profannawatts.bsky.social
Professor of Astrophysics @ Univ. of Amsterdam. Neutron stars, pulsars, dense matter, X-ray astronomy. Mom, Migrant. "Particles in stars blah blah". Does science with NICER on the International Space Station and schemes for EVEN BIGGER SPACE TELESCOPES.
457 posts
8,481 followers
410 following
Regular Contributor
Active Commenter
comment in response to
post
G...grading?
comment in response to
post
I'm just going to wait the crisis out in the library. With my project paperwork and grading safely up to date, I'll be last on the hit list.
comment in response to
post
Those poor fools. They have no idea what is coming in the next budget round....
comment in response to
post
I only have a tiny budget line. If I keep very quiet maybe no one will notice it.
comment in response to
post
The coffee budget has been obliterated. Chaos ensues.
comment in response to
post
Omg how do you not know this, Bob? IT'S CARNAGE OUT HERE!!
comment in response to
post
Dude. Why do you think we look this frazzled? YOU HAVE TO STOP NAPPING DURING STAFF MEETINGS.
comment in response to
post
It was absolutely brilliant, and 💯 approved by the Obsessive PJ fan in the house. Can definitely recommend!
comment in response to
post
The Github issue was finally closed in 2024. 😅
comment in response to
post
This work was funded by both NWO (as part of an ENW-XL) and the ERC (as part of a Consolidator Grant), and we've made use of time on the national supercomputer Snellius. So big shout-out to those folks/facilities!!
comment in response to
post
And now that we have Ultranest up and running as a standard part of X-PSI we can also cross-check this for future analyses (and are doing so). Ultranest is more computationally expensive than Multinest, but it's fantastic to have to this to cross-check one potential source of systematic error.
comment in response to
post
For the tests that we have done (and one can ALWAYS do more tests!!) both samplers appear to recover parameters reliably - and both deliver consistent credible intervals for the radius of PSR J0740+6620. Which gives us more confidence in our results.
comment in response to
post
So Mariska bravely jumped in to implement Ultranest with X-PSI, and to run a series of analyses using both synthetic and real data for both samplers to investigate performance.
comment in response to
post
There are a lot of reasons why this might be the case, but the two groups also use different samplers. And we know from the excellent work done by our gravitational wave colleagues (see e.g. the Bilby paper by Ashton et al. 2019) that this can affect results.
comment in response to
post
Finally, in the fifth half of 2021 🫣 we got to work.
Part of the renewed motivation was that in our analysis of the massive pulsar PSR J0740+6620, X-PSI (Salmi et al. 2024) typically finds narrower credible intervals than those reported by the Maryland team (Dittmann et al. 2024).
comment in response to
post
Time passed.
comment in response to
post
Ultranest (johannesbuchner.github.io/UltraNest/in...) has different smart ways of doing things that avoid some of the known pitfalls of Multinest.
We agreed that it would be great to try it out, and brave statements were made that we would get to it in the second half of 2021.
comment in response to
post
But it's also good to try out completely different samplers to see if a completely different method delivers the same results! And back in 2021 @johannesbuchner.bsky.social suggested in an X-PSI github issue that we try using the sampler Ultranest as an alternative.
comment in response to
post
Different samplers have different ways of exploring the space, and various settings that can be tweaked. Tuning these to ensure accurate parameter recovery takes time and effort, and we've done a lot of work on this with Multinest for our application (see e.g. Vinciguerra et al. 2023).
comment in response to
post
We also tried emcee back when we were developing X-PSI but it struggled a bit with the very multi-modal spaces we are typically exploring.
comment in response to
post
We rely on off-the-shelf samplers to sift through the large parameter space and deliver posterior probability distributions on all of our parameters of interest. For our simulation code X-PSI, results published to date have relied on the well-used sampler Multinest.
comment in response to
post
Pulse profile modeling, the relativistic ray-tracing technique that we use to model X-ray emission from rotation-powered millisecond pulsars (mostly with NICER data) - hence measuring their mass, radius and the properties of the hot magnetic poles - involves Bayesian inference.
comment in response to
post
A swashbuckling tale of samplers, sizing up neutron stars correctly and THE LONGEST-LIVED GITHUB ISSUE EVER.
comment in response to
post
What?!
comment in response to
post
Meow Tower. Very demanding cats + nonograms.
comment in response to
post
They did it in class so I'm hoping that wasn't an issue 🤣
comment in response to
post
So disappointing: 👩🔬
"Fermilab suspended its DEI office and women in engineering group in January. Meanwhile, the Fermilab LBGTQ+ group, spectrum, was ordered to cease all activities and its mailing list deleted. Even the rainbow Pride flag was removed from the lab’s iconic Wilson Hall."
comment in response to
post
Also apparently for Group 1/2 VO (high school) level. Yikes. Feels like something went awry with the qualifying test design this year.....
Or our kids are just much smarter than 12 months ago 😅
comment in response to
post
Arial every time!