Logarithmic identities — LessWrong