Papers

Performing measurements on gapless quantum ground states can give rise to new universal phenomena, whose origins remain poorly understood even for the paradigmatic Ising critical point. Replica methods offer a natural framework to study such post-measurement ensembles, with the replica limit $Q\to 1$ describing physical Born-averaged observables. Here we show that, for the monitored critical Ising chain, the replicated post-measurement ensemble of states itself undergoes a phase transition precisely at the physically relevant replica limit $Q \to 1$, preventing standard replica treatments. We use boundary conformal field theory (BCFT) methods to explain the underlying mechanism of this transition. Our methods yield a perturbative analytic calculation of the universal $O(1)$ constant in the Shannon entropy of the measurement record, previously known only numerically. We also apply our tools to study measurement-averaged spin correlators, providing a unified description of the universal properties of measurement outcomes and post-measurement states in the critical Ising chain.

We complete the proof that every compact 3-manifold group is linear. Specifically, we construct a faithful representation in GL5(Q[u]) for a finite-index subgroup of every closed non-geometric graph-manifold group.

We study the Euler-Maruyama approximation of the degenerate stochastic differential equation $dX_t = |X_t|^\alpha\,dW_t$, $0 < \alpha < 1/2$, $X_0 = 0$, for which uniqueness in law fails. The Euler scheme started exactly from zero is trapped by the absorbing solution, whereas a nonzero initialization $x_h \to 0$ can select a different weak solution. We identify the initialization-dependent limiting regimes. The critical spatial scale is $r_h = h^{1/[2(1-\alpha)]}$, and the relevant depth parameter is $\Theta_h = h\log(1+\log^+(r_h/|x_h|))$. If $\Theta_h \to 0$, the scheme converges to the non-sticky weak solution. If $\Theta_h \to \theta \in (0,\infty)$, it converges to the solution that stays at zero for the deterministic time $\theta/|\log\alpha|$ and then evolves as the non-sticky solution. If $\Theta_h \to \infty$, the scheme converges to the absorbing solution. The proof is based on an exact scale-invariant Euler chain, a logarithmic exit-time law of large numbers, a discrete Tanaka occupation estimate, and a compact-uniform restart argument. Every finite deterministic delay can be obtained by a suitable vanishing initialization, while oscillating depth parameters yield process-level nonconvergence. Thus a vanishing initialization can survive in the limit as a macroscopic deterministic delay and thereby act as a solution selector for the Euler-Maruyama approximation.

Variance reduction improves the convergence rate of stochastic non-convex optimization under the average smoothness condition. Existing adaptive variance reduction methods are especially designed for average smoothness and do not adapt to standard smoothness. In this paper, we study how one algorithm can adapt to these two settings and the noiseless setting without knowing the smoothness constants or the variance bound. Given the gradient budget $T$, we develop a two-phase method. The first phase compares two independent gradients and switches to normalized momentum when their difference exceeds a threshold. This phase provides both a candidate with standard smooth and noiseless guarantees and also ensures a bounded expected function gap for the second variance reduction phase. The resulting rates are $O(T^{-1/3})$ under average smoothness, $O(T^{-1/4})$ under standard smoothness, and $O(T^{-1/2})$ without noise. When $T$ is unknown, we propose a method with two paths that attain the corresponding rates with logarithmic factors in the stochastic settings. Here, we use a variance reduction path to deal with average smoothness and employ the AdaGrad-Norm for standard smoothness and noiseless setting. Neither method restarts from the initial point or requires a growing number of oracle calls in each round.

