Goal Driven Learning of Quantum Chemical Energy Surfaces using Multi-Fidelity Bayesian Optimization

AI_CDT_DecisionMaking

Details

Perhaps the most important task in computational materials modelling is efficient sampling of the complex interatomic potential energy surface (PES), dictated by chemical formula unit, and relative orientation of those species [1]. While exploring this vast configuration space is an important problem from a practical perspective, where, for example, minima correspond to synthesisable compounds, it remains a particularly formidable computational task. This is due to the combination of a high-dimensional search space, mixed continuous and discrete variables and a highly non-convex energy landscape. This problem is compounded by the high computational cost and poor scaling of the most accurate, quantum mechanical methods, which provide “ground truth” values of relative energies.

Over the past decade, Artificial Intelligence (AI) techniques have provided a potential resolution to this problem. Modern Machine Learning Interatomic Potentials (MLIPs) are capable of accurately and efficiently interpolating the full PES, matching the predictions of high-fidelity electronic structure models at a tiny fraction of the computational cost. Underlying many of these improvements is the Atomic Cluster Expansion (ACE) parameterisation [2-5], which provides a physically transparent and universal descriptor of local atomic environments through a cluster-based expansion of local atomic energies. When combined with message-passing graph neural networks (GNN-IPs), where non-linear activations are physically constrained by equivariance under translation and rotation, they constitute the current state of the art MLIP architecture [6]. Implementation of GNN-IP Foundation models (FMs), which exploit transfer learning of “chemical intuition” when trained over large chemical databases, would provide an extremely promising solution to the problems outlined above [7,8].

A key barrier to the development of GNN-IP FMs is the lack of high-quality, expressive data. While historical data is available, typically from parametrisation of older MLIPs, this will not generically benefit a different algorithm, as models trained via active learning fail for different configurations, and in different ways, which limits generalisation to out-of-distribution molecules [9-11]. Therefore, current GNN-IP FMs often require substantial retraining when applied to new systems and may not generalize well due to these limitations.

This project will develop new methods for single- and mutli-fidelity Bayesian optimization (MFBO) of complex chemical PES surfaces, balancing the cost and utility of each QM calculation to efficiently search the PES [12, 13]. A key step in achieving this will be the implementation of a probabilistic measure on top of current GNN-IP models, which would allow implementation of appropriate acquisition functions to guide PES sampling. This will be achieved through the extension of the linear ACE framework, by implementing “Bayesian ACE” descriptors [14], which naturally incorporate uncertainty into cluster expansion coefficients.

This has numerous practical advantages, such as improved FM training time, more efficient data usage, and significantly improved accuracy. For example, an MFBO approach can feasibly allow MLIP parameterisation using expensive, highest-fidelity “gold standard” QM data, which would open up crucial new applications. While this project focuses on applying the Bayesian ACE framework to MLIP models, cluster expansions provide a formally complete expansion of the statistical partition function of generic multicomponent systems [15-18]. Rigorous implementation and understanding of the proposed approach is therefore highly likely to find applications beyond atomistic materials modelling. Additionally, the hierarchy of possible MLIP models also provide a theoretically well understood set of benchmark systems on which to quantify transfer learning between fidelity levels, and across models of different complexity [19].

Before you apply

We strongly recommend that you contact the supervisor(s) for this project before you apply. Please include details of your current level of study, academic background and any relevant experience and include a paragraph about your motivation to study this PhD project.

How to apply:

Please apply through the below link for the PhD Artificial Intelligence CDT:

https://pgapplication.manchester.ac.uk/psc/apply/EMPLOYEE/SA/s/WEBLIB_ONL_ADM.CIBAA_LOGIN_BT.FieldFormula.IScript_Direct_Login?Key=UMANC1251000021489F

When applying, you’ll need to specify the full name of this project, the name of your supervisor, if you already having funding or if you wish to be considered for available funding through the university, details of your previous study, and names and contact details of two referees.

Your application will not be processed without all of the required documents submitted at the time of application, and we cannot accept responsibility for late or missed deadlines. Incomplete applications will not be considered.

After you have applied you will be asked to upload the following supporting documents:

• Final Transcript and certificates of all awarded university level qualifications
• Interim Transcript of any university level qualifications in progress
• CV
• Supporting statement: A one or two page statement outlining your motivation to pursue postgraduate research and why you want to undertake postgraduate research at Manchester, any relevant research or work experience, the key findings of your previous research experience, and techniques and skills you’ve developed. (This is mandatory for all applicants and the application will be put on hold without it).
• Contact details for two referees (please make sure that the contact email you provide is an official university/work email address as we may need to verify the reference)
• English Language Certificate (if applicable)

Equality, diversity and inclusion is fundamental to the success of The University of Manchester, and is at the heart of all of our activities. We know that diversity strengthens our research community, leading to enhanced research creativity, productivity and quality, and societal and economic impact. (Equality, diversity and inclusion | The University of Manchester)

We actively encourage applicants from diverse career paths and backgrounds and from all sections of the community, regardless of age, disability, ethnicity, gender, gender expression, sexual orientation and transgender status.

We also support applications from those returning from a career break or other roles. We are dedicated to supporting work-life balance and offer flexible working arrangements to accommodate individual needs. Our selection process is free from bias, and we are committed to ensuring fair and equal opportunities for all applicants.

_{featuredproject22_nov24}