Searcharxiv⌕ Search

arXiv subjects

Wenjing Wu

Publications and source records attributed to Wenjing Wu.

18 recordsLinked to original sources
arXiv2026

What Aggregate Scores Miss: Measuring Item-Level Regressions in Commercial LLM API Migrations

Context: Software systems that depend on commercial large language model APIs must migrate to successor versions when vendors deprecate older models. Migration decisions typically rely on aggregate benchmark scores, which compress heterogeneous item-level behaviour into a single net figure. Objective: We measure what that compression conceals. Method: On three pairwise upgrades in the GPT-5.4 to GPT-5.6 Sol product sequence, we query 900 public benchmark items (graduate-level knowledge, olympiad mathematics, instruction following) 50 times per item per model, classify each item as reliably improved, reliably regressed, practically equivalent, or inconclusive under false-discovery-rate control and a practical-significance threshold, and calibrate the results against a label-permutation null. Results: Across all nine migration-benchmark cells, reliable improvements and reliable regressions coexist. Edges with aggregate gains of up to 7.3 percentage points contain up to 8.3% reliably regressed items; edges with aggregate losses contain up to 10.7% reliably improved items. On the instruction-following benchmark, the gap between strict and loose scoring widens by 3.9 percentage points on the latest migration: a 3.9-point regression under strict scoring shrinks to 0.04 points under loose scoring. Conclusion: Migration decisions based on aggregate scores alone miss substantial bidirectional item-level change. The complete response-level archive and per-item scoring outputs are released.

cs.SE↗
arXiv2026

Test-time reasoning effort and unauthorized tool use in language-model agents: a prespecified equivalence study

Language-model agents that execute multi-step workflows through tool calls operate under access-control policies that restrict which operations each role may perform. The APIs serving these agents expose a reasoning-effort parameter that operators adjust for cost and latency. Whether this parameter also changes the rate of unauthorized tool use has not been tested by direct manipulation within a single model. We vary reasoning effort (low, max) inside GPT-5.6 across the 14 confirmatory scenarios of TRIO-20, a suite of 20 matched workplace triads in which a policy-prohibited tool call is effective and its effect on the target metric is stated in the environment, effective but discoverable only through rule inspection, or ineffective. The three conditions derive from one code base and differ in two configuration fields, with identical prompts and tool sets. All analyses were prespecified in a frozen plan before confirmatory collection. Across 840 trajectories and two model tiers, no unauthorized tool call occurred. Exact one-sided 95% limits place each arm's violation rate below 3.50% (Terra, n = 84) and 5.21% (Sol, n = 56). The interaction estimand, with a simultaneous exact 95% interval of $\pm 4.34$ percentage points on Terra, lies inside the $\pm 7.01$-point equivalence margin. Raising effort did change behaviour, but only in inspection: rule-probe rates rose in all conditions, most where probing carried no instrumental payoff, a pattern inconsistent with the hypothesis of targeted search ($-14.3$ points, 95% CI $-27.4$ to $+1.2$). Raw trajectories are released at https://github.com/WenJing95/trio-20.

cs.CR↗
arXiv2026

Validation Evidence in LLM Repair Agents: How Much of What Passes Actually Tests the Bug?

When a repair agent runs a test and sees it pass, the result is treated as evidence about the reported defect. We measure how often that treatment is warranted. BSG-VA (buggy-state/candidate-state/gold-fix validation analysis) captures each validation command at its exact working-tree state, extracts a test-only patch, and replays the command on the original buggy code (B), the candidate state (S), and the developer gold fix (G). The captured outcome and the replay results assign every event an evidence role, from gold-aligned bug-discriminating through regression-only to misleading. Across 3,730 events in 643 rollouts on 110 tasks, 46.0% of positive comparable events carry no bug-discriminating information; 23.8% of baseline rollouts, with no feedback injected, close with a patch whose entire positive evidence base is of this kind. A three-arm experiment tests whether returning the B-replay outcome to the agent changes this pattern. Bug-contrast feedback reduces evidence-inadequate closure by 7.8 percentage points relative to an attention-matched reminder (p = 0.0029) and raises bug-discriminating evidence by 7.4 points (p = 0.011), with no detectable cost to repair success. Both estimates fall below the prespecified 10-percentage-point smallest effect size of interest, so practical magnitude remains uncertain. Roughly a third of the improvement traces to the reminder alone; across two exploratory replications, varying the scaffold and the model, the B-replay content adds a detectable increment only with gpt-5.6-sol under the unconstrained tool-use loop. BSG-VA applies post hoc to any replayable repair trajectory that preserves the required code states and execution environment. Keywords: program repair agents, validation evidence, test adequacy, large language models, software quality, controlled experiment.

