[Octave-bug-tracker] [bug #65075] Tolerances for complementary error function are higher with UCRT

[Markus Mützel]

URL:
  <https://savannah.gnu.org/bugs/?65075>

                 Summary: Tolerances for complementary error function are
higher with UCRT
                   Group: GNU Octave
               Submitter: mmuetzel
               Submitted: Fri 22 Dec 2023 10:50:55 PM CET
                Category: Libraries
                Severity: 3 - Normal
                Priority: 5 - Normal
              Item Group: Inaccurate Result
                  Status: None
             Assigned to: None
         Originator Name: 
        Originator Email: 
             Open/Closed: Open
                 Release: dev
         Discussion Lock: Any
        Operating System: Microsoft Windows
           Fixed Release: None
         Planned Release: None


    _______________________________________________________

Follow-up Comments:


-------------------------------------------------------
Date: Fri 22 Dec 2023 10:50:55 PM CET By: Markus Mützel <mmuetzel>
When using functions from the Universal C runtime (UCRT) on Windows, a test
for the complementary error function with single precision floating point
values has higher deviations than on other platforms:

***** assert (erfcinv (erfc (single ([-3 -1 -0.4 0.7 1.3 2.8]))),
        single ([-3 -1 -0.4 0.7 1.3 2.8]), -1e-4)
!!!!! test failed
ASSERT errors for:  assert (erfcinv (erfc (single ([-3, -1, -0.4, 0.7, 1.3,
2.8]))),single ([-3, -1, -0.4, 0.7, 1.3, 2.8]),-1e-4)

  Location  |  Observed  |  Expected  |  Reason
    (1)        -2.9994         -3        Rel err 0.00019693 exceeds tol 0.0001
by 0.0001


I'm not sure if there is anything Octave can do about that.

Opening this report so we can tag this as a known issue in the test suite.








    _______________________________________________________

Reply to this item at:

  <https://savannah.gnu.org/bugs/?65075>

_______________________________________________
Message sent via Savannah
https://savannah.gnu.org/