DONLP2  with AMPL 


This directory contains the  DONLP2 executable for

  *********************
  * Intel/Linux 32bit *
  *********************



This executable works together with the respective AMPL student binary
"sampl"  which is included for
convenience here. In this the maximum number of variables and constraints
is limited to 300. This means that the total number of bounds and general
constraints is <= 300 for DONLP2.




 

    

  • 


 donlp2.gz  is the gzip compressed binary of donlp2
with the ampl interface in the student's version. 
     

  • 


 sampl.gz   is  the students ampl gzip compressed binary. 
    

  • 


 donlp2_options  describes the options for the solver, see also below.
 
    

  • 

models  is a subdirectory with many examples coded in ampl. 
    

  •  

test  is a directory with a standard test for donlp2 with results.

    

  • 

results is the directory with the results from the examples in the models 
   directory. 
     

  • 

amplbook chapter1
 is chapter one from the AMPL textbook showing how to 
   work with ampl respectively with sampl. 
    








The directory 
  models 
contains AMPL models for some well known nonlinear optimization problems. They
may be used as examples how to code a model in AMPL. 

 

To use sampl with donlp2 one may proceed in two ways:



Either one writes the sequence of ampl-commands in a file file.mod and invokes 

      sampl [options] file.mod  

with prior exporting of the environment variable 
   
   export OPTIONS_IN=filename ;  (in the ksh) 

or 

   setenv options donlp2_options "<....your desired settings>" ( in csh)


or one invokes 

   sampl - 

and then types commands interactively. 
 

for example this: 

 model  {filename with .mod extension: the ampl coded problem} ;
 
 option solver donlp2; 

 option donlp2_options 'tau0=1.e4 silent=0 del0=1.0 outlev=1 wantsol=7'; 

 solve; 

 display {AMPL variables you want to see, if not in the output} ; 

 quit; 


don't forget the semicolon !!! 




In order to use DONLP2 with sampl the options to be set are 


option solver donlp2; 

option donlp2_options ".....your desired settings"; 



donlp2_options are the following:







maxfwd   max vars in fwd automatic differentiation of common exprs (default 5)




tau0     sum of constraint violations allowed during iteration (default 1.d0)




del0     initially an inequality g is binding if g/max{1,||grad g||}<=del0 (defa
ult 1.d0)




epsx     required precision in primal solution (default 0.00001)




maxit    iteration limit (<= 4000)




nreset   allowed number of successive steps with insignificant changes in x or f 




intakt   write iteration protocol also to stdout =0 off (default) =1 on 




silent   write results and messages on special events to file *.pro and *.mes (d
efault =1: no) 




outlev   composed from 4 bits:

         bit 1 ="0" means no output , "1" one line per step,

         bit 2 ="1" short iteration protocol

         bit 3 ="1" detailed iteration protocol (primal and dual sol's, stepsize
s etc.)
 
         bit 4 ="1" print also gradients and hessian estimate




prou     unit number for output protocol




meu      unit number for special events messages protocol




wantsol  solution report without -AMPL: 4 bit integer namely sum of

                1 ==> write .sol file

                2 ==> print primal variable values

                4 ==> print dual variable values

                8 ==> do not print solution message








For example, one may set 

option donlp2_options "tau0=0.1 del0=0.1 silent=0 outlev=4 wantsol=7"; 



Possible solve_result_num values for donlp2: 






0       Success! KKT conditions satisfied 




100     nearly feasible with small directional deriv.




101     relaxed KKT conditions satisfied: singular point




110     small QP correction: nearly feasible and singular




200     tiny correction from QP at infeasible point




400     reached maxit steps




500     N too large (the Fortran donlp2 routines must be recompiled)




501     unknown termination reason




510     cannot evaluate constraints




511     cannot evaluate objective




520     dual extended QP failure: singular working set




521     infeasible QP (theoretically impossible) 




522     no descent in QP (theoretically impossible) 




523     tiny QP step from infeasible point




524     nondescent direction from QP




525     no acceptable step size




527     tiny step




528     more than NRESET small primal corrections




529     max(N,10) small penalty function differences




540     slow primal progress: singular or ill-conditioned problem?





