Energy landscape exploration and global optimisation of hydrated molecular clusters

Hey, John Christopher ORCID: 0000-0002-2828-1006 (2024). Energy landscape exploration and global optimisation of hydrated molecular clusters. University of Birmingham. Ph.D.

Preview

Hey2024PhD.pdf
Text
Available under License All rights reserved.
Download (15MB) | Preview

Abstract

The Hofmeister series is a ranking of ions based on their ability to affect various properties of proteins and other biological macro-molecules in aqueous solutions. Initially described by Franz Hofmeister in the late 19\(^{th}\) century, the series categorises ions according to their effects on the solubility, stability, and activity of proteins. It is often used to understand and predict the behaviour of proteins in different ionic environments, particularly in the context of protein precipitation, folding, and stability. In this thesis I present a systematic study of the stepwise hydration of simple ions within the Hoffmeister series with the aim of elucidating both their structures and their energy landscapes. This study was undertaken using parallel basin hopping global optimisation with a hybrid of potentials, a cheap OPLS-like (Optimised Potentials for Liquid Simulations) potential was used to generate large databases of candidate minima. These putative minima were then reoptimised to a much tighter tolerance using a dispersion-corrected DFT scheme, to yield good minima that are shown to reproduce various experimental results. This technique has allowed us to provide good putative minima to larger sizes of clusters in a systematic manner than would otherwise be available. We establish common structural motifs across a variety of ions from across the Hoffmeister series, and show the effects of interpolating charges between two candidate ions has on the structural motifs present.

We then extend this study to encompass metal ions. We study the stepwise hydration of AuAg nanoclusters as cations, anions and uncharged clusters. This part of the study was conducted using the MEGA/GIGA genetic algorithms, directly at the DFT level. I reproduce some previous experimental infrared spectra for the anionic clusters, and then go on to present novel structures for the cationic system and the neutral system. I then categorise the structural motifs present in the clusters and compare these to those seen in the Hoffmeister ion study and present some computed infrared spectra.

Finally, I present a novel fusion of a parallel pool genetic algorithm with a Thresholding Algorithm I am calling BMPGA. The genetic algorithm is an entirely new implementation of the old BPGA written specifically with the optimisation of molecular clusters in mind. BMPGA implements a modified Deavan and Ho crossover scheme optimised for molecular systems, has massive thread-safe parallelism built-in by design, and implements a number of common and less common mutation schemes.1 The mutation schemes include the old classics such as single and multiple translations and rotations, as well as some more computationally demanding mutations such as short, high temperature Monte Carlo and molecular dynamics simulations. The hybridisation with the Thresholding scheme comes in to play in two different ways; it is available as an optional mutation scheme in the form of Threshold Monte Carlo. These mutations allow for the system to undergo infinite-temperature MC runs with a total-system energy cap applied the final configuration obtained is then subjected to a number of stochastic quenches. The other way that Thresholding is incorporated into the GA is that it can be applied to the population of minima within the GA itself, this leads to ensemble Thresholding and I show that this method allows for very rapid exploration of the landscape by allowing the population to escape wells that it may otherwise become trapped in. By lowering the energy cap over the course of the optimisation, good performance of this method is seen. I present benchmark results for this new GA using a variety of systems and potentials, from simple Lennard-Jones clusters, through some DFTB results to direct landscape exploration at the DFT level. I show that the parallel performance of this method is extremely good, and scales very well to a large number of CPU cores. This new GA method shows promise for allowing very large-scale optimisations to be performed on a large range of systems, only limited by the availability of HPC resources.

Type of Work:

Thesis (Doctorates > Ph.D.)

Award Type:

Doctorates > Ph.D.

Supervisor(s):