Papers

Climate downscaling aims to reconstruct fine scale spatial fields from coarse resolution inputs. Evaluating the quality of these reconstructions is challenging: low pointwise error can come at the cost of fine scale variability, while realistic spatial variability can be achieved with inaccurate local structures. The evaluation metric can therefore change which method appears to perform best. This work presents a multi metric benchmark comparing five spatial downscaling methods on ERA5 temperature, wind, and precipitation fields. Five criteria assess complementary properties: pointwise error, structural similarity, distribution error, spectral error, and gradient error. The results reveal a systematic trade off between spatial fidelity and fine scale variability. Some methods perform best on pointwise and spatially aligned metrics, but lose high frequency content, while others preserve substantially more spectral variability at the cost of less accurately positioned local structures. Consequently, method rankings change across metrics and variables. These results show that there is no single best downscaling method. Multi metric evaluation is therefore essential for assessing which properties of a climate field are preserved.

Why can masked prediction learn useful representations that unmasked reconstruction misses? We study this question in a high-dimensional model of a masked autoencoder (MAE) trained on data with shared latent structure and heterogeneous noise. We prove that masked linear reconstruction can recover the latent feature at linear sample complexity in regimes where unmasked linear reconstruction, equivalent to PCA, fails. The analysis also quantifies the statistical advantage of mask resampling, an established ingredient of masked pretraining. By introducing a fixed collection of $K$ masks per sample, we characterize its effect on feature recovery and downstream performance, identifying regimes where greater mask diversity lowers sample complexity. Guided by this prediction, we find that random cropping and flipping in standard image-training pipelines can obscure the advantage of mask resampling by renewing the prediction task even when the patch mask is fixed. Removing these transformations reveals a downstream advantage for dynamic over static masking in CNN autoencoders and vision transformers. A complementary BERT pilot finds benefits from greater mask diversity on downstream language tasks. Our results separate the benefit of the masked prediction objective from that of mask diversity, and show how a tractable theory can guide experiments that uncover advantages hidden by standard training practices.

We study the properties of hot rapidly rotating neutron stars using various equations of state based on the Brueckner-Hartree-Fock approach at finite temperature. The temperature dependence of gravitational mass, equatorial radius, Kepler frequency, moment of inertia are analysed in detail and universal relations are examined.

Phyllosilicate clays are widely used in polymer nanocomposites owing to their high anisotropy and tunable surface polarity. Their distribution and interface localization in polymer blends can be used to tune the properties of polymer-clay nanocomposites (PCNCs). A coarse-grained (CG) force field for clays can aid molecular simulations in PCNC development. Here, we developed MARTINI-3 parameters, a CG force field with high chemical specificity, for phyllosilicate clays with diverse surface polarities. Initial interactions for the clay functional groups were determined from hydration free energies, obtained by applying the Lifshitz theory to experimental surface tension data. These were fine-tuned using the structural, thermodynamic, and dynamic properties of thermoplastic starch (TPS)-clay composites from all-atom (AA) molecular dynamics (MD) simulations. The radial distribution function and two-body excess entropy of TPS components, not used in CG parameterization, were accurately estimated, establishing the robustness of the parameters. We investigated the effect of clay surface polarity on polymer segmental dynamics and structure-property relationships. The CG parameters were then used to study the effect of dodecyltrimethylammonium (C12TAB)-modified montmorillonite (MMT), an organically-modified MMT, on TPS-polyethylene (PE) blend morphology using large time- and length-scale MD simulations. We observed compatibilization of the TPS-PE interphase by the amphiphilic clay particle, reducing the TPS-PE interfacial tension from 45 mN/m to 13.06 mN/m. We found good agreement between MARTINI-3 estimates for properties of model MMT-based PCNCs and those from AA simulations and experimental data, establishing grounds for the transferability of the parameters to other systems.

High-stability frequency stabilization between lasers is essential for applications such as precision metrology, frequency dissemination, and quantum communications. We demonstrate the use of code-based multiplexing of a passive fiber-based interferometer for frequency stabilization and transfer between lasers separated by 85 GHz in frequency using a single shared passive reference. Out-of-loop characterization places an upper bound of sub-kHz/$\sqrt{\text{Hz}}$ on the transferred frequency noise at Fourier frequencies above 40 mHz, with a differential fractional stability of $7\times10^{-13}$ at 1 second. We also demonstrate continuous laser tunability and characterize the residual length noise coupling. Finally, we provide a framework for scaling the architecture to higher laser counts, offering a pathway to scalable, frequency agile stabilization of multiple light sources without the overhead of an optical frequency comb.

