inverted ordering (IO), with all Pull values satisfying $|\mathcal{P}| < 0.8$. The model further yields specific predictions for yet-unmeasured parameters, including the absolute neutrino masses, the Dirac CP phase, the Majorana phases, and the effective Majorana mass $m_{\beta\beta}$. We then study thermal leptogenesis by solving the density matrix equations. For a given benchmark point, the final baryon asymmetry depends solely on the single free parameter $b$ that sets the right-handed neutrino mass scale, and an appropriate choice of this parameter reproduces the observed value $\eta_B \simeq 6.12 \times 10^{-10}$. Finally, the correlation between this scale parameter and specific low-energy observables provides a potential bridge between the fermion flavor structure and early-Universe cosmology.
Read the full abstract
Quantum Algebra
Stretched near-groups
We study stretched near-group fusion categories over $\mathbb{C}$ of type $G+k|G|$ for abelian groups $G$ and positive integers $k$, describing the structure of the double including fusion rules, modular data, and Frobenius–Schur indicators under a variety of assumptions. For cyclic $G$, when enough invertible objects exist in the double, we describe all associated de-equivariantizations which connect the existence of stretched near-group fusion categories to the existence of other stretched quadratic categories. All of our results apply to the case $k=1$. In particular, the known existence of $C_{m^2}+m^2$ near-group categories for all positive integers $m$ implies the existence of Haagerup–Izumi fusion categories for all cyclic groups. As a further application, we provide a general conductor bound severely restricting the existence of stretched near-group categories, and for infinitely many abelian groups $G$ of square order, among them $C_4\times C_2^{2j}$ with $j\geq2$, we show that a near-group fusion category $G+k|G|$ exists if and only if $k=0$.
Read the full abstract
Computation and Language
A Layer-wise Study of How Style Rewriting Breaks the Calibration of Zero-Shot LLM Fake-News Detectors
Large language models (LLMs) are increasingly asked to judge whether a news article is true, and their stated confidence is often taken at face value. We test what happens to that confidence when an article's writing style, but not its facts, is manipulated, as in recent style attacks on fake-news detectors. Tracing how the prediction and its confidence form layer by layer inside the network, we study three model families (LLaMA-3-8B, Qwen2.5-7B, Mistral-7B) on three fake-news benchmarks, and find the same three patterns in every model–dataset pairing. First, on the original articles the models are competent and largely honest judges: they settle on an answer, briefly become overconfident, and then correct their confidence in the final layers. Second, style-rewritten articles break this self-correction: the models become far less accurate, under the strongest attack, worse than random guessing, yet remain just as confident, so they are confidently wrong precisely when a warning is most needed. Third, nearly all of the damage comes from rewriting into sensational or emotionally charged styles, while sober, neutral rewrites leave the models' honesty largely intact. The practical implication is that an LLM's confidence about veracity cannot be trusted as a safety signal when an article's style may have been manipulated.
Read the full abstract
Machine Learning
FRESCO: A Novel Consistency Control for Asynchronous Pipeline Parallel Training
Asynchronous pipeline parallelism can accelerate distributed model training, but suffers from instability caused by inconsistent gradient computations. We identify a fundamental source of this inconsistency: error signals are coupled to forward-time features, not just parameter states. This means that parameters may evolve as long as they preserve forward-equivalence. Leveraging this insight, we introduce FRESCO, which frames asynchronous updates as a constrained optimization problem: it finds the minimal update modification that acts trivially on the activation subspace. We demonstrate that FRESCO consistently outperforms representative asynchronous baselines in LLM pre-training, with the gains most pronounced under deep pipelines and high inconsistency levels: at pipeline depth 32, it reaches the same loss as the strongest baseline with 43% fewer iterations and 30% less wall-clock time, while requiring 8~921$\times$ less memory overhead. Moving beyond the current limitations, FRESCO establishes a new foundation for high-utilization pipeline parallel training, providing a scalable path for large-model development on diverse, resource-constrained systems.
