Defects Index | Documentation Table of Contents
97/08/01 09:30 DEFECT in program page4
SYMPTOM: IF site coordinates' estimated shifts are greater than a few
decimeters and one attempts to fix integers, then solution degrades.
PROBLEM: Program does two tasks completely ignorant of each other:
a) add fixed integer and turn off estimate of that bias,
b) remove a priori bias to keep observed-computed (O-C) phase
small eliminating loss of precision in sum of squares during
matrix manipulation (basement window).
With infinit precision, the program could:
1) without fixing integers or removing a priori bias, create an
O-C and estimate complete bias.
2) fixing integers but still without removing a priori bias,
create an (O-C)' and estimate complete bias where necessary.
Key point is that the O-C between the two cases is and should be
different because the fixed integers are a modification to the
models and cause a corresponding change in the estimation strategy.
The program was:
1) without fixing integers but removing a priori bias, correctly
creating an (O-C), estimating a small bias, adding the a priori
bias back recreating the complete bias.
2) fixing integers but by still removing a priori bias, incorrectly
recreating an (O-C) indentical to case 1. However, the
estimation strategy was now different and the results were
incorrect estimates.
Finally, if the positions were close enough, the a priori bias
removed would exactly match the fixed integer. This would
inadvertently degenerate into a correct solution.
CORRECTION: Modified obsdd, where the double-difference phase O-C is created,
to handle these two cases:
1) integer is not fixed; a priori bias is removed as before,
2) integer is fixed; a priori bias is not removed giving
O-C that is different that case 1, but correct for this case.
FOUND BY: L Hall and S Hilla
FIXED BY: M Schenewerk
VERSION: Beta
SOURCE: obsdd.f
NOTES: Will go out with next upgrade.
Hilla has but in place for Hall.
970801.page4
June 25, 1999
Steve Hilla