gmane.comp.mathematics.globsol.user

GlobSol: Changing first line of *.DT1 file does the trick Re: Stopping criteria problem

Ralph Baker Kearfott — 2011-03-09T16:58:48

Pete,

I've attached my result for your problem, along with the
corresponding alteration to the *.DT? file and your original
*.f90 file.  Although this was done with my experimental version
of GlobSol, I am fairly confident it will do the trick for you.

Baker

=============================================================

Pete,


Try changing 1D-5 in the first line of HeH.321G.DT1 to 1D-10.

Also, you may wish to send me your output file.  (I am presently
checking your problem against a later version of GlobSol, although I
can also check it against the "windows" version as appropriate.)

Finally, please call on me if there is any question concerning 
interpreting the output.  (The best upper bound found on the
global optimum is BEST_ESTIMATE.)

Best regards,

Baker

On 03/09/2011 07:06 AM, Pete Janes wrote:

GlobSol: Re: Stopping criteria problem

Ralph Baker Kearfott — 2011-03-09T14:23:45

(Posted to the GlobSol users' forum list)

Pete,


Try changing 1D-5 in the first line of HeH.321G.DT1 to 1D-10.

Also, you may wish to send me your output file.  (I am presently
checking your problem against a later version of GlobSol, although I
can also check it against the "windows" version as appropriate.)

Finally, please call on me if there is any question concerning 
interpreting the output.  (The best upper bound found on the
global optimum is BEST_ESTIMATE.)

Best regards,

Baker

On 03/09/2011 07:06 AM, Pete Janes wrote:

GlobSol: Stopping criteria problem

Pete Janes — 2011-03-09T13:06:54

Dear Baker,

Beginner question.

I am using GlobSol to solve some energy optimization problems
(constrained polynomial programming type). I am having problems
configuring the stopping criteria so that it only stops at an absolute
gap of 1.6D-3 between the lower bound and upper bound of the global
minima. 

I am essentially trying to meet the epsilon-optimality condition.

Right now it is stopping when the box coordinates, and PHI is still
fairly wide. I have attached my f90, DT1, and .CFG files as a reference.

Kind regards,

Pete
1D-5
 -1   1
 -1   1
 -1   1
 -1   1
T T
T T
T T
T T
! File: GlobSol.CFG  Configuration parameters for the GlobSol package
!
! =============================================================

! OVERLOAD.CFG -- Section to configure how the code list is
!                 created and used.

! This section defines default values for the code list(s) configuration
! You may edit the values to suit your application.  We suggest using
! values about 3-5 times larger than you might guess, bu

Re: GlobSol: NAGARCH (1,1) using GlobSol

Ralph Baker Kearfott — 2011-01-10T15:36:16

(Posted to the GlobSol users' forum list)

Berislav,

One problem I see immediately is that you have dimensioned volpr with m locations, but
you set volpr(m+1) in the "DO" loop "DO i = 1,m".

Once you get this fixed, I'd be interested in how it proceeds.  In particular, it
is a least squares problem with a large number of data points, and your formulation
may lead to a significant amount of overestimation in the ranges, due to the
"interval dependency" phenomenon.  Thus, I'm not particularly optimistic about
success on this particular problem with this particular formulation.  However,
I may be able to work with you to reformulate the problem more appropriately.
(Think in terms of "factoring out" the X(I)'s wherever possible.)

Sincerely,

Baker

On 1/3/2011 07:14, Berislav Buca wrote:

GlobSol: NAGARCH (1,1) using GlobSol

Berislav Buca — 2011-01-03T13:14:04

Hello!

I'm using GlobSol to fit a NAGARCH (1,1) of S&P 500 returns. I ran the
program on both a Windows XP machine and a Windows 7 machine.

I keep getting an error message:

 "  Gradient code list ...
        1 file(s) copied.
Exception: Access Violation
At line 48 of file c:\GlobSol\overload\intlist.f90
Traceback: not available, compile with -ftrace=frame or -ftrace=full
 Error in opening the code list file in subroutine GET_CODELIST.
 The supplied code list file name is: nagarch1G.CDL
 Does this file exist?
Could Not Find C:\GlobSol\integration_test_data\nagarch1G.CDL
The system cannot find the file specified.
   Find global minima for  nagarch1 ...
 Error in opening the code list file in subroutine GET_CODELIST.
 The supplied code list file name is: nagarch1G.CDL
 Does this file exist?
   To see results, edit  nagarch1.OT1
"

The code is here:
PROGRAM nagarch
USE CODELIST_CREATION

IMPLICIT NONE
INTEGER :: I, J, m
parameter (m=1000)
DOUBLE PRECISION, DIMENSION(m) :: P, r
TYPE(CDLVAR)       :: f, delta,

GlobSol: Windows version not updated Re: Problem identified and partially solved

Ralph Baker Kearfott — 2010-04-14T15:00:00

(Posted to the GlobSol users' forum list)

GlobSol users,

Although I have fixed the problem in my development version, I have noticed
that I need to do some more work to update the Windows version. Thus, the
Windows version will not be updated until I modify the packaging process.

