Gurobi or Tomlab?

[Terry]

I'm looking into commercial license options for Gurobi and TomLab. It
looks like all tests in Matlab (>>testAll) can be run using Gurobi,
except for ">>testC13Fitting". We'll probably be doing C13 tracing
experiments, so I'm guessing I'll need ">>testC13Fitting" running at
some point. My questions are:

-Anyone has an idea how performance compares when using Gurobi or
TomLab? If there isn't a big difference then Tomlab that can run all
tests is the better choice?

- To get everything running in COBRA with Tomlab, do I need "TOMLAB
Base Module", "TOMLAB /CPLEX", and "TOMLAB /SNOPT" only? If so then
the Tomlab commercial license would cost half that of Gurobi's and
again Tomlab is the better choice.

- I'm probably getting evaluation licenses first, but wanted to get
people's opinions here since I'm new to the COBRA toolbox.

[Ronan M.T. Fleming]

Hi Terry,

to my knowledge, a comparison of

"TOMLAB Base Module", "TOMLAB /CPLEX",
and

Gurobi has not yet been done.

A little bird told me that the developers sold CPLEX to IBM for a packet, then setup Gurobi. Could this story hold the clue?

There will be lots of algorithmic developments for 13C in the next year. SNOPT is available directly if you want it
http://www.sbsi-sol-optimize.com/asp/sol_product_snopt.htm
Most 13C algorithms try to solve an optimization problem that is not globally convergent so to make it really work you'd need to get your fingers right on the buttons of the solver. My experience with the tomlab interface for matlab is that it provides a simplified interface to CPLEX, it may be the same for SNOPT, though Jan Schellenberger is the one who knows more about SNOPT/Tomlab

Ronan