Logarithms invert exponentials — LessWrong