# solver type and penalty factors

Hi,

I’ve a question about the type of solver and the penalty factors of the
objective function.

If I’ve got it right, the objective function is as follows:
Act_max+e1sum(Act)+e2sum(act^2) --> min

In the AnyBody software there are at least two types of solvers
available:

MinMaxOOSolSimplex and MinMaxOOSolQP

In my understanding both solvers could be used to opimize the system in
terms of the MinMax Theory with both penalty factors set to zero.

If the factor e1 is set to a large value (e.g. e1=1000, e2=0) both
solvers should optimize the system in terms of the sum of all muscle
activities.

Only the MinMaxOOSolQP could be used for minimizing the sum of squared
muscle activities.

If my assumtions are right so far, the MinMaxOOSolQP seems to be the
most powerful solver which could be used as a pure MinMax, pure linear
and a pure quadratic solver just depending on the penalty factors.

That means, there should be no difference between the results when
using:

1. MinMaxOOSolSimplex with e1=0 & e2=0 OR MinMaxOOSolQP e1=0 & e2=0
(pure MinMax)

2. MinMaxOOSolSimplex with e1=1000 & e2=0 OR MinMaxOOSolQP e1=1000 &
e2=0 (pure linear)

Is that correct or do I have to take care about the solver type as well
as the penalty factors?

Thanks a lot in advance,

Tim

Hi Tim,

Your assumptions are not exactly right. First with both penalties set
to zero (pure MinMax) it is not certain that the result will be the
same. You can have a difference between using the MinMaxSimplex
solver and the MinMaxOOSolQP. And the second point is that the
MinMaxOOSolQP solver does not take in account the linear penalty e1.
So here there is definitely a difference. MinMaxOOSolQP solver cannot
be used as pure linear.
In conclusion both solvers are needed because they can do different
approaches, and you have to take care about which solver to use
depending on the approach you choose for your problem.

Best regards,
Sylvain, AnyBody Support.

— In anyscript@yahoogroups.com, “timwehner” <timwehner@…> wrote:
>
> Hi,
>
> I’ve a question about the type of solver and the penalty factors of
the
> objective function.
>
> If I’ve got it right, the objective function is as follows:
> Act_max+e1sum(Act)+e2sum(act^2) --> min
>
> In the AnyBody software there are at least two types of solvers
> available:
>
> MinMaxOOSolSimplex and MinMaxOOSolQP
>
> In my understanding both solvers could be used to opimize the
system in
> terms of the MinMax Theory with both penalty factors set to zero.
>
> If the factor e1 is set to a large value (e.g. e1=1000, e2=0) both
> solvers should optimize the system in terms of the sum of all
muscle
> activities.
>
> Only the MinMaxOOSolQP could be used for minimizing the sum of
squared
> muscle activities.
>
> If my assumtions are right so far, the MinMaxOOSolQP seems to be
the
> most powerful solver which could be used as a pure MinMax, pure
linear
> and a pure quadratic solver just depending on the penalty factors.
>
> That means, there should be no difference between the results when
> using:
>
> 1. MinMaxOOSolSimplex with e1=0 & e2=0 OR MinMaxOOSolQP e1=0 & e2=0
> (pure MinMax)
>
> 2. MinMaxOOSolSimplex with e1=1000 & e2=0 OR MinMaxOOSolQP e1=1000
&
> e2=0 (pure linear)
>
> Is that correct or do I have to take care about the solver type as
well
> as the penalty factors?
>
> Thanks a lot in advance,
>
> Tim
>