Searcharxiv⌕ Search

arXiv subjects

Iris Ye

Publications and source records attributed to Iris Ye.

8 recordsLinked to original sources
arXiv2026

Structuring Survey Histories for LLM Digital Twins: A Controlled Study of Background, Decision Procedure, and Evaluation Profiles

LLM-based digital twins aim to predict how individuals would respond in new settings from representations of their prior responses. Existing evidence that compression of survey histories need not reduce predictive accuracy suggests that representation structure, rather than information volume alone, may be an important design choice. We study BDE, a fixed representation that organizes each respondent's survey history into Background, Decision procedure, and Evaluation. We evaluate it on Twin-2K-500 (Toubia et al., 2025) by predicting each respondent's held-out answers on 17 tasks. In a paired evaluation with 100 respondents, a fully inferred BDE profile improves overall accuracy over the raw survey transcript by 1.27 percentage points (pp) on gpt-5.4-nano (p < 0.01) and 1.19 pp on gpt-5.4-mini (p < 0.001). Three analyses then examine the schema, extraction depth, and component composition. Against a length-matched free-form profile written from the same transcript by the same extractor, BDE improves pricing accuracy on both simulators (2.15 pp on nano, p < 0.01; 1.38 pp on mini, p < 0.05), while its overall advantage is significant only on nano (1.05 pp, p < 0.05). Allowing deeper inference during extraction yields no statistically significant gain on either simulator. By contrast, adding Evaluation to a Background--Decision procedure profile improves overall accuracy by 1.44 pp on nano (p < 0.001) and 1.22 pp on mini (p < 0.001). These results indicate that the advantage over raw transcripts is not explained by compression or length alone, and attribute it primarily to profile composition rather than deeper extraction.

cs.CL↗
arXiv2026

Automatic indexing of wide-band quasi-Laue neutron diffraction patterns of large unit cells

With a continuous spectrum of incident neutron wavelengths, quasi-Laue diffraction records many Bragg reflections from a stationary crystal, each satisfying the diffraction condition at a particular wavelength. Because the wavelength of each reflection is not measured, the crystal orientation must be determined from the scattering directions alone. At the new IMAGINE-X instrument, instant-readout detectors and many short exposures per experiment require this indexing to run automatically, yet for a protein crystal at the full 2-10 Angstrom band a single peak can have approximately $10^4$ candidate reflections, and a random orientation indexes a large fraction of the observed peaks. We describe an algorithm that indexes such patterns for a known unit cell and wavelength band. Differential evolution searches uniformly sampled rotations, matching observed directions to band-limited shells of candidate directions, and each solution is tested against the indexing expected for a random orientation. In a benchmark search, a k-d tree and a persistent pool of 32 worker processes reduce the wall time 270-fold relative to a serial scan of all candidates. The algorithm is first validated on the time-of-flight diffractometer MANDI, where the measured wavelengths give a reference solution by standard indexing. With the wavelengths discarded, every indexed reflection from 93 rare-earth garnet patterns agrees up to harmonic order. The algorithm then reliably indexes the natural mineral mesolite, measured on the original image plate at IMAGINE before the upgrade, and the protein T4 lysozyme, measured on the new IMAGINE-X instrument over the full band.

physics.ins-det↗
arXiv2026

NeuXtalViz: Interactive Three-Dimensional Visualization and Analysis for Single-Crystal Neutron Diffraction

NeuXtalViz (Neutron Single-Crystal Visualization) is a Python-based software package developed at Oak Ridge National Laboratory to provide interactive three-dimensional visualization and analysis tools for single-crystal neutron diffraction experiments. Built on the Mantid framework for data reduction, and leveraging PyVista and Matplotlib within a Python Qt environment, NeuXtalViz adopts a model-view-presenter architecture that separates the user interface from the core processing components. The software provides a unified interface for tasks central to single-crystal diffraction, including UB-matrix determination, experiment planning, visualization of normalized reciprocal-space volumes, and real-space crystal-structure calculations. It also integrates with widely used community tools and has been deployed on instrument and analysis servers, where it is now being adopted by instrument teams and users. By embedding advanced three-dimensional visualization directly into the experimental workflow, NeuXtalViz enhances the planning, execution, and analysis cycle for single-crystal neutron diffraction experiments, while providing a flexible framework for future development.

physics.ins-det↗
arXiv2026

Variations of Helly's theorem for convex splinters

A convex splinter $K$ is a union of convex sets in $\mathbb{R}^d$ such that every minimal affine dependent set of $\mathbb{R}^d$ contained in $K$ is contained in one of the sets. The study of convex splinters was motivated by the study of flat transversals to convex sets. We extend several variations of Helly's theorem from convex geometry to convex splinters. These include fractional and colorful variations of Helly's theorem. We also extend Tverberg's theorem to convex splinters.

