Re: CST branch -- October 22, 2010 - 18:26

Hi,
I am continuing in the discussion started in my post https://mail.gna.org/public/relax-devel/2010-09/msg00020.html
It covers changes of functions func_mf.py, func_local_tm , func_diff, func_all and their
equivalents for a first and second derivatives (class Mf, file maths_fns/mf.py).

I
would like to include into next patch also treatment of the fact, that
it is necessary to sum together contributions of all interactions. It
seems to me that the most suitable way is to do that by the
modification just revised functions (func_mf.py, func_local_tm ...)
I would suggest to initialize ri_prime before loop over interactions and then step by step add the contributions into the ri_prime:

        data = ""                                # the cases when `i` is replaced by `0` were discussed in your last mail
        ri_prime=0

        for j in xrange(self.num_interactions[i]):                   

              ...
              ...

              # Calculate the R1, R2, and sigma_noe values.
              ri_prime = ri_prime + data[j].create_ri_prime(data[j])

         data[0].ri_prime = ri_prime


When the loop over interactions is finished the accumulated relaxation rate is copied into the data storage of the first interaction. Then it is possible to call functions, where total ri_prime is needed:

        # Calculate the NOE values.
        data[0].ri = data[0].ri_prime * 1.0
        for m in xrange(data[0].num_ri):
            if data[0].create_ri[m]:
                data[0].create_ri[m](data[0], m, data[0].remap_table[m], data[0].get_r1, params)

        # Calculate the chi-squared value.
        data[0].chi2 = chi2(data[0].relax_data, data[0].ri, data[0].errors)

Regards,
Pavel

On 19 October 2010 13:49, Edward d'Auvergne <edward@xxxxxxxxxxxxx> wrote:

Hi,

Sorry for the delay, I just came back from a 2 week holiday.  This is
correct, the func_mf, func_local_tm, etc. methods are working on a
single spin.  This is stored in self.data[0].  The other functions
work on multiple spin data located in self.data[0], self.data[1], etc.

