Papers

We investigate the microwave response of sputtered Nb films as a function of biaxial film stress. Although transport measurements show nearly identical superconducting transition temperatures ($9.0\pm 0.2$K) and residual resistance ratios across the stress series, microwave resonators fabricated from highly compressive films exhibit enhanced temperature-dependent loss and larger frequency shifts. A conventional single-phase Mattis–Bardeen description yields an effective critical temperature of $2.4\pm0.2$K, approximately a factor of four below the value obtained from transport measurements. We show that the data are instead consistent with the presence of an additional low-gap component whose contribution increases with compressive stress. These results show that conventional transport measurements can overlook stress-dependent low-energy excitations that strongly influence microwave performance.

We completely characterize when a general curve of genus $g$ admits a non-degenerate degree $d$ map to projective space $\mathbb{P}^r$, that is $k$-secant along a linear space $\mathbb{P}^s \subset \mathbb{P}^r$ and deforms in a smooth family of expected dimension. This gives an optimal improvement of a theorem of Farkas, who gave necessary conditions, and proved sufficiency under additional numerical assumptions. Our main theorem shows that the natural necessary conditions are in fact sufficient, yielding a generalization of the classical Brill-Noether existence theorem. The proof relies on deformation theory of stable maps, and is valid in arbitrary characteristic.

In the welded tree problem, an algorithm is tasked with navigating a graph formed from two binary trees joined at the leaves through a “weld” of connecting edges. A quantum walk can navigate from root to root exponentially faster than any classical algorithm. However, known efficient quantum algorithms cannot find a path between the roots, as recording the path destroys constructive interference and thus the speedup. We prove that this is inherent: any quantum algorithm needs exponentially many queries to find a path between the roots of an independently matched welded tree graph. This provides an example of a problem that a quantum computer can solve exponentially faster than any classical algorithm by exploring exponentially many paths in superposition, but where it is provably intractable to find any such path. The proof uses compressed permutation oracles to record the progress of a quantum algorithm as it queries the graph. We show that the compressed database remains path-free up to a small error. By controlling such errors and bounding the progress of the algorithm with each compressed oracle query, we show that $\Omega(2^{n/12})$ queries are required to find a path in a height-$n$ tree with constant success probability.

Software vulnerabilities are often discovered long after they are introduced, making it difficult to identify the vulnerability-inducing commit (VIC) responsible for introducing the underlying vulnerable condition. Existing VIC identification techniques largely rely on git blame to trace vulnerable code through revision history and use positional heuristics, such as selecting its earliest or most recent modification. However, the true VIC may occur anywhere within this history, and vulnerable behavior may depend on code that evolves across multiple revisions. We therefore argue that VIC identification requires reasoning about how vulnerability-relevant code evolves, rather than simply where a candidate commit appears in the revision history. We present TraceVIC, a temporal graph-based approach for identifying and ranking VICs by reasoning over code evolution. TraceVIC first localizes likely root-cause lines and traces their histories across revisions, constructing graph representations that capture program structure within each revision and the evolution of vulnerability-relevant code across the history. It reasons over the resulting revision history, using temporal edges to preserve correspondences between program elements across consecutive revisions, and directly ranks candidate commits according to their contribution to the vulnerable condition. Ablation results show that modeling the full revision history improves F2 from 0.637 to 0.814. TraceVIC improves F2 by up to 28.7% over state-of-the-art methods and identifies a valid VIC for 78 of 79 vulnerabilities across four unseen C/C++ projects.

The Hirsch index (introduced in 2005) has been a very successful scientometric author-level measure for the publication impact inequalities among the authors. However, it does not have any upper bound and statistically it grows in proportion to the square root of the number of publications by the author. Social scientists, on the other hand, measure the economic performance of different countries (having different wealth amounts and populations etc.) using inequality indices like the Gini index (since 1912) etc., which are normalized quantities (independent of the country's total wealth or the population size). We show here that a normalized fractional Hirsch index, giving the fraction of total citations earned by fraction of those successful papers published by the author, can even indicate that typically about $14\%$ of the top cited papers earn about $86\%$ of citations for very successful scientists (like the 95 Nobel Laureates considered here). This is similar to Pareto's 80-20 law (since 1896) and agrees more precisely with those observed for self-organized sandpile models, where typically $86\%$ of the avalanche masses are dissipated through $14\%$ of the largest avalanches as the piles approach their respective self-organized critical points.

