Dear weak lensers and other interested parties,
The LSST CCB has recently approved a change to one of the project’s two main requirements that cover shape measurement for weak lensing. The old requirement was very much targeted as a specific kind of measurement (forward modeling) that no longer seems to be the way forward for shear estimation (more precisely, it’s at least not the only way forward). The new requirement attempts to be more agnostic to the actual methodology, and instead reflects the way shear algorithms are evaluated today. The new text is:
Specification: The Object catalog shall include shape measurements that permit the recovery of constant applied shears between 0.01 and 0.05 from an ensemble of isolated galaxies, with a multiplicative bias less than 0.001 and additive biases less than 0.0001.
Discussion: See e.g. Mandelbaum et al (2014) for a description of shear recovery biases. While LSST is responsible for performing these pixel-level measurements, estimating shear and verifying shear recovery accuracy under realistic conditions is considered a science activity and is hence out of scope. This requirement is thus expected to be met by providing an implementation of an externally-developed algorithm that has been demonstrated to meet these specifications. At least one such algorithm already exists (Sheldon & Huff, 2017).
Careful readers will note that the context of the specification is somewhat artificial (“isolated galaxies”, “constant applied shear”), and the numerical thresholds themselves reflect what state-of-the-art codes (specifically Metacalibration; see above reference) can do today under those conditions, rather than a top-down analysis starting from the systematic error budget in specific science analyses. That matches the spirit of the old requirement, and the understanding that I think most people in both the Project and the Science Collaborations had of where the boundary has always been for shear estimation: DM will run either the best community-developed pixel-level code or something no worse, and the community is responsible for deriving accurate shears from its outputs under realistic conditions. That’s still our plan, and this change should be interpreted as an update to and generalization of our understanding of what that pixel-level code will produce.
At the beginning of this post I noted that this was one of two requirements that very directly impact shape measurements. The other concerns PSF residual ellipticity correlations (section 3.3.3.3 of the SRD), and is not affected by this change, though meeting it is probably a necessary (but definitely not sufficient) condition for meeting the new requirement.