Serendipity in Astrophysics

Dr. Patrick Kelly1 is an assistant astronomy professor in Minnesota Institute for Astrophysics. During his postdoctoral time at Berkeley he made his first serendipitous discovery of the first gravitationally lensed distant supernova, using the Hubble Space Telescope. This, in turn, led to further discoveries with tremendous scientific impact in the field. This interview discusses a personal story of serendipity in astrophysics which illustrates that luck comes to those who are prepared to welcome it.


JUnQ: Let’s start from the beginning. Can you tell us in simple words what the discovery from your Science paper1 in 2015 was about?
Patrick Kelly: Actually, the discovery was in 2014 and the paper came out in 2015. When I was a graduate student, I studied weak gravitational lensing by galaxy clusters. Weak gravitational lensing refers to the case where you distort the images of background galaxies, but not multiply image them. But you still can measure the masses of galaxy clusters, and then matter distribution in a unique way, which is a very useful tool to study the distribution and total amount of dark matter in galaxy clusters. As you know, we cannot detect dark matter directly, but we can detect its gravitational effect. Before that, I studied supernovae as an undergraduate student at Harvard working with Bob Kirshner. At Stanford, a postdoctoral scholar I was working with suggested that I could help out with a new project with the Hubble Space Telescope that Prof. Tommaso Treu was starting. The focus of the project, called the Grism Lens-Amplified Survey from Space (GLASS), was to take spectra of galaxies that were gravitationally lensed using the instrument on Hubble. I was the transient supernovae person on the team as I was looking for supernovae. So, we found a couple and, then, we happened to find this one which was very special. We could immediately see that there were three bright images (Fig. 1).

Of course, for supernovae everything to some extent is serendipitous because you cannot in general predict when stars will explode. Although, in the case of this supernova we were actually able to predict for the first time that a supernova would appear in the sky, when we predicted the reappearance. So, it is a very interesting twist on your serendipity. Because with this very serendipitous case we were able to make an actual prediction.2

After that we spent a lot of time in understanding what was happening. In fact, related to this discovery was another serendipitous discovery in the same field which I also made. We were taking images of this multiply imaged supernova over a long period of time. It was something that hadn’t really been done before with the Hubble space telescope looking at the same cluster lens for a year every couple of weeks. So, we actually found a new phenomenon – the microlensing of an individual star at a cosmological distance, which in this case was more than half-way across the universe. And that was another serendipitous discovery.


JUnQ: What was the original plan for that project?
Patrick Kelly: The data were taken by the Hubble space telescope as a part of the GLASS program. Every image was taken once a month or something like that, and each field we’d returned to once. So, we had multiple images of them, and they had already been imaged by other programs of Hubble. And the goal was… Well, we all knew that it would be wonderful to find one of these supernovae, but the probability was low. So, in that sense we knew we were looking for something like this, but we didn’t have any hope we would actually find it. Whereas actually the discovery of the star – no one was really looking for that. That was even, I would say, more serendipitous. So, I guess we knew that there was a possibility this could happen. There is another crazy thing about this supernova. If you have a gravitational lens, you can get multiple images of the source if it is aligned in a certain way. But here we get multiply imaged both by the galaxy cluster and by a galaxy in the galaxy cluster. So that was even less likely and this is probably why it is so spectacular. The image looks amazing.


JUnQ: How did it feel, the moment you saw the image? Was it difficult to realize, persuade yourself that it was something real?
Patrick Kelly: In this case it was immediately clear that it was very exciting. There were multiple images and I think I emailed everybody right away. Also, on the team, they were all very excited. And then we didn’t get much sleep for the next couple of weeks. The data all become public immediately for this particular program. And you try to analyse them quickly because you know that other people may have seen them too, so you want to publish them quickly. But it was a lot of fun.


