priori given. The search of these values and the preprocessing of the 
measurements are allowed by the ITM programs package /19/, /20/; how- 
ever the software compatibility between this data management and the 
adjustment is not yet fully achieved, : 
The solution method is performed by using two libraries of subroutines 
based on graph theory and linear algebra, that have been previously 
implemented. The former concerns a reordering algorithm of reverse Cut 
Hill, McKee type /13/ and its application to the least squares problems 
/6/, /5/; the latter concerns the algorithms necessary to calculate the 
direct solution. 
A comparison between direct and iterative algorithms in least Squares 
problems /18/, /15/ has given good results /7/, /8/; therefore a second 
version of the same program has been performed by using the iterative 
Solution. A ng 
All the observationsare treated as independent in the statistical 
Sense and the stochastic model is assumed as given a-priori. The pro- 
gram furnishes the standardized residuals, thus allowing the applica- 
tion of the Baarda snooping /2/, /3/, /16/; however the sequential al- 
The program CALGE is written in the language FORTRAN 77, it runs on the 
computer UNIVAC 1100/90 of the CILEA, all the array variables are decla 
red as virtual memory and the numerical array ones are in double preci- 
sion. 
2. ACCELERATION OF THE INVERSION ROUTINE /12/ 
This research start from the remark that the computing time for inver- 
sion of sparse large matrices is much more than two times the computing 
time of Cholesky factorization, while the number of operations involved 
in the inversion is asymptotically about two times the number of the 
operations for the factorization. 
It is to be observed that the execution time for inversion is bursting 
time, spoiled in making tests to find the required non-zero elements. 
So a new strategy is wanted to optimize the adjustment package and to 
recover the correct ratio. : 
The basic formulas to compute the inverse elements are presented in 
this following expression: | 
n 
adi 7 
Y En kei*l "ik uj, t, 
iri 2 
= we L i. e . 
Y d'est "uu ti te itis dui; 
part A part B 
The elements of the triangular Cholesky factor run only on rows: this 
requires a testing procedure which cannot be.eliminated.The procedure 
needs of n sequences of tests; each sequence admits, starting from 
the bottom, at most two alternatives: