Accurate and Ultrafast Simulation of Molecular Recognition and Assembly on Metal Surfaces in Four Dimensions

Abstract

Understanding molecular interactions with metal surfaces in high reliability is critical for the development of catalysts, sensors, and therapeutics. Obtaining accurate experimental data for a wide range of surfaces remains a critical bottleneck and quantum-mechanical data remain speculative due to high uncertainties and limitations in scale. We report molecular dynamics simulations of adsorption energies and assembly of organic molecules on elemental metal surfaces using the INTERFACE force field (IFF). The force field-based simulations reach up to 8 times higher accuracy than density functional calculations at a million-fold faster speed, as well as more than 1 order of magnitude higher accuracy than other force fields relative to accurate measurements by single-crystal adsorption calorimetry. Uncertainties of prior computational methods are effectively reduced from on the order of 100% to less than 10% and validated by experimental data from multiple sources. Specifically, we describe the molecular interactions of benzene and naphthalene with even and defective platinum surfaces across a wide range of surface coverage in depth. We discuss molecular-scale influences on the heat of adsorption and clarify the definition of surface coverage. The methods can be applied to 18 metals to accurately predict binding and assembly of organic molecules, ligands, electrolytes, biological molecules, and gases without additional fit parameters.

Accurate Force Fields for Atomistic Simulations of Oxides, Hydroxides, and Organic Hybrid Materials up to the Micrometer Scale

“…Similar agreement of computed and experimentally measured adsorption energies and conformations of organic molecules, biopolymers, and gases on nanostructured metals, minerals, and 2D materials on the order of 10% or better was previously demonstrated using IFF parameters and combinations with CHARMM, AMBER, OPLS-AA, PCFF, and other (bio)­organic force fields. ,,,− ,,,,, …”

“…DFT has had a profound impact in materials design and will continue to drive many exciting computational research areas. , At the same time, we find much better reliability of IFF compared to DFT across the board, including the computed densities (<0.3% vs 4% error), surface properties (8% vs 25%, sometimes up to 50%), and bulk modulus (6% vs 15%) (Table , Table , Table S2, and Section S4 in the Supporting Information). Similar improvements can be expected for computed defect energies and adsorption energies at interfaces . Therefore, using training data from IFF molecular dynamics simulations for machine learned (ML) potentials will lead to several times higher reliability than using training data from ab initio MD simulations. − The differences will be particularly significant when considering a wider set of properties, such as structures, energy differences, elastic properties, adsorption, electrolyte interfaces, and organic and biological interfaces that are critical to design real materials and devices.…”

“…65,92,93,133,134 Similar agreement of computed and experimentally measured adsorption energies and conformations of organic molecules, biopolymers, and gases on nanostructured metals, minerals, and 2D materials on the order of 10% or better was previously demonstrated using IFF parameters and combinations with CHARMM, AMBER, OPLS-AA, PCFF, and other (bio)organic force fields. 25,29,30,[34][35][36]56,57,59,135,136 Simulations of biomolecule−oxide and biomolecule−hydroxide assemblies in solution use the same types of models as for water and organic molecules and are expected to perform with similar accuracy. As shown in prior work, IFF is well suited to characterize peptide and protein interactions with various minerals.…”

“…Similar improvements can be expected for computed defect energies and adsorption energies at interfaces. 135 Therefore, using training data from IFF molecular dynamics simulations for machine learned (ML) potentials will lead to several times higher reliability than using training data from ab initio MD simulations. 162−164 The differences will be particularly significant when considering a wider set of properties, such as structures, energy differences, elastic properties, adsorption, electrolyte interfaces, and organic and biological interfaces that are critical to design real materials and devices.…”

“…The proposition of "accurate" global machine learning force fields, so-called "modern machine learning force fields (MLFFs)" 164 therefore so far remains a myth. This work and earlier studies [26][27][28][29][30][31][32][33][34][35][36][37]57,135,168,169 demonstrate that IFF performs several times better. MLFFs lack systematic benchmarks in virtually any domain, including inorganic compounds as demonstrated here as well as biomolecular systems, e.g., DNA and proteins in solution, in comparison to biomolecular force fields.…”