We already know what’s topological sort of a directed acyclic graph. So why do we need a revision of this algorithm? First of all I never mentioned its complexity, thus to understand why we do need a revision let’s get again on the algorithm.

We have a directed acyclic graph (DAG). There are no cycles so we must go for some kind of order putting all the vertices of the graph in such an order, that if there’s a directed edge (u, v), u must precede v in that order.

The process of putting all the vertices of the DAG in such an order is called topological sorting. It’s commonly used in task scheduling or while finding the shortest paths in a DAG.

The algorithm itself is pretty simple to understand and code. We must start from the vertex (vertices) that don’t have predecessors.

We put them in our sorted list in random order. Since they don’t depend on each other we can assume they are equally sorted already. Indeed thinking of a task schedule if there are tasks that don’t have predecessors (they don’t depend on other tasks before them) and that don’t depend on each other we can put them in random order (and execute them in random order).

Once we have the vertices with no predecessors we must remove the edges starting from them. Then – go again with the vertices with no predecessors.

It’s as simple as that, so why do we need a revision of this algorithm? Well, basically because of its efficiency.

Overview

As we know most of the graph algorithms depend on the way the graph is represented in our application. We consider as the two main representations the adjacency matrix …

… and adjacency lists.

Let’s first take a look of some of the main approaches to get the topologically sorted list at the end of the algorithm.

What can we do in order to find the vertices with no predecessors? We can only scan the entire list of vertices.

Adjacency Matrix

In case we’re using adjacency matrix we need|V|^2 space to store the graph. To find the vertices with no predecessors we have to scan the entire graph, which will cost us O(|V|^2) time. And we’ll have to do that |V| times. This will be |V|^3 time consuming algorithm and for dense graphs this will be quite an ineffective algorithm.

Adjacency Lists

What about the adjacency list? There we need |E| space to store a directed graph. How fast can we find a node with no predecessor? Practically we’ll need O(|E|) time. Thus in the worst case we have again O(|V|^2) time consuming programs.

So what can be done in order to optimize this algorithm?

Practically we can start by picking up a random vertex and “go back” until we get a node with no predecessors. This approach can be very effective yet also very ineffective. First of all if we have to scan all the way back to a node with no predecessors this will cost us |V| time, but if we stuck on a node that don’t have a preceding node then we’ll have a constant speed.

This means that we can modify the algorithm a bit in order to improve a lot the algorithm. We just need to store both incoming and outgoing edges and slightly modify the adjacency lists.

What’s the algorithm now?

First we easily find the nodes with no predecessors. Then, using a queue, we can keep the nodes with no predecessors and on each dequeue we can remove the edges from the node to all other nodes.

Pseudo Code

1. Represent the graph with two lists on each vertex (incoming edges and outgoing edges)
2. Make an empty queue Q;
3. Make an empty topologically sorted list T;
4. Push all items with no predecessors in Q;
5. While Q is not empty
   a. Dequeue from Q into u;
   b. Push u in T;
   c. Remove all outgoing edges from u;
6. Return T;

This approach will give us a better performance than the “brute force” approach. The running time complexity is O(|V| + |E|). The problem is that we need additional space and an operational queue, but this approach is a perfect example of how by using additional space you can get a better performing algorithm.

A very common problem in computer programming is finding the longest increasing (decreasing) subsequence in a sequence of numbers (usually integers). Actually this is a typical dynamic programming problem.

Dynamic programming can be described as a huge area of computer science problems that can be categorized by the way they can be solved. Unlike divide and conquer, where we were able to merge the fairly equal sub-solutions in order to receive one single solution of the problem, in dynamic programming we usually try to find an optimal sub-solution and then grow it.

Once we have an optimal sub-solution on each step we try to upgrade it in order to cover the whole problem. Thus a typical member of the dynamic programming class is finding the longest subsequence.

However this problem is interesting because it can be related to graph theory. Let’s find out how.

Overview

We already know various ways to calculate the shortest paths in a graph. Indeed finding the single-source shortest path is a typical graph problem. To model such kind of solutions we definitely need a graph represented in our solution.

However the single-source shortest path isn’t a straight-forward problem. It depends on many factors. Thus for positive edges the algorithm of Edsger Dijkstra can be a perfect solution, but when the graph contains negative edges his algorithm is no longer useful.

In the presence of negative edges the Dijkstra’s algorithm doesn’t work and we’d better use the Bellman-Ford algorithm. It is interesting to note, that it was exactly Richard Bellman who first introduced the term “dynamic programming” in the 1940s.

