Image and Video Processing
Papers
Fresh from arXiv, typeset properly. Highlighted words trade on the market.
All of arXiv
Friday 6 November
Plug-and-play (PnP) methods reconstruct images by alternating a data-fitting step with an image denoiser. Their performance depends on two parameters: the step size $\gamma$ and the measurement noise level $\sigma$. In practice, $\gamma$ is often tuned by grid search over validation images and must be retuned whenever the problem setting changes, and $\sigma$ is assumed known or estimated separately. We propose AdaptivePnP, a PnP framework that jointly estimates $\sigma$ and $\gamma$ at every iteration from the data-fitting residual. To do so, we apply the PnP virtual model to the data-fitting proximal map and show that the residual is zero-mean Gaussian with covariance parameterized by $\gamma$ and $\sigma$. However, when one parameter is much larger than the other, the residual carries little information about the smaller one and using the maximum likelihood (ML) estimator can degrade the PnP iterations. We therefore introduce the identifiable maximum likelihood (IML) estimator, which restricts the ML estimator to an identifiable window in which the normalized Cramér–Rao bound of every parameter is bounded by a user-specified tolerance. For the virtual model, this window is an interval of ratios $\rho = \gamma/\sigma$, so computing the IML estimate reduces to a one-dimensional search over $\rho$. We apply AdaptivePnP to two standard PnP methods, yielding algorithms that require neither a tuned $\gamma$ nor a known $\sigma$. On image deblurring and super-resolution, AdaptivePnP matches or outperforms PnP baselines tuned with the true $\sigma$. Code is available at https://github.com/hdubasullivan/AdaptivePnP.
Combinatorics
Local fluctuations, majorization, and nontransitivity in cube tilings of $\mathbb{Z}^d$
Let $\mathcal{P}$ be a partition of $\mathbb{Z}^d$ into translates of the discrete unit cube $\{0,1\}^d$. For a uniformly chosen $\varepsilon\in\{0,1\}^d$, we count the cubes of $\mathcal{P}$ entirely contained in the observation window $[0,3]^d+\varepsilon$. The resulting score has mean $(3/2)^d$ for every partition, while its distribution reflects the geometry of the tiling. Comparing two such scores, with the shifts chosen independently, defines a game in which the larger score wins. In dimension $d=4$, we construct a geometric realization of the Steinhaus–Trybuła paradox: three periodic partitions form a cycle in which each beats the next with probability strictly greater than $1/2$. We describe local configurations by finite cube tiling codes and prove that every resulting list of counts is majorized by that of the simple code $C_d=\{0,2\}^d$. Nevertheless, every non-simple code differing from $C_d$ in only two coordinate directions beats it, with a sharp lower bound on the winning margin. We also show that the ratio of wins to losses of $C_d$ has limit superior at most $1$ as the dimension tends to infinity, uniformly over opposing codes. For $4\mathbb{Z}^d$-periodic tilings, changing the window size can repeatedly reverse the winner, but the winning margin eventually stabilizes separately for side lengths $4m+1$ and $4m+3$.
Read the full abstract
Machine Learning
Fifty-Two Shades of Pig: What the UCR Archive Really Asks Classifiers to Do
The pigs were bleeding. That was the point of the experiment: could airway pressure reveal hemorrhage before it became critical? In the UCR Time Series Classification Archive, the de facto benchmark for Time Series Classification (TSC), the same recordings ask a different question: "which of 52 pigs produced this trace?". The dataset PigAirwayPressure is not an outlier. Although the archive is the standard benchmark, whether its tasks resemble the problems TSC is meant to solve has rarely been examined. We find that several datasets are repurposed from their original task, built from toy setups, or have flawed test splits. We further question how representative the archive's ordered non-temporal sequences (e.g. spectra, shape outlines) are of real TSC, and the potential downsides from the decomposition of multivariate problems. Our findings call for more careful interpretation of archive-wide scores and greater attention to deployment-relevant evaluation. The audit is at https://github.com/automl/ucr-provenance-audit.
