Dr. Lorenz Gärtner

Constraints on invisible B→K X decays

2026-02-01T00:00:00+01:00

Applying the public \(B^+\to K^+\nu\bar\nu\) likelihood to constrain a generic invisible particle \(X\) in \(B^+\to K^+X\), including a heavy dark-sector mediator interpretation.

Published as: Constraints on invisible \(B^+\to K^+X\) decays from the Belle II \(B^+\to K^+\nu\bar\nu\) measurement, Phys. Rev. D 114, 032003 (2026).

Reinterpreting the measurement as a search for a dark-sector mediator \(X\): posterior on its mass \(m_X\) and coupling \(\mu_X\), for two assumed decay widths \(\Gamma_X\).

Try your own reinterpretation — a minimal starting point for fitting the model at two fixed widths.

pyhf-tutorial

2025-09-26T00:00:00+02:00

Three self-contained tutorial notebooks on statistical inference with pyhf: building histogram-based models and including uncertainties, a realistic statistical model for the B⁺→K⁺π⁰ decay, and hypothesis testing. Used for teaching at the Belle II StarterKit, Belle II Physics Week, the Belle II Summer Workshop, the Belle II Academy, and the PUNCH Young Academy.

Have a look at the repository here:

github.com/lorenzennio/pyhf-tutorial

BibSpire

2025-08-27T00:00:00+02:00

BibSpire is a Python tool that cleans up bibliography files. It reads a .bib file, searches each entry on INSPIRE-HEP, and replaces the entry with the official INSPIRE citation while keeping the original reference key. It uses title, author, eprint, and DOI information to find matches, and handles nested-brace BibTeX formatting and rate limiting to stay within the API.

Have a look at the repository here:

github.com/lorenzennio/bibspire

Public likelihood for B→Kνν

2025-07-01T00:00:00+02:00

The official Belle II collaboration paper releasing a public, model-agnostic likelihood for the \(B^+\to K^+\nu\bar\nu\) measurement, built with the general method described above. It lets anyone test their own new-physics model against the measurement's data directly, without access to the internal analysis.

Published as: Model-agnostic likelihood for the reinterpretation of the \(B^+\to K^+\nu\bar\nu\) measurement at Belle II, Phys. Rev. D 112, 092016 (2025).

Posterior constraints on the effective Wilson coefficients of a general \(b\to s\nu\bar\nu\) contact interaction, from the public likelihood applied to the full \(B^+\to K^+\nu\bar\nu\) measurement.

Try your own reinterpretation — a minimal worked example, from loading the public likelihood to a full WET fit.

pyhfcorr

2024-11-12T00:00:00+01:00

pyhfcorr adds support for arbitrarily correlated uncertainties to pyhf, whose HistFactory-based models natively support only fully correlated or uncorrelated systematics. It pre-processes a model’s correlations block — a modifier name, the correlated parameters, and their correlation matrix — using singular value decomposition to produce a standard, pyhf-compatible specification with the correlation structure built in.

Have a look at the repository here:

github.com/lorenzennio/pyhfcorr

redist

2024-02-06T00:00:00+01:00

REDIST is a reweighting method for reinterpreting binned analyses in high-energy physics: it recomputes how an observable’s distribution changes under alterations to the kinematic distributions of a decay channel. It’s built on pyhf for statistical inference, with an optional JAX backend for differentiable likelihood calculations, and integrates with EOS for theoretical predictions. The method is described in arXiv:2402.08417.

Have a look at the repository here:

github.com/lorenzennio/redist

Constructing model-agnostic likelihoods

2024-02-01T00:00:00+01:00

The general method behind a line of work on reinterpreting particle-physics measurements without redoing the full experimental analysis for every new-physics model: a likelihood built directly from a measurement's signal-region yields, that lets anyone test their own model against the data by supplying only its predicted event distribution. Introduced and validated on simulated data.