Let $D$ be an oriented graph. For $X \subseteq V(D)$, the inversion of $X$ in $D$ is the oriented graph that is obtained via changing the direction of all edges which have both endpoints in $X$. The inversion number of $D$, $\operatorname{inv}(D)$, is the minimum number of inversions necessary in order to remove all cycles from $D$. A dijoin $D_1 \to D_2$ is the oriented graph obtained by taking disjoint copies of $D_1,D_2$ and adding edges from all $v \in V(D_1)$ to each $u \in V(D_2)$. It was conjectured by Bang-Jensen, da Silva and Havet that $\operatorname{inv}(D_1 \to D_2) = \operatorname{inv}(D_1) + \operatorname{inv}(D_2)$ for all $D_1, D_2$. This was disproved by simultaneous constructions of various $D_1$, $D_2$ with $\operatorname{inv}(D_1 \to D_2) = \operatorname{inv}(D_1)+\operatorname{inv}(D_2)-1$. However, no known constructions achieved more than this gain of 1. Moreover, all of the constructions exploit a gap of 1 between $\operatorname{inv}(D)$ and the minimum rank of a family of related matrices, a gap that can never be more than 1, suggesting that perhaps something close to the original conjecture might be true. We prove that in fact additivity fails badly. In particular, for all $\varepsilon>0$ we show that there exist tournaments $D_1$, $D_2$ with arbitrarily large inversion numbers such that $\operatorname{inv}(D_1 \to D_2) \leq \left(\frac{1}{2} +\varepsilon\right)(\operatorname{inv}(D_1)+\operatorname{inv}(D_2))$.

Detecting communicative meaning in dialogue requires modeling not only lexical content but also the participant-role structures through which sentiment, emotion, politeness, and formality (SEPF) are expressed. Existing approaches rely on contextual language representations while leaving semantic roles implicit. We introduce a role-aware transformer architecture that maps the four SEPF dimensions to theory-grounded Semantic Role Labeling (SRL) configurations, incorporating these configurations directly into the transformer's self-attention mechanism. Role-aware attention models non-directional internal states (sentiment and emotion), while directional role-aware attention captures initiator-to-recipient relations for external cues (politeness and formality). The resulting role-aware representations are fused with contextual embeddings through a learned gate that adaptively weights how much role information each input contributes, rather than appending semantic roles as an auxiliary feature vector. Across four benchmark datasets, the framework yields statistically reliable gains over strong BERT-based baselines for emotion, politeness, and formality, with gains concentrated in theoretically predicted cases such as low-frequency, complex emotions and highly polite expressions; sentiment improvements are marginal and non-significant, suggesting that sentiment is conveyed more through lexical choice than participant-role structure. Additional evaluation on real-world data demonstrates that the framework improves both predictive performance and model traceability. These results demonstrate that linguistically grounded modifications to transformer self-attention improve the detection of communicative meaning, generalize beyond benchmark datasets to real-world mental-health data, and yield traceable, role-grounded explanations that extend beyond predictive performance to interpretability.

The motion of inertial particles suspended in a carrier fluid, together with the dynamics of the underlying vortical structures, is investigated in a square microfluidic T-junction. The coherent structures forming near the branching strongly affect the particle distribution and may promote particle trapping around the T-region. Single-phase numerical simulations of the unperturbed base flow are combined with integration of the particle force balance equation, which, unlike most previous studies, includes the Basset-Boussinesq (BB) history force. The flow topology is analysed for $Re=500$ and $570$, while particle properties are varied through the Stokes number $St$ and the particle-to-fluid density ratio $\rho$. At $Re=500$ the base flow exhibits a steady vortex pair, whereas at $Re=570$ a Hopf bifurcation gives rise to a helical oscillation of the primary vortex pair characterized by periodic breathing. The BB history force has a non-negligible impact over the whole range of $(\rho, St)$ pairs investigated. It acts as an additional viscous force which, together with the drag of the relative motion, opposes the pressure-gradient force, widening particle spiral trajectories radially and elongating them axially. Its inclusion reduces the basin of attraction, indicating that memory effects promote particle displacement away from the vortex core and weaken trapping. At $Re=570$, spectral proper orthogonal decomposition characterizes the breathing of the primary vortex pair. Particles injected at a given location interact with the vortical structure at different phases of the cycle depending on the injection time, producing substantially different trajectories. This phase dependence is reflected in the shape and extent of the basin of attraction, quantified through the trapping probability over the inlet section and the probability density functions of the force components.

