Matrix Calculation The Heisenberg Model

Matrix calculation The Heisenberg model
Matrix calculation The Heisenberg model

Interactions in a long chain of N quantistic spins There are real systems that can be studied by this simple model The quantum state of the chain can be expressed in terms (a product) of the single spin states, eigenvectors of z , with eigenvalues 1.
Each spin points up or down.
E.g., typical conguration for 9 spins:
There are 2N congurations (states for the whole chain)
The hamiltonian of the Heisenberg model is
1 H = 2 S j S j+1 = 2 j=1
j=1
j j+1
where S j = 1 2 j is the spin operator acting on spin j .
The energy of a chain depends on whether adjacent spins are parallel

or antiparallel
1 1 A pair like or contributes 2 . Pairs like or contribute + 2
(x , y , z ) (1 , 2 , 3 ) are the 2 2 Pauli spin matrices.

1 x where l 2 (l il ) ip the l -th spin up (or down). y + ) + z z , + It can be shown that a b = 2(+ a b a b a b
Notice that the spin states as we dened are a complete set of states (a basis) but are not eigenstates of H . So H is not diagonal on this basis.
H is can be represented as a 2N 2N matrix H

The energy levels of the quantum system are the eigenvalues of H. Given two spin congurations of the string (Cl and Cm ), the (l , m) element of H (Hl ,m ) is the matrix element Cl | H | Cm Lets try for N = 2... The general practical rule to build H is:
Each row and column of H is labelled by a pattern of spins (spins state) and an element of H can be seen as linking 2 patterns. The diagonal elements of the matrix are computed as + 1 2 for each or , and 1 2 for every or . The only non-zero entries off the diagonal correspond to patterns that can be generated by changing into or into . In this case, the off-diagonal entry is then 1.
Example: conguration

( 1) Entry 2 + 6 2 = 2 on the diagonal of the matrix
1 2
Example: conguration

( 1) Entry 2 + 6 2 = 2 on the diagonal of the matrix
1 2
Entries equal to 1 for congurations
For example, for N = 4 the dimension of the matrix is 2N 2N = 16 16 (= 256 elements!).
The matrix form of the Hamiltonian of the quantum y z x Heisenberg model of spins S j = (S j , S j , S j ) on a one-dimensional lattice of N = 4 sites with periodic boundary conditions SN +1 = S1 is given by

2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 1 0 0 0 0 0 0 0 0 1 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 1 0 0 0 0 0 0 0 1 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 1 0 -2 1 0 0 1 0 0 1 0 0 0 0 0 0 0 0 1 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 1 0 0 1 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 1 0 0 0 0 0 0 0 0 1 0 0 1 0 0 1 -2 0 1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 1 0 0 0 0 0 0 0 1 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 1 0 0 0 0 0 0 0 0 1 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2
Matrix size 2N 2N .
Can easily get N = 16
Physical systems have large values of N so we need N .
Matrix size 2N 2N .
Physical systems have large values of N so we need N . For one-dimensional chains we can get N large enough to extrapolate to
N=
For two or three-dimensional lattices of spins and the size of the lattice is then N 1/2 N 1/2 or N 1/3 N 1/3 N 1/3 It is impossible to get N large enough to approximate real systems.
Matrix size 2N 2N .
Physical systems have large values of N so we need N . For one-dimensional chains we can get N large enough to extrapolate to
N=
For two or three-dimensional lattices of spins and the size of the lattice is then N 1/2 N 1/2 or N 1/3 N 1/3 N 1/3 It is impossible to get N large enough to approximate real systems. This is a simple problem whose accurate solution is beyond current computers. Using symmetry and hard work will get N 24 For larger matrices Monte Carlo is the only possibility.
Computing Eigenvalues of Matrices

Many physics problems require the calculation of the eigenvalues of a matrix.
Computing Eigenvalues of Matrices

Many physics problems require the calculation of the eigenvalues of a matrix. Find N vectors un and numbers n associated with the N N matrix A such that
Aun = n un .
The general methods are beyond this course and you would normally use a library program which converts the matrix into a tridiagonal one with the same eigenvalues. A tridiagonal matrix has all elements zero except on the diagonal and one stripe on either side. This step requires all the matrix in memory at once, so limits the maximum size of matrix.
One method for BIG matrices

Suppose x(0) is a random unit vector Sequence of vectors, also of unit length, x(k) calculated by
x(k+1) =
1 Ax(k) , Lk
Lk is the scaling constant needed to make x(k+1) a unit vector.

