blob: 400e92538bc9466e76252e74304117e5cbdb9518 (
plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
|
Thu Dec 21 21:46:54 2000 Brian Gough <bjg@network-theory.co.uk>
* changed err to abserr, to make it clear that it is an absolute
error in line with other modules.
Thu Nov 2 20:08:14 2000 Brian Gough <bjg@network-theory.co.uk>
* added support for a workspace so that the user does not have to
allocate memory
* made the names of the functions consistent as either levin_u,
for the full u transform with error estimate, or levin_utrunc for
the u transform with only a truncation error estimate.
Mon Apr 24 21:15:27 2000 Brian Gough <bjg@network-theory.co.uk>
* gsl_sum.h: added #include <stdlib.h> for size_t
Mon Nov 1 12:50:17 1999 Brian Gough <bjg@network-theory.co.uk>
* test.c (main): added tests using series for the Euler constant
and eta(1/2)
* resolved problems with spurious failures by replacing the
directly computed truncation error by an estimate which varies
more smoothly. I have used the average of the previous two values,
which seems to give a reliable estimate of the truncation
error. The direct evaluation of the truncation error sometimes
fluctuated wildly, due to cancellation effects.
Thu Oct 28 12:05:47 1999 Brian Gough <bjg@network-theory.co.uk>
* test.c: cleaned up tests, now find that everything works in
double-precision but not extended-precision where there are two
failures depending on the optimization level
* levin_uerr.c (gsl_sum_levin_u_accel_minmax): changed loop
maximum from <=n to <n to avoid access of uninitialized memory,
need further correctness checks on algorithm though.
Thu Oct 7 12:03:36 1999 Brian Gough <bjg@network-theory.co.uk>
* levin_u.c levin_uerr.c: changed DBL_MAX to GSL_DBL_MAX since we
don't rely on DBL_MAX
Sat Feb 6 20:35:26 1999 Brian Gough <bjg@netsci.freeserve.co.uk>
* test.c: adjusted the precision check to allow for "infinite
accuracy" which occurs when two results agree to machine precision
Thu Nov 19 13:10:19 1998 Brian Gough <bjg@vvv.lanl.gov>
* added an n_used parameter to all routines which gives the number
of terms actually used
Tue Nov 17 12:31:03 1998 Brian Gough <bjg@vvv.lanl.gov>
* test.c: added #include <config.h>
* renamed test_sum.c to test.c
* renamed all the functions so that _with_derivs is now the
default and _trunc is the case of no error estimates from the
derivatives
* test_sum.c (main): cleaned up tests
Mon Nov 9 22:05:45 1998 Brian Gough <bjg@vvv.lanl.gov>
* levin_u.c (gsl_sum_levin_u_accel_minmax): got rid of noise
variables since they aren't used in the non-derivative case
1998-11-06 <bjg@ancho.lanl.gov>
* test_sum.c: replace variable N by macro to avoid variable length
array warning
Tue Oct 27 18:06:16 1998 Brian Gough <bjg@vvv.lanl.gov>
* levin_u.c: added in noise but it does not seem to be giving the
right answer for the error estimate. The actual value for the
accelerated sum is correct though. Check toms/602 for the original
algorithm.
|