This paper investigates the performance scalability and adversarial robustness of Quantum Machine Learning (QML) models deployed on noise-model-based fake simulators. We conduct a dual-phased study, first benchmarking the computational throughput of Qiskit's Aer simulation engine across varying hardware architectures, and second, evaluating the effectiveness of Projected Gradient Descent (PGD) attacks and adversarial retraining strategies under realistic noise conditions. Our results quantify the runtime tradeoffs and scaling behavior for medium-scale simulations (projected up to 8 qubits) and demonstrate that high adversarial-to-benign retraining ratios (50/50) are essential for achieving practical model robustness for 4-qubit classifiers under realistic noise conditions.

This paper presents a guidance algorithm for micro aerial vehicles operating in unknown, cluttered environments using only onboard sensing. The method is based on a panel formulation originally derived from aerodynamic potential-flow theory and generates smooth, collision-free guidance vectors from locally perceived obstacles. The approach is extended to unknown environments by constructing and updating the obstacle representation online from onboard LiDAR measurements. The resulting obstacle-avoidance field is integrated with a nominal guiding vector field to produce the final control input. The system is experimentally validated in indoor flight tests under two scenarios: waypoint navigation and directional guidance. In both cases, the vehicle successfully completes its task while avoiding all obstacles in real time using only onboard perception. The results demonstrate that the method is computationally lightweight and suitable for onboard implementation, with pointcloud processing identified as the main practical limitation. These results support the feasibility of lightweight onboard guidance in unknown environments.

This paper investigates stochastic multi-level optimization where the objective is a nested composition of several smooth non-convex functions. We assume that only stochastic estimates of the gradient and function values for each level are accessible. Consequently, obtaining an accurate estimate of the overall gradient is challenging due to the nested structure. To address this, we employ a momentum-based estimator with mini-batches to track the function values of each level, which are subsequently used to construct momentum gradient estimators. We establish an optimal sample complexity of $\mathcal{O}(\epsilon^{-4})$ for finding an $\epsilon$-stationary point, avoiding the stronger average smoothness assumption commonly relied upon in prior literature. Furthermore, by employing a normalization technique, we attain the same rate without requiring problem-dependent constants to set hyperparameters. To achieve the optimal rate without mini-batches, we further develop a batch-free method that incorporates a first-order approximation and a clipping technique for function value estimation. Finally, we validate the effectiveness of our proposed methods through experiments on risk-averse portfolio optimization and hierarchical tilted empirical risk minimization.

In the quantum inverse scattering method, the sine-Gordon model is solved on a lattice, and its continuum and infinite-volume limits become uniform only after the length of the box is rescaled. We show that this rescaling, supplemented by an on-shell condition, defines a renormalization scheme. The whole dependence on the cut-off is absorbed into a factor $Z_L$ multiplying the length of the box, with an exponent fixed by the scaling dimension of the vertex operator; neither the field nor the coupling $\beta$ is renormalized, no subtraction scale is introduced, and the mass parameter receives a finite renormalization, fixed by requiring that the lightest breather has the mass of the boson in the linearized theory. The soliton and breather masses follow from the eigenvalues of the monodromy operator as functions of parameters that do not run, and reduce to the classical and semiclassical results as $\beta\rightarrow0$. In particular, the mass ratios are produced directly from the regularized construction, while the relation between the lattice parameters and the physical scale is absorbed into $Z_L$ and never needs to be computed. The scheme is a reparameterization of the theory which makes explicit the interplay between integrability and renormalization. We compare it with Coleman's normal ordering and with conformal perturbation theory, and we argue that it extends to ultralocal integrable theories with a massive spectrum, possibly including asymptotically free models that admit an ultralocal lattice regularization.

