Papers

We prove that, for $2\leq p<n$, the optimal $L^p$-Sobolev constant on every complete $n$-dimensional Euclidean minimal immersion without boundary is the Euclidean Aubin–Talenti constant. The proof combines optimal transport to a suitably chosen target with a two-kernel comparison linking the tangential transport moment to deficits in the normal-fibre and Jacobian estimates.

Some backdoor attacks never touch an image. FLIP only changes labels: on CIFAR-10, flipping 3% of them makes a ResNet-32 send 99.7% of triggered source-class images to the attacker's class. The usual defenses look for the poison inside a model trained on those same labels. So when they struggle, is the problem the test they run, or the space they run it in? We answer with a controlled swap: we keep the test, the labels and the metric fixed, change only the feature space, and judge each defense by retraining the victim on what it keeps. The space decides. On dog$\to$cat the same nearest-neighbor check scores AUROC 0.546 on the poisoned model's features, 0.967 on frozen DINOv2 features that never saw a label, and 0.995 on the same network trained on clean labels, an oracle control rather than a defense. We reproduce SPECTRE, the defense FLIP's authors recommend, to 90.94 / 0.97 against its published 91.47 / 1.7, then price it: on truck$\to$deer it deletes 20% of every class, while a prototype audit on the frozen features deletes 6% for higher clean accuracy and lower attack success. On three harder pairs that audit leaves about a quarter of the attack standing. A count read off encoder geometry and clean labels, before any attack is run, separates the pair where the audit removes the attack from those where it only partly does, though on a single zero-count pair. Three results warn against the easy story. A better detection score does not mean a safer model: on two of four pairs Confident Learning scores the higher AUROC, and on dog$\to$cat leaves 2.6 times the attack success. The prototype family, our best static performer, is the first an adaptive attacker slips past. And past a point, catching more poisons stops reducing attack success. Check labels in a space that never learned them, and judge a defense by the retrained model, not by its detection score.

Recent image-to-3D models generate complete, high-quality assets from a single image, with global shape and appearance that are already convincing; what remains is the fine scale, where small, thin, elongated, or detail-rich regions are reconstructed softly and under-resolved. This gap can arise in generators with fixed-resolution spatial grids: the same spatial sampling density is used for a large, smooth, low-curvature surface and for a thin, high-curvature structure that carries most of the perceived detail. To this end, we introduce NOFACE3D, a training-free framework that refines these detail-critical regions while preserving the rest of the asset. Given a generated mesh, its input image, and a refinement box, NOFACE3D extracts the selected submesh, scales it up to fill the generator's input volume, and re-encodes it into the full-resolution latent grid. Conditioned on the image crop, it refines the region through flow-based sampling with a spatially varying noise schedule that anchors boundary latents. The refined region is then integrated back into the original asset, preserving the surrounding geometry and appearance. We implement NOFACE3D on TRELLIS.2 and its sparse voxel representation; all of it operates purely at inference time on the frozen pretrained generator. We evaluate NOFACE3D on PAniC-3D, TexVerse and BigDetail, a new benchmark of 339 assets with annotated detail-critical regions. On all three, it generates sharper geometry and finer texture detail than the baselines. Project page: https://idkcouscous.github.io/NoFACE3D/

Multi-robot coordination is essential in applications such as coverage, target tracking, and information gathering, where robots must jointly select actions to maximize team performance. Submodular maximization provides an effective framework for such coordination, and resilient variants further account for robot failures or adversarial removals. However, existing methods typically overlook the impact of robot loss on the communication network, which may disconnect the surviving team and prevent continued coordination. To address this issue, we introduce Resilient $k$-Connected Submodular Maximization, which jointly considers task performance, resilience, and communication connectivity. The resulting feasible family is not a matroid, preventing the direct application of classical curvature-greedy guarantees. We overcome this challenge through a Menger-based connectivity reduction, a connectivity-aware greedy algorithm with a feasibility-preserving fallback, and a new curvature-based analysis that captures the cost of enforcing connectivity through a single factor $\rho$. Our method achieves a worst-case post-failure objective within a factor of $[\max(1-\nu_f(\mathcal{I}),h(N,\beta))]^2 \rho \frac{1}{\kappa_f}(1-e^{-\kappa_f})$ of the optimal connectivity-respecting resilient assignment, while recovering the classical unconstrained guarantee in the appropriate limits. Experiments on multi-robot target tracking validate the proposed method and characterize the practical impact of connectivity constraints.

