Simulated annealing range problem

On line 199 in siman/siman.c is:

   } else if (gsl_rng_uniform(r) < exp (-(new_E - E)/(params.k * T)) ) {

which is the line that decides to take a step in the direction of higher
objective function energy sometimes.  The problem is that my objective
function sometimes returns very large new_E for a small change in parameters,
leading to exp(-750) for example, and errno is set to ERANGE.  Should this
condition be checked for?  

gsl_sf_hyperg_1F1 bug

This function erroneously negates its return values for some specific 
arguments.

Specifically when
a is a negative integer
b = -.5

The return value is incorrect.  Presumably this is due to special case 
handling in the function as for negative integral a the 1F1 
hypergeometric function reduces to a Laguerre polynomial.

This error can also be seen by plotting the function value versus a, 
and noticing a discontinuity at negative integer values.

Re: Simulated annealing range problem

Daniel Webb writes:
 > On line 199 in siman/siman.c is:
 > 
 >    } else if (gsl_rng_uniform(r) < exp (-(new_E - E)/(params.k * T)) ) {
 > 
 > which is the line that decides to take a step in the direction of higher
 > objective function energy sometimes.  The problem is that my objective
 > function sometimes returns very large new_E for a small change in parameters,
 > leading to exp(-750) for example, and errno is set to ERANGE.  Should this
 > condition be checked for?  

Thanks for the bug report.  I will add a check to avoid that.

--

-- 
Brian Gough

Network Theory Ltd,
Publishing Free Software Manuals --- http://www.network-theory.co.uk/

Re: Missing gsl_sf_sin_err and gsl_sf_cos_err

Alex Tartakovsky writes:
 > These two functions, gsl_sf_sin_err and gsl_sf_cos_err, mentioned in the manual (p. 66), are missing.
 >  

Thanks for the bug report. I'll fix the documentation (they shouldn't be defined as they
don't make sense for error propagating functions).

--

-- 
Brian Gough

Network Theory Ltd,
Publishing Free Software Manuals --- http://www.network-theory.co.uk/

incorrect documentation for gsl_matrix_minmax_index()

Hi,

the declaration string in the documentation is identical to that of
gsl_matrix_min_index() which means that two parameters are missing.

Patch attached (taken against CVS as of 30 Nov, but not tested to
compile!)

It might also be a good idea to change imax and jmax in the docu of
gsl_matrix_min_index() to imin and jmin, though the documentation is
consistent.

Regards
	Thomas

_______________________________________________
Bug-gsl mailing list
Bug-gsl <at> gnu.org
http://lists.gnu.org/mailman/listinfo/bug-gsl

Re: incorrect documentation for gsl_matrix_minmax_index()

Thomas Weber writes:
 > the declaration string in the documentation is identical to that of
 > gsl_matrix_min_index() which means that two parameters are missing.
 > 
 > Patch attached (taken against CVS as of 30 Nov, but not tested to
 > compile!)
 > 
 > It might also be a good idea to change imax and jmax in the docu of
 > gsl_matrix_min_index() to imin and jmin, though the documentation is
 > consistent.

Thanks for the bug report, I'll fix that for the next release.

--

-- 
Brian Gough