cs.SE↗
arXiv2026

Quantum Dot Moiré from Crossed MoS2 Nanoribbons

Twisted atomically thin layers have attracted much attention for Moiré potential and correlated quantum phenomena. However, existing Moiré superlattices have largely been limited to extensive wavefunction without lateral confinement. Here we introduce a new platform where 1D nanoribbons of 2D MoS2 grown by vapor deposition can be easily superposed at various angles from stacking and transferring, to form Moiré quantum dots at their intersections with unique exciton physics. Angle-dependent Moiré intersections show enhanced exciton emission at commensurate angle 22 deg, which demonstrates faster relaxation at the cryogenic temperature. A size-dependent study further exhibits a reduced exciton energy and soften out-of-plane interlayer coupling for smaller Moiré areas. Our results reveal exciton physics turnability via precise overlapping of 1D nanoribbons.

cond-mat.mtrl-sci↗
arXiv2026

Compression, structure, and executor capability: a controlled real-cost decomposition of language-model agent skill optimisation

Agent skills, reusable instruction artefacts supplied to a tool-using language model, are increasingly optimised by shortening, structural rewriting, stronger-model compilation, and scoped loading, on the assumption that a smaller or better-organised skill lowers cost while preserving success. That assumption is rarely tested with quality and real monetary cost measured on the same runs and the contributing factors separated. This study reports a controlled decomposition over ten skill-delivery conditions, 40 software-engineering tasks, and three repetitions per cell (1,200 rollouts), separating no-skill execution, raw skills, deterministic shortening, linear and structured rendering from a shared semantic ledger, scoped loading, and the compiler and executor model tiers. Quality is the verifier pass rate at task level; cost is solve-stage token cost at standard provider prices, with a token-volume-normalised view for robustness and compilation cost amortised separately. The task is the unit of inference, intervals are task-clustered, and the contrast family is multiplicity-controlled. Deterministic shortening is close to the raw baseline but does not establish non-inferiority within the preset margin. Structured rendering and scoped loading lower pass rate on the compact executor without lowering cost, and structured rendering is indistinguishable from linear text at matched content. The only contrast surviving correction is executor capability, which raises pass rate by 27 percentage points at roughly five times the real cost, with compiler tier showing no robust effect. Under real prices no optimised representation reaches a practical break-even. The evidence indicates that executor capability is the dominant lever and that no representation strategy improves over the raw skill on either executor tier.

cs.SE↗
arXiv2026

Skill Availability and Presentation Granularity in Large-Language-Model Agents: A Controlled SkillsBench Study

Skill documents provide procedural knowledge to large-language-model agents at inference time. This article studies whether the presentation granularity of controlled skill knowledge changes downstream task success. The experiment uses a pinned SkillsBench version, a 30-task domain-balanced subset validated by official oracle runs, two reasoning-enabled model configurations, six skill conditions, and five trials per task-condition-model cell. Skill availability is the clearest empirical signal. Relative to no skill, skill conditions increase task-mean pass rate by 26.7 to 36.0 percentage points for GPT-5.5 and by 18.0 to 26.0 percentage points for DeepSeek V4-Flash. The final data contain 1,800 rows, with 900 rows for each model. The task is the inference unit. Five trials are aggregated within each task-condition-model cell before paired contrasts are estimated over 30 tasks. The primary presentation contrasts are smaller and uncertain. Low-abstraction guidance differs from high-abstraction guidance by +0.7 percentage points for GPT-5.5 and -6.7 percentage points for DeepSeek V4-Flash, with both 95% bootstrap confidence intervals crossing zero. Adding one worked example to medium-abstraction guidance differs from the no-example variant by +0.7 and +1.3 percentage points. Mean-reward robustness checks preserve the same substantive conclusion. In this controlled subset, skill availability is associated with higher success than no skill, while the tested presentation-granularity changes yield small, uncertain, and model-dependent effects.

