[Halld-physics] [EXTERNAL] Fwd: AmpTools v0.12.2 and SET EPS

[Shepherd, Matthew]

Should have sent to this list also..

From: "Shepherd, Matthew" <mashephe at indiana.edu>
Date: September 20, 2021 at 4:54:28 PM EDT
To: halld-offline at jlab.org
Subject: AmpTools v0.12.2 and SET EPS


Hi all,

At this mornings amplitude analysis meeting I mentioned this problem where manually setting the machine precision (which is done in AmpTools by default) sometimes MINUIT to terminate with a statement like "machine precision limits further improvement."  This is considered an error condition and as such the results of this fit will not be recorded.  Thanks to Malte for bringing this up and digging into the log files find the exact error messages from MINUIT.

This bit of code might just now be a historical relic.  It was necessary when we were doing likelihood computation on older single precision GPUs in order to keep MINUIT from chasing apparent significant variations in the likelihood that were only due to GPU computational precision.

A number of you seem interested in in this change.  I've made a new release:  v0.12.2

https://urldefense.proofpoint.com/v2/url?u=https-3A__github.com_mashephe_AmpTools_releases_tag_v0.12.2&d=DwIGaQ&c=CJqEzB1piLOyyvZjb8YUQw&r=p3b5acpwYNkmZ3alLqoe1DpcHi6vq5JvK_7ohJe1WqI&m=I0duoSKJlH79r0EB_svxjWm9NjJmlTOSwzj2Bp2jlKw&s=JiQuLv3oeV5O6i1LC9--55jBQ-aFATNkNZ6-BDr6v20&e= 

This removes the manual setting of machine precision in the AmpToolsInterface and defaults to letting MINUIT determine machine precision.  (It can still be overridden by the user if desired, but it is probably better not to override unless absolutely necessary.)

If you want to know whether you are successful in making the change you can search your log file for the acknowledgement of MINUIT command:  "SET EPS".  MINUIT will report to screen when precision is set manually.  And I expect all of you can find line in your log that says something like:  SET EPS 1E-13.   This should go away with the new version.

Matt

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mailman.jlab.org/pipermail/halld-physics/attachments/20210920/c7edfc86/attachment.html>