[[precision-step]]
=== `precision_step`

Most <<number,numeric>> datatypes index extra terms representing numeric
ranges for each number to make <<query-dsl-range-query,`range` queries>>
faster.  For instance, this `range` query:

[source,js]
--------------------------------------------------
  "range": {
    "number": {
      "gte": 0
      "lte": 321
    }
  }
--------------------------------------------------

might be executed internally as a <<query-dsl-terms-query,`terms` query>> that
looks something like this:

[source,js]
--------------------------------------------------
  "terms": {
    "number": [
      "0-255",
      "256-319"
      "320",
      "321"
    ]
  }
--------------------------------------------------

These extra terms greatly reduce the number of terms that have to be examined,
at the cost of increased disk space.

The default value for `precision_step` depends on the `type` of the numeric field:

[horizontal]
`long`, `double`, `date`, `ip`::  `16` (3 extra terms)
`integer`, `float`, `short`::     `8` (3 extra terms)
`byte`::                          `2147483647` (0 extra terms)
`token_count`::                   `32` (0 extra terms)

The value of the `precision_step` setting indicates the number of bits that
should be compressed into an extra term.  A `long` value consists of 64 bits,
so a `precision_step` of 16 results in the following terms:

[horizontal]
Bits 0-15:: `value & 1111111111111111 0000000000000000 0000000000000000 0000000000000000`
Bits 0-31:: `value & 1111111111111111 1111111111111111 0000000000000000 0000000000000000`
Bits 0-47:: `value & 1111111111111111 1111111111111111 1111111111111111 0000000000000000`
Bits 0-63:: `value`