When functions with $n$ input bits and $m$ output bits are used for post-processing the output of a physical random number generator, which is assumed to produce statistically independent bits with a fixed bias $\epsilon$, the probabilities of the $2^m$ outputs can be represented as polynomials in $\epsilon$. For a good post-processing function all low powers of $\epsilon$ should have coefficients of zero. K. Suzuki and T. Iwata showed in the paper 
"Bounds on Fixed Input/Output Length Post-Processing Functions for Biased Physical Random Number Generators"
presented at SAC 2008  for most cases with $1 \leq m \leq n \leq 16$ how many low powers of $\epsilon$ can be maximally eliminated, but there were 13 cases remaining open. This paper solves these 13 cases.