• 1 Introduction
• 2 2025 LSST Data Rights Call
• 3 Short-term LSST Data Rights for project students
• 4 New paper: The quenching of star formation in dwarf galaxies – new perspectives from deep-wide surveys
• 5 Testing the LSST Stack and imSim performance on Isambard 3 
• 6 LSST:UK Project Leader position
• 7 Unbiased galaxy clustering measurements directly from catalogues
• 8 Final call for nominations for Executive Group
• 9 Rubin at Cheltenham Science Festival
• 10 LSST:UK communications update
  • 10.1 Shape how the public understands LSST:UK and Rubin
  • 10.2 Video resource: Rubin Seminar for Science Writers
• 11 Leadership positions held by LSST:UK members
• 12 Forthcoming meetings of interest

Open

Introduction

As most of you will know already, 15 April saw the first on-sky engineering data taken with LSSTCam. This major milestone (see right for jubilant responses from Rubin staff) marked the start of a very successful first week of on-sky observing with LSSTCam, building on the successes of the on-sky ComCam observations before Christmas. So, for example, the system managed to deliver 0.7-0.8 arcsec image quality across the whole field of view on only the third night of LSSTCam observing, thanks to the optical alignment accuracy that had been obtained with ComCam. As during the ComCam observing, the Commissioning team will be posting weekly updates on their progress to the Community forum - see https://community.lsst.org/tag/commissioning-update - and all are encouraged to keep an eye on those over the next few months.

There are also notable events to report within LSST:UK. May 1st sees the opening of our next data rights call, which will run until the end of July, to catch new applicants inspired by the Rubin First Look celebrations due in June or July and by our sessions at NAM. As @Terry Sloan also notes below, we now have a separate mechanism through which undergraduate and masters students can apply for data rights for one year, to cover a specific project that they are working on as part of their degree. Since the need for such students to have data rights can arise at various points in the year, applications can be made at any point in the year and will be processed very promptly.

Finally, our Consortium Board Chair, @Mike Watson , describes how we are launching appointment processes for a number of leadership positions in the LSST:UK Consortium. As previously mentioned in the April Newsletter, and in a mailing to the lusc-announce list, the current terms of three of our five elected Executive Group members are ending; Mike notes below that 2 May is the deadline for applications to fill those three positions. Similarly, my current term as Project Leader is soon to end, and Mike describes below the process by which the Consortium will fill the Project Leader position for the next three years, starting with a call for expressions of interest by 20 June. Please consider applying for these positions, which are important for the smooth running of LSST:UK.

Open

Open

 

2025 LSST Data Rights Call

On 1 May 2025, a new call for applications for LSST Data Rights will be issued via the lusc-announce email list.

Among other things, LSST data rights allow you to access data previews – e.g., on the Project Science Platform – and to participate in Rubin International Science Collaborations. The call will close at 4pm on 29 July 2025. Applicants will be informed of the outcome of their applications on 1 September 2025.

Staff and students from any organisation eligible to hold an STFC grant may apply. Since the UK’s Data Rights Agreement is awaiting signature by STFC and the US Department of Energy, this call will be for LSST Data Rights up to 30 September 2026.

Most existing data rights holders already have data rights to at least 30 September 2026 and hence need not re-apply in this selection round. If you are unsure of the end-date of your data rights term, you can check this on the current UK data rights holders list. If your end-date is BEFORE 30 September 2026 and you wish to extend your term to that date, then please contact contact the LUSC Project Manager, Terry Sloan (t.m.sloan@ed.ac.uk).

More detailed information about the call will be provided in the email to lusc-announce and more information on data rights is available on the Rubin Data Policy website.

Short-term LSST Data Rights for project students

LSST:UK is now also able to award – in a responsive mode – LSST Data Rights to MSc and undergraduate students undertaking short-term projects. These can be applied for at any time of the year and are for short-term Junior Associate (JA) data rights of length up to one year.

Further information on the application process and the link to the application form are available on this page.

@Terry Sloan

New paper: The quenching of star formation in dwarf galaxies – new perspectives from deep-wide surveys

MNRAS has recently published an article by Sugata Kaviraj and collaborators on the topic of how star formation in dwarf galaxies is quenched.

