the heap-based algorithm could be implemented like this:
init empty heap H of vertices sorted by dist
for each vertex v set v.visited = false, v.dist = infinity
set goal.dist = 0
add all vertices to H # O(N) amortized with Fibonacci heap
while H not empty # O(N)
v = extract-min H # O(log N) amortized with Fibonacci heap
set v.visited = true
# this inner loop iterates a *total* of O(E) times
for each neighbor vertex n of v with n.visited = false and n.dist > (v.dist + dist(v,n))
set n.next = v, n.dist = v.dist + dist(v,n)
decrease-key of n in H # O(1) amortized with Fibonacci heap
decrease-key
operationdecrease-key
actually often perform better in practice.Here is one way to modify the above algorithm so that it does not require decrease-key
, and so that it does not mutate the sort key of any object that has already been added to the priority queue:
init empty priority queue Q of (vertex, dist) pairs sorted by increasing dist
# the dist part of each pair is immutable; more than one pair can be created for a single vertex
for each vertex v set v.visited = false, v.dist = infinity
set goal.dist = 0
add make-pair(goal, 0) to Q
while Q not empty # O(E) since inner loop adds a pair for every edge
(v, d) = pop Q # O(log E) with binary heap (Java PriorityQueue)
if not v.visited
set v.visited = true
# this inner loop iterates a *total* of O(E) times
for each neighbor vertex n of v with n.visited = false and n.dist > (v.dist + dist(v,n))
set n.next = v, n.dist = v.dist + dist(v,n)
add make-pair(n, n.dist) to Q # O(log E) with binary heap (Java PriorityQueue)
The asymptotic running time increases to
in this implementation.