Bert's blog

The St Andrews Monte Carlo summer school

The St Andrews Monte Carlo Summer School or SAMCSS is a biyearly summer school that is primarily aimed at PhD students based in the UK, and has been organised since 2013. I have been involved in its organisation for both the 2017 and the 2019 edition, the latter of which was held this week.

The original topic of SAMCSS have been the use of Monte Carlo methods to study radiative transfer in astronomical problems. Since the community at large seems to have a current interest in combining all kinds of radiative transfer with hydrodynamics to perform radiation hydrodynamics (RHD), the 2017 and especially the 2019 edition expanded on this original topic by incorporating more hydrodynamics and lectures on how to couple both techniques. Below is a very personally biased summary of the summer school and its main topics.

Monday: MCRT and hydro

Since the two main topics of the summer school were Monte Carlo radiative transfer (MCRT) and its combination with hydrodynamics, it made sense to devote the first day to a low-level introduction to both MCRT and numerical hydrodynamics. The former was introduced by Kenny Wood, the main organiser of the school, and was a reiteration of the introductory lecture he gave for previous editions of the school. Next up was a lecture by school alumnus Antonia Bevan; she gave a very down to Earth overview of what she had to do to develop her own MCRT code based on what she learned during the summer school she attended.

Having worked very intensively with Kenny over the past three years on the development of my own MCRT code, these lectures did not really provide me with any shocking new insights, but that couldn’t really be expected either. I think my main lesson of the day was something very practical: Antonia mentioned how it is a really good idea to make the variables you use in loop counters (typically i, j and k) two characters instead of a single character (so ii, jj and kk), so that they become searchable. This is a very simple trick, but I think this could have saved me quite a lot of time in the past. I had seen this kind of loop counters before, but it had never occurred to me there was actually a good reason to do this.

The afternoon was entirely devoted to an introduction to numerical hydrodynamics that I lectured myself, so I will not comment any further on it.

Tuesday: RHD

Having spent the first day learning all about MCRT and hydrodynamics, the second day provided the logical continuation: the combination of both for RHD. We started the day with a talk from Tim Harries about this very topic, although he did spent the first half of his lecture talking about MCRT optimisation techniques as well (you want to make sure you can do the radiative transfer very efficiently before even considering coupling it to a real-time hydrodynamics scheme). This lecture contained some techniques I hadn’t encountered before (the modified random walk for example). But it most of all contained a lot of very clear images that showed the result of various optimisation techniques. So I think the most important lesson there was to use clear images when you are trying to make a point.

The morning session was closed by Stuart Sim, who introduced the notion of performing MCRT simulations in a highly dynamic, time-dependent environment: the study of supernova explosions. Again, the things I remember most of this lecture might be less relevant to the topic of the school, as they are more about the use case of these simulations. Stuart mentioned that most of the light from a supernova that we see is not actually created by the supernova explosion itself: it is partially due to the thermal afterglow of the explosion from the hot gas and other factors, but is primarily caused by the decay of highly unstable radioactive elements that are excited during the explosion. I guess this is something I simply didn’t know.

The afternoon session was all about Lyman (Ly) MCRT. Ly is a particular frequency of light that gets emitted when an electron in a hydrogen atom falls back from its first excited state into the ground state. Aaron Smith explained in great detail how this light has great difficulty to actually escape the gas clouds from which it is emitted, as this light (obviously) has the exact right frequency to excite an electron from the ground state of a hydrogen atom to its first excited state, so that it is absorbed again. Ly can only escape by (very) slowly shifting in frequency due to all kinds of small random Doppler shifts that happen within a real turbulent gas cloud. Modelling this slow escape process is extremely challenging, and Aaron discussed a lot of optimisation techniques that are required to model it nonetheless.

As a small aside: the Balmer (H) line is emitted when an electron falls back from the second excited state to the first excited state of the hydrogen atom. Unlike Ly, this line has not much trouble escaping a gas cloud, since it is reasonably unlikely that another atom in the gas cloud will have an electron in its first excited state that can absorb this light to jump to the second excited state. This is why modelling H is much easier; this is something I have done in the past.

