Environmental Split Yields 2-3x Tighter f_NL Bounds in DESI LRG Forecasts

The paper derives the f_NL response for tracers conditioned on reconstructed dark-matter environment and validates it against separate-universe simulations. A simple linear bias model suffices; no assembly-bias parameters or halo formation time are required. In Quijote mocks the method matches direct power-spectrum inference within statistical errors.

Forecasts for the DESI LRG sample show the environmental split outperforms conventional single-tracer analyses by the same margin achieved when perfect halo-age information is assumed. Gains are smaller for lower-density samples because the environment split correlates more weakly with the underlying density field. An analytic model explains the reduced leverage.

Unlike mass- or age-based splits, the approach relies only on observable linear reconstruction, making it immediately applicable to photometric and spectroscopic surveys. The chief limitation is that the quoted factor-of-2–3 improvement is a forecast, not a measurement on real data.

Next steps include application to existing BOSS or DESI early data and joint analysis with CMB lensing to test residual systematics in the reconstruction.