共轭梯度法编程（Conjugate gradient programming）.doc

共轭梯度法编程（Conjugate gradient programming） P90 second, 11 (2) Solving with large M method Min w=2x1+x2-x3-x4 S.T x1-x2+2x3-x4=2 2x1+x2-3x3+x4=6 X1+x2+x3+x4=7 Xi = 0, I = 1, 2, 3, 4 Matlab is solved as follows: F=[2,1, -1, -1200200200]; A=[1 -1 2-1100; 21-31010; 111100 1]; B=[2 6, 7]; Lb=zeros (7,1); [x, Fval, exitflag, output, lambda]=linprog (F, [], [], a, B, LB) Optimization terminated. The running results are as follows: = x Three Zero One Three Zero Zero Zero = Fval Two Exitflag = One Output = Iterations: 7 Algorithm:large-scale: interior point Cgiterations: 0 Message:Optimization terminated. Lambda = Ineqlin: [0x1 double] Eqlin: [3x1 double] Upper: [7x1 double] Lower: [7x1 double] From the above results, it can be concluded that the optimal solution is x= and the minimum is about f*=2. P151, chapter third, 26 Conjugate gradient algorithm is used to find the minimal point of F (x) = (x1-1) ^2+5* (x2-x1^2) ^2, and take the initial point x0=. Matlab is solved as follows: Function, mg=MG () % Solving problem three and twenty-sixth with% conjugate gradient method % Clc; Clear; N=2; X=[2 0]; Max_k=100; Count_k=1; Trace (1,1) =x (1); Trace (2,1) =x (2); Trace (3,1) =f_fun (x); K=0; G1=f_dfun (x); S=-g1; While count_k=max_k If k==n G0=f_dfun (x); S=-g0; K=0; Else R_min=fminbnd (@ (T), f_fun (x+t*s), -100100); X=x+r_min*s; G0=g1; G1=f_dfun (x); If norm (G1) 10^ (-6) Break; End M= (norm (G1) ^2) / (norm (G0) ^2); S=-g1+m*s; Count_k=count_k+1; Trace (1, count_k) =x (1); Trace (2, count_k) =x (2); Trace (3, count_k) =f_fun (x); K=k+1; End End Count_k X F=f_fun (x) Function g=f_dfun (x) G (1,1), =20*x (1), ^3-20*x (1), *x (2), +2*x (1), -2; G (2,1) =10*x (2) -10*x (1) ^2; Function f=f_fun (x) F=5*x (1), ^4-10*x (1), ^2*x (2), +x (1), ^2-2*x (1), +1+5*x (2) ^2; The running results are as follows: = k Fifty-nine = X0 One One = F0 Zero From the above results, it can be concluded that the optimal solution is x= and the minimum is about f*=0. P151, chapter third, 26 Using BFGS algo