MSIS 685: Linear Programming Lecture 8 10/29/98 Scribe: Konstantinos Moyssiadis In this lecture will see the simplex method in matrix notation. Let's recall first the matrix notation for the Prime and the Dual Problem. Primal Dual max c T x st . Ax b min b T y st . AT y c y0 x0 With the addition of the slack variables we get: max c T x st . Ax b x 0 min b T y st . AT y c y0 Where: x x , w y y, z A [ A, I ] A [ AT / I ] c [c ,0] b [b ,0] Let B represent the set with the basic variables: B={j1,j2,…,jm} Then: x=[xB,xN] y=[yB,yN] and To solve the linear Prime Problem, we just follow the same steps but using the matrix notation instead: Ax=b BxB+NxN=b xB=B-1-B-1NxN Hence, for the objective function: ζ=cTx=cBTxB+cNTxN =cBT(B-1b-B-1NxN)+cNTxN =cBTB-1b-(cBTB-1N-cNT)xN We secure feasibility when: B-1b>0 and optimality when: cBTB-1N-cNT > 0 Working similarly for the dual problem we get: (yN is the basic dual matched with the xN prime non basic, and yB the non basic dual matched with the xB prime basic) yTA=cT If we use the prime matrix, then for the dual matrix we get by converting the lines to columns and negating: c TB B 1b 1 T (B N) c B c N - (B -1 b) T - (B -1 N) T Therefore, for the objective function and the variables we have respectively: ζ= c TB B 1b - (B -1 b) T yN= (B 1 N) T c B c N - (B -1 N) T To secure feasibility: (B 1 N) T c B c N 0 And optimality: B-1b>0