Let
be an
matrix of matrix rank
over a field.
A rank factorization, also called a full-rank factorization or rank decomposition,
is a factorization
where
is an
matrix of full column rank and
is an
matrix of full row rank.
A rank factorization can be constructed from the reduced row-echelon form of . Let
consist of its nonzero rows, and let
consist of the columns of the original matrix
whose indices are the columns containing the pivot
elements of
. The rows of
form a basis for the row
space, the columns of
form a basis for the column space,
and
(Piziak and Odell 1999).
A rank factorization is generally not unique. For any invertible matrix
of size
,
is another rank factorization.