xref: /petsc/src/dm/dt/tests/ex14.c (revision d6685f554fbda8d96c6a5d73ab0e7a4e21a05c51)

const char help[] = "Tests properties of probability distributions";

#include <petscdt.h>

/* Checks that
   - the PDF integrates to 1
   - the incomplete integral of the PDF is the CDF at many points
*/
static PetscErrorCode VerifyDistribution(const char name[], PetscBool pos, PetscProbFunc pdf, PetscProbFunc cdf)
{
  const PetscInt digits = 14;
  PetscReal      lower = pos ? 0. : -10., upper = 10.;
  PetscReal      integral, integral2;
  PetscInt       i;
  PetscErrorCode ierr;

  PetscFunctionBeginUser;
  ierr = PetscDTTanhSinhIntegrate((void (*)(const PetscReal[], void *, PetscReal *)) pdf, lower, upper, digits, NULL, &integral);CHKERRQ(ierr);
  PetscCheck(PetscAbsReal(integral - 1.0) < 100*PETSC_MACHINE_EPSILON, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONG, "PDF %s must integrate to 1, not %g", name, integral);
  for (i = 0; i <= 10; ++i) {
    const PetscReal x = i;

    ierr = PetscDTTanhSinhIntegrate((void (*)(const PetscReal[], void *, PetscReal *)) pdf, lower, x, digits, NULL, &integral);CHKERRQ(ierr);
    ierr = cdf(&x, NULL, &integral2);CHKERRQ(ierr);
    PetscCheck(PetscAbsReal(integral - integral2) < PETSC_SQRT_MACHINE_EPSILON, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONG, "Integral of PDF %s %g != %g CDF at x = %g", name, integral, integral2, x);
  }
  PetscFunctionReturn(0);
}

static PetscErrorCode TestDistributions()
{
  PetscProbFunc  pdf[]  = {PetscPDFMaxwellBoltzmann1D, PetscPDFMaxwellBoltzmann2D, PetscPDFMaxwellBoltzmann3D, PetscPDFGaussian1D};
  PetscProbFunc  cdf[]  = {PetscCDFMaxwellBoltzmann1D, PetscCDFMaxwellBoltzmann2D, PetscCDFMaxwellBoltzmann3D, PetscCDFGaussian1D};
  PetscBool      pos[]  = {PETSC_TRUE,                 PETSC_TRUE,                 PETSC_TRUE,                 PETSC_FALSE};
  const char    *name[] = {"Maxwell-Boltzmann 1D",     "Maxwell-Boltzmann 2D",     "Maxwell-Boltzmann 3D",     "Gaussian"};
  PetscErrorCode ierr;

  PetscFunctionBeginUser;
  for (PetscInt i = 0; i < (PetscInt) (sizeof(pdf)/sizeof(PetscProbFunc)); ++i) {
    ierr = VerifyDistribution(name[i], pos[i], pdf[i], cdf[i]);CHKERRQ(ierr);
  }
  PetscFunctionReturn(0);
}

static PetscErrorCode TestSampling()
{
  PetscProbFunc  cdf[2]     = {PetscCDFMaxwellBoltzmann1D, PetscCDFMaxwellBoltzmann2D};
  PetscProbFunc  sampler[2] = {PetscPDFSampleGaussian1D,   PetscPDFSampleGaussian2D};
  PetscInt       dim[2]     = {1, 2};
  PetscRandom    rnd;
  Vec            v;
  PetscScalar   *a;
  PetscReal      alpha, confidenceLevel = 0.05;
  PetscInt       n = 1000, s, i, d;
  PetscErrorCode ierr;

  PetscFunctionBeginUser;
  ierr = PetscRandomCreate(PETSC_COMM_SELF, &rnd);CHKERRQ(ierr);
  ierr = PetscRandomSetInterval(rnd, 0, 1.);CHKERRQ(ierr);
  ierr = PetscRandomSetFromOptions(rnd);CHKERRQ(ierr);

  for (s = 0; s < 2; ++s) {
    ierr = VecCreateSeq(PETSC_COMM_SELF, n*dim[s], &v);CHKERRQ(ierr);
    ierr = VecSetBlockSize(v, dim[s]);CHKERRQ(ierr);
    ierr = VecGetArray(v, &a);CHKERRQ(ierr);
    for (i = 0; i < n; ++i) {
      PetscReal r[3], o[3];

      for (d = 0; d < dim[s]; ++d) {ierr = PetscRandomGetValueReal(rnd, &r[d]);CHKERRQ(ierr);}
      ierr = sampler[s](r, NULL, o);CHKERRQ(ierr);
      for (d = 0; d < dim[s]; ++d) a[i*dim[s]+d] = o[d];
    }
    ierr = VecRestoreArray(v, &a);CHKERRQ(ierr);
    ierr = PetscProbComputeKSStatistic(v, cdf[s], &alpha);CHKERRQ(ierr);
    PetscCheck(alpha < confidenceLevel, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONG, "KS finds sampling does not match the distribution at confidence level %.2g", confidenceLevel);
    ierr = VecDestroy(&v);CHKERRQ(ierr);
  }
  ierr = PetscRandomDestroy(&rnd);CHKERRQ(ierr);
  PetscFunctionReturn(0);
}

int main(int argc, char **argv)
{
  PetscErrorCode ierr;

  ierr = PetscInitialize(&argc, &argv, NULL, help); if (ierr) return ierr;
  ierr = TestDistributions();CHKERRQ(ierr);
  ierr = TestSampling();CHKERRQ(ierr);
  ierr = PetscFinalize();
  return ierr;
}

/*TEST

  test:
    suffix: 0
    requires: ks
    args:

TEST*/