Two options exist for the least-squares solution of the network adjustment, Numpy or Gravnet (Hwang et al., 2002). Gravnet is a Fortran program, and a compiled executable for Windows is provided with GSadjust. Gravnet uses Cholesky decomposition to invert the design matrix and solve the network adjustment system of equations. The numpy inverse solution (using numpy.linalg.inv) uses the LAPACK linear algebra package's Gaussian elimination routine.
For most network adjustment solutions the Numpy and Gravnet solutions will be identical, and run equally fast. There are three main differences in functionality:
Gravnet only accomodates a single relative-gravity meter when calculating a calibration coefficient; the Numpy solution will calculate a unique calibration coefficient for each relative-gravity meter.
When drift is included as a polynomial to be solved for in the least squares solution, Gravnet is limited to a single polynomial for all observations in a Survey. That is, if the drift method on the Drift tab is set to "Network adjustment", all Loops in the same Survey must also be set to "Network adjustment", all with identical polynomial degrees. With the Numpy solution, the drift method can be set independently for each loop. If necessary, the "Network adjustment" method may be used for some loops, and an alternative method used for other loops within the same survey.
Gravnet includes Pope’s t-test method to test for outliers; Numpy inversion does not perform outlier tests.
Gravnet can also solve for circular (screw) error, but this option is not currently implemented in GSadjust.
Gravnet is provided as a Windows executable; it will not run on non-Windows computers unless the Fortran source code is compiled for the specific platform.
Hwang, C., C. Wang, and L. Lee (2002), Adjustment of relative gravity measurements using weighted and datum-free constraints, Comput. Geosci., 28, 1005–1015.
