16
I + (N - B 1 * U B) L
I
b 1
B t
U B N" 1
U B N" 1
L" 1 = N - B 1 U B
L" 1 = NB" 1 U" 1 (B*)" 1 (N - B 1 U B)
= NB” 1 U" 1 (B 1 )" 1 N - NB" 1 B
= NB" 1 (U" 1 - BN" 1 B t )(B t )" 1 N
then
L = - N" 1 B t (U" 1 - BN" 1 ^ 1 )" 1 BN" 1
• • • •
4.26
Substituting 4.26 into 4.25 we get:
M" 1 = N" 1 + N" 1 6 t (U" 1 - BN" 1 ^)" 1 BN" 1 .... 4.27
. . . • -1
Equation 4.27 is the relationship sought. The inverses U and
(U" 1 - BN" 1 ^ 1 )" 1 are of sizes that range from 2x2 to a maximum of
16 x 16 in a 3x3 sub-block. These inverses are still smaller than the
size of the entire system (47 x 47). In addition, the use of equation
4.27 reduces considerably the amount of computations that would be
required to evaluate M - ^ from the retained observation equations. Also,
equation 4.27 can be applied in an iterative process, eliminating only
one point in each cycle, then modifying the new inverse. The modified
vector of constant terms Q can be easily computed from equation 4.22 as
Q=P-B t UF ....4.28
which also possesses the characteristic of stepwise application.
It is interesting to note that equations 4.27 and 4.28 may be
of considerable value for the general problem of least squares, particularly
for large systems of linear equations. Assuming that the general form of
the observation equations (corresponding to 4.01) is:
KV + GE + F = 0
.... 4.29
the generalized equations corresponding to 4.27 and 4.28 become
M' 1 = N" 1 + N“ 1 G t (KW” 1 K t - GN“ 1 G t ) -1 GN" 1 .... 4.30