Hugh McDougall Astro

Importance Sampling (recycling posterior samples) has a reputation as crude brute force, but this underappreciated method can be a surprisingly powerful tool in many practical cases. Find my new article about the importance of importance sampling here: 🔭
hughmcdougall.github.io/blog/03_stat...

Call me old fashioned but sometimes my python IDE questions my grammar choices and frankly I think it should mind its own business

Everytime I read LVK my brain tells me it stands for Ligo Virgo Kirgo

🔭

🔭 Excited to be starting my 3 month visiting research placement at @monashastro.bsky.social working on some numerical specifics in GW population fitting!

🔭 It's called bluesky because of Rayleigh Scattering.

🔭
"I don't know why so many grad students have imposter syndrome. Where does the fear of failure come from?"

Meanwhile, in the arXiv submission form:

Watch the solar plasma curve along magnetic field lines like iron filings near a bar magnet. #stellarastro 🧪🔭

This work was funded in part by NSF.

You might have heard of the cool chaotic patterns you get from the logistic map, including the islands of stability that appear at certain values of 'r'.

Here's a fun animation of a different recursion relation that bounds the output as we change the itteration gain, wrapped around a polar axis.

Wild that Adobe's forced AI features means that you can open a program called "pdf reader", and then be inflicted with a popup asking if you'd like to not read a pdf

🔭 You've got two sets of data: can you combine them, or are the in tension? How do you know?

Answer: the Bayesian suspiciousness statistic! This blog (+examples) shows how you can consistently tell if two data sets are in tension with a full Bayes approach!

hughmcdougall.github.io/blog/03_stat...

Actual slide from a public outreach talk last week

No AI image will ever match the charm of a poorly photoshopped stock image with the watermark still in it

🔭If you work in stats you've probably heard of nested sampling: MCMCs fancy cousin. Enough people have asked me how it works that I've written an article (including examples!) explaining:
1. What NS is
2. Why we need it over MCMC
3. How to know if it's working
hughmcdougall.github.io/blog/03_stat...

I'm adding new features every week and would love to hear from AGN scientists who would like to see new tools added. Reach out by email or BlueSky, or check my website out at hughmcdougall.github.io.

LITMUS is built for measuring lags in industrial scale RM surveys like OzDES, but its extendable modelling means you can adapt it to _any_ delay estimation problem (Quasar lensing delay, accretion disk RM, even BLR transfer function recovery). For you traditionalists: it also has nested sampling!

LITMUS manages this by using JAX's autodiff for a new approach the Bayesian fitting- using the Laplace Approximation (or SVI) to Slice the posterior into Gaussians. This is a new technique for RM, and gives us all the good Bayesian stuff while fitting lags in a fraction of the time of older tools

LITMUS offers AGN scientists a suite of new stats tools to sort the signal from the noise, including the established Gaussian Process model for AGN _and_ a flexible framework that lets you write your own more advanced physics model if you need!

If you've never heard of RM before - it's a trick where we use time series data to infer the geometry of Quasars. Like the gap in lighting and thunder telling us how far away a storm is, we can listen for the way light "echoes" around the AGN as a ruler to map their size.

Not only does it massively improve on earlier MCMC based RM tools tendency to over-report half yearly lags (180 days, 540 days etc), it also, for the first time in RM, offers Bayesian model evidences for proper model comparison! Check out the docs example at hughmcdougall.github.io/litmus/examp...

LITMUS is a new tool purpose built to solve the “aliasing” problem that has plagued reverberation mapping in high redshift surveys like OzDES and SDSS – using a new approach to the Bayesian fitting to properly handle the window function and to principled false positive rejection!

Extremely excited to announce the first paper / project of my PhD: LITMUS – a new Bayesian framework for reverberation mapping lag recovery! The paper is on arXiv at arxiv.org/abs/2505.09832, or you can find the code at hughmcdougall.github.io/litmus/ and the docs at hughmcdougall.github.io/litmus/

For my first post on bluesky, I present to the world an astronomer friendly guide to taking your first steps with #NumPyro:
hughmcdougall.github.io/blog/02_nump...