Read the full abstract
Information Theory
Asymptotic Analysis of Data Deduplication Under a Constant Edit Model
Data deduplication has become increasingly important in large-scale storage systems, driven by the rapid growth of digital data. By identifying and removing repeats in the data stream, deduplication saves storage space more efficiently than traditional compression methods. In this paper, we study the performance of variable-length deduplication and multi-chunk deduplication algorithms from an information-theoretic perspective, focusing on data streams where repeats are approximate rather than exact. We consider two probabilistic source models in which source strings consist of repeated blocks that are modified by either a constant number of substitution edits or burst-insertion edits per block, the latter capturing the localized edit patterns commonly observed in real-world data. For both models, we show that variable-length deduplication achieves asymptotically arbitrarily large compression ratios, and that multi-chunk deduplication achieves compression within a constant factor of the source entropy under mild conditions.
Read the full abstract
Adaptation and Self-Organizing Systems
Discrete time crystals and heteroclinic switching in actively mode-locked semiconductor lasers
We investigate the mechanisms underlying the emergence of discrete time crystals in actively mode-locked semiconductor lasers. Our analysis is based on a Haus master equation generalized to retain carrier memory between round-trips. We perform a detailed bifurcation analysis that reveals a well-defined hierarchy of subharmonic states, illustrating how pulses may suppress their neighbors when the modulation runs faster than the pulse round-trip frequency. To identify the underlying mechanism, we derive effective equations of motion that couple the pulse energies, their positions relative to the modulation transmission maxima, and the local time-dependent gain. These reveal that transitions between time-crystalline states occur via transcritical global bifurcations, opening a regime of heteroclinic winnerless competition between symmetry-degenerate configurations.
Read the full abstract
Human-Computer Interaction
GroundWork: Bridging Human and LLM Reasoning Through a Structured Argumentation Model
Large language models (LLMs) are increasingly integrated into workflows that require reasoning over text documents, yet interaction remains largely conversational. This leaves argumentative structure implicit or embedded in text, making it difficult to maintain, compare, and revise arguments as analyses evolve. We present GroundWork, a mixed-initiative system that makes argument structure a shared, editable state for humans and LLMs. GroundWork operationalizes the Toulmin model of argumentation as a typed graph with explicit operations for human and LLM edits. It exposes this structure through coordinated corpus views and an editable workbench for constructing, comparing, testing, and tracing source-grounded arguments. We evaluate GroundWork through an expert study with four participants spanning consulting, product development, law, and immunology, and three use-case scenarios in intelligence, law, and medicine. Our findings show that GroundWork makes source-grounded arguments inspectable, comparable, and revisable as shared human-LLM state.
Read the full abstract
Quantum Physics
Gibbsian Entropy for General Wigner Distributions
We propose a definition of non-equilibrium Gibbsian entropy applicable to general quantum systems, computable from Wigner functions, for both Gaussian and non-Gaussian states. We show that while the Wigner function is a quasi-probability distribution, fundamental results linking observables in repeated observations to the function, such as the law of large numbers, the central limit theorem and, crucially, the relation between the entropy and the cumulant generating function continue to apply. This ideas are the core of the large deviation theory, which allow to formulate a positive-definite definition of entropy based on our ignorance of the underlying state structure after repeated measurements of cumulants. This new definition is a parallel to the Gibbs entropy formulation, but it can also be used in out-of-equilibrium regimes. Some important properties of this entropy are discussed. In addition we derive some information theory quantities such as relative entropy, conditional entropy, mutual information and Rényi entropy. We then compare the results with Wehrl entropy for the 1D harmonic oscillator, for pure and mixed states. We conclude the the work discussing potential applications of this new formalism, ranging from quantum correlations beyond entanglement to thermalization in closed quantum systems.
Read the full abstract
Optimization and Control
Instantaneous control for stabilization of interacting particle systems: convergence rates and dominant frequencies
We study a class of Cucker-Smale-type systems with a binary interaction potential subject to an external control input. We apply an instantaneous rolling horizon control (IRHC) feedback to address two different control problems: full set-point stabilization of a spatial configuration with zero velocity and stabilization of the center of mass at a desired position with zero mean velocity using a single external agent. The IRHC feedback has been shown to preserve the port-Hamiltonian structure of the corresponding open-loop systems. We analyze how the control parameters and the time discretization affect convergence rates and oscillation frequencies and compare the IRHC feedback with the corresponding IDA-PBC feedback obtained by omitting an additional correction term. For the reduced center of mass dynamics, we derive explicit predictions for continuous- and discrete-time spectral decay rates and oscillation frequencies and verify these predictions numerically. For the full nonlinear particle system, the controller comparison is done by numerical experiments.
