[Csdp] A mystery
Brian Borchers
borchers at nmt.edu
Fri Aug 7 14:02:25 EDT 2009
I agree with Imre's response- this problem has multiple optimal solutions
(all with optimal value 2/3), and CSDP happens to be returning one of those
solutions, although it isn't the solution that Ron would prefer.
Let me just add the following.
Ron wrote:
>while for the phony solution we have
>Tr[F[0].Y]=1/2
...
>and for the true solution
>Tr[F[0].Y]=2/3
The F[0] from the data file is
F[0]= (pardon the fixed font here!)
-1 1 0
1 0 0
0 0 0
0 0 0 0
0 0 0 0
0 0 0 0
0 0 0 0
Ron's Y (The one that Ron thinks is optimal) is
Y=
1/2 7/12 sqrt(23)/12
7/12 1 0
sqrt(23)/12 0 1
7/2 0 0 0
0 0 0 0
0 0 0 0
0 0 0 7/2
CSDP's Y is
Y=
1/2 7/12 0
7/12 1 0
0 0 1
7/2 0 0 0
0 0 0 0
0 0 0 0
0 0 0 7/2
In both cases, when I compute it by hand, I get that Tr[F[0].Y]=2/3, not 1/2!
When I run my copy of CSDP on this problem, I get the solution listed above
as "CSDP's Y", and the optimal value reported by CSDP is 2/3.
It isn't clear whether Ron's CSDPGMP came up with the incorrect value of
Tr[F[0].Y]=1/2 or whether this was a mistake that was made outside of CSDP.
In the former case, that would imply a bug in CSDPGMP.
More information about the Csdp
mailing list