Résumé
We perform a joint analysis of the power spectrum and the bispectrum of the CMB temperature and polarization anisotropies to improve the constraints on isocurvature modes. We construct joint likelihoods, both for the existing Planck data, and to make forecasts for the future LiteBIRD and CMB-S4 experiments. We assume a general two-field inflation model with five free parameters, leading to one isocurvature mode (which can be CDM density, neutrino density or neutrino velocity) arbitrarily correlated with the adiabatic mode. We theoretically assess in which cases (of detecting and/or fixing parameters) improvements can be expected, to guide our subsequent numerical analyses. We find that for Planck, which detected neither isocurvature modes nor primordial non-Gaussianity, the joint analysis does not improve the constraints in the general case. However, if we fix additional parameters in the model, the improvements can be highly significant depending on the chosen parameter values. For LiteBIRD+CMB-S4 we study in which regions of parameter space compatible with the Planck results the joint analysis will improve the constraints or the significance of a detection. We find that, while for CDM isocurvature this region is very small, for the neutrino isocurvature modes it is much larger. In particular for neutrino velocity it can be about half of the Planck-allowed region, where the joint analysis reduces the isocurvature error bars by up to 70%. In addition the joint analysis can also improve the error bars of some of the standard cosmological parameters, by up to 30% for θMC for example, by breaking the degeneracies with the correlation parameter between adiabatic and isocurvature modes.