arXiv subjects

Jaehyung Lee

Publications and source records attributed to Jaehyung Lee.

12 recordsLinked to original sources

arXiv2026

ALIGNN 2.0: A Unified Line-Graph Neural Network Framework for Materials Screening, Force Fields, Inverse Design, Spectroscopy, and Microscopy

Graph neural networks are central to materials property prediction and machine-learning interatomic potentials, yet their reliance on specialized graph libraries hampers portability and reproducibility, and property and force-field models have historically required separate graph pipelines. We present ALIGNN 2.0, a dependency-free, pure-PyTorch reimplementation of the Atomistic Line Graph Neural Network, with the line graph and its batching built from scratch, running on current-generation accelerators and unifying scalar, spectral, tensorial, per-atom, and force-field prediction behind a single graph, a combination that to our knowledge no existing framework provides. Comparing radius and k-nearest-neighbor (kNN) graphs, the wider kNN graph is more accurate for properties while the smoothly varying radius graph is required for energy-conserving molecular dynamics. On the JARVIS-Leaderboard, ALIGNN 2.0 leads on 26 of 30 single-property benchmarks against the original ALIGNN, with large gains for piezoelectric and dielectric maxima, exfoliation energy, moduli, and superconducting Tc. The LAMMPS- and OpenMM-compatible ALIGNN-FF matches leading universal potentials on the Matbench-Discovery and CHIPS-FF benchmarks at a small fraction of their parameters while scaling to hundred-thousand-atom cells. We further use ALIGNN 2.0 as the denoiser in a conditional crystal-diffusion model, where explicit line-graph message passing consistently lowers structural denoising loss. We also show, as work in progress, that an independently diffused, redundant bond-angle state is learnable but does not uniformly improve reconstruction or combine additively with the line graph. Finally, from a single relaxed structure the same framework reconstructs infrared, Raman, optical-dielectric, and neutron spectra in agreement with experiment and DFT, and drives frozen-phonon electron-microscopy image simulation.

cond-mat.mtrl-sci↗

arXiv2026

Hybrid DiffractGPT-Rietveld Refinement Framework for Automated X-ray Diffraction Analysis

X-ray diffraction (XRD) is fundamental to structural materials characterization, yet transforming a raw powder pattern into a refined crystal structure still demands considerable domain expertise. We present AGAPI-XRD, a hybrid framework integrating DiffractGPT generative structure prediction, database pattern matching against JARVIS-DFT and COD, and automated Rietveld refinement and ALIGNN-FF relaxation through a unified API at https://atomgpt.org/xrd. First, we used the AGAPI-XRD pipeline to evaluate the crystal structure of a variety of minerals in the RRUFF database that were experimentally characterized using powder x-ray diffraction. Next, we benchmarked the lattice parameter prediction fidelity of the AGAPI-XRD pipeline using a subset of the Alexandria PBE-hull dataset and the subset of RRUFF minerals that have known lattice parameters. AGAPI-XRD returns valid lattice parameters for 79.7\% of the RRUFF benchmark minerals and for 94.8--98.1\% of the Alexandria subset, while identifying a candidate structure for 93.8\% of RRUFF minerals. For this benchmark, pattern matching delivers the highest accuracy for known phases, while DiffractGPT extends structure generation to complex materials absent from existing databases. Together, AGAPI-XRD advances accessible, end-to-end automated crystal structure determination from powder XRD data.

cond-mat.mtrl-sci↗

arXiv2026

AGAPI-Agents: An Open-Access Agentic AI Platform for Accelerated Materials Design on AtomGPT.org

Agentic AI systems increasingly connect large language models (LLMs) to external scientific tools, yet whether and when tool access improves prediction accuracy remains uncharacterized. We present AGAPI (AtomGPT.org API), an open access platform integrating eight open-source LLMs with 18 REST endpoints (28 agent tools, 50 web apps) spanning materials databases, force fields, tight-binding band structures, X-ray diffraction, and protein structure. A three-evaluation residual decomposition on JARVIS-Leaderboard electronic-structure test sets separates agent pipeline fidelity from inherited density functional theory (DFT) functional bias. For bulk modulus and bandgap the agent reproduces JARVIS-DFT entries to numerical precision, so the experimental-reference degradation is functional bias, not agentic malfunction. On memorization-resistant test sets (57 defective supercells, 60 hypothetical compositions), tool-augmented mean absolute error (MAE) is below 0.005 eV versus 1.25 to 1.86 eV tool-free, confirming tools are indispensable where parametric knowledge is unavailable. We further demonstrate autonomous multi-step workflows including 10-operation defect-engineering pipelines. AGAPI is available at https://github.com/atomgptlab/agapi.