Read the full abstract
High Energy Physics (Theory)
99.9...% of the SU(2) BFSS ground state
We solve for the ground-state wavefunction of the SU(2) BFSS matrix model by applying the Rayleigh-Ritz method to an auxiliary operator $H_{\mathrm{aux}}$. We show that this operator agrees with the Hamiltonian on the ground-state subspace and has a spectral gap. For SU(2) specifically, the representation theory of the singlet sector gives an astronomical reduction in the number of degrees of freedom. This allows us to construct an efficient variational basis whose inner products we compute exactly. We demonstrate the method by computing a state that satisfies rigorous bounds on its energy $E\,g_{\mathrm{YM}}^{-2/3}\leq 4.18\times10^{-8}$ and ground-subspace overlap $\|P_{\ker H}\psi\|^2\geq 0.999997$. We compute ground-state moments with high precision, finding values consistent with the bounds from our matrix bootstrap. Additionally, the state accurately reproduces both the exponent and coefficient of the asymptotic radial probability density predicted by the three-graviton amplitude.
Read the full abstract
Artificial Intelligence
Can Agentic Systems Close the Gap in Radiology with Production AI Tools?
Radiology AI has developed along two paths: specialized models that perform well on narrow tasks but cover few clinical findings, and generalist models that address a broader range of tasks but remain prone to hallucination and inaccurate measurements. Combining the two through agentic systems offers a route towards comprehensive radiology reporting, but published systems are tied to one modality or body part. We describe a radiology agentic framework that standardizes how an orchestrator calls modular tools to produce a structured report, and use it to measure what such a system can do with the tools from FDA-cleared radiology AI devices. Report generation is simulated on a large multi-modality report corpus, with the orchestrator given either a catalog of finding classes consolidated from the cleared devices, each available on the modalities it is cleared for, or an unlimited catalog in which any class can be requested, and planning either from the clinical indication or from the target report. Tool outputs are simulated from the reference report, so the comparison isolates tool availability and orchestration from tool error. With the cleared catalog and the indication alone, the orchestrator recovers under a third of the clinically actionable findings; recovery improves substantially with the unlimited catalog, and further when the plan is written from the target report. Attributing every missed finding to a cause shows that most are classes no cleared device offers for the study's modality, and that a third are classes or tools that were offered and not called. Adding the most frequently missing tools closes most of the gap to the unlimited catalog on held-out reports, whereas a smaller orchestrator with the unlimited catalog gives up the whole gain. Tool availability and effective orchestration are thus complementary constraints on comprehensive AI-assisted radiology reporting.
Read the full abstract
Solar and Stellar Astrophysics
Ensemble Individual Mode Frequency Modeling of Kepler Giants - Extending the IACgrid towards the red-giant branch
Precise stellar parameters, in particular ages, are vital for understanding the formation and history of the Milky Way. Usually, this is achieved by fitting stellar evolution models to observational constraints such as spectroscopic and astrometric parameters. Adding global asteroseismic properties to the classical parameters (global seismic modeling) breaks some degeneracies and improves the results. However, this approach has its limitations, with red-giant random age uncertainties usually $\sim$ 20-50%, and often lacking proper quantification of the systematics. We aim to extend seismic modeling to include individual mode frequencies for red-giant branch stars, increasing the number of evolved solar-like oscillators with precise stellar parameters (ages, masses, and radii), and allowing us to quantify the systematics in global seismic modeling. We developed a grid of stellar evolutionary models and computed the individual oscillation frequencies, including a figure of merit to find the best-fitting model. We validate our methodology on a sample of red giants with individual frequency modeling from the literature, finding our masses, radii, and ages in close agreement. We then apply our method to around 100 Kepler red-giant branch stars of near-solar mass and metallicity, which previously have only been modeled with global seismic parameters. We obtain random average uncertainties of 3.4%, 1.3%, and 13.2%, for stellar masses, radii, and ages, respectively. We find that ages from individual mode frequency modeling are significantly, 14% on average, larger than those from global seismic mass-constrained modeling. The modeled luminosities agree within 1.0% with those derived from Gaia parallaxes. We conclude that modeling individual mode frequencies improves age precision by a factor of two for red-giant branch stars, providing stronger constraints for galactic archaeology.
