Create 1D fitting function for discrete rotational diffusion in a circle

[Jose Borreguero]

A 1D fitting function to help quasi-elastic analysis of discrete rotational diffusion on N sites located in a circle. Typical of elongated molecules, like alkanes, but also applicable to methyl rotations.

A more detailed description in http://www.mantidproject.org/DiffRotDiscreteCircle

============================================================== NOTES FOR THE TESTER:

Download attached files to /tmp. In Mantidplot.Script_Window, load test.py and execute. A fit of the data should yield the followings parameters:

Intensity = 0.745, Radius = 0.876(Angstroms), Decay = 763.6(pico-seconds)

You can plot workspace data_Workspace to visually inspect the fit

You can also use the MantidPlot's fit wizard to perform the fit. Here's the starting fit-string guess that you load to the wizard via Setup->Manage_Setup->Load_From_String:

(composite=Convolution,FixResolution=true,NumDeriv=true;name=TabulatedFunction,FileName="",Workspace=resolution,Scaling=1;(name=DiffRotDiscreteCircle,N=3,NumDeriv=true,Q=1.9,Intensity=1,Radius=1,Decay=10));name=LinearBackground,A0=0.001,A1=0.0001 ==============================================================

[Jose Borreguero]

Refs #8508 Little more than the header

Changeset: 4f75e86c6035f8c4829f25d4ff9bd74d42472f4b

[Jose Borreguero]

Refs #8508 Moved to a C++ algorithm

modified: Framework/CurveFitting/CMakeLists.txt new file: Framework/CurveFitting/inc/MantidCurveFitting/DiffRotDiscreteCircle.h new file: Framework/CurveFitting/src/DiffRotDiscreteCircle.cpp new file: Framework/CurveFitting/test/DiffRotDiscreteCircleTest.h deleted: Framework/PythonInterface/plugins/functions/DiffRotDiscreteCircle.py

Changeset: 38ec9a3534c7ff73853a98315c987f61f2dce220

[Jose Borreguero]

Refs #8508 Cosmetic changes in the code

Changeset: 8af64157f97c1ad42cf47632e3da3294bd93dcef

[Jose Borreguero]

Refs #8508 Inserted WIKI page in source docstring

Changeset: a5889007a57b9c8b7d8b1186192d67ba0ca0ff33

[Jose Borreguero]

Refs #8508 Include limits header for numeric_limits

Changeset: 8cf4875a34a0961a37de0337619408b4c0e0dc01

[Jose Borreguero]

Refs #8508 Avoid declaring statically sized arrays

Windows compiler complains when declaring a statically sized array while passing in a size variable which is evaluated at run-time

Changeset: 47014c8e8db1fc8fcb7c3e107bd83e81c0e60a11

[Jose Borreguero]

Refs #8508 declaring a statically sized array v2

I had another one of those :(

Changeset: 8ca3319cb0c42ca0b88142ef85c974c6ea064fcd

[Jose Borreguero]

Notes to the tester:

Download attached files to /tmp. In Mantidplot.Script_Window, load test.py and execute. A fit of the data should yield the followings parameters:

Intensity = 47.014, Radius = 1.567, Decay = 7.567

You can plot workspace data_Workspace to visually inspect the fit

You can also use the Fit GUI in MantidPlot to perform the fit. Here's the starting fit-string guess that you can load from Setup->Manage_Setup->Load_From_String:

(composite=Convolution,FixResolution=true,NumDeriv=true;name=Gaussian,Height=1,PeakCentre=0,Sigma=20,ties=(Height=1,PeakCentre=0,Sigma=20);(name=DiffRotDiscreteCircle,N=3,NumDeriv=true,Q=0.5,Intensity=10,Radius=1,Decay=20))

[Jose Borreguero]

Refs #8508 Changed the WIKI docstring

Changeset: bd821fd40f198786562fcdad3c29ab628e614055

[Jose Borreguero]

Refs #8508 Substitute array with vector

Changeset: d7c67d8c1efaf2614a8ad91da4e76c789f114cd6

[Jose Borreguero]

Refs #8508 Upper limit in one for loop was wrong

Changeset: 595f5903d86ae7901c365f3a3bdc3cb6864b3067

[Jose Borreguero]

Refs #8508 Test simple lorentzian case N=3

Changeset: e4727397bfeefffcc90f38fc0e606e0c14daec3f

[Jose Borreguero]

Refs #8508 change asserts for Chi2

Changeset: 457861313d7a8e18ad6e85d3aaeaa2fc6335a75f

[Jose Borreguero]

Refs #8508 More detailed description of the fitting function

Changeset: da798fd6df66f754b6f718d4055c5e04083bc1ba

[Jose Borreguero]

Refs #8508 Even more detailed description

modified: DiffRotDiscreteCircle.cpp

Changeset: 3e89ba209aac57ef1ed09bdaecc3e12798f92777

[Roman Tolchenov]

I decided to reopen this ticket with these comments:

• The errors in the test data file make fit very sensitive to the [StartX,EndX] interval. The result parameters vary in a wide range.
• If errors are set to 1 the test data can be fitted with a simple gaussian to a good accuracy. Is it possible to construct data that cannot be fitted by other functions?
• It looks like a couple of commits are missing from develop, although they contain changes to a comment block only and shouldn't break any builds or tests.

[Jose Borreguero]

data to run the test

[Jose Borreguero]

python script to be run in MantidPlot

[Jose Borreguero]

Replaced the example with data from a real experiment. In the previous example, the resolution function (the Gaussian) was broader than the quasielastic signal. This was the reason why one could fit the data with a simple Gaussian when one allowed for bigger errors.

[Roman Tolchenov]

The function fits the data very well.

[Roman Tolchenov]

Merge remote-tracking branch 'origin/feature/8508_discrete_rotational_diffusion'

Full changeset: ​83940618dcd27e8897972eb7bde4f01b26770ece

[Stuart Campbell]

This ticket has been transferred to github issue ​9352