If you think this is important, you should try this data in other
dispersion software for a comparison. But you'll find that these
cases are close to impossible to optimise to the correct solution.
The dispersion curves are essentially flat - therefore there are many,
many ways that the dispersion equations can compensate to produce
semi-flat lines. This is what you are seeing in the mapping of the
optimisation space. Note that even if you could find the true
solution, a statistical comparison to the 'No Rex' model (either AIC
or F-test) will say that dispersion is statistically insignificant.
Therefore spending a huge amount of time fine tuning the optimisation
for such cases will not be worth the effort. You should consider
setting pA to 0.95 rather than 0.993 to obtain reasonable amounts of
dispersion.
Note that relax uses the relax_disp.insignificance user function for
the auto-analysis. This defaults to a R2eff/R1rho' difference of 2.0
rad.s^-1 (or 1.0 in the GUI and 0.0 in the auto-analysis). Therefore
if the maximuma dispersion curve difference is less than this value,
the spin will be deselected and not optimised (the R2eff model being
an exception). In the case of this bug report, this user function
will probably deselect the spin.
Regards,
Edward
On 10 May 2014 18:29, Troels E. Linnet <NO-REPLY.INVALID-ADDRESS@xxxxxxx>
wrote:
Follow-up Comment #8, bug #22024 (project relax):
And for this setup, where you have only 1 field, and dw=1.2
Why is it finding a solution under the chi2 surface sheet???
Params for dx map is
['dw', 'pA', 'kex']
Point creating dx map is
[1.2, 0.99299999999999999, 2200.0]
Minimises point in dx map is
[5.5418405143389711, 0.50000101059958624, 0.96403927774135734]
########################
Generated data with MODEL:NS CPMG 2-site expanded
Analysing with MODEL:CR72.
########################
For spin: ':1@N'
CR72 Ala :1@N r2 599.9 GRID=25.031 MIN=25.081 CLUST=25.081 SET=25.000
RELC=0.003
CR72 Ala :1@N pA GRID=0.500 MIN=0.500 CLUST=0.500 SET=0.993 RELC=0.986
###################################
WARNING: pA Have relative change above 0.05, and is 0.9860.
###################################
CR72 Ala :1@N dw GRID=6.500 MIN=5.542 CLUST=5.542 SET=1.200 RELC=0.783
###################################
WARNING: dw Have relative change above 0.05, and is 0.7835.
###################################
CR72 Ala :1@N kex GRID=1.000 MIN=0.964 CLUST=0.964 SET=2200.000
RELC=2281.065
###################################
WARNING: kex Have relative change above 0.05, and is 2281.0647.
###################################
(file #20704, file #20705, file #20706)
_______________________________________________________
Additional Item Attachment:
File name: CR72_one_field_dw_1.2_map_1_N.png Size:191 KB
File name: CR72_fail_one_field.py Size:24 KB
File name: CR72_one_field_dw_1.2_disp_1_N.png Size:6 KB
_______________________________________________________
Reply to this item at:
<http://gna.org/bugs/?22024>
_______________________________________________
Message sent via/by Gna!
http://gna.org/
Re: [bug #22024] Minimisation space for CR72 is catastrophic. The chi2 surface over dw and pA is bounded.