On Fri, Feb 22, 2008 at 6:03 PM, Sébastien Morin
<sebastien.morin.1@xxxxxxxxx> wrote:
Hi,
I am trying to use the full_analysis.py script for optimizing model m11
to m15 (m1 to m5 with CSA) for a system for which I have datasets at
three magnetic fields (500, 600, 800 MHz).
The optimization is quite long, but this could be due to the algorithm
used. I'll thus try to optimize models with another algorithm and see
how it goes...
Be careful with these models. Although they exist, they are untested.
By untested, think synthetic noisy data covering much of the
model-free space to see if you can get the correct CSA values back
(for example the grids I used in my two publications listed at
http://www.nmr-relax.com/refs.html). The same goes for the bond
length optimisation, and the simultaneous CSA and bond length
optimisation.
I have another problem which is that, for model tm12, I get a lot of
warnings of the form 'Mu_too_small.' This warning is related to the
Lagrange Multiplier.
I get the same exact problem when trying to work with models m21 to m25
and m31 to m35...
Why does this happen only for model tm12 (model tm15 optimization has
not finished yet) ?
Is this issue a problem ?
How could I solve it ?
This warning is not much of an issue due to the iterative nature of
the Method of Multipliers algorithm. It occurs for difficult
optimisation problems, but the final result should be very close to
the true minimum. I've seen the warning many times before, but it has
never resulted in incorrect parameter values as far as I have seen.
The warning is nevertheless preserved so that the user can be wary of
that model, especially in its interpretation.
Regards,
Edward
Re: Lagrange multiplier warning