Preprint: Crystal Representation in the Reciprocal Space
Jan 20264D reciprocal space representation with scattering factors for symmetry-aware crystal structure matching and reconstruction now on arXiv.
Osman Goni Ridwan
AI + Materials Discovery • Generative Models • Crystal Structure Prediction
PhD candidate in Mechanical Engineering (UNC Charlotte). I build symmetry-aware generative models for crystals (LEGO-xtal) and high-throughput optimization workflows using ASE, VASP, and ML force fields (MACE / ALIGNN-FF). I collaborate across Computer Science and Computational Chemistry to accelerate functional materials discovery.
University of North Carolina at Charlotte · Materials Modeling and Informatics (MMI) Lab
Research Interests
- Atomistic materials modeling & crystal structure prediction
- AI generative models for materials (diffusion, LLMs, VAE with symmetry-aware representations)
- Energy, molecular and battery materials
News
AI-Assisted Crystal Generation Paper Published in npj Computational Materials
Jan 2026AI-Assisted Rapid Crystal Structure Generation Towards a Target Local Environment is now online. The work presents LEGO-xtal, a symmetry-aware generative framework capable of discovering 1,700+ novel sp²-carbon allotropes. Published in npj Comput. Mater., 12, 7.
New Preprint: Fully Differentiable Pipeline for Crystal Generation
Jan 2026Crystal Generation using the Fully Differentiable Pipeline and Latent Space Optimization now available on arXiv. This work extends crystal design with GPU-accelerated optimization in latent space.
Peptide Crystal Topology Work Accepted in Matter
Jan 2026Context-Adaptive Nanotopology in Peptide Crystals has been accepted for publication in Matter. Co-authored work exploring adaptive structure-property relationships in peptide systems.
Master's Degree Completed
Dec 2025Completed M.S. in Mechanical Engineering at UNC Charlotte with perfect 4.00/4.00 GPA. Research focused on symmetry-aware generative models and crystal design.
Awarded Best Poster at CITRANS Symposium
Aug 2025Recognition for research presented at the CITRANS Symposium, University of North Carolina at Charlotte.
LEGO-xtal Preprint and Code Released
Jul 2025Major update to LEGO-xtal preprint with complete code refactor and release on arXiv. Open-source implementation now available for reproducibility.
Education
- Ph.D., Mechanical Engineering, University of North Carolina at Charlotte — 2024–present (3rd year), GPA 4.00/4.00; Advisor: Prof. Qiang Zhu
- M.S., Mechanical Engineering, University of North Carolina at Charlotte — December 2025, GPA 4.00/4.00
- B.Sc., Naval Architecture & Marine Engineering, BUET — 2017–2022; Thesis on earth-pipe cooling system thermal performance (MARTEC 2022 proceedings, 2023).
Research Projects
LEGO-xtal: Symmetry-Aware Generative System for Crystal Design
PyTorch, Deep Generative Models, PyXtal, ASE, VASP, HPC
- Developed a symmetry-informed generative model that respects space-group constraints and periodicity for crystal structure generation.
- Expanded from 25 known low-energy sp²-carbon allotropes to over 1,700 new candidate structures within 0.5 eV/atom of graphite.
- Enabled targeted generation toward local chemical environments using geometry-aware latent representations.
LEGO-xtal-GPU: Differentiable CSP and Latent Optimization
GPU parallelization and batching, CVAE, SO(3) descriptors, differentiable optimization
- Implemented GPU-parallel batched evaluation of structure descriptors and objectives to scale candidate generation and refinement.
- Built a fully differentiable pipeline that backpropagates through representation-space objectives and couples them with latent-space refinement.
- Achieved ~5× speedup via batching and GPU acceleration, with improved success in meeting target local-environment constraints.
PyXtal-DFTB: High-Throughput Property Screening
ASE, DFTB+, automated workflows
- Built an automated pipeline for elastic-constant computation of organic crystals.
- Enabled mechanical screening of peptide hydrates in collaboration with experimental partners.
ML-Assisted X-ray Absorption and Optical Modeling of Boron Nitride
DFT (VASP), first-principles spectroscopy
- Conducted DFT simulations and collaborated with LLNL on ML-assisted X-ray absorption spectroscopy and optical modeling of boron nitride.
Reciprocal-Space Representations for Structure Matching
Scattering features, geometric embeddings, PyTorch
- Developed 4D scattering-informed representations for structure matching and reconstruction.
- Enabled quantitative comparison between simulation outputs and experimental signals.
xtal-builder: Agentic Crystal Design Assistant
LLM tool-calling, Ollama, LangChain/LangGraph, ASE, PyXtal, MLIP
- Building an agent that orchestrates structure generation, simulation, and refinement.
- Implements log parsing, failure detection, and auto-retry loops on HPC.
- Goal: autonomous crystal design from minimal user input.
Publications & Preprints
-
AI-Assisted Rapid Crystal Structure Generation Towards a Target Local Environment (LEGO-xtal) — 2026.
O. G. Ridwan, S. Pitié, M. S. Raj, D. Dai, G. Frapper, H‑F. Xue, Q. Zhu. Published in npj Computational Materials, 12, 7.
Journal | arXiv
published BibTeX
-
Crystal Generation using the Fully Differentiable Pipeline and Latent Space Optimization — 2026.
O. G. Ridwan, G. Frapper, H‑F. Xue, Q. Zhu.
arXiv
preprint BibTeX
-
Unveiling X‑ray Absorption Signatures of Boron Nitride via First‑Principles Simulation and Machine Learning — 2025.
W‑Y Sun, O. G. Ridwan, S. O. Kucheyev, Q. Zhu, L‑W Wan. Accepted for publication in Next Materials.
Journal
preprint BibTeX
-
Context‑Adaptive Nanotopology in Peptide Crystals — 2026.
V. Athiyarath, E. Naranjo, D. Dave, O. G. Ridwan, …, R. V. Ulijn, X. Chen. Accepted in Matter.
ChemRxiv
accepted BibTeX
-
Crystal Representation in the Reciprocal Space — 2026.
O. G. Ridwan, H. Xue, Y. Chen, H. Cherukuri, Q. Zhu. Preprint available on arXiv.
arXiv
preprint BibTeX
-
Numerical Computation of Thermal Performance of Earth Pipe Cooling Systems — 2023.
O. G. Ridwan, S. S. Sifat, F. E. Idrish, M. M. Karim. Proceedings of the 13th International Conference on Marine Technology (MARTEC 2022).
SSRN
conference BibTeX
Blog & Notes
I write short, reproducible posts showing how to rebuild my results (datasets, code, parameters) and commentary on ML for materials.
- 2025‑09‑01 Rebuilding LEGO‑xtal: from seed structures to 1,700 sp²‑allotropes End‑to‑end tutorial with data, config files, and common pitfalls.
- 2025‑08‑15 MACE vs ALIGNN‑FF for rapid CSP workflows When to pre‑relax with ALIGNN‑FF and refine with MACE (benchmarks + scripts).
- 2025‑07‑20 Crystal representations in reciprocal space Physics‑informed 4D scattering descriptors — intuition and code.
Contact
Email: oridwan@charlotte.edu