User issue: multifit/spinw

[cmarooney-stfc]

David Tam, instrument scientist at ILL has emailed:

Hello,

I am an instrument scientist at ILL. I have previously used Horace quite a lot and it is wonderful. But this is my first time trying to implement multifit and I am running into some problems.

My goal: I want to do (1) a simultaneous fit of several d2d objects, (2) after masking them, using (3) my SpinW model, and (4) with parallelization locally on a Mac Studio desktop. I am using:
Matlab R2025a
Herbert 4.1.0.2a820f49a
Horace 4.1.0.2a820f49a

At the moment I can get multifit to run, it takes some time running but in the end it is not fitting anything and returns the original parameters. I wonder if it is related to using d2d objects, since at the moment I am using

kk.set_fun(@spinwobj.horace_sqw)
Whereas I think it should be
kk.set_fun(@spinwobj.horace)
But this is generating errors.

Another question is about the masking of the d2d objects, I want to control the masking a bit more rigidly than is done automatically, but although this command works fine:

[y,var,msk] = sigvar_get(w11)
This one throws an error:
sigvar_getx(w11)
Incorrect number or types of inputs or outputs for function sigvar_getx.

I am attaching my files for reference. The SpinW one works fine and generates the model. The trouble is in the trymultifit script. Thanks a lot in advance for your feedback.

Thanks,
David Tam

Attached files. .m files converted to .m.txt as Github does not take .m raw
respaghetti.zip
Spinw_h4_lite.m.txt
trymultifit_lite.m.txt

[cmarooney-stfc]

Reply from Alex: (8/8/25 13.24)

Hi, David,

I am investigating. Will try to reply something more certain later today.

What Horace version you are using? Current 3.02 version from release have number of known bugs fixed on Github master, but not included in release. Two of them are related to multifit, though it does not look like your issue.

I will try to reproduce your issue and come back to you with more info.

Regards,
Alex

[cmarooney-stfc]

Reply from David 8/8/25 13.27

Hi Alex,

Thanks a lot for your help. Yes, I was also confused about the version, it is saying version 4.1.0 but on Github the maximum is now only 4.0.2. Was it rolled back at some point?

Thanks,
David

horace_on
!==================================================================!
! ISIS utilities for visualization and analysis !
! of neutron spectroscopy data !
! Herbert 4.1.0.2a820f49a !
!------------------------------------------------------------------!
!==================================================================!
! Horace 4.1.0.2a820f49a !
!------------------------------------------------------------------!
! Visualisation of multi-dimensional neutron spectroscopy data !
! !
! R.A. Ewings, A. Buts, M.D. Le, J van Duijn, !
! I. Bustinduy, and T.G. Perring !
! !
! Nucl. Inst. Meth. A 834, 132-142 (2016) !
! !
! http://dx.doi.org/10.1016/j.nima.2016.07.036 !
!------------------------------------------------------------------!
! If you found a bug or have technical problem with Horace, !
! please contact our team at HoraceHelp@stfc.ac.uk !
! Somebody will be back trying to help you. !
!------------------------------------------------------------------!

[cmarooney-stfc]

Alex replied 10/8/25 19.23

Hi, David,

It took me a bit longer than expected as the issue is in between spinW and multifit and I am not an expert in spinW (will still talk to the experts on Monday)

You seems doing everything right but the SpinW model you are using have problem. In the state as provided, it generates multiple NaN values and weakly sensitive to changes in input parameters. Tobyfit does not know what to do about so many NaN-s, all singular values which define step are nan, so it assumes that the initial point is the best fit.

I am attaching Tobyfit core function modified to ignore Nan-s so it works but the model is weakly sensitive to changes in input parameters, so that
fit_dat.p-[-0.4 -0.4 1.0 -0.2 1 1e-16]

ans = 1.0e-06 *[ 0.0654 -0.1098 -0.3711 -0.1476 0 0]

Put attached function to your Horace distributive instead of standard one. As user model normally should not generate nan-s, I am not distributing it to Horace.

I think Nan-s are the feature of spinW model, not Tobyfit. I will talk with spinW guys on why spinW generates NaNs and how to make it more reliable.

As for sigvar_getx -- its old function, which have not been modified from days it was used in Horace-3. It just does not defined on dnd objects. Do not know if it was lost or never defined. This is probably minor ticket to fix, which I can not do now as working on another project.
Similar result can be obtained by using method:

w11.axes.get_bin_nodes('-bin_center',’-hull’)
If you need to mask data for fitting, just set w11.e(bins to mask) = 0 and Tobyfit will do it for you.

Attachments (.m --> .m.txt, .m --> .m.2.txt for replacements for previous attachments)

multifit_lsqr_ser.m.txt
trymultifit_lite.m.2.txt