In fact the Bellman-Ford algorithm was able to detect negative cycles. That’s too important, because in presence of negative cycles the shortest path problem is no longer well defined.

In the other hand when we’re talking about shortest paths in a DAG (Directed Acyclic Graph) we can find a faster (linear) solution. That’s because we’re sure that there are no cycles (not even negative cycles)!

Finding the shortest paths in a DAG is closely related to the topological sorting of the DAG. This gives us a linear representation of the vertices of the DAG and we can clearly calculate the distances from the starting node to all other nodes. Note that in a DAG we have one or more nodes that can be considered as starting nodes – which means they don’t have predecessors (incoming edges).

Following the words above is pretty hard to find what all these graph algorithms have to do with finding the longest increasing (decreasing) subsequence. Actually this problem is very closely related to the toplogical sort of the DAG and the problem of finding shortest paths in a DAG.

That’s because we can represent our sequence as a DAG. The only thing we must care about is to “connect” with directed edges those elements that form an increasing (decreasing) pair.

Thus the sequence from our example [1, 8, 2, 7, 3, 4, 1, 6] is going to look like this.

Another important thing to note is that we don’t search for shortest, but for longest path, since our task is to find the longest subsequence.

Pseudo Code

Our pseudo code for finding the shortest paths in a DAG was something like that.

1. Get the toplogically sorted list L of the DAG;
2. The starting node is s;
3. The distance to s equals to 0;
4. All other distances are initialized to ∞
5. For each node (v) in L\{s} do:
5.1. If dist(v) > dist(u) + w(u, v) then
5.1.1.  dist(v) := dist(u) + w(u, v)

In the above pseudo code “u” is every predecessor of “v”!

Now we must “reverse” the solution above in order to find the longest increasing subsequence. Note that we don’t care any more about the weight of the edges, thus we can simply substitute them with 1.

1. Get the sequence (L) as a toplogically sorted DAG;
2. For each (i) in L do:
2.1. S(i) := 1 + max(S(j), where (i, j) is an edge from the DAG);

Application

Finding the longest increasing subsequence can be very useful not only at the Google/Yahoo/Facebook interview, but also in various fields of statistics.

Along with the Kruskal’s minimum spanning tree algorithm, there’s another general algorithm that solves the problem. The algorithm of Prim.

As we already know the algorithm of Kruskal works in a pretty natural and logical way. Since we’re trying to build a MST, which is naturally build by the minimal edges of the graph (G), we sort them in a non-descending order and we start building the tree.

During the whole process of building the final minimum spanning tree Kruskal’s algorithm keeps a forest of trees. The number of trees in that forest decreases on each step and finally we get the minimum weight spanning tree.

A key point in the Kruskal’s approach is the way we get the “next” edge from G that should be added to one of the trees of the forest (or to connect two trees from the forest). The only thing we should be aware of is to choose an edge that’s connecting two vertices – u and v and these two shouldn’t be in the same tree. That’s all.

An important feature of the Kruskal’s algorithm is that it builds the MST just by sorting the edges by their weight and doesn’t care about a particular starting vertex.

In the same time there’s another algorithm that builds a MST – the algorithm of Prim designed by Robert Prim in 1957.

Overview

The idea behind the Prim’s algorithm is rather different from Kruskal’s approach. During the process of building the MST this algorithm keeps a single tree, which is finally sub-tree of the final minimum weight spanning tree.

On each step we chose an edge which we add to the growing tree that finally forms the MST.

It is somehow unnatural approach! We start from a given vertex and initially we don’t choose the lightest edge. Thus during the whole process the tree grows, but outside the tree (T) there might be edges that are lighter than those in the tree (i.e. the edge (5, 1) from the tree above is lighter than (2, 5) but (2, 5) is added to the growing tree before the edge (5, 1)).

Compared to the Kruskal’s algorithm this time everything seems to be really unnatural. How we should be sure the final tree (T) will be a minimum spanning tree since we don’t get the lightest edge on each step?

Actually we are sure that the final tree is a MST because of another obvious feature of the minimum spanning trees. They should “connect” all the vertices of G, thus somehow at least one edge reaching each vertex will appear in the MST. Thus we shouldn’t care where do we start, the only important thing is to choose the lightest edge that’s visible so far.

This algorithm looks much like Dijkstra’s shortest path in a graph, because we start from a vertex, we push all the edges starting from this node to a priority queue and we chose the lightest edge. Going to the next node connected by this edge we append to the queue all the edges that aren’t in the queue.

