arXiv subjects

Yiqun Wang

Publications and source records attributed to Yiqun Wang.

58 records · Page 4Linked to original sources

arXiv2020

Featureless adaptive optimization accelerates functional electronic materials design

Electronic materials exhibiting phase transitions between metastable states (e.g., metal-insulator transition materials with abrupt electrical resistivity transformations) are challenging to decode. For these materials, conventional machine learning methods display limited predictive capability due to data scarcity and the absence of features impeding model training. In this article, we demonstrate a discovery strategy based on multi-objective Bayesian optimization to directly circumvent these bottlenecks by utilizing latent variable Gaussian processes combined with high-fidelity electronic structure calculations for validation in the chalcogenide lacunar spinel family. We directly and simultaneously learn phase stability and band gap tunability from chemical composition alone to efficiently discover all superior compositions on the design Pareto front. Previously unidentified electronic transitions also emerge from our featureless adaptive optimization engine. Our methodology readily generalizes to optimization of multiple properties, enabling co-design of complex multifunctional materials, especially where prior data is sparse.

cond-mat.mtrl-sci↗

arXiv2020

MGCN: Descriptor Learning using Multiscale GCNs

We propose a novel framework for computing descriptors for characterizing points on three-dimensional surfaces. First, we present a new non-learned feature that uses graph wavelets to decompose the Dirichlet energy on a surface. We call this new feature wavelet energy decomposition signature (WEDS). Second, we propose a new multiscale graph convolutional network (MGCN) to transform a non-learned feature to a more discriminative descriptor. Our results show that the new descriptor WEDS is more discriminative than the current state-of-the-art non-learned descriptors and that the combination of WEDS and MGCN is better than the state-of-the-art learned descriptors. An important design criterion for our descriptor is the robustness to different surface discretizations including triangulations with varying numbers of vertices. Our results demonstrate that previous graph convolutional networks significantly overfit to a particular resolution or even a particular triangulation, but MGCN generalizes well to different surface discretizations. In addition, MGCN is compatible with previous descriptors and it can also be used to improve the performance of other descriptors, such as the heat kernel signature, the wave kernel signature, or the local point signature.

cs.GR↗

arXiv2019

Assessing exchange-correlation functional performance in the chalcogenide lacunar spinels GaM$_4$Q$_8$ (M = Mo, V, Nb, Ta; Q = S, Se)

We perform systematic density functional theory (DFT) calculations to assess the performance of various exchange-correlation potentials $V_{xc}$ in describing the chalcogenide GaM$_4$Q$_8$ lacunar spinels (M=Mo, V, Nb, Ta; Q=S, Se). We examine the dependency of crystal structure (in cubic and rhombohedral symmetries), electronic structure, magnetism, optical conductivity, and lattice dynamics in lacunar spinels at four different levels of $V_{xc}$: the local density approximation (LDA), generalized gradient approximation (GGA), meta-GGA, and hybrid with fractional Fock exchange. We find that LDA significantly underperforms GGA and higher level functionals in predicting lattice constants. LDA also fails to properly describe the magnetism in the family, and for some compositions it does not find the ground state distorted crystal structure. The Perdew-Burke-Ernzerhof (PBE) and PBE revised for solids (PBEsol) GGA functionals perform reasonably well in predicting lattice constants as well as the electronic structures. We find that the GGA with an on-site Coulomb interaction (GGA$+U$) is unnecessary to produce a semiconducting state in the distorted polar $R3m$ phase. Plus Hubbard $U$ values ranging from 1$\sim$2 eV, however, improve the quantitative performance of the GGA functionals. The meta-GGA functional SCAN predicts reasonable lattice constants and electronic structures; it exhibits behavior similar to the GGA$+U$ functionals for small $U$ values. The hybrid functional HSE06 is accurate in predicting the lattice constants, but leads to a band gap of $\approx$1 eV in the rhombohedral phase, which is higher than the experimental estimation of 0.2 eV. Our findings suggest that accurate qualitative and quantitative simulations of the lacunar spinel family with DFT requires careful attention to the nuances of the exchange-correlation functional and considered spin structures.

cond-mat.mtrl-sci↗

arXiv2019

Symbolic Regression in Materials Science

We showcase the potential of symbolic regression as an analytic method for use in materials research. First, we briefly describe the current state-of-the-art method, genetic programming-based symbolic regression (GPSR), and recent advances in symbolic regression techniques. Next, we discuss industrial applications of symbolic regression and its potential applications in materials science. We then present two GPSR use-cases: formulating a transformation kinetics law and showing the learning scheme discovers the well-known Johnson-Mehl-Avrami-Kolmogorov (JMAK) form, and learning the Landau free energy functional form for the displacive tilt transition in perovskite LaNiO$_3$. Finally, we propose that symbolic regression techniques should be considered by materials scientists as an alternative to other machine-learning-based regression models for learning from data.

cond-mat.mtrl-sci↗