Wednesday: more codes and processes

After the introductory lectures of the first two days, the third day featured some more advanced topics. Kees Dullemond gave an overview of his widely used RADMC-3D code, from which I mainly remember that choosing a somewhat obscure name for a code (RADMC-RD, CMacIonize) instead of a nice sounding acronym (like Shadowfax, Gadget, SWIFT…) is actually a good idea, as it makes it a lot easier to find your code through a search engine. Again, a good point that I had not realised before.

Tom Haworth gave a very accessible introduction to the ALMA interferometer, and explained how to make model images of ALMA observations to support applications for telescope time. Nothing I particularly fancied, but definitely a good lecture.

In the afternoon, Stuart Sim introduced the notion of macro atoms to deal with atomic transitions in gas clouds. Some applications of MCRT try to model the spectrum that escapes from radiatively excited gas, and these applications need to model the atomic processes responsible for emitting and absorbing specific spectral lines. An important aspect of these models is energy conservation: you want to make sure that all the light that is emitted by the cloud eventually escapes from it; atoms are not exactly very good at storing energy.

The issue with this is that light that gets absorbed at one frequency might actually be reemitted in stages at different frequencies: light can excite an electron from the ground state of the hydrogen atom into its second state. The electron will then emit an H photon to fall back to the first excited state, and only then will it fall back to the original ground state by emitting another Ly photon.

One way of tracking this would be to actually split photons within the MCRT algorithm, but this becomes intractable very quickly. The macro atom formalism offers an alternative method by modelling the excitation and decay process for a single atom as a traffic flow problem, where various transition channels have probabilities based on the underlying atomic properties (that can be measured). For every absorption event, a single ingoing photon will set off a random walk through the subsequent excitation and deexcitation process using these probabilities, until a new photon is emitted. By modelling the traffic flow within the atom itself using a Monte Carlo technique, you can guarantee energy conservation, even though the frequencies of ingoing and outgoing photons are not necessarily the same.

The last lecture of the day was given by Kenny Wood and dealt with Monte Carlo photoionization. Since this is exactly what I have been working on for the past three years, there was nothing new there for me.

Thursday: parallelisation and confidence

This last day of the summer school did not really introduce any new science topics, but instead dealt with some more practical and personal aspects of numerical science, as well as an interdisciplinary application of MCRT. During the morning session, Tim Harries introduced OpenMP and MPI as main techniques to parallelise a numerical algorithm, after which I explained why parallelisation is so important and why we should aim to develop our codes in a more inherently parallel way, by changing the way we think about algorithms. After this, Lewis McMillan gave a nice overview of the work he did using MCRT to model light propagation and tissue damage in skin.

In the afternoon, Antonia Bevan gave a very personal (and brave) talk about the struggles of academia and how the very competitive nature of academia can very easily cause confidence issues (the so called imposter syndrome). She explained what imposter syndrome is, how it manifests itself and affects you (and your work), and what you can do to help yourself and others to cope with it. She then also touched upon the diversity challenges science faces, since science is still dominated by white, middle-class males. This led to a very interesting open discussion in which various points were made about gender biases in academia and how we can work to improve the gender balance within numerical astrophysics (why do so little girls code? as Kenny would put it). Antonia made a very good point during the discussion about how it is up to white, middle-class males to actively counteract any unconscious biases that still exist, as it is mainly these biases that make it hard for any minority to feel at ease within the scientific community. For me she managed to drive home the message that any kind of stereotyping is really damaging, even if it is by no means intended to be harmful. And she convinced me that I should try to be more aware of these kinds of stereotyping and try to actively speak out against them.

Summary

Although the scientific contents of the summer school was not really relevant to me personally (as I was already very familiar with the relevant topics), I still think this was a very interesting summer school. There was a very good interaction between the participants, and especially Antonia Bevan’s final lecture made it very easy for me to interact with some of the participants on a more personal level. I saw a lot of very motivated and nice people that were willing to look at the future with a very open mind. This does give some hope that academia is changing for the better, although we still have a long way to go.

Scientifically, I had many interesting discussions with Aaron Smith, and for the first time really managed to discuss my new task-based MCRT algorithm with him and Tim Harries. I hope to have some time in the near future to write things up and get this algorithm out there; it is not really ready, but is innovative enough as it is to be very helpful for people in the field.


Professional astronomer.