The focus of this note is to highlight two relatively simple approache
s to determine the matrix projection first introduced by Crowe et al.
(1983) to solve data reconciliation problems when unmeasured variables
exist. The first method uses recursive matrix inversion by partition
where the second uses a modified Cholesky factorization. The purpose o
f the two algorithms is to identify dependent columns and rows in the
topology matrix of the unmeasured variables where the matrix projectio
n is then formulated. Although other methods are available to determin
e the nullity of a matrix such as QR factorization and singular value
decomposition, it is preferred to use this identification procedure al
ong with Crowe's matrix projection formulation because of its numerica
l efficiency, simplicity and interpretation. Two mass reconciliation e
xamples are presented, one small and one large, to clarify and verify
the techniques. (C) 1998 Elsevier Science Ltd. All rights reserved.