104 lines
No EOL
2.9 KiB
Matlab
104 lines
No EOL
2.9 KiB
Matlab
function [B,D,c_B,c_D,x_B,x_D,index_B,index_D] = auxiliary (A_aug,c_aug,h,m,n)
|
|
% The auxiliary problem is always a minimization problem
|
|
|
|
% The output will be: B and D (basic and nonbasic matrices), c_B and c_D (subdivision of the coefficient vector in basic and nonbasic parts), x_B
|
|
% and x_D (basic and nonbasic variables) and index_B and index_D (to keep track of the variables indices)
|
|
|
|
% Redefine the problem by introducing the artificial variables required by
|
|
% the auxiliary problem (the objective function has to reach value 0)
|
|
A_aug = [A_aug, eye(m)];
|
|
c_aug = [c_aug, zeros(1,m)]; % original objective function coefficients
|
|
c_aux = [zeros(1,n+m), ones(1,m)]; % auxiliary c of minimization problem
|
|
index = 1:n+2*m; % index to keep track of the different variables
|
|
|
|
% Defining the basic elements and the nonbasic elements
|
|
B = A_aug(:,(n+m+1):(n+2*m)); % basic variables
|
|
D = A_aug(:,1:(n+m)); % nonbasic variables
|
|
c_Baux = c_aux(1,(n+m+1):(n+2*m));
|
|
c_Daux = c_aux(1,1:(n+m));
|
|
c_B = c_aug(1,(n+m+1):(n+2*m));
|
|
c_D = c_aug(1,1:(n+m));
|
|
x_B = h;
|
|
x_D = zeros((n+m),1);
|
|
index_B = index(1,(n+m+1):(n+2*m));
|
|
index_D = index(1,1:(n+m));
|
|
|
|
nIter = 0;
|
|
z = c_Baux*x_B;
|
|
itMax = factorial(2*m+n)/(factorial(n+m)*factorial(m));
|
|
|
|
% Compute B^{-1}*D and B^{-1}*h
|
|
BiD = B\D;
|
|
Bih = B\h;
|
|
|
|
% Compute reduced cost coefficients
|
|
r_D = c_Daux - c_Baux*BiD;
|
|
|
|
while(z~=0)
|
|
|
|
% Find nonnegative index
|
|
idxIN = find(r_D==min(r_D));
|
|
% Using Bland's rule to avoid cycling
|
|
if(size(idxIN,2)>1)
|
|
idxIN = min(idxIN);
|
|
end
|
|
in = D(:,idxIN);
|
|
c_inaux = c_Daux(1,idxIN);
|
|
c_in = c_D(1,idxIN);
|
|
index_in = index_D(1,idxIN);
|
|
|
|
% Evaluating the coefficients ratio
|
|
inRatio = BiD(:,idxIN);
|
|
ratio = Bih./inRatio;
|
|
|
|
% Find the smallest ratio
|
|
for i = 1:size(ratio,1) % Eliminating negative ratios
|
|
if(ratio(i,1)<0)
|
|
ratio(i,1) = Inf;
|
|
end
|
|
end
|
|
idxOUT = find(ratio==min(ratio));
|
|
% Using Bland's rule to avoid cycling
|
|
if(size(idxOUT,1)>1)
|
|
idxOUT = min(idxOUT);
|
|
end
|
|
out = B(:,idxOUT);
|
|
c_outaux = c_Baux(1,idxOUT);
|
|
c_out = c_B(1,idxOUT);
|
|
index_out = index_B(1,idxOUT);
|
|
|
|
% Update the matrices by exchanging the columns
|
|
B(:,idxOUT) = in;
|
|
D(:,idxIN) = out;
|
|
c_Baux(1,idxOUT) = c_inaux;
|
|
c_Daux(1,idxIN) = c_outaux;
|
|
c_B(1,idxOUT) = c_in;
|
|
c_D(1,idxIN) = c_out;
|
|
index_B(1,idxOUT) = index_in;
|
|
index_D(1,idxIN) = index_out;
|
|
|
|
% Compute B^{-1}*D and B^{-1}*h
|
|
BiD = B\D;
|
|
Bih = B\h;
|
|
|
|
% Compute reduced cost coefficients
|
|
r_D = c_Daux - c_Baux*BiD;
|
|
|
|
% Detect inefficient loop if nIter > total number of basic solutions
|
|
nIter = nIter + 1;
|
|
if(nIter>itMax)
|
|
error('The original LP problem does not admit a feasible solution.');
|
|
end
|
|
|
|
x_B = Bih - BiD*x_D;
|
|
z = c_Baux*x_B;
|
|
|
|
end
|
|
|
|
check = index_D<(n+m+1);
|
|
D = D(:,check);
|
|
index_D = index_D(1,check);
|
|
c_D = c_D(1,check);
|
|
x_D = x_D(check,1);
|
|
|
|
end |