math.CO↗
arXiv2026

Beyond Raw Transcripts: Structured Persona Extraction for LLM-Based Digital Twins

LLM-based "digital twins" aim to simulate how an individual would behavein new environments or respond to novel questions, given some representation of that individual's prior responses. A common approach constructs this representation from survey transcripts or summaries responses. Prior work shows that compressing long transcripts into shorter LLM-generated summaries does not significantly reduce predictive accuracy, suggesting that information volume is not the primary bottleneck. In this work, we argue that the key limitation is instead structural:how persona information is organized before being provided to thesimulator model. We study this by comparing unstructured summaries with structured persona representations. First, we introduce a hand-craftedschema (BDE: Background, Decision procedure, Evaluation), grounded in consumer-behavior theory, and show that it improves predictive accuracy over raw transcripts by +1.91 percentage points on a homogeneous benchmark (Twin-2K-500), with similar gains on gpt-5.4-mini and Qwen3-8B as robustness checks. However, this fixed structure does not generalizeacross more heterogeneous tasks, where performance is statistically indistinguishable from the raw transcript baseline. To address this limitation, we propose an automatic structure-discovery pipeline in which an LLM iteratively proposes and refines task-specific persona structures and extraction prompts. On a benchmark of 13 diverse sub-studies, this approach restores performance, improving mean accuracy by +1.91 percentage points over the raw transcript baseline and eliminating significant losses observed with the fixed schema. Overall, our results suggest that the main constraint in LLM-based digital twins is not how much information is provided, but how it is structured -- and that the optimal structure depends on the task.

cs.CL↗
arXiv2026

Contrasting structural reversibility and magnetic correlations in isostructural honeycomb magnets CrCl$_3$ and $α$-RuCl$_3$

We report a comparative neutron single crystal diffraction study of the structural and magnetic properties of layered halides CrCl$_3$ and $α$-RuCl$_3$, which host a honeycomb arrangement of transition metal ions with distinct electronic configurations and undergo a first-order structural transition between high-temperature \textit{C}2/\textit{m} and low-temperature \textit{R}$\bar{3}$. Both compounds show a step-like change in the $c$-lattice, consistent with an expected stacking rearrangement. In contrast, the in-plane lattice response is quite different: $α$-RuCl$_3$ exhibits an abrupt hysteretic change across the transition accompanied by progressive crystalline degradation upon thermal cycling, whereas CrCl$_3$ shows a smooth in-plane lattice evolution and remains structurally robust. Magnetically, CrCl$_3$ orders into an A-type antiferromagnetic structure at T$_N$=14\,K and exhibits pronounced diffuse magnetic scattering extending up to about 40\,K. $α$-RuCl$_3$ shows no observable magnetic diffuse scattering above its zig-zag antiferromagnetic ordering temperature T$_N$=7.6\,K. These results suggest that the contrasting structural and magnetic behaviors arise from an interplay between interlayer sliding energetics and the fundamentally different electronic configurations of the two compounds.

cond-mat.str-el↗
arXiv2023

Structure transition and zigzag magnetic order in Ir/Rh-substituted honeycomb lattice RuCl3

We report magnetization and neutron diffraction studies on crystal and magnetic structures of Ir- and Rh-substituted honeycomb lattice $α$-RuCl$_3$. The iridium or rhodium atoms are distributed at the Ru site with little structural modification. Both systems undergo a room-temperature monoclinic $C2/m$ to low-temperature trigonal $R\bar{3}$ phase transformation with a large recoverable hysteresis. At low temperature, a zigzag spin order is observed with the same characteristic wavevector $(0,0.5,1)$ as in the parent $α$-RuCl$_3$. Detailed magnetic structure refinement reveals an ordered moment of $\rm 0.32(5) μ_B/Ru$ and a upper boundary of canting angle of $15(4)^\circ$ away from the basal plane at 5~K for the 10\% Ir-substituted $α$-RuCl$_3$, which is different from the 0.45-0.73~$\rm μ_B/Ru$ and $32^\circ$-$48^\circ$ canting angle reported in the parent compound $α$-RuCl$_3$. The observation of unchanged RuCl$_6$ local octahedral environment, reduced magnetic moment size and canting angle highlights the potential to study quantum spin liquid behavior through non-magnetic ion doping.

cond-mat.mtrl-sci↗