That way the queue grows and we get always the lightest edge – thus forming a priority queue.

Now let’s summarize the algorithm of Prim

Pseudo Code

As an initial input we have the graph (G) and a starting vertex (s).

1.  Make a queue (Q) with all the vertices of G (V);
2.  For each member of Q set the priority to INFINITY;
3.  Only for the starting vertex (s) set the priority to 0;
4.  The parent of (s) should be NULL;
5.  While Q isn’t empty
6.     Get the minimum from Q – let’s say (u); (priority queue);
7.     For each adjacent vertex to (v) to (u)
8.        If (v) is in Q and weight of (u, v) < priority of (v) then
9.           The parent of (v) is set to be (u)
10.          The priority of (v) is the weight of (u, v)

Indeed it looks much like the Dijkstra’s algorithm.

Code

Here’s a PHP implementation of the algorithm of Prim, which directly follows the pseudo code.

// Prim's algorithm

define('INFINITY', 100000000);

// the graph
$G = array(
    0 => array( 0,  4,  0,  0,  0,  0,  0,  0,  8),
    1 => array( 4,  0,  8,  0,  0,  0,  0,  0,  11),
    2 => array( 0,  8,  0,  7,  0,  4,  2,  0,  0),
    3 => array( 0,  0,  7,  0,  9,  14,  0,  0,  0),
    4 => array( 0,  0,  0,  9,  0,  10,  0,  0,  0),
    5 => array( 0,  0,  4,  14,  10,  0,  0,  2,  0),
    6 => array( 0,  0,  2,  0,  0,  0,  0,  6,  7),
    7 => array( 0,  0,  0,  0,  0,  2,  6,  0,  1),
    8 => array( 8,  11,  0,  0,  0,  0,  7,  1,  0),
);

function prim(&$graph, $start)
{
    $q = array(); // queue
    $p = array(); // parent
    
    foreach (array_keys($graph) as $k) {
        $q[$k] = INFINITY;
    }
    
    $q[$start] = 0;
    $p[$start] = NULL;
    
    asort($q);
    
    while ($q) {
        // get the minimum value
        $keys = array_keys($q);
        $u = $keys[0];
        
        foreach ($graph[$u] as $v => $weight) {
            if ($weight > 0 && in_array($v, $keys) && $weight < $q[$v]) {
                $p[$v] = $u;
                $q[$v] = $weight;
            }
        }
        
        unset($q[$u]);
        asort($q);
    }
    
    return $p;
}

prim($G, 5);

History

It’s curious to say that the algorithm developed by Robert Prim isn’t developed by him. It’s considered that a Czech mathematician Vojtech Jarnik discovered back in 1930. However now we know this algorithm as the algorithm of Prim, which independently discovered it in 1957 as I said above, and finally Edsger Dijkstra described it in 1959. That’s why his algorithm on finding the single-source shortest paths in a graph looks so much to this algorithm. Perhaps by finding this algorithm on minimum spanning tree Dijkstra discovered how we can find the shortest paths to all vertices using a priority queue. Indeed the paths to all other vertices use the edges of the minimum spanning tree.

Just because Jarnik found and described this algorithm 27 years earlier than Robert Prim, today it’s more convenient to call this algorithm the Prim-Jarnik algorithm.

One of the two main algorithms in finding the minimum spanning tree algorithms is the algorithm of Kruskal. Before getting into the details, let’s get back to the principles of the minimum spanning tree.

We have a weighted graph and of all spanning trees we’d like to find the one with minimal weight. As an example on the picture above you see a spanning tree (T) on the graph (G), but that isn’t the minimum weight spanning tree!

We can think of a group of islands and the possible connections of bridges connecting them. Of course building bridges is expensive and time consuming, so we must be aware of what kind of bridges we want to build. Nevertheless there is an important question, what’s the minimum price we’d like to pay to build such set of bridges connecting all the islands.

Thus we practically need to build a minimum spanning tree, where the vertices will be the islands, while the edges will be the possible bridges between them. Every possible bridge has a weight (the price or the time we need to build it, etc.).

This scenario is only one of possible use cases of where minimum spanning trees can be used in practice.

The two main approaches – the Kruskal’s and the Prim’s algorithms however differ.

Overview

The algorithm of Kruskal starts by initializing a set of |V| trees.

During the process of building the final spanning tree we keep a forest. Obviously we start with a forest with |V| trees, where each tree is a single node tree.

