Solving a matrix equation with fixed point iteration method

5 visualizaciones (últimos 30 días)
I want to solve the following equation for
where
I believe Equ.1 can be solved using fixed iteration method. The twist here is that the term itself have a self-consistent equation:
So, to solve Equ.1, I have to solve Equ.3 for and then put the value of in Equ.2 and calculate and finally put in Equ.1 to solve it.
In Equ.1, And H and are 2-by-2 matrices given in below code. is identity matrix and E is a scalar parameter.
I'm working on this code. The loop for is converging well, but the loop for is very slow for certain u values, and it never converges for others. Any tips to speed up convergence or alternative solution methods?
clear; clc;
% parameters of equations:
E = 1;
n = 0.1;
u = 0.2;
% parameters of this script:
Nk = 1000; % number of points for integrating over kx and ky
max_iter = 2000; % # of maximum iterations
convergence_threshold = 1e-6;
% k-points and limits
xmin = -2*pi/(3*sqrt(3));
xmax = 4*pi/(3*sqrt(3));
ymin = -2*pi/3;
ymax = 2*pi/3;
kxs = linspace(xmin,xmax,Nk);
dkx = kxs(2) - kxs(1);
kys = linspace(ymin,ymax,Nk);
dky = kys(2) - kys(1);
%%%%%%%%%%%%%%%%%%%%%%%% Calculation of Sigma_0 %%%%%%%%%%%%%%%%%%%%%%%%
Sigma_0 = (0.1 + 0.1i)*eye(2); % initial guess
for iter = 1:max_iter
% Calculation of integration in Sigma_0 (Equ.3) via sum:
G_0 = @(kx,ky) inv( E*eye(2) - H(kx,ky) - Sigma_0 );
integral_term_Equ3 = zeros(2);
parfor ikx = 1:Nk
qx = kxs(ikx);
for iky = 1:Nk
qy = kys(iky);
integral_term_Equ3 = integral_term_Equ3 + G_0(qx,qy) * dkx * dky;
end
end
%new value of Sigma_0 (Equ.3):
new_Sigma_0 = n * u * inv( eye(2) - 1/(4*pi^2) * integral_term_Equ3 * u);
diff = norm(new_Sigma_0 - Sigma_0); %difference
fprintf('G_0 Iteration: %d, Difference: %0.9f\n', iter, diff);
if diff < convergence_threshold
fprintf('G_0 converged after %d iterations\n', iter);
break;
end
Sigma_0 = new_Sigma_0; % update Solution
end
G_0 = @(kx,ky) inv( E*eye(2) - H(kx,ky) - Sigma_0 ); %the chosen G_0
%%%%%%%%%%%%%%%%%%%%%%%% Calculation of Sigma_x %%%%%%%%%%%%%%%%%%%%%%%%
Sigma_x = Sigma_0; %taking Sigma_0 as initial guess
for iter = 1:max_iter
% Calculating the integration in Sigma (Equ.1) via sum:
integral_term_Equ1 = zeros(2);
parfor ikx = 1:Nk
qx = kxs(ikx);
for iky = 1:Nk
qy = kys(iky);
integrant = G_0(qx,qy) * (Sigma_x - 1i*E*v_x(qx,qy)) * G_0(qx,qy)' + 1i*E/2 * (G_0(qx,qy)*v_x(qx,qy)*G_0(qx,qy) + G_0(qx,qy)'*v_x(qx,qy)*G_0(qx,qy)');
integral_term_Equ1 = integral_term_Equ1 + integrant * dkx * dky;
end
end
%new value of Sigma_x (Equ.1):
new_Sigma_x = -1/(4*pi^2*n) * Sigma_0 * integral_term_Equ1 * Sigma_0';
diff = norm(new_Sigma_x - Sigma_x); %difference
fprintf('G Iteration: %d, Difference: %0.9f\n', iter, diff);
if diff < convergence_threshold
fprintf('G converged after %d iterations\n', iter);
break;
end
Sigma_x = new_Sigma_x; % update Solution
end
%%%%%%%%%%%%%%%%%%%%%%%% H and v_x functions %%%%%%%%%%%%%%%%%%%%%%%%
function H = H(kx,ky)
J = 1;
D = 0.5;
S = 1;
a1 = [0,-1]';
a2 = [sqrt(3)/2,1/2]';
a3 = [-sqrt(3)/2,1/2]';
b1 = [sqrt(3)/2,-3/2]';
b2 = [sqrt(3)/2,3/2]';
b3 = [-sqrt(3),0]';
s0 = eye(2,2);
sx = [0,1; 1,0];
sy = [0, -1i; 1i, 0];
sz = [1, 0; 0, -1];
k = [kx,ky];
h0 = 3*J*S;
hx = -J*S*( cos(k*a1) + cos(k*a2) + cos(k*a3) );
hy = -J*S*( sin(k*a1) + sin(k*a2) + sin(k*a3) );
