void py_coulomb_set_hubbard_u(PyObject *self, void *p, double *U, int *error)
{
particles_t *py_p;
PyObject *py_U, *r;
npy_intp dims[1];
INIT_ERROR(error);
#ifdef DEBUG
printf("[py_coulomb_set_Hubbard_U] %s %p %p\n",
PyString_AsString(PyObject_Str(self)), U, error);
#endif
f_particles_get_tag(p, (void**) &py_p);
assert(py_p->f90obj == p);
dims[0] = f_particles_get_nel(py_p->f90obj);
py_U = PyArray_SimpleNewFromData(1, dims, NPY_DOUBLE, U);
r = PyObject_CallMethod(self, "set_Hubbard_U", "(OO)", py_p, py_U);
Py_DECREF(py_U);
PASS_PYTHON_ERROR(error, r);
Py_DECREF(r);
}
void iterative_matrix_inverse(double *matptr, double *invmatptr, int n,
_Bool prev, double epsilon, double *work1,
double *work2, int *error,
cublasHandle_t cublas_handle, int *nit_out)
{
INIT_ERROR(error);
mat<double> matr(n, matptr, cublas_handle);
mat<double> invmat(n, invmatptr, cublas_handle);
/* Will allocate and release upon destruction if work1, work2 == NULL */
mat<double> help1(n, work1, cublas_handle);
mat<double> help2(n, work2, cublas_handle);
/*
* - Initialize inverse matrix if previous not used
* The starting invmat has to be small enough so that the iteration
* won't start running to infinity
*/
#if 0
mat<double> dummy(n);
dummy = matr;
printf("dummy.data() = %p\n", dummy.data());
printf("matr.data() = %p\n", matr.data());
printf("dummy.on_host() = %i\n", dummy.on_host(error));
PASS_ERROR(error);
printf("matr.on_host() = %i\n", matr.on_host(error));
PASS_ERROR(error);
printf("sum = %f %f\n", dummy.sum(), matr.sum());
printf("max = %f %f\n", dummy.max(), matr.max());
printf("min = %f %f\n", dummy.min(), matr.min());
printf("amax = %f %f\n", dummy.amax(), matr.amax());
printf("amin = %f %f\n", dummy.amin(), matr.amin());
#endif
if (!prev) {
double smin, smax;
ev_bounds(n, matptr, &smin, &smax, error);
PASS_ERROR(error);
mat_mul_sca(1.0/(n*MAX(fabs(smin), fabs(smax))), matr, invmat, error);
PASS_ERROR(error);
}
/*
* Find inverse via S^-1 = 2 S^1 - S^-1 S S^-1
*/
double sigma = epsilon + 1.0;
int i = 0;
while (sigma > epsilon) {
/*
* help1 = matr.invmat
*/
gemm(OP_N, OP_N, 1.0, matr, invmat, 0.0, help1, error);
PASS_ERROR(error);
help2 = invmat;
/*
* invmat = -help2.help1 + 2*invmat
*/
gemm(OP_N, OP_N, -1.0, help2, help1, 2.0, invmat, error);
PASS_ERROR(error);
mat_mul_sca(1.0, help2, -1.0, invmat, help1, error);
PASS_ERROR(error);
sigma = help1.amax(error);
PASS_ERROR(error);
i = i+1;
if (i % 100 == 0) {
prscrlog("iterative_matrix_inverse: No convergence after %i iterations.",
i);
}
}
if (nit_out) {
*nit_out = i;
}
}
