arXiv subjects

Hyeonjun An

Publications and source records attributed to Hyeonjun An.

5 recordsLinked to original sources

arXiv2026

Polymer-Regulated Freezing of Water Droplets Revealed by Synchrotron X-ray Imaging and Raman Spectroscopy

Adding a polymer to a sessile water droplet not only lowers its freezing point but also suppresses the tip singularity that forms during its freezing on cold substrates. Here, we employ synchrotron X-ray and Raman imaging to elucidate the spatiotemporal mechanism underlying tip suppression in an aqueous polyvinyl alcohol (PVA) solution, a model polymer solution. As the polymer concentration increases, we observe slower propagation of the freezing front, reduced bubble entrapment, and a progressively more rounded apex across the volumes and molecular weights examined. X-ray tomography reveals that frozen PVA droplets retain low X-ray transmittance domains in their interiors and at the surface, and Raman spectral mapping confirms that these domains correspond to PVA-enriched regions, providing direct evidence of freeze-induced polymer segregation. These findings indicate that PVA is redistributed heterogeneously during water solidification rather than shifting bulk properties homogeneously, providing a spatially resolved framework for interpreting the observed tip blunting and the suppression of discrete bubble entrapment. Our work identifies freeze-induced polymer segregation as a pathway by which a dissolved polymer regulates both the external shape and the internal structure of a freezing droplet, and these findings shed light on potential applications in freezing-based processes such as freeze-casting and cryopreservation.

cond-mat.soft↗

arXiv2026

DFLOP: A Data-driven Framework for Multimodal LLM Training Pipeline Optimization

Multimodal Large Language Models (MLLMs) have achieved remarkable advances by integrating text, image, and audio understanding within a unified architecture. However, existing distributed training frameworks remain fundamentally data-blind: they parallelize computation without accounting for variations in input data characteristics. This data unawareness leads to severe computation skew across stages and microbatches, where heterogeneous multimodal inputs incur different processing costs. Consequently, GPU resources are unevenly utilized, synchronization delays accumulate, and overall training efficiency degrades. To address this limitation, we present DFLOP, a data-driven framework for multimodal LLM training pipeline optimization. DFLOP continuously profiles runtime behavior to capture data-induced computation variance and employs predictive scheduling to balance workloads across stages and microbatches. By coupling data characteristics with execution planning, DFLOP substantially improves GPU utilization and throughput. Extensive experiments on large-scale multimodal benchmarks show that DFLOP achieves up to 3.6x faster training compared to state-of-the-art distributed training frameworks.

cs.DC↗

arXiv2025

Exqutor: Extended Query Optimizer for Vector-augmented Analytical Queries

Vector similarity search is becoming increasingly important for data science pipelines, particularly in Retrieval-Augmented Generation (RAG), where it enhances large language model inference by enabling efficient retrieval of relevant external knowledge. As RAG expands with table-augmented generation to incorporate structured data, workloads integrating table and vector search are becoming more prevalent. However, efficiently executing such queries remains challenging due to inaccurate cardinality estimation for vector search components, leading to suboptimal query plans. In this paper, we propose Exqutor, an extended query optimizer for vector-augmented analytical queries. Exqutor is a pluggable cardinality estimation framework designed to address this issue, leveraging exact cardinality query optimization techniques to enhance estimation accuracy when vector indexes (e.g., HNSW, IVF) are available. In scenarios lacking these indexes, we employ a sampling-based approach with adaptive sampling size adjustment, dynamically tuning the sample size to balance estimation accuracy and sampling overhead. This allows Exqutor to efficiently approximate vector search cardinalities while minimizing computational costs. We integrate our framework into pgvector, VBASE, and DuckDB, demonstrating performance improvements of up to four orders of magnitude on vector-augmented analytical queries.

cs.DB↗

arXiv2025

X-ray Microscopy Study of Freezing Sessile Droplets

A sessile water droplet on a cold substrate freezes into a shape with a sharp apex because of water's expansion upon freezing, yielding a universal tip angle across various conditions. Using \textit{in situ} X-ray imaging, we report that this angle changes with substrate temperature, and the deviation originates from bubble formation during freezing. Three-dimensional tomography enables direct quantification of the effective ice-water density ratio, accounting for trapped bubbles. Incorporating this effective density ratio reconciles the temperature-dependent tip angles. We also confirm that a bubble-free frozen droplet in a vacuum chamber exhibits the universal tip angle. Furthermore, X-ray imaging allows us to measure the three-phase boundary angles \textit{in situ}, thereby validating the geometric theory behind tip formation. These findings advance our understanding of the freezing dynamics associated with multiphase systems and highlight the capabilities of high-resolution X-ray imaging in ice research.

cond-mat.soft↗

arXiv2023

Suppressed terahertz dynamics of water confined in nanometer gaps

Nanoconfined waters have been extensively studied within various systems, demonstrating low permittivity under static conditions; however, their dynamics have been largely unexplored due to the lack of a robust platform, particularly in the terahertz (THz) regime where hydrogen bond dynamics occur. We report the THz complex refractive index of nanoconfined water within metal gaps ranging in width from 2 to 20 nanometers, spanning mostly interfacial waters all the way to quasi-bulk waters. These loop nanogaps, encasing water molecules, sharply enhance light-matter interactions, enabling precise measurements of refractive index, both real and imaginary parts, of nanometer-thick layers of water. Under extreme confinement, the suppressed dynamics of the long-range correlation of hydrogen bond networks corresponding to the THz frequency regime result in a significant reduction in the terahertz permittivity of even 'non-interfacial' water. This platform provides valuable insights into the long-range collective dynamics of water molecules which is crucial to understanding water-mediated processes such as protein folding, lipid rafts, and molecular recognition.

physics.optics↗