A 1 microhertz grid with shared phases gives false noise correlations on three ESPRESSO observing schedules

We tested a Fourier noise model on public ESPRESSO observing dates for HD 10700, HD 102365, and HD 304636. We used 109, 47, and 61 nightly dates, respectively. The test kept one set of phases across each full series. Frequencies ran from 0.1 microhertz to below 6 millihertz in 1 microhertz steps. We used a published solar alpha = 2 supergranulation spectrum with fixed parameters. The reference was its continuous integral over the same frequency limits. The grid forces a correlation of 0.809 at 11.574 days and minus one at 57.870 days. These correlations are absent from the reference. At a fixed signal period of 20 days, a generalized least-squares fit reported sine/cosine coefficient errors that were too small by factors of 2.64, 1.52, and 1.68. Under Gaussian reference noise, the nominal 95% joint ellipses had coverage of 35.8%, 73.1%, and 67.1%. The package gives exact covariance calculations, simulations, controls, figures, source data, and provenance. The tested shared-phase condition is an extension of a published recipe for observations within a night. This test does not assess that paper's planet fits. We did not fit the measured velocities or detect a planet. The result is a reproducible check for noise simulations used in planet searches.

radial velocity · ESPRESSO · supergranulation · Fourier simulation · covariance · confidence coverage · reproducible benchmark