Re: [ublas] [patch] Fixes for compressed matrices

From: Paul C. Leopardi (leopardi_at_[hidden])
Date: 2005-03-21 16:31:45

On Tue, 22 Mar 2005 03:30 am, Michael Stevens wrote:
> I was thinking of how to describe how the two index arrays are built and
> the invariant that 'complete_index1_data' enforces. That is to say if we
> keep the implementation where index1 may be incomplete we should somewhere
> document it. Do you know if any other compressed matrix libraries yield
> similarly incomplete index1 array?

Hi all,
Sorry to chime in at this late stage, but I have been quite confused by this 
thread of discussion. 

Is it possible/valid to have a compressed matrix where the index1 array is 
complete and yet there are rows (or columns depending on type of compressed 
matrix) which contain no non-zeros? A simple example would be the zero 
matrix. Is is possible/valid to have a compressed matrix with no non-zeros 
and yet a complete index1 array? If not, wouldn't this badly break the 
expected linear algebra semantics of compressed matrices, eg. you would need 
to force some of the non-zeros to have the value zero?
Thanks