Impacts of octupole correlations on low-energy excited $0^+$ states and two-neutron transfer intensities in Ra and Th isotopes are analyzed within the interacting boson model, using the Hamiltonian parameters that are obtained from microscopic nuclear density functional theory calculations. Incorporating the octupole degree of freedom in the model is shown to give an overall reasonable description of the ground-state-band energy levels and to lower the $0^+_2$ ones significantly. The calculation yields a number of excited $0^+$ states, most of which are formed by the configurations of more than one octupole boson. Octupole correlations make significant contributions to the $0^+_1 \to 0^+_2$ $(p,t)$ and $(t,p)$ transfer intensities in those nuclei in which the octupole collectivity is most enhanced. Behaviors of the predicted transfer intensities with the increasing neutron number along the isotopic chain reflect nuclear structural evolution.

AI data centers (AIDCs) are emerging as large power-system loads, making their operational flexibility increasingly valuable. Existing studies often treat AIDC demand as continuously adjustable or generically delay-tolerant, which overstating the flexibility of large language model (LLM) pretraining. This paper proposes an execution-feasible dispatch framework for LLM pre-training in renewable-dominant AIDCs. LLM descriptors are first converted into checkpoint-domain daily workloads, then mapped to feasible 3D-parallel GPU modes with explicit throughput, GPU-capacity, memory, and IT-power limits. The resulting checkpoint- and mode-constrained load states are embedded into a day-ahead power dispatch model. Case studies on a 2048 GPUs AIDC with two LLaMA pretraining tasks show that checkpoint-aware 3D dispatch reduces operating cost by 20.02% and renewable curtailment from 1.077 to 0.014 MWh relative to fixed continuous training. Related results also show that checkpoint records alone do not create load flexibility,practical grid-interactive flexibility arises from checkpoint waiting combined with execution-feasible 3D-mode compensation.

Energetic partons traversing the quark-gluon plasma in relativistic heavy-ion collisions induce a medium response that modifies the production of associated hadrons, whose chemical composition provides a sensitive probe of this response. We study jet-associated hadrochemistry using the coupled linear Boltzmann transport and hydrodynamics model with a hybrid hadronization scheme that combines Cooper–Frye particlization, quark coalescence and string fragmentation. For inclusive charged jets in $p+p$ and 0–10% central Pb+Pb collisions at $\sqrt{s_{\mathrm{NN}}}=5.02$~TeV, the calculated in-cone $K/\pi$ and $p/\pi$ ratios are enhanced in Pb+Pb relative to $p+p$ collisions, consistent with preliminary ALICE measurements, with coalescence between shower and thermal partons providing a substantial contribution. Extending the analysis to dijet events, we predict that both ratios are enhanced around the leading and subleading jets, more strongly around the subleading jet and more pronounced for larger radial size. After mixed-event acceptance correction and sideband background subtraction, the yields of associated pions, kaons and protons all become negative near $|\Delta\phi|\simeq\pi/2$, signaling a depletion caused by the diffusion wake. The $K/\pi$ and $p/\pi$ ratios of these deficits remain finite and exceed the $p+p$ values around the jet peaks, indicating that the missing hadrons originate from jet-excited medium. These results establish dijet hadrochemistry as an experimentally accessible tomographic probe of jet-induced medium excitation and the diffusion wake.

On-policy distillation (OPD) transfers capabilities through teacher predictions on student-generated sequences, as studied by GKD and MiniLLM for language models and Video-OPD for video tasks. We ask how internal supervision can further support temporal learning. Direct hidden-state alignment does not explicitly use the student's temporal responses to choose a supervision direction. We propose Temporal-Response Guidance (TRG), which measures representation responses to local temporal perturbations, adds the teacher correction's projection onto their span back to the original correction, and restores its original norm. The projection is not normalized separately, and correction components outside the response space are retained. With an orthogonal tangent projector and a detached target, TRG changes the supervision direction while preserving the cosine-loss value and aligned-representation gradient norm at target construction. On TempCompass, Video-MME, MVBench, and Video-MMMU, TRG improves the 4B student's macro accuracy by 2.86 percentage points over the reported output-only OPD baseline.