JUnQ: What kind of impact did this discovery have on your career and the field in general?
Patrick Kelly: The scientific impact hasn’t been realized yet. I was working on a couple of papers to measure the expansion rate of the Universe using the supernova. And hopefully those will be very exciting for people. There are also some interesting things that we figured out with a very careful analysis. But they are not published yet.
For career – there were a couple of us. For sure it helped me to find a permanent job, I have no doubt. So, it was a great thing for me. And, especially for the second author on that paper, Steven Rodney. He is also an assistant professor now. He made a huge contribution as well so, I am sure, it was very helpful for him to have found it.
It is just kind of fun for the field. Because a lot of people got very excited about it, making the predictions. There is a paper by Tomasso Treu who collected the predictions from the community.3 There are different ways to model the dark matter distribution. So, this discovery gives us a unique opportunity to test the quality of the current models by conducting a textbook-like falsifiable experiment. That is the Karl Popper’s idea4 of falsifiable prediction, which you can do in science especially in astronomy very often. So, we are still working through the implications and we are trying to be very careful.

JUnQ: Did you have more of these lucky discoveries?
Patrick Kelly: Certainly, the discovery of the star. I never thought we could see individual stars across the Universe. It seemed totally crazy. So that was lucky. And then for other papers I feel like we were lucky to make some kind of discoveries, but it was not like we found an individual object. Probably a lot of the things are lucky, because you look at the right time and you have the right skills.


JUnQ: Making successful discoveries in astronomy is not comparable to buying a lottery ticket, right? What is generally needed for luck to strike someone? How can one be prepared?
Patrick Kelly: Part of being prepared is to have a good team of collaborators. Because everyone has different expertise. For us, there were people who were supernovae experts, people who were modellers of the dark matter, and gravitational lensing experts and galaxy experts. Pulling all those people is important. Because then you can rapidly interpret what is going on. So, I think it is a really important piece.


JUnQ: There are hundreds of research papers having “serendipitous” in the title. But yours does not, why is that?
Patrick Kelly: I think it is because in part one can never predict (with the exception of our own prediction) when a supernova or a star will die or explode. That’s how transient astronomy works – you never know when something happens. So, you try to put yourself in a position where you could find something exciting. And that was what we were trying to do. It wasn’t even the main driver of the science program. We were just lucky in that it happened while we were observing, but it was not completely unexpected I would say.


JUnQ: Astronomy seems to be the field of science where chance discoveries certainly play an important role. How do you think the advances in machine learning and artificial intelligence will impact the chances of serendipitous discoveries?
Patrick Kelly: We are looking at various things that are exciting and intrinsically rare and that haven’t been seen before. You certainly are going to find more of those if you do more searches and collect more data.


Figure 1. The image shows a part of the deep field observation of the galaxy cluster from the Hubble Space Telescope. The many red galaxies are members of the massive MACS J1149.6+2223 cluster, which creates distorted and highly magnified images of the galaxies behind it. A large cluster galaxy (center of the box) has split the light from an exploding supernova in a magnified background galaxy into four yellow images (arrows). Image credit: NASA, ESA, and S. Rodney (JHU) and the FrontierSN team; T. Treu (UCLA), P. Kelly (UC Berkeley) and the GLASS team; J. Lotz (STScI) and the Frontier Fields Team; M. Postman (STScI) and the CLASH team; and Z. Levay (STScI).

So, I think we will start finding more crazy things for sure. And even if each of them has low probability, if we look enough, hopefully we will find more of them.

In astronomy, there is a movement towards wider and wider field telescopes. One of the new NASA missions is called the Roman space telescope, previously called W1. It has a field of view of a hundred times Hubble’s. So, it should be able to discover a hundred times more transients.

I was searching the data from Hubble by eye. When we scale it up like this, we’ll certainly need some machine learning and certainly software to help. But I think it is a challenge for sure. We will see, we should find interesting things.


JUnQ: Do you have some advice on how to be lucky?
Patrick Kelly: I guess you never know what could be around the corner and it is a reason to be optimistic. But maybe I am just lucky in this case. It seems like luck is an important part of science for experimentalists. So, you just try your best.


JUnQ: Thank you very much for the interview!


— Mariia Filianina
Read more:

1 P. L. Kelly et al. Science, 2015, 47, 1123–1126. DOI: 10.1126/science.aaa3350.
2 https://en.wikipedia.org/wiki/SN_Refsdal.
3 T. Treu et al. The Astrophysical Journal, 2016, 817, 60. DOI: 10.3847/0004-637X/817/1/60.
4 K. R. Popper, The logic of scientific discovery. (Translated from German original: Logik der Forschung. Zur Erkenntnistheorie der Modernen Naturwissenschaft). 2002. London: Routledge.