hz = 2*D*S*( sin(k*b1) + sin(k*b2) + sin(k*b3) );
H = s0*h0 + sx*hx + sy*hy + sz*hz;
end
function v_x = v_x(kx,ky)
J = 1;
D = 0.5;
S = 1;
sx = [0,1; 1,0];
sy = [0, -1i; 1i, 0];
sz = [1, 0; 0, -1];
dkx_hx = -J*S*((3^(1/2)*sin(ky/2 - (3^(1/2)*kx)/2))/2 - (3^(1/2)*sin(ky/2 + (3^(1/2)*kx)/2))/2);
dkx_hy = J*S*((3^(1/2)*cos(ky/2 - (3^(1/2)*kx)/2))/2 - (3^(1/2)*cos(ky/2 + (3^(1/2)*kx)/2))/2);
dkx_hz = 2*D*S*((3^(1/2)*cos((3*ky)/2 - (3^(1/2)*kx)/2))/2 - 3^(1/2)*cos(3^(1/2)*kx) + (3^(1/2)*cos((3*ky)/2 + (3^(1/2)*kx)/2))/2);
v_x = sx*dkx_hx + sy*dkx_hy + sz*dkx_hz;
end
The result of above code are:
G_0 Iteration: 1, Difference: 0.127690191
G_0 Iteration: 2, Difference: 0.001871435
G_0 Iteration: 3, Difference: 0.000038569
G_0 Iteration: 4, Difference: 0.000000739
G_0 converged after 4 iterations
G Iteration: 1, Difference: 0.046150821
G Iteration: 2, Difference: 0.050105083
G Iteration: 3, Difference: 0.050493792
G Iteration: 4, Difference: 0.050542706
G Iteration: 5, Difference: 0.050547627
G Iteration: 6, Difference: 0.050543314
G Iteration: 7, Difference: 0.050536343
G Iteration: 8, Difference: 0.050528515
G Iteration: 9, Difference: 0.050520402
G Iteration: 10, Difference: 0.050512195
G Iteration: 11, Difference: 0.050503958
G Iteration: 12, Difference: 0.050495711
G Iteration: 13, Difference: 0.050487461
G Iteration: 14, Difference: 0.050479212
G Iteration: 15, Difference: 0.050470964
G Iteration: 16, Difference: 0.050462717
G Iteration: 17, Difference: 0.050454471
G Iteration: 18, Difference: 0.050446226
G Iteration: 19, Difference: 0.050437983
G Iteration: 20, Difference: 0.050429741
G Iteration: 21, Difference: 0.050421501
G Iteration: 22, Difference: 0.050413262
G Iteration: 23, Difference: 0.050405024
G Iteration: 24, Difference: 0.050396788
G Iteration: 25, Difference: 0.050388553
G Iteration: 26, Difference: 0.050380319
G Iteration: 27, Difference: 0.050372087
G Iteration: 28, Difference: 0.050363856
G Iteration: 29, Difference: 0.050355626
G Iteration: 30, Difference: 0.050347398
G Iteration: 31, Difference: 0.050339171
G Iteration: 32, Difference: 0.050330945
G Iteration: 33, Difference: 0.050322721
G Iteration: 34, Difference: 0.050314498
G Iteration: 35, Difference: 0.050306276
G Iteration: 36, Difference: 0.050298056
G Iteration: 37, Difference: 0.050289837
G Iteration: 38, Difference: 0.050281619
G Iteration: 39, Difference: 0.050273403
G Iteration: 40, Difference: 0.050265188
G Iteration: 41, Difference: 0.050256975
G Iteration: 42, Difference: 0.050248762
G Iteration: 43, Difference: 0.050240552
G Iteration: 44, Difference: 0.050232342
G Iteration: 45, Difference: 0.050224134
G Iteration: 46, Difference: 0.050215927
G Iteration: 47, Difference: 0.050207721
G Iteration: 48, Difference: 0.050199517
G Iteration: 49, Difference: 0.050191315
G Iteration: 50, Difference: 0.050183113
G Iteration: 51, Difference: 0.050174913
G Iteration: 52, Difference: 0.050166714
G Iteration: 53, Difference: 0.050158517
G Iteration: 54, Difference: 0.050150321
G Iteration: 55, Difference: 0.050142126
G Iteration: 56, Difference: 0.050133932
G Iteration: 57, Difference: 0.050125740
G Iteration: 58, Difference: 0.050117549
G Iteration: 59, Difference: 0.050109360
G Iteration: 60, Difference: 0.050101172
G Iteration: 61, Difference: 0.050092985
G Iteration: 62, Difference: 0.050084800

