Project Nayuki

Binary array set


The binary array set is a data structure that allows adding elements in amortized \(O(1)\) time and testing set membership in worst-case \(O((\log n)^2)\) time. The set is orderless and the data structure does not support removing elements.

It’s useful for applications that need to add elements to a set and test for duplicates quickly but don’t need to remove items (though the entire set can be discarded). One application of this structure is to remember the set of nodes already visited in a breadth-first search of a graph, which only grows as time goes on.

Also this data structure is space-efficient, having only \(O(\log n)\) overhead for pointers, unlike balanced tree structures that use \(O(n)\) extra space for pointers. So this is a fast set data structure that occupies practically the same amount of space as a flat array. This is especially helpful in languages like Java, where the virtual machine gives each object (such as a tree node) a significant amount of storage overhead.

Internal details

How this data structure works is that it has a sequence of \(\lfloor \log_2 n \rfloor + 1\) slots, where the slot at index \(k\) (0-based counting) is either null or an ascending-sorted array of length \(2^k\). For example, this is one way of how the abstract set {1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13} is represented as a binary array set:

Slot 0: [5] (length 1)
Slot 1: Null
Slot 2: [2, 3, 9, 13] (length 4)
Slot 3: [1, 4, 6, 7, 8, 10, 11, 12] (length 8)

Note that the size of the set is 13, and its binary representation is 11012. The binary representation corresponds with which slots are filled. Also note that the union of all the arrays in the binary array set is exactly equal to the set of elements being represented, no more, no less, no duplicates.

To test if a given value is an member of the set, we simply binary search in each array (in the non-null slots) and return whether a result was found.

To add a new element, we wrap the element in a new array of length 1. If slot 0 is empty, we simply put down the new array. Otherwise we merge the new array with the array at slot 1, in ascending order, to form a new sorted array of length 2. We check if slot 1 is empty; we repeat this merge/check process on increasing slot indexes until we find an empty slot.

Source code


The class implements the java.util.Set interface, so it can be used as a drop-in replacement of HashSet or TreeSet as long as remove() is not called. The iterator is not fail-fast on concurrent modifications. The maximum set size is Integer.MAX_VALUE (i.e. \(2^{31} - 1\)).


This implementation only supports the methods {add(), clear()}, built-in functions {len(), iter()}, and operator in – so it only supports a fraction of the methods that the built-in set type has. Compatible with Python 2 and 3.


The class is modeled after the Java version, supporting the methods size(), add(), contains(), and clear(). It has far fewer features than std::set. Requires C++11.

License: MIT (open source)