FLEX options are customized exchange-traded contracts allowing participants to choose nonstandard strikes, expirations, exercise styles, and settlement terms while retaining central clearing. We investigate whether single-name FLEX activity leaves measurable footprints in regular listed-option markets and underlying equities. Using a linked dataset of FLEX consolidations, daily open-interest files, listed-option chains, equity returns, and corporate actions over 250 trading days (March 2025-March 2026) across 41 underlyings and 53 event dates, we document three primary empirical findings: (1) Matched-control difference-in-differences estimates reveal a robust post-event increase in listed-option open interest (beta = 0.3061, p = 0.0042, Benjamini-Hochberg p = 0.0168). (2) FLEX activity is strongly selected into high implied-volatility (IV) environments, concentrated in mid-price underlyings (date-fixed-effect ATM IV premium beta = 0.1235, p = 0.0030). (3) Event-day IV-minus-realized-volatility spreads predict execution-stressed short-straddle returns (beta = 0.0366, p = 0.0100). Short-lived equity return effects (CAR_0 difference of -0.85%, p = 0.0442) do not persist over 1-to-5-day horizons. Overall, FLEX activity serves primarily as an observable marker of institutional volatility demand and open-interest positioning rather than a directional stock-return predictor.

We show that a variant of the Wagner/BKW-style algorithm for LWE of Kirchner and Fouque [KF15] solves decision and search LWE with secrets of size $n$ over prime moduli $q=n^{\kappa+o(1)}$, for any fixed $\kappa>0$, with errors bounded by $\exp((\ln n)^{o(1)})$, in time $2^{O(n/\ln\ln n)}$ given $n+O(n/\ln\ln n)$ original samples. Further, if each coordinate of the secret is also bounded by $\exp((\ln n)^{o(1)})$, the same time bound holds with errors bounded by $n^{o(1)}$ and only $O(n/\ln\ln q)$ samples. Earlier, it was known that Kirchner and Fouque's algorithm could solve LWE with these parameters in subexponential time with a much larger number of samples $m = \Omega(n\log{n})$ [HKM18]. Ducas et al. [DEL25] showed that an adaptation of this algorithm could solve the corresponding dual, the $\mathrm{SIS}^{\infty}$ problem, with a suitable norm bound $q/\operatorname{polylog}(n)$ and $m=n+\omega(n/\ln\ln n)$ variables in subexponential time. Whereas their analysis involves interpreting the algorithm in terms of chains of lattices, our approach is quite elementary and is based on bounding the lower moments of a Fourier statistic.

We prove that, for every fixed $d\geq 2$ and $\varepsilon>0$, the $\mathbb{F}_q$-points of a geometrically irreducible reflexive plane curve of degree $d$ over $\mathbb{F}_q$ can be covered by at most $\left(\frac{1}{d}+\varepsilon\right)q$ many $\mathbb{F}_q$-lines, provided that $q$ is sufficiently large. The same conclusion holds for $\mathbb{F}_q$-irreducible curves that are not geometrically irreducible, since they have only $O(d^2)$ $\mathbb{F}_q$-points. In particular, for sufficiently large $q$, the bound applies to every $\mathbb{F}_q$-irreducible plane curve of degree $d$ when $\operatorname{char}(\mathbb{F}_q)>d$. Our bound is asymptotically sharp. The proof combines Entin's Chebotarev density theorem for varieties over finite fields with the Pippenger–Spencer theorem on covers of nearly regular hypergraphs.

Recurrent-depth language models improve inference-time reasoning by repeatedly applying a shared network to refine internal representations, but the additional recurrent computation also increases inference cost. This overhead is particularly pronounced during long-context prefill, where the full input sequence is repeatedly processed across recurrent depth. Weight-and-activation quantization offers a natural way to accelerate this computation through lower-precision arithmetic, yet repeatedly applying quantized computation introduces numerical perturbations that accumulate along the recurrent trajectory. We systematically analyze how such errors propagate across recurrent depth and show that quantization errors introduced earlier are attenuated by more subsequent recurrent steps and therefore have a smaller effect on the final representation. Matched-budget experiments strongly support this prediction, showing that placing quantized computation earlier in recurrent depth consistently yields the best fidelity. Based on this insight, we introduce RAMP (Recurrent Adaptive Mixed Precision), which performs early recurrent refinement using low-precision weight-and-activation computation and restores high precision for terminal refinement. Across context lengths up to 4K, RAMP preserves near-full-precision prefill quality while achieving up to 1.16x end-to-end first-prefill speedup on an NVIDIA H100, corresponding to a 13.9% latency reduction.

