RSD version 0.1
RSD (written in MATLAB) is an SDP solver that aims to find solutions with
high accuracy. The current version is 0.1. There are two solvers:
-
GRSD for general (full) matrices
-
SRSD for sparse matrices
The corresponding codes are grssdp.m and srssdp.m.
The main inputs are:
-
C - given real symmetric n by n matrix
-
A - given m real symmetric n by n matrices in the form of
MATLAB cell structure
-
b - given real m by 1 vector
Two main parameters are:
-
toler - relative tolerance for stopping criteria of the pd-ip algorithm
-
precond - indicator for type of preconditioning (0, 1(diagonal), or 2(block))
In order to generate random inputs, two MATLAB codes can be used:
testgrssdp.m; and testsrssdp.m. These two codes need [n, m], the size of inputs,
and two main parameters, toler and precond. They generate random instances of
(C, A, and b) and run the corresponding solver, grssdp.m or srssdp.m.
There are some more parameters and inputs, please use 'help grssdp' or
'help srssdp' to learn more about these additional parameters and inputs.
The current version is written completely in MATLAB; therefore, it is suitable
only for problems of small size (n ~ 100, m < n(n+1)/2). There is a
long to-do list regarding the implementation/improvement of the code
and we will try to put them into the next version.
If you have any comments or suggestions, please let us know by sending
emails to Xuan.Doan@wbs.ac.uk or hwolkowicz@uwaterloo.ca
November 2010:
Xuan Vinh Doan,
Serge Kruk,
Henry Wolkowicz