The influence of rear-face inclination on turbulence around a surface-mounted protrusion depends on the separated flow approaching its trailing edge. This dependence is examined experimentally using stereoscopic particle image velocimetry over twelve trapezoidal protrusions with upper-surface length-to-height ratios $L/h=1,2,3,4$ and rear-face angles of $30^\circ$, $45^\circ$, and $90^\circ$. Measurements were conducted at a height-based Reynolds number of approximately $5\times10^4$, with an incoming boundary-layer thickness comparable to the protrusion height. For $L/h=1$ and 2, a connected reverse-flow region extends over the upper surface into the wake. For $L/h=3$ and 4, upper-surface reattachment separates the two regions. Length governs this change in topology and the variation in turbulence levels at common spatial locations, whereas rear-face inclination has its strongest influence close to the surface. From $L/h=2$ onward, shallower faces have higher near-face turbulent kinetic energy, while vertical faces have lower energy but a larger spanwise fraction. A decomposition of the in-plane deviatoric production separates the effects of Reynolds-stress and mean-strain magnitudes from those of their relative principal-axis orientation. At the inclined faces, this contribution changes from negative to positive between $L/h=1$ and 2, before upper-surface reattachment occurs. Its subsequent increase with length is driven mainly by increasing tensor magnitudes, with the largest contribution at $L/h=4$ and $30^\circ$, where strong production remains close to the rear face. These results show that local turbulence production depends jointly on the flow state approaching the trailing edge and the position of the energetic shear layer relative to the rear surface.

Unmanned aerial vehicle (UAV) swarms increasingly rely on language-model agents to provide adaptive mission-level reasoning in uncertain environments. Fully distributed control, in which each UAV hosts an independent Small Language Model (SLM), removes reliance on a centralized coordinator but introduces an information-management problem: long-running interaction histories can degrade the reasoning context, while indiscriminate information dissemination increases communication and inference overhead. We address these challenges with a distributed UAV-agent architecture that enables continuous local SLM control through an event-driven reason-act-observe lifecycle. Runtime knowledge is represented as structured atomic notes and organized into core, local, and peer-specific memory. A deterministic interest-aware gossip engine selectively disseminates these notes according to recipient-specific semantic novelty and recency. We evaluate the architecture using ten UAVs in a simulated search-and-rescue mission. Our approach completes all experimental runs, whereas unrestricted flooding messages completes only 70-85%, and delegating forwarding decisions to the SLM prevents mission completion in every run. Compared with unrestricted flooding, our approach approximately halves inference-token consumption, reduces transmitted data, and achieves lower survivor-count error.

Magnetic skyrmions are promising nanoscale information carriers because their position can be manipulated by electrical currents. The additional ability of deterministic control of the spin structure would allow the use of magnetic skyrmions as multi-state bits, expanding the capabilities of skyrmion devices further. Such control could be realized by coupling all-optical helicity independent switching of magnetization (AO-HIS) to magnetic skyrmions. This is realized in a Pt/Ir/CoB/Gd/Pt multilayer that is engineered to stabilize magnetic skyrmions and, at the same time, enable AO-HIS by single ultrafast laser pulse. Laser excitation is used to write skyrmionic textures, whose polarity is selected by a small applied out-of-plane field. A single 30 fs laser pulse is then used to toggle the magnetization of an illuminated region containing skyrmionic textures, reversing their polarities without requiring any magnetic field or current pulses. This establishes deterministic optical switching of the skyrmion spin structure as a new manipulation channel, distinct from previously reported optical nucleation and annihilation. The material is then patterned into a wire, to combine the optical manipulation with current-induced skyrmion motion. These results establish a route toward opto-spintronic skyrmion devices in which light programs the internal state of a skyrmion and electrical currents control its position.

Large-cap or large-ensemble decoupling inequalities were initially studied by Demeter. These are intimately connected to the reverse square-function conjecture and restriction conjecture in Fourier restriction theory. By strips, we mean a partition of the $R^{-1}$-neighborhood of the paraboloid into $1\times R^{-1/2}\times R^{-1}$ curved large-caps. In this article, we prove the sharp $\ell^2(L^{10/3})$ decoupling estimate for strips in $\mathbb{R}^3$. We reduce the strip decoupling inequalities to a planar incidence problem, which is dealt with via two-ends Furstenberg incidence estimates. Our method is inspired by Wang–Wu on the restriction conjecture.

