Matrix methods in computational quantum mechanics
Useful for atoms, molecules, crystals
Example: Gaussian basis functions
Outline
- Representing the wavefunction in a basis
- Vector in Hilbert space
- Norms
- Orthonormal bases
- transformations of basis - unitary transformations
- Overlap matrices for non-orthonormal bases
- Matrix representation of operators
- unitary transformation matrices
- Hermitian matrices for physical observables
- Examples of Hamiltonian in often used bases
- Plane waves
- Fourier analysis - complete - orthonormal
- Eigenstates for free particles
- Kinetic energy operator
- Potential energy - convolution in Fourier space
- Gaussians
- Localized - complete - eigenstates for harmonic oscillator
- Non-orthogonal and overcomplete if expanded around
multiple sites and/or multiple widths
- All needed matrix elements are analytic
- Detailed derivations of Gaussian matrix elements in Thijssen, 4.8
- Analytic advantages NOT so elegant for DFT calculations
- GTOs and STOs
- GTO - sum of Gaussians
- STO- "fully contracted", i.e., one sum of Gaussians for an
atomic-like orbital
- STO-NG - standard notation for N Gausians in an STO
- Numerical atomic orbitals
Localized - realistic - difficult to use
- Finding the energy eigenstates
- Hamiltonian matrix for simple potential - Kohn-Sham Eqs.
- Determinants
- Eigenvalues and eigenvectors
- Variational theorems
- McDonald's theorem for each eigenvalue
- Hartree-Fock more complex - Roothan Eqs.
(Referred to as the "Pople-Nesbet" equations for the
Gaussian basis in Thijssen)
- Computer good for representation of finite matrices
- Appropriate for discrete spectrum of eigenstates
- By using conservation laws, we can often block
diagonalize the Hamiltonian, leaving discrete spectra for each block
- Algorithms appropriate for computation
- See "Numerical Recipes", ch. 11; Koonin, ch 5
- Inversion; Determinants
- Eigenvalues of tridiagonal matrix - O(N)
- Reducing matrices to tridiagonal form - O(N3)
- Eigenvectors - inverse iteration
- Libraries - e.g., Netlib
- lapack, eispack
- examples in lapack: dsyevx; cheevx
- Special approaches - Lanczos - later
- Greens Functions in matrix form
- Inverse of Hamiltonian matrix (with energy shift)
- Diagonal representation in eigenstates of energy
Next time: Calculations in a Gaussian basis: The GAUSSIAN package
Note Gaussian web site: http://www.gaussian.com/
Last Modified Feb. 24
Email question/comments/corrections to
rmartin@uiuc.edu
.