Implementing smelli as an external likelihood

[ElieHammou]

This PR adresses issue #218

The goal is to load from SMEFiT the appropriate datasets and likehood names and the name of the SMEFT/LEFT operators, to pass them one to smelli for a chi2 evaluation which would be taken into account during SMEFiT nested sampling.

[Copilot]

Pull request overview

Introduces an initial (placeholder) implementation for integrating smelli as an external likelihood/chi² contribution within SMEFiT’s existing external_chi2 plugin mechanism (Issue #218).

Changes:

• Adds a new external_chi2_smelli.py module intended to translate SMEFiT coefficients (via RGE.RGEbasis) into a format suitable for smelli likelihood evaluation.
• Sketches a compute_chi2 entrypoint compatible with SMEFiT’s external chi² API (callable accepting a coefficient-value array).

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

You can also share your feedback on Copilot code review. Take the survey.

[ElieHammou]

I am not familiar with the ultranest.py implementation so my changes there should be examined very carefully

[ElieHammou]

With this implementation, running the following runcard:

result_path: "/Users/eliehammou/Software/smefit/smefit_release/results"
data_path: "/Users/eliehammou/Software/smefit/smefit_database/commondata"
theory_path: "/Users/eliehammou/Software/smefit/smefit_database/theory"

use_quad: False

nlive: 5
n_samples: 10
maxiter: 100

rge:
   init_scale: 1000.0
   obs_scale: dynamic # float or "dynamic"
   smeft_accuracy: integrate # options: integrate, leadinglog
   yukawa: top # options: top, full or none
   adm_QCD: False # if true, the EW couplings are set to zero

coefficients:
  OpD: {'min': -10.0, 'max' : 10.0}

external_chi2:
  SmelliChi2:
    # path: /abs/path/to/smefit/smefit_release/src/smefit/external_chi2_smelli.py
    path: /Users/eliehammou/Software/smefit/smefit_release/src/smefit/external_chi2_smelli.py
    include_likelihoods:
      - likelihood_ewpt.yaml
    # optional override:
    # initial_scale: 1000.0

allows to run the following fit:

smefit NS runcard_smelli.yaml
........
[15:19:43] DEBUG    iteration=532, ncalls=785, regioncalls=29320, ndraw=40, logz=-5.40, remainder_fraction=22.1536%, Lmin=1.63, Lmax=1.66                                        integrator.py:2791
[15:19:44] DEBUG    iteration=533, ncalls=786, regioncalls=29360, ndraw=40, logz=-5.40, remainder_fraction=21.8194%, Lmin=1.63, Lmax=1.66                                        integrator.py:2791
[15:19:45] DEBUG    iteration=534, ncalls=787, regioncalls=29400, ndraw=40, logz=-5.40, remainder_fraction=21.4877%, Lmin=1.63, Lmax=1.66                                        integrator.py:2791
[15:19:46] DEBUG    iteration=535, ncalls=788, regioncalls=29440, ndraw=40, logz=-5.39, remainder_fraction=21.1650%, Lmin=1.63, Lmax=1.66                                        integrator.py:2791
[ultranest] Explored until L=2  
           INFO     Explored until L=2                                                                                                                                           integrator.py:2835
[ultranest] Likelihood function evaluations: 788
           INFO     Likelihood function evaluations: 788                                                                                                                         integrator.py:2936
           DEBUG    did a run_iter pass!                                                                                                                                         integrator.py:2478
[ultranest] Reached maximum number of improvement loops.
           INFO     Reached maximum number of improvement loops.                                                                                                                 integrator.py:2852
[ultranest] done iterating.
           INFO     done iterating.                                                                                                                                              integrator.py:2481
[15:19:46] Time : 16.751 minutes                                                                                                                                                   ultranest.py:470
           Number of samples: 10                                                                                                                                                   ultranest.py:471
┏━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━┓
┃ Parameter ┃ Best value ┃ Error ┃
┡━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━┩
│ OpD       │ -0.014     │ 0.009 │
└───────────┴────────────┴───────┘

[ElieHammou]

The tests/build errors comes from the fact that smelli must be installed in the environment for the fit to run, which is not currently taken cared of in the github implementation.