Read the full abstract
Portfolio Management
The Logic of Factor Portfolios: Boolean Rules, Interactions, and Trading Costs
Investors turn signals into holdings through decision rules: buy the top decile, hold stocks with high momentum and low volatility, exclude stocks that fail a screen. Many such rules are Boolean functions of threshold statements about signal ranks, and writing them this way shows both what a rule holds and how often it trades. A Walsh decomposition splits the return of any rule into exposures to individual signals and to interactions among signals; an intersection and a union of the same two screens, for example, have opposite interaction exposures. The same coefficients, together with the persistence of the underlying statements, determine turnover. Interactions are less persistent than their components, so rules that load on higher-order interactions trade more, and thresholds discard information, so at equal risk threshold implementations of interactions trade more than smooth ones. In U.S. stocks from 1973 to 2024, these predictions hold closely: predicted and realized persistence have correlations of at least 0.999, predicted turnover matches realized turnover with a slope of 1.02, and the same interactions are more persistent as products of ranks than as threshold rules. Trading costs rise with order, and because premia fall with order, 58% of single-signal portfolios have absolute premia above their quoted-spread costs over 1995-2024, against 24% of two-signal and 6% of three-signal interaction portfolios. Out of sample, interaction portfolios raise the Sharpe ratio of single-signal tangency portfolios, and over 1995-2024 they reject a six-factor model that conventional sorts on the same signals do not reject.
Read the full abstract
Mathematical Physics
A scalable computational framework for PDEs on physical networks
Transport, reaction and wave propagation on physical networks depend on dynamics along edges and interactions at junctions. We present a flexible framework for modeling physical networks at scale. A network calculus expresses edge dynamics and vertex contributions as fields over the whole network. Differential operators satisfy integration by parts-identities that connect local dynamics to global conservation laws, and explicit vertex constraints allow a variety of physical junction conditions. We implement this formalism using the open-source Dedalus spectral framework. This accommodates heterogeneous edge lengths, material properties and spectral resolutions. Initial-value, boundary-value and eigenvalue problems share a parallel solution method that eliminates internal unknowns independently on subgraphs and couples their solutions through shared boundary vertex interfaces. We verify numerical convergence and demonstrate strong and weak scaling on networks with millions of edges. We present applications to localized optical resonances in photonic networks, preferential channel formation during mineral replacement in evolving pore networks, and nonlinear dispersive waves for KdV on a periodic network.
Read the full abstract
Numerical Analysis
Mass-lumped finite elements with maximum-bound preservation and original-energy dissipation for the degenerate-mobility Allen-Cahn equation
In this work, we propose a mass-lumped finite element method for the Allen-Cahn equation that incorporates nonlinear mobility which may degenerate. The spatial approximation uses continuous, piecewise linear, mass-lumped finite elements, while convex splitting is employed for the double-well potential in time. Under the standard non-obtuse mesh condition, it is shown that the resulting fully discrete method preserves the maximum-bound principle unconditionally: if the initial approximation takes values in $[-1, 1]$, then so does the numerical solution at every time level. Additionally, the method dissipates the original discrete Ginzburg-Landau energy without any time-step restriction, even when the mobility vanishes at pure phases. This analysis relies on the monotonicity of the lumped finite element operator and the convexity of the split potential. Numerical experiments confirm these theoretical properties and demonstrate the robustness of the method in phase-separation dynamics involving degenerate mobility.
Read the full abstract
Fluid Dynamics
Quantifying the importance of equation of state and transport property compatibility in supercritical fluids
High-fidelity simulation of supercritical flows requires accurate values of thermodynamic properties including the density (rho), specific heat cp, enthalpy h, and entropy s as well as transport properties such as the viscosity (eta) and thermal conductivity (lambda) as functions of the temperature T and pressure P. In simulations of mixtures, such as supercritical carbon dioxide (sCO2) with contaminants, a generalized equation of state (EoS) is often used, leading to errors in rho(T,P) when compared to results from correlations developed specifically for individual species. Similarly, generalized correlations for eta and lambda are often used, with a corresponding decrease in accuracy. However, a second source of error is introduced in calculations of transport properties. These correlations are created and computed as functions of T and rho instead of as functions of T and P. Because of this, when eta(T,rho )and lambda(T,rho) are calculated, errors in rho(T,P) propagate through the calculation. This work creates a procedure for tracking both errors caused by using a generalized correlation for eta(T,rho) and lambda(T,rho), and for tracking the errors caused by using a generalized correlation for rho(T,P) in these calculations. The viscosity and thermal conductivity of supercritical CO2, N2, andCH4 are used as examples to demonstrate how to separate the errors due to the generalized property correlations and the errors due to the generalized EoS. Both sources of error are largest near the critical temperature Tc. These errors are large enough to impact both the numeric stability and the accuracy of simulations.
