A false positive is an incorrect positive result: a lookup reports that an absent key is present. In a Bloom filter, a negative result is certain, but a positive result requires verification because several keys may set the same bits. section 4.4.5 section 4.8.2
Collisions create false positives. With four bits and two hash functions, inserting values 1 and 2 can set bits 0 and 1; a lookup for absent value 3 returns true when its hashes point to those already-set positions. section 4.8.2
Handle positive results according to their cost and required accuracy:
- Double-check a positive Bloom-filter result against the authoritative server or dictionary when correctness matters. section 4.6.1 section 4.7.3
- Tune the filter’s memory allocation and number of hash functions for the expected capacity and tolerated false-positive probability. section 4.6.5
- Accept false positives when their consequence is limited, such as a small reduction in cache performance. section 4.7.1
Email contact lookup
A local Bloom filter can reject contacts that are definitely absent. When it returns true, the application asks the server, because the local result may be a false positive. section 4.6.1
The false-positive rate rises as the filter holds more elements or uses less memory. A constructor can estimate the rate from the current element count, bit count, and number of hash functions; exceeding the planned capacity invalidates the expected precision. section 4.6.5 section 4.6.8

























