Newton Raphson method for system of nonlinear equations
3 visualizaciones (últimos 30 días)
Mostrar comentarios más antiguos
Hi, can someone explain what is wrong with the code as the answers are different from the expected answer which I have attached:
x0 = [3; -1.5];
maxIter = 50;
tolX = 1e-14;
x = x0;
xold = x0;
format long
xv = nan(maxIter,1);
yv = xv;
for i = 1:maxIter
[f,j] = newtraph(x);
x = x- j\f;
err(:,i) = abs(x-xold);
xold = x;
if (err(:,i)<tolX)
break;
end
xv(i,:) = x(1);
yv(i,:) = x(2);
end
OutputValues = table(xv(~isnan(xv)), yv(~isnan(yv)), 'VariableNames',{'X','Y'})
display(x)
fprintf('Number of iterations: %d \n',i)
fprintf('Final tolerance before accepted tolerance: %d \n' ,err(i))
function [fval,jac] = newtraph(X)
x = X(1);
y = X(2);
fval(1,1)=x+exp(-x)+y^3;
fval(2,1)=x^2+2*x*y-y^2+tand(x);
jac = [1-exp(-x), 3*y^2;
2*x+2*y+1+(tand(x))^2, 2*x-2*y];
end
The output of the code is:
OutputValues =
23×2 table
X Y
________________ _________________
3.64477577226406 -1.54258686595204
3.71768338513548 -1.55256145261782
3.7349599350235 -1.55483810676808
3.73922377944533 -1.5554091461244
3.74028579368963 -1.55555181328355
3.74055090524151 -1.55558745542312
3.74061712194685 -1.55559635948701
3.74063366312717 -1.55559858386503
3.74063779531838 -1.55559913954898
3.74063882759969 -1.55559927836732
3.7406390854791 -1.55559931304626
3.74063914990129 -1.55559932170959
3.74063916599493 -1.55559932387382
3.74063917001537 -1.55559932441447
3.74063917101973 -1.55559932454954
3.74063917127064 -1.55559932458328
3.74063917133332 -1.55559932459171
3.74063917134898 -1.55559932459382
3.74063917135289 -1.55559932459434
3.74063917135387 -1.55559932459447
3.74063917135411 -1.55559932459451
3.74063917135417 -1.55559932459451
3.74063917135419 -1.55559932459452
x =
3.740639171354190
-1.555599324594516
Number of iterations: 24
Final tolerance before accepted tolerance: 8.663327e-09
The steps is the iteration number, the delta(xk-1) is the tolerance which should be less than
tolX = 1e-14;
and the xk and yk are of course the x and y values.
The code works for a previous example that I tried, where the function had different fval and jac equations, but not for this instance.
0 comentarios
Respuestas (1)
Ayush
el 2 de Abr. de 2024
Hi,
It seems that your output tolerance is not under the specified "tolX". To improve this, you can use a "norm" (like the Euclidean norm) function to get a single value representing the "distance" between your old and new "x" values and then compare this against your tolerance. Currently, the tolerance check is performed against the vector "err(:,i)" having absolute error calculation directly; instead of it, update the error using the "norm" of "deltaX" Refer to the modified code:
x0 = [3; -1.5];
maxIter = 50;
tolX = 1e-14;
x = x0;
xold = x0;
format long
xv = nan(maxIter,1);
yv = xv;
err = zeros(1, maxIter); % Initialize the error array
for i = 1:maxIter
[f,j] = newtraph(x);
deltaX = j\f; % Calculate the change in x
x = x - deltaX; % Update x
err(i) = norm(deltaX); % Update the error using the norm of deltaX
if err(i) < tolX
break;
end
xv(i) = x(1); % Store the x and y values
yv(i) = x(2);
end
% Remove NaN values from xv and yv
xv = xv(~isnan(xv));
yv = yv(~isnan(yv));
OutputValues = table(xv, yv, 'VariableNames',{'X','Y'})
display(x)
fprintf('Number of iterations: %d \n',i)
fprintf('Final tolerance before accepted tolerance: %d \n' ,err(i)) % Adjusted the print format to show more precision
function [fval,jac] = newtraph(X)
x = X(1);
y = X(2);
fval(1,1)=x+exp(-x)+y^3;
fval(2,1)=x^2+2*x*y-y^2+tand(x);
jac = [1-exp(-x), 3*y^2;
2*x+2*y+1+(tand(x))^2, 2*x-2*y];
end
This resulted in the error of "3.722328e-15", which is less than your defined tolerance. For more information on the "norm" function, refer to the below documentation:
0 comentarios
Ver también
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!