arXiv subjects

Hyunchul Kim

Publications and source records attributed to Hyunchul Kim.

3 recordsLinked to original sources

arXiv2026

Compositional Spectral Prompts for LLM-based Online Time Series Forecasting

To address the sequential and evolving nature of time series, the Online Time Series Forecasting (OTSF) task has been extensively studied in multiple domains. Existing research focuses on adapting to non-stationary environments by employing memory buffer-based retrieval strategies. However, we observe that such frameworks struggle with long-term adaptation and fail to generalize to unseen patterns. To this end, we introduce CoSPOT, an LLM-based online time series forecasting framework that leverages a pre-trained LLM as the backbone online forecaster, motivated by its strong few-shot capabilities. For efficient online adaptation, CoSPOT keeps the LLM frozen and employs compositional spectral prompts grounded in frequency-domain bases to guide the model with the overall distribution of the input, thereby substantially reducing the number of parameters updated during the online phase. Specifically, CoSPOT decomposes time series into frequency bases and composes the corresponding spectral basis prompts according to their amplitudes, allowing unseen patterns to be represented as new combinations of learned basis prompts. Our extensive experiments on real-world datasets demonstrate the superiority and practicality of CoSPOT across challenging online scenarios, including extended online phases and cross-dataset settings with substantial distribution shifts. Our code is available at https://github.com/seungyoon-Choi/CoSPOT.

cs.LG↗

arXiv2026

A Modulated Electron Lattice (MEL) Criterion for Metallic Superconductivity

A central unresolved question in the theory of superconductivity is why only a small subset of metallic elements exhibit a superconducting state, whereas many others remain strictly normal. Neither the conventional Bardeen Cooper Schrieffer (BCS) framework nor its extensions involving charge density wave (CDW) or pair density wave (PDW) order provide a predictive or material-selective criterion capable of distinguishing superconducting metals from non-superconducting ones. In particular, the persistent absence of superconductivity in simple noble metals with well-defined Fermi surfaces poses a challenge for all traditional approaches. Here we address this problem using the Modulated Electron Lattice (MEL) Ginzburg Landau (GL) framework introduced in our previous work. In this formulation, a coarse-grained MEL charge field $\rho_{\mathrm{MEL}}(\mathbf{r})$ with momentum dependent stiffness $\alpha(q)$ is coupled to the superconducting (SC) order parameter $\psi(\mathbf{r})$. We show that metallic superconductivity emerges only when the system satisfies a specific ``MEL enhancement window,'' characterized by a negative minimum of $\alpha(q)$ at either a finite modulation wave vector $q^{\ast}$ or at $q=0$, together with sufficiently strong coupling between $\rho_{\mathrm{MEL}}$ and $\psi$. This unified criterion naturally partitions metallic elements into three universal classes: (i) MEL-enhanced superconductors with a finite-$q^{\ast}$ charge mode, (ii) conventional BCS superconductors as the homogeneous $q^{\ast}=0$ limit of the MEL framework, and (iii) metals for which $\alpha(q)$ remains positive for all $q$, suppressing all MEL modes and preventing any superconducting instability. By applying this criterion to simple metallic elements, we identify why some metals develop superconductivity while others do not, possibly resolving a selection problem long open within the BCS paradigm.

cond-mat.supr-con↗

arXiv2025

Short-Range Modulated Electron Lattice and d-Wave Superconductivity in Cuprates: A Phenomenological Ginzburg-Landau Framework

We develop a phenomenological Ginzburg-Landau (GL) framework for high-$T_c$ cuprates in which a short-range modulation of the electronic charge density couples to a $d$-wave superconducting condensate. The resulting modulated electron lattice (MEL) state is distinct from long-range static charge density wave order: it is short range, partially phase coherent, and linked to superconducting coherence. A preferred wave vector $q^{\ast} \approx 0.3$ reciprocal lattice units along the Cu-O bond direction emerges from the interplay between a momentum-dependent susceptibility and bond-stretching phonons, consistent with neutron and x-ray data on YBa$_2$Cu$_3$O$_{7-\delta}$ and related cuprates. The GL free energy contains coupled $d$-wave superconducting and charge sectors with parameters constrained by optimally doped YBa$_2$Cu$_3$O$_{7-\delta}$. We identify an MEL enhancement window in doping, temperature, MEL correlation length, and disorder where a coherence linked modulation enhances the superfluid stiffness. Classical Monte Carlo simulations yield an in-plane stiffness enhancement of order ten percent, which we treat as a qualitative prediction to be tested by self-consistent Bogoliubov de Gennes calculations. The MEL framework yields falsifiable experimental signatures. For scanning tunneling spectroscopy in Bi-based cuprates we highlight two predictions: the Fourier-transformed local density of states should exhibit a $q^{\ast} \approx 0.3$ peak whose spectral weight sharpens as superconducting phase coherence develops below $T_c$, in contrast to static charge scenarios, and the local gap magnitude $\Delta(r)$ should correlate positively with the local MEL amplitude. The framework implies correlations between MEL correlation length, superfluid stiffness, disorder, and vortex pinning, and organizes cuprate observations into testable STM/STS predictions.

cond-mat.supr-con↗