2008-07-03  Brian Gough  <bjg@hp2.network-theory.co.uk>

	* Makefile.am (INCLUDES): use top_srcdir instead of top_builddir

2007-05-30  Brian Gough  <bjg@network-theory.co.uk>

	* siman.c (boltzmann): moved safe_exp into boltzmann

	(copy_state): put all copying into one function

2006-03-08  Brian Gough  <bjg@network-theory.co.uk>

	* test.c (square): removed inline since it causes problems with

	some compilers

2005-11-14  Brian Gough  <bjg@network-theory.co.uk>

	* siman.c (safe_exp): added a safe_exp function to avoid underflow

	for large uphill steps

2003-03-31  Brian Gough  <bjg@network-theory.co.uk>

	* siman.c (gsl_siman_solve): avoid reevaluation for best_E

	(gsl_siman_solve): loop over param.iters_fixed_T not

	params.n_tries

	(gsl_siman_solve): initialise energy at start

Sat Aug  3 20:32:38 2002  Brian Gough  <bjg@network-theory.co.uk>

	* siman.c (gsl_siman_solve): fix acceptance criterion to match

	documentation (Peter S. Christopher)

Thu Jun 13 20:57:00 2002  Brian Gough  <bjg@network-theory.co.uk>

	* siman.c (gsl_siman_solve): keep track of the best result

2002-02-07  Mark Galassi  <rosalia@galassi.org>

	* siman.c (gsl_siman_solve): bug fix in the destructor for x and

	new_x which was being called on &x and &new_x instead of x and

	new_x; thanks to Karsten Howes <karsten@videotron.ca>

Thu Jul 12 21:50:07 2001  Brian Gough  <bjg@network-theory.co.uk>

	* gsl_siman.h: changed renamed gsl_Efunc_t to gsl_siman_Efunc_t,

	in accordance with namespace conventions

2000-12-15  Mark Galassi  <rosalia@galassi.org>

	* siman.c (gsl_siman_solve): reversed a small change I had made

	earlier and went back to taking Boltzmann-conditional steps when

	the new energy is equal to the previous one.  This allows you to

	move around if you are stuck on a plateau.

	* gsl_siman.h, siman.c, siman_test.c, siman_tsp.c, test.c: changed

	the siman_solve API to allow for more general search spaces.  The

	problem was that we assumed that points in the search space were

	data structures that were allocated in continguous memory, so they

	could not be linked structures.  I replaced the malloc(), memcpy()

	and free() calls with copy_constructor(), copyfunc() and

	copy_destructor() functions.  The user passes these functions,

	which means that siman_solve() now takes three more arguments of

	type gsl_siman_copy_t, gsl_siman_copy_construct_t and

	gsl_siman_destroy_t.  If these arguments are NULL (and all three

	of them have to be NULL together), the traditional memcpy()

	approach is used.

1999-02-14  Mark Galassi  <rosalia@cygnus.com>

	* minor fixes.

Tue Nov 17 17:22:14 1998  Brian Gough  <bjg@vvv.lanl.gov>

	* added #include <config.h> to all top-level source files

Sun Nov  8 20:40:28 1998  Brian Gough  <bjg@vvv.lanl.gov>

	* siman_tsp.c: clean up for make strict

1998-11-06    <bjg@ancho.lanl.gov>

	* test.c: added prototype for memcpy using #include <string.h>

	* siman_test.c: added prototype for memcpy using #include <string.h>

Wed Oct 28 15:06:58 1998  Brian Gough  <bjg@vvv.lanl.gov>

	* siman.c: added #include <string.h> for memcpy

Thu Aug 20 12:22:28 1998  Brian Gough  <bjg@vvv.lanl.gov>

	* siman.c: use (char *) judiciously to avoid warnings about void

 	pointer arithmetic (see randist/shuffle.c for similar examples)

	* siman_test.c: perform several tests, using the exact answer as

 	the comparison value, rather than checking for stationarity.

Sun Jun 28 14:11:04 1998  Brian Gough  <bjg@vvv.lanl.gov>

	* Converted to work with rng-style random number generators

	* gsl_siman.h: gsl_siman_step_t type functions now take a gsl_rng

 	random number generator as their first argument

	* siman.c (gsl_siman_solve): Now takes a gsl_rng random number

 	generator as the first argument

Fri Jun 19 11:17:24 1998  Brian Gough  <bjg@vvv.lanl.gov>

	* siman.c (gsl_siman_solve_many): changed the variable 'throw' to

 	'u' (for uniform-random-number) so that we can compile with c++

 	where throw is a reserved word.

Sat May 23 13:59:55 1998  Brian Gough  <bjg@vvv.lanl.gov>

	* siman.c: made the solving functions deterministic by removing

 	the random seed, gsl_ran_seed(time(0L)). When the function is

 	non-deterministic it is hard to debug and test (about 1 time in 20

 	the test would fail due to the randomness). We can let the user do

 	the seeding if they need that.

1998-02-09  Mark Galassi  <rosalia@nis.lanl.gov>

	* siman_test_driver.sh (LAST_ENERGY): fixed a typo; the tests now

	report well when they converge.

1998-01-30  Mark Galassi  <rosalia@cygnus.com>

	* siman_test_driver.sh, Makefile.am (TESTS): added a test driver

	so that now "make check" does something interesting.