[Ipopt-tickets] [Ipopt] #60: Support for the Trilinos Amesos sparse linear solver library

[Ipopt]

#60: Support for the Trilinos Amesos sparse linear solver library
--------------------------------+-------------------------------------------
  Reporter:  nowozin            |       Owner:  andreasw
      Type:  enhancement        |      Status:  assigned
  Priority:  normal             |   Component:  Ipopt   
   Version:  3.3 (C++ Version)  |    Severity:  normal  
Resolution:                     |    Keywords:          
--------------------------------+-------------------------------------------
Changes (by andreasw):

  * owner:  ipopt-team => andreasw
  * status:  new => assigned

Comment:

 Hi,

 Thank you for your suggestion to improve Ipopt by using alternative linear
 solvers.

 One very important item to consider is the following:  In order to deal
 with nonconvex problems, Ipopt requires that the linear solver is able to
 report the number of positive and negative eigenvalues of the symmetric
 matrix in the augmented system so that it can modify the Hessian part if
 necessary in order to guarantee that a descent direction is generated (see
 section 3.1 in
 http://www.research.ibm.com/people/a/andreasw/papers/ipopt.pdf).  This
 considerably limits the possible linear solvers that can be used with
 Ipopt.  As far as I know, only symmetric linear solvers using the Bunch-
 Kaufman-type of approach are able to provide this inertia information
 (this is for example the reason we can't use SuperLU in Ipopt).  I had a
 quick glance at the links you included, and it sees to me that the linear
 solvers you discuss are for general, non-symmetric matrices, and would
 therefore not provide the required inertia information.  If that is
 correct, I can't see how Amesos can be used from Ipopt.  This is very
 unfortunate, and has always been a drag, and made it difficult to find a
 good free version of Ipopt.

 Does that make sense?  Did I misunderstand something?

 Thanks

 Andreas

-- 
Ticket URL: <https://projects.coin-or.org/Ipopt/ticket/60#comment:1>
Ipopt <http://projects.coin-or.org/Ipopt>
Interior-point optimizer for nonlinear programs.