Topic
  • 4 replies
  • Latest Post - ‏2014-03-20T07:11:47Z by ouaigooo
ouaigooo
ouaigooo
3 Posts

Pinned topic roadmap

‏2014-03-13T11:26:00Z |

Hey guys,

Is there any roadmap available about planned improvements of CPLEX ?

For instance, do you guys plan to make CPLEX solve QCP programs without any special requirements on the Q matrix ?

Regards

  • MaryFenelon
    MaryFenelon
    8 Posts

    Re: roadmap

    ‏2014-03-17T02:53:39Z  

    We are not able to discuss plans in this forum but do want to hear what features you would like to see.  In particular, can you give more detail on what sort of nonconvex QCPs you want to solve, such as how many nonconvex quadratic constraints, size of Q's, size of problem, etc.

    --Mary Fenelon, Development Manager

  • ouaigooo
    ouaigooo
    3 Posts

    Re: roadmap

    ‏2014-03-17T07:26:38Z  

    We are not able to discuss plans in this forum but do want to hear what features you would like to see.  In particular, can you give more detail on what sort of nonconvex QCPs you want to solve, such as how many nonconvex quadratic constraints, size of Q's, size of problem, etc.

    --Mary Fenelon, Development Manager

    Hi Mary,

    It would be an optimization of an underlying physical model of about 50 variables (involved in quadratic eqs) over let's say 200 timesteps (i.e. Q would be 1000x1000)

    Q is thus very sparse (variables are mostly "connected" with variables of the same timestep)

    Can you please also  consider the case where the problem also have

    . lots of other linear inequalities and variables (50'000x50'000)

    . some integer variables (100)

    Thanks for your help

  • MaryFenelon
    MaryFenelon
    8 Posts

    Re: roadmap

    ‏2014-03-18T21:00:01Z  
    • ouaigooo
    • ‏2014-03-17T07:26:38Z

    Hi Mary,

    It would be an optimization of an underlying physical model of about 50 variables (involved in quadratic eqs) over let's say 200 timesteps (i.e. Q would be 1000x1000)

    Q is thus very sparse (variables are mostly "connected" with variables of the same timestep)

    Can you please also  consider the case where the problem also have

    . lots of other linear inequalities and variables (50'000x50'000)

    . some integer variables (100)

    Thanks for your help

    For the case with integer variables, are you interested in solutions which are integral but not globally optimal which is what Bonmin does?  Or do you need global optimality, which is what we implemented for our nonconvex MIQP solver?

  • ouaigooo
    ouaigooo
    3 Posts

    Re: roadmap

    ‏2014-03-20T07:11:47Z  

    For the case with integer variables, are you interested in solutions which are integral but not globally optimal which is what Bonmin does?  Or do you need global optimality, which is what we implemented for our nonconvex MIQP solver?

    Global optimality would be a plus (but not a must)