서브메뉴
검색
Finite Size Effects, Machine Learning DFT Functionals and Intermolecular Interaction Energies From Self-Consistent GW
Finite Size Effects, Machine Learning DFT Functionals and Intermolecular Interaction Energies From Self-Consistent GW
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20250211153016
- ISBN
- 9798384045847
- DDC
- 541
- 저자명
- Chen, Yuting.
- 서명/저자
- Finite Size Effects, Machine Learning DFT Functionals and Intermolecular Interaction Energies From Self-Consistent GW
- 발행사항
- [Sl] : University of Michigan, 2024
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2024
- 형태사항
- 78 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 86-04, Section: B.
- 주기사항
- Advisor: Zgid, Dominika Kamila.
- 학위논문주기
- Thesis (Ph.D.)--University of Michigan, 2024.
- 초록/해제
- 요약Accurate treatment of electron correlation is crucial for computational and theoretical chemists as it influences reaction mechanisms, structural properties, and various spectroscopic quantities. Traditionally, the chemistry community has relied on Density Functional Theory (DFT) and wavefunction-based methods to address the problem of electron correlation. While relatively successful, these methods have limitations either in computational scalability or in the degree of electron correlation described.Green's functions provide an alternative formalism to study electron correlation. This thesis focuses on the self-consistent GW (scGW) approximation, which has been shown to describe a higher degree of electron correlation at a cost comparable to other more traditional wavefunction and Density Functional Theory methods.In Chapter 2 of this thesis, we show a way to account for finite size effects in periodic systems and demonstrate its applications to band structure diagrams. This work also demonstrates that Fock and Self-Energy matrix quantities are able to be extrapolated to the thermodynamic limit.In Chapter 3, this thesis presents a novel application of using Green's functions to train a DFT exchange-correlation functional that recovers the more strongly correlated scGW result at the cheaper DFT cost. It is found that the machine-learning-trained functional outperformed manually created functionals, showcasing the applicability of this approach. This work also showcases the limitations of machine learning by training just on scGW energies instead of enforcing exact conditions, as we find the lack of exact conditions causes the paramaterization to fail in regimes with fewer data points. These two works both focus on reproducing the accuracy of scGW calculations at lower computational costs.Chapter 4 presents an application of the self-consistent GW approximation to studying interaction energies in high-spin open-shell dimers. These systems have traditionally only been evaluated using DFT and wavefunction methods because it was believed scGW could not resolve the small quantity of interaction energies due to the usage of a numerical grid, and this work demonstrates that scGW is capable of studying these complex systems effectively while also highlighting some problems in the benchmark datasets that arise from using a restricted formalism compared to an unrestricted method.
- 일반주제명
- Physical chemistry
- 일반주제명
- Physics
- 일반주제명
- Analytical chemistry
- 일반주제명
- Computational physics
- 일반주제명
- Computer science
- 기타저자
- University of Michigan Chemistry
- 기본자료저록
- Dissertations Abstracts International. 86-04B.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008250123s2024 us c eng d■001000017164559
■00520250211153016
■006m o d
■007cr#unu||||||||
■020 ▼a9798384045847
■035 ▼a(MiAaPQ)AAI31631535
■035 ▼a(MiAaPQ)umichrackham005856
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a541
■1001 ▼aChen, Yuting.
■24510▼aFinite Size Effects, Machine Learning DFT Functionals and Intermolecular Interaction Energies From Self-Consistent GW
■260 ▼a[Sl]▼bUniversity of Michigan▼c2024
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2024
■300 ▼a78 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 86-04, Section: B.
■500 ▼aAdvisor: Zgid, Dominika Kamila.
■5021 ▼aThesis (Ph.D.)--University of Michigan, 2024.
■520 ▼aAccurate treatment of electron correlation is crucial for computational and theoretical chemists as it influences reaction mechanisms, structural properties, and various spectroscopic quantities. Traditionally, the chemistry community has relied on Density Functional Theory (DFT) and wavefunction-based methods to address the problem of electron correlation. While relatively successful, these methods have limitations either in computational scalability or in the degree of electron correlation described.Green's functions provide an alternative formalism to study electron correlation. This thesis focuses on the self-consistent GW (scGW) approximation, which has been shown to describe a higher degree of electron correlation at a cost comparable to other more traditional wavefunction and Density Functional Theory methods.In Chapter 2 of this thesis, we show a way to account for finite size effects in periodic systems and demonstrate its applications to band structure diagrams. This work also demonstrates that Fock and Self-Energy matrix quantities are able to be extrapolated to the thermodynamic limit.In Chapter 3, this thesis presents a novel application of using Green's functions to train a DFT exchange-correlation functional that recovers the more strongly correlated scGW result at the cheaper DFT cost. It is found that the machine-learning-trained functional outperformed manually created functionals, showcasing the applicability of this approach. This work also showcases the limitations of machine learning by training just on scGW energies instead of enforcing exact conditions, as we find the lack of exact conditions causes the paramaterization to fail in regimes with fewer data points. These two works both focus on reproducing the accuracy of scGW calculations at lower computational costs.Chapter 4 presents an application of the self-consistent GW approximation to studying interaction energies in high-spin open-shell dimers. These systems have traditionally only been evaluated using DFT and wavefunction methods because it was believed scGW could not resolve the small quantity of interaction energies due to the usage of a numerical grid, and this work demonstrates that scGW is capable of studying these complex systems effectively while also highlighting some problems in the benchmark datasets that arise from using a restricted formalism compared to an unrestricted method.
■590 ▼aSchool code: 0127.
■650 4▼aPhysical chemistry
■650 4▼aPhysics
■650 4▼aAnalytical chemistry
■650 4▼aComputational physics
■650 4▼aComputer science
■653 ▼aElectronic structure theory
■653 ▼aGreen's functions
■653 ▼aComputational scalability
■653 ▼aSelf-consistent GW
■653 ▼aDensity Functional Theory
■690 ▼a0494
■690 ▼a0486
■690 ▼a0216
■690 ▼a0605
■690 ▼a0984
■690 ▼a0800
■71020▼aUniversity of Michigan▼bChemistry.
■7730 ▼tDissertations Abstracts International▼g86-04B.
■790 ▼a0127
■791 ▼aPh.D.
■792 ▼a2024
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17164559▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.
