Accelerating Quantum Many Body Calculations for Strongly Correlated Systems

Student thesis: Doctoral ThesisDoctor of Philosophy

Abstract

Harnessing strong electron correlations at the atomic scale presents tantalising opportunities to
revolutionise commercial electronics by going beyond silicon-based technologies. Their precise
control is also of fundamental scientific interest, through which the remarkable manifestations
of the quantum nature of matter can be understood more thoroughly. However, due to the
computational intractability of the quantum many-body problem, modelling strongly correlated
electron systems is notoriously difficult, and is often entirely out of reach for those without
considerable supercomputing resources. Moreover, for materials systems with correlated driven
mulitfunctional properties the application of standard approximations can fail and necessitates
the use of numerical approximations with a considerable computational footprint.

In this thesis, we introduce a series of advances, for both strongly correlated methodologies
and materials systems alike, that focus on the accurate modelling of many-body phenomena
using moderate computational resources. We introduce two novel error-free and data-driven
algorithms to efficiently solve the Anderson Impurity Solver (AIM) for Dynamical Mean Field
Theory (DMFT) calculations. Firstly, we present the highly extensible data-driven DMFT (d3mft)
software ecosystem that uses machine learning approaches to learn the solution of the singleband
Hubbard Model and predict its Mott Transition. Then, we propose the Quasicontinuoustime
Quantum Monte Carlo (QCTQMC) approach for solving materials specific multiorbital
models using a data-driven interpolation and extrapolation protocol. We show the success of this
method by modelling the prototypical correlated materials systems of SrVO3 and elemental g-
Cerium. Both procedures extrapolate from approximate to error-free solutions by systematically
predicting and then removing their inherent error, enabling faster calculations by only requiring
approximate solutions to the many-body problem, instead of computationally prohibitive exact
ones.

Subsequently, we go beyond prototypical correlated materials systems by examining the
correlated driven emergent phenomena at oxide interfaces. We first explore the phase-diagram
of interfacial archetypal lanthanum derived Mott insulators using Density Functional Theory
(DFT) and DMFT with spin-assisted random structure searches. We find a robust high-spin
antiferromagnetic ground state, accompanied by a family of similar metastable states, that persists
up to room temperature where spin-state transitions can be triggered by ThZ illumination,
and illustrates the utility of correlated interfaces for Mottronic-based technological devices. On
the other hand, we then reveal the detrimental effect that magnetic interfaces have on thin-film
Nb superconductors that are used in superconducting circuits for quantum computers. By
developing the ab initio theory of surface impedance we show that off-stoichiometry at the
surface of a superconductor introduces magnetic loss channels that have a deteriorating effect
on its overall functional properties.

The results presented in this thesis pave the way towards the development of complementary
approaches to the quantum many-body problem, from those that are data-driven to entirely quantum
mechanical, that are time-efficient, have a low computational footprint and are applicable
across a range of materials systems.
Date of Award1 Dec 2021
Original languageEnglish
Awarding Institution
  • King's College London
SupervisorCedric Weber (Supervisor)

Cite this

'