Anna Rosen, Ph.D.Computational Astrophysicist

Jaxstro · Observables

fluxax

Physical states → survey observables.

A differentiable Gaia and Rubin/LSST catalogue forward model — photometry, astrometry, selection, crowding, blending and extinction inside one gradient-safe likelihood — with a differentiable pixel renderer alongside it.

Readiness

Far along

Publications

Methods paperplanned

A survey does not measure a population; it measures whatever its selection let through. fluxax puts those effects — magnitude limits, crowding, blending, footprint, extinction, the parallax zero-point — inside a single differentiable likelihood, where each one can be shown to bias a recovered IMF slope when it is ignored, and to come back out when it is modelled.

The error models are checked against the published ones and against the official instrument curves, not against sky data. Every demonstration here uses synthetic populations with known answers; fitting real catalogues is informax’s job.

What's working

Six panels in two columns. Top row: a simulated g-band image of a crowded cluster, first as an expected source-limited scene and then as a nominal 30-second Rubin-like exposure with sky and noise. Middle row: maps of how much information each pixel carries about the IMF slope. Bottom row: the same for the binary mass-ratio slope. The information maps are concentrated on a sparse subset of stars, and differ between the two parameters.
PreliminaryWhich Pixels Know the Answer. This is what differentiable codes buy you. The same truth-known cluster is rendered as an idealised scene and as a nominal 30-second Rubin-like exposure with PSF, sky, Poisson and read noise (top). Differentiating that render with respect to the population parameters gives the Fisher information carried by every individual pixel — for the IMF slope (middle) and for the binary mass-ratio slope (bottom). Note that the two are not the same pixels: different physics is constrained by different parts of the image. You cannot make this map by running a simulation forward; it exists only because the whole pipeline carries gradients. Population-to-pixel differentiation is implemented; validated detector and survey transformations, and global recovery, are proposed work.
Three colour-magnitude diagrams of the same synthetic cluster. Left: ground truth in Rubin bands, singles and unresolved binaries both visible along the main sequence. Centre: the Gaia view, where faint stars fall below the detection limit and drop out. Right: the Rubin view, reaching deeper but saturating at the bright end and scattering near the faint limit.
PreliminaryWhat Gaia and Rubin Actually See. One synthetic population, three views. Ground truth in Rubin bands (left); the same stars as Gaia would record them, with everything faint falling out below the detection limit (centre); and as Rubin would record them, reaching far deeper but saturating at the bright end (right). Unresolved binaries sit systematically above the single-star sequence in every panel. Selection, saturation and blending are part of the forward model, not corrections applied afterwards.

What's coming

Methods paper planned. It will cover the differentiable path from physical state to Gaia and Rubin catalogue observables, including selection, crowding and noise.