Deterministic electrical switching of the Néel order remains challenging in centrosymmetric altermagnets, where linear current-induced spin polarization is forbidden by inversion symmetry. Here, we establish a nonlinear route to Néel order control. Spin group symmetry shows that, in the nonrelativistic limit, the nonlinear spin polarization is constrained along the Néel vector, with a staggered response allowed in $8$ of the $10$ nontrivial spin-Laue groups. Although this Néel spin polarization is torque-inactive by itself, finite spin-orbit coupling can generate an additional uniform component transverse to the Néel vector. Using a minimal model for the $d$-wave altermagnet FeSb$_2$, we demonstrate the coexistence of the nonrelativistic staggered response with an SOC-induced uniform transverse component. Landau-Lifshitz-Gilbert equation-based macro-spin simulations reveal that the transverse component alone produces oscillatory dynamics, whereas its cooperative action with the staggered response enables deterministic $180^\circ$ reversal. We further show that changing the electric field orientation reverses the staggered spin response and selects the switching direction, while the opposite Néel states can be distinguished by their anomalous Hall response. Our results establish a nonlinear electrical write-read scheme for centrosymmetric altermagnets.

Quantization-aware distillation (QAD) restores much of the short-form question-answering performance lost to sub-3-bit quantization, yet leaves mathematical and code reasoning substantially impaired. Long generations often degenerate into repetitive loops, exhausting the decoding budget without completing a solution. We trace this gap to quantization-amplified exposure bias: QAD trains on fixed corpus prefixes, while quantization-induced deviations compound along the model's own autoregressive trajectories. To address this mismatch, we introduce an on-policy distillation (OPD) stage that places teacher supervision where the quantized model actually goes. Starting from a QAD checkpoint, the student generates through the quantized forward path used at deployment and receives feedback from a frozen full-precision teacher on its own prefixes, combining dense token-level guidance with task-verifier rewards. Across four models at 2.79 and 1.88 effective bits, OPD raises average BF16 performance retention from 35% to 70% on MATH-500 and from 66% to 91% on HumanEval while preserving short-form performance, with reasoning gains substantially exceeding those of continued teacher-forced QAD in matched-budget comparisons. By coupling QAD's stable low-bit initialization with OPD's on-policy reasoning recovery, our framework provides a comprehensive sub-3-bit solution that preserves broad capabilities while restoring long-form reasoning.

Offline reinforcement learning and off-policy evaluation evaluate dynamic decision rules based on historical data, without the ability to deploy online. In recent artificial intelligence (AI) applications, simple state and reward information is recorded as complex text or image, expert annotation may be available but at a higher cost. Recent AI and machine learning (ML) predictors can label data cheaply but with unknown bias. We show how a limited budget for ground-truth data-annotation can safely make use of plentiful ML predictions in a new estimator for doubly-robust off-policy evaluation with missing rewards. Our method directs where data should be collected by optimizing annotation probabilities for the best downstream estimation variance, which we implement in a feasible batch-adaptive approach. Our work is motivated by a collaboration with a homelessness services nonprofit that writes casenotes for individuals over time. Our method can be used to unlock trustworthy inference from casenote data and answer new inferential questions such as: how does expanding outreach effort over time affect progress towards a housing application and improvement in housing placement? In simulations and on two real datasets – casenotes from the nonprofit and human-preference votes from LMArena – we see reductions in RMSE of 37-65% for housing placement and 21-67% for progress towards a housing application at budgets of 50% of full annotation and above, and by 55-62% at every budget on LMArena under an equal-exposure policy.

Alpöge's construction produces a period family of compact complex threefolds fibred over $\mathbb{P}^1$, with complex two-tori as general fibres. We give a direct proof that every admissible member is Oka, using only the complex-analytic construction. Du–Guo–Kusakabe–Xie independently established the Oka property and determined the exact biholomorphism classes in this family. Assuming Alpöge's identification of the underlying smooth manifolds with $S^6$, these results yield uncountably many pairwise nonbiholomorphic Oka complex structures on the six-sphere.