Since dwarf galaxies dominate the galaxy number density, they are critical to our understanding of galaxy evolution. However, typical dwarfs are too faint to be visible outside the very local Universe in past surveys like the SDSS, which offer large footprints but are shallow. Dwarfs in such surveys have relatively high star formation rates, which boost their luminosity, making them detectable in shallow surveys, but also biased and potentially unrepresentative of dwarfs as a whole. Crucially, this makes it virtually impossible to draw conclusions about the star formation properties of dwarf galaxies using the small subsets of dwarfs that are detected by shallow surveys. 

 In this study Kaviraj et al. use deep data to perform an unbiased statistical study of around 7000 nearby (z < 0.25) dwarfs (8 < log M/M_SUN < 9.5), in the COSMOS field which, at these redshifts, is a relatively low-density region. Most importantly, the detection image used for producing the catalogue employed by this study utilises ultra-deep imaging from the Hyper Suprime-Cam, which is the pre-cursor instrument to LSST.

 Kaviraj et al. show that, at z ~ 0.05, around 40 per cent of dwarfs in low-density environments are red/quenched, falling to around 30 per cent by z ~ 0.25. Red dwarfs reside closer to nodes, filaments and massive galaxies. Proximity to a massive galaxy appears to be more important in determining whether a dwarf is red, rather than simply its distance from nodes and filaments or the mean density of its local environment. Interestingly, around half of the red dwarfs reside outside the virial radii of massive galaxies and around a third of those also inhabit regions in the lower 50 per cent in density percentile (i.e. regions of very low ambient density). Around half of the red dwarf population is therefore quenched by mechanisms unrelated to environment, which are likely to be internal processes such as stellar and active galactic nucleus feedback. Many of these conclusions are in sharp contrast with what has been suggested in the past using shallow surveys like the SDSS and demonstrates the crucial need for using deep data (similar to what will soon be available from LSST) to explore the dwarf galaxy regime.  

Open

The rest-frame (g−i) colour of galaxies in this study (which uses the COSMOS2020 catalogue), shown as a heatmap, with the location of SDSS spectroscopic objects within the COSMOS2020 footprint shown using blue contours. A random sample of SDSS galaxies is overplotted using the blue open circles. The detection limit of the COSMOS2020 dataset is around 10 magnitudes deeper than that of the SDSS spectroscopic sample and comparable to the depth of 10-year LSST data. While massive galaxies appear in the SDSS spectroscopic sample regardless of whether they are red or blue, galaxies become progressively bluer with decreasing stellar mass, as red objects preferentially fall out of the selection because they are fainter at a given stellar mass. The dwarf population in shallow surveys like the SDSS are therefore strongly biased towards highly star forming systems and cannot be used to draw accurate conclusions about the star formation properties of galaxies in the dwarf regime. Deep data is critical for an accurate and unbiased exploration of the dwarf-galaxy population.

Testing the LSST Stack and imSim performance on Isambard 3 

For the past decade or so, the majority of HPC systems have been built with nodes containing Intel or AMD processors using the x86 instruction set. For example, CSD3 in Cambridge uses Intel Xeon CPUs, Perlmutter at NERSC uses AMD EPYC, and S3DF, the machine at SLAC hosting the USDF, has both Intel and AMD nodes. In a change of tack, NVIDIA’s new platforms are using its Grace CPUs which use ARM’s aarch64 instruction set. 

NVIDIA is producing several designs using its new ARM CPUs, but of particular interest to HPC and scientific computing are what it calls its superchip designs: either two 72-core Grace CPUs (a Grace superchip) or a 72-core Grace CPU and a Hopper H100 or H200 GPU (a Grace Hopper superchip), both sitting on the same board with a very high bandwidth memory pipeline between the two components and enough software paging help behind the scenes to make it look as though the two halves’ memory is unified. Of course, the Grace Hopper superchip is named after the pioneering computer scientist while NVIDIA have recently announced that the next generation design will be called Vera Rubin. 