Read the full abstract
Quantum Physics
Deterministic Quantum Search for NP Problems Through Verifier-Based Oracle Construction
Problems in the class NP are important in computational science, where a given candidate solution can be verified in polynomial time, while finding a solution may require evaluating the entire search space in the worst case. Classical exhaustive approaches generally solve such problems by evaluating the verifier function over all possible inputs, resulting in a worst-case query complexity of O(N), where N = d^n represents the size of the search space, d is the number of possible values for each variable, and n is the number of variables. Quantum search algorithms can solve unstructured search problems more efficiently by reducing the query complexity to O(sqrt (N)), providing a quadratic speedup over classical exhaustive search. However, conventional quantum search formulations typically assume access to an oracle that identifies valid solutions, while standard implementations generally yield probabilistic success. In this work, we propose a generalised quantum oracle construction method for NP problems, together with a deterministic quantum search framework that identifies satisfying assignments using their corresponding classical verifier functions. The proposed framework retains the quadratic query speedup of quantum search over classical exhaustive search. We further establish that constructed oracle can be implemented using a number of gates polynomial in the size (L) of the binary encoding of verification function, consistent with the polynomial-time evaluation of the corresponding classical verifier, while the total qubit requirement remains O(poly(L)). The construction is demonstrated on multiple N-SAT instances, with circuit-level simulations confirming deterministic identification of satisfying assignments. These results establish the applicability of deterministic quantum search to NP problems while retaining the quadratic query advantage over classical exhaustive search.
Read the full abstract
Cosmology and Nongalactic Astrophysics
Odd Loops: What Structures Can Produce a Parity-Violating Inflationary Trispectrum?
Non-Gaussian perturbations generated during inflation can often be described perturbatively via integrals of kernels acting on Gaussian fields. In this work, we investigate, in a model-independent manner, whether these kernels can produce parity-violating (PV) contributions to the trispectrum, and which Wick's theorem contraction configurations of the Gaussian fields lead to them. In detail, we evaluate all possible contraction configurations arising from the third- and second-order density perturbations. We find that, for the third-order perturbations, only one configuration can support PV contributions, as it retains three independent wave-vectors. In contrast, none of the four second-order perturbation contraction configurations can produce a PV contribution under statistical isotropy. These results offer a first step toward deriving kernels from inflationary models capable of generating a PV trispectrum. Such kernels could then be used to perform PV searches with current and upcoming galaxy redshift surveys such as DESI, Roman, SPHEREx, and Euclid, and photometric efforts such as LSST. Our results also offer a guide for exploration of PV with CMB data-sets such as Planck.
Read the full abstract
Machine Learning
Designing Effective Monitor-Based Interventions for Mitigating Reward Hacking During RL
Reinforcement learning (RL) rewards are notoriously difficult to design and control, often leading to the model learning unintended behaviors such as reward hacking. One potential solution is to monitor for reward hacking and penalize it when detected; however, training against a monitor could lead to evasive behavior, and our general understanding of how to apply monitors effectively during training is limited. To study how best to use monitors to mitigate reward hacking, we introduce and open source three realistic environments where Qwen3-4B reward hacks: a coding environment hackable via test overwriting, a medical chat environment hackable via sycophancy, and a biography generation environment hackable via hallucination. We first focus on the coding environment, where we find that: (1) models can learn to evade highly accurate monitors by exploiting systemic flaws in probes and LLM judges; (2) monitors that leak more learning signal during RL suppress reward hacking but are more often evaded; and (3) including easier problems in training can decrease reward hacking. We apply our findings to build better reward hacking monitors for the medical chat and biography generation environments that improve upon naive baselines to reduce reward hacking rates across seeds to before-training levels in some interventions. Our results demonstrate that our takeaways translate to new settings and that better monitor intervention designs are possible.