A fundamental question in physics is: When does classical behavior emerge from quantum systems? Bosonic Gaussian states provide a natural setting to explore this quantum-classical boundary, as they capture both the classical field behavior and the intrinsic quantum nature of light. Here, we address this problem from a learning-theoretic perspective by asking: When are bosonic Gaussian states classical to learn? That is, under what conditions (if any) can an n-mode bosonic Gaussian state be learned with as few samples, and with operations as simple, as are needed to learn a classical 2n-variate Gaussian distribution? We establish a smooth crossover in learnability governed by the state's thermal fluctuations: - Cold Gaussian states are non-classical to learn: When the covariance matrix satisfies $\Sigma\le(\frac12+O(\frac1n))I$, i.e. close to the vacuum covariance, tomography under single-copy (i.e., non-entangled) measurements fundamentally requires $\Omega(n^3)$ copies, strictly exceeding the sample complexity $\Theta(n^2)$ of learning classical Gaussian distributions. We show that this hardness persists even when few-copy entangled measurements are allowed. - Warm Gaussian states are classical to learn: When thermal fluctuations exceed the vacuum noise, parameterized by $\Sigma\ge(\frac12+\nu)I$ for any parameter $\nu>0$, we prove that single-copy tomography requires $N=\Theta\left(n^2\min(n,1+\nu^{-1})\right)$ copies. This bound is tight and is achieved by simple, non-adaptive, unentangled heterodyne measurements. Crucially, for $\nu=\Omega(1)$, the sample complexity drops to $\Theta(n^2)$, matching the classical case. Our results tightly characterize a quantum-to-classical crossover in the learnability of bosonic Gaussian states, reveal a novel connection between fundamental physics and statistical learning theory, and have implications for real-world sensing experiments.

Large language models increasingly tackle hard reasoning problems by spending more test-time compute, yet the dominant strategy remains naive repeated sampling: draw many independent solutions and hope one is correct. Because such sampling explores only through local decoding noise, it tends to produce many near duplicate attempts rather than genuinely different ideas. We ask whether exploration can instead be steered at a semantic level, by first sampling problem specific concepts, hints, or strategies and then conditioning answer generation on them. We refine this into a simple, more exploratory procedure that emits many diverse concepts in a single trajectory, and evaluate it on hard problems where repeated sampling struggles. We then go a step further and make concept generation trainable: a small concept generator is optimized with reinforcement learning so that its concepts maximize the downstream success of a larger, frozen answer generator. On hard mathematical reasoning problems, the trained concept generator substantially improves the answer generator's pass@k over naive repeated sampling at the same answer generation allocation, surpasses concepts drawn from much larger untuned models, and transfers to answer generators it was never trained against, including a model from a different family. A small model can thus be trained into an effective, reusable search policy for a much larger one.

Finite-temperature many-body theories are typically formulated in imaginary time, where they are amenable to efficient numerical treatment. However, the extraction of spectral properties from the imaginary axis is difficult and requires numerical analytic continuation, while a direct formulation on the real axis requires prohibitively expensive numerical quadrature. In this paper, we represent the objects of many-body perturbation theory using systematically improvable pole and moment expansions and develop the corresponding framework for evaluating diagrammatic contributions directly on the real axis. We introduce a controlled recompression procedure that prevents the proliferation of poles under algebraic and diagrammatic operations. We benchmark the approach on paradigmatic quantum impurity problems and apply it to self-consistent calculations of the uniform electron gas. The resulting spectral quantities converge systematically towards numerical accuracy. This work provides a practical framework for controlled and systematically improvable calculations of spectral quantities within diagrammatic many-body theories, including self-consistent conserving approaches.

