[Csdp] Sizes of matrices

[Nestor Aguilera]

I am using CSDP in a model that has many small matrices constrained  
to be PSD, typically 2x2 up to 5x5.

For instance, a 3x3 matrix restriction models an inequality of the  
form x1^2 + x2^2 <= t (x1, x2, t unknowns). I could reformulate this  
as x1^2 <= t1, x2^2 <= t2, replacing the 3x3 matrix by two 2x2  
matrices and adding one unknown.

What is best (in time, and space?) for CSDP: having fewer but larger  
matrices, or more but smaller matrices?

Thanks in advanced.

                                                  Nestor Aguilera