Consumer rotating devices, including washing machines, air conditioners, electric fans, robotic vacuum cleaners, and home fitness equipment, increasingly rely on vibration sensing for intelligent condition monitoring and predictive maintenance. Ensuring reliable operation in these resource-constrained appliances requires efficient, lightweight artificial intelligence (AI) solutions capable of processing non-stationary signals under real-world operational interference. This paper proposes a lightweight AI-assisted vibration sensing framework that employs continuous wavelet transform (CWT) and Hilbert-Huang transform (HHT) to convert one-dimensional (1-D) vibration time-series signals into two-dimensional (2-D) time-frequency representations. These representations are subsequently analyzed using lightweight deep learning architectures, specifically MobileNetV3-Small, ConvNeXt-Tiny, and the You Only Look Once version 11 classification model (YOLOv11-cls). Public bearing benchmark datasets, including Case Western Reserve University (CWRU), Paderborn University (PU), and Intelligent Maintenance Systems (IMS), are utilized as representative rotating components to investigate model generalization. The proposed framework is designed with eventual embedded deployment in consumer electronics in mind. Experimental results demonstrate that CWT combined with YOLOv11-cls achieves high diagnostic accuracy, while MobileNetV3-Small provides an attractive low-latency alternative for computationally constrained edge platforms. Conventional random splitting is also compared with dataset-specific structured isolation. The results highlight the importance of avoiding data leakage and quantify tradeoffs in classification accuracy, parameter count, floating-point operations (FLOPs), model footprint, preprocessing cost, and central processing unit (CPU) inference latency.

We report redshifted cool gas toward the central star-forming region of the nearby starburst dwarf galaxy NGC 4214. HST/COS spectra show low-ionization absorbers redshifted relative to the systemic velocity, most clearly in SII ($\Delta v=19.8\pm2.6\ {\rm km\,s^{-1}}$) and supported by MgII ($\Delta v=30.3\pm3.0\ {\rm km\,s^{-1}}$), while PII remains tentative ($\Delta v=7.5\pm3.4\ {\rm km\,s^{-1}}$). The Keck/HIRES spectra also reveal a redshifted CaII absorption component ($\Delta v=25.6\pm1.5\ {\rm km\,s^{-1}}$). Because the absorption is detected against the central UV or optical continuum, the gas must lie in the foreground of the star-forming region and is moving away from the observer and toward the background UV source in projection. A VLA–THINGS HI spectrum extracted at the COS position within one VLA beam also shows a redshifted neutral component at $v_{\rm hel}=329.2\pm0.9\ {\rm km\ s^{-1}}$, offset by $38.2\pm1.3\ {\rm km\ s^{-1}}$ from systemic and by $27.6\pm1.0\ {\rm km\ s^{-1}}$ from the dominant local HI component. The FAST HI mapping further reveals an extended, kinematically disturbed neutral disk, with residual structures relative to a warped-disk model. Together, the redshifted UV absorption, the redshifted HI component, and the disturbed large-scale HI velocity field show that the central anomalous cool gas is observed in a galaxy with a disturbed extended neutral disk. This geometry suggests that the redshifted cool gas may trace inflow or recycled fallback toward the central region, although warped-disk or non-circular disk motions cannot be fully excluded.

The indirect-control paradigm offers a scalable approach to quantum computation by restricting external operations to a small subset of qubits, whereas the entire system evolves under its intrinsic many-body Hamiltonian. This framework reduces hardware complexity and limits the noise introduced through control channels. Recently, Anan et al. proposed an implementation of the variational quantum eigensolver (VQE) within this indirect-control architecture. However, the impact and integration of quantum error mitigation techniques in such a setting remain largely unexplored. In particular, zero-noise extrapolation (ZNE), a widely used error mitigation strategy, has not yet been systematically adapted to indirect-control VQE. A key requirement of ZNE is the controlled amplification of noise, which is commonly achieved through circuit-level techniques such as unitary folding. In the indirect-control framework, however, quantum operations are largely governed by continuous time-evolution under the system Hamiltonian, making the construction of physically realizable inverse operations non-trivial. In this study, we investigate the feasibility of implementing ZNE within the indirect-controlVQEframework.We address the challenges associated with unitary folding in the presence of time-evolution operators and, with a minimal relaxation of the indirect-control constraint, we propose a ZNE-based error-mitigation strategy that exploits the symmetry of the one-dimensional XY spin chain.