RGB-D 3D instance detectors benefit from visual semantics, but the task-specific Faster R-CNN/ResNet branch used by IIFNet3D couples feature extraction to a separately trained 2D detector and its image-domain labels. Replacing that branch with a frozen vision foundation model removes this task-specific dependency, but may introduce occlusion noise and a mismatch between patch features and geometry-aware detection features. In this work, we investigate this replacement through an adaptation of DINOv3 to the instance-level fusion pipeline of IIFNet3D. At the core of our approach is a depth-consistent feature pipeline that projects scene points into calibrated RGB-D frames, applies a metric depth-residual check, averages the accepted DINOv3 features into an offline point cache, and aggregates the cached features inside proposal-aligned RoI grids. The geometric and bidirectional instance-fusion paths are preserved, while Conservative VAID is evaluated as a low-strength, support-weighted semantic distillation recipe applied only to positive RoIs. We conduct extensive evaluations on ScanNetV2 to assess the proposed transfer recipes. On ScanNetV2, our DINOv3 control achieves mAP scores of 76.15 and 60.93 at IoU thresholds of 0.25 and 0.50, respectively. The Conservative VAID setting achieves mAP scores of 76.59 and 62.16, corresponding to numerical gains of 0.44 and 1.23 points over the control, respectively, in this checkpoint-level recipe comparison. The reported IIFNet3D result of 75.7/63.8 is used only as an external reference because the visual branch and processing protocol differ. Accordingly, we interpret these results as evidence for a controlled transfer recipe rather than as a causal estimate of the individual contributions of VAID or depth filtering.

We show that a great circle in $\mathbb{S}^3$ bounds embedded non-orientable minimal surfaces of unbounded genera and all possible non-zero Euler numbers. This answers a question of R. Hardt and H. Rosenberg (1990). It also gives the first example of a real analytic Jordan curve bounding infinitely many embedded minimal surfaces of distinct topological types. Such surfaces may arise as the links of boundary singularities of non-orientable minimal hypersurfaces in $\mathbb{R}^4$.

Aldous and Fill (2002) conjectured the asymptotic maximum relaxation time of a random walk on a connected regular graph. Since the relaxation time of a $d$-regular graph $G$ is $d/\mu(G)$, where $\mu(G)$ denotes its algebraic connectivity, this conjecture is closely related to the problem of minimizing algebraic connectivity among regular graphs. Guiduli and Mohar (1996) conjectured that, for every fixed minimum degree $\delta=d\ge3$ and all sufficiently large orders, graphs with minimum algebraic connectivity are path-like and, apart from bounded portions near their two ends, have a prescribed block structure. Abdi and Ghorbani (2024) proposed an analogous structural conjecture for $d$-regular graphs with minimum algebraic connectivity and fixed degree $d\ge3$. In Part~I, we proved the Aldous–Fill conjecture, the Guiduli–Mohar conjecture, and for odd degrees, the Abdi Ghorbani conjecture. In this paper, we settle the remaining even-degree case, thereby completing the structural characterization of regular graphs with minimum algebraic connectivity. We also prove that, for every fixed even $d\ge4$, the minimum algebraic connectivity at order $n$ is $2(d-2)\pi^2/n^2+O_d(n^{-3})$, and that every minimizing graph has diameter $3n/(d+1)+O_d(1)$. For every fixed even $d\ge6$, $d$-regular graphs whose algebraic connectivity is asymptotically minimum have asymptotically maximum diameter. Finally, we obtain a sharp normalized-gap bound for all even regular degrees, including degrees that grow with $n$.

Aldous and Fill (2002) conjectured that the maximum relaxation time of a random walk on a connected regular graph with $n$ vertices is bounded above by $(1+o(1))\frac{3n^2}{2\pi^2}$, with asymptotic equality for even $n$. Since the relaxation time of a $d$-regular graph $G$ is $d/\mu(G)$, where $\mu(G)$ denotes its algebraic connectivity, this conjecture is closely related to the problem of minimizing algebraic connectivity among regular graphs. Guiduli and Mohar (1996) conjectured that, for every fixed minimum degree $\delta=d\ge 3$ and all sufficiently large orders, graphs with minimum algebraic connectivity are path-like and, apart from bounded portions near their two ends, have a prescribed block structure. For fixed odd degree $d\ge 3$, Abdi and Ghorbani (2004) conjectured that $d$-regular graphs with minimum algebraic connectivity have the same structure. We prove the Aldous–Fill conjecture and the Guiduli–Mohar conjecture, as well as the corresponding conjecture for $d$-regular graphs of fixed odd degree. Finally, we prove that, for every fixed odd degree $d\ge 3$, $d$-regular graphs, as well as graphs of fixed minimum degree $d\ge 3$, whose algebraic connectivity is asymptotically minimum have asymptotically maximum diameter. This establishes the corresponding cases of another conjecture of Abdi and Ghorbani.

