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
|
From 8306771281f9b2883c5200ee241b0c0767bfe1b4 Mon Sep 17 00:00:00 2001
From: Michael Orlitzky <michael@orlitzky.com>
Date: Sun, 30 Jun 2024 19:45:45 -0400
Subject: [PATCH 1/1] eigen/test.c: loosen tolerance in eigenvalue norm tests
One test for the (unit) norm of a random hermitian matrix's
eigenvalues is victim to tolerance issues, e.g.
FAIL: herm random, normalized(1), unsorted (0.999999999999999112
observed vs 1 expected) [117761]
The tolerance used is N*GSL_DBL_EPSILON, where N is the size of the
matrix. On my machine, the actual error is four times GSL_DBL_EPSILON,
which means that the tolernace is close but not quite right. Despite
having done a PhD in numerical analysis, I am choosing to fix this
with a shotgun:
* By a factor of 10
* Then also by a factor of N
That's still extremely close, and hopefully will sweep the problem
under the rug whether it's off by a constant or off by a function
of N.
Bug: https://bugs.gentoo.org/664936
Bug: https://git.adelielinux.org/adelie/packages/-/issues/504
Bug: https://savannah.gnu.org/bugs/?56843
---
eigen/test.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/eigen/test.c b/eigen/test.c
index b0bdcc7..2681721 100644
--- a/eigen/test.c
+++ b/eigen/test.c
@@ -506,7 +506,7 @@ test_eigen_herm_results (const gsl_matrix_complex * A,
{
gsl_vector_complex_const_view vi = gsl_matrix_complex_const_column(evec, i);
double nrm_v = gsl_blas_dznrm2(&vi.vector);
- gsl_test_rel (nrm_v, 1.0, N * GSL_DBL_EPSILON, "%s, normalized(%d), %s",
+ gsl_test_rel (nrm_v, 1.0, N*N * 10.0 * GSL_DBL_EPSILON, "%s, normalized(%d), %s",
desc, i, desc2);
}
--
2.44.2
|
