The day in conversation for 2026-11-12

discussion

Transcript

The discussion: Vera: So, looking at that 71 thousand calibrated NIR images with a root mean square astrometric error of 10 mas, it feels like a solid baseline for DR1, but I’m curious how much of that residual uncertainty is truly systematic versus random noise? Jocelyn: That systematic component is the nightmare. When we talk about persistence issues in the astrometric calibration, are we talking about instrumental drifts over long exposures or something inherent to how the data was downlinked from L2? Subrahmanyan: I think it's likely a combination of both, but if you look at the paper on manifold alignment, those variance shrinkage problems in galaxy property inference—like shape measurements—that directly feed into how noisy our positional data is interpreted. How does that calibration uncertainty specifically propagate into the statistical manifold fitting? Vera: If the astrometry is off by ten milliarcseconds across a field, that translates to significant angular smearing when trying to constrain subtle morphological features in those deep surveys. Did we run simulations on how much variance shrinkage occurs if we use a standard error model versus incorporating that manifold alignment correction? Jocelyn: We did run preliminary tests, but I think the key difference is whether the alignment method assumes Gaussian noise or something more complex related to the detector artifacts themselves. If it’s not Gaussian, applying a simple shrinkage factor might actually exacerbate the bias. Subrahmanyan: That's where my focus shifts slightly; for cosmology, we often

More episodes

← Home