MATLAB Function Reference Previous page   Next Page

Column approximate minimum degree permutation



p = colamd(S) returns the column approximate minimum degree permutation vector for the sparse matrix S. For a non-symmetric matrix S, S(:,p) tends to have sparser LU factors than S. The Cholesky factorization of S(:,p)' * S(:,p) also tends to be sparser than that of S'*S.

knobs is a two-element vector. If S is m-by-n, then rows with more than (knobs(1))*n entries are ignored. Columns with more than (knobs(2))*m entries are removed prior to ordering, and ordered last in the output permutation p. If the knobs parameter is not present, then knobs(1) = knobs(2) = spparms('wh_frac').

stats is an optional vector that provides data about the ordering and the validity of the matrix S.

Number of dense or empty rows ignored by colamd
Number of dense or empty columns ignored by colamd
Number of garbage collections performed on the internal data structure used by colamd (roughly of size 2.2*nnz(S) + 4*m + 7*n integers)
0 if the matrix is valid, or 1 if invalid
Rightmost column index that is unsorted or contains duplicate entries, or 0 if no such column exists
Last seen duplicate or out-of-order row index in the column index given by stats(5), or 0 if no such row index exists
Number of duplicate and out-of-order row indices

Although, MATLAB built-in functions generate valid sparse matrices, a user may construct an invalid sparse matrix using the MATLAB C or Fortran APIs and pass it to colamd. For this reason, colamd verifies that S is valid:

The ordering is followed by a column elimination tree post-ordering.


The Harwell-Boeing collection of sparse matrices and the MATLAB demos directory include a test matrix west0479. It is a matrix of order 479 resulting from a model due to Westerberg of an eight-stage chemical distillation column. The spy plot shows evidence of the eight stages. The colamd ordering scrambles this structure.

Comparing the spy plot of the LU factorization of the original matrix with that of the reordered matrix shows that minimum degree reduces the time and storage requirements by better than a factor of 2.8. The nonzero counts are 16777 and 5904, respectively.

See Also

colmmd, colperm, spparms, symamd, symmmd, symrcm


[1]  The authors of the code for colamd are Stefan I. Larimore and Timothy A. Davis (, University of Florida. The algorithm was developed in collaboration with John Gilbert, Xerox PARC, and Esmond Ng, Oak Ridge National Laboratory. Sparse Matrix Algorithms Research at the University of Florida:

Previous page  cmopts colmmd Next page

© 1994-2005 The MathWorks, Inc.