PDE solver I made for the laplace equation fails when I insert a poisson equation

1 回表示 (過去 30 日間)
Matan
Matan 2024 年 1 月 9 日
回答済み: UDAYA PEDDIRAJU 2024 年 1 月 15 日
Hello all, I wrote a solver to the poisson equation, it actually works fine for the laplace equation (e.g, when the laplacian equals to 0), but it screws by a lot when the laplacian isn't equal to 0.
function Sol = LaplaceSolver(xLimit, yLimit, boundaryFunction, rectFunction, N_x, N_y,tol, LinearSolverType)
%% in this function we solve the Laplace Equation using Jacobi's method
%% first we define our grid
x_s = linspace(xLimit(1), xLimit(2), N_x);
y_s = linspace(yLimit(1), yLimit(2), N_y);
delta_x = x_s(2)-x_s(1);
delta_y = y_s(2)-y_s(1);
%% We essentially solve the linear system of equation Ax = b where:
%% b = boundary
%% A the matrix
%% we begin with the bottom layer
b = DefineBoundaryLaplace5(x_s,y_s, boundaryFunction, rectFunction);
%% Next we determine our system of equations
delta_x_sq = delta_x^2;
delta_y_sq = delta_y^2;
FiniteDifferenceEquationLaplace = DefineEquationLaplace4(N_x,N_y,delta_x_sq,delta_y_sq);
%% Finally we use the Jacobi's method to solve our (sparse) linear system of equation
xInitial = zeros(length(b),1);
switch LinearSolverType
case "Jacobi"
VectorSol = JacobiMethod(FiniteDifferenceEquationLaplace,b,xInitial, tol);
case "Gauss"
VectorSol = GaussSiedelMethod(FiniteDifferenceEquationLaplace,b,xInitial, tol);
case "SOR"
omega = 4 / (2 + sqrt(4 - (cos(pi / N_y) + cos(pi/N_x)^2)));
VectorSol = SORMethod(FiniteDifferenceEquationLaplace,b,xInitial,omega,tol);
case "ConjGrad"
VectorSol = ConjGrad(FiniteDifferenceEquationLaplace,b,tol);
case "weird"
VectorSol = Weirdy(b,tol)
end
%% and trully finally, we reshape our VectorSol into a matrix:
Sol = reshape(VectorSol, [N_x-2,N_y-2]);
end
The function DefineBoundaryLaplace5:
function b = DefineBoundaryLaplace5(x,y, boundaryFunction, rectFunction)
N_x = length(x);
N_y = length(y);
delta_x = x(2) - x(1);
delta_y = x(2) - x(1);
b = zeros((N_x-2),(N_y-2));
delta_x_sq = delta_x^2;
delta_y_sq = delta_y^2;
arr_x = 1:(N_x-2);
arr_y = 1:(N_y-2);
x_coeffs = 1;
y_coeffs = delta_x_sq / delta_y_sq;
rect_coeffs = delta_x_sq;
b(arr_x,arr_y) = rect_coeffs * rectFunction(x(arr_x+1), y(arr_y+1));
b(arr_x,1) = b(arr_x,1) + y_coeffs * boundaryFunction(x(arr_x+1), y(1))';
b(arr_x, N_y-2) = b(arr_x,N_y-2) + y_coeffs * boundaryFunction(x(arr_x+1), y(N_y))';
b(1, arr_y) = b(1,arr_y) + x_coeffs * boundaryFunction(x(1), y(arr_y+1));
b(N_x-2, arr_y) = b(N_x-2,arr_y) + x_coeffs * boundaryFunction(x(N_x), y(arr_y+1));
b = reshape(b, [(N_x-2)*(N_y-2),1]);
end
The function DefineEquationLaplace4:
function FiniteDifferenceLaplace = DefineEquationLaplace4(N_x,N_y,delta_x_sq,delta_y_sq)
FiniteDifferenceLaplace = cell((N_x-2) * (N_y-2),1);
L = CalcL(1,1,N_y);
x_coeff = -1;
y_coeff = -delta_x_sq / delta_y_sq;
Center_coeff = 2 * (delta_x_sq / delta_y_sq + 1);
FiniteDifferenceLaplace{L} = [Center_coeff,1, x_coeff,2, y_coeff,N_x-1]; %% bottom left
L = CalcL(N_x-2,1,N_y);
FiniteDifferenceLaplace{L} = [x_coeff,L-1,Center_coeff,L,y_coeff,L+(N_x-2)]; % bottom right
L = CalcL(1, N_y-2, N_y);
FiniteDifferenceLaplace{L} = [y_coeff,L-(N_x-2),Center_coeff,L,x_coeff,L+1]; % upper left
L = CalcL(N_x-2, N_y-2, N_y);
FiniteDifferenceLaplace{L} = [y_coeff,L-(N_x-2),x_coeff,L-1,Center_coeff,L]; % upper right
for i = 2:(N_x-3) % i iterates on x
L = CalcL(i,1,N_y); % bottom line
FiniteDifferenceLaplace{L} = [x_coeff,L-1,Center_coeff,L,x_coeff,L+1,y_coeff, L + (N_x-2)]; % bottom line
L = CalcL(i,N_y-2,N_y); % upper line
FiniteDifferenceLaplace{L} = [y_coeff, L - (N_x-2),x_coeff,L-1,Center_coeff,L,x_coeff,L+1]; % upper line
end
for j = 2:(N_x-3) % j iterates on y
L = CalcL(1,j,N_y); % left column
FiniteDifferenceLaplace{L} = [y_coeff,L-(N_x-2),Center_coeff,L,x_coeff,L+1,y_coeff, L + (N_x-2)]; % left line
L = CalcL(N_x-2,j,N_y); % upper line
FiniteDifferenceLaplace{L} = [y_coeff,L-(N_x-2),x_coeff,L-1,Center_coeff,L,y_coeff, L + (N_x-2)]; % right line
end
for i = 2:(N_x-3) % i iterates on x
for j = 2:(N_y-3) % j iterates on y
L = CalcL(i,j,N_y);
FiniteDifferenceLaplace{L} = [y_coeff,L - (N_x-2),x_coeff,L-1,Center_coeff,L,x_coeff,L+1,y_coeff, L + (N_x-2)]; % all other circumstances
end
end
end
end
And for instance, the Jacobi method which I use to iteratively solve the equation Ax = b:
function x = JacobiMethod(Matrix,b, xInitial, tol)
%% this function solves a sparse linear system of equations
%% it receives a cell array A{1},...,A{n}, each cell represents a line in
%% the matrix in the following way - A{1} = {A(1i1, i1),i1, A(1i2),i2, A1(i3),i3),...}
%% this is an efficent way to represent a sparse matrix, since we simply dont represent the zeros.
%% A = [[a1_i1, i1 , a_i2, i2, ], [ai_i, i,...]
N = length(xInitial);
x = zeros(N,1);
for i = 1:N
CurrLine = Matrix{i};
SumDotProduct = sum(CurrLine(1:2:length(CurrLine)) * xInitial(CurrLine(2:2:length(CurrLine))) );
diagonal_id = find(CurrLine(2:2:length(CurrLine)) == i);
a_ii = CurrLine(diagonal_id * 2 - 1);
SumDotProduct = SumDotProduct - a_ii * xInitial(i);
x(i) = ( -SumDotProduct + b(i)) / a_ii;
end
norm_to_check = max(abs(x - xInitial)) / max(abs(x));
if (norm_to_check < tol)
return
else
x = JacobiMethod(Matrix,b,x,tol);
end
end
Thanks in advance for anyone willing to help, and I will also appericate any other advises on how to make my code even more optimized (even though it runs pretty smooth when I use the Conj grad)