On some point we have a forest of “k” trees which are all a sub-trees of the minimum spanning tree.

Finally one step before building the final MST we have two trees and we connect them with the less weighted edge left that connects them.

It’s important to note that during the process of building the tree we sort the edges in ascending order by their weight.

Than we start getting edges and check whether their ends (the two vertices making the edge) belong to a different sub-trees.

Pseudo Code

1. T (the final spanning tree) is defined to be the empty set;
2. For each vertex v of G, make the empty set out of v;
3. Sort the edges of G in ascending (non-decreasing) order;
4. For each edge (u, v) from the sored list of step 3.
      If u and v belong to different sets
         Add (u,v) to T;
         Get together u and v in one single set;
5. Return T

A great feature about the Kruskal’s algorithm is that it also work on disconnected graphs.

History

Kruskal’s algorithm is named after Joseph Kruskal, who wasn’t only computer scientist, but also prominent mathematician and statistician. Although he is best known for its algorithm for computing the minimum spanning tree, described in this post, he’s also known with his work as a statistician and his contribution to the formulation of multidimensional scaling.

Kruskal also explored the Indo-European languages contributing the studies of the linguistics along with other scientists. His “Indo-European Lexicographical List” (http://www.wordgumbo.com/ie/cmp/) is still widely used.

Here’s a classical task on graphs. We have a group of cities and we must wire them to provide them all with electricity. Out of all possible connections we can make, which one is using minimum amount of wire.

To wire N cities, it’s clear that, you need to use at least N-1 wires connecting a pair of cities. The problem is that sometimes you have more than one choice to do it. Even for small number of cities there must be more than one solution as shown on the image bellow.

Here we can wire these four nodes in several ways, but the question is, which one is the best one. By the way defining the term “best one” is also tricky. Most often this means which uses least wire, but it can be anything else depending on the circumstances.

As we talk on weighted graphs we can generally speak of a minimum weight solution through all the vertices of the graph.

By the way there might be more the one equally optimal (minimal) solutions.

Overview

Obviously we must choose those edges that are enough to connect all the vertices of the graph and whose sum of weights is minimal. Since we can’t have cycles in our final solution it must form a tree. Thus we’re speaking on a minimum weight spanning tree, as the tree spans over the whole graph.

Does each connected and weighted graph have a minimum spanning tree? The answer is yes! By removing the cycles from the graph G we get a spanning tree, since it’s connected. From all possible spanning trees one or more are minimal.

If w(u, v) is the weight of the edge (u, v), we can speak of weight of any spanning tree T – w(T) which is the sum of all the edges forming that tree.

Thus the weight of the minimum spanning tree is less than the weight of whatever other spanning tree of G.

After we’re sure that there is at least one minimum spanning tree for all connected and weighted graphs we only need to find it somehow.

We can go with an incremental approach. At the end we’ll have the minimum spanning tree (MST), but before that on each step of our algorithm we’ll have a sub-set of this final tree, which will grow and grow until it becomes the real MST. This subset of edges we’ll keep in one additional set A.

So far we know that on each step we have a subset of the final MST, but first we need to answer a couple of questions.

How do we start?

Well, we’ll start with the empty set of edges. Clearly the empty set is a subset of any other set, thus it will be also a subset of the MST.

How do we grow the tree?

Another question we must answer is how to grow the tree. Since we have a MST sub-set (A) on each step how do we add an edge to this set in order to get another (bigger than the previous one) subset of edges, which will be again a subset of the minimum spanning tree?

Clearly we must make a decision which edge to add to the growing subset and this is the tricky part of this algorithm.

Chose the lowest weight edge!

To find the minimum spanning tree on each step we must get the lowest weighted edge that connects our subset (A) with the rest of the vertices.

However can we be sure that by choosing the less weighted edge we’ll get the MST? Well, let’s assume that isn’t right in order to prove that wrong!

OK so on some step of our growing sub-tree we don’t get the lightest edge (u, v), because we somehow doubt this rule, and we get another edge – let’s say (x, y). Mind that w(x, y) >= w(u, v).

Thus our final MST will contain somewhere in its set of edges the edge (x, y), but the weight of MST w(T) is minimal, and if we get another spanning tree that contains the exact same edges as T but instead of (x, y) contains (u, v) we’ll get a smaller weight!

That isn’t possible! Thus we proved that on each step we must get the less weighted edge.