If the sequence converges to a xed vector so that x(k+1) = x(k) then x(k) is one of the eigenvectors and Lk converges to the corresponding eigenvalue. The sequence does converge to the vector with the largest value of ||.
This algorithm involves operating with the matrix A on a vector. Only the vectors x(k+1) and x(k) need be in memory (A can be on disk). In the important applications most of the elements of the matrix are zero and do not appear in the calculation. Multiplication by A need not be evaluated as a matrix operation provided we have an algorithm for getting Ax from x. The limit on the matrix size is the memory to hold the two vectors. Beyond this Monte Carlo methods are possible!
Computer Exercise: The Heisenberg model

In the computer we use 0 and 1 to represent and ; a spin conguration is simply an binary integer. The matrix H is very sparse we do not store it but use a procedure takes in a vector v and returns the vector u equal to Hv. 1. Diagonalisation with the whole matrix in memory Possible for very small N . 2. Iterative method for the ground state Compute /N . You will run out of memory somewhere before N = 20. 3. Extrapolate to N = Record the ground state energy for various N , including both odd and even values, and extrapolate to N = .
4. First excited state
Eigenvectors are orthogonal. Find the ground state eigenvector u0 . Repeat the iteration including an extra step to force v to be orthogonal to u0 . If u0 is a unit vector then v (u0 v)u0 is orthogonal to u0 . (A piece
of code is included in the program to do this.) Extrapolate as before.
Overview of program (megaeig.py)

1. Create bit patterns corresponding to adjacent pairs of spins 2. Create whole 2N 2N matrix in memory and get energies 3. Iterative method for ground state 4. Space to ll in extension for rst excited state 5. Code to carry out operation u = H v 6. Code to normalise a vector 7. Code to make v orthogonal to u0 (by removing its component parallel to u0 ).
Operate with matrix on vector

def matrix(u, v): "Computes u=Hv where H is hamiltonian" for i in range(0,M): sum=0.0 for j in range(0,N): p=pair[j] m=i & p # & is bitwise AND if not ( (m==0) or (m==p) ): sum=sum + v[ip] - 0.5*v[i] # is bitwise XOR else: sum=sum + 0.5*v[i] u[i]=sum
# Main Program N=10 M=2**N pair=array([3<<i for i in range(0,20)], int) ; pair[N-1]=(M/2)+1 ## << left shift operator: add a trailing 0 in binary rapresentation # Compute eigenvalues using full matrix in memory H=zeros((M,M), float) u=zeros(M, float) for j in range(0,M): # Construct full matrix v=zeros(M, float) v[j]=1.0 matrix(u, v) H[:,j]=u[:] (H, val)=mateig(H) #compute eigenvalues print val/N
# Iterative method for ground state using 2 vectors only in memory u=zeros(M, float) v=array([uniform(1) for i in range(0,M)], float) normalise(v) E=1e10 lastE=0 while abs(E-lastE) > 1e-5*abs(E): lastE=E matrix(u, v) E=normalise(u) v[:]=u[:] # copy array u to v print E/N
Main Points
How nite difference approximate rst and second derivatives. The merits of nite difference solutions compared with theoretical
ones.
Lots of physics problems translate into linear equations. Some techniques for handling linear equations on computers. Simple physics problems can easily be beyond current average
computers.
Finding eigenvalues of very big matrices. Extrapolation to innite systems from nite ones.

Matrix Calculation The Heisenberg Model

Uploaded by

Matrix Calculation The Heisenberg Model

Uploaded by

Matrix calculation The Heisenberg model

Matrix calculation The Heisenberg model

E.g., typical conguration for 9 spins:

There are 2N congurations (states for the whole chain)

The hamiltonian of the Heisenberg model is

where S j = 1 2 j is the spin operator acting on spin j .

The energy of a chain depends on whether adjacent spins are parallel

(x , y , z ) (1 , 2 , 3 ) are the 2 2 Pauli spin matrices.

H is can be represented as a 2N 2N matrix H

Entries equal to 1 for congurations

For example, for N = 4 the dimension of the matrix is 2N 2N = 16 16 (= 256 elements!).

Can easily get N = 16

Physical systems have large values of N so we need N .

Can easily get N = 16

Can easily get N = 16

Computing Eigenvalues of Matrices

Computing Eigenvalues of Matrices

One method for BIG matrices

Lk is the scaling constant needed to make x(k+1) a unit vector.

Computer Exercise: The Heisenberg model

4. First excited state

Overview of program (megaeig.py)

Operate with matrix on vector

You might also like