Figure 1: This graph shows the drop in the value of the function that
is minimized in refinement over 20 cycles of refinement. Four different
methods of minimization are compared. In some test runs (solid lines) only the
positional parameters were varied while in the rest (broken lines) both the
positional and thermal parameters were varied.
The function is 
after the 
's and 
's
have been scaled to each other, plus the sum of the geometry deviation terms.
The methods represented with triangles
required 18.5 minutes of CPU time per cycle on a VAX 3600 computer. The
methods represented with squares required the additional calculation of
curvatures and took 22 minutes per cycle.
This plot demonstrates that the conjugate direction method produces a lower
function value for a given number of cycles of refinement.