Dopants can tune the activity of zirconia-based CO$_2$ hydrogenation catalysts through changes in adsorbate and transition-state (TS) energetics. Computing these quantities with density functional theory (DFT) becomes expensive when several dopants, reaction intermediates, and TSs must be screened. Here, we further develop the graph-based machine learning method Wasserstein Weisfeiler-Lehman Gaussian Process Regression (WWL-GPR): we introduce additional electronic-structure and local-environment features relevant for oxide surfaces and use the resulting model to predict adsorption and TS energies on ZrO$_2$(101) surfaces doped with Al$^{3+}$, Cd$^{2+}$, Ga$^{3+}$, In$^{3+}$, and Zn$^{2+}$. Adsorbate-surface structures are represented as attributed graphs that combine bonds connectivity and features. We benchmark WWL-GPR against thermodynamic scaling relations (TSR) for reaction intermediates and Bronsted-Evans-Polanyi relations (BEP) for TSs, and then propagate the predicted energies to microkinetic CH$_3$OH-yield estimates. WWL-GPR consistently improves the prediction of reaction intermediate energetics relative to linear scaling approaches and provides TS estimates that are competitive with BEP-based models. These improvements become especially relevant when the predicted energetics are propagated through the reaction network. Microkinetic calculations show that WWL-GPR-derived energies give CH$_3$OH yields closer to DFT-derived yields than the linear-relation workflow. Degree-of-rate-control analysis links the activity errors to dopant-dependent rate-controlling steps, including HCOO* formation, H$_2$COOH* dissociation, and H$_3$CO* formation. These results support WWL-GPR as a data-efficient surrogate for connecting atomistic energetics with catalyst-screening observables on doped oxide surfaces.

Discourse particles subtly shape the flow of discourse by reflecting conversational context and interlocutor attitudes, often leaving the truth-conditional meaning untouched. They are also often polyfunctional, with distinct senses which partially overlap with other particles; for example, the exclusive use of English "just" and "only" ("John only/just drinks water"). Despite the frequency of these words in natural language, they have gone largely unstudied due to the subtle nature of their effects. We investigate LLM sensitivity to "just" and "only" in natural conversations using behavioral and mechanistic analyses. First, we find that LLMs and humans similarly prefer the presence of these particles in context, and both are more sensitive to "just" compared to "only". Second, when sampling from LLMs with and without these particles, we find LLMs consistently rate generations as more likely, given their source context, demonstrating their particle sensitivity. In both experiments, instruction tuning increases sensitivity. Finally, we localize particle-sensitive units within LLMs and find the overlap between "just" and "only" is asymmetric: "only" units transfer selectively to exclusive "just", while all "just" units transfer to "only" equally well regardless of sense, suggesting LLM lexical-level particle representations are partially organized by sense. We release our code, sense-labeled dialogue stimuli, and human preference annotations at https://github.com/JiahuiKChen/Not_Just_an_Evaluation.

We consider an inexact version of the Schwarz domain decomposition method in $3$D to solve a Laplace problem in a domain that is the union of two overlapping spheres. We solve each subproblem approximately by projecting its boundary data $L^2$-orthogonally onto the space of spherical harmonics of degree at most $N$. The model problem arises from the ddCOSMO solver used for implicit solvation in computational chemistry. We analyze the Schwarz-spherical domain decomposition method by establishing a new variant of the maximum principle.

We show that certain classes of unital, simple, separable C*-algebras with comparison are singly generated, including tracially AF algebras and monotracial crossed products of actions on the Cantor space. In particular, we show that simple, separable, tracially AF algebras have generator rank one.

We study the spectrum of two-particle bound states of Kähler-Dirac fermions coupled via gravity in Euclidean Dynamical Triangulations (EDT). We work in the quenched approximation, where the geometry is allowed to fluctuate, but there is no back-reaction from the matter on the geometry. The Kähler-Dirac action respects an exact $U(1)$ symmetry that is related to continuum chiral symmetry, and this $U(1)$ symmetry ensures that some of the bound-state masses vanish in the chiral limit of vanishing fermion bare mass. The bound-state masses that are not protected by the $U(1)$ symmetry do not vanish in the chiral limit, but the scale associated with the symmetry breaking does not appear to be a physical scale like the Planck mass or the curvature scale of the emergent de Sitter background. Rather, the mass scale associated with this non-degeneracy appears to be a positive power of the lattice spacing, decreasing as the continuum limit is approached. This suggests that Kähler-Dirac fermions in EDT can be interpreted as the gravity analog of staggered fermions in flat-space lattice field theories, with similar behavior.