Read the full abstract
Physics and Society
Effects of a Nuclear Detonation in Low Earth Orbit
More than 13,000 satellites operate in low Earth orbit (LEO), and concern has grown that nuclear weapons could be used in space to target them. Beyond the initial X-ray burst, electrons released by radioactive fission products can become trapped in Earth's magnetic field, repeatedly exposing satellites as an artificial radiation belt. Here we couple prompt radiation, electron trapping and belt evolution to satellite failures, collisions and atmospheric drag. For a 1-Mt detonation at 500km over the South Atlantic, repeated belt exposure disables approximately 98% of modeled satellites within one week in our central case. Tripling the assumed radiation tolerance delays widespread failure by only days: approximately 97% are still disabled within two weeks. The loss of maneuverability sharply increases collision risk, with approximately 20 collisions expected within two years. Drag removes many disabled spacecraft from lower altitudes, but replacement satellites could remain exposed to radiation and debris for years. These findings show how a single detonation could devastate orbital infrastructure and support efforts to prevent the deployment and use of nuclear weapons in space.
Read the full abstract
Differential Geometry
Radial reductions of variable exponent Laplacian equations on warped products with compact symmetry
We study nonlinear elliptic equations of variable exponent type, \[ -\Delta_{q(x)}u+|u|^{q(x)-2}u=f(x,u), \] on non-compact Riemannian manifolds with symmetry. We consider complete warped products \[ M=\mathcal R\times_\phi Y,\qquad g_\phi=dr^2\oplus\phi(r)^2g_Y, \] where $\mathcal R$ is a non-compact interval, possibly with boundary, and $Y$ is a compact homogeneous manifold. We establish sufficient conditions on the warping function ensuring a Ricci lower bound, a uniform non-collapsing volume condition, and the orbit-growth hypothesis of the abstract existence theorem proved in \cite{JFA}. We also obtain comparison bounds for the admissible warping functions. We analyze the corresponding radial equations for the admissible exponential and linear models $\phi(r)=e^{ar}$ and $\phi(r)=ar$, with $r\geq r_0>0$. For a constant diffusion exponent $q=k$, we consider explicit decaying profiles $u=\phi^{-\sigma}$, which belong to $L_1^k(M)$ provided $\sigma k>\dim Y$ in the exponential case and $\sigma k>\dim Y+1$ in the linear case. Under additional explicit parameter restrictions, these profiles solve a resonant equation with a positive bounded coefficient, a construction distinct from the superlinear regime of the abstract existence theorem. Finally, we establish logarithmic reductions for constant and variable radial exponents, and justify the passage from radial identities to weak formulations against arbitrary compactly supported interior test functions.
Read the full abstract
Artificial Intelligence
BitDecoding-FP4: Native FP4 KV Cache Decoding via Algorithm-System Co-Design
The key-value (KV) cache dominates memory capacity and bandwidth in long-context LLM serving. While 4-bit microscaling formats (NVFP4, MXFP4) shrink the cache and run attention on FP4 Tensor Cores, a fundamental gap remains between algorithmic quantization and system-aware acceleration: existing FP4 KV systems use vanilla quantization that causes severe accuracy drops, whereas accuracy-oriented 4-bit algorithms rely on vector codebooks whose lookup overhead degrades decoding throughput. We present BitDecoding-FP4, an algorithm–system co-design for accurate and efficient native FP4 KV decoding. We mitigate three error sources: Dual-Residual Query Quantization splits queries into two FP4 tensors, cutting query logit error from 60% to 1%; Underflow-Aware Probability Scaling prevents E4M3 scales from rounding long-context probabilities to zero; and Outlier-Aware Key Quantization combines static smoothing, Hadamard rotation, and scale search to suppress key outliers under token-wise scaling. At the system level, we remove the softmax bottleneck via transposed softmax with fused probability quantization and stacked query MMA on paged KV caches. Across dense and MoE models with GQA and MLA, BitDecoding-FP4 stays within 0.5 points of BF16 KV accuracy with NVFP4 (1 point with MXFP4), and tracks BF16 KV in RL training rollouts. In end-to-end serving on NVIDIA B300 with Qwen3-4B and Qwen3-30B-A3B, it delivers 2.4-3.0$\times$ higher peak serving throughput than BF16 KV. The implementation is available at https://github.com/OpenBitSys/BitDecoding-FP4.
