From 294d82ca9b287cefa7f782967e42631f6006a995 Mon Sep 17 00:00:00 2001 From: Guido Kanschat Date: Sat, 13 Sep 2014 11:50:29 +0200 Subject: [PATCH] Further polishing of documentation after comments --- include/deal.II/lac/constraint_matrix.h | 42 +++++++++++++------------ 1 file changed, 22 insertions(+), 20 deletions(-) diff --git a/include/deal.II/lac/constraint_matrix.h b/include/deal.II/lac/constraint_matrix.h index 0424bfdcb8..6a38facb66 100644 --- a/include/deal.II/lac/constraint_matrix.h +++ b/include/deal.II/lac/constraint_matrix.h @@ -879,18 +879,20 @@ public: * corresponding to constrained nodes. Thus, if a degree of freedom in @p * local_dof_indices is constrained, we distribute the corresponding entries * in the matrix, but also add the absolute value of the diagonal entry of - * the local matrix to the corresponding entry in the global matrix. Since - * the exact value of the diagonal element is not important (the value of - * the respective degree of freedom will be overwritten by the distribute() - * call later on anyway), this guarantees that the diagonal entry is always + * the local matrix to the corresponding entry in the global + * matrix. Assuming the discretized operator is positive definite, + * this guarantees that the diagonal entry is always * non-zero, positive, and of the same order of magnitude as the other - * entries of the matrix. + * entries of the matrix. On the other hand, when solving a source + * problem $Au=f$ the exact value of the diagonal element is not + * important, since the value of + * the respective degree of freedom will be overwritten by the distribute() + * call later on anyway. * - * @note While not changing the condition number, the procedure - * described above adds an unforeseeable number of artificial - * eigenvalues to the spectrum of the matrix. Therefore, it is - * recommended to use the equivalent function with two local index - * vectors in such a case. + * @note The procedure described above adds an unforeseeable number + * of artificial eigenvalues to the spectrum of the + * matrix. Therefore, it is recommended to use the equivalent + * function with two local index vectors in such a case. * * By using this function to distribute local contributions to the * global object, one saves the call to the condense function after the @@ -911,17 +913,17 @@ public: MatrixType &global_matrix) const; /** - * Does almost the same as the function above but can treat non - * quadratic matrices. The main difference to achieve this is that - * the diagonal entries in constrained rows are left untouched + * Does almost the same as the function above but can treat general + * rectangular matrices. The main difference to achieve this is + * that the diagonal entries in constrained rows are left untouched * instead of being filled with arbitrary values. * * Since the diagonal entries corresponding to eliminated degrees of * freedom are not set, the result may have a zero eigenvalue, if - * applied to a quadratic matrix. This has to be considered when - * solving the resulting problems. For solving a source problem, it - * is possible to set the diagonal entry after building the matrix - * by a piece of code of the form + * applied to a square matrix. This has to be considered when + * solving the resulting problems. For solving a source problem + * $Au=f$, it is possible to set the diagonal entry after building + * the matrix by a piece of code of the form * * @code * for (unsigned int i=0;i void -- 2.39.5