A generative model-driven pipeline for post-hoc prioritization of candidates in nonlinear optical materials discovery
Resumo
Discovering nonlinear optical (NLO) materials for quantum photonic applications remains challenging because of the vastness of chemical space and the high cost of first-principles calculations. Generative machine learning models, such as the Conditional Variational Autoencoder (CVAE-χ2), enable targeted generation of novel candidates, but their outputs do not guarantee physical feasibility. In this work, we propose a generative model-driven pipeline to post-hoc prioritize 1,000 synthetic candidates previously generated by a CVAE-χ2 conditioned to a band gap range of 2–4 eV. The pipeline sequentially applies band gap filtering, thermodynamic stability estimation via the energy above the convex hull (ΔEhull) obtained from the nearest neighbor in the Materials Project database, non-centrosymmetric symmetry selection, and a heuristic SPDC figure of merit. This multi-criteria data-mining strategy reduced the original candidate set to a single physically consistent material (Cristal_CVAE_382), which showed competitive performance compared with commercial NLO crystals such as PPLN, PPKTP, and BBO. The proposed approach provides an effective and reproducible screening strategy that bridges generative artificial intelligence and materials science, significantly narrowing the search space for subsequent DFT calculations and experimental synthesis in the pursuit of future quantum technologies.
Referências
Horton, M. K., Huck, P., Yang, R. X., et al. Accelerated data-driven materials science with the materials project. Nature Materials vol. 24, pp. 1522–1532, 2025.
Kingma, D. P. and Welling, M. Auto-encoding variational bayes, 2022.
Kwiat, P. G., Waks, E., White, A. G., Appelbaum, I., and Eberhard, P. H. Ultrabright source of polarization-entangled photons. Physical Review A 60 (2): R773–R776, Aug., 1999.
Lopes, A., Santos, C. E., and Freitas, R. Uma abordagem com autoencoders variacionais condicionais para geração de materiais nlo com bandgap alvo. In Anais do I Simpósio Brasileiro de Computação e Comunicação Quânticas. SBC, Porto Alegre, RS, Brasil, pp. 164–175, 2026.
Lopes, A., Santos, C. E. d. S., and Freitas-Rodrigues, R. d. Geração de materiais nlo para computação quântica usando aprendizado de máquina. In Anais do
Matsunoshita, K., Yamaguchi, Y., Hamaie, M., Horibe, M., Tanibata, N., Takeda, H., Nakayama, M., Karasuyama, M., and Kobayashi, R. Optimization of force-field potential parameters using conditional variational autoencoder. Science and Technology of Advanced Materials: Methods 3 (1): 2253713, 2023.
Sanchez-Lengeling, B. and Aspuru-Guzik, A. Inverse molecular design using machine learning: Generative models for matter engineering. Science 361 (6400): 360–365, 2018.
Sohn, K., Lee, H., and Yan, X. Learning structured output representation using deep conditional generative models. In Advances in Neural Information Processing Systems, C. Cortes, N. Lawrence, D. Lee, M. Sugiyama, and R. Garnett (Eds.). Vol. 28. Curran Associates, Inc., 2015.
Ward, L., Dunn, A., Faghaninia, A., Zimmermann, N. E., Bajaj, S., Wang, Q., Montoya, J., Chen, J., Bystrom, K., Dylla, M., Chard, K., Asta, M., Persson, K. A., Snyder, G. J., Foster, I., and Jain, A. Matminer: An open source toolkit for materials data mining. Computational Materials Science vol. 152, pp. 60–69, 2018.
Xie, C., Tikhonov, E., Chu, D., et al. A prediction–driven database to enable rapid discovery of nonlinear optical materials. Science China Materials vol. 66, pp. 4473–4479, 2023.
