Title: Baryonic effects in cosmic shear tomography: PCA parametrization and importance of extreme baryonic models

Abstract: Baryonic effects are amongst the most severe systematics to the tomographic analysis of weak lensing data which is the principal probe in many future generations of cosmological surveys like LSST, Euclid etc.. Modeling or parameterizing these effects is essential in order to extract valuable constraints on cosmological parameters. In a recent paper, Eifler et al. (2015) suggested a reduction technique for baryonic effects by conducting a principal component analysis (PCA) and removing the largest baryonic eigenmodes from the data. In this article, we conducted the investigation further and addressed two critical aspects. Firstly, we performed the analysis by separating the simulations into training and test sets, computing a minimal set of principle components from the training set and examining the fits on the test set. We found that using only four parameters, corresponding to the four largest eigenmodes of the training set, the test sets can be fitted thoroughly with an RMS $\sim 0.0011$. Secondly, we explored the significance of outliers, the most exotic/extreme baryonic scenarios, in this method. We found that excluding the outliers from the training set results in a relatively bad fit and degraded the RMS by nearly a factor of 3. Therefore, for a direct employment of this method to the tomographic analysis of the weak lensing data, the principle components should be derived from a training set that comprises adequately exotic but reasonable models such that the reality is included inside the parameter domain sampled by the training set. The baryonic effects can be parameterized as the coefficients of these principle components and should be marginalized over the cosmological parameter space.