700

Chapter 25 All-Pairs Shortest Paths

25.2-6
How can we use the output of the Floyd-Warshall algorithm to detect the presence
of a negative-weight cycle?
25.2-7
Another way to reconstruct shortest paths in the Floyd-Warshall algorithm uses
values ij.k/ for i; j; k D 1; 2; : : : ; n, where ij.k/ is the highest-numbered intermediate vertex of a shortest path from i to j in which all intermediate vertices are
in the set f1; 2; : : : ; kg. Give a recursive formulation for ij.k/ , modify the F LOYD WARSHALL procedure to compute the ij.k/ values, and rewrite the P RINT-A LL 

PAIRS -S HORTEST-PATH procedure to take the matrix D ij.n/ as an input.
How is the matrix like the s table in the matrix-chain multiplication problem of
Section 15.2?
25.2-8
Give an O.VE/-time algorithm for computing the transitive closure of a directed
graph G D .V; E/.
25.2-9
Suppose that we can compute the transitive closure of a directed acyclic graph in
f .jV j ; jEj/ time, where f is a monotonically increasing function of jV j and jEj.
Show that the time to compute the transitive closure G  D .V; E  / of a general
directed graph G D .V; E/ is then f .jV j ; jEj/ C O.V C E  /.

25.3 Johnsons algorithm for sparse graphs

Johnsons algorithm nds shortest paths between all pairs in O.V 2 lg V C VE/
time. For sparse graphs, it is asymptotically faster than either repeated squaring of
matrices or the Floyd-Warshall algorithm. The algorithm either returns a matrix of
shortest-path weights for all pairs of vertices or reports that the input graph contains
a negative-weight cycle. Johnsons algorithm uses as subroutines both Dijkstras
algorithm and the Bellman-Ford algorithm, which Chapter 24 describes.
Johnsons algorithm uses the technique of reweighting, which works as follows.
If all edge weights w in a graph G D .V; E/ are nonnegative, we can nd shortest paths between all pairs of vertices by running Dijkstras algorithm once from
each vertex; with the Fibonacci-heap min-priority queue, the running time of this
all-pairs algorithm is O.V 2 lg V C VE/. If G has negative-weight edges but no
negative-weight cycles, we simply compute a new set of nonnegative edge weights

25.3 Johnsons algorithm for sparse graphs

701

that allows us to use the same method. The new set of edge weights w
y must satisfy
two important properties:
1. For all pairs of vertices u;  2 V , a path p is a shortest path from u to  using
weight function w if and only if p is also a shortest path from u to  using
weight function w.
y
2. For all edges .u; /, the new weight w.u;
y / is nonnegative.
As we shall see in a moment, we can preprocess G to determine the new weight
function w
y in O.VE/ time.
Preserving shortest paths by reweighting
The following lemma shows how easily we can reweight the edges to satisfy the
rst property above. We use to denote shortest-path weights derived from weight
function w and y to denote shortest-path weights derived from weight function w.
y
Lemma 25.1 (Reweighting does not change shortest paths)
Given a weighted, directed graph G D .V; E/ with weight function w W E ! R,
let h W V ! R be any function mapping vertices to real numbers. For each edge
.u; / 2 E, dene
w.u;
y / D w.u; / C h.u/  h./ :

(25.9)

Let p D h0 ; 1 ; : : : ; k i be any path from vertex 0 to vertex k . Then p is a

shortest path from 0 to k with weight function w if and only if it is a shortest path
y 0 ; k /.
y
D .
with weight function w.
y That is, w.p/ D .0 ; k / if and only if w.p/
Furthermore, G has a negative-weight cycle using weight function w if and only
if G has a negative-weight cycle using weight function w.
y
Proof

We start by showing that

w.p/
y
D w.p/ C h.0 /  h.k / :

(25.10)

We have
w.p/
y
D

k
X

w.
y i 1 ; i /

i D1

k
X

.w.i 1 ; i / C h.i 1 /  h.i //

i D1

k
X

w.i 1 ; i / C h.0 /  h.k /

i D1

D w.p/ C h.0 /  h.k / :

(because the sum telescopes)