cs.CL↗
arXiv2026

Hot biexcitons driven by extreme optical confinement

A powerful means to understanding condensed matter that possesses a multi-constituent, non-isolated, and complex nature, with a preeminent example being two-dimensional (2D) materials, is studying many-body interactions. However, experimentally observing high-order many-body interactions is a daunting task due to its heavy reliance on the abundance of low-order complexes. Here, we report the observation of four-body hot biexcitons in an energetically unfavorable bilayer of tungsten disulfide (WS2) through creating extreme optical confinement. Specifically, we integrate a non-radiative bound state in the continuum (BIC) into a photonic crystal (PhC) defect cavity, forming a quasi-three-dimensional (q-3D) but open confinement for photons at the driving frequency. The extremely confined photons in both reciprocal and physical spaces then excite inherently unproductive two-body hot excitons situated slightly above the indirect bandgap so efficiently that they form overwhelmed higher-order four-body hot biexcitons. Distinctively, these hot biexcitons exhibit substantial valley polarization and coherence at room temperature, which we attribute to the topological nature of BICs and the associated q-3D confinement with an orbital angular momentum. Besides achieving room-temperature biexcitons, the q-3D confinement could be valuable for higher-order interactions, such as triexcitons, and many other many-body phenomena, including Bose-Einstein condensation.

physics.optics↗
arXiv2026

Brain network science modelling of sparse neural networks enables Transformers and LLMs to perform as fully connected

Dynamic sparse training (DST) can reduce the computational demands in ANNs, but faces difficulties in keeping peak performance at high sparsity levels. The Cannistraci-Hebb training (CHT) is a brain-inspired method for growing connectivity in DST. CHT leverages a gradient-free, topology-driven link regrowth, which has shown ultra-sparse (less than 1% connectivity) advantage across various tasks compared to fully connected networks. Yet, CHT suffers two main drawbacks: (i) its time complexity is $O(Nd^3)$ - N node network size, d node degree - restricting it to ultra-sparse regimes. (ii) it selects top link prediction scores, which is inappropriate for the early training epochs, when the network presents unreliable connections. Here, we design the first brain-inspired network model - termed bipartite receptive field (BRF) - to initialize the connectivity of sparse artificial neural networks. We further introduce a GPU-friendly matrix-based approximation of CH link prediction, reducing complexity to $O(N^3)$. We introduce the Cannistraci-Hebb training soft rule (CHTs), which adopts a flexible strategy for sampling connections in both link removal and regrowth, balancing the exploration and exploitation of network topology. Additionally, we integrate CHTs with a sigmoid gradual density decay (CHTss). Empirical results show that BRF offers performance advantages over previous network science models. Using 1% of connections, CHTs outperforms fully connected networks in MLP architectures on image classification tasks, compressing some networks to less than 30% of the nodes. Using 5% of the connections, CHTss outperforms fully connected networks in two Transformer-based machine translation tasks. Finally, at 30% connectivity, both CHTs and CHTss outperform other DST methods in language modeling task.

cs.LG↗
arXiv2025

Multimode Phonon-Polaritons in Lead-Halide Perovskites in the Ultrastrong Coupling Regime

Phonons play a central role in fundamental solid-state phenomena, including superconductivity, Raman scattering, and symmetry-breaking phases. Harnessing phonons to control these effects and enable quantum technologies is therefore of great interest. However, most existing phonon control strategies rely on external driving fields or anharmonic interactions, limiting their applicability. Here, we realize multimode ultrastrong light--matter coupling and theoretically show the modulation of phonon emission. This regime is realized by coupling two optical phonon modes in lead halide perovskites to a nanoslot array functioning as a single-mode cavity. The small mode volume of the nanoslots enables high coupling strengths in the phonon-polariton system. We show theoretically that the nanoslot resonator mediates an effective interaction between phonon modes, leading to superthermal phonon bunching in thermal equilibrium between distinct modes. Our findings are well described by a multimode Hopfield model. This work establishes a pathway for engineering phononic properties for light-harvesting and light-emitting technologies.

