On the number of arithmetic formulas