Regards,

Edward

On 14 October 2010 19:38, Pavel Kaderavek <pavel.kaderavek@xxxxxxxxx> wrote:
> Hi,
>
> I would like to announce a small clarification of my previous mail about
> changes in functions func_mf, func_local_tm, func_diff, func_all, and their
> derivatives (class Mf, file maths_fns/mf.py). Loop suggested in my last
> post:
>
> https://mail.gna.org/public/relax-devel/2010-09/msg00020.html
>
> is valid just for the functions func_diff, func_all, and their equivalents
> for a first and second derivatives.
> While for the functions  func_mf, func_local_tm, and corresponding
> derivatives the index of self.num_interactions should be set to `0` instead
> of index `i`
>
>         for j in xrange(self.num_interactions[0]):
>
> (The rest of the loop remains the same)
> That comes from the fact that these functions (func_mf, func_local_tm, ...
> ) are called for each spin separately. As it is indicated by the preceding
> statement:
>
>         data = ""> >
> Regards,
> Pavel
>
> On 29 September 2010 10:53, Edward d'Auvergne <edward@xxxxxxxxxxxxx> wrote:
>>
>> Hi,
>>
>> This is the perfect approach.  It will abstract the calculations so
>> that we will not need to touch many of maths_fns modules.  With this
>> code in place, I would aim at then making the test suite pass again by
>> having the correct data structures pass into maths_fns.mf.  The last
>> step would be to input CSA tensors and the multi-dipole interactions
>> via user functions.  If you make a patch for all of the func_*(),
>> dfunc_*(), and d2func_*() methods, I can check and apply it quickly.
>>
>> Regards,
>>
>> Edward
>>
>>
>> On 28 September 2010 12:07, Pavel Kaderavek <pavel.kaderavek@xxxxxxxxx>
>> wrote:
>> > Hi,
>> > we were thinking about next necessary changes in the CST branch.
>> >
>> > Now we need to break through the problem, which implies the fact we
>> > split
>> > the relaxation rate calculation into contributions of individual
>> > interactions. Each of them has its own data class to store its
>> > parameters
>> > (so far called data[i][j], where the [i] was a spin index and [j] was
>> > the
>> > interaction index).
>> >
>> > It seems to us, that it the best way to deal with it is to edit
>> > functions:
>> > func_mf, func_local_tm , func_diff, func_all and their equivalents for a
>> > first and second derivatives (defined in mf.py file).
>> >
>> > Within these functions the calculations of direction cosines, diffusion
>> > tensor weight calculations, components of the spectral densities and so
>> > on
>> > are performed. All these must be calculated for each interaction
>> > separately,
>> > because each interaction has its own data storage (which replaced
>> > previously
>> > used one data class container for the whole IS spin system).
>> >
>> > Instead of:
>> >
>> >
>> >         # Direction cosine calculations.
>> >         if self.diff_data.calc_di:
>> >             self.diff_data.calc_di(data, self.diff_data)
>> >
>> >         # Diffusion tensor weight calculations.
>> >         self.diff_data.calc_ci(data, self.diff_data)
>> >
>> >         # Diffusion tensor correlation times.
>> >         self.diff_data.calc_ti(data, self.diff_data)
>> >
>> >         # Calculate the components of the spectral densities.
>> >         if data.calc_jw_comps:
>> >             data.calc_jw_comps(data, params)
>> >
>> >         # Calculate the R1, R2, and sigma_noe values.
>> >         data.ri_prime = data.create_ri_prime(data)
>> >
>> >
>> > we would suggest to introduce a loop over interacations
>> >         for j in xrange(self.num_interactions[i]):
>> >             # Direction cosine calculations.
>> >             if self.diff_data.calc_di:
>> >                 self.diff_data.calc_di(data[j], self.diff_data)
>> >
>> >             # Diffusion tensor weight calculations.
>> >             self.diff_data.calc_ci(data[j], self.diff_data)
>> >
>> >             # Diffusion tensor correlation times.
>> >             self.diff_data.calc_ti(data[j], self.diff_data)
>> >
>> >             # Calculate the components of the spectral densities.
>> >             if data.calc_jw_comps:
>> >                 data.calc_jw_comps(data[j], params)
>> >
>> >             # Calculate the R1, R2, and sigma_noe components.
>> >             data.ri_prime = data.create_ri_prime(data[j])
>> >
>> >
>> > It must be accompanied in the next step by a change the ri_prime
>> > function so
>> > that it just calculate only a product of specific interaction constant
>> > and
>> > corresponding linear combination of spectral densities. While the final
>> > sumation over all interactions should be done in a separate step.
>> >
>> > Moreover it will be also necessary to distinguish within the function
>> > setup_equation the type of equation used for contribution of individual
>> > interactions according to their type.
>> >
>> > Best
>> > Pavel
>> >
>> >
>> > On 10 September 2010 15:43, Edward d'Auvergne <edward@xxxxxxxxxxxxx>
>> > wrote:
>> >>
>> >> Hi Pavel,
>> >>
>> >> I missed it in the patches, but there were tab characters '\t' causing
>> >> problems.  These are now fixed.  relax requires that a tab is replaced
>> >> by 4 spaces.  I have also added you to the copyright notices
>> >> (http://svn.gna.org/viewcvs/relax?view=rev&revision=11543) of the
>> >> files you have modified.
>> >>
>> >> Regards,
>> >>
>> >> Edward
>> >>
>> >> On 10 September 2010 15:36, Edward d'Auvergne <edward@xxxxxxxxxxxxx>
>> >> wrote:
>> >> > Hi,
>> >> >
>> >> > I've carefully checked the patches and committed them with the
>> >> > messages you wrote.  Sorry again for the delays.  It should be faster
>> >> > now that I am no longer in the tropical wilderness of far north
>> >> > Australia.
>> >> >
>> >> > Regards,
>> >> >
>> >> > Edward
>> >> >
>> >> >
>> >> > On 6 September 2010 13:23, Edward d'Auvergne <edward@xxxxxxxxxxxxx>
>> >> > wrote:
>> >> >> Hi Pavel,
>> >> >>
>> >> >> Sorry for the delayed response.  I was at the ICMRBS conference in
>> >> >> Australia and then travelled through the tropical north end of
>> >> >> Australia afterwards.  I came back yesterday out of the remote
>> >> >> wilderness and can soon start looking at this patches.
>> >> >>
>> >> >> Regards,
>> >> >>
>> >> >> Edward
>> >> >>
>> >> >>
>> >> >> On 31 August 2010 18:00, Pavel Kaderavek <pavel.kaderavek@xxxxxxxxx>
>> >> >> wrote:
>> >> >>> Hi,
>> >> >>>
>> >> >>> some time ago, we submitted two patches regarding CST branch. We
>> >> >>> are
>> >> >>> not
>> >> >>> sure if we should wait for some additional comment from your side,
>> >> >>> or
>> >> >>> we can
>> >> >>> continue with introducing further changes in the code.
>> >> >>> Next step would be a splitting of the relaxation equation so that
>> >> >>> contribution to the relaxation due to the individual types of
>> >> >>> interaction
>> >> >>> (dipole-dipole, CSA) can be calculated separately.
>> >> >>>
>> >> >>>
>> >> >>> Regars,
>> >> >>>
>> >> >>> Pavel, Petr
>> >> >>>
>> >> >>
>> >> >
>> >
>> >
>
>