Depth-first search


Depth-first search is an algorithm for traversing or searching tree or graph data structures. The algorithm starts at the root node and explores as far as possible along each branch before backtracking.
A version of depth-first search was investigated in the 19th century by French mathematician Charles Pierre Trémaux as a strategy for solving mazes.

Properties

The time and space analysis of DFS differs according to its application area. In theoretical computer science, DFS is typically used to traverse an entire graph, and takes time, linear in the size of the graph. In these applications it also uses space in the worst case to store the stack of vertices on the current search path as well as the set of already-visited vertices. Thus, in this setting, the time and space bounds are the same as for breadth-first search and the choice of which of these two algorithms to use depends less on their complexity and more on the different properties of the vertex orderings the two algorithms produce.
For applications of DFS in relation to specific domains, such as searching for solutions in artificial intelligence or web-crawling, the graph to be traversed is often either too large to visit in its entirety or infinite. In such cases, search is only performed to a limited depth; due to limited resources, such as memory or disk space, one typically does not use data structures to keep track of the set of all previously visited vertices. When search is performed to a limited depth, the time is still linear in terms of the number of expanded vertices and edges but the space complexity of this variant of DFS is only proportional to the depth limit, and as a result, is much smaller than the space needed for searching to the same depth using breadth-first search. For such applications, DFS also lends itself much better to heuristic methods for choosing a likely-looking branch. When an appropriate depth limit is not known a priori, iterative deepening depth-first search applies DFS repeatedly with a sequence of increasing limits. In the artificial intelligence mode of analysis, with a branching factor greater than one, iterative deepening increases the running time by only a constant factor over the case in which the correct depth limit is known due to the geometric growth of the number of nodes per level.
DFS may also be used to collect a sample of graph nodes. However, incomplete DFS, similarly to incomplete BFS, is biased towards nodes of high degree.

Example

For the following graph:
a depth-first search starting at A, assuming that the left edges in the shown graph are chosen before right edges, and assuming the search remembers previously visited nodes and will not repeat them, will visit the nodes in the following order: A, B, D, F, E, C, G. The edges traversed in this search form a Trémaux tree, a structure with important applications in graph theory.
Performing the same search without remembering previously visited nodes results in visiting nodes in the order A, B, D, F, E, A, B, D, F, E, etc. forever, caught in the A, B, D, F, E cycle and never reaching C or G.
Iterative deepening is one technique to avoid this infinite loop and would reach all nodes.

Output of a depth-first search

A convenient description of a depth-first search of a graph is in terms of a spanning tree of the vertices reached during the search. Based on this spanning tree, the edges of the original graph can be divided into three classes: forward edges, which point from a node of the tree to one of its descendants, back edges, which point from a node to one of its ancestors, and cross edges, which do neither. Sometimes tree edges, edges which belong to the spanning tree itself, are classified separately from forward edges. If the original graph is undirected then all of its edges are tree edges or back edges.

DFS ordering

An enumeration of the vertices of a graph is said to be a DFS ordering if it is the possible output of the application of DFS to this graph.
Let be a graph with vertices.
For be a list of distinct elements of, for, let be the greatest such that is a neighbor of, if such a exists, and be otherwise.
Let be an enumeration of the vertices of.
The enumeration is said to be a DFS ordering if, for all, is the vertex such that is maximal.
Recall that is the set of neighbors of. Equivalently, is a DFS ordering if, for all with, there exists a neighbor of such that.

Vertex orderings

It is also possible to use depth-first search to linearly order the vertices of a graph or tree. There are four possible ways of doing this:
For binary trees there is additionally in-ordering and reverse in-ordering.
For example, when searching the directed graph below beginning at node A, the sequence of traversals is either A B D B A C A or A C D C A B A. Note that repeat visits in the form of backtracking to a node, to check if it has still unvisited neighbors, are included here. Thus the possible preorderings are A B D C and A C D B, while the possible postorderings are D B C A and D C B A, and the possible reverse postorderings are A C B D and A B C D.
Reverse postordering produces a topological sorting of any directed acyclic graph. This ordering is also useful in control flow analysis as it often represents a natural linearization of the control flows. The graph above might represent the flow of control in the code fragment below, and it is natural to consider this code in the order A B C D or A C B D but not natural to use the order A B D C or A C D B.
if then else
D

Pseudocode

Input: A graph G and a vertex v of G
Output: All vertices reachable from v labeled as discovered
A recursive implementation of DFS:
procedure DFS is
label v as discovered
for all directed edges from v to w that are in G.adjacentEdges do
if vertex w is not labeled as discovered then
recursively call DFS
The order in which the vertices are discovered by this algorithm is called the lexicographic order.
A non-recursive implementation of DFS with worst-case space complexity, with the possibility of duplicate vertices on the stack:
procedure DFS_iterative is
let S be a stack
S.push
while S is not empty do
v = S.pop
if v is not labeled as discovered then
label v as discovered
for all edges from v to w in G.adjacentEdges do
S.push
These two variations of DFS visit the neighbors of each vertex in the opposite order from each other: the first neighbor of v visited by the recursive variation is the first one in the list of adjacent edges, while in the iterative variation the first visited neighbor is the last one in the list of adjacent edges. The recursive implementation will visit the nodes from the example graph in the following order: A, B, D, F, E, C, G. The non-recursive implementation will visit the nodes as: A, E, F, B, D, C, G.
The non-recursive implementation is similar to breadth-first search but differs from it in two ways:
  1. it uses a stack instead of a queue, and
  2. it delays checking whether a vertex has been discovered until the vertex is popped from the stack rather than making this check before adding the vertex.
If is a tree, replacing the queue of the breadth-first search algorithm with a stack will yield a depth-first search algorithm. For general graphs, replacing the stack of the iterative depth-first search implementation with a queue would also produce a breadth-first search algorithm, although a somewhat nonstandard one.
Another possible implementation of iterative depth-first search uses a stack of iterators of the list of neighbors of a node, instead of a stack of nodes. This yields the same traversal as recursive DFS.
procedure DFS_iterative is
let S be a stack
S.push
while S is not empty do
if S.peek.hasNext then
w = S.peek.next
if w is not labeled as discovered then
label w as discovered
S.push
else
S.pop

Applications

Algorithms that use depth-first search as a building block include:
The computational complexity of DFS was investigated by John Reif. More precisely, given a graph, let be the ordering computed by the standard recursive DFS algorithm. This ordering is called the lexicographic depth-first search ordering. John Reif considered the complexity of computing the lexicographic depth-first search ordering, given a graph and a source. A decision version of the problem is P-complete, meaning that it is "a nightmare for parallel processing".
A depth-first search ordering, can be computed by a randomized parallel algorithm in the complexity class RNC. As of 1997, it remained unknown whether a depth-first traversal could be constructed by a deterministic parallel algorithm, in the complexity class NC.