Clusters using NVIDIA’s new hardware are now coming online around the world. Of note for us in the UK are two HPE Cray EX clusters operated by the Bristol Centre for Supercomputing: Isambard 3, a Grace superchip-based CPU-only machine, which is now online, and Isambard-AI using the GH200 Grace Hopper superchip, which is planned to begin service soon. 

In autumn 2024, we were made aware of the opportunity to apply for preparatory access to Isambard 3 before the system opened more generally through the usual UKRI Access to HPC channel. We were interested in seeing the performance of the Grace superchips and testing the ease of use of the Cray Programming Environment and Cray OS which, while mature on Intel and AMD CPUs, are very new and perhaps an unknown factor on the ARM platform. We applied for and were granted access to the machine on the basis of testing the LSST Science Pipelines and imSim. 

Access to Isambard 3 opened to us in September 2024. The pipeline installation came first. Using the GNU-based Cray Programming Environment, it took less than a week to create a procedure for installation with lsstinstall. Only two issues were encountered: the first was related to a bad JAX package release that caused tests in the meas_algorithms package to fail, which was encountered and fixed by the pipelines team at the same time. The second was a larger hurdle related to a bad page size configuration in the conda release of the ruff linting and formatting tool, that was also being experienced simultaneously by other Grace CPU users around the world. Now fixed upstream, our initial workaround was to build ruff and its jemalloc dependency manually. Otherwise, the pipelines installation was extremely smooth and a good demonstration of the software’s reliability. 

imSim sits on top of the stack. It runs as a GalSim module and principally interacts with skyCatalogs, using the first for the bulk of the computational work and the latter to obtain observational details and information on the objects to be drawn based on visit information from OpSim. We built GalSim manually to be sure it was as optimised as possible for the hardware, and with three different FFT backends to test that they worked and if there was any difference in performance. To benchmark imSim we ran a full focal plane simulation of all 189 LSSTCam sensors for a single visit. The field drawn contains roughly 40,000 objects in each sensor and is shown in Figure [1]. 

Open

Figure 1: The full set of 189 output images from the imSim LSSTCam simulation used in the benchmarks. The only processing on the output .fits files has been scaling to make the objects visible. Note the vignetting towards the edges of the image.

Single core performance results were very impressive. On Perlmutter, running the same simulation on the same field takes a mean time of 3790 seconds. Isambard 3 was able to generate the image on each sensor in a mean of 1900 seconds. With so few stars, which are the only objects that are bright enough to be drawn with FFTs rather than photon shooting, performance across the three FFT libraries was found to be broadly identical.  

GalSim is actually GPU accelerated in the C++ silicon sensor model, work previously carried out by James Perry. Here at EPCC in Edinburgh, we have two Grace-Hopper superchip nodes – each with only one Grace CPU, but alongside it a Hopper H100 GPU. Compiling and running the same benchmark on one of these nodes, and using GPU acceleration, improved the performance even further to the point that a single sensor could be drawn in a mean time of 860 seconds. The distribution of times to draw the sensors across the three systems benchmarked is given in Figure [2].

Open

Figure 2: The times required to draw the sensors in imSim are given here in box-and-whisker plots for the three systems. The Isambard 3 results are broken down by the FFT backend used, which, as can be seen, had essentially no effect.

The memory on each node hindered the draw it took to draw all 189 sensors on Isambard 3. GalSim spawns tasks to draw each sensor in parallel, but, with only 240 GB on a node, we were not able to run with all 144 cores and instead had to underpopulate to 72. In contrast, Perlmutter’s 512 GB of on-node memory means it can run with 128 tasks across all 128 cores. As such, the mean time to draw the full set of images for 189 sensors was measured to be 7240 seconds on Perlmutter but not much less at 6000 seconds on Isambard 3. On the Grace-Hopper node at EPCC, placing 72 tasks on the CPU and making use of the GPU would require – in an ideal world – perhaps as little as 2600 seconds, however there are complications in accessing the GPU in parallel meaning that currently it has to be run in serial. Testing whether a full parallel run is possible using the NVIDIA MPS tool is on the to-do list. 

