서브메뉴
검색
Hybrid Functionals in Real-Space Density Functional Theory
Hybrid Functionals in Real-Space Density Functional Theory
Detailed Information
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20260202105601
- ISBN
- 9798265404374
- DDC
- 620.118
- 저자명
- Jing, Xin.
- 서명/저자
- Hybrid Functionals in Real-Space Density Functional Theory
- 발행사항
- [Sl] : Georgia Institute of Technology, 2024
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2024
- 형태사항
- 118 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 87-05, Section: A.
- 주기사항
- Advisor: Suryanarayana, Phanish.
- 학위논문주기
- Thesis (Ph.D.)--Georgia Institute of Technology, 2024.
- 초록/해제
- 요약In this thesis, we introduce a series of advancements in hybrid exchange-correlation functionals within the generalized Kohn-Sham Density Functional Theory (DFT) [1, 2] framework, optimized for efficient real-space implementations[3, 4, 5]. Initially, we develop a real-space formalism that leverages the Kronecker product structure [6, 7] of the finite-difference Laplacian matrix, enabling efficient solutions for hybrid functionals in both isolated and periodic systems[8]. By integrating Fast Fourier Transform (FFT)schemes and supporting various boundary conditions, this formalism achieves up to an order-ofmagnitude speedup compared to traditional plane-wave methods, as verified through benchmarks with established codes. We further apply this method to an Ab Initio Molecular Dynamics (AIMD) study of liquid water, demonstrating accuracy in alignment with existing literature[9, 10, 11, 12, 13, 14].The work also includes a Graphics Processing Unit (GPU)-accelerated implementation of this formalism, enhanced by a multi-column Kronecker product solver. Detailed analysis of GPU-Central Processing Unit (CPU) and GPU-GPU communication identifies bottlenecks in GPU-GPU communication, yet achieves additional speedups of 6x to 8x with minimal computational nodes.Additionally, we present a linear scaling O(N) Spectral Quadrature (SQ) [15, 16, 17, 18] hybrid functional method, which calculates energy and stress using the density matrix rather than Kohn-Sham orbitals. Through a weak scaling test on various sizes of carbon systems, SQ hybrid demonstrates superiority over the Simulation Package for Ab-initio Real-space Calculations (SPARC) [3, 19, 4, 5] hybrid approach starting at approximately 32 atoms. A strong scaling analysis highlights SQ hybrid's efficient parallelism, distributing workload independently across grid points [15].For cases requiring higher accuracy, we introduce a second-order scaling O(N2) method using the Discrete Discontinuous Basis Projection (DDBP) [20] approach, which reduces the required Poisson solutions from O(N2) to O(N). Though this method achieves overall O(N2) complexity, additional operations involving face-splitting [21, 22] and matrixvector products contribute a large prefactor, making it less efficient than SPARC hybrid for smaller systems. Weak scaling tests on 1D Carbon Nanotube (CNT), 2D Silicene, and 3D Aluminum confirm a practical complexity of approximately O(N1.92), with face-splitting and matrix-vector operations identified as the main computational expenses.Overall, these contributions demonstrate enhanced computational efficiency for hybrid functionals, offering scalable solutions suited for both small and large system sizes.
- 일반주제명
- Nanotubes
- 일반주제명
- Fourier transforms
- 일반주제명
- Communication
- 일반주제명
- Carbon
- 일반주제명
- Decomposition
- 일반주제명
- Aluminum
- 일반주제명
- Energy
- 일반주제명
- Boundary conditions
- 일반주제명
- Linear algebra
- 일반주제명
- Mathematics
- 일반주제명
- Nanotechnology
- 기본자료저록
- Dissertations Abstracts International. 87-05A.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008260126s2024 us c eng d■001000017360653
■00520260202105601
■006m o d
■007cr#unu||||||||
■020 ▼a9798265404374
■035 ▼a(MiAaPQ)AAI32315981
■035 ▼a(MiAaPQ)GeorgiaTech76925
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a620.118
■1001 ▼aJing, Xin.
■24510▼aHybrid Functionals in Real-Space Density Functional Theory
■260 ▼a[Sl]▼bGeorgia Institute of Technology▼c2024
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2024
■300 ▼a118 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 87-05, Section: A.
■500 ▼aAdvisor: Suryanarayana, Phanish.
■5021 ▼aThesis (Ph.D.)--Georgia Institute of Technology, 2024.
■520 ▼aIn this thesis, we introduce a series of advancements in hybrid exchange-correlation functionals within the generalized Kohn-Sham Density Functional Theory (DFT) [1, 2] framework, optimized for efficient real-space implementations[3, 4, 5]. Initially, we develop a real-space formalism that leverages the Kronecker product structure [6, 7] of the finite-difference Laplacian matrix, enabling efficient solutions for hybrid functionals in both isolated and periodic systems[8]. By integrating Fast Fourier Transform (FFT)schemes and supporting various boundary conditions, this formalism achieves up to an order-ofmagnitude speedup compared to traditional plane-wave methods, as verified through benchmarks with established codes. We further apply this method to an Ab Initio Molecular Dynamics (AIMD) study of liquid water, demonstrating accuracy in alignment with existing literature[9, 10, 11, 12, 13, 14].The work also includes a Graphics Processing Unit (GPU)-accelerated implementation of this formalism, enhanced by a multi-column Kronecker product solver. Detailed analysis of GPU-Central Processing Unit (CPU) and GPU-GPU communication identifies bottlenecks in GPU-GPU communication, yet achieves additional speedups of 6x to 8x with minimal computational nodes.Additionally, we present a linear scaling O(N) Spectral Quadrature (SQ) [15, 16, 17, 18] hybrid functional method, which calculates energy and stress using the density matrix rather than Kohn-Sham orbitals. Through a weak scaling test on various sizes of carbon systems, SQ hybrid demonstrates superiority over the Simulation Package for Ab-initio Real-space Calculations (SPARC) [3, 19, 4, 5] hybrid approach starting at approximately 32 atoms. A strong scaling analysis highlights SQ hybrid's efficient parallelism, distributing workload independently across grid points [15].For cases requiring higher accuracy, we introduce a second-order scaling O(N2) method using the Discrete Discontinuous Basis Projection (DDBP) [20] approach, which reduces the required Poisson solutions from O(N2) to O(N). Though this method achieves overall O(N2) complexity, additional operations involving face-splitting [21, 22] and matrixvector products contribute a large prefactor, making it less efficient than SPARC hybrid for smaller systems. Weak scaling tests on 1D Carbon Nanotube (CNT), 2D Silicene, and 3D Aluminum confirm a practical complexity of approximately O(N1.92), with face-splitting and matrix-vector operations identified as the main computational expenses.Overall, these contributions demonstrate enhanced computational efficiency for hybrid functionals, offering scalable solutions suited for both small and large system sizes.
■590 ▼aSchool code: 0078.
■650 4▼aNanotubes
■650 4▼aFourier transforms
■650 4▼aCommunication
■650 4▼aCarbon
■650 4▼aDecomposition
■650 4▼aAluminum
■650 4▼aEnergy
■650 4▼aBoundary conditions
■650 4▼aLinear algebra
■650 4▼aMathematics
■650 4▼aNanotechnology
■690 ▼a0791
■690 ▼a0459
■690 ▼a0800
■690 ▼a0405
■690 ▼a0652
■71020▼aGeorgia Institute of Technology.
■7730 ▼tDissertations Abstracts International▼g87-05A.
■790 ▼a0078
■791 ▼aPh.D.
■792 ▼a2024
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17360653▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.