702

Chapter 25 All-Pairs Shortest Paths

Therefore, any path p from 0 to k has w.p/

y
D w.p/ C h.0 /  h.k /. Because h.0 / and h.k / do not depend on the path, if one path from 0 to k is
shorter than another using weight function w, then it is also shorter using w.
y Thus,
y
y
D .0 ; k /.
w.p/ D .0 ; k / if and only if w.p/
Finally, we show that G has a negative-weight cycle using weight function w if
and only if G has a negative-weight cycle using weight function w.
y Consider any
cycle c D h0 ; 1 ; : : : ; k i, where 0 D k . By equation (25.10),
w.c/
y
D w.c/ C h.0 /  h.k /
D w.c/ ;
and thus c has negative weight using w if and only if it has negative weight using w.
y
Producing nonnegative weights by reweighting
Our next goal is to ensure that the second property holds: we want w.u;
y / to be
nonnegative for all edges .u; / 2 E. Given a weighted, directed graph G D
.V; E/ with weight function w W E ! R, we make a new graph G 0 D .V 0 ; E 0 /,
where V 0 D V [ fsg for some new vertex s 62 V and E 0 D E [ f.s; / W  2 V g.
We extend the weight function w so that w.s; / D 0 for all  2 V . Note that
because s has no edges that enter it, no shortest paths in G 0 , other than those with
source s, contain s. Moreover, G 0 has no negative-weight cycles if and only if G
has no negative-weight cycles. Figure 25.6(a) shows the graph G 0 corresponding
to the graph G of Figure 25.1.
Now suppose that G and G 0 have no negative-weight cycles. Let us dene
h./ D .s; / for all  2 V 0 . By the triangle inequality (Lemma 24.10),
we have h./  h.u/ C w.u; / for all edges .u; / 2 E 0 . Thus, if we dene the new weights w
y by reweighting according to equation (25.9), we have
w.u;
y / D w.u; / C h.u/  h./  0, and we have satised the second property.
Figure 25.6(b) shows the graph G 0 from Figure 25.6(a) with reweighted edges.
Computing all-pairs shortest paths
Johnsons algorithm to compute all-pairs shortest paths uses the Bellman-Ford algorithm (Section 24.1) and Dijkstras algorithm (Section 24.3) as subroutines. It
assumes implicitly that the edges are stored in adjacency lists. The algorithm returns the usual jV j jV j matrix D D dij , where dij D .i; j /, or it reports that
the input graph contains a negative-weight cycle. As is typical for an all-pairs
shortest-paths algorithm, we assume that the vertices are numbered from 1 to jV j.

25.3 Johnsons algorithm for sparse graphs

0
3
0

4
0

5 3

1
6

3
2/3

13

10

10

2/2

2/1

13

2
0

2/2
5

3
0/5

2
(f)

3
0/4

0 4

2
(d)

4
1
2/7

13

2
0

10

0/1
4

2/3
5

3
0/0
0

0
2
(e)

0/5
4

4
1
4/8

13

2
0

10

2
2/5

2
0/1
1
2/2

13

2
0

(c)

10

5 3

2
0/4
0

1
2/3

13

2
0/0
0

0/4

(b)

2
2/1
4

0 4

(a)

1
0/0

4
1
0

0
5

2
1

1
4

1
0
2

2
1

703

10

0/0

0/0

3
2/1
0

0
2
(g)

2/6
4

Figure 25.6 Johnsons all-pairs shortest-paths algorithm run on the graph of Figure 25.1. Vertex numbers appear outside the vertices. (a) The graph G 0 with the original weight function w.
The new vertex s is black. Within each vertex  is h./ D .s; /. (b) After reweighting each
edge .u; / with weight function w.u;
y / D w.u; / C h.u/  h./. (c)(g) The result of running
Dijkstras algorithm on each vertex of G using weight function w
y. In each part, the source vertex u
is black, and shaded edges are in the shortest-paths tree computed by the algorithm. Within each
y / and .u; /, separated by a slash. The value du D .u; / is equal to
vertex  are the values .u;
y u; / C h./  h.u/.
.

704

Chapter 25 All-Pairs Shortest Paths

J OHNSON .G; w/
1 compute G 0 , where G 0 :V D G:V [ fsg,
G 0 :E D G:E [ f.s; / W  2 G:Vg, and
w.s; / D 0 for all  2 G:V
2 if B ELLMAN -F ORD .G 0 ; w; s/ == FALSE
3
print the input graph contains a negative-weight cycle
4 else for each vertex  2 G 0 :V
5
set h./ to the value of .s; /
computed by the Bellman-Ford algorithm
6
for each edge .u; / 2 G 0 :E
7
w.u;
y / D w.u; / C h.u/  h./
8
let D D .du / be a new n n matrix
9
for each vertex u 2 G:V
y / for all  2 G:V
y u/ to compute .u;
10
run D IJKSTRA .G; w;
11
for each vertex  2 G:V
y / C h./  h.u/
12
du D .u;
13
return D
This code simply performs the actions we specied earlier. Line 1 produces G 0 .
Line 2 runs the Bellman-Ford algorithm on G 0 with weight function w and source
vertex s. If G 0 , and hence G, contains a negative-weight cycle, line 3 reports the
problem. Lines 412 assume that G 0 contains no negative-weight cycles. Lines 45
set h./ to the shortest-path weight .s; / computed by the Bellman-Ford algoy For each pair of verrithm for all  2 V 0 . Lines 67 compute the new weights w.
y /
tices u;  2 V , the for loop of lines 912 computes the shortest-path weight .u;
by calling Dijkstras algorithm once from each vertex in V . Line 12 stores in
matrix entry du the correct shortest-path weight .u; /, calculated using equation (25.10). Finally, line 13 returns the completed D matrix. Figure 25.6 depicts
the execution of Johnsons algorithm.
If we implement the min-priority queue in Dijkstras algorithm by a Fibonacci
heap, Johnsons algorithm runs in O.V 2 lg V CVE/ time. The simpler binary minheap implementation yields a running time of O.VE lg V /, which is still asymptotically faster than the Floyd-Warshall algorithm if the graph is sparse.
Exercises
25.3-1
Use Johnsons algorithm to nd the shortest paths between all pairs of vertices in
the graph of Figure 25.2. Show the values of h and w
y computed by the algorithm.