摘要: New formulas are given for the minimax linear risk in estimating a functional of an unknown object from indirect data contaminated with random Gaussian noise. The cover variety loss functions, and do not require symmetry convex priori class. It is shown that affine rules within few percent even among nonlinear rules, functions. also difficulty estimation measured by modulus continuity to be estimated. method proof exposes correspondence between estimates statistical problem optimal algorithms theory recovery.