In the mean time, the Windows version should work on problems that do not
have denominators that can contain 0.

If anyone is interested in using the development version or helping with it,
contact me.

Baker
=============================================================================

(Posted to the GlobSol users' forum list)

Jorge and globsol:

The reason that GlobSol stopped subdividing before finding a small box containing
the optimizers was due to a comparison of quantities where both quantities
were infinite.  In particular, the problem was in CHECK_SMEAR_DIAMETER in
f90intbi/rigorous_global_search.f90:

The problem is in CHECK_SMEAR_DIAMETER, both sides of the inequality
test here are infinite, so it doesn't ever test "t

Problem identified and partially solved Re: GlobSol: Problem: "SEVERITY 2 ERROR: ZERO IN DENOMINATOR IN ORDINARY INTERVAL DIVISION."

Ralph Baker Kearfott — 2010-04-13T00:36:14

(Posted to the GlobSol users' forum list)

Jorge and globsol:

The reason that GlobSol stopped subdividing before finding a small box containing
the optimizers was due to a comparison of quantities where both quantities
were infinite.  In particular, the problem was in CHECK_SMEAR_DIAMETER in
f90intbi/rigorous_global_search.f90:

The problem is in CHECK_SMEAR_DIAMETER, both sides of the inequality
test here are infinite, so it doesn't ever test "true".
       DO I = 1,NSMALL
          IF ( SMEARS(I) * WID(B%X(I)) > EPS_DOMAIN * SMEAR ) THEN
             TOO_SMALL = .FALSE.
             RETURN
          END IF
This can be fixed by replacing > by >=

I have done this, and have fixed a similar problem in the selection of which
coordinate to bisect.  I have updated GlobSol-win-gnu.correct_IO.exe and
have posted it on the GlobSol download site.

However, now with your "case 2," GlobSol must do many bisections (i.e.
many branchings in the branch and bound process) to complete.  I have run
GlobSol to completion on

Re: GlobSol: Problem: "SEVERITY 2 ERROR: ZERO IN DENOMINATOR IN ORDINARY INTERVAL DIVISION."

Ralph Baker Kearfott — 2010-04-09T01:25:21

(Posted to the GlobSol users' forum list)

Jorge,

The "severity 2 error" is a warning message from the interval arithmetic that
one of the denominators in the interval division contains zero.  I notice
that you have several places where you do a division.  For example,
you appear to divide by (sqrt(er_eff)-sin(X(2))*cos(X(3))) in various
places and by 
((cos(2*k0*sqrt(er_eff)*L)-(p*r))**2 +(sin(2*k0*sqrt(er_eff)*L))**2)
in others.  The error could be due to the range of the expression over
the interval values actually passed to it actually containing zero,
or due to overestimation of the range because of interval
dependencies.  I note the actual value of the denominator is
[-2.225073858507203E-308 0.027133418584390872], so it looks like 
the lower bound would be equal to zero except for "roundout
error".

The actual division occurs in routine IDIV (part of ACM TOMS 
Algorithm 737), and, upon occurrence of the error, the
result is set to [-HUGE,HUGE], the best depiction we have of
the whole real line on mach

GlobSol: Problem: "SEVERITY 2 ERROR: ZERO IN DENOMINATOR IN ORDINARY INTERVAL DIVISION."

Jorge Luis Lagos B. — 2010-04-08T23:32:34

Hello! This is the first time I write to the list; I hope it's the right
place for asking for help.

 

I have been trying to use Globsol to find the global optima of a function
representing the magnitude of an induced voltage in an electric circuit due
to an electromagnetic interfering wave. The wave is characterized by three
angles, which correspond to the independent variables of the problem X(1),
X(2), X(3), and the goal is to find which angles produce the worst induced
voltage. The details of the program are included for an example case at the
end of this message (file example.f90). The problem is that Globsol fails to
find the global maximum, and I get several times the  "SEVERITY 2 ERROR:
ZERO IN DENOMINATOR IN ORDINARY INTERVAL DIVISION." error.