quant-ph↗
arXiv2024

A Talenti-type comparison theorem for the $p$-Laplacian on $\mathrm{RCD}(K,N)$ spaces and some applications

In this paper, we prove a Talenti-type comparison theorem for the $p$-Laplacian with Dirichlet boundary conditions on open subsets of a $\mathrm{RCD}(K,N)$ space with $K>0$ and $N\in (1,\infty)$. The obtained Talenti-type comparison theorem is sharp, rigid and stable with respect to measured Gromov-Hausdorff topology. As an application of such Talenti-type comparison, we establish a sharp and rigid reverse Hölder inequality for first eigenfunctions of the $p$-Laplacian and a related quantitative stability result.

math.DG↗
arXiv2023

Subgradient estimates for a nonlinear subparablic equation on complete pseudo-Hermitian manifolds

Let $(M,J,θ)$ be a complete pseudo-Hermitian manifold which satisfies the CR sub-Laplacian comparison property. In this paper, we derive the local subgradient estimates for positive solutions to the following nonlinear subparabolic equation: \begin{equation*} u_t=Δ_\mathrm{b}u+au\log u+bu, \end{equation*} on $M\times [0,+\infty)$, where $a,b\in \mathbb{R}$. As a application, we derive a priori estimate and a Harnack inequality for positive solutions to the subelliptic equation $Δ_\mathrm{b}u+au\log u=0$.

math.DG↗
arXiv2022

EllipsoNet: Deep-learning-enabled optical ellipsometry for complex thin films

Optical spectroscopy is indispensable for research and development in nanoscience and nanotechnology, microelectronics, energy, and advanced manufacturing. Advanced optical spectroscopy tools often require both specifically designed high-end instrumentation and intricate data analysis techniques. Beyond the common analytical tools, deep learning methods are well suited for interpreting high-dimensional and complicated spectroscopy data. They offer great opportunities to extract subtle and deep information about optical properties of materials with simpler optical setups, which would otherwise require sophisticated instrumentation. In this work, we propose a computational ellipsometry approach based on a conventional tabletop optical microscope and a deep learning model called EllipsoNet. Without any prior knowledge about the multilayer substrates, EllipsoNet can predict the complex refractive indices of thin films on top of these nontrivial substrates from experimentally measured optical reflectance spectra with high accuracies. This task was not feasible previously with traditional reflectometry or ellipsometry methods. Fundamental physical principles, such as the Kramers-Kronig relations, are spontaneously learned by the model without any further training. This approach enables in-operando optical characterization of functional materials within complex photonic structures or optoelectronic devices.

physics.optics↗
arXiv2022

Ultra Localized Optoelectronic Properties of Nanobubbles in 2D Semiconductors

The optical properties of transition metal dichalcogenides have previously been modified at the nanoscale by using mechanical and electrical nanostructuring. However, a clear experimental picture relating the local electronic structure with emission properties in such structures has so far been lacking. Here, we use a combination of scanning tunneling microscopy (STM) and near-field photoluminescence (nano-PL) to probe the electronic and optical properties of single nano-bubbles in bilayer heterostructures of WSe2 on MoSe2. We show from tunneling spectroscopy that there are electronic states deeply localized in the gap at the edge of such bubbles, which are independent of the presence of chemical defects in the layers. We also show a significant change in the local bandgap on the bubble, with a continuous evolution to the edge of the bubble over a length scale of ~20 nm. Nano-PL measurements observe a continuous redshift of the interlayer exciton on entering the bubble, in agreement with the band to band transitions measured by STM. We use self-consistent Schrödinger-Poisson (SP) simulations to capture the essence of the experimental results and find that strong doping in the bubble region is a key ingredient to achieving the observed localized states, together with mechanical strain.

cond-mat.mes-hall↗
arXiv2020

Diffusivity Reveals Three Distinct Phases of Interlayer Excitons in MoSe2/WSe2 Heterobilayers