Practical Reinforcement Learning (RL) algorithms learn to solve Markov Decision Processes (MDPs) through iterative policy improvement in the presence of approximate evaluation. We study policy improvement from first principles, defining optimal policy improvement as producing the best policy attainable in a single update under specified constraints. We show that optimal improvement restricted to a set of states is equivalent to solving an induced MDP, characterizing planning with an explicit or implicit model as a path towards optimal policy improvement. Because practical methods commonly solve such induced problems through iterative improvement in the form of greedification, we take steps towards optimal greedification under the central practical constraint of approximate evaluation. We formulate greedification under this constraint as probabilistic decision-making under uncertainty and derive a novel operator that is optimal with respect to the resulting objective. Empirically, the operator and its practical gradient-based approximations improve aggregate performance across GumbelAlphaZero, SAC, ReBRAC and Generalized Policy Iteration, in experiments spanning discrete and continuous actions, model-based and model-free, online and offline RL.

We develop a conservative micro-continuum-cellular automaton method for simulating multispecies biofilm dynamics in complex flows. The proposed method couples the Darcy-Brinkman-Stokes equations, reactive transport, suspended bacteria, and biofilm dynamics with a two-stage cellular automaton algorithm for biofilm redistribution and interface evolution. While treating biofilms as evolving porous media, we ensure conservative redistribution of multispecies biomass across partially occupied cells. Donor and recipient cell volumes are explicitly accounted for to conserve biomass and preserve species composition on non-uniform meshes. The proposed method is assessed against diffusion-dominated benchmark cases, including single-species fingering and multispecies stratification, and is further evaluated through a mesh-convergence study for flow and growth over a rectangular bump. The framework is then applied to counter-diffusional biofilms in a membrane-aerated biofilm reactor as a canonical example. The results demonstrate that the framework captures the expected biofilm morphology and stratification in systems involving coupled flow, substrate transport, and biofilm dynamics. The proposed method provides a flexible computational approach for simulating multispecies biofilm dynamics in complex flows and geometries.

LLM agents use skills to improve performance on specialized tasks. To complete a user request, an agent may invoke several skills in sequence, allowing information produced under one skill to guide the next. Because skills may come from open-source repositories, this handoff can also carry attacker-controlled claims into later decisions. In this paper, we introduce APEX, which constructs and refines adversarial skill chains tailored to a user task and an attacker-selected action. The key insight is that an agent-written record of genuine task progress can carry a false claim of user approval across skills: an upstream skill induces the agent to create the record, and a downstream skill uses it to direct the attacker-selected action. Across four targeted-action families and six models on SkillsBench, the chains induce the selected action in 512 of 690 attempts (74.2%). On GPT-5.4, the full chain succeeds in 84.3% of attempts, compared with 17.4% when the workflow is merged into one skill. We further evaluate a prompting defense that asks the agent to check skill-produced files against the original request. On GPT-5.4, it lowers targeted-action success from 84.3% to 59.1%, while the verifier test-pass rate across 72 benign native-skill tasks falls from 86.7% to 56.3%. These results highlight the need for defenses that prevent attacker-directed actions while preserving legitimate task performance.

Universal coding of integers (UCI) provides prefix codes for positive integers such that, for every nonincreasing source distribution $P$, the average codeword length under $P$ stays within $K$ times $\max\{1,H(P)\}$. For a given UCI $\mathcal{C}$, the smallest such constant $K$ is called the minimum expansion factor of $\mathcal{C}$, denoted $C_{\mathcal{C}}^{*}$. Let $C^{*}\triangleq \inf \{C_{\mathcal{C}}^{*}\}$ over all UCIs. A UCI attaining $C^{*}$, or a family of UCIs attaining $C^{*}$ in the limit, is called optimal. It is currently known that $2\le C^{*}\le 2.0386$. In this paper, we construct a family of one-point plus uniform-tail distributions and prove that, for every UCI, a distribution in this family attains the worst-case ratio; thus, the family is least favorable for the UCI problem. We further establish an inequality, called the UCI inequality, which plays the same role for UCI as the Kraft inequality does for prefix codes: for any real number $B\ge 1$, it decides whether $B$ lies below or above $C^*$. The UCI inequality gives an equivalent definition of $C^*$. By numerical computation, we determine $C^* = 2.00012475703610197861701176297579886254\cdots$, the first $38$ decimal digits being certified. Once $C^*$ is known, we can theoretically construct the optimal UCI.

The tendency of the double auction mechanism to drive prices to competitive equilibrium has been well documented in laboratory experiments, but the phenomenon has lacked a theoretical explanation. This paper studies dynamic double auctions in a pure exchange economy where agents bid their indifference prices implied by their current holdings and preferences. We show that Walras equilibria coincide with the fixed points of the double auction and that repeated double auctions generate bounded sequences of allocations and prices whose cluster points are Walras equilibria with transfers.

