Hessian matrix 3D plot help

7 views (last 30 days)
Max
Max on 25 Apr 2015
Commented: Max on 4 May 2015
I am writing a script to find/plot the maxs and mins of a 3D function using the steepest ascent method, golden search method and confirmed by the Hessian matrix, I have almost finished this question for a school project but am having trouble having with the script storing the appropriate max/min/saddle values to plot on the 3D surface. The script plots the 3D surface its self, but no max or min points. I have attached what I have created so far, I just need some pointers in allocating the max/min/saddle points so they will be plotted on the 3D surface.
clear;clc
Fun = @(q,r)1.8.^(-1.5*sqrt(q.^2+r.^2)).*cos(0.5*r).*sin(q);
Guess = [2,2; 2,-2; -2,2; -2,-2]; %Give intial guesses
XRange = [-2 2]; YRange = [-2 2];
%------------Establish a matrix of four starting guesses based on the range-------------%
x = XRange(1,1) : 0.25 : XRange(1,2); %(-2, -1.75, -1.5,...,1.5, 1.75, 2)%
y = YRange(1,1) : 0.25 : YRange(1,2);
[X,Y] = meshgrid(x,y); %Form a mesh of base points that the surf plot will use
Z = 1.8.^(-1.5*sqrt(X.^2+Y.^2)).*cos(0.5*Y).*sin(X);
deltax = 0.01; % Establish delta's for central difference method
deltay = 0.01;
iMin = 1; iMax = 1; iSad = 1; %Keep the number of mins, maxs, and saddle points found
Min = zeros(1,5); Max = zeros(1,5); Saddle = zeros(1,5);
d = zeros(1,5); h_opt = zeros(1,5); xopt = zeros(1,5);
f1 = zeros(1,5); f2 = zeros(1,5);
x1 = zeros(1,5); x2 = zeros(1,5);
xU = zeros(1,5); xL = zeros(1,5);
GS_Error = zeros(1,5); SA_Error = zeros(1,5);
R = (5^.5-1)/2;
%---------Steepest Ascent Method-------%
for w = 1:4 %Loop over the four intial guesses
x_new = Guess(w,1); y_new = Guess(w,2); %Location of intial guesses
OptZ_new = Fun(x_new, y_new); %Funcion value at the point
while SA_Error>.01
x_old = x_new;
y_old = y_new;
OptZ_old = OptZ_new;
InitialGuess=Fun(x_new,y_new);
dfdx = (Fun(x_old+deltax,y_old)-Fun(x_old-deltax,y_old))/(2*deltax);
dfdy = (Fun(x_old,y_old+deltay)-Fun(x_old,y_old-deltay))/(2*deltay);
Funh = @(h)Fun(x_new+dfdx*h,y_new+dfdy*h);
%Begin Golden Section method
while abs(GS_Error)>.01
d(i) = ((sqrt(5)-1)/2)*(xU(i)-xL(i));
x1(i)=xL(i)+d(i);
x2(i)=xU(i)-d(i);
f1(i)=Funh(x1(i));
f2(i)=Funh(x2(i));
if abs(f1(i))>abs(f2(i))
OptZ_high = x1;
x1 = x2;
x2 = OptZ_high - d(i);
elseif abs(f2(i))>abs(f1(i))
OptZ_low= x2;
x2 = x1;
x1 = OptZ_low +d(i);
GS_Error(i) = (1-R)*abs(xU(i)-xL(i))/xopt(i);
x_new=x_new+dfdx*h_opt;
y_new=y_new+dfdy*h_opt;
SA_Error = sqrt((x_new-x_old)^2+(y_new-y_old)^2);
end
i = i+1;
end
end
end
%Use Hessian matrix to determine local minimum or maximum
d2fdx2 = (Fun(x_new+deltax,y_new)-2*Fun(x_new,y_new)+Fun(x_new-deltax,y_new))/(2*deltax)^2;
d2fdy2 = (Fun(x_new,y_new+deltay)-2*Fun(y_new,x_new)+Fun(x_new,y_new-deltay))/(2*deltay)^2;
d2fdxdy = (Fun(x_new+deltax,y_new+deltay)-Fun(x_new+deltax,y_new-deltay)-Fun(x_new-deltax,y_new+deltay)+Fun(x_new-deltax,y_new-deltay))/(4*deltax*deltay);
Hdet=d2fdx2*d2fdy2-d2fdxdy^2;
if abs(Hdet) > 0 && d2fdx2 > 0;
Max = OptZ_new;
elseif abs(Hdet) > 0 && d2fdx2 < 0;
Min = OptZ_new;
elseif abs(Hdet) <0;
saddle = OptZ_new;
end
%Plot for found mins and maxs
hold on
surf(X,Y,Z)
plot3(Max(1,1),Max(1,2),Min(1,1),Min(1,2),Saddle(1,1),Saddle(1,2),'m*','MarkerSize',10)
Thank you in advance for any help or advice, I feel like it is something very simple I am just not seeing.

Answers (2)

James Tursa
James Tursa on 25 Apr 2015
Edited: James Tursa on 25 Apr 2015
If you just need help with plotting the point, maybe something like this:
if abs(Hdet) > 0 && d2fdx2 > 0;
Max = [x_new y_new OptZ_new];
Extreme = Max;
elseif abs(Hdet) > 0 && d2fdx2 < 0;
Min = [x_new y_new OptZ_new];
Extreme = Min;
elseif abs(Hdet) <0;
Saddle = [x_new y_new OptZ_new];
Extreme = Saddle;
end
%Plot for found mins and maxs
hold on
surf(X,Y,Z)
plot3(Extreme(1),Extreme(2),Extreme(3),'m*','MarkerSize',10)
Also, it seems to me that these two lines should be done once prior to the for loop (i.e., moved out of this section of code):
hold on
surf(X,Y,Z)
And also it seems that the rest of the code should be inside and at the end of the for loop, so you get four points plotted ... one for each iteration of the loop.
  5 Comments
James Tursa
James Tursa on 25 Apr 2015
Edited: James Tursa on 25 Apr 2015
OK, now you are asking for more than just plotting help. I haven't looked in detail at all of your code, but here are some obvious problems you need to fix:
You initialize these variables as 0's:
GS_Error = zeros(1,5); SA_Error = zeros(1,5);
And then test them as follows:
while SA_Error>.01
:
while abs(GS_Error)>.01
First, testing a vector is not what you probably wanted ... I am guessing you wanted to use SA_Error(i) and GS_Error(i) here.
Second, these vectors start at 0's, so on the first pass they fail the test! You never get inside to execute the code you painstakingly wrote! SO maybe this instead:
GS_Error = ones(1,5); SA_Error = ones(1,5);
And I am not sure what to make of your indexing. Your for loop uses w but inside you use i. So something looks like it needs to be fixed there as well.
Max
Max on 25 Apr 2015
it calls on w to iterate x_new and y_new through the array guess values defined as x_new = Guess(w,1), y_new = Guess(w,2)
I will mess around with what you have pointed out though, thank you

Sign in to comment.


Umair Asif
Umair Asif on 4 May 2015
Max were you able to complete the code? I am working on something similar to what you have displayed so by any chance you got that program to work? Plz respond asap
  1 Comment
Max
Max on 4 May 2015
Yes, I was able to get it to work. What are you stuck with?

Sign in to comment.

Tags

Community Treasure Hunt

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

Start Hunting!