version 1.1 (1997)
F. A. Q.
- Linux users: Will your C code run under Linux?
- Yes, but comment out the line "#include <math.h>", because
Linux C uses the HP C math macros as functions.
- The ps-files are not readable by ghostview.
- GSview32 (version 2.1 and higher for Windows) and
ghostview (version 1.5 and higher for Unix, properly installed) read the files correctly.
- The ps-files are not printable.
- The files can be printed on any HP Postscript printer. We did not try to print these
files on other printers.
- I don't like to pass the parameters to the functions fun and grad by GLOBAL declarations. Why don't you use the same technique as the one used in the Matlab Optimization Toolbox?
- Actually, the evaluating strings
feature provided by the Matlab function eval can be used in SolvOpt, but we avoid using it. The reasons are: (i) this
causes logical (structural) changes in the code and (ii) we cannot see any danger of using global declarations.
However, we shall definitely change this in the next version.
- The solver fails to find the minimum for some known test functions.
- Yes, this may happen with the current version too. There are some nasty cases, where every
optimization routine fails. To learn more about these cases visit the page
Moré set of test functions.
In general, the worst case for the solver is a "ravine" with very steep and almost
parallel walls and a flat bottom.
- Is SolvOpt an interior point solver?
- Certainly, it is not. SolvOpt uses penalization of the constraints, therefore, it is required that an objective
function is defined everywhere. (See the next item).
- The objective function is undefined outside the feasible set. Is it still possible to
apply SolvOpt for the case?
- Yes, it is. However, this requires the use of the user defined penalty function. At every point where the
objective function is not defined, the user has to complete the function definition by setting it to a linear
penalty function with sufficiently large (but finite) penalty coefficient. The user has to choose a suitable penalty
- Solvopt fails quite frequently when using it for solving a
linear program because of "function is extremely steep at the optimum" (which is
not the case).
- This message concerns the penalty function which is an objective for the solver.
Actually, this happens when the LP (NLP) problem has no feasible point. The warning
message comes to indicate an irregular case, however, one can trust the solution
obtained. It minimizes constraint violations in the case).
Performance and tests.
Alexei Kuntsevich |
& Franz Kappel |
August 15, 1998