Respuesta aceptada

Torsten
Torsten el 29 de Dic. de 2023
Editada: Torsten el 29 de Dic. de 2023
main()
Equation solved. fsolve completed because the vector of function values is near zero as measured by the value of the function tolerance, and the problem appears regular as measured by the gradient.
Sigma_0 =
0.020586935783072 + 0.002472080743781i 0.000584326409329 + 0.001608817728934i 0.000584326409329 + 0.001608817728934i 0.020586736770647 + 0.002650780039849i
Equation solved. fsolve completed because the vector of function values is near zero as measured by the value of the function tolerance, and the problem appears regular as measured by the gradient.
Sigma_x =
-0.000000000000000 + 0.024841208312721i 0.000769082661170 + 0.019717976813217i -0.000769082661170 + 0.019717976813218i 0.000000000000000 + 0.019770775411212i
function main
clear; clc;
format long
% parameters of equations:
E = 1;
n = 0.1;
u = 0.2;
% parameters of this script:
Nk = 300; % number of points for integrating over kx and ky
% k-points and limits
xmin = -2*pi/(3*sqrt(3));
xmax = 4*pi/(3*sqrt(3));
ymin = -2*pi/3;
ymax = 2*pi/3;
kxs = linspace(xmin,xmax,Nk);
dkx = kxs(2) - kxs(1);
kys = linspace(ymin,ymax,Nk);
dky = kys(2) - kys(1);
%%%%%%%%%%%%%%%%%%%%%%%% Calculation of Sigma_0 %%%%%%%%%%%%%%%%%%%%%%%%
Sigma_0 = (0.1 + 0.1i)*eye(2); % initial guess
sigma0 = [Sigma_0(:,1);Sigma_0(:,2)];
sigma0 = fsolve(@fun_Sigma0,sigma0,optimset('TolFun',1e-12,'TolX',1e-12));
Sigma_0 = [sigma0(1:2),sigma0(3:4)]
%%%%%%%%%%%%%%%%%%%%%%%% Calculation of Sigma_x %%%%%%%%%%%%%%%%%%%%%%%%
Sigma_x = Sigma_0;
sigmax = [Sigma_x(:,1);Sigma_x(:,2)];
sigmax = fsolve(@fun_Sigmax,sigmax,optimset('TolFun',1e-12,'TolX',1e-12));
Sigma_x = [sigmax(1:2),sigmax(3:4)]
function res = fun_Sigma0(sigma0)
Sigma_0 = [sigma0(1:2),sigma0(3:4)];
% Calculation of integration in Sigma_0 (Equ.3) via sum:
G_0 = @(kx,ky) inv( E*eye(2) - H(kx,ky) - Sigma_0 );
integral_term_Equ3 = zeros(2);
for ikx = 1:Nk
qx = kxs(ikx);
for iky = 1:Nk
qy = kys(iky);
integral_term_Equ3 = integral_term_Equ3 + G_0(qx,qy) * dkx * dky;
end
end
Res = Sigma_0 - n * u * inv( eye(2) - 1/(4*pi^2) * integral_term_Equ3 * u);
res = [Res(:,1);Res(:,2)];
end
function res = fun_Sigmax(sigmax)
Sigma_x = [sigmax(1:2),sigmax(3:4)];
% Calculation of integration in Sigma_0 (Equ.3) via sum:
G_0 = @(kx,ky) inv( E*eye(2) - H(kx,ky) - Sigma_0 );