This particular approach is called “greedy”, because on each step we get the best possible choice. However greedy algorithms don’t always get the right or optimal solution. Fortunately for MST this isn’t true so we can be greedy as much as we can!

OK let’s make a summary of our algorithm in the following pseudo code.

Pseudo Code

1. We start with an  empty set (A) subset of the final MST;
2. Until A does not form T:
      a. Get the less weighted edge u from G;	
      b. Add u to A;
3. Return A

Application

Actually this algorithm is used firstly by Borůvka which started to wire Moravia in 1926. Even without knowing that the “greedy” approach will lead him to the right solution he optimally covered Moravia with electricity.

However this algorithm is too general and there are two main algorithms – the Prim’s algorithm and the Kruskal’s algorithm that we shall see in future posts.

The thing is that on each step we must get the less weighted edge and both algorithms use different approaches to do that.

We saw how to find the shortest path in a graph with positive edges using the Dijkstra’s algorithm. We also know how to find the shortest paths from a given source node to all other nodes even when there are negative edges using the Bellman-Ford algorithm. Now we’ll see that there’s a faster algorithm running in linear time that can find the shortest paths from a given source node to all other reachable vertices in a directed acyclic graph, also known as a DAG.

Because the DAG is acyclic we don’t have to worry about negative cycles. As we already know it’s pointless to speak about shortest path in the presence of negative cycles because we can “loop” over these cycles and practically our path will become shorter and shorter.

Thus we have two problems to overcome with Dijkstra and the Bellman-Ford algorithms. First of all we needed only positive weights and on the second place we didn’t want cycles. Well, we can handle both cases in this algorithm.

Overview

The first thing we know about DAGs is that they can easily be topologically sorted. Topological sort can be used in many practical cases, but perhaps the mostly used one is when trying to schedule dependent tasks.

After a topological sort we end with a list of vertices of the DAG and we’re sure that if there’s an edge (u, v), u will precede v in the topologically sorted list.

This information is precious and the only thing we need to do is to pass through this sorted list and to calculate distances for a shortest paths just like the algorithm of Dijkstra.

OK, so let’s summarize this algorithm:
– First we must topologically sort the DAG;
– As a second step we set the distance to the source to 0 and infinity to all other vertices;
– Then for each vertex from the list we pass through all its neighbors and we check for shortest path;

It’s pretty much like the Dijkstra’s algorithm with the main difference that we used a priority queue then, while this time we use the list from the topological sort.

Code

This time the code is actually a pseudocode. Altough all the examples so far was in PHP, perhaps pseudocode is easier to understand and doesn’t bind you in a specific language implementation. Also if you don’t feel comforatable with the given programming language it can be more difficult for you to understand the code than by reading pseudocode.

1. Topologically sort G into L;
2. Set the distance to the source to 0;
3. Set the distances to all other vertices to infinity;
4. For each vertex u in L
5.    - Walk through all neighbors v of u;
6.    - If dist(v) > dist(u) + w(u, v) 
7.       - Set dist(v) <- dist(u) + w(u, v);

Application

It’s clear why and where we must use this algorithm. The only problem is that we must be sure that the graph doesn’t have cycles. However if we’re aware of how the graph is created we may have some additional information if there are cycles or not – then this linear time algorithm can be very applicable.

As we saw in the previous post, the algorithm of Dijkstra is very useful when it comes to find all the shortest paths in a weighted graph. However it has one major problem! Obviously it doesn’t work correctly when dealing with negative lengths of the edges.

We know that the algorithm works perfectly when it comes to positive edges, and that is absolutely normal because we try to optimize the inequality of the triangle.

Since Dijkstra’s algorithm make use of a priority queue normally we get first the shortest adjacent edge to the starting point. In our very basic example we’ll get first the edge with the length of 3 -> (S, A).

However when it comes to negative edges we can’t use any more priority queues, so we need a different, yet working solution.

Overview

The solution was published by Richard E. Bellman and Lester Ford, Jr. in 1958 in their publication “On a Routing Problem” and it is quite simple to explain and understand. Since we can prioritize the edges by its lengths the only thing we should do is to calculate all the paths. And to be sure that our algorithm will find all the paths correctly we repeat that N-1 times, where N is the number of vertices (|V| = N)!

In this very basic image we can see how Bellman-Ford solves the problem. First we get the distances from S to A and B, which are respectively 3 and 4, but there is a shorter path to A, which passes through B and it is (S, B) + (B, A) = 4 – 2 = 2.

