Test VMC

Run the test suite after building on a new compiler or system and after changing the simulation, wavefunction, energy, optimizer, or output code.

TODO — UWHPC test-job policy

Missing from the repository: whether tests may run on a login node, the required partition and resource request if they must be submitted, and the UWHPC module commands. Link to the UWHPC-wide testing/job policy when it exists.

Run all tests

From the repository root:

./scripts/test.sh

The script creates build-tests, configures a Debug FP64 CPU build, builds vmc_tests, and runs CTest with --output-on-failure. It always configures VMC_ENABLE_CUDA=OFF; this script does not test the CUDA backend.

On a clean build, CMake fetches Catch2 3.8.1 from GitHub, so the first configuration needs network access. The final portion of a successful run was:

85/94 Test #85: Finite-size convergence: N=7 and N=19 agree at r_s=5 .............................................   Passed    3.27 sec
      Start 86: Blocking analysis produces finite standard error
86/94 Test #86: Blocking analysis produces finite standard error .................................................   Passed    1.31 sec
      Start 87: Warmup produces reasonable acceptance rate
87/94 Test #87: Warmup produces reasonable acceptance rate .......................................................   Passed    0.95 sec
      Start 88: Energy per particle at r_s=1 is positive and bounded
88/94 Test #88: Energy per particle at r_s=1 is positive and bounded .............................................   Passed    1.72 sec
      Start 89: Standard error is small relative to energy scale
89/94 Test #89: Standard error is small relative to energy scale .................................................   Passed    1.75 sec
      Start 90: Same seed produces identical energy
90/94 Test #90: Same seed produces identical energy ..............................................................   Passed    0.73 sec
      Start 91: Independent seeds produce consistent energies
91/94 Test #91: Independent seeds produce consistent energies ....................................................   Passed    2.89 sec
      Start 92: WaveFunction evaluateDerivatives clears buffers and delegates to Jastrow
92/94 Test #92: WaveFunction evaluateDerivatives clears buffers and delegates to Jastrow .........................   Passed    0.01 sec
      Start 93: WaveFunction evaluate_log_psi returns finite for N=1
93/94 Test #93: WaveFunction evaluate_log_psi returns finite for N=1 ..............................................   Passed    0.01 sec
      Start 94: WaveFunction default Jastrow parameter matches fully polarized cusp choice
94/94 Test #94: WaveFunction default Jastrow parameter matches fully polarized cusp choice .......................   Passed    0.01 sec

100% tests passed, 0 tests failed out of 94

Total Test time (real) =  52.82 sec

Run a group of tests

CTest discovers individual Catch2 test cases. The test executable also accepts Catch2 tag filters. For example:

./build-tests/vmc_tests '[config]'
./build-tests/vmc_tests '[energy]'
./build-tests/vmc_tests '[validation]'

To list the available tags:

./build-tests/vmc_tests --list-tags

Observed output:

All available tags:
   3  [aligned_soa]
   3  [blocking]
  11  [config]
   6  [energy]
  10  [interacting]
   6  [jastrow]
   3  [matrix]
   3  [optimizer]
   2  [output_writer]
   2  [particles]
   8  [published]
   9  [rigorous]
   7  [simulation]
  19  [slater]
   7  [validation]
   4  [wavefunction]
16 tags

Test another precision mode

The test script accepts the same precision-related flags as the build script:

./scripts/test.sh --fp32
./scripts/test.sh --vmc-fast-math

You may also provide a build type as a positional argument, such as ./scripts/test.sh Release.

TODO — CUDA validation

The supplied test script forces a CPU build. Missing: an approved CUDA test procedure, the GPU resource request, the CUDA test results, and any CPU-versus-GPU numerical acceptance criteria.

TODO — authoritative scientific validation

The repository contains tests named against PZ81 and several expected-energy checks, but the docs still need the full paper citation, derivation of the reference values, unit conventions, approved tolerances, and a maintained table of validated production results.

Source of truth

The test workflow is defined by scripts/test.sh and the testing section of CMakeLists.txt. Test cases are under tests/.