I’m using ROOT 4.00/03.
I’m inverting a TMatrix with UL method; sometimes it happens that inversion fails, then I get an invalid (nRow = -1) matrix.
Now, I want to recover that matrix to reinitialize it with other data (life still goes on).
Which (fast) method can I use? The dimension of that matrix must be the same as the old one, so no memory reallocation is needed; anyway, ResizeTo() fails (there is an assertion on IsValid()) and I can’t figure which other method I’m supposed to use to resurrect it.
I have installed your request to be able to reset an invalidate matrix .
It should be in CVS soon .
For this, a new data member in the vector/matrix classes has been
introduced : fStatus . Now we will not set the fNrows = -1 in order to
invalidate the matrix (which made us lose the number of rows) .
Use “MakeValid()” to reset the status bit .
On another note . If you are inverting matrices of the same
dimensions, invoking in place inversion through “Invert()” might not
be the best way to go . The method “Invert()” makes each time an
instance of the class TDecompLU for a matrix of a certain dimension .
Better (faster) is to do a
TDecompLU lu(n); outside your loop . The decomposition class is now
setup for matrices of n x n . Inside your loop you now do a
lu.SetMatrix(a); thereby assigning your matrix a . If matrix a is
of dimension n x n, no memery allocation has to be performed by lu .
Now you can call either lu.Invert(a), which replaces a by its inversion
or TMatrixD a_inv = lu.Invert();
I’m still a little confused about your suggestion; reading TMatrixD::Invert() method it seems to me that TMatrixD does not use any real TDecompLU object, using TDecompLU static members instead and passing them itself as the data on which to act.
Oops, start forgetting my own code . yes, you are right, I changed the
Invert code so that it only uses some static functions . It is still true
for the Deteminant() functionality