Code

Here’s the code on PHP. Note that this time we use an adjacency matrix and an additional array of distances. It’s important (for directed graphs, and our graph this time is directed) to put the positive value of A[j][i] if A[i][j] is negative. Note the case for A[1][2]!

define('INFINITY', 10000000);

$matrix = array(
    0 => array( 0,  3,  4),
    1 => array( 0,  0,  2),
    2 => array( 0,  -2, 0),
);

$len = count($matrix);

$dist = array();

function BellmanFord(&$matrix, &$dist, $start)
{
    global $len;
    
    foreach (array_keys($matrix) as $vertex) {
        $dist[$vertex] = INFINITY;
        if ($vertex == $start) {
            $dist[$vertex] = 0;
        }
    }
    
    for ($k = 0; $k < $len - 1; $k++) {
        for ($i = 0; $i < $len; $i++) {
            for ($j = 0; $j < $len; $j++) {
                if ($dist[$i] > $dist[$j] + $matrix[$j][$i]) {
                    $dist[$i] = $dist[$j] + $matrix[$j][$i];
                }
            }
        }
    }
}

BellmanFord($matrix, $dist, 0);

// [0, 2, 4]
print_r($dist);

Complexity

The complexity is clearly O(n³) which follows directly from the code above.

Application

Actually this algorithm is very useful and it not only works with negative weights, but also can help us find negative cycles in the graph.

This is done with the simple check after the main loop.

    for ($i = 0; $i < $len; $i++) {
        for ($j = 0; $j < $len; $j++) {
            if ($dist[$i] > $dist[$j] + $matrix[$j][$i]) {
                echo 'The graph contains a negative cycle!';
            }
        }
    }

And here’s the full code.

$matrix = array(
    0 => array( 0,  3,  4),
    1 => array( 0,  0,  2),
    2 => array( 0,  -2, 0),
);

$len = count($matrix);

$dist = array();

function BellmanFord(&$matrix, &$dist, $start)
{
    global $len;
    
    foreach (array_keys($matrix) as $vertex) {
        $dist[$vertex] = INFINITY;
        if ($vertex == $start) {
            $dist[$vertex] = 0;
        }
    }
    
    for ($k = 0; $k < $len - 1; $k++) {
        for ($i = 0; $i < $len; $i++) {
            for ($j = 0; $j < $len; $j++) {
                if ($dist[$i] > $dist[$j] + $matrix[$j][$i]) {
                    $dist[$i] = $dist[$j] + $matrix[$j][$i];
                }
            }
        }
    }
    
    for ($i = 0; $i < $len; $i++) {
        for ($j = 0; $j < $len; $j++) {
            if ($dist[$i] > $dist[$j] + $matrix[$j][$i]) {
                echo 'The graph contains a negative cycle!';
            }
        }
    }
}

BellmanFord($matrix, $dist, 0);

// [0, 2, 4]
print_r($dist);

We already know how we can find the shortest paths in a graph starting from a given vertex. Practically we modified breadth-first search in order to calculate the distances from s to all other nodes reachable from s. We know that this works because BFS walks through the graph level by level.

Some sources give a very simple explanation of how BFS finds the shortest paths in a graph. We must just think of the graph as a set of balls connected through strings.

As we can see by lifting the ball called “S” all other balls fall down. The closest balls are directly connected to “s” and this is the first level, while the outermost balls are those with longest paths.

Clearly edges like those between A and B doesn’t matter for our BFS algorithm because they don’t make the path from S to C through B shorter. This is also known as the triangle inequality, where the sum of the lengths of two of the sides of the triangle is always greater than the length of the third side.

We must only answer the question is BFS the best algorithm that finds the shortest path between any two nodes of the graph? This is a reasonable question because as we know by using BFS we don’t find only the shortest path between given vertices i and j, but we also get the shortest paths between i and all other vertices of G. This is an information that we actually don’t need, but can we find the shortest path between i and j without that info?

The answer is simply “no”! Practically depth-first search can’t help us. Even worse – we can find paths that are far not the shortest ones.

In the image above using DFS the distance between 1 and 7 is 7 while practically there is an edge between them.

So BFS is the optimal algorithm for finding shortest paths in a graph. But there’s a catch! This algorithm works fine when we assume that all the edges are the same length. In the examples so far each edge has the value of 1. So N edges between s and i made the distance between them of a length N.

Overview

As we know in practice different edges can have different values. Exactly that was the case in weighted graphs. Going back to the road map example the distances between different cities are commonly evaluated in miles or kilometers. Of course we can associate any other meaningful value to this edges. This can be either time in hours to travel between cities, money for fuel or anything else.

