Minisymposium

​With its ease of use combined with high performance, the Julia programming language is increasingly “closing the gap” between developers and users of scientific software. In this talk, I will briefly introduce the Julia language and motivate its use in computational science generally and atomistic modeling in particular. One of the key selling points of Julia is its potential for seamless interoperability across packages. While much of this comes “for free” from aspects of the language structure and type system, in a landscape with as many parameters, assumptions, and approximations as atomistic simulation, there still must be community effort to build consensus around a shared set of functions and formats for exchanging things like system geometry, energies, or forces. In recent years, the JuliaMolSim community has been working to develop a set of interfaces to serve this purpose. In this talk, I will give an overview of the history and current state of these efforts, starting with the AtomsBase interface and moving into more recent endeavors including AtomsCalculators and GeometryOptimization, and showcase a few examples where these interfaces have already seen success, as well as discuss plans for the future of these packages and the community more broadly.

This talk will introduce Julia-language tools for generating interatomic potentials from quantum mechanical reference data using the Atomic Cluster Expansion (ACE). The ACE construction provides a complete description of atomic environments, including invariance to overall translation and rotation as well as permutation of like atoms, making the resulting potentials systematically improvable and data efficient. The presentation will include demonstration of active learning strategies, as well as example applications.

Transport coefficients are computationally challenging quantities, measuring sensitivities in various fluxes for equilibrium systems subject to nonequilibrium perturbations, and govern first-order dynamical properties such as diffusivity, shear viscosity or thermal conductivity.

We propose a dual approach to compute these coefficients, based on fixing the magnitude of the flux exactly, and estimating the reciprocal sensitivity by measuring the average magnitude of the perturbation or forcing needed to induce it, reversing the role of the forcing and the flux compared to standard nonequilibrium methods.

We implement this approach using the Julia package for MD simulations Molly.jl, allowing for first examples of realistic computations giving promising results.

Physical data is increasingly openly accessible though it may be challenging to definitively rank the accuracy of different information sources. We demonstrate that multitask Gaussian process regression can leverage “datasets of opportunity” to efficiently construct surrogate models. In particular, we consider training sets constructed from coupled-cluster (CC) and density functional theory (DFT) data generated with multiple exchange-correlation functional approximations. The cost of CC calculation scales at a rate of N to the power of seven where N is the number of atoms in the system while DFT demonstrates relatively tractable N cubed scaling. We report that multitask surrogates can predict at CC level accuracy with a reduction to data generation cost by over an order of magnitude. This interdisciplinary effort has been facilitated by Julia packages for atomistic computation and for the custom design of optimization and Gaussian process models. If time permits, we will discuss the extension of our computational models to produce calibrated uncertainty indicators for each prediction.