The predictive performance of neural networks for receptor functional potency modeling depends on both molecular representations and prediction-head design. Using GLP-1R cAMP-related pEC50 prediction as the primary task, we jointly train a classical molecular encoder with a linear base head and a shallow quantum-circuit residual head that share the learned molecular representation. We evaluate the models using random and structure-group splits, include tuned Chemprop as a domain reference, and extend the evaluation to ADRB2 and MC4R. Under the GLP-1R structure-group split, the quantum-residual models reduce mean absolute error relative to the corresponding base models from 0.677 to 0.639 with a multilayer perceptron (MLP) encoder and from 0.680 to 0.641 with a graph neural network (GNN) encoder. In structural control experiments, a parameter-count-matched real-orthogonal control achieves similar predictive performance, suggesting that some quantum-residual gains may be associated with structured quadratic feature interactions. By combining receptor functional potency prediction with structural analysis of the quantum residual, this study provides empirical evidence to inform compact residual-head design and highlights the potential of quantum-circuit parameterizations for this task.

Evaluating deep learning optimizer is challenging because confounding factors such as hyperparameter tuning and optimizer-model co-design often obscure true algorithmic performance. We present results of the updated AlgoPerf V1 benchmark, in which we evaluate a new set of training algorithms on wall-clock time-to-target across nine workloads on fixed hardware. This report focuses on the results of the benchmark's self-tuning ruleset, which requires submissions to use a single configuration across all workloads. Out of the 13 submissions in PyTorch and JAX, Schedule-Free AdamW ranked first. In addition to the wall-clock scores, we also analyze the trade-offs between step efficiency and wall-clock execution, revealing that matrix-preconditioned and two-loop methods achieve good sample efficiency but require more aggressive computational optimizations to be competitive in runtime. Finally, our hyperparameter saliency and retention analyses show that the learning rate and momentum are highly sensitive across domains and algorithms, highlighting the need for methods that are more robust against hyperparameter tuning and can generalize well to new tasks.

Multi-view representations enable pretrained image and video vision-language models to understand 3D scenes, but their long visual token sequences make inference costly. Redundancy arises both from repeated observations of the same scene regions across views and from processing all visual tokens at every LLM layer, regardless of their relevance to the current computation. We propose GeoGR, a two-stage framework that jointly compresses scene representations and visual computation. At the projector level, GeoSemZip consolidates cross-view observations into 3D voxel tokens and combines semantic saliency with spatial coverage to construct a compact scene representation. Unselected voxel features are aggregated into contextual anchors to retain scene information that would otherwise be discarded. At the LLM level, we identify an effective visual computation window and observe that query-conditioned token relevance remains locally stable but changes over longer layer intervals. Guided by these findings, GroupRoute combines late entry and early exit with recoverable group-wise skipping. Low-relevance tokens bypass complete LLM layers while retaining their hidden states and can rejoin the active group at subsequent anchor layers selected under a computation budget. For compression-aware post-training, we construct a compact dataset of 67K samples by retaining 30% of the original training pool. With this protocol, GeoGR retains 95.2% and 96.8% of the dense post-trained models' performance on LLaVA-OneVision-7B and Video-3D LLM-7B, respectively, while accelerating LLM prefill by $6.8\times$ and $6.1\times$. These results demonstrate the consistent effectiveness of GeoGR across 3D VLMs with and without explicit 3D spatial encoding, highlighting a favorable performance-efficiency trade-off. The code is available at https://github.com/EIT-NLP/GeoGR.

Trajectory extrapolation from a partial target trajectory is underdetermined, but completed related trajectories provide information about the unresolved continuation. We propose target-constrained pooling (TCP), which separates the finite-horizon endpoint from continuation shape and estimates that endpoint from two complementary source signals. Uniform pooling captures final response levels, while matched-increment pooling captures the remaining change after aligning source and target states. TCP selects representative sources, projects both endpoint estimates onto a target-consistent region, converts them into corrections of the free forecast, and controls their influence through strength calibration and degradation-aware weighting. Across 25,245 targets from five domains and both forecasting directions, TCP reduces mean RMSE by 35.8% to 42.1% for all ten target-only forecasters and improves all 100 forecaster, dataset, and direction comparisons. The two endpoint signals consistently outperform either alone, while representative selection remains within 0.64% to 1.02% of the full donor pool using 12.7x fewer donor inputs and 138.8x fewer matching distance computations.

