test_vsa_no_mass fails

[Flamefire]

We are installing Tamkin 1.2.6 using Python 3.11.3, molmod 1.4.8, nose3 1.3.8 using GCC 12.3.0

When running the tests export MATPLOTLIBRC=$PWD; echo "backend: agg" > $MATPLOTLIBRC/matplotlibrc; export OMP_NUM_THREADS=1; nosetests -v tamkin we see the test_vsa_no_mass failing on 2 different architectures (power9le & x86 haswell) with similar differences:

FAIL: test_vsa_no_mass (tamkin.test.test_nma.NMATestCase.test_vsa_no_mass)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/dev/shm/s3248973-EasyBuild/TAMkin/1.2.6/foss-2023a/tamkin-1.2.6/tamkin/test/test_nma.py", line 532, in test_vsa_no_mass
    self.check_freqs(expected_freqs, nma, 0, check_zeros=True)
  File "/dev/shm/s3248973-EasyBuild/TAMkin/1.2.6/foss-2023a/tamkin-1.2.6/tamkin/test/test_nma.py", line 77, in check_freqs
    self.assertAlmostEqual(
AssertionError: 0.13774798 != 0.6608735502232888 within 0 places (0.5231255702232888 difference) : Frequency 2 does not match, expected - computed = 0.1 - 0.7 = -0.523126

======================================================================
FAIL: test_vsa_no_mass (tamkin.test.test_nma.NMATestCase.test_vsa_no_mass)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/tmp/boegelbot/TAMkin/1.2.6/foss-2023a/tamkin-1.2.6/tamkin/test/test_nma.py", line 532, in test_vsa_no_mass
    self.check_freqs(expected_freqs, nma, 0, check_zeros=True)
  File "/tmp/boegelbot/TAMkin/1.2.6/foss-2023a/tamkin-1.2.6/tamkin/test/test_nma.py", line 77, in check_freqs
    self.assertAlmostEqual(
AssertionError: 0.13774798 != 0.6460767049351539 within 0 places (0.5083287249351539 difference) : Frequency 2 does not match, expected - computed = 0.1 - 0.6 = -0.508329

Is this something to worry about or is this somehow "expected"?

[Flamefire]

It doesn't seem to be related to the compiler or Python versions as it happens for other versions too. But it only happens on haswell machines as it seems. Any ideas?

[Flamefire]

I was also able to replicate this locally (on my laptop) on an Intel(R) Core(TM) i7-7600U

• The expected frequencies are: -0.4205594, 0.03940166, 0.13774798
• On a machine where it passes: -0.4383613, -0.13210359, 0.43330176
• On a machine where it fails: -0.3107732, 0.176978305, 0.64607670

So to me this looks like the expected frequency is off and it just happens to match for a difference of 0.3 but not anymore for 0.5