Read the full abstract
Differential Geometry
Neutral Ricci spectra of homogeneous Riemannian four-manifolds
It is shown that every quadruple of real numbers with two positive and two negative entries is the Ricci spectrum of some left-invariant Riemannian metric on a Lie group of dimension four. Consequently, for eight out of the fifteen possible Ricci signatures of four-dimensional homogeneous Riemannian manifolds, one can determine precisely which spectra in each given signature actually occur. We also describe the Ricci spectra realized by left-invariant metrics on a certain class of 2-step solvable Lie groups of all dimensions $n>2$.
Read the full abstract
Complex Variables
Regular multi-types and the Bloom conjecture II
Let $M\subset\mathbb{C}^n$ $(n\geq2)$ be a smooth pseudoconvex real hypersurface. We prove that, for every $1\leq s\leq n-1$, the regular $s$-contact type, the $s$-commutator type, and the $s$-Levi type coincide at every point of $M$, thereby resolving Bloom's conjecture in any dimension. The solution in dimension three was obtained in 2021 by the second and the third authors. Moreover, for any smooth rank $s$ subbundle $B\subset T^{1,0}M$, we establish the equality of the commutator type and the Levi type associated with $B$, giving an affirmative answer to the generalized D'Angelo conjecture.
Read the full abstract
Quantum Physics
Discovering formally verified quantum programs with AI
The timeline by which quantum computers can solve breakthrough problems in science depends on improved quantum algorithms as much as on better hardware. Such algorithmic advances are typically achieved through many years of expert human collaboration. In mathematics and computer science, large language models have accelerated progress through automated discovery loops. However, quantum algorithms can only be evaluated by classical computers on restricted or small instances. Here we demonstrate automated discovery of quantum programs beyond these regimes including order-of-magnitude improvements on scientific applications. To achieve this, we build a quantum program library in the Lean theorem prover in which Gemini 4 agents synthesize novel programs against formal problem specifications and verify their correctness and cost. Agent teams collaborate, sharing results and knowledge through wikis and discussing ideas at organized conferences. Our results improve on prior baselines for parallel search (up to 2.57x), quantum arithmetic (290x), and simulations relevant to fusion, photochemistry and drug-metabolism by 12x, 8x and 2.5x respectively. Combining human and artificial intelligence at scale, under the rigor of mathematical proof, lets researchers design algorithms that today's classical and quantum computers cannot test, shortening the path to practical quantum computation.
Read the full abstract
Number Theory
Mellin Transform formulas for special values of Goss $L$-functions
In the present paper, we provide formulas for the special values of dual Goss $L$-functions of Drinfeld modules at any non-negative integer in terms of the determinant of the values of a motivic map, extensively studied by both authors in 2026, for a large class of Anderson modules. Our result may be seen as a function field analogue of conjectural formulas for special values of motivic $L$-functions in terms of algebraic Mellin integrals.
Read the full abstract
High Energy Physics (Theory)
On Superconformal Index Degenerations and Half-Indices in S-Confining Gauge Theories
We study the $p \rightarrow 0$ limit of four-dimensional $\mathcal N=1$ superconformal indices and its relation to half-indices of three-dimensional $\mathcal N=2$ theories. At the level of special functions, this limit reduces elliptic hypergeometric integrals to basic hypergeometric integrals. The scaling of the flavor fugacities determines the Neumann and Dirichlet boundary conditions of the chiral multiplets. The gauge multiplet has Neumann boundary conditions. We apply this prescription to s-confining theories with gauge groups $SU(2)$, $USp(2N)$, $G_2$, and $SU(N)$, and recover the half-index identities of several boundary confining dualities, expressed through the Nassrallah-Rahman and Gustafson integrals. For $SU(N)$ with a rank-two antisymmetric chiral, four fundamentals and $N$ antifundamentals, we obtain an analytic proof of the half-index identities proposed by Okazaki and Smith for all $N \geq 5$.