The real-time dynamics of quantum phase transitions in open systems coupled to finite-temperature environments sits at the heart of contemporary condensed-matter and quantum-information science. Here, invoking a Schwinger-Keldysh formalism, we uncover that, within the coherence-dominated regime of critical real-time evolution, a dissipative coupling deemed naively irrelevant determines, counterintuitively, the longest timescale that holds sway. Possessing a well-defined temporal scaling dimension, it engenders a novel dissipation-regression scaling that systematically reproduces the unusual Kibble-Zurek scaling and relaxation observed in recent numerical studies. By contrast, dissipation-dominated dynamics behaves qualitatively differently. There, furthermore, we show that the discretized and diagonalized formulations of the same open quantum Ising chain realize distinct universality classes.

The rapid growth of electricity demand from AI data centers is creating a dual mismatch in power systems: a temporal mismatch between fast-rising demand and slow transmission expansion, and a spatial mismatch between concentrated data-center loads and geographically dispersed renewable resources. These mismatches can exacerbate transmission congestion, resulting in unserved data-center load and renewable-energy curtailment. This study models geographic workload migration as a virtual transmission line (VTL) that changes nodal data-center demand without transferring physical electricity. A mixed-integer linear programming model is then developed to co-optimize multi-year candidate-line construction and workload migration, subject to construction delays, renewable-energy curtailment, unserved data-center load, workload-balance requirements, communication-bandwidth constraints, and link-latency limits. The model is evaluated over an eight-year planning horizon on a modified IEEE 14-bus system. The results show that coordinating VTL with physical transmission expansion reduces total system cost, unserved data-center load, and renewable-energy curtailment while alleviating transmission congestion. VTL-aware planning also requires one fewer candidate line than build-only planning. However, VTL alone cannot meet system demand and does not eliminate the need for physical transmission reinforcement. Sensitivity analysis indicates that VTL provides the greatest value when transfer capacity is tight, construction costs are high, load growth is rapid, and data-center flexibility is greater. Additional validation on the IEEE 118-bus system over a five-year planning horizon shows that coordinated planning retains its cost and reliability benefits while reducing physical expansion requirements.

Simulation-based inference (SBI) enables Bayesian inference when a simulator is available but the likelihood is intractable. Modern SBI methods train neural networks on simulated data to obtain an amortized approximation of the posterior, making inference for new observations fast. A central challenge in SBI is to extract informative, low-dimensional summary statistics; without them, performance can degrade rapidly as the observation dimension grows. We therefore highlight summary learning as a critical component of SBI and propose a general framework that jointly learns a neural-network-parameterized summary and an amortized posterior estimator. We further introduce a local refinement strategy that adapts the amortized estimator to a specific observation, thereby reducing the amortization gap. Both amortized training and local refinement accommodate implicit posterior estimators without tractable densities. On the theoretical side, we analyze the finite-sample behavior of the joint learning procedure through the lens of nonparametric conditional distribution estimation, and show on representative examples how low-dimensional summaries can mitigate the classical curse of dimensionality. Empirically, our results support this motivation and demonstrate favorable performance relative to existing methods.

We study the relationship between strong data-processing inequalities for Rényi divergences and hypercontractivity. For Rényi divergences of order greater than one, we characterize the contraction of Rényi divergences through a property of the dual Markov operator that we name power contraction. Using the contraction coefficient, we derive an upper bound on rectangle probabilities in terms of their marginals that is at least as strong as the corresponding bound obtained via hypercontractivity. Combining this result with inversion of the binary Rényi divergence, we derive communication lower bounds for one-way communication-assisted agreement distillation. Examples involving binary symmetric and rare-event Z-channels illustrate the improvements over standard hypercontractive converses.