Scaling (new 19980606):  in an AMPL session, if you declare 


        option suffix vsclog10; 


then variable.vsclog10 (default 0) is passed to donlp2, which scales
the variable internally by 10^-variable.vsclog10.  (x is the ampl name for the
variable) Thus if variables x and y are distances in meters that have typical values of a few
centimeters and a few kilometers, respectively, then

        let x.vsclog10 := -2; 

        let y.vsclog10 := 3; 

would probably be appropriate, since it would give the internal
variables typical values around 1. 






                ----------------------------------


Test example:  As an example the file weapon.mod from the models directory
was modified to the one stored in the directory test and the commands


     sampl 

     model weapon.mod 


were executed. As a result the ouput in weapon.out was printed to the
screen and the two files   XtmpXat1.MES  XtmpXat1.PRO  were generated. 

This is contained in 
test
                ----------------------------------

The entire set of problems in the directory models was run with the
exported option of intakt=1. The output is in the directory results
.


 
The  sampl command:
 

Usage: sampl [options] [file [file...]] 



No file arguments means read from standard input, as does - by itself.
that is you invoke sampl and then type your commands interactively



Options:







    

  •  -Cn {0 = suppress Cautions; 1 = default; 2 = treat as error;}*

  •         -L {fully eliminate linear definitional constraints and var = decls}*

  •         -P {skip presolve -- same as "option presolve 0;" }
 
  •        -S {substitute out definitional constraints (var = expression)}*
 
  •        -T {show genmod times for each item}*

  •         -enn {exit (if nn > 0) or abort command at |nn|-th error; 0 = no exit}*

  •         -f {do not treat unavailable functions of constant args as variable}*

  •         -ooutopt {specify -o? for details}*

  •         -s[seed] {seed for random numbers; -s means current time}*

  •         -t {show times}*

  •         -v {show version; -v? shows other -v options}

  •         -?? {show debug options}




* Equivalent option settings:






    

  •    -Cn      option Cautions n;  

  •     -L      option linelim 1;

  •     -P      option presolve 0;    

  •     -S      option substout 1;

  •     -T      option gentimes 1; 

  •     -enn    option eexit nn;

  •     -f      option funcwarn 1;

  •     -ofstub option outopt fstub;

  •     -s      option randseed '';

  •     -snn    option randseed nn;

  •     -t      option times 1;



-v options:






    

  •     -v? {show -v options}

  •     -v2 {merge stderr with stdout (subprocesses as well as ampl)}

  •         -vd[nnn] {debug level nnn (default 3)}

  •         -ve {write ampl error messages on stdout rather than stderr}

  •         -vf {show available user-defined functions}

  •         -vp {write prompts on stdout rather than stderr}

  •         -vq {quit now}

  •         -vr {show builtin random functions}

  •         -vs {show statistics}

  •         -vv {show version}





sampl -o?






    

  •     -o0 {no files written}

  •         -o! {no genmod and no output; not for use with write command}

  •         -obstub {generic binary format -- line -og, but binary

  •                 after the first 10 lines}

  •         -ogstub {generic ASCII format: no MPS file, no .spc file,

  •                 nonlinearities in .nl file, otherwise like -omstub}

  •         -om {MPS format to stdout, no other files written}

  •         -omstub {MPS format to stub.mps,

  •                 names to stub.row & stub.col,

  •                 fixed vars to stub.fix,

  •                 unused vars to stub.unv

  •                 slack constraints to stub.slc

  •                 objective adjustments to stub.adj,

  •                 MINOS SPECS to stub.spc}

  •         -onstub {MPS format to , otherwise like -omstub}





debug options:








    

  •         -# {precede -MDG output with #}

  •         -D {print data read in}

  •         -F {force generation of := sets}

  •         -G {print generated data}

  •         -M {print model}

  •         -O {print compiled model}

  •         -b {block I/O mode}

  •         -pnn {use nn decimal places in converting numbers to symbols}

  •         -q {always quote output literals (under -D, -G, -M, -O)}

  •         -r {restricted mode}

  •         -z {lazy mode -- treat := as default (evaluate only as needed);
                same as "option lazy 1;" }




+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++


  A M P L 



AMPL is a language and system for formulating, solving, and helping
understand mathematical programming problems (of minimizing or
maximizing a function subject to constraints).  The AMPL book
explains AMPL in much more detail and provides many examples:
It is available in the departments library
for more information see 


AMPL  The AMPL company.