38 #ifndef GMM_PRECOND_ILDLTT_H 
   39 #define GMM_PRECOND_ILDLTT_H 
   50   template <
typename Matrix>
 
   53     typedef typename linalg_traits<Matrix>::value_type value_type;
 
   54     typedef typename number_traits<value_type>::magnitude_type magnitude_type;
 
   58     row_matrix<svector> U;
 
   59     std::vector<magnitude_type> indiag;
 
   65     template<
typename M> 
void do_ildltt(
const M&, row_major);
 
   66     void do_ildltt(
const Matrix&, col_major);
 
   69     void build_with(
const Matrix& A, 
int k_ = -1, 
double eps_ = 
double(-1)) {
 
   71       if (eps_ >= 
double(0)) eps = eps_;
 
   73       indiag.resize(std::min(mat_nrows(A), mat_ncols(A)));
 
   74       do_ildltt(A, 
typename principal_orientation_type<
typename 
   75                 linalg_traits<Matrix>::sub_orientation>::potype());
 
   78       : U(mat_nrows(A),mat_ncols(A)), K(k_), eps(eps_) { build_with(A); }
 
   81     size_type memsize()
 const { 
 
   82       return sizeof(*this) + 
nnz(U)*
sizeof(value_type) + indiag.size() * 
sizeof(magnitude_type);
 
   86   template<
typename Matrix> 
template<
typename M> 
 
   89     typedef typename number_traits<T>::magnitude_type R;
 
   91     size_type n = mat_nrows(A);
 
   95     R prec = default_tol(R()), max_pivot = gmm::abs(A(0,0)) * prec;
 
   98     for (size_type i = 0; i < n; ++i) {
 
  102       for (size_type krow = 0, k; krow < w.nb_stored(); ++krow) {
 
  103         typename svector::iterator wk = w.begin() + krow;
 
  104         if ((k = wk->c) >= i) 
break;
 
  105         if (gmm::is_complex(wk->e)) {
 
  106           tmp = gmm::conj(U(k, i))/indiag[k]; 
 
  107           gmm::add(scaled(mat_row(U, k), -tmp), w);
 
  111           if (gmm::abs(tmp) < eps * norm_row) { w.sup(k); --krow; } 
 
  112           else { wk->e += tmp; 
gmm::add(scaled(mat_row(U, k), -tmp), w); }
 
  117       if (gmm::abs(gmm::real(tmp)) <= max_pivot)
 
  118         { GMM_WARNING2(
"pivot " << i << 
" is too small"); tmp = T(1); }
 
  120       max_pivot = std::max(max_pivot, std::min(gmm::abs(tmp) * prec, R(1)));
 
  121       indiag[i] = R(1) / gmm::real(tmp);
 
  123       gmm::scale(w, T(indiag[i]));
 
  124       std::sort(w.begin(), w.end(), elt_rsvector_value_less_<T>());
 
  125       typename svector::const_iterator wit = w.begin(), wite = w.end();
 
  126       for (
size_type nnu = 0; wit != wite; ++wit)  
 
  127         if (wit->c > i) { 
if (nnu < K) { U(i, wit->c) = wit->e; ++nnu; } }
 
  131   template<
typename Matrix> 
 
  132   void ildltt_precond<Matrix>::do_ildltt(
const Matrix& A, col_major)
 
  133   { do_ildltt(gmm::conjugated(A), row_major()); }
 
  135   template <
typename Matrix, 
typename V1, 
typename V2> 
inline 
  136   void mult(
const ildltt_precond<Matrix>& P, 
const V1 &v1, V2 &v2) {
 
  138     gmm::lower_tri_solve(gmm::conjugated(P.U), v2, 
true);
 
  139     for (
size_type i = 0; i < P.indiag.size(); ++i) v2[i] *= P.indiag[i];
 
  140     gmm::upper_tri_solve(P.U, v2, 
true);
 
  143   template <
typename Matrix, 
typename V1, 
typename V2> 
inline 
  144   void transposed_mult(
const ildltt_precond<Matrix>& P,
const V1 &v1, V2 &v2)
 
  147   template <
typename Matrix, 
typename V1, 
typename V2> 
inline 
  148   void left_mult(
const ildltt_precond<Matrix>& P, 
const V1 &v1, V2 &v2) {
 
  150     gmm::lower_tri_solve(gmm::conjugated(P.U), v2, 
true);
 
  151     for (
size_type i = 0; i < P.indiag.size(); ++i) v2[i] *= P.indiag[i];
 
  154   template <
typename Matrix, 
typename V1, 
typename V2> 
inline 
  155   void right_mult(
const ildltt_precond<Matrix>& P, 
const V1 &v1, V2 &v2)
 
  156   { 
copy(v1, v2); gmm::upper_tri_solve(P.U, v2, 
true); }
 
  158   template <
typename Matrix, 
typename V1, 
typename V2> 
inline 
  159   void transposed_left_mult(
const ildltt_precond<Matrix>& P, 
const V1 &v1,
 
  162     gmm::upper_tri_solve(P.U, v2, 
true);
 
  163     for (
size_type i = 0; i < P.indiag.size(); ++i) v2[i] *= P.indiag[i];
 
  166   template <
typename Matrix, 
typename V1, 
typename V2> 
inline 
  167   void transposed_right_mult(
const ildltt_precond<Matrix>& P, 
const V1 &v1,
 
  169   { 
copy(v1, v2); gmm::lower_tri_solve(gmm::conjugated(P.U), v2, 
true); }
 
incomplete LDL^t (cholesky) preconditioner with fill-in and threshold.
 
sparse vector built upon std::vector.
 
size_type nnz(const L &l)
count the number of non-zero entries of a vector or matrix.
 
void copy(const L1 &l1, L2 &l2)
*/
 
number_traits< typename linalg_traits< V >::value_type >::magnitude_type vect_norm2(const V &v)
Euclidean norm of a vector.
 
void clear(L &l)
clear (fill with zeros) a vector or matrix.
 
void resize(V &v, size_type n)
*/
 
void clean(L &l, double threshold)
Clean a vector or matrix (replace near-zero entries with zeroes).
 
void mult(const L1 &l1, const L2 &l2, L3 &l3)
*/
 
void add(const L1 &l1, L2 &l2)
*/
 
ILUT: Incomplete LU with threshold and K fill-in Preconditioner.
 
size_t size_type
used as the common size type in the library