Fit test of DiffSphereTest fails when tried on MantidPlot

[Jose Borreguero]

Resolution of this ticket is dependent on resolution of ticket #4394

Resolution of this ticket is dependent on resolution of ticket #5603

The fit test in DiffSphereTest test the convolution of a Gaussian resolution function with a DiffSphere function. The target parameters are:
Gaussian c=0 h=3.0 sigma=0.62
DiffSphere I=2.9 Q=0.70 R=2.3 D=0.45

The initial parameters are:
Gaussian c=0 h=3.0 sigma=0.62
DiffSphere I=1.3 Q=0.70 R=1.4 D=1.5
Fitting in MantidPlot produces the following error:
Error in execution of algorithm Fit:
GSLVector index is out of range

Starting with different initial parameters, very close to the target ones:
Gaussian c=0 h=3.0 sigma=0.62
DiffSphere I=2.7 Q=0.70 R=2.5 D=0.46
results in another error:
Error in execution of algorithm Fit:
Error in function boost::math::sph_bessel_j<e>(e,e): Got x = -29910.0084641663015645, but function requires x > 0.

[Jose Borreguero]

DiffSphere now correctly inherits from ImmutableCompositeFunction and outputs a string representation akin to a non-composite function

[Roman Tolchenov]

Fitted in MantidPlot to the data extracted from DiffSphereTest. If the initial values are close to the given it works fine. Although the following function gives the error message of the second type.

(composite=Convolution;name=Gaussian,Height=3,PeakCentre=0,Sigma=0.62;
(name=DiffSphere,NumDeriv=true,Q=0.69999999999999996,Intensity=16.0614,
Radius=5.97427,Diffusion=1908.43))

[Jose Borreguero]

Refs #5597 Substitute pointer by string representation

modified: DiffSphere.h modified: DiffSphere.cpp modified: DiffSphereTest.h

Changeset: 7d1348d31656f6ee4aba72e9344a2728832a2770

[Jose Borreguero]

Refs #5597 Substitute pointer by string representation

modified: DiffSphere.h modified: DiffSphere.cpp modified: DiffSphereTest.h

Changeset: 7d1348d31656f6ee4aba72e9344a2728832a2770

[Nick Draper]

Moved to the Backlog after the code freeze for R2.6

[Nick Draper]

moved to Release 3.0 as these all seem to be active

[Jose Borreguero]

The unit tests in DiffSphereTest are working now, after closing of ticket #4394.

Negative arguments being passed on to the spherical Bessel functions are now prevented by assigning an "infinite" Chi-square to the fit. Negative arguments are unphysical.

TESTER:

There's nothing to test, but you can review the code and note in particular that numeric_limits<double>::infinity() is assigned to the output of DiffSphere.cpp:InelasticDiffSphere::function1D() when either of the fit parameters become negative.

[Russell Taylor]

Looked over the mentioned code, though it's been in master for ages.

[Stuart Campbell]

This ticket has been transferred to github issue ​6443