Smelli

New feature: compute the minimum of Δχ²

The new method `chi2_min` of the `GlobalLikelihood` class can be used to compute the minimum of Δχ²=-2 log likelihood for all individual likelihoods.

Updated measurements

All measurement have been updated to those included in flavio v2.2.0 (see https://github.com/flav-io/flavio/releases/tag/v2.2.0)

End of support for Python 3.5
This release requires at least Python 3.6. Python 3.5 has reached [end-of-life](https://devguide.python.org/devcycle/#end-of-life-branches) on 30 September 2020. If you are on an outdated system and cannot update Python, have a look at [conda](https://docs.conda.io/). It allows you to install the newest version of Python in a virtual environment without root privileges. Conda is also the best choice if you are on Windows.

New feature - 2D plot data
The new method `plot_data_2d` of the `GlobalLikelihood` class now allows convenient generation of plot data for *all* individual likelihoods at the same time. This new feature requires `flavio` v2.1.0.

Observables
The two CKM input observables `BR(B+->taunu)` and `DeltaM_d` have been removed from the likelihood in v2.0.0 since they are used to constrain the CKM elements. If the option `fix_ckm` is set to `True`, they are now included again in the likelihood.
Bugs fixed
- An error is now raised if an unknown likelihood is used in one of the arguments `include_likelihoods` or `exclude_likelihoods` on instantiation of `GlobalLikelihood` (thanks to MJKirk).
- The text encoding of the readme file is now specified in the setup script to avoid errors on an OS not using UTF-8 as default.

New observables: Higgs production and decay, beta decays, e⁺e^- scattering, kaon and pion decays
This release adds a large number of new observables that have been recently implemented in [flavio 2.0](https://github.com/flav-io/flavio/releases/tag/v2.0.0) (based on [arXiv:1911.07866](https://arxiv.org/abs/1911.07866) by Adam Falkowski and David Straub and a collaboration of Martín Gonzalez-Alonso, Adam Falkowski, and David Straub):
- 32 new observables in the likelihood `fast_likelihood_quarks.yaml`: kaon decays, pion decays, nuclear and neutron beta decays,
- 48 observables in the new likelihood `likelihood_eeww.yaml`: total and differential cross sections of e⁺e^-→W⁺W^-,
- 33 observables in the new likelihood `likelihood_higgs.yaml`: Higgs signal strengths.

This raises the total number of observables taken into account in the global likelihood to 399.

New feature
The CKM scheme used by a `GlobalLikelihood` instance can now be selected on instantiation using the optional argument `ckm_scheme`. This will facilitate the use of different CKM schemes in the future (currently, only one default CKM scheme is defined).

Bugs fixed
- Removed the branching ratio of Λ_b→Λμμ in the [1.1, 6] GeV² bin.
- If a `GlobalLikelihood` instance is created with the argument `fix_ckm=True`, now the default CKM elements from flavio are used instead of those defined by the default CKM scheme.

Other updates
The required flavio version is now [v2.0](https://github.com/flav-io/flavio/releases/tag/v2.0.0). The flavio version is not pinned anymore to allow updating flavio from pip before updating smelli.

New feature: new physics in CKM

A major new feature is the consistent treatment of dimension-6 contributions affecting CKM elements. Similarly to the mechanism described recently in [this paper](https://arxiv.org/abs/1812.08163), smelli now uses CKM elements in SM predictions of observables that correctly account for the presence of new physics effects in the observables that are used to measure them. This means one of the major limitations mentioned [in the smelli paper](https://arxiv.org/pdf/1810.07698.pdf) is now removed, without any change to the user interface or any scan over nuisance parameters!

The four input observables which are used to determine the CKM elements are: `RKpi(P+->munu)`, `BR(B+->taunu)`, `BR(B->Xcenu)`, and `DeltaM_d/DeltaM_s`. Consequently, `DeltaM_d` and `BR(B+->taunu)` have been removed as direct contributions to the likelihood. New physics effects in them will appear indirectly through modifying CKM elements.

The old behaviour can be reinstated by passing the option `fix_ckm=True` to `GlobalLikelihood`.

New observables

- [5 new](7e885897055ee4d740495fe4eb989176e8052390) Z pole electroweak precision observables
- New bins in R_K measured recently by Belle

Other updates

The required flavio version is now [v1.6](https://github.com/flav-io/flavio/releases/tag/v1.6.0), which contains various updated parameters and measurements (e.g. K_S→μμ and B_s→ll).

New features

- The new concept of "custom likelihoods" allows to define arbitrary subsets of observables for which the likelihood will be computed in addition to the global likelihood, correctly taking into account correlations, and at almost no computational cost thanks to caching. The list of observables has to be passed as argument on instantiation of the global likelihood, e.g.:

python
smelli.GlobalLikelihood(custom_likelihoods={'bsgamma': ['BR(B->Xsgamma)', 'BR(B0->K*gamma)'])

In this example, a key `custom_bsgamma` will be added to the log-likelihood dictionary that correponds to the contribution from these two observables only.
- In addition to the `log_likelihood_dict` method of the `GlobalLikelihoodPoint`, which returns the log-likelihood difference with respect to the SM point, there are now two new methods that allow a quantification of the overall fit quality in the SM and beyond: `chi2_dict` returning the absolute χ² and `pvalue_dict` returning the p value.

New observables in the likelihood

- time-dependent CP violation and rate asymmetries in B_s→φγ
- B_s→τμ
- τ→3e
- μ-e differences in B→K*ll angular observables P'_4,5

Updated measurements

- Updated Belle measurement of R_D^(*)
- Updated Belle measurement of R_K^(*)
- Updated LHCb measurement of R_K
- Some more measurements updated in flavio [1.4](https://github.com/flav-io/flavio/releases/tag/v1.5.0) and [1.5](https://github.com/flav-io/flavio/releases/tag/v1.5.0)

Other changes

- A warning will now be issued if the `flavio` package installed does not correspond to the version required for `smelli`.

Changes to the likelihood

- The flavio dependency has been updated to [v1.3](https://github.com/flav-io/flavio/releases/tag/v1.3), involving some (minor) changes in hadronic parameters.
- The global combination of BR(B_s→μμ) presented in [arXiv:1903.10434](https://arxiv.org/abs/1903.10434) and included in flavio v1.3 is now used, while the individual measurements are explicitly excluded to prevent double counting.

New features

- The theoretical and experimental covariances can now be saved and loaded on instantiation, see 4.
- Observable prediction calculations are now cached on individual `GlobalLikelihoodPoint` instances. This means in particular that `obstable()` can be computed extremely fast if `log_likelihood_global()` has already been called and vice versa.