Beyond Fisher forecasting for cosmology

The planning and design of future experiments rely heavily on forecasting to assess the potential scientific value provided by a hypothetical set measurements. Fisher information matrix, due its convenient properties low computational cost, provides an especially useful tool. However, matrix only reasonable approximation true likelihood when data are nearly Gaussian distributed observables have linear dependence parameters interest. Also, techniques alone cannot be used their own validity. Thorough sampling exact or mock can definitively determine whether forecast is valid, though such often prohibitively expensive. We propose simple test, based Derivative Approximation for LIkelihoods (DALI) technique, good likelihood. show that becomes poor in regions where two-dimensional slices level surfaces DALI differ from demonstrate our method accurately predicts situations which deviates various cosmological models several combinations, with modest increase cost compared standard forecasts.