Charge separated interlayer excitons in transition metal dichalcogenide (TMDC) heterobilayers are being explored for moiré exciton lattices and exciton condensates. The presence of permanent dipole moments and the poorly screened Coulomb interaction make many body interactions particularly strong for interlayer excitons. Here we reveal two distinct phase transitions for interlayer excitons in the MoSe2/WSe2 heterobilayer using time and spatially resolved photoluminescence imaging: from trapped excitons in the moiré-potential to the modestly mobile exciton gas as exciton density increases to ne/h ~ 1011 cm-2 and from the exciton gas to the highly mobile charge separated electron/hole plasma for ne/h > 1012 cm-2. The latter is the Mott transition and is confirmed in photoconductivity measurements. These findings set fundamental limits for achieving quantum states of interlayer excitons.

cond-mat.mes-hall↗
arXiv2020

Deep moiré potentials in twisted transition metal dichalcogenide bilayers

In twisted bilayers of semiconducting transition metal dichalcogenides (TMDs), a combination of structural rippling and electronic coupling gives rise to periodic moiré potentials that can confine charged and neutral excitations. Here, we report experimental measurements of the structure and spectroscopic properties of twisted bilayers of WSe2 and MoSe2 in the H-stacking configuration using scanning tunneling microscopy (STM). Our experiments reveal that the moiré potential in these bilayers at small angles is unexpectedly large, reaching values of above 300 meV for the valence band and 150 meV for the conduction band - an order of magnitude larger than theoretical estimates based on interlayer coupling alone. We further demonstrate that the moiré potential is a non-monotonic function of moiré wavelength, reaching a maximum at around a 13nm moiré period. This non-monotonicity coincides with a drastic change in the structure of the moiré pattern from a continuous variation of stacking order at small moiré wavelengths to a one-dimensional soliton dominated structure at large moiré wavelengths. We show that the in-plane structure of the moiré pattern is captured well by a continuous mechanical relaxation model, and find that the moiré structure and internal strain rather than the interlayer coupling is the dominant factor in determining the moiré potential. Our results demonstrate the potential of using precision moiré structures to create deeply trapped carriers or excitations for quantum electronics and optoelectronics.

cond-mat.mes-hall↗
arXiv2020

One-Dimensional Moiré Excitons in Transition-Metal Dichalcogenide Heterobilayers

The formation of interfacial moiré patterns from angular and/or lattice mismatch has become a powerful approach to engineer a range of quantum phenomena in van der Waals heterostructures. For long-lived and valley-polarized interlayer excitons in transition-metal dichalcogenide (TMDC) heterobilayers, signatures of quantum confinement by the moiré landscape have been reported in recent experimental studies. Such moiré confinement has offered the exciting possibility to tailor new excitonic systems, such as ordered arrays of zero-dimensional (0D) quantum emitters and their coupling into topological superlattices. A remarkable nature of the moiré potential is its dramatic response to strain, where a small uniaxial strain can tune the array of quantum-dot-like 0D traps into parallel stripes of one-dimensional (1D) quantum wires. Here, we present direct evidence for the 1D moiré potentials from real space imaging and the corresponding 1D moiré excitons from photoluminescence (PL) emission in MoSe2/WSe2 heterobilayers. Whereas the 0D moiré excitons display quantum emitter-like sharp PL peaks with circular polarization, the PL emission from 1D moiré excitons has linear polarization and two orders of magnitude higher intensity. The results presented here establish strain engineering as a powerful new method to tailor moiré potentials as well as their optical and electronic responses on demand.

cond-mat.mtrl-sci↗
arXiv2018

Using ATLAS@Home to exploit extra CPU from busy grid sites

Grid computing typically provides most of the data processing resources for large High Energy Physics experiments. However typical grid sites are not fully utilized by regular workloads. In order to increase the CPU utilization of these grid sites, the ATLAS@Home volunteer computing framework can be used as a backfilling mechanism. Results show an extra 15% to 42% of CPU cycles can be exploited by backfilling grid sites running regular workloads while the overall CPU utilization can remain over 90%. Backfilling has no impact on the failure rate of the grid jobs, and the impact on the CPU efficiency of grid jobs varies from 1% to 11% depending on the configuration of the site. In addition the throughput of backfill jobs in terms of CPU time per simulated event is the same as for resources dedicated to ATLAS@Home. This approach is sufficiently generic that it can easily be extended to other clusters.

physics.comp-ph↗