cs.AI↗

arXiv2026

BatteryMat: a hierarchical machine-learning and DFT framework for average-voltage screening of lithium-ion cathode materials

Density functional theory (DFT) predicts cathode voltages accurately but does not scale to the combinatorial chemical spaces of modern materials databases, while pure machine-learning surrogates are fast but cannot guarantee thermodynamic consistency. We introduce BatteryMat, a three-tier framework that promotes single-pass average-voltage prediction with the Atomistic Line Graph Neural Network (ALIGNN) as the primary screening signal across JARVIS-DFT, then validates survivors with ALIGNN-FF force-field delithiation profiles and automated PBE+U or optB88-vdW+U supercell DFT. The exchange-correlation functional is selected automatically by spacegroup, and the lithium metal reference is recomputed in the same plane-wave basis as the cathode runs, removing a systematic offset of about 1 V present in tabulated values. Trained on 7,610 ALIGNN-FF delithiation voltages, the ALIGNN predictor reproduces the force-field labels with a mean absolute error of 0.17 V and a coefficient of determination of 0.94; this measures distillation fidelity to the force-field protocol, not agreement with DFT or experiment. On four commercial chemistries (LiFePO4, LiMnPO4, LiMn2O4, LiCoO2) the DFT tier reproduces the experimental average voltage to within 0.3 V and the theoretical volumetric capacity to within 5%; a fifth, non-stoichiometric layered entry is carried as an edge case. The pipeline prioritises, rather than generates, existing structures: it ranks the lithium-containing JARVIS-DFT pool into 71 candidates and a scan of about 4.49 million Alexandria structures into 213, all surrogate-level leads awaiting DFT validation rather than confirmed cathodes. BatteryMat is available at https://github.com/atomgptlab/batterymat with a demo at https://atomgpt.org/battery.

cond-mat.mtrl-sci↗

arXiv2026

Hallucination Detector: A hybrid LLM and Semantic Scholar tool calling for detecting hallucination in scientific literature on AtomGPT.org

Large language models are now commonly used as partners in scientific writing, and this shift has brought a subtler type of failure: made-up references. Fabricated authors, bogus DOIs, wrongly assigned identifiers, and citations that merge elements from multiple genuine articles are now being inserted into manuscripts at a volume that traditional peer review was never meant to handle. Recent audits reveal that such references have already slipped through the review process and made their way into the published literature, including leading journals and conferences. Automated verification that operates at the speed and scale of modern content production has therefore become a necessary safeguard rather than a convenience. This work presents and evaluates the AtomGPT reference checker (https://atomgpt.org/hallucination_detector), an open, web-accessible tool that verifies citations against the scholarly literature by combining large-language-model field extraction with structured retrieval from Semantic Scholar. For each reference, the tool extracts the bibliographic fields, retrieves the closest matching real papers, and scores the agreement across title, authorship, and venue to produce a graded judgment of whether a citation is trustworthy, partially supported, or likely fabricated. We benchmark the tool against an externally curated set of confirmed hallucinated citations from accepted NeurIPS 2025 papers and find that it reliably flags the great majority of them.

cs.DL↗

arXiv2026

SlaKoNet-VQD: A universal Slater-Koster tight-binding Hamiltonian for variational quantum band-structure calculations on near-term hardware