How well can vision-language models (VLMs) reason about what is missing from an image? We study this question with OccludeBench, a manually curated benchmark for recovering fully occluded objects from the visual evidence that remains. OccludeBench distinguishes direct cases, where the hidden object must be recovered from a specific physical cue such as a shadow or reflection, from inferential cases, where it can be inferred from broader scene context. Across five VLMs, accuracy averages 42.7% on direct cases versus 62.7% on inferential cases, exhibiting a 20.0-point gap. Humans attain higher accuracies of 63.4% versus 68.5% for direct and inferential images, respectively, with no significant gap. To locate this failure, we introduce OccludeTrace, a four-stage diagnostic that isolates where the visual reasoning process breaks down. The bottleneck emerges when models must recover the relevant object from the projection within the full image: they succeed on only 50% of failed examples, compared with 91% of examples they answer correctly. Extended reasoning, tool use, and few-shot prompting yield inconsistent improvements and do not reliably eliminate the gap. In contrast, an inpaint-then-confront intervention, which prompts models to first generate a candidate hidden object and then evaluate that hypothesis against the visible physical cue, reduces the gap to a statistically non-significant difference for both models with image-generation capabilities. Together, our results establish a tractable, controlled diagnostic for isolating how VLMs reason about visual content that is missing from an image.

Momentum methods such as STORM and second-order momentum track the gradient by adding a transport correction to momentum, at the cost of an extra oracle call per iteration. We introduce randomized momentum transport, which evaluates the correction with probability $q$ and otherwise uses a cheap prediction, with importance weighting that preserves the conditional mean and adds $(q^{-1}-1)$ times the prediction loss to the variance. Under transport-bias, second-moment, and residual-control assumptions, linear minimization oracle (LMO) updates, including normalized, sign, and exact spectral (Muon) directions, retain an $O(T^{-1/3})$ stationarity bound for each fixed $q>0$ with a single-gradient initialization. For unnormalized STORM, some $q<1$ lowers the leading coefficient of our sufficient gradient budget if and only if the gradient-difference noise modulus exceeds the smoothness constant, by at most about $10\%$. An online diagonal predictor admits a regret-to-convergence guarantee. At approximately equal oracle cost, querying a quarter of the corrections lowers the final training loss by $18$ to $43\%$ on MNIST1D and raises the test accuracy of Hessian-vector transport on CIFAR-10 by $1.5$ points.

LiDAR-inertial odometry (LIO) is widely used for real-time localization and mapping, but its reliability can degrade substantially in subterranean environments affected by smoke, dust, and other airborne obscurants, whereas radar remains comparatively robust. This article presents a one-class temporal learning framework for LiDAR degradation detection and reliability-aware localization under such conditions. A broad set of degradation-sensitive cues, including spatial, cross-modal, geometric/motion, intensity, and measurement-recency features, is evaluated to characterize changes in LiDAR sensing quality. Their temporal evolution is modelled over sliding windows using an LSTM autoencoder trained exclusively on nominal operation, enabling deviations from healthy sensor behaviour to be detected without requiring labelled degradation data. The resulting LiDAR reliability estimate governs adaptive selection between LIO and radar inertial odometry (RIO), retaining LIO during reliable sensing and transferring localization authority to RIO when degradation is detected. The framework is evaluated using real-world ground robot data collected in GNSS-denied subterranean tunnels under nominal and dense-smoke conditions. Experimental results demonstrate degradation detection, adaptive modality switching, and continued localization output under severe LiDAR degradation. A feature family ablation study further shows that a reduced feature representation can improve degradation-aware localization performance in the evaluated dense-smoke scenarios, highlighting the importance of feature parsimony in one-class sensor degradation detection. The code will be released as open source to support reproducibility and further research.