Published as: Constructing model-agnostic likelihoods, a method for the reinterpretation of particle physics results, Eur. Phys. J. C 84, 693 (2024).

Posterior samples on a set of effective Wilson coefficients from combining the method with a projected future dataset (50 ab\(^{-1}\)): constraints sharpen substantially with more data, resolving degeneracies present in a smaller, 362 fb\(^{-1}\) sample.

Circumbinary disk simulations

2021-12-13T00:00:00+01:00

I simulate the gas disk of a binary star system to investigate temperature distributions and compare them to the specific physical system IRAS 16293-2422 A. I run a parallelized, static mesh refined simulation of the system using the magnetohydrodynamics code Athena++. To identify the relevant physical processes to be included in the simulation, I have adapted the original code to the specific needs. I have learned a lot about numerical solvers and mesh refinement. I run the code on the institute cluster with Slurm. The output of the simulation then has to be analysed and visualised, for which I use Python specific tools.

On the backreaction of quantum scalar fields on the de Sitter spacetime

2020-09-30T00:00:00+02:00

The questions I wanted to address in my master thesis were: What effect do the particles, created during the expansion of our universe, have on the cosmological evolution?

In this work we analyse the behaviour of a quantised scalar field in a de Sitter background (a model of an exponentially expanding spacetime, which is often used as an approximation for the inflationary phase in the early universe) and the backreaction on the geometry. Due to the dynamic nature of the background no unique vacuum, diagonalising the Hamiltonian at all times, exists on the basis of which we can build a global Fock space. Imposing de Sitter invariance of our vacuum we find a two parameter class of vacua known in the literature as Mottola-Allen vacua, which are invariant under the time-preserving part of the de Sitter symmetry group. Furthermore, fixing one of the mentioned parameters results in a one parameter group of vacua, leaving argument-symmetric Green functions invariant under the full de Sitter group. Lastly, matching our result to the flat Minkowski solutions on very small scales where curvature is expected to be negligible, we can fix the last parameter and obtain what is referred to as the Bunch-Davies (BD) vacuum. We use our obtained insight to compute the expectation value of the regularised energy momentum tensor for a general choice of de Sitter invariant vacua. We conclude that as long as we respect de Sitter invariance, we do not get any dynamic backreaction and only obtain a constant shift in the cosmological constant. We then look for the breaking of this isometry in the loop corrections of a self interacting \( \lambda \phi^4 \) theory, but find that it is generally also respected in loops, as long we restrict to a certain coordinate patch of de Sitter spacetime. Finally, we break de Sitter isometry explicitly by introducing a scalar metric perturbation to the de Sitter geometry. We introduce a free massive scalar field to our spacetime and estimate the backreaction by solving for the perturbation. Our results show that in the short wavelength (UV) regime the largest contribution to the metric perturbation decays while oscillating in a similar way to gravitational wave modes. In the long wavelength (IR) regime we find that one part of the solution significantly grows and these perturbations can no longer be considered small.

Have a look at my thesis here:

This browser does not support PDF display. Please download the PDF to view it.

</embed>

Background event classification for Belle II

2018-07-31T00:00:00+02:00

In this project I used neural networks to classify EvtGen simulated data into signal, \(B \to K^* \nu \bar{\nu} \), and background. The goal was to investigate the possibility of bypassing the expensive Geant4 detector simulation and following analysis. This project was done in the realms of the Belle-II experiment, but is applicable to all of experimental particle physics. It was a very new direction in the collaboration, where I helped to develop one of the first convolutional neutral networks for classification. By restructuring the network layers and managing to constructively input additional data (such as the decay string), I managed to improve the accuracy by over 10% and set the groundwork for future contributions. The project is still ongoing and being constantly improved.

During this project I learned about neural networks, some common design practises and options for handling and pre-processing data. I used Keras and other parts of Tensorflow to embed relevant data from the simulation and develop a convolutional neural network for classification of signal and background in particle collisions. Furthermore, I learned about the simulation side of particle physics and some details about general data handling.