A new analysis of the false positive rate of a Bloom filter

A Bloom filter is a space-efficient data structure used for probabilistic set membership testing. When testing an object for set membership, a Bloom filter may give a false positive. The analysis of the false positive rate is a key to understanding the Bloom filter and applications that use it. We show experimentally that the classic analysis for false positive rate is wrong. We formally derive a correct formula using a balls-and-bins model and show how to numerically compute the new, correct formula in a stable manner. We also prove that the new formula always results in a predicted greater false positive rate than the classic formula. This correct formula is numerically compared to the classic formula for relative error – for a small Bloom filter the prediction of false positive rate will be in error when the classic formula is used.