回答 (1 件)

UDAYA PEDDIRAJU
UDAYA PEDDIRAJU 2024 年 1 月 15 日
Hi Matan,
To address the issue with poison equation solver you can try the following:
  1. Ensure Proper Source Term Integration: Check that the source term is correctly added to the “b” vector.
  2. Check Boundary Conditions: Confirm that “DefineBoundaryLaplace5” applies the boundary conditions accurately for both equations.
  3. Verify Matrix Coefficients: Make sure that “DefineEquationLaplace4” accurately represents the discretized Poisson equation.
  4. Review Convergence Criteria: Adjust the tolerance and iteration count for the Poisson equation as needed.
  5. Test Solver Stability: Experiment with different initial guesses and consider using a more robust solver if necessary.
  6. Optimize Code: Use vectorization and efficient sparse matrix operations to enhance performance.
  7. Debugging: Use debugging tools to examine intermediate computations, especially after incorporating the source term.
  8. Assess Numerical Stability: Check if the discretization scheme requires modification for stability with the source term.
Additionally, you can refer to the following toolbox, that can solve the Poisson equation: https://www.mathworks.com/matlabcentral/fileexchange/38090-2d-poisson-equation.
I hope this helps!

カテゴリ

Help Center および File ExchangeBoundary Conditions についてさらに検索

Community Treasure Hunt

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

Start Hunting!

Translated by