In this paper, we propose a linear and decoupled fully discrete scheme that preserves the entropy production property for the dynamic van der Waals model. This model provides a thermodynamically consistent diffuse-interface framework for describing non-isothermal phase transitions in single-component liquid-vapor systems, containing the coupled conservation laws of mass, momentum and energy. We first introduce an equivalent reformulation of the governing equations that avoids the use of chain-rule identities, which generally fail under discrete differentiation. Further, a component-wise convex-splitting method is employed to calculate the discrete chemical potential and temperature. This treatment provides a discrete entropy-stable inequality without a time step restriction. Based on the explicitly treated velocity, the scheme allows the three conservation equations to be solved linearly and sequentially in a decoupled manner. Since the momentum equation does not directly affect the entropy-stability analysis, we use a classical linear semi-implicit scheme to solve it. Numerical experiments verify the expected numerical accuracy. Simulations of droplet evaporation and two- and three-dimensional Leidenfrost effects further demonstrate the proposed method can capture complex non-isothermal phase-transition dynamics while numerically reproducing the discrete entropy production behavior.

Neural operators that learn physical laws from data should generalize beyond their training distribution and expose the learned law in closed form. The Kolmogorov-Arnold neural operator achieves both for position-dependent dynamics by learning a dense pseudo-differential symbol over all positions and frequencies, at a cost that grows with the square of the resolution and with closed forms that must be read out by hand. This dense symbol is unnecessary: a finite-order variable-coefficient differential operator has a symbol whose separation rank is bounded by the number of its coefficient functions, independently of the resolution, and the symbols that this network learns on its own benchmarks have rank two. We introduce the Kolmogorov-Arnold flow (KAFlow), which learns the generator of a physical operator as a low-rank symbol built from univariate Kolmogorov-Arnold functions, fixes the gauge of these functions so that every term has a unique place, integrates the dynamics with exact exponential integrators, and distills its closed form automatically. On the synthetic benchmarks of the Kolmogorov-Arnold neural operator, KAFlow lowers the out-of-distribution loss by 3.4 to 11 times and recovers every coefficient to within 10^-4 without symbolic templates, reaching the precision floor of the data. On long-horizon quantum dynamics, it lowers the state infidelity by up to 6 orders of magnitude, and by 30 times when trained only on position and momentum measurements. KAFlow also learns a variable-order fractional operator 15 times more accurately than existing operator networks and discovers nonlocal, fractional and non-polynomial dispersion relations on which sparse regression fails.

Frozen features of brain MRI foundation models lose accuracy on new datasets, a failure usually blamed on the features encoding where each scan came from. We show that the failure lies elsewhere. Across 17 public datasets and eight frozen representations, every representation recognizes the MRI sequence within each dataset, yet most fail on an unseen one. The cause is neither dataset encoding nor a shift of each dataset: the content itself is encoded along different directions in each dataset. This explains which datasets fail, for the sequence and for sex, and why removing the dataset signal with INLP, LEACE, CORAL or adversarial training never improves sequence transfer. Only the broadly pretrained 3DINO and SAM-Med3D encode the sequence consistently enough to transfer; the brain-specific BrainIAC and BrainFM do no better than an untrained network. Code is available at https://github.com/BrainFM/brainfm-profiler/tree/main/representation_utility.

Matter's first post-quantum change protects device attestation with ML-DSA, but its operational handshake, CASE, still derives session keys from P-256 ECDH, so recorded traffic can be decrypted once a quantum computer exists. We quantify the cost of moving CASE to ML-KEM and post-quantum signatures over Thread, where frames carry at most 127 bytes and a Matter message at most 1178 bytes. A byte-exact model of the CASE messages shows that a drop-in ML-KEM-768/ML-DSA-44 handshake grows from 1.6 kB to 22.8 kB and that no fully post-quantum configuration built from FIPS 203/204/205 algorithms fits the message limit. A Monte Carlo model of multi-hop Thread, calibrated to OpenThread and Matter source constants, shows that sending these messages as fragmented IPv6 datagrams causes spurious retransmissions whenever a message and its reply outlast Matter's retransmission timer. At three to four hops the drop-in handshake then takes 1.8 to 2.6 times longer than with a selective-repeat block transfer, with or without interference, and up to 33% of handshakes with sleepy devices fail; raising the timers does not remove the effect. Caching certificate chains and authenticating the device with a pre-distributed ML-KEM key reduce the handshake to 6.1 kB and 44 ms of device computation. As a first step, hybrid P-256+ML-KEM-512 key exchange with a cached device chain protects session keys against harvest-now, decrypt-later attacks while every message still fits the current limit.

