Re: Constraints -- March 25, 2008 - 19:25

Hi,

Thanks for the professional comment :)  The program has been carefully
designed to have as little limitations as possible - so that the only
limit is imagination and CPU time.  If you do encounter limits (not
the constraint type though), these can be removed from relax straight
away.  I would like to have these constraints, as well as the diagonal
scaling values, more accessible to the user, but unfortunately I have
never managed to design a flexible system for this.  The constraint
values for model-free analysis are documented in my model-free model
elimination paper
(http://www.nmr-relax.com/refs.html#dAuvergneGooley06), but these may
one day change.  For example an upper limit on tm of 200 ns may need
to be changed for studying extremely large systems a la Kay.

Regards,

Edward


On Tue, Mar 25, 2008 at 6:31 PM, Sébastien Morin
<sebastien.morin.1@xxxxxxxxx> wrote:
>  Hi Ed,
>
>  Thanks for these informations... I was afraid that you could have limited
> relax to small proteins in the first drafts of the program with forgetting
> to change these limits afterwards... thus creating artefacts when used with
> bigger systems.
>
>  However, as I see, the limits are all general and the size of the protein
> (up to the huge correlation time of 200 ns !) does not influence the
> results...
>
>  As usual, relax is extremely professional...
>  Thanks Ed !
>  Cheers !
>
>
>  Séb
>
>
>
>
>
>
>  Edward d'Auvergne wrote:
>  Hi,
>
> Unless turned off, the constraints are used all of the time. In the
> minimisation this turns on the Method of Multipliers (also known as
> the Augmented Lagrangian) algorithm. In the grid search, any points
> outside of the limits are dropped. Unfortunately these constraints
> are hard coded as I couldn't, at the time, come up with a flexible way
> for the user to modify the default values. There are a number of
> methods for applying constraints, but currently only linear
> constraints are supported. These are in the form:
>
> A.x >= b,
>
> where A is an matrix of coefficients, x is an array of parameter
> values, and b is a vector of scalars. These translate into
> constraints such as:
>
> S2 >= 0,
> -S2 >= -1,
> etc.
>
> The full list of constraints can be seen in the documentation string
> for the linear_constraints() method in the specific_fns/model_free.py
> file (relax-1.2). The diffusion tensor parameter constraints aren't
> yet documented, but can be seen in the comments in the code of
> linear_constraints(). I hope this info helps.
>
> Regards,
>
> Edward
>
>
> On Fri, Feb 22, 2008 at 5:38 PM, Sébastien Morin
> <sebastien.morin.1@xxxxxxxxx> wrote:
>
>
>  Hi,
>
>  I have a question about constraints in relax...
>
>  I would like to know what are the built-in constraints in relax,
>  especially for what concerns 'tau' (tau_m, tau_e, tau_s, tau_f) as well
>  as 'r' and 'csa' (for models m1x and m2x).
>
>  What I'd like to know is if those parameters are contrained during grid
>  search, optimization, elimination and selection, except for
>  'tau_(e,f,s)' values which should not exceed '1.5 x tau_m'...
>
>  I understand that constraint are used by default within the method of
>  multipliers algorithm, but don't really get what are those constraint
>  and on which variables they act...
>
>  Thanks for help !
>
>  Cheers,
>
>
>  Séb :)
>
>
>
>  _______________________________________________
>  relax (http://nmr-relax.com)
>
>  This is the relax-users mailing list
>  relax-users@xxxxxxx
>
>  To unsubscribe from this list, get a password
>  reminder, or change your subscription options,
>  visit the list information page at
>  https://mail.gna.org/listinfo/relax-users