We present new algorithms for stochastic optimization problems in which the objective is neither smooth nor convex, subject to a convex constraint set. Our algorithms return a $(\delta,\varepsilon)$-Goldstein stationary point, as recently defined for constrained problems by Liu et al. (2024), with improved complexities, and are the first to achieve optimal convergence rates for this task in several settings. In particular, we design projection-based and projection-free algorithms that achieve a $\widetilde O(\delta^{-1}\varepsilon^{-3})$ rate using stochastic gradients, which we show to be optimal, and a $\widetilde O(d\delta^{-1}\varepsilon^{-3})$ rate using zeroth-order queries, which is also optimal in its dependence on the dimension $d$ and improves over prior work. Notably, our lower bound is given in terms of the Frank-Wolfe gap and holds even for smooth problems, which is of independent interest. These results show that in nonconvex stochastic optimization, the convergence rate is degraded neither by the presence of constraints nor by nonsmoothness. Our analysis is based on a reduction to online linear optimization, extending a related technique from unconstrained optimization.

Discrete graph diffusion models (GDMs) corrupt the graph structure by adding noise independently to nodes and edges under a Markovian noise model. In the unconditional setting without node features, an equivariant denoiser trained on a single graph is able to generate novel graphs and sample-level metrics such as edge overlap and WL kernel similarity find no evidence of memorization. In this work, we challenge this view and show for the first time that graph diffusion models indeed exhibit memorization. Analyzing the reverse diffusion trajectories of GDMs trained on two-block stochastic block models (SBMs), we find a strong train-data bias: the model recovers a corrupted training graph almost perfectly compared to a held-out test graph. We term this conditional memorization, as it surfaces only when the reverse trajectory is conditioned on a corrupted training graph and is not apparent during inference. It coincides with a growing generalization gap, where the training loss improves as the loss on held-out graphs degrades, and it cannot be mitigated by early stopping. The onset of memorization always precedes the emergence of community structure in the generated graphs, thus every checkpoint earlier produces structureless graphs. This behavior holds across all node sizes we test.

Aggregate linear demand systems can be micro-founded by a population of heterogeneous consumers making discrete choices. They can also be extended to incorporate income effects. I prove these results and show how to structurally estimate this class of models using only market-level data. The distribution of consumer preferences is recovered by disciplining the corner solutions with aggregate terms only. This approach allows for the same richness of Berry, Levinsohn, and Pakes (1995) [BLP] while retaining the analytical and computational advantages of linear demand: it does not require exogenously fixing market size, allows for heterogeneity in choice sets, and results in a flexible substitution matrix which varies endogenously as different consumers enter/exit the market. The discrete-choice microfoundation facilitates welfare analysis and the flexible measurement of markups an cost pass-throughs, among a wide range of other applications.

Dissipative state preparation engineers quantum many-body states as steady states of an open-system dynamics. The spatial range of the required dissipation sets both the feasibility of analog implementations and the connectivity and control complexity of digital ones. Yet how this locality quantitatively constrains preparation times remains largely unknown. Here, we establish an exact locality-preparation time relation for quadratic fermionic systems and develop a construction that optimizes the locality of the dissipative dynamics while preserving the target state. Applied to the Kitaev chain, our framework reveals sharply different locality-preparation time tradeoffs in critical and gapped regimes. At criticality, suppressing the algebraic tails of the jump operators necessarily forces the preparation time to grow algebraically with system size. By contrast, both trivial and topological gapped phases admit exponentially localized dissipation with a finite preparation time in the thermodynamic limit. Our results identify when fast dissipative preparation is compatible with local control and when criticality imposes an unavoidable dynamical cost.

Memes play a central role in modern online communication, especially in contexts of political conflict where visual and textual cues combine to convey persuasive messages. We study persuasive technique detection in Ukrainian memes using vision-language models (VLMs). We introduce UkrMeme, a dataset of 200 organically produced Ukrainian wartime memes annotated for persuasion techniques, and PTM-UA, a Ukrainian translation of a subset of the Propaganda 950 benchmark that preserves original visual content while translating meme text. We evaluate VLMs in a zero-shot prompting setup across image, text, and multimodal inputs using hierarchical F1 metrics. Finally, we compare Layer-wise Relevance Propagation (LRP) and model-generated token attributions and analyze the semantic categories of evidence emphasized by each explanation method.

