Convergence rate dependent on constants in objective function.

[1ozturkbe]

As of commit 1bc2440, we can 'kind of' get the correct connectivity. However, the reltol has an effect on the convergence rate, as well as the magnitude of the penalty on the slack. I need to implement a way to normalize slack penalty.
Ideas:

• normalize by maximum flow cost among all edges times number of edges
• total cost is the sum(edgeCost*flow) * prod(slack).
  To be brainstormed...

[1ozturkbe]

ITEM: total cost is the sum(edgeCost * flow) * prod(slack)
COMMENTS: Doesn't work. Slacks blow up so that edgeCost*flow goes to zero in the case when the self-flow has no cost.

[1ozturkbe]

ITEM: normalize by maximum flow cost among all edges times number of edges
COMMENTS: Doesn't work. Same as above (should have known...)