1%***************************************************************************** 2% DSDP5: Dual-Scaling Algorithm for Positive Semidefinite Programming 3% Copyright (c) 2004 by 4% S. J. Benson, Y. Ye 5% Last modified: 20 August 2004 6%***************************************************************************** 7% 8% > [STAT,y] = DSDP() returns a structure STAT with relevant information 9% concerning the performance of the solver and an approximate 10% dual solution y . 11% 12% The fields of STAT are: 13% 14% Objective Value: 15% stype = 'PDFeasible' if an feasible primal and dual solutions 16% were computed, 'Infeasible' if dual 17% infeasibility was detected, and 'Unbounded' if primal 18% infeasibility is detected. 19% obj = objective value at solution 20% pobj = an approximately optimal objective value to (P) 21% dobj = an approximately optimal objective value to (D) 22% stopcode = 0: convergence to prescribed accuracy, 23% ~0: termination for other reasons 24% 25% Characteristics of Solution 26% tracex = if X was returned, this is the trace of it. This number 27% also corresponds to the minimum penalty parameter that 28% could solve this problem. IMPORTANT: For improved 29% performance, consider using penalty parameter (see DOPTIONS) 30% other than the default. 31% penalty = the penalty parameter used by the solver, which must be 32% greater than the trace of the primal solution (see above). 33% errors = several error estimates to the solution. (See DERROR) 34% ynorm = the largest element of y (infinity norm). 35% boundy = the bounds placed on the magnitude of each variable y. 36% mu = final barrier parameter. 37% r = the multiple of the identity matrix added to 38% C-A'(y) in the final solution to make S positive definite. 39% That is, S = C - A'y + r*I. 40% xy, xdy, xmu = values used to compute X. 41% 42% Solver Statistics 43% iterates = number of iterations used by the algorithm. 44% pstep = final primal step size. 45% dstep = final dual step size. 46% pnorm = final norm of distance to central path. 47% gaphist = a history of the duality gap. 48% infhist = a history of the dual infeasibility. 49% datanorm = the Frobenius norm of C, A and b, 50% 51% See also: DSDP 52% 53% DSDP5 54% Copyright (c) 2004 by 55% S. Benson and Y. Ye 56% Last modified: October 2004 57%***************************************************************************** 58 59