Flavio

This is a relatively minor release, but major changes are in the pipeline which also require some future changes in dependencies, please read on below.

New processes

- The longitudinal polarization fraction in B→K*νν, `'FL(B0->K*nunu)'`
- The branching ratio of the inclusive semi-tauonic decay `'BR(B->Xctaunu)'` as well as lepton flavour ratios like `'Rtaul(B->Xclnu)'`

Other improvements

- The default values for masses and lifetimes have been updated to PDG 2017 (from PDG 2016).

Bugs fixed

- Several typos in the QCD factorization contributions to B→Vll decays spotted by SatoshiMishima. The numerical impact is small.

Dropping support for Python 3.4

**This will be the last version of flavio working with Python 3.4.** The next version will require at least Python 3.5.
Python 3.5 was released in September 2015; the current version is 3.6. If you have a reasonably recent Linux distribution (e.g. Ubuntu 16.04 LTS or newer), you should be fine (you can check your Python version by running `python3 --version`). If you are on an outdated system and cannot update Python, have a look at [conda](https://conda.io/docs/). 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.

What's new in this release:

Parallel likelihood profiling

This release features major improvements to the likelihood profiler introduced in v0.22. Most notably, it can now make use of parallel processing, by simply passing the argument `threads=N` to the `run` method (see the [API docs](https://flav-io.github.io/apidoc/flavio/statistics/fitters/profiler.m.html) for details).

Other improvements

- The B->D and B->π form factors have been updated to use the [FLAG 2017](http://itpwiki.unibe.ch/flag/images/e/e1/FLAG_webupdate.pdf) lattice average by default
- The `best_fit` method to determine the best-fit point of a `FastFit` is now more robust (i.e., faster and less likely to fail)

Breaking changes

- The BCL parametrization of B->P form factors has been changed such that the kinematic relation f₊(0)=f₀(0) is automatically fulfilled, by fixing the highest z-expansion parameter of f₀

Bug fixes

- The Isgur-Wise relation for the B->P tensor form factor has been corrected
- Avoid Interpolation artifacts in contour plots (37), thanks to Peter Stangl (peterstangl)
- Fix to effective Wilson coefficient C₈ (38), thanks to haolinli1991

This minor release fixes two bugs,

- a bug in the newly released frequentist fit class that erroneously included constraints on fit parameters in the likelihood (rather than ignoring them),
- interchanged B⁰ matrix elements for non-SM operators (thanks to Zeren Simon Wang for reporting & fixing it). The numerical impact of this bug was (accidentally) tiny.

New feature: frequentist fits & profile likelihoods

The major new feature in the release is the addition of **frequentist fits** to complement the Bayesian fits that have existed ever since v0.5. This allows to do analyses with both Bayesian and frequentist statistics in a single framework.

To determine frequentist confidence levels in one or two dimensions, there is a new **profile likelihood fitter**. In two dimensions, the results can be visualized using the existing `contour` plot function, while for one-dimensional p-value distributions, there is a **new plot function** `pvalue_plot` (which was inspired by the pretty plots generated by the [GammaCombo](https://gammacombo.github.io/) package).

On this occasion, the documentation of fits & fitters has been overhauled completely. See in particular the new [introduction to inference in flavio](https://flav-io.github.io/docs/fits.html), the extended [discussion of Bayesian fits](https://flav-io.github.io/docs/bayesian.html), and the new [discussion of frequentist fits](https://flav-io.github.io/docs/frequentist.html).

There is also a [new notebook](https://github.com/flav-io/flavio-examples/blob/master/FrequentistFit_VubVcb_Profiler.ipynb) in the flavio examples repository demonstrating the use of the new fitter and plots.

Other improvements

- A major limitation of the `sm_error_budget` function and its visualization `error_budget_pie` to display error budgets of SM predictions has been removed: previously, the error budget contributions were obtained by individually varying each parameter one by one; this lead to misleading results in case some of the parameters are strongly correlated. This applies for instance to angular observables in B→Vll decays, where uncertainties partially drop out in ratios. The new version now groups together all correlated parameters into common slices of the pie.
- The top quark mass dependence in the matching condition of the Wilson coefficient C₁₀ is now
approximately taken into account (previously, it was taken as a constant).

Bug fixes

- missing conjugation in QCDF corrections (thanks Jacky Kumar jackypheno)
- wrong warning when defining several fits (thanks Peter Stangl peterstangl for reporting & fixing it)

This minor release brings two small improvements and corresponds to the numerics used in the published version of [arXiv:1703.09189](https://arxiv.org/abs/1703.09189).

Improvements

- Correlations between LHCb measurements of B_s→φll angular observables added (thanks to Javier Virto for suggesting and to peterstangl for implementing it).
- In the [last release](https://github.com/flav-io/flavio/releases/tag/v0.21.1), the integration precision for theoretically clean lepton flavour universality tests was increased; this was now implemented also for B_s→φll.

This minor release fixes two problems:

- It increases the numerical integration precision for observables that have a very small SM uncertainty (like the lepton flavour universality ratios in B decays), where this precision is relevant
- It corrects the renormalization scale used for the computation of the SM contribution to meson-antimeson mixing, which previously was not consistent with the QCD correction factors used (thanks to Jacky Kumar for pointing out this problem)