We establish a class of sharp higher-order inequalities related to the uncertainty principle. First, by means of the expanding the squares method, we give a concise proof of sharp higher-order Heisenberg uncertainty principles. We provide explicit expressions for the optimal constants and establish the existence of extremal functions. Furthermore, we obtain sharp higher-order Heisenberg uncertainty principles for curl-free vector fields. In particular, when $N=2$, we give an answer to the higher-order version of Open Problem 9 raised by Maz'ya in [Integr. Equ. Oper. Theory (2018) 90:25]. Second, we establish sharp higher-order Heisenberg uncertainty principles that do not involve higher-order tensors. We also prove sharp higher-order Heisenberg uncertainty principles and higher-order hydrogen uncertainty principles for radially symmetric functions.

Model intelligence and fast response jointly shape the quality of interaction with speech language models, yet remain difficult to achieve together. Explicit chain-of-thought (CoT) improves reasoning and audio understanding, but generating intermediate reasoning tokens delays responses. Describing fine-grained acoustic cues further lengthens CoT and increases latency. Latent reasoning can reduce this overhead, yet existing methods often trail CoT and remain limited by single-path supervision and reasoning budgets that do not adapt to problem difficulty. We introduce AURAL, which models a distribution over multiple plausible reasoning continuations in latent space and jointly predicts chunks of future states to reduce sequential forward passes and reasoning latency. To provide initial supervision for latent reasoning, we construct AuralReason-683K: 683K bilingual speech utterances (about 1,000 hours) with concise CoT for emotion recognition, empathetic dialogue, and general reasoning. AURAL-RL then explores beyond these traces, rewarding concise reasoning that yields high-quality answers and adapting reasoning effort to each problem. Across two backbones, AURAL-RL achieves performance comparable to CoT-RL, with larger gains over the respective supervised checkpoints on most metrics. Analysis further shows that harder questions elicit more latent reasoning steps. On Qwen2.5-Omni, it reduces time to the first answer token by 11.8x, from 1.22 to 0.10 s, versus 0.05 s for direct answering.

World Action Models (WAMs) predict visual futures and robot actions, yet they remain susceptible to task-incomplete imagination, where plausible, action-consistent predictions omit the transition needed for task completion. In this paper, we show that this failure is not inherent to the world model backbone, but emerges when adapted for short-chunk control, which can repeatedly favor plausible local continuations over task-completing transitions. To address this, we introduce Completion Aware Guidance (CAG), a training-free sampling method that guides generation toward task completion. Across representative WAMs, CAG improves success from 64% to 70% on a RoboTwin 2.0 subset and from 69% to 75% in zero-shot simulation, while reducing task-incomplete imagination from 79% to 40%.

Spiking Neural Networks (SNNs) have attracted increasing attention due to their impressive temporal dynamics, energy efficiency, and brain-inspired mechanisms. Although SNNs have demonstrated promising performance in image classification tasks, recent studies have shown that they remain vulnerable to adversarial attacks, where imperceptible perturbations are added to input images to mislead model predictions. Existing defense methods mainly focus on training strategies, while the role of input encoding remains less explored. An observation is that the robustness advantage of Poisson encoding over direct encoding may benefit from its inherent randomness. Motivated by this, we propose a stochastic quantization encoding method that encodes the input image with controllable randomness adjusted by the quantization scale, thereby improving the adversarial robustness of SNNs. We further show that this method constitutes a general framework that reduces to both Poisson encoding and direct encoding under different choices of the quantization scale. Since it enhances robustness at the input encoding stage, it can be combined with existing training-based defenses for further gains. Experimental results on CIFAR-10 and CIFAR-100 demonstrate the effectiveness of the proposed stochastic quantization encoding method. To sum up, this work highlights the importance of input encoding for the adversarial robustness of SNNs, providing a new perspective for understanding and improving it.

