Speaker:	Nicolas Payot (University of Montreal, Mila, Ciela)　and Conghao Zhou (UC Santa Cruz)
Title:	Scaling Up Cosmological Simulations with Differentiable Particle Mesh Models ; No Cluster Left Behind: Multiwavelength Cluster Analysis with All the Data
Date (JST):	Tue, Oct 06, 2026, 11:00 - 12:00
Place:	Seminar Room A
Abstract:	Speaker 1: Nicolas Payot (University of Montreal, Mila, Ciela) Scaling Up Cosmological Simulations with Differentiable Particle Mesh Models New generations of telescopes are delivering increasingly detailed observations and larger datasets, creating a need for cosmological simulations that combine scale, accuracy, and computational efficiency. In this seminar, I will present my work on fast and differentiable cosmological forward models for large-scale structure. I will introduce PM++, a particle-mesh simulator that distributes both the evolving particles and gravitational calculations across multiple GPUs, enabling cosmological simulations with billions of particles while retaining end-to-end gradients. I will discuss its parallelization strategy, computational scaling, and applications to cosmological inference. I will then present ongoing work on learned corrections to approximate gravitational dynamics, designed to improve small-scale accuracy while preserving the computational efficiency of particle-mesh simulations. Together, these developments aim to make large, accurate, and differentiable cosmological simulations practical for inference with current and future surveys. Speaker 2: Conghao Zhou (UC Santa Cruz) No Cluster Left Behind: Multiwavelength Cluster Analysis with All the Data Cluster abundance is among the most powerful cosmological probes, but its constraining power is currently limited by systematic uncertainties in mass calibration. Optically selected samples in particular suffer from line-of-sight projection effects that bias both richness and lensing measurements. Jointly modeling clusters observed across multiple wavelengths offers a path to self-calibrating these systematics ― yet previous multiwavelength analyses have typically discarded clusters lacking tSZ or X-ray counterparts, sacrificing much of the sample's statistical power. In this talk, I will present an analysis framework that retains every cluster detected in at least one survey: no cluster left behind. Applying it to Dark Energy Survey and South Pole Telescope data, I will show joint constraints on cluster mass calibration and optical projection effects. I will conclude by looking forward to multiwavelength cluster cosmology in the era of Rubin, Roman, SPT-3G, and eROSITA.

← Back