2016MNRAS.459.3551B

Understanding the sources of instrumental systematic noise is a must to improve the design of future spectrographs. In this study, we alternated observations of the well-suited pair of M-stars GJ 725A+B to delve into the sub-night High Accuracy Radial Velocity Planet Searcher for the Northern hemisphere (HARPS-N) response. Besides the possible presence of a low-mass planet orbiting GJ 725B, our observations reveal changes in the spectral energy distribution (SED) correlated with measurements of the width of the instrumental line profile and, to a lower degree, with the Doppler measurements. To study the origin of these effects, we searched for correlations among several quantities defined and measured on the spectra and on the acquisition images. We find that the changes in apparent SED are very likely related to flux losses at the fibre input. Further tests indicate that such flux losses do not seriously affect the shape of the instrumental point spread function of HARPS-N, but identify an inefficient fitting of the continuum as the most likely source of the systematic variability observed in the full width at half-maximum. This index, accounting for the HARPS-N cross-correlation profiles width, is often used to decorrelate Doppler time series. We show that the Doppler measurement obtained by a parametric least-squares fitting of the spectrum accounting for continuum variability is insensitive to changes in the slope of the SED, suggesting that forward modelling techniques to measure moments of the line profile are the optimal way to achieve higher accuracy. Remaining residual variability at ∼1 m/s suggests that for M-stars Doppler surveys the current noise floor still has an instrumental origin.