Now BFS can’t help us any more. Why? Because using non-equal values for the edges the triangle inequality is no longer true. Now the edge (the direct path) between A and B can be greater than the sum of the two edges (A, C) + (C, B)!

In other words, assuming the same abstraction with balls and wires the hanging wires can’t be discarded so easily.

So now how can we solve this problem? A very dummy approach is to break apart each edge with dummy vertices in order to make BFS work again.

However this approach has several weak points. The major one is that we’ll have to keep much more information, which means more memory usage, for even small graphs. This is done in case we break each edge on too many parts.

The solution of this problem was given by Edsger Dijkstra in 1956 and published in 1959. The only thing we should do now is to be sure that even discarding the triangle inequality we have the shortest paths. The first thing to do is to keep information for the distance from s to the parent (previous) node of i in the graph in order to calculate which distance is shorter.

In BFS we used a queue in order to walk through all the ancestors of a node. This was made consecutively. Thus for the graph G on the next image the order of enqueuing the ancestors of S was A, B, C.

The Dijkstra’s algorithm make use of a priority queue, also know as a heap. This fact combined by the fact we keep info for the shortest path so far help us find shortest paths in a weighted graphs.

Why this works? To answer this question let’s see the next very basic example, assuming the graph G from the next image. As we can see the triangle inequality isn’t true.

OK, we see that the path [S, B, A] is shorter than [S, A] although the edge (S, A) exists. How the Dijkstra algorithm overcomes this problem.

First we have no information about the distances (S, A) and (S, B), the only thing we know is that S is the starting point, its distance is 0 and its path so far is the empty set. So first we enqueue in a priority the distances from S to A and B.

Now we dequeue the minimum (first in the heap) element from the queue – the closest node to S, which is B. Then all the nodes adjacent to S in the queue are tested for adjacency to B, thus if we have already the distance between S and A now we can test if its longer than (S, B) + (B, A) – the triangle inequality!

So far we know that we must change a bit BFS to get the Dijkstra algorithm. The only thing to do is to keep info for each node for the path through its parent and to use a priority queue.

Code

Implementing this algorithms isn’t much more difficult than BFS, so here’s the code in PHP. However this example make use of the standard php library SPL and the PriorityQueue data structure, but any developer can code his own heap.

Here’s the graph from the code:

class vertex
{
    public $key         = null;
    public $visited     = 0;
    public $distance    = 1000000;  // infinite
    public $parent      = null;
    public $path        = null;
    
    public function __construct($key) 
    {
        $this->key  = $key;
    }
}

class PriorityQueue extends SplPriorityQueue
{
    public function compare($a, $b)
    {
        if ($a === $b) return 0;
        return $a > $b ? -1 : 1;
    }
}

$v0 = new vertex(0);
$v1 = new vertex(1);
$v2 = new vertex(2);
$v3 = new vertex(3);
$v4 = new vertex(4);
$v5 = new vertex(5);

$list0 = new SplDoublyLinkedList();
$list0->push(array('vertex' => $v1, 'distance' => 3));
$list0->push(array('vertex' => $v3, 'distance' => 1));
$list0->rewind();

$list1 = new SplDoublyLinkedList();
$list1->push(array('vertex' => $v0, 'distance' => 3));
$list1->push(array('vertex' => $v2, 'distance' => 7));
$list1->rewind();

$list2 = new SplDoublyLinkedList();
$list2->push(array('vertex' => $v1, 'distance' => 7));
$list2->push(array('vertex' => $v3, 'distance' => 8));
$list2->push(array('vertex' => $v4, 'distance' => 12));
$list2->rewind();

$list3 = new SplDoublyLinkedList();
$list3->push(array('vertex' => $v0, 'distance' => 1));
$list3->push(array('vertex' => $v2, 'distance' => 8));
$list3->rewind();

$list4 = new SplDoublyLinkedList();
$list4->push(array('vertex' => $v2, 'distance' => 12));
$list4->push(array('vertex' => $v5, 'distance' => 3));
$list4->rewind();

$list5 = new SplDoublyLinkedList();
$list5->push(array('vertex' => $v4, 'distance' => 3));
$list5->rewind();

$adjacencyList = array(
    $list0,
    $list1,
    $list2,
    $list3,
    $list4,
    $list5,
);

function calcShortestPaths(vertex $start, &$adjLists)
{
    // define an empty queue
    $q = new PriorityQueue();
    
