Isaac Tamblyn

Identifying materials with strong performance in a specific application, particularly when the origin of that performance is unclear or difficult to compute, remains a central challenge in materials science. Trial-and-error exploration is prohibitively expensive due to the vastness of the materials space. A more practical strategy is to search for new materials within the proximity of known compounds exhibiting the desired property. This requires defining a meaningful representation to assess materials similarity. Structural fingerprints are commonly used, yet structural similarity often fails to translate into similarity in properties. Electronic fingerprints, such as density of states or band structure, were proposed as a better alternative, but computing them for >100 000 materials is still too costly for rapid screening. Here, we present ProDosNet, a graph convolutional network trained on atom- and orbital-resolved projected density of states (PDOS) data, capable of predicting electronic structure at extremely low computational cost. Using this model, we generated PDOS fingerprints for all compounds in the Materials Project database and clustered them by orbital-resolved PDOS similarity. We demonstrate that these electronic fingerprints allow finding materials with similar electronic properties but drastically different structures for applications in photovoltaics, catalysis, and batteries.

Radiation-induced segregation (RIS) is the redistribution of alloying elements near defect sinks under irradiation. It is commonly modeled using rate theory (RT) to describe the spatial and temporal evolution of solute concentrations. The accuracy of RT predictions strongly depends on the reliability of the constants and parameters used in the model. However, inconsistencies in reported parameter values across the literature, stemming from differences in calculation methods, approximations, and the complexity of the studied systems, limit the generalizability of RT models. In this work, we employ atom probe tomography to measure solute distributions at grain boundaries in 304L stainless steel irradiated with 2 MeV protons to a dose of 1.5 displacements per atom (dpa) at 100$^{\circ}$C and 360$^{\circ}$C. These measurements are complemented by additional RIS data from the literature, encompassing a range of temperatures, irradiation types, doses, and alloy chemistries. Two conventional optimization techniques, a gradient-based constrained nonlinear optimizer and a Genetic Algorithm, which is a population-based approach, along with physics-informed neural network (PINN) were employed in this work. Using these methods we extract the migration energies of Fe, Cr, Ni, and Si via vacancy and interstitial mechanisms in irradiated austenitic stainless steels. Our results show that, for a single dataset, PINNs and constrained gradient-based optimization converge to similar migration energy values. Notably, PINNs achieve this convergence approximately 25 times faster. In contrast, genetic algorithms are two orders of magnitude slower and yield less reliable migration energy estimates, with notable discrepancies observed.

Identifying the diffusion path of point defects is a critical step in understanding their evolution and the mechanisms of related phenomena. Defect diffusion occurs at small length and time scales, with impacts on material properties that may continue to evolve over ns to μs, ms, and the continuum scale (s, min, etc., and cm, m, etc.). The time scale accessible to molecular dynamics (MD) simulations is limited by small step sizes, typically in the fs range. Thus, surrogate models of MD simulations through machine learning (ML)-based algorithms are of great interest, especially for complex systems such as high entropy alloys (HEAs). In this work, dynamics governing vacancy migration in HEA were approximated with graph convolutional network (GCN) models as ansatzes for kinetic Monte Carlo (KMC) rate catalogs. Network design considered that diffusion in crystalline solids generally depends on interactions between defects and their immediate neighbor atoms. Graphs represented the vacancy surroundings, MD-generated trajectories provided training and comparison datasets, and unsupervised GCN models approximated interatomic dynamics governing vacancy migration in HEAs as ansatzes for KMC. A proof-of-concept model trained on MD data for the Fe, Ni, Cr, Co, and Cu HEA environment was used with two different neighbor interactions to assess the feasibility of training a GCN to predict vacancy defect transition rates in the HEA environment. The resulting setup rapidly generated MD-formatted synthetic trajectories based on dynamics learned from the MD training set, with a time acceleration of roughly two orders of magnitude and a similar diffusion coefficient to MD observations. Additionally, Nudged Elastic Band (NEB) calculations were performed on randomly generated FeNiCrCoCu HEA structures to determine vacancy migration barriers across nearest-neighbor sites. Transition probabilities for each jump, categorized by atomic type, were extracted from these calculations. NEB-based and GCN-based approaches led to similar outcomes.