The LUX-ZEPLIN experiment has reported one nuclear-recoil event at $248\pm23_{\rm stat}\pm23_{\rm sys}\,{\rm keV}$ in a $2.84\,{\rm tonne\text{-}year}$ exposure, with a local significance of $3.4\sigma$. Inelastic (endothermic) dark matter is a natural candidate: its kinematic threshold suppresses low-energy recoils and shifts the signal toward higher recoil energies. We perform a model-independent analysis of the LZ data using two sidebands, without specifying an ultraviolet completion, so our results apply broadly to inelastic dark matter with coherent spin-independent nuclear scattering. For each $(m_\chi,\delta)$ we compute $N_{\rm low}/N_{\rm sig}$ and $N_{\rm hi}/N_{\rm sig}$—predicted count ratios in the sidebands $[14,225)\,{\rm keV}$ and $(271,800]\,{\rm keV}$ relative to the signal region $[225,271]\,{\rm keV}$. The two ratios vary in opposite directions with $\delta$ and are simultaneously optimized at a unique splitting $\delta^\star$. For $m_\chi=1\,{\rm TeV}$, $\delta^\star=339\,{\rm keV}$ with a combined Poisson tension of $\approx1.9\sigma$; over $500\,{\rm GeV}$–$3\,{\rm TeV}$, the tension at $\delta^\star$ rises from $\lesssim1.3\sigma$ to $\approx2.5\sigma$. Additional suppression arises from the Helm form factor: $|F_{\rm Xe}^{\rm Helm}|^2\simeq3\times10^{-4}$ at $248\,{\rm keV}$, near its second diffraction zero. A dedicated LZ search over $271<E_R<800\,{\rm keV}$ would provide a decisive test of inelastic dark matter with $\delta\gtrsim330\,{\rm keV}$.

We present new CO(1-0) measurements from ALMA and new HI measurements from MeerKAT in four ram pressure stripped (RPS) galaxies in the core of the Hydra I cluster: NGC 3312, NGC 3314a, NGC 3314b and LEDA 753342. These deep data, with compatible spatial resolution in CO and HI ($\sim15''$, $\sim$4 kpc), help increase the low statistics of galaxies observed in CO at extreme infall stage for probing how molecular and atomic gas behave under high ICM pressures. We studied the cold gas morphologies of these galaxies, and found asymmetries in their detected CO disks with respect to their HI and optical disks. No stripped CO was found in the tails, which could support the scenario that H$_2$ displays morphological and velocity asymmetries in the disks before being stripped. We then compared the H$_{2}$/HI ratios and H$_{2}$ vs. HI deficiencies between the four Hydra I galaxies, the xCOLDGASS sample dominated by field galaxies and VIVA-VERTICO Virgo RPS galaxies. LEDA 753342 ($\rm log~M_{\star}/M_{\odot} = 8.6$) is below the stellar mass limit of the reference samples ($\rm log~M_{\star}/M_{\odot} = 9$), but its $\rm M_{H_2}/M_{HI}$ and cold gas deficiencies are compatible with lower stellar mass field and Virgo cluster galaxies. NGC 3312 and NGC 3314a are in an intermediate stage between field galaxies and more advanced RPS Virgo galaxies. NGC 3314b is more consistent with Virgo galaxies more advanced in RPS, than with field galaxies. We reason that LEDA 753342 is at a very early stage of RPS, that NGC 3314a and NGC 3312 are at a similar timescale of early but very active RPS with differences potentially due to unequal initial total baryonic mass and infall angles, and that NGC 3314b is the latest stage in RPS in the sample. These galaxies display RPS across different evolutionary stages, and suggest that RPS starts affecting the CO reservoir before it starts being stripped from the galaxy.