Self-evolving agents are increasingly used to improve language model systems without human intervention through either weight updates or harness improvements. Existing approaches typically pursue these directions separately: some methods update model weights using self-generated experience, while others rewrite the harness that defines how an agent uses prompts, tools, context, and execution feedback. However, these approaches overlook the relationship between model weights and the agent harness. To connect these processes and ensure effective and smooth agent self-evolving, we propose Evo$^2$, a framework for weight-harness co-evolution. For harness updates, we design a unified dual-role pipeline that a single policy serves as both a Proposer, which writes harness edits, and a Reflector, which diagnoses their effects and guides revision. For weight updates, harness modifications and their execution outcomes guide the training of both the Proposer and the Reflector. We further develop EvoTIDE, a training pipeline that recovers usable trajectories through edit extraction and prefix continuation, then combines SFT with interleaved GRPO across the two roles. Extensive experiments across three coding benchmarks demonstrate that Evo$^2$ significantly outperforms existing self-evolution methods. It achieves 82.1% on Polyglot and 48.3% on SWE-bench Verified, compared with 62.6% and 41.0% for the strongest baseline, respectively. We also show that the evolved harnesses transfer to different tasks without further evolution and rank first in cross-benchmark evaluations. Our dual-role harness improvement framework is also able to extend to self-evolution with proprietary models, improving a GPT-5-mini agent's performance from 24.4% to 48.3%. These results demonstrate the applicability of the framework Evo$^2$ across domains and its broad transferability across models and tasks.

World Action Models (WAMs) inherit powerful priors from pretrained video generation models, but their variational autoencoder (VAE) representations are optimized for visual reconstruction rather than manipulation. External visual representations can improve manipulation, but existing methods couple representation choices with different integration mechanisms, obscuring their individual effects. We systematically compare visual representations, five integration mechanisms, and feature combinations under controlled settings. Our study yields three main findings. First, strong general-purpose visual representations provide broader robustness gains and can match or outperform representations pretrained for specific visual capabilities. Second, channel-wise joint prediction achieves the highest average success rate; further analyses suggest that future-feature prediction and the action expert's effective use of added visual information contribute to its gains. Third, additional sources should be selected for the task-relevant information they contribute beyond existing representations, as the same source can help one combination but harm another. However, even the best evaluated composition provides a smaller additional gain than the initial improvement from adding a strong general-purpose representation to the VAE-only baseline. We instantiate the resulting Representation Recipe as R$^2$-WAM. It achieves state-of-the-art performance on RoboCasa GR1 Tabletop with an 84.7% success rate and reaches 92.8% on RoboTwin 2.0, improving over the VAE-only baselines by 5.9 and 2.2 percentage points, respectively. Taken together, these findings suggest that future work should prioritize stronger general-purpose visual representations and more effective mechanisms for integrating them into WAMs, with feature composition serving as a supplementary route to improvement.

Video diffusion models can reconstruct visually compelling content from severely compressed representations at ultra-low bitrates, but they impose a rigid deployment trade-off. Bidirectional (BI) models achieve high quality through dense full-sequence attention, yet require the entire sequence before reconstruction and incur substantial latency. Autoregressive (AR) models enable low-delay streaming, but their restricted temporal context reduces compression efficiency. This tension is particularly relevant when a video stream must be decoded in real time yet is also stored for later, quality-oriented playback, making AR and BI models natural choices for the live and replay stages, respectively. To support both stages, we present UniGVC, a Unified Generative Video Codec that supports two decoding modes from a single bitstream. UniGVC-BI employs a BI-DiT with dense attention for quality-oriented random-access decoding, whereas UniGVC-AR uses windowed AR-DiT for low-delay streaming. Both paths share a hierarchical latent codec that recurrently produces the same conditions with ultra-low bitrates, allowing UniGVC-BI to serve as a conditional teacher for single-step distillation of UniGVC-AR. A tiny conditional video decoder further reduces streaming latency while preserving most of the reconstruction quality. On standard 1080p benchmarks, UniGVC-BI achieves the best overall rate-distortion-perception trade-off among the evaluated methods and produces temporally stable video at rates as low as 0.0005 bits per pixel. Meanwhile, UniGVC-AR remains competitive while decoding 1080p video at 36.85 FPS, over $250\times$ faster than UniGVC-BI. Code and models will be released later.