This is a list of known issues in the printed version that I handed in. The box indicates whether these have been fixed.

• [ ] <2024-02-01 Thu 18:26> The background rate plots, over the entire chip, which use filterNoisyPixels use the list of pixels from the Timepix3 detector and not the small noise region for the Septemboard. That implies the background rates might be a small bit higher in the full chip background rate plots than in reality.
• [X] The appendix about the software and how to reconstruct the entire results, still lists the data reference as XXX instead of the correct link. -> Update <2024-05-10 Fri 12:39>. Now links to the correct Zenodo dataset.
• [X] The text below the figure explaining the FADC as a means to close the Timepix shutter mentions a time difference of 50μs. I think that should be 50 clock cycles. Need to cross check. -> Update <2024-02-07 Wed 11:34>. Indeed, the time should have been 5 μs and 200 clock cycles. See ./../../org/journal.html and this definition here: ./../../CastData/ExternCode/TOS/include/fpga.hpp https://github.com/Vindaar/TOS/blob/master/include/fpga.hpp#L49-L51 (which is used in the fpga constructor and set in EnableFADCShutter without being changed).
• [X] <2024-03-27 Wed 13:35> The observed axion-photon limit given in the summary is an outdated value I forgot to update (the value in the limit section about it is the correct value).
• [X] <2024-06-25 Tue 13:13> Whoops, just noticed that Araq's name in the bibliography is wrongly as 'Rump' instead of 'Rumpf'. Sorry Araq!

If you would like to read the full thesis, you have multiple options.

• the official publication (PDF as of before my defense; minor issues may still be present, see errata)
• an extended HTML version (up to date)
• an updated PDF version
• the extended version as a PDF

One of the most important parts for my thesis has always been being able to replicate results. Part of that is providing access to my data, including the raw data taken with our detector at CAST.

The raw data, as well as reconstructed data and further miscellaneous files can be found on Zenodo:

https://zenodo.org/records/10521887

The Standard Model of particle physics (SM) is extremely successful at describing three out of the four known fundamental forces of nature. However, there are several open questions even within the SM. One of those is termed the 'strong CP problem'. In a simplified way it asks why the neutron does not have an electric dipole moment. While only a mathematical problem of sorts, it can be neatly solved by the introduction of a new hypothetical particle, the axion. This particle can interact with photons and leptons, even if extremely feebly. If they exist, they will be produced in the core of the Sun at very high rates, but easily escape similar to neutrinos. On the Earth we can attempt to detect them using strong magnetic fields. The latest such experiment, a solar helioscope, is the CERN Axion Solar Telescope (CAST). Axions carry the energy of the photons, which initially produced them, placing them in the soft X-ray energy range.

In the course of this thesis, a gaseous detector using 7-GridPixes as its readout was deployed at the CAST experiment in 2017/18 to follow up on previous measurements with a single GridPix detector in 2014/15. \(\SI{3150}{h}\) of background data and \(\SI{160}{h}\) of solar tracking data was taken. The data taken at such an experiment is dominated by cosmic radiation, radioactive background and X-ray fluorescence. As such methods to suppress this background and extract the few possible axion signals are needed. A software library TimepixAnalysis was developed in the course of this thesis to reconstruct and analyze the data taken at CAST.

For the classification of event data into background-like or signal-like (X-ray) data, a machine learning approach is used. A multilayer perceptron (MLP, simple type of an artificial neural network) was trained on synthetic X-ray data and CAST background data of 6 of the 7 chips (leaving out the chip sensitive to axions). This method significantly improves the signal efficiency at comparable background rates, when compared to the likelihood cut method used for the old detector.

Further, as no excess signal was detected in the solar tracking dataset, a limit calculation method was developed. This method builds on the unbinned Bayesian likelihood method used in CAST's Nature paper in 2017. It is extended to allow the inclusion of systematic uncertainties as nuisance parameters. Due to the very expensive evaluation of such a likelihood function, a Markov Chain Monte Carlo (MCMC) approach is used.

One of the limit calculation inputs is the 'axion image' produced by the Lawrence Livermore National Laboratory (LLNL) telescope at CAST. To properly characterize this, a raytracing simulation taking into account the axion production rates in the Sun and reflection through the X-ray optic was developed. It was verified against PANTER measurements of the telescope.

With the software advances and improved detector features, world best limits could be set on the axion-electron coupling and on the chameleon coupling (a separate hypothetical particle). The previous best limit on the axion-electron coupling is improved from \(g_{ae}·g_{aγ} \lesssim \SI{8.1e-23}{GeV⁻¹}\) to \(g_{ae}·g_{aγ} \lesssim \SI{7.35e-23}{GeV⁻¹}\) and the chameleon coupling from \(β_γ < \num{5.74e10}\) to \(β_γ \lesssim \num{3.1e+10}\).

In addition, the software developed during the course of this thesis was written with future detectors in mind. It is in principle ready for data analysis for a GridPix3 based future detector.

The main repository used in my thesis is TimepixAnalysis. See the appendix about the software and how to reconstruct all data in the thesis for more information.

The results in the thesis were all produced with the git tag phd_rc0. It pins all important libraries to fixed versions for reproducibility. Once the final thesis is published I'll create a phd tag (and make sure I didn't break anything in between).

There are 2 major files potentially of interest to the kind of reader who may stumble on this website.

The first is a sort-of journal I wrote over the years. It was used more as a day-to-day journal in 2023 and contains some other 'journaley' notes from before.

The second is my "status and progress" type document. It contains a large amount of notes related to all sorts of researching ideas / results etc. Almost all things that ended up in the thesis were written at one point or another in one of the two documents. So they may be worthwhile to look into.

• The 'journal': http://files.vindaar.de/journal/journal.html
• The status and progress notes: http://files.vindaar.de/status/StatusAndProgress.html

Note that both of these documents are very large. At least on my laptop running Brave these two pages are no fun. Firefox however deals with them very well. I'll probably end up splitting them into different pages at some point. But if you are more seriously interested in them, I recommend downloading the Org file of the document (just replace the .html ending by .org; note: currently not available for the journal, because I use that journal for stuff unrelated to the PhD as well and I haven't produced a standalone Org file without those…).

Aside from these, another 2 files may be of interest:

• The calculations related to a possible gas phase for (Baby)IAXO: http://files.vindaar.de/phd/docs/bufferGasIAXO/v1/axionMass.html or http://files.vindaar.de/phd/docs/bufferGasIAXO/v1/index.html
• The original document of the notes containing the calculation of the median and mean conversion point of axions from the Sun in our detector: http://files.vindaar.de/phd/docs/SolarAxionConversionPoint/axion_conversion_point.html These are almost identical to the appendix about the same in the thesis.

A few other documents exist that I might upload at some other time (related to the LLNL telescope, raytracing etc.). These are mostly already contained in some form or another in the above though.

This main page is hosted on Github using Github Pages. But all the linked pages mentioned in the section above are just files publicly linked on Backblaze B2. I might migrate them elsewhere at some point to support TLS there too, but as all the plots alone sum up to almost 1GB of data, I'm not sure where at the moment.

But then again, who cares for this kind of stuff?