Opened 10 years ago

Closed 15 months ago

#4 closed defect (migrated)

balanced branching strategies in non-convex univariate functions

Reported by: pbelotti Owned by: pbelotti
Priority: major Milestone:
Component: component1 Version:
Keywords: branching, balanced, odd power Cc:

Description

The balanced branching strategy might use wrong values or raise math exceptions when branching on odd power operators. In minMaxDelta.cpp, computing curvDistance() does not take into account the fact that

1) there can be two result of the inverse of a derivative (FpInv?()), which is a square root in the case that the function is x3

2) the resulting approximation of subproblem area for one of the two branches is erroneous.

Until this ticket is closed, it is suggested that the branching strategy to be used is the default, and can be set with

branch_pt_select mid-point

in the couenne.opt option file.

Change History (2)

comment:1 Changed 10 years ago by pbelotti

  • Status changed from new to assigned

comment:2 Changed 15 months ago by stefan

  • Resolution set to migrated
  • Status changed from assigned to closed

This ticket has been migrated to GitHub and will be resolved there: https://github.com/coin-or/Couenne/issues/4

Note: See TracTickets for help on using tickets.