In this paper, we introduce Gnomon, a physics-informed system that forecasts user-perceived low-Earth-orbit (LEO) downlink throughput, uplink throughput, and round-trip time (RTT) under different levels of trace availability. Gnomon's physics layer reconstructs the serving geometry and four-leg bent-pipe attenuation from public weather, orbital, routing, and licensing data. Based on what is available, Gnomon conditions on the target terminal's own history (Mode 1), measurements from nearby publicly reachable dishes (Mode 2), or the physical covariates alone (Mode 3) to predict the link state: Modes 1 and 2 share a fine-tuned time-series foundation model, while Mode 3 uses a compact boosted-tree estimator. All three modes expose a common output interface and can be selected without retraining. We evaluate Gnomon using 8,260 minutes of 1 Hz measurements collected at nine sites across five states in the U.S. We train on three sites and hold out the remaining six sites and their serving beams. On these unseen sites, the own-trace mode reduces downlink-throughput and RTT prediction error by 17% and 11% relative to the strongest published baseline and, to our knowledge, provides the first LEO uplink forecasts. The neighbor-trace mode requires no on-site hardware, while the covariate-only mode reduces downlink-throughput and RTT error by 24.6% and 78.8% relative to the only prior covariate-only forecaster. Moreover, experiments show that Gnomon provides calibrated quantile bands and improves adaptive-bitrate streaming driven over real TCP flows on replayed Starlink links.

The upgrade of the Large Hadron Collider (LHC) to the High-Luminosity LHC (HL-LHC) will greatly increase the average number of simultaneous proton-proton interactions per bunch-crossing (pileup), and will place significant demands on computing resources for the reconstruction of charged-particle tracks. Line Segment Tracking (LST) is a novel, highly parallelizable algorithm that can run efficiently on GPUs and has been integrated into the CMS software to enable the reconstruction of displaced tracks in the high pileup scenarios of the HL-LHC. The LST algorithm is designed to build track candidates using only outer tracker hits, naturally allowing for high acceptance of displaced track signatures. In the HL-LHC tracking reconstruction sequence without LST, displaced tracking ends at a radial displacement of the track origin of approximately 8 cm. The usage of LST in track reconstruction enhanced efficiency for displaced tracks, extending the acceptance in radial displacement from the track origin by more than a factor of 4. In this work, we introduce a new LST object, the quadruplet (T4), defined as a linked pair of triplets with a common line segment, targeting displaced track signatures. We present the performance of this improved implementation of the LST algorithm in several physics scenarios, extending the displaced-track acceptance from 40 cm to 60 cm in radial displacement compared to LST without T4s.

The generalized Kadanoff-Baym ansatz (GKBA) [P. Lipavský et al., Phys. Rev. B 34, 6933 (1986)] has emerged as a popular and numerically efficient algorithm for simplification of nonequilibrium Green's function (NEGF)-based calculations of time-dependent quantum transport. For systems that can be split into classical and quantum degrees of freedom, another popular simplifying strategy is adiabatic expansion (AE) of NEGF [N. Bode et al., Phys. Rev. Lett. 107, 036804 (2011); S. Deghi et al., Phys. Rev. B 110, 115409 (2024)] in terms of the velocity of classical degrees of freedom, such as localized magnetic moments (LMMs) in spintronics or coordinates of nuclei in nanoelectronics. Here we compare GKBA and second-order AE with numerically exact benchmarks for two-terminal junctions whose central region hosting quantum electrons and classical LMMs is attached to two semi-infinite normal metal leads. Three simple models are employed to exhibit cornerstone time-dependent effects in spintronics—spin pumping and spin-transfer torque (STT), as well as magnetic inertia as a recently explored phenomenon. We find that GKBA fails to describe pumping of spin current by precessing LMMs, or STT vectors, and thereby induced LMM dynamics. Conversely, the second-order AE matches numerically exact benchmarks for both effects remarkably well, thereby also revealing the essentially nonadiabatic nature of spin pumping. Thus, AE opens a path toward an accurate description of STT-driven magnetization dynamics, including combination with first-principles Hamiltonians, while incurring a fraction of the cost of time evolution of full NEGF. However, despite including terms up to the second time derivatives of LMMs into AE, this approach fails to capture fast nutational oscillations on top of the precessional motion of LMMs as the hallmark of magnetic inertia.

A coding agent must emit a valid tool call–a parseable invocation of a tool in the provided schema–before the harness can execute its chosen action. We study how local serving stacks affect this protocol step and show that measured outcomes can depend on the serving layer rather than model behavior alone. In Ollama, the default tools= request is gated per model by a static template flag: some models are accepted and return calls as text, some return native tool_calls, while Phi-3 and Gemma-3 are rejected before inference. In our harness, rejection and retry exhaustion are not preserved as structured failure metadata, so downstream analysis can misclassify them as model non-calls and naively report 0% fidelity. Adding a text tool list while retaining the native channel recovers much of the measured fidelity for accepted models, whereas a uniform text protocol reduces fidelity for Llama-3.2, which has native tool-call support. Cross-stack probes on Ollama, llama.cpp, vLLM, and SGLang show different handling of the same request. Constrained decoding removes parse failures but can induce non-termination, and turn-pooled versus per-instance estimates differ by up to about 55 points. We conclude with a checklist for treating serving behavior as part of the evaluation protocol.

