Re: tru64 failures

[Brian Gough <bjg at network-theory dot co dot uk>]

 > FAIL: qagiu(myfn1) smooth rlist (4.3793924773509602e-20 observed vs 4.3793924773509536e-20 expected)

This looks ok, just a slightly different rounding error, so I'll
increase the tolerance for the test to accept it.

The other errors are strange ---- the simulated annealing does not
move from its initial configuration.  The patch below will print out
debugging information from monte/test.c which might indicate what is
going wrong.

--- siman/test.c.~1~	Thu May  4 12:15:16 2000
+++ siman/test.c	Mon May 15 14:13:22 2000
@@ -77,6 +77,12 @@
   printf("%12g", x.D1);
 }
 
+void debug_pos_1D(void * x)
+{
+  printf("%12g", ((Element *)x)->D1);
+}
+
+
 /* a metric for the 2D space */
 double distance_1D(Element x, Element y)
 {
@@ -217,6 +223,8 @@
 /*   new_x = new_x*2*step_size; */
 /*   new_x = new_x - step_size + old_x; */
 
+  printf("test step from old_x = %g to new_x = %g\n", old_x, new_x);
+
   memcpy(xp, &new_x, sizeof(new_x));
 }
 
@@ -237,7 +245,7 @@
      There is a local minimum at x = 0.60146196, (f = -0.84893) */
 
   x = -10.0 ;
-  gsl_siman_solve(r, &x, E1, S1, M1, 0, sizeof(double), params);
+  gsl_siman_solve(r, &x, E1, S1, M1, &debug_pos_1D, sizeof(double), params);
   gsl_test_rel(x, x_min, 1e-3, "f(x)= exp(-(x-1)^2) sin(8x), x0=-10") ;
 
   x = +10.0 ;