Read the full abstract
High Energy Physics (Phenomenology)
Precision Fragmentation Functions from Modern COMPASS SIDIS Data
We present a comparative account of three global determinations of collinear fragmentation functions (FFs) for charged pions, charged kaons, and unidentified charged hadrons. The HAPS-PiFF1.0, HAPS-KaFF1.0, and HAPS-hFF1.0 analyses combine single-inclusive $e^+e^-$ annihilation with charge-separated semi-inclusive deep-inelastic scattering (SIDIS), including the 2025 COMPASS proton-target measurements and the revised 2026 isoscalar-target results. Flexible neural-network parametrizations, Monte Carlo replicas, and NLO/NNLO perturbative setups are employed. The proton and isoscalar samples sharpen the flavor sensitivity of SIDIS, while gluon fragmentation remains comparatively weakly constrained. We discuss the effects of perturbative accuracy, the principal limitations, and prospects for future measurements.
Read the full abstract
Data Structures and Algorithms
Algorithms for Soft Tree Containment on Tree-Child Networks
Phylogenetic networks generalize evolutionary trees by allowing lineages to merge. The Tree Containment problem asks whether a network $N$ displays a tree $T$ on the same taxa. When the polytomies of $T$ reflect uncertainty rather than simultaneous divergence, the relevant notion is soft display, in which $N$ may refine the polytomies of $T$. We take it in its asymmetric reading. Soft Tree Containment is NP-hard on reticulation-visible networks even at reticulation indegree three, and it was open for a non-binary tree and a tree-child network with multifurcating tree nodes or reticulations of indegree above two. We solve this case. A cluster characterization reduces soft display to realizing all clusters of $T$ by one switching, and in a tree-child network the vertices that can realize a cluster lie on a single chain. The per-cluster requirements then form a constraint network of width one, which arc consistency decides without search, and a $2$-SAT instance when $N$ is binary. This yields, to our knowledge, the first polynomial-time algorithm for soft Tree Containment on single-labelled tree-child networks of arbitrary reticulation indegree, and a linear-time algorithm when $T$ is binary. The same method decides the symmetric variant of the problem, in which the polytomies of $N$ may be resolved as well.
Read the full abstract
Analysis of PDEs
Continuity of solutions to nonlocal nonlinear diffusion equations
We establish an explicit modulus of continuity for locally bounded weak solutions to nonlocal porous medium type equations with measurable kernels. The nonlocal tail is only assumed to be integrable in time. The result lays a solid foundation for the intrinsic Harnack estimates of these equations.
Read the full abstract
Artificial Intelligence
LoMASA: A Local Multi-Agent Sales Assistant with Confidence-Gated Delegation for B2B Quoting
Frontline sales staff spend much of their day re-coordinating quotes after customers change quantities, ask for discounts or new payment terms, or costs move. We present LoMASA (Local Multi-Agent Sales Assistant), a training-free multi-agent system that runs entirely on one desktop computer. A 35B mixture-of-experts model (System-1) makes each decision in a single forward pass and reads its confidence from the probabilities of the answer options. A 27B dense model (System-2) writes customer replies and escalation briefs and is consulted when System-1 is unsure. Code computes every price, margin and option, and hard pricing rules cannot be overridden. On a test split of 500 simulated episodes with 1512 decisions built from public supply-chain data, the salesperson has to decide only 19.5% of decisions: 38.0% are settled automatically, with a median latency of 2.67 seconds, and 42.5% are escalated to a manager by the hard rules. On the automatically settled decisions LoMASA is as accurate as a rule baseline (0.894 vs.\ 0.894) and makes fewer commitments whose margin later falls below the floor (2.61% vs 4.53%); the price is additional escalations to the manager. Option probabilities are well calibrated without any training, unlike confidence stated in words by the same model. We also report where the design does not help: the 27B model rarely changes a decision.