integral_term_Equ1 = zeros(2);
for ikx = 1:Nk
qx = kxs(ikx);
for iky = 1:Nk
qy = kys(iky);
G_0_num = G_0(qx,qy);
v_x_num = v_x(qx,qy);
integrant = G_0_num * (Sigma_x - 1i*E*v_x_num) * G_0_num' + 1i*E/2 * (G_0_num*v_x_num*G_0_num + G_0_num'*v_x_num*G_0_num');
integral_term_Equ1 = integral_term_Equ1 + integrant * dkx * dky;
end
end
Res = Sigma_x - (-1/(4*pi^2*n) * Sigma_0 * integral_term_Equ1 * Sigma_0');
res = [Res(:,1);Res(:,2)];
end
%%%%%%%%%%%%%%%%%%%%%%%% H and v_x functions %%%%%%%%%%%%%%%%%%%%%%%%
function H = H(kx,ky)
J = 1;
D = 0.5;
S = 1;
a1 = [0,-1]';
a2 = [sqrt(3)/2,1/2]';
a3 = [-sqrt(3)/2,1/2]';
b1 = [sqrt(3)/2,-3/2]';
b2 = [sqrt(3)/2,3/2]';
b3 = [-sqrt(3),0]';
s0 = eye(2,2);
sx = [0,1; 1,0];
sy = [0, -1i; 1i, 0];
sz = [1, 0; 0, -1];
k = [kx,ky];
h0 = 3*J*S;
hx = -J*S*( cos(k*a1) + cos(k*a2) + cos(k*a3) );
hy = -J*S*( sin(k*a1) + sin(k*a2) + sin(k*a3) );
hz = 2*D*S*( sin(k*b1) + sin(k*b2) + sin(k*b3) );
H = s0*h0 + sx*hx + sy*hy + sz*hz;
end
function v_x = v_x(kx,ky)
J = 1;
D = 0.5;
S = 1;
sx = [0,1; 1,0];
sy = [0, -1i; 1i, 0];
sz = [1, 0; 0, -1];
dkx_hx = -J*S*((3^(1/2)*sin(ky/2 - (3^(1/2)*kx)/2))/2 - (3^(1/2)*sin(ky/2 + (3^(1/2)*kx)/2))/2);
dkx_hy = J*S*((3^(1/2)*cos(ky/2 - (3^(1/2)*kx)/2))/2 - (3^(1/2)*cos(ky/2 + (3^(1/2)*kx)/2))/2);
dkx_hz = 2*D*S*((3^(1/2)*cos((3*ky)/2 - (3^(1/2)*kx)/2))/2 - 3^(1/2)*cos(3^(1/2)*kx) + (3^(1/2)*cos((3*ky)/2 + (3^(1/2)*kx)/2))/2);
v_x = sx*dkx_hx + sy*dkx_hy + sz*dkx_hz;
end
end
  15 comentarios
Torsten
Torsten el 16 de En. de 2024
Editada: Torsten el 16 de En. de 2024
I believe that if I could write a code that takes a lot of points near these points, we can achieve accurate integration.
That's exactly what an adaptive ODE integrator like ode45 does. If it didn't succeed, I doubt you will find a way to handle this problem with existing MATLAB codes.
Are you sure that the matrix G00 has no singularities in the domain of integration ?
Luqman Saleem
Luqman Saleem el 16 de En. de 2024
I am sure that as long as η is non-zero there are no singularities, however, G00 can be very very very huge for some kx-ky points (I have mentioned those points in the new question that I've posted). I am aiming at case.

Iniciar sesión para comentar.

Más respuestas (0)

Categorías

Más información sobre MATLAB en Help Center y File Exchange.

Productos


Versión

R2023a

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by