We introduce a deep neural network (DNN) framework called the Real-space Atomic Decomposition NETwork (radnet), which is capable of making accurate predictions of polarization and of electronic dielectric permittivity tensors in solids and aims to address limitations of previously available machine learning models for Raman predictions in periodic systems. This framework builds on previous, atom-centered approaches while utilizing deep convolutional neural networks. We report excellent accuracies on direct predictions for two prototypical examples: GaAs and BN. We then use automatic differentiation to efficiently calculate the Born-effective charges, longitudinal optical-transverse optical (LO-TO) splitting frequencies, and Raman tensors of these materials. We compute the Raman spectra, and find agreement with ab initio results. Lastly, we explore ways to generalize the predictions of polarization while taking into account periodic boundary conditions and symmetries.

Finding materials with good performance in a specific application, especially when the origin of good performance is not well understood or not easily computable, is a major challenge in materials science. Trial-and-error random exploration is prohibitively expensive due to the vastness of the materials space. A more practical approach is to search for new materials within the proximity of known compounds that possess the desired property. In such an approach, assessing materials’ similarity requires deriving some fingerprint relevant for material’s performance. Typically, material’s structure is used as the fingerprint, which often does not translate into similarity in properties. Electronic structure fingerprints, e.g., density of states (DOS) or electronic band structure, were proposed as a better alternative, however, the computational cost of their calculation on the scale of 100,000 materials remains too high for rapid exploration. In this work, we developed a Graph Convolutional Network (GCN) ProDosNet which is trained on orbital-resolved and atom-resolved projected density of states (PDOS) data and is capable of predicting the electronic structure of materials at extremely low computational cost. With this model, we were able to generate PDOS fingerprints for all compounds present in the Materials Projects database and cluster them by similarity of their orbital-resolved PDOS. We demonstrate that these electronic fingerprints allow finding materials with similar electronic properties but drastically different structures for applications in photovoltaics, catalysis, and batteries.

Reinforcement learning (RL) has been shown to learn sophisticated control policies for complex tasks including games, robotics, heating and cooling systems and text generation. The action-perception cycle in RL, however, generally assumes that a measurement of the state of the environment is available at each time step without a cost. In applications such as materials design, deep-sea and planetary robot exploration and medicine, however, there can be a high cost associated with measuring, or even approximating, the state of the environment. In this paper, we survey the recently growing literature that adopts the perspective that an RL agent might not need, or even want, a costly measurement at each time step. Within this context, we propose the Deep Dynamic Multi-Step Observationless Agent (DMSOA), contrast it with the literature and empirically evaluate it on OpenAI gym and Atari Pong environments. Our results, show that DMSOA learns a better policy with fewer decision steps and measurements than the considered alternative from the literature.

We show that a neural network originally designed for language processing can learn the dynamical rules of a stochastic system by observation of a single dynamical trajectory of the system, and can accurately predict its emergent behavior under conditions not observed during training. We consider a lattice model of active matter undergoing continuous-time Monte Carlo dynamics, simulated at a density at which its steady state comprises small, dispersed clusters. We train a neural network called a transformer on a single trajectory of the model. The transformer, which we show has the capacity to represent dynamical rules that are numerous and nonlocal, learns that the dynamics of this model consists of a small number of processes. Forward-propagated trajectories of the trained transformer, at densities not encountered during training, exhibit motility-induced phase separation and so predict the existence of a nonequilibrium phase transition. Transformers have the flexibility to learn dynamical rules from observation without explicit enumeration of rates or coarse-graining of configuration space, and so the procedure used here can be applied to a wide range of physical systems, including those with large and complex dynamical generators.

Demonstration of a new open source Python library for simulating chemistry experiments as a gymnasium-API, reinforcement learning environment. Allowing learning policies for material design tasks or pipelines using a modular, extendable design.

We show that cellular automata can classify data by inducing a form of dynamical phase coexistence. We use Monte Carlo methods to search for general two-dimensional deterministic automata that classify images on the basis of activity, the number of state changes that occur in a trajectory initiated from the image. When the number of time steps of the automaton is a trainable parameter, the search scheme identifies automata that generate a population of dynamical trajectories displaying high or low activity, depending on initial conditions. Automata of this nature behave as nonlinear activation functions with an output that is effectively binary, resembling an emergent version of a spiking neuron.