Opened 14 years ago

Closed 14 years ago

#131 closed bug (fixed)

possible bug in USANS fitting

Reported by: Lionel Owned by: srkline
Priority: major Milestone: Analysis_4.0_Beta
Component: Analysis Keywords:
Cc: Blocking:
Task: Resolution

Description

OK, problem again...
I'm trying to refit some old data with the new release as I struggle with
my modified version of the function.. take forever to smear...
but getting used to the new interface....

Coming back to a strict comparison, you can see that both fits do not give
the same answer.
With the new version, I cut the data as the masking does not work.

What I am concerned is the USANS.
One fit has the model below the data at high Q and the other is the reverse.
I'm sure that you checked that the new method for smearing USANS works as
well as the old one, but could you give me a possible reason why it is
different??

That is the only difference I see.

THanks

Lionel

* NOTE *
Lionel sent test data files on 4/10/08. I (SRK) couldn't find anything obviously wrong with what Lionel was doing. The masking is an issue. I think I've partially fixed it in a version he doesn't have yet, and the rest of the USANS masking/ partial matrix is still an open ticket.

The mis-fit that he shows may be simple to reproduce on a single data set (use his), possibly with a bad slope.

If there is a well hidden way in the code to switch back to the trapezoid code, that would be helpful for debugging.

Change History (1)

comment:1 Changed 14 years ago by srkline

  • Resolution set to fixed
  • Status changed from new to closed

Andrew researched this and found that Lionel was using an incorrect slope for the matrix calculation.

No bug, but more needs to be done to ensure that the users choose a proper slope, or allow reverting back to a numerical integration (look for something more robust than qtrap() )

Note: See TracTickets for help on using tickets.