Opened 16 years ago

Closed 15 years ago

#4 closed bug (fixed)

Very low Q behavior of some models

Reported by: srkline Owned by: srkline
Priority: minor Milestone: Analysis_3.1_Release
Component: Analysis Keywords:
Cc: Blocking:
Task:

Description

Fri Dec 22 16:46:49 2006 steven.kline@… - Ticket created

Tate Owen (again, one of roger Pynn's students) noticed that some of the polydisperse sphere models gave errtic numerical results at very low Q (< 0.001 A-1) for very small sphere radius (< 50 A).

This perverse limit was not tested, but is important that the USANS q-range gives proper results - since it may be combined with SANS data where 50 A makes sense.

Upon investigation, numerically, it seems that trig functions are correct at these limits. What fails is that the terms that add up to the form factor net to zero within machine precision (10-20). What's left is junk.

For these models (spheres with any type of polydispersity) it is suggested to pick an appropriate QR cutoff (QR<0.1 seems to work) and replace the model function with the (polydisperse) Guinier approximation which is quite valid at these low QR.

Any thoughts of whether this is Kosher?

More testing will be needed to determine which models do actually need this "fix", especially the anisotropic form factors.

Then - fix and propogate to the release distribution.

-Steve

reference RT: http://rt.ncnr.nist.gov/Ticket/Display.html?id=681

Attachments (1)

LowQR_FormFactor_Approx.pdf (96.8 KB) - added by srkline 15 years ago.
Plots of the low QR model with/without the Guinier approximation

Download all attachments as: .zip

Change History (4)

comment:1 Changed 16 years ago by srkline

  • Owner changed from kline to srkline

comment:2 Changed 15 years ago by ajj

  • Milestone set to Analysis_3.1_Release

comment:3 Changed 15 years ago by srkline

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

Fixed the low QR behavior of the Schulz and Rectangular distributed spheres to use the (polydisperse) Guinier approximation at QR<0.1 where machine precision is an issue. The XOP code now duplicates the Igor code.

Only these two models need/can be fixed. see the attached pdf for the results.

Changed 15 years ago by srkline

Plots of the low QR model with/without the Guinier approximation

Note: See TracTickets for help on using tickets.