We establish that bond Hamiltonian densities, up to local difference terms whose sum vanishes under periodic boundary conditions, lying in the commutant of bond operators, $w_j\pm w_{j+1}$, is a necessary and sufficient condition for a one-dimensional nearest-neighbour Hamiltonian to conserve a uniform or staggered sum of the one-site operators $w_j$. Starting from these local commutants, we show that every such Hamiltonian and its conserved charge can be obtained from monodromy matrices constructed using regular and non-regular members of scalene Yang-Baxter triples. The results hold on periodic chains of length $L\geq3$, with even length required in the staggered case. The results do not require translation invariance, and hence the interaction on each bond can be chosen independently from the full corresponding commutant. Our representation-independent construction applies in arbitrary finite local dimension and exhausts the Hamiltonians with either prescribed additive symmetry, without requiring ordinary Yang-Baxter integrability. We illustrate the construction with spin-$\tfrac12$ and spin-1 systems and discuss its relation to the double-commutant theorem. Classifying the one-site generators $w$ and their associated bond commutants provides a route to classifying such Hamiltonians with additive conserved charges.

Driving world action models (WAMs) learn from paired trajectories and future observations. Post-training with reinforcement learning allows the policy to explore alternative trajectories and improve planning using driving rewards. However, driving logs contain future videos only for the recorded trajectories, leaving these alternatives without corresponding visual supervision. Thus, we introduce SimVTA, a simulation-grounded post-training framework that combines trajectory optimization with video supervision. First, we sample candidate trajectories for each observation and use a real-world simulator (e.g., World Engine) to render future videos for high-reward candidates in real-world scenes reconstructed with 3D Gaussian Splatting (3DGS). This provides trajectory-matched supervision independent of the model's own predictions. Under a limited rendering budget, we prioritize video supervision for higher-error futures among these rendered candidates. Finally, we optimize trajectory planning with Group Relative Policy Optimization (GRPO) and use a flow-matching loss on the matched videos to update shared representations. At inference time, the model predicts trajectories directly, without simulator calls. On NAVSIM v1, applying SimVTA to a WAM initialization improves PDMS from 86.2 to 91.3 (+5.1). It achieves 88.2 EPDMS on NAVSIM v2 and an average HD-Score of 38.8 in closed-loop evaluation on HUGSIM. Ablations show that updating shared representations with video supervision improves planning, while prioritizing higher-error futures among high-reward trajectories further improves performance.

Dead cell detection in fluorescence microscopy is essential for phenotypic screening and drug efficacy assessment. However, automated identification remains challenging due to signal variability, optical artifacts, and colocalization ambiguities across multiple channels. Markers that define nuclear boundaries do not neatly coincide spatially or temporally with apoptotic markers. Furthermore, marker accumulation kinetics vary across individual cells and apoptosis stages, making universal thresholds unreliable. We present a systematic comparison of several computational approaches for detecting apoptotic cells in nucleus-stained specimens: (1) heuristic methods using intensity thresholds, distributions, correlations, and domain guidance; (2) classification using numerical features (Logistic Regression, Random Forest, XGBoost and LightGBM) and image features (ResNet-18 and EfficientNet-B0); and (3) anomaly detection, also using numerical features (Isolation Forest, one-class SVM and PCA reconstruction) and image features (convolutional autoencoders and Deep SVDD). Using a dataset of 258 images containing 65,527 cells/nuclei across multiple scenes with 243 manually annotated ground truth dead cells, we evaluate each method on metrics such as PR-AUC, Precision, Recall, F1-score, and macro-F1. Our results indicate that classification using image features (deep learning) and numerical features (traditional) achieved the best performance, with ResNet-18 and XGBoost significantly outperforming manually designed heuristics and anomaly detection methods. Our findings demonstrate that such biomedical problems remain highly challenging to solve even with the state-of-the-art computational and learning techniques. We aim to provide an effective decision-support tool to researchers while tackling such problems either fully-automatically or in a human-in-the-loop setting as an aide to domain scientists.

