Changeset 1478

Timestamp:

Sep 6, 2008, 8:55:46 PM (15 years ago)

Author:

Peter

Message:

reverting [1474] as original docs were correct. Clarified that SVD only works when rows>=columns for all algorithms and added an assert checking this.

Location:

trunk/yat/utility

Files:

• SVD.cc (modified) (2 diffs)
• SVD.h (modified) (3 diffs)

trunk/yat/utility/SVD.cc

@@ -31 +31 @@
 #include "VectorBase.h"
 
+#include <cassert>
 #include <sstream>
 
@@ -51 +52 @@
   void SVD::decompose(SVDalgorithm algo)
   {
+    assert(U_.rows()>=U_.columns());
     int status=0;
     switch (algo) {

trunk/yat/utility/SVD.h

@@ -47 +47 @@
      decomposition, SVD.
 
-     A = U S V' = (MxM)(MxN)(NxN) = (MxN)\n   
+     A = U S V' = (MxN)(NxN)(NxN) = (MxN)\n   
 
      A = Matrix to be decomposed, size MxN\n
-     U = Orthogonal matrix, size MxM\n
-     S = Diagonal matrix of singular values, size MxN\n
+     U = Orthogonal matrix, size MxN\n
+     S = Diagonal matrix of singular values, size NxN\n
      V = Orthogonal matrix, size NxN\n
   */
@@ -60 +60 @@
     /**
        A number of SVD algorithms are implemented in GSL. They have
-       their strengths and weaknesses, check the GSL documentation.
+       their strengths and weaknesses.
     
-       There are restrictions on the matrix dimensions depending on
-       which SVD algorithm is used. From the GSL's SVD source code one
-       finds that the Golub-Reinsch algorithm implementation will not
-       work on matrices with fewer rows than columns, the same is also
-       true for the modified Golub-Reinsch algorithm.
-    
-       \see GSL's SVD documentation.
+       \see GSL's SVD documentation. 
     */
     enum SVDalgorithm {
@@ -80 +74 @@
        input. The input matrix is copied for further use in the
        object.
+
+       \note Number of rows must be equal or larger than number of columns.
     */
     SVD(const utility::Matrix& Ain);