Variational quantum algorithms such as VQE and VQD are promising for near-term electronic structure calculations, but for periodic solids their reach is limited by the cost of building a faithful second-quantized Hamiltonian, typically via DFT plus Wannierization or hand-fit tight-binding parameters. SlaKoNet addresses this by combining deep learning with the Slater-Koster tight-binding formalism to fit hopping and overlap parameters across 65 elements, enabling deterministic Hamiltonian construction for any crystal built from these elements. Here we couple a SlaKoNet model trained on JARVIS-TBmBJ with a Qiskit-based VQD algorithm, replacing costly Hamiltonian construction with a universal neural Hamiltonian generator. The resulting SlaKoNet-VQD workflow is structure-agnostic, differentiable, and suited to high-throughput bandstructure screening. We benchmark on silicon, recovering the full eight-band structure along the standard k-path with mean absolute deviation of 1.78 meV from exact diagonalization on a 3-qubit simulator, and extend to five conventional superconductors (Al, Ta, Nb, V, ZrN) with similar accuracy. We demonstrate execution on IBM Quantum hardware for a k-point ground-state calculation on aluminum (MAE ~0.37 eV). We further promote the Hamiltonian to a correlated Hubbard model solved via dynamical mean-field theory, recovering weakening correlations across group-5 metals and strong quasiparticle renormalization in La2CuO4, identifying the impurity problem as a natural quantum solver target. This pipeline enables high-throughput VQA benchmarking across the periodic table and gradient-based ansatz-Hamiltonian co-optimization for materials discovery. Web app: https://atomgpt.org/quantum.

cond-mat.mtrl-sci↗

arXiv2026

RamanGPT: Bidirectional Mapping Between Crystal Structures and Raman Spectra with Graph Neural Networks and Generative Transformers

Raman spectroscopy is one of the most accessible vibrational probes in materials laboratories, but its forward problem (structure to spectrum) is bottlenecked by the cost of density functional perturbation theory, and its inverse problem (spectrum to structure) typically relies on retrieval against curated references. We introduce RamanGPT, a deep-learning framework that addresses both directions for crystalline inorganic materials. The forward model, an Atomistic Line Graph Neural Network (ALIGNN), is trained on the 5{,}099-material Computational Raman Database and predicts 200-bin spectra over 50-1000~cm$^{-1}$ with 42.5\% having a cosine similarity greater than or equal to 0.354 suggesting qualitative features of the target spectrum. The model also shows some qualitative agreement with the approximate features and appearance of similar relative intensity of the modes to an experimental measurement of metallic 1T VSe$_{2}$, a system absent from the training set. The inverse model fine-tunes a large language model via Quantized Low-Rank Adaptation on Raman-plus-formula prompts, recovering lattice parameters with mean absolute errors of 1.14-2.16~Å and reduced-formula consistency of 86.8\% on 508 held-out materials. A cosine-similarity matcher and an inverse$\rightarrow$relax$\rightarrow$forward consistency loop are deployed at https://atomgpt.org/raman.

cond-mat.mtrl-sci↗

arXiv2026

From Photons to Electrons: Accelerated Materials Discovery via Random Libraries and Automated Scanning Transmission Electron Microscopy

The real-world implementation of materials prediction algorithms remains limited by persistent characterization bottlenecks in materials discovery, where photon-based probe techniques (e.g., XRD or Raman) impose long acquisition times and access latencies, restricting exploration to quasi-ternary composition spaces typically realized as compositional libraries. Here, we argue that a paradigm shift from photon- to electron-based characterization can realign materials characterization with modern high-throughput synthesis. We formulate cost functions and exploration strategies for STEM-based chemical and structural characterization and use Monte Carlo simulations to show that random chemical libraries, where compositionally distinct regions are co-located within a single specimen and interrogated in situ by electron spectroscopies, can sample high-dimensional composition and phase spaces with orders-of-magnitude greater effective coverage than conventional spread-library/X-ray approaches. We further demonstrate autonomous discovery on a laboratory STEM platform, where ML-based autotuning and scripted control enable iterative region selection and characterization without human intervention. Finally, we outline extensions to labeled or position-encoded libraries that preserve compositional and processing metadata, enabling joint exploration of composition and process spaces. Together, these results establish electron-based, ML-enabled STEM as a scalable pathway toward combinatorially rich materials discovery.

cond-mat.mtrl-sci↗

arXiv2025

DSAC: Low-Cost RowHammer Mitigation Using In-DRAM Stochastic and Approximate Counting Algorithm