BARseq enables spatially resolved gene-expression profiling across intact tissue, but its sensitivity depends on accurate detection and decoding of dense fluorescent rolonies. Existing spot-calling approaches can be conservative where signals overlap, while supervised deep learning is limited by the lack of reliable manual ground truth in real tissue images. Here we present PhySeqNet, a physics-informed deep learning framework for annotation-free fluorescent foci detection in BARseq. PhySeqNet trains U-Net detectors entirely on semi-synthetic images generated from an instrument-matched optical model combined with real sample-free experimental noise, and is integrated into pyBARseq, a modular Python pipeline for BARseq analysis. PhySeqNet achieved high detection performance on held-out simulated data across wavelength bands. On real tissue images, where ground truth is unavailable, we evaluated detections by forward-model reconstruction: PhySeqNet with conservative barcode-aware non-maxima suppression (NMSB) achieved higher reconstruction fidelity (NCC and MS-SSIM) than the BarDensr single-shot estimator, tuned classical spot detectors, and a pretrained deep-learning detector. In the analyzed brain section, PhySeqNet with NMSB yielded a median of 51 counts per cell versus 47 for BarDensr, with a similar number of genes detected per cell. The recovered transcripts produced spatial expression patterns qualitatively consistent with Allen MERFISH reference data and preserved cell-type composition after scANVI label transfer. These results show that physics-informed simulation can train practical, annotation-free detectors that improve BARseq sensitivity without disrupting downstream biological interpretation.

State-space models (SSMs) maintain a fixed-size recurrent state, but reading and writing it at every decoding step can bottleneck hybrid models. This eager state management causes large state I/O with low arithmetic intensity, leaves speculative decoding no cheap rollback, and feeds state quantization error into subsequent recurrent updates. We introduce ReplaySSM, which unrolls the recurrence to represent the state as a checkpoint and a buffer of recent SSM inputs. This separates output computation from checkpoint updates. By updating the checkpoint periodically, ReplaySSM avoids state writes on most steps and roughly halves the dominant state memory traffic. Keeping recent inputs also makes speculative rollback a buffer-pointer operation and enables output-only verification without materializing intermediate states. The same separation allows a low-precision checkpoint to serve output computation on most steps while a higher-precision checkpoint serves recurrent updates, so rounding error in the low-precision copy never enters later updates. We implement ReplaySSM in vLLM for Mamba-2 and Gated DeltaNet. At serving batch size 256, it accelerates standard decoding by $1.20$–$1.47\times$ over a tuned vLLM baseline across four hybrid models, and delivers about $1.8\times$ the throughput of vLLM's speculative decoding on the two 120B-class models. Separate state precisions also reduce output KL divergence and downstream accuracy losses compared with a shared low-precision checkpoint.

We investigate the local gravitational source and charges obtained by introducing translations as an internal Stückelberg redundancy in Weyl gauge theory. The internal-$T(4)$-invariant dressed coframe identifies the source of the translational potential with the existing coframe variational derivative, without introducing new dynamics. We derive an excitation/source equation and balance law whose structure is independent of the constitutive parameters of the parity-even quadratic torsion family. The gravitational self-current, defined relative to a specified gravitational/matter split and boundary representative, gives a density invariant under the internal gauge factors once its frame index is contracted with the coframe; this invariance neither establishes ordinary conservation nor identifies the density with the external translation Noether current. A surface charge follows by contraction with a covariantly constant internal direction, and we separately derive the three-sector Noether superpotential of generalized reducibility, with the asymptotic conditions for conservation of its flux. At the teleparallel equivalent of general relativity (TEGR) the excitation reduces to the teleparallel superpotential and the covariant coframe–connection pair gives the Schwarzschild ADM energy; flat cosmology and transverse-traceless waves give further checks. Away from TEGR, an exact solution of New General Relativity gives a charge whose fractional difference from the Kepler mass is half the deviation of the post-Newtonian spatial-curvature parameter from unity, at the adopted normalisation; independent Hamiltonian charge matching remains open. Finally, the integrable Hamiltonian charge is invariant under covariant Lagrangian boundary terms at fixed reference normalisation and presymplectic representative.