We address the problems of giving a semantics to a relational database (RDB) that has missing values (MVs). The causes for the latter are governed by a Missingness Mechanism that is modelled as a Bayesian Network (BN) that involves the DB attributes as variables. The BN is called a Missingness Graph (MG). Our approach considerable departs from the treatment of RDBs with NULL (values). The combination of the MG and the observed DB allows us to build a block-independent probabilistic DB. We identify two optimal classes of its possible worlds on which QA can be performed. Those classes jointly capture probabilistic uncertainty and statistical plausibility of the implicit imputation of MVs. We obtain tractability results for the computation of some optimal classes; and we also obtain complexity results that characterize the computational feasibility of our approach.

We achieve linearly many parallelizable and addressable transversal logical multi-controlled-Z gates with asymptotically optimal parameters simultaneously on good quantum low-density parity check codes and quantum locally testable codes, by applying the gate framework of [arXiv:2604.01874] to good qLTC constructions. To establish the nontriviality of the logical operation, we express the cup product pairing as a coefficient in a product of Moore determinants and prove that this coefficient is nonzero. Moreover, we use the symmetries of the covering spaces to construct large spaces of cocycles. By multiplying the initial cocycles by suitable elements of these spaces, we obtain linearly many independent pairings, each selected by a corresponding cycle constructed using Poincaré duality. This yields linearly many parallelizable and addressable multi-controlled-Z gates and enables asymptotically constant-overhead magic state distillation with good qLDPC codes.

In a point-scanning microscope, detection through a pinhole improves resolution and replacing the pinhole with a detector array provides redundant encoding of the object spectrum, enabling correction of nonidealities in the imaging system or further resolution improvements, for instance. Here, we show that the mathematical principles underlying structured detection are not intrinsically spatial and therefore structured detection may be used for imaging distributions in other physical domains, such as time, frequency, and velocity space. We numerically illustrate this generality by applying structured detection to an electromagnetic distribution in time using four-wave mixing (FWM). The resulting image reconstructions outperform the temporal equivalent of confocal detection, even as the FWM models are made more realistic and shot noise is considered.

This work examines sub-threshold triadic responses to a vortex-ring-induced perturbation of an internal gravity-wave field, using sparsity-promoting dynamic mode decomposition (spDMD) to diagnose wave structures from experimental observations. Compared with a standard approach to dynamic mode decomposition (DMD) that selects modes by truncation of the singular value spectrum, spDMD, an alternative approach that instead seeks to minimise reconstruction error, is shown here to robustly identify previously indistinguishable oscillatory signals of high frequency in cases of both triadic resonance instability (TRI) and sub-threshold sustained triadic response (STR). These signals are consistent with wave–wave interaction products composed of a primary internal wave with frequency $\omega_0$, and secondary TRI waves with frequencies $\omega_1$ and $\omega_2$ respectively, leading to signals with frequencies $\omega_3 = \omega_0 + \omega_1$ and $\omega_4 = \omega_0 + \omega_2$. We show that these signals satisfy conditions for resonance, but fail to satisfy the dispersion relation for internal waves, indicating that they are confined, non-propagating oscillatory signals rather than freely propagating internal waves. A hierarchical analysis of wave–wave interactions provides evidence that feedback between both branches of the TRI secondary waves gives rise to these signals. Whilst we expect our observations to generalise to many classes of wave–wave interaction, we cannot exclude the possibility that the particular observations we make in the laboratory may arise due to reflections of the primary wave from a free surface. However, beneath strongly density-stratified regions of the ocean we would expect to find these interaction pathways faced with the same geometric configuration, so we are confident our findings will transfer directly to the natural environment.