This paper provides the fundamental mechanisms of two types of row activation-induced bit flips and proposes in-DRAM protection techniques. RowBleed occurs when a victim row experiences charge leakage due to transistor's threshold voltage lowering induced by long activation of a neighboring aggressor row. Therefore, this paper proposes Time-Weighted Counting for RowBleed mitigation, which assigns greater counter weights to rows that are activated for longer durations. On the other hand, RowHammer occurs when a victim row experiences electron injection due to frequent activation of a neighboring aggressor row. Similarly, Extended RowHammer, the phenomenon where victim rows are two rows beyond aggressor rows, is also caused by electron injection due to frequent activation of a neighboring aggressor row. Consequently, accurate detection of aggressor rows is crucial. Therefore, this paper proposes RowHammer mitigation algorithm named DSAC (in-DRAM Stochastic and Approximate Counting algorithm), which utilizes a replacement probability that adjusts based on the count of the old row. This paper introduces a RowHammer protection index called Maximum Disturbance, which measures the maximum accumulated number of row activations within an observation period. The experimental results demonstrate that DSAC can achieve 133x lower Maximum Disturbance than the state-of-the-art counter-based algorithm.

cs.CR↗

arXiv2018

Exeum: A Decentralized Financial Platform for Price-Stable Cryptocurrencies

Price stability has often been cited as a key reason that cryptocurrencies have not gained widespread adoption as a medium of exchange and continue to prove incapable of powering the economy of decentralized applications (DApps) efficiently. Exeum proposes a novel method to provide price stable digital tokens whose values are pegged to real world assets, serving as a bridge between the real world and the decentralized economy. Pegged tokens issued by Exeum - for example, USDE refers to a stable token issued by the system whose value is pegged to USD - are backed by virtual assets in a virtual asset exchange where users can deposit the base token of the system and take long or short positions. Guaranteeing the stability of the pegged tokens boils down to the problem of maintaining the peg of the virtual assets to real world assets, and the main mechanism used by Exeum is controlling the swap rate of assets. If the swap rate is fully controlled by the system, arbitrageurs can be incentivized enough to restore a broken peg; Exeum distributes statistical arbitrage trading software to decentralize this type of market making activity. The last major component of the system is a central bank equivalent that determines the long term interest rate of the base token, pays interest on the deposit by inflating the supply if necessary, and removes the need for stability fees on pegged tokens, improving their usability. To the best of our knowledge, Exeum is the first to propose a truly decentralized method for developing a stablecoin that enables 1:1 value conversion between the base token and pegged assets, completely removing the mismatch between supply and demand. In this paper, we will also discuss its applications, such as improving staking based DApp token models, price stable gas fees, pegging to an index of DApp tokens, and performing cross-chain asset transfer of legacy crypto assets.

cs.CR↗

arXiv2017

Riemannian approach to batch normalization

Batch Normalization (BN) has proven to be an effective algorithm for deep neural network training by normalizing the input to each neuron and reducing the internal covariate shift. The space of weight vectors in the BN layer can be naturally interpreted as a Riemannian manifold, which is invariant to linear scaling of weights. Following the intrinsic geometry of this manifold provides a new learning rule that is more efficient and easier to analyze. We also propose intuitive and effective gradient clipping and regularization methods for the proposed algorithm by utilizing the geometry of the manifold. The resulting algorithm consistently outperforms the original BN on various types of network architectures and datasets.

cs.LG↗

arXiv2015

Hessian-free Optimization for Learning Deep Multidimensional Recurrent Neural Networks

Multidimensional recurrent neural networks (MDRNNs) have shown a remarkable performance in the area of speech and handwriting recognition. The performance of an MDRNN is improved by further increasing its depth, and the difficulty of learning the deeper network is overcome by using Hessian-free (HF) optimization. Given that connectionist temporal classification (CTC) is utilized as an objective of learning an MDRNN for sequence labeling, the non-convexity of CTC poses a problem when applying HF to the network. As a solution, a convex approximation of CTC is formulated and its relationship with the EM algorithm and the Fisher information matrix is discussed. An MDRNN up to a depth of 15 layers is successfully trained using HF, resulting in an improved performance for sequence labeling.

cs.LG↗