No abstract

We show that the Gaussian Approximation Potential machine learning framework can describe complex magnetic potential energy surfaces, taking ferromagnetic iron as a paradigmatic challenging case. The training database includes total energies, forces, and stresses obtained from densityfunctional theory in the generalized-gradient approximation, and comprises approximately 150,000 local atomic environments, ranging from pristine and defected bulk configurations to surfaces and generalized stacking faults with different crystallographic orientations. We find the structural, vibrational and thermodynamic properties of the GAP model to be in excellent agreement with those obtained directly from first-principles electronic-structure calculations. There is good transferability to quantities, such as Peierls energy barriers, which are determined to a large extent by atomic configurations that were not part of the training set. We observe the benefit and the need of using highly converged electronic-structure calculations to sample a target potential energy surface. The end result is a systematically improvable potential that can achieve the same accuracy of densityfunctional theory calculations, but at a fraction of the computational cost.

In this work, we have developed quantitative structure-property relationship (QSPR) models using advanced machine learning algorithms that can rapidly and accurately recognize high-performing metal organic framework (MOF) materials for CO2 capture. More specifically, QSPR classifiers have been developed that can, in a fraction of a section, identify candidate MOFs with enhanced CO2 adsorption capacity (>1 mmol/g at 0.15 bar and >4 mmol/g at 1 bar). The models were tested on a large set of 292 050 MOFs that were not part of the training set. The QSPR classifier could recover 945 of the top 1000 MOFs in the test set while flagging only 10% of the whole library for compute intensive screening. Thus, using the machine learning classifiers as part of a high-throughput screening protocol would result in an order of magnitude reduction in compute time and allow intractably large structure libraries and search spaces to be screened.

The charge equilibration (QEq) method has been parametrized to reproduce the ab-initio-derived electrostatic potential in a large and diverse training set of 543 metal organic frameworks (MOFs) containing the most popular Zn, Cu, and V structural building units (SBUs), 52 different organic carboxylate- and nitrogen-capped SBUs, and the 17 functional groups. The MOF electrostatic-potential-optimized charge scheme, or MEPO-QEq, was validated by evaluating the CO2 uptake and heats of adsorption in the 543 member training set and a nonoverlapping 693 member validation set. Compared with the results obtained from ab-initio-derived charges, the MEPO-QEq charges give Pearson (linear) and Spearman (rank-order) correlation coefficients of >0.97 for these two sets. MEPO-QEq enables near-ab-initio quality nonbonded electrostatic interactions to be evaluated using the fast QEq method for fast and accurate virtual high-throughput screening of gas-adsorption properties in MOFs.