We prove a factorization theorem for ordinary topological recursion near a regular nodal degeneration of a spectral curve, allowing logarithmic spectral coordinates. The partition function for closed genera $g\ge2$ factors into the partition function of the normalization, a universal Gaussian vacuum, and the exponential of a connected graph sum. This graph sum has strictly positive order in the vanishing period, and thus the factorization recovers the conifold gap, identifies the term of degree zero in the vanishing period with the free energy of the normalization, and gives finite graph formulas for the coefficients of positive powers of the period. At fixed genus, the regular series converges jointly in the period and the parameters along the nodal locus. The Gaussian neck tensors can be expressed in terms of relative Gromov–Witten invariants of a parametrized $\mathbb P^1$. As examples, we use remodeling to obtain Gromov–Witten factorizations for local $\mathbb P^2$ and local $\mathbb F_0$ in the conifold frame, with backgrounds $\mathbb C^3$ and the resolved conifold, respectively. The analogous local $\mathbb F_1$ factorization has background $\mathcal O(1)\oplus\mathcal O(-3)$ over $\mathbb P^1$.

This article provides a complete algebraic description of $\frac{1}{2}$-derivations, local $\frac{1}{2}$-derivations, and 2-local $\frac{1}{2}$-derivations on $n$-dimensional totally graded complex filiform Lie algebras of maximum length. Based on the foundational classification framework established by Janez Bernik (2020), we systematically determine the vector spaces of $\frac{1}{2}$-derivations for the six infinite structural sequences ($m_0(n)$, $m_2(n)$, $W^+(n)$, $m_{0,1}(n)$, $m_{0,2}(n)$, $m_{0,3}(n)$) and the five exceptional one-parameter families ($g_{7,\alpha}$ through $g_{11,\alpha}$). By analyzing the pointwise local evaluation equations via parametric matrix systems, we establish the structural linearity and rigidity of local $\frac{1}{2}$-derivations. In contrast, we demonstrate that the independent parameters residing in the boundary rows of the $\frac{1}{2}$-derivation matrices provide sufficient degrees of freedom to bypass linearity constraints. Exploiting these boundary configurations, we explicitly construct pure non-linear and non-additive 2-local $\frac{1}{2}$-derivations leveraging the homogeneous function of degree one, $f(z_1, z_2) = z_1^3 / (z_1^2 + z_2^2)$, thereby defining the exact boundary where local rigidity fails.

Time-reversal symmetry is a fundamental principle of physics describing the invariance of physical laws under reversal of the direction of time. We formulate a Bell-inequality-like test of this antiunitary symmetry using only forward access and trusted quantum operations: entanglement converts temporal input–output relations into measurable spatial exchange symmetry. For $n$-qubit unitary dynamics, we prove that reliably distinguishing the time-reversal-symmetric circular ensembles from Haar-random dynamics requires $\Omega(\min\{2^{n/2},2^{n-e}\})$ queries for any classically adaptive protocol. Here, $e=\min\{e_{\mathrm s},e_{\mathrm m}\}$ with $e_{\mathrm s}$ and $e_{\mathrm m}$ representing the probe and measurement logarithmic entanglement negativities, respectively. Maximally entangled probes and SWAP measurements reduce this cost to a constant number of queries. Furthermore, we develop a time-reversal symmetry test for arbitrary fixed, compatible probes and measurements, relate its query complexity to their logarithmic negativities, and match the lower-bound scaling in the high-entanglement regime by optimizing the probe and measurement. Our results establish a quantitative connection between entanglement and time-reversal symmetry, bridging two central concepts in quantum information science and fundamental physics.

Wanda (Sun et al., 2024) prunes large language models by scoring weights independently within each linear projection, although queries and keys interact through dot products. We introduce QK-Wanda, which scores query and key weights by their individual deletion costs under an unmasked pre-RoPE reconstruction objective. It augments Wanda scores with information from the opposite projection (keys for query weights, and queries for key weights), allowing both projections to share a pruning budget. Its closed-form scores require no gradients or weight updates; full pruning takes 1.3% longer than Wanda on A100 and 3.1% longer on H200 with the calibration used in our main experiments. We evaluate QK-only pruning across 15 models from TinyLlama, Llama 2, Llama 3, and Qwen2.5, spanning 0.5B-72B parameters. Relative to Wanda, QK-Wanda reduces QK reconstruction error by an average of 60% at 50% sparsity and 45% at 80%. Downstream gains depend on the model. At 80% sparsity on Llama 2 70B, WikiText-2 and C4 perplexity decrease by 20.3% and 13.5%, while mean zero-shot accuracy rises by 5.94 percentage points. Qwen2.5-72B also improves, but Llama-3.1-70B has substantially higher perplexity despite lower reconstruction error. These results show both the promise of coupled pruning criteria and the limits of local reconstruction as a predictor of model quality.