Infancy is a critical developmental window characterized by rapid functional brain reorganization, during which large-scale networks emerge, individualized connectome signatures continue to form, and early deviations may shape long-term cognitive and clinical outcomes. Functional MRI (fMRI) offers an opportunity to study these processes in vivo, yet extracting developmentally meaningful information from it remains challenging due to comparatively short scan duration, structured motion artifacts, variable scan states, and rapid brain maturation. Amid these challenges, deep learning has expanded the capacity of computational neuroimaging by learning robust representations from noisy, high-dimensional data, integrating complex spatial and temporal information, and capturing the nonlinear and rapidly evolving organization of the developing brain. Here, we review recent advances in deep learning for infant functional neuroimaging, synthesizing progress across input representation formatting, population and individualized brain mapping, longitudinal trajectory forecasting, robust and explainable model evaluation, and biological translation. Collectively, these methodological advances mark a paradigm shift in infant functional neuroimaging from descriptive, group-level analyses toward reliable, individualized, and developmentally grounded models. Future progress will depend on larger and more diverse longitudinal datasets, developmentally appropriate model designs, rigorous and standardized evaluation, and integration of computational predictions with biological mechanisms towards clinically meaningful outcomes. Addressing these priorities will help establish deep learning as a robust framework for understanding early functional brain development, identifying developmental variation at the individual level, and ultimately supporting earlier and precise assessment of neurodevelopmental risk.

Distinguishing GPT-assisted from independently authored student writing has become a critical challenge in academia. This paper evaluates the discriminative capability of interpretable stylometric features extracted solely from submitted text. Using data from 90 participants who wrote both independently and with ChatGPT assistance, we evaluate eight machine learning classifiers while keeping data from the same participant together during validation. On the held-out test set, Random Forest achieved an ROC-AUC of 0.87 and an F1-score of 0.84, with False Positive and False Negative rates of 22.2% and 11.1%, respectively. SHAP analysis shows that lexical and grammatical characteristics drive the resulting predictions. The findings suggest that transparent, text-intrinsic features provide measurable signal for detecting GPT-assisted writing.

We prove that, over a commutative Noetherian ring, every finitely generated module of finite complete intersection dimension has finite quasi-projective dimension. Our proof adapts Bergh's technique of reducing complexity, originally used to establish virtual smallness for complexes of finite complete intersection dimension over local rings, by successively enlarging the perfect locus of the cones and controlling their homology. This gives an affirmative answer to part of a question of Jorgensen, Takahashi, and the third author. We also establish a dual injective version of our result for rings admitting a dualizing complex. Finally, we show that, over such rings, the three injective analogues of complete intersection dimension appearing in the literature coincide for bounded complexes with finitely generated homology, while two of them coincide even without a dualizing complex. We also give an example showing that the finite-generation hypothesis is necessary, thereby providing a full answer to a question of Sather-Wagstaff. We conclude by studying the localization behavior of these dimensions, obtaining a partial answer to a question of Sather-Wagstaff and Totushek.

The structure of interfacial water at electrified metal electrodes is known to affect electrocatalysis, and experiments highlight its sensitivity to electrolyte ion identity and applied bias. To disentangle generic charge-controlled structural changes from ion-specific effects, we compare ab initio molecular dynamics water structures at electrified Pt(111) generated with two very different biasing schemes: an explicit ion imbalance in the double layer and homogeneously distributed partially charged hydrogen atoms. Despite these distinct counter-charge representations, both approaches yield a consistent average response of the first water bilayer, in particular of the chemisorbed first layer, when compared on a common surface-charge scale. Method-dependent differences, in particular those associated with explicit electrolyte ions, become apparent only in more local structural descriptors. The hydrogen-bond topology reveals charge-dependent chain-to-ring rearrangements, with explicit ions enhancing ring populations near the potential of zero charge. Layer-resolved vibrational density of states (VDOS) assigns the strongest O-H stretching perturbation to chemisorbed first-layer water and identifies high-frequency signatures of ion-coordinated water, while computed vibrational sum-frequency generation (VSFG) spectra show that the physisorbed, electrolyte-facing region is particularly sensitive to the counter-charge representation. These results suggest that surface charge controls the average structural response, whereas electrolyte ions and their solvation shells become visible only when locally refined descriptors are investigated or probed, e.g. the H-bond network topology and vibrational fingerprints of the interfacial bilayer.