 

For instance, when I run Globsol with the following contents for the
example.DTECH file:

 

1D-5                      

0.0 6.283185307179586   ! range for X(1)

0.0 1.570796326794897   ! range for X(2)

0.0 6.283185307179586   ! range for X(3)

T T

T T

Re: GlobSol: User functions in codelist

Ralph Baker Kearfott — 2009-10-01T14:34:13

(Posted to the GlobSol users' forum list)

Aleksandr,

That capability (implementing "user" functions) described
in the book is still there, and hasn't been removed.  However,
I haven't used it or tested it recently.  Have you  tried it?

By the way, you might contact Mark Stadtherr concerning molecular
energy calculations.  He has had success in that area.

Best regards,

Baker

P.S. Do I have your permission to post this to the entire
      "globsol" list?

Aleksandr Yakimov wrote:

Re: GlobSol: User functions in codelist

R. Baker Kearfott — 2009-09-30T15:40:36

(Posted to the GlobSol users' forum list)

Aleksandr,

First of all, thank you for your interest.

To know whether or not you can implement your goal function
without adding to GlobSol's library of functions, I would first
need to have more information about it.

To implement mid, inf, and sup of a "codelist variable"
(type CDLVAR), you should first understand what GlobSol does:

1. When you compile and run the program containing your goal function,
    a list of addresses and operation codes is generated, for interpretation
    when you actually run the minimizer.

2. When you run the minimizer, various functions interpret the code list
    from step 1 to evaluate, for each operation, interval values, derivatives,
    second derivatives, and corresponding floating point values.

To add an operation to GlobSol, you need to

1. Add an opcode in module CODELIST_CREATION (which, due to historical
    reasons, is in the file overload/overload.f90)

2. Add support for the new opcode in the functions that interpre

GlobSol: User functions in codelist

Aleksandr Yakimov — 2009-09-30T12:44:23

(Posted to the GlobSol users' forum list)

I'm working on custom GlobSol application and my goal function needs
to use 3 specific functions which are not present in the list of
standard CDL-functions.
Whose include:
1) middle point of an interval: F1([a,b])=(a+b)/2
2) lower boundary: F2([a,b])=a
3) upper boundary: F3([a,b])=b

What is the correct way to implement it in the GlobSol environment for
global minimization purposes?

Re: GlobSol: User functions in codelist

Ralph Baker Kearfott — 2009-10-01T14:34:13

(Posted to the GlobSol users' forum list)

Aleksandr,

That capability (implementing "user" functions) described
in the book is still there, and hasn't been removed.  However,
I haven't used it or tested it recently.  Have you  tried it?

By the way, you might contact Mark Stadtherr concerning molecular
energy calculations.  He has had success in that area.

Best regards,

Baker

P.S. Do I have your permission to post this to the entire
      "globsol" list?

Aleksandr Yakimov wrote:

Re: GlobSol: User functions in codelist

R. Baker Kearfott — 2009-09-30T15:40:36

(Posted to the GlobSol users' forum list)

Aleksandr,

First of all, thank you for your interest.

To know whether or not you can implement your goal function
without adding to GlobSol's library of functions, I would first
need to have more information about it.

To implement mid, inf, and sup of a "codelist variable"
(type CDLVAR), you should first understand what GlobSol does:

1. When you compile and run the program containing your goal function,
    a list of addresses and operation codes is generated, for interpretation
    when you actually run the minimizer.

2. When you run the minimizer, various functions interpret the code list
    from step 1 to evaluate, for each operation, interval values, derivatives,
    second derivatives, and corresponding floating point values.

To add an operation to GlobSol, you need to

1. Add an opcode in module CODELIST_CREATION (which, due to historical
    reasons, is in the file overload/overload.f90)

2. Add support for the new opcode in the functions that interpre

GlobSol: User functions in codelist

Aleksandr Yakimov — 2009-09-30T12:44:23

(Posted to the GlobSol users' forum list)

I'm working on custom GlobSol application and my goal function needs
to use 3 specific functions which are not present in the list of
standard CDL-functions.
Whose include:
1) middle point of an interval: F1([a,b])=(a+b)/2
2) lower boundary: F2([a,b])=a
3) upper boundary: F3([a,b])=b

What is the correct way to implement it in the GlobSol environment for
global minimization purposes?