[Ipopt] are block diagonal jacobians better for ma86?

[Greg Horn]

Hello,

I'm solving optimal control problems similar to

  min           f(x1,x2..,xn,p)
x1..xn,p
s.t.
       g1(x1,p) == 0
       g2(x2,p) == 0
       ....
       gn(xn,p) == 0

This gives a constraint jacobian with a dense column for variable p. I was
wondering if it would be more efficient to reformulate this to:

  min                    f(x1,x2..,xn,p)
x1..xn,p,p1..pn
s.t.
       g1(x1,p1) == 0
       g2(x2,p2) == 0
       ....
       gn(xn,pn) == 0
       p - p1 == 0
       p - p2 == 0
       ....
       p - pn == 0

or

  min                    f(x1,x2..,xn,p1)
x1..xn,p1..pn
s.t.
       g1(x1,p1) == 0
       g2(x2,p2) == 0
       ....
       gn(xn,pn) == 0
       p2 - p1 == 0
       p3 - p2 == 0
       ....
       pn - p{n-1} == 0

These would make a more block-diagonal jacobian. Does anyone know if the
sparse linear solvers solve these problems more efficiently? I usually use
HSL ma86.

Thanks!
Greg
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.coin-or.org/pipermail/ipopt/attachments/20150313/539f3c4c/attachment.html>