The project and our access to Isambard 3 ended in January of this year. We were very impressed with how easy it was to set up and install all the software we needed, the maturity of the new Cray environment on the hardware, and the per-core performance of the Grace CPUs. We were also able to share these results in a talk at the Isambard Day event in March, which Ritchie Sommerville, Deputy Director at EPCC, kindly presented on behalf of LSST:UK. As far as imSim is concerned, we would be able to get to work tomorrow if a very sudden migration to a new platform were forced on us! 

 @William Lucas and @George Beckett

LSST:UK Project Leader position

The LSST:UK Consortium is inviting applications for the role of Project Leader as the term of office of the current Leader, Bob Mann, is coming to an end. This key position is funded at a level of 0.2 FTE and that indicates the level of time commitment the position requires. Note that the current Leader is permitted to apply, i.e. essentially for re-appointment. It is available from ~October 2025, with an initial appointment for three years, although the position and its funding are expected to continue beyond that date.

The application process will run in two phases:

Phase 1: submission of letters of intent

Any candidates interested in being considered for the Project Leader role are invited to submit a letter of intent via email to Mike Watson (email: mgw@le.ac.uk) by 17.00 on 20 June 2025. Letters of intent need only contain applicants name, institution and position.

In the second phase, full applications will be solicited (see webpage for details).

Informal questions from potential applicants may be addressed to any of the following:

• Mike Watson (mgw@le.ac.uk): LSST:UK Consortium Board Chair

• Bob Mann (rgm@roe.ac.uk): current LSST:UK Project Leader

• Graham Smith (gps@star.sr.bham.ac.uk): LSST:UK Project Scientist. 

The further details and the role description for the Project Leader appears on this webpage.

@Mike Watson

Unbiased galaxy clustering measurements directly from catalogues

Open

Comparison of the map-based and catalogue-based approaches to measuring angular power spectra when applied to a sample of galaxies from the Hyper Suprime-Cam Subaru Strategic Program. The bottom x-axis shows the multipole moment, which approximately scales inversely with the angular separation between galaxies as shown in the top axis. The y-axis thus represents the clustering strength at a given angular scale. Blue circles show the results from the map-based approach, having removed contributions from systematic effects using mode deprojection. Red triangles show the same for the catalogue-based approach, including our new adaptation of mode deprojection to this framework. Translucent red triangles show the catalogue-based measurements if these systematic effects are not accounted for, which has a noticeable impact at low multipoles (large scales). The map-based and catalogue-based approaches are consistent up to multipoles of ~2000, beyond which pixelisation effects are expected to bias the map-based measurements; this regime depends on the resolution of the maps used, and is highlighted by the shaded region for the map used here (resolution ~ 3.4 arcmin/pixel).

Galaxy clustering is an important tool for understanding the large-scale structure (LSS) of the Universe, as galaxies are biased tracers of the underlying dark matter distribution. By measuring the positions of billions of galaxies across the Southern hemisphere sky, LSST will thus obtain tight constraints on the amount and distribution of dark matter in the Universe.

A useful summary statistic for galaxy clustering is the angular power spectrum, which effectively quantifies the amount of clustering that occurs at a given angular scale (see figure). These power spectra are typically computed by first constructing maps showing the local galaxy overdensity, which involves binning galaxies into pixels on the map and comparing to the average density across the map. However, this introduces a bias at scales comparable to, or smaller than, the pixel size of the maps. Probing the LSS on small scales therefore requires maps of sufficiently high resolution, while also being low-resolution enough to avoid being dominated by Poisson noise. The former condition can be computationally expensive, and the latter cannot always be met – e.g. if a population of galaxies is sparsely distributed. This is a hindrance to any science reliant on constraining the gravity-dominated, nonlinear regime of cosmic structure formation.

Meanwhile, galaxy clustering measurements are also hampered by systematic effects. These include observational or instrumental phenomena that cause the observed galaxy density to fluctuate in ways that are not cosmological in origin (e.g. the attenuation of light from distant objects by dust in the Milky way), thereby biasing the clustering measurements. These effects are typically dominant on large scales, and the development of optimal mitigation techniques is an active area of research in cosmology.

Obtaining accurate, unbiased clustering measurements therefore requires overcoming the difficulties of probing both the largest and smallest scales. In recent work, Wolz et al. (2025) developed a method of computing angular power spectra which bypasses the need for pixelised maps, instead computing the power spectra directly from the discrete positions of the individual sources. This circumvents the biases introduced by pixelisation at small scales.