Patent search requires processing documents routinely exceeding tens of thousands of tokens. Most neural retrieval approaches operate on truncated inputs, limiting their effectiveness. Graph-based retrieval addresses this by representing each patent as a structured invention graph, but constructing these graphs relies on brittle rule-based parsers. We present the neural parser, which adapts biaffine attention from dependency parsing to predict invention graphs directly from patent text. Our local biaffine attention restricts pairwise scoring to a sliding window, reducing complexity from $O(n^2)$ to $O(n \cdot w)$. Since local and global scoring share the same weights, the model trains on short sequences and deploys on documents exceeding 40,000 tokens without retraining. Distilled from 1 million rule-parsed documents, it surpasses its teacher at 3$\times$ lower inference cost: neural graphs improve citation recall by 0.5% on short queries and 1.1% on full documents in a downstream Graph Transformer retrieval system.

In this work, a comprehensive theoretical and numerical study of the quantum-chromodynamic and quantum-electrodynamic radiative corrections: 1. $q\bar{q} \rightarrow q\bar{q}\gamma$, 2. $q\bar{q} \rightarrow gg\gamma$, and 3. $q\bar{q} \rightarrow g\gamma\gamma$ to the annihilation of a quark-antiquark pair $q\bar{q} \rightarrow g\gamma$ of prompt photon production in proton-proton collisions at NICA energies $\sqrt{s}$ = 10 GeV was performed. The differential cross-sections and double spin asymmetries $A_{LL}$ of the radiative corrections were investigated analytically using the FeynCalc and modulated in PYTHIA 8.316. Radiative corrections are significant at low collision energies of protons $\sqrt{s}$ and small transverse momentum of photons $p_{T}$. It was shown that radiative corrections to the annihilation of a quark-antiquark pair process satisfy the following relations: $$ 2. \sigma_{\sqrt{s}}(q\bar{q} \rightarrow gg\gamma) > 1. \sigma_{\sqrt{s}}(q\bar{q} \rightarrow q\bar{q}\gamma) > 3. \sigma_{\sqrt{s}}(q\bar{q} \rightarrow g\gamma\gamma) $$ $$ 1. \frac{d\sigma(q\bar{q} \rightarrow q\bar{q}\gamma)}{dp_{T}^{2}} > 3. \frac{d\sigma(q\bar{q} \rightarrow g\gamma\gamma)}{dp_{T}^{2}} > 2. \frac{d\sigma(q\bar{q} \rightarrow gg\gamma)}{dp_{T}^{2}} $$ The polarization of colliding protons has almost the same effect on processes 1. $q\bar{q} \rightarrow q\bar{q}\gamma$, 2. $q\bar{q} \rightarrow gg\gamma$, and 3. $q\bar{q} \rightarrow g\gamma\gamma$ and in this case above indicated relation is also satisfied. A direct comparison between strict leading order analytical calculations in FeynCalc and PYTHIA 8.316 modulations revealed excellent agreement in the central kinematic regions.

The interplay between electronic and structural degrees of freedom underpins many emergent phenomena in quantum materials, yet magnetoelastic coupling in metallic systems is typically weak and confined to low temperatures or microscopic length scales. Here, we report giant magnetoelastic coupling slightly above room temperature in the uranium-based intermetallic compound UIr$_4$Al$_{15}$. Using temperature-dependent single-crystal X-ray diffraction together with anisotropic magnetic susceptibility measurements, we directly resolve subtle but reproducible structural distortions coupled to magnetic alignment. Despite the absence of crystallographic symmetry breaking, pronounced anomalies emerge in lattice parameters and selected chemical bond distances near the magnetic transition region, revealing an unusual sensitivity of the crystal structure to magnetic orientation. The coupling enables direct probing of how atomic distances and local chemical bonding govern the electronic and magnetic states in a bulk intermetallic material. Our results establish UIr$_4$Al$_{15}$ as a rare platform in which magnetism and lattice distortions are strongly intertwined near room temperature, opening new opportunities for magnetically responsive quantum materials and functional magnetic sensing applications.

Among Archimedes' many celebrated discoveries is a striking fact: the region of the unit sphere between two parallel planes, both meeting the sphere and separated by a distance $h$, has area $2\pi h$. We prove that, among connected smooth closed embedded surfaces in $\mathbb R^3$, this property characterizes the unit sphere for a single fixed width.