    // push the starting vertex into the queue
    $q->insert($start, 0);
    $q->rewind();
    
    // mark the distance to it 0
    $start->distance = 0;
    
    // the path to the starting vertex
    $start->path = array($start->key);
    
    while ($q->valid()) {
        $t = $q->extract();
        $t->visited = 1;
        
        $l = $adjLists[$t->key];
        while ($l->valid()) {
            $item = $l->current();
            
            if (!$item['vertex']->visited) {
                if ($item['vertex']->distance > $t->distance + $item['distance']) {
                    $item['vertex']->distance = $t->distance + $item['distance'];
                    $item['vertex']->parent = $t;
                }
                
                $item['vertex']->path = array_merge($t->path, array($item['vertex']->key));
                
                $q->insert($item["vertex"], $item["vertex"]->distance);
            }
            $l->next();
        }
        $q->recoverFromCorruption();
        $q->rewind();
    }
}

calcShortestPaths($v0, $adjacencyList);

// The path from node 0 to node 5
// [0, 1, 2, 4, 5]
echo '[' . implode(', ', $v5->path) . ']';

Complexity

The complexity of that code is based on the complexity of BFS with the main difference that we keep a priority queue. For BFS we knew that the complexity was O(|V| + |E|), while Dijkstra’s algorithm has running time of O((|V| + |E|).log(|V|)). That is quite natural since the heapsort’s complexity is O(n.log(n))!

Application

Since the basic BFS can’t help us for weighted graphs and there are plenty of problems designed with weighted graphs obviously Dijkstra’s algorithm can be very handy. The only thing we should be aware of is the positive values of the edges.

Since with graphs we can represent real-life problems it’s almost clear why we would need an efficient algorithm that calculates the shortest path between two vertices. Getting back to our example of a road map we can use such an algorithm in order to find the shortest path between two cities. This example, of course, is very basic indeed, but it can give us a clear example of where shortest path can be applied.

In the other hand, we can model an enormous field of real-life problems using graphs – not only road maps. As we already know, whenever we have relations between different abstract objects we can refer an efficient graph algorithm.

OK, so we need a shortest path algorithm, but before we proceed with the exact algorithm first we’ll need to answer some questions and give some definitions.

Overview

First we need a definition of the terms distance and path between two nodes. A path is considered to be the sequence of vertices (or edges if you wish) between two vertices i and j. Of course we assume that there might be no path between any to vertices in the graph! Also we assume that this definition relates both for directed and undirected graphs. After we have the definition of a path we can proceed by defining a “distance”, which is said to be the number of edges in the path between i and j.

Using this terms, if there’s an edge between i and j, the path between them is [i, j], while the distance is 1. Of course, for an undirected graph (i, j) equals to (j, i) and the path [i, j] equals the path [j, i], but that isn’t true for directed graphs where the path [i, j] differs in general from [j, i].

Here we talk about the path between two adjacent vertices, but we can go with the more general case of a path between two vertices that aren’t adjacent.

Now, getting back to the road map example, there might be many paths between city A and city B.

What we actually need to find is the shortest one. This can be very important, because we often want to get from A to B as quickly as possible using the shortest path.

So first, what is a shortest path between i and j. Well, besides the strict definition, I’ll give a simplified one that might be clearer. The shortest path between i and j is such a path, which has the lowest distance compared to all other paths between i and j.

In our algorithm we will use breadth-first search. Why? That is because by using BFS by starting at a given point we expand our search consecutively starting with the closest vertices.

Is breadth-first search enough and will it give us the correct answer – the shortest path between i and j. Actually breadth-first search will gives us even more – the shortest paths to each reachable vertex from a given starting point – the staring vertex.

Why this is correct? Well, because of the nature of the breadth-first search algorithm. As we already know BFS uses a queue in order to store the front of the expansion. Usually as an abstraction BFS colors the vertices in white, gray and black, where the white vertices are those that aren’t visited yet, the gray are in the queue and the black vertices are already visited.

However how can be sure that BFS will give us the shortest paths to each vertex? To answer this question and to be sure that BFS will work for us we must take a closer look at the queue. Clearly by starting at a given point the algorithm is correct – the distance is 0.

Now the second step is to put into the queue all the vertices adjacent to s (where s is the starting point). Clearly this will give us the shortest paths to all adjacent vertices of s.

Continuing by induction we can assume that at level k we have all the shortest paths from s to all the vertices at the level k. It is clear the path between s and the