Persistent data structures preserve all historical versions of themselves, unlike ephemeral structures that discard old state on update. Three categories exist: partial persistence (read-only history, only latest is mutable), full persistence (all versions readable and writable), and confluent persistence (historical versions can be merged). Real-world applications include functional programming languages (Haskell, Clojure, Scala), computational geometry (point location via persistent red-black trees), and text editors (undo/redo). Implementation techniques for partial persistence include Copy-on-Write (naive deep copy), the Fat Node method (each node stores multiple versioned values), the Node-Copying method (fixed-size nodes with pointers to newer copies), and the Path-Copying method (copies only the path from root to modified node). Memory management via reference counting or mark-and-sweep is essential to prevent leaks.
