Cvxopt

Cvxopt provides many routines for solving convex optimization problems such as linear and quadratic programming packages. It also has a very nice sparse matrix library that provides an interface to umfpack (the same sparse matrix solver that matlab uses), it also has a nice interface to lapack. For more details on cvxopt please refer to its documentation at http://cvxopt.org/userguide/index.html

Sparse matrices are represented in triplet notation that is as a list of nonzero values, row indices and column indices. This is internally converted to compressed sparse column format. So for example we would enter the matrix

\[\begin{split}\left( \begin{array}{ccccc} 2&3&0&0&0\\ 3&0&4&0&6\\ 0&-1&-3&2&0\\ 0&0&1&0&0\\ 0&4&2&0&1 \end{array}\right)\end{split}\]

by

sage: import numpy
sage: from cvxopt.base import spmatrix
sage: from cvxopt.base import matrix as m
sage: from cvxopt import umfpack
sage: Integer=int
sage: V = [2,3, 3,-1,4, 4,-3,1,2, 2, 6,1]
sage: I = [0,1, 0, 2,4, 1, 2,3,4, 2, 1,4]
sage: J = [0,0, 1, 1,1, 2, 2,2,2, 3, 4,4]
sage: A = spmatrix(V,I,J)

To solve an equation \(AX=B\), with \(B=[1,1,1,1,1]\), we could do the following.

sage: B = numpy.array([1.0]*5)
sage: B.shape=(5,1)
sage: print(B)
[[1.]
 [1.]
 [1.]
 [1.]
 [1.]]


sage: print(A)
[ 2.00e+00  3.00e+00     0         0         0    ]
[ 3.00e+00     0      4.00e+00     0      6.00e+00]
[    0     -1.00e+00 -3.00e+00  2.00e+00     0    ]
[    0         0      1.00e+00     0         0    ]
[    0      4.00e+00  2.00e+00     0      1.00e+00]
sage: C=m(B)
sage: umfpack.linsolve(A,C)
sage: print(C)
[ 5.79e-01]
[-5.26e-02]
[ 1.00e+00]
[ 1.97e+00]
[-7.89e-01]

Note the solution is stored in \(B\) afterward. also note the m(B), this turns our numpy array into a format cvxopt understands. You can directly create a cvxopt matrix using cvxopt’s own matrix command, but I personally find numpy arrays nicer. Also note we explicitly set the shape of the numpy array to make it clear it was a column vector.

We could compute the approximate minimum degree ordering by doing

sage: RealNumber=float
sage: Integer=int
sage: from cvxopt.base import spmatrix
sage: from cvxopt import amd
sage: A=spmatrix([10,3,5,-2,5,2],[0,2,1,2,2,3],[0,0,1,1,2,3])
sage: P=amd.order(A)
sage: print(P)
[ 1]
[ 0]
[ 2]
[ 3]

For a simple linear programming example, if we want to solve

\[\begin{split}\begin{array}{cc} \text{minimze} & -4x_1-5x_2\\ \text{subject to} & 2x_1 +x_2\le 3\\ & x_1+2x_2\le 3\\ & x_1 \ge 0 \\ & x_2 \ge 0\\ \end{array}\end{split}\]
sage: RealNumber=float
sage: Integer=int
sage: from cvxopt.base import matrix as m
sage: from cvxopt import solvers
sage: c = m([-4., -5.])
sage: G = m([[2., 1., -1., 0.], [1., 2., 0., -1.]])
sage: h = m([3., 3., 0., 0.])
sage: sol = solvers.lp(c,G,h) #random
     pcost       dcost       gap    pres   dres   k/t
 0: -8.1000e+00 -1.8300e+01  4e+00  0e+00  8e-01  1e+00
 1: -8.8055e+00 -9.4357e+00  2e-01  1e-16  4e-02  3e-02
 2: -8.9981e+00 -9.0049e+00  2e-03  1e-16  5e-04  4e-04
 3: -9.0000e+00 -9.0000e+00  2e-05  3e-16  5e-06  4e-06
 4: -9.0000e+00 -9.0000e+00  2e-07  1e-16  5e-08  4e-08

sage: print(sol['x'])    # ... below since can get -00 or +00 depending on architecture
[ 1.00e...00]
[ 1.00e+00]