Now, we have extended this catalogue-based framework to include systematics mitigation, by adapting an existing technique implemented in the map-based approach. This technique, known as mode deprojection (or template deprojection), involves assuming that each systematic effect has a linear contribution to the observed galaxy density, then determining best-fit linear coefficients for each systematic and using those to remove their contribution from the measured signal. We have implemented this new method in the power spectrum estimation code NaMaster, and have verified that it gives results consistent with the map-based version by applying it to real data from the Hyper Suprime-Cam Subaru Strategic Program (HSC-SSP; see figure), as well as to simulated data.

This implementation will be described in more technical detail in Cornish et al. (in prep). We are also in the process of incorporating it into the LSST:DESC pipeline for joint galaxy clustering and weak lensing analysis, TXPipe. With this update, it will be possible not only to perform clustering analysis at small angular scales without the need for making high-resolution maps (thereby saving on computing resources and avoiding pixelisation effects), but also to mitigate the systematics that plague clustering measurements at large scales. This will extend the regime of angular scales that LSST will be able to reliably probe for galaxy clustering, and is an important step towards obtaining precise, unbiased constraints on the dark matter distribution in the Universe.

@Thomas Cornish and @David Alonso

Final call for nominations for Executive Group

The March issue featured a news story about the forthcoming LSST:UK elections for Executive Group. While the LSST:UK Board elects the main officers of the consortium, anyone is free to nominate for those roles and self-nominations are welcome. Exec Group members may serve for more than one term, i.e. re-election is allowed. So could I ask you to consider carefully if you know of anyone, or would yourself be willing to serve as a member of the Executive Group.  Please keep in mind the consortium’s commitment to equality and diversity. Three posts, running for 3 years, are available in this election.

The role of the Executive Group is outlined in our governance document. There are a few limitations on who can be a member of the Executive Group, designed to maintain separation of powers.  Thus Local Managers (i.e. academics in universities funded through the STFC grant to fund developers) may not be Executive Group members.  There are also limitations on Board members who are also Executive Group members voting on the Board.  If you need more information, please feel free to contact @Mike Watson.

@Mike Watson

Rubin at Cheltenham Science Festival

The Rubin Observatory is the focus of a public event at Cheltenham Science Festival on 7 June 2025. Cosmologist Andrew Pontzen will be joined by astrophysicist Hiranya Peiris and @Chris Lintott to discuss Rubin’s plans to create a high-definition time-lapse movie of the Universe.

Find out more and book tickets for the event

@Eleanor O'Kane

Open

 

LSST:UK communications update

Shape how the public understands LSST:UK and Rubin

As we receive more interest from the UK media as excitement around Rubin builds, I’d like to hear from anyone who is happy to speak to the media about their area of science and work in LSST:UK.

Journalists rely on experts to help shape their stories and, by being involved, you can influence what the public understands about astronomy and the impact of LSST:UK. If you’re curious to know more, please email Eleanor O’Kane, Communications Officer: eokane@roe.ac.uk

Video resource: Rubin Seminar for Science Writers

If you receive interest from members of the press (or anyone else curious about Rubin), the Seminar for Science Writers from Rubin is an excellent resource for non-specialists who wants to know more about the project. As an accessible guide to Rubin’s groundbreaking aims and achievements, the one-hour session was held at the 245th meeting of the American Astronomical Society in January. The recording is a useful counterpart to the Rubin Media Kit, which was highlighted in the March issue.

Watch the Seminar for Science Writers on YouTube

 

@Eleanor O'Kane

Leadership positions held by LSST:UK members

Here is the latest list of significant leadership positions held by members of the LSST:UK consortium in the project and international Science Collaborations. If you are aware of any corrections or additions, please contact the LSST:UK Project Managers (@George Beckett and @Terry Sloan: lusc_pm@mlist.is.ed.ac.uk).

Forthcoming meetings of interest

Dates, locations and links… The current list of forthcoming meetings is always available on the Relevant Meetings page. You may also wish to check information held on the LSST organisation website LSST-organised events and the LSST Corporation website.