Selective state-space models are an efficient approach to long-sequence video modeling. However, their recurrent computation makes it difficult to determine which video tokens drive a prediction. Existing attribution methods often require additional backward passes or repeated model evaluations, making them substantially more expensive than the forward inference they aim to explain. In this work, we introduce READ, a parameter-free attribution method for selective state-space models that unrolls the recurrence at the position read by the classifier. READ traces the contribution accumulated in the recurrent state back to the video tokens that contribute to the final prediction. By anchoring attribution at the classifier's readout, it produces a single spatiotemporal relevance map from which frame- and region-level relevance can be obtained directly. Despite requiring only one forward pass, READ achieves competitive faithfulness and localisation compared with substantially more expensive attribution methods. Experimental results show that READ avoids the systematic relevance biases of existing methods while recovering meaningful temporal and class-specific structure across datasets and backbones. More broadly, our results show that the recurrence underlying selective state-space models is not only an efficient mechanism for inference, but also a direct source of interpretable evidence for their predictions.

Pandora's problem is a fundamental model of sequential search in which opening each box incurs a cost. We study the variant with deadlines, where each box has a publicly known deadline for both opening and selection. We prove that a simple threshold strategy achieves a $1/2$-approximation to the optimal expected utility, improving the previous guarantee arbitrarily close to $1/4$ due to Amanatidis et al. (2025).

The brightest 6.7 GHz methanol maser in the Galaxy is associated with the high-mass star-forming region (HMSFR) G009.62+0.20E. Additionally, this methanol maser shows periodic flaring observed at 6.7 and 12.2 GHz. Several models have been hypothesized to explain the periodicity, including periodic accretion, spiral shocks in the gaps of circumbinary disks, pulsating stars, dicke's superradiance, and a colliding wind binary (CWB) system. We analyzed the time series of the 1.22 km s-1 feature of the 6.7 GHz methanol maser spectrum and the 1.25 km s-1 feature of the 12.2 GHz methanol maser spectrum over the monitoring duration of roughly 20 years, to determine an updated period, analyze the stability and repeatability of the flares, the temporal behavior of the flare profile, and the duty cycle of the flare profile based on the updated period. From these results, it is argued for or against any of the mechanisms that have been proposed to explain the periodicity. Lomb-Scargle (LS) and phase dispersion minimization (PDM) methods were used to determine the methanol maser period at 6.7 and 12.2 GHz. Once the period has been determined, phase-folding or period-folding was used to analyze the flare shape, flare stability, and duty cycle of the flares. The periods of the 6.7 and 12.2 GHz methanol masers were determined to be ~ 243.0 +- 0.3 days. The flares are highly repeatable, stable, and show almost identical shapes, as well as their duty cycles. A replicated kernel density estimator (KDE) of the flare shape reproduces the time series very well over a time span of ~ 8000 days. After careful consideration of the proposed mechanisms for periodic methanol masers, we conclude that all but one, the CWB model, have been ruled out as a possible driver of this flare profile. It has also been established that the CWB model can qualitatively explain the flare profile of the periodic methanol masers.

Multiple-gradient descent is typically formulated in Euclidean geometry, whereas matrix-aware optimizers use directions induced by other norms. We analyze a single-loop multi-objective method combining a norm-constrained linear minimization oracle with entropic updates of task weights. For any finite-dimensional norm pair, we establish an $O(T^{-1/2})$ ergodic stationarity rate in the deterministic setting, deriving the required trajectory gradient bound from smoothness and lower-boundedness in the unregularized case. With stochastic gradients and a bounded-gradient assumption, one persistent blended-gradient momentum buffer yields an $O(T^{-1/4})$ rate whose leading term has no explicit dependence on the number of tasks. We also establish static regret for the inner weighting objective on arbitrary bounded gradient sequences and, with entropy regularization and a weight floor, iteration-wise inner-gap control after burn-in while retaining ergodic stationarity, conditional on a trajectory gradient bound. The optimal blended gradient can be nonunique in spectral and sign geometries, motivating a selection-independent inner gap. Our analysis extends the spectral construction studied by concurrent MOON to general norms. Experiments compare the method with existing multi-task optimizers and examine its oracle, weighting, and momentum choices.