Read the full abstract
Medical Physics
Joint Poisson Reconstruction for Prompt-Gamma-Enabled 68Ga/18F Dual-Isotope Quantitative PET Imaging
Objective:To enable simultaneous quantitative 68Ga/18F PET imaging despite low prompt-gamma triple-coincidence counts and random-triple contamination. Methods: We developed an expectation-maximization reconstruction method (PNR-EM) that jointly estimates pure-emitter, non-pure-emitter, and random-triple components using a Poisson model of prompt double, prompt triple, and delayed triple measurements. A consistent-delay strategy estimates random triples while preserving true-pair temporal correlations. Prompt doubles additionally constrain the non-pure component, and total variation regularization suppresses reconstruction noise. The method was evaluated with quantitative and Derenzo phantoms, a nine-well separation phantom, and tumor-bearing mice. Results: Phantom quantitation showed strong linearity (r2>0.99) and a maximum relative deviation below 9.9% relative to single-isotope references. The separated images largely preserved rod resolvability relative to their corresponding single-isotope references. The best PSNR, SSIM, and RMSE were 33.72 dB, 0.975, and 0.0206 for 18F, and 34.86 dB, 0.955, and 0.0181 for 68Ga in the 9-well phantom. The 68Ga RMSE was 41.4% lower than that of the best image-domain separation baseline. Tumor-bearing mouse imaging demonstrated in vivo separation of 68Ga-DOTA-NOC and 18F-FDG distributions. Conclusion: Joint estimation improved dual-isotope separation under the evaluated low-count conditions. Significance: To our knowledge, this study provides the first experimental demonstration of simultaneous 68Ga/18F PET separation using prompt-gamma coincidences, supporting simultaneous imaging of complementary metabolic and receptor-targeted tracer distributions.
Read the full abstract
Machine Learning
TLRS-DPO: Token-Level Risk Scoring for Better Direct Preference Optimization
Direct Preference Optimization (DPO) is a simple and widely used alternative to RLHF for aligning large language models (LLMs), but its sequence-level objective cannot precisely locate or suppress the local patterns that make an otherwise acceptable response unsafe or undesirable. We propose TLRS-DPO (Token-Level Risk Scoring for Better Direct Preference Optimization), which augments DPO with fine-grained token-level control while remaining fully offline and reference-based. TLRS-DPO introduces a token-level robust compensation term that converts the policy-reference log-ratio on rejected responses into localized risk scores, a curriculum-style schedule that gradually increases the strength of this compensation during training, and a risk-adaptive screening mechanism that selectively penalizes only high-risk tokens instead of entire sequences. This design bridges sequence-level preference fitting and token-level behavior control, and can be implemented as a drop-in modification to standard DPO trainers. Empirically, TLRS-DPO improves alignment quality and safety-relevant behavior with minimal impact on helpfulness compared to strong DPO baselines.
Read the full abstract
Signal Processing
Nonlinear Interference Estimation in Coherent Optical Systems: Principles and Applications
Accurate estimation of nonlinear interference (NLI) is important for monitoring and optimizing coherent optical networks. This invited paper reviews the main techniques proposed for receiver-side NLI estimation, with particular focus on longitudinal power monitoring (LPM). We discuss the principles, advantages and limitations of LPM-based NLI estimation, including recent experimental demonstrations. We then experimentally investigate the impact of the spatial step size and the number of averaged constellation blocks on estimation accuracy. The results show that accurate NLI estimation does not necessarily require a highly accurate power profile, with limited improvements obtained by increasing the number of averaged blocks. Finally, we discuss the main practical limitations and applications of NLI estimation, with particular attention to end-to-end and span-wise optical power optimization.
Read the full abstract
Analysis of PDEs
An elementary observation on the Robin problem for NLS
We present an infinite set of explicit solutions of the integrable focusing NLS equation on the half-line that are rapidly decaying and satisfy a homogeneous Robin condition but have very different behaviors on the wall $x=0$. We then argue that a similar phenomenon exists for a small non-homogeneous perturbation of the Robin problem. This renders a potential application of the unified transform method to non-linearizable Robin problems impossible.