Predicting the dissolution kinetics of silicate glasses by topology-informed machine learning

Abstract

Machine learning (ML) regression methods are promising tools to develop models predicting the properties of materials by learning from existing databases. However, although ML models are usually good at interpolating data, they often do not offer reliable extrapolations and can violate the laws of physics. Here, to address the limitations of traditional ML, we introduce a "topologyinformed ML" paradigm-wherein some features of the network topology (rather than traditional descriptors) are used as fingerprint for ML models-and apply this method to predict the forward (stage I) dissolution rate of a series of silicate glasses. We demonstrate that relying on a topological description of the atomic network (i) increases the accuracy of the predictions, (ii) enhances the simplicity and interpretability of the predictive models, (iii) reduces the need for large training sets, and (iv) improves the ability of the models to extrapolate predictions far from their training sets. As such, topology-informed ML can overcome the limitations facing traditional ML (e.g., accuracy vs. simplicity tradeoff) and offers a promising route to predict the properties of materials in a robust fashion.

Analytical model of the network topology and rigidity of calcium aluminosilicate glasses

“…19,20 Various TCT-based models have been proposed over the past decades to predict glass-forming ability, glass-transition temperature, liquid fragility, hardness, stiffness, dissolution rate, etc. 19,[21][22][23][24][25][26][27][28] The success of TCT is based on the fact that many macroscopic properties of disordered materials primarily depend on the topology of the atomic structure, whereas other structural details only have a second-order effect. 29 As such, TCT reduces complex disordered atomic networks into simpler structural trusses, 22 wherein some nodes (the atoms) are connected to each other by some topological constraints (the chemical bonds).…”

Literature mining for alternative cementitious precursors and dissolution rate modeling of glassy phases

“…Reported rates (n = 802 rates from 105 distinct glasses) are extracted and aggregated from 33 different papers-pulled from cement, 29,30,[33][34][35][36] geochemistry [44][45][46][47][48]50,[52][53][54][55][56] and nuclear waste and glass science 57,[60][61][62][63][64][65][66][67][68][69][70][71][72][73][74] literature. While other studies have analyzed published data to discover trends and ascertain relationships between driving factors and dissolution rates 32,81 or other material properties; [82][83][84] as far as the authors are aware, this is both the largest data set specifically focused on glass dissolution and the first time such a study has tested cross-discipline relationships.…”

Accelerated design and characterization of non-uniform cellular materials via a machine-learning based framework

“…Thanks to recent advances in computational power and robust algorithms, we have witnessed the rise of an interdisciplinary field in computational material science [6][7][8][9] . Specifically, researchers have already shown success in utilizing ML-based or other AI methods in two major categories: (1) to accelerate the prediction of material properties for specific applications [10][11][12][13][14][15][16][17][18][19][20][21][22][23][24] , and (2) to accelerate the on-demand design and the optimization of material microstructure and composition for targeted properties [25][26][27][28][29][30][31][32][33][34][35][36][37][38][39][40][41][42] . These promising studies have shown superior effectiveness of ML techniques compared to traditional computational modeling or experimental measurements on a variety of materials.…”