Read the full abstract
Mesoscale and Nanoscale Physics
Microscopic theory of exciton polarons in transition metal dichalcogenides
Polarons are composite quasiparticles formed when charge carriers strongly couple to lattice vibrations, manifesting as carriers dressed by a surrounding cloud of phonons. Although this dressing fundamentally renormalizes the effective mass and mobility of carriers, its impact on electrically neutral exciton polarons in 2D semiconductors has received limited attention. In particular, despite the growing interest in polaronic effects in atomically thin semiconductors, only a few theoretical studies have addressed spectral signatures and renormalized properties of exciton polarons in these materials. Here, we develop a microscopic model to describe exciton polarons in transition metal dichalcogenides. Applying this model to $1s$ acoustic polarons in an MoSe$_2$ monolayer, we reveal a distinct kink in the polaron dispersion, accompanied by a pronounced broadening of its spectral function. These spectral signatures mark a critical transition from a regime of well-defined bound exciton polarons, characterized by a renormalized energy, reduced group velocity, and enhanced effective mass, to a regime dominated by exciton-phonon scattering. Furthermore, by incorporating intervalley exciton-phonon interaction, we demonstrate that the energetic position of momentum-dark excitons strongly modifies the exciton-polaron lifetime. Overall, our work provides a microscopic framework for determining renormalized properties of exciton polarons and establishes experimentally accessible polaron signatures in 2D semiconductors.
Read the full abstract
Artificial Intelligence
Generative Latent Priors for Interpreting and Steering Protein Language Models
Dominant methods for interpreting large models impose strong structural assumptions. Generative Latent Priors (GLP) are a complementary approach: a diffusion model that learns the distribution of a frozen model's activations directly, imposing no assumptions. We port GLP to ESM-2 activations (UniRef50) and evaluate it at two host scales (ESM-2-8M and -650M), both as a lens for protein representations and as a prior for editing them. Generation quality scales smoothly with depth and compute. As a lens, single GLP "meta-neurons" are the strongest probes of discrete Swiss-Prot concepts at both hosts, beating InterPLM's released SAE features, and lead on continuous properties at 8M though not at 650M; falsification tests show this is mostly concentration, not new information. As an editing prior, the on-manifold correction makes DiffMean steering of three biochemical properties effective and restores on-manifoldness, but only in isolation: re-embedding the sequences actually produced after several rounds of decoding, the structural plausibility is compromised. This is mainly reproduced in additional SAE-based clamping structural steering experiments, where only Helix motif succeeds. The naturalness failure is traced to the rewriting coverage and the decoder rather than the prior, highlighting what the method must clear before it can serve for both discovery-by-editing and discovery-by-exploration. Trained GLPs are released on huggingface.co/collections/HPAI-BSC/glprot.
Read the full abstract
Lattice Field Theory
Machine learning for lattice QCD
Machine learning (ML) methods have been explored as a means of optimizing lattice field theory calculations, with increasing interest in recent years. This chapter gives an introduction and summary to the use of ML methods to optimize lattice field theory without introducing any uncontrolled approximations. We focus on those directions which are clearly established as of late 2025, marking the 50th anniversary of the beginning of the study of lattice gauge theory. This chapter is a contribution to the online book “Lattice QCD at 50 years,” (LQCD@50), edited by Tanmoy Bhattacharya, Maarten Golterman, Rajan Gupta, Laurent Lellouch, and Steve Sharpe.
Read the full abstract
Computer Vision
BEV3SR: Evidence-Aware 3D Spatial Reasoning from Video
3D spatial reasoning from video remains challenging for VLMs because videos provide sequential, view-local observations rather than persistent metric scene records. Many questions must be answered in a query-specified perspective rather than the observed camera viewpoint, and prompting with a fixed set of RGB views may miss sparse or occluded visual cues. To address these gaps, we introduce BEV3SR, a training-free and backbone-agnostic framework for RGB-D videos with camera parameters. BEV3SR turns video observations into queryable metric bird's-eye-view (BEV) scene memory and formalizes each query into an evidence-aware spatial task. The Spatial Task Formalizer specifies the query's reference frame, spatial constraints, objective, and evidence requirement. The evidence requirement determines the appropriate reasoning pathway. A query treated as requiring geometric evidence is handled by the Geometric Reasoner, which deterministically computes the answer using the metric BEV records in the query-specified reference frame. Otherwise, a query treated as requiring visual evidence is passed to the Visual Evidence Aligner, which uses BEV-indexed object records and their associated supporting views to retrieve object-consistent and scene-context RGB frames for VLM grounding. Within a standard 16-frame budget, BEV3SR consistently improves different VLM backbones across multiple 3D scene QA benchmarks, delivering large gains on spatially intensive questions and significant improvements on open-ended scene QA. These results demonstrate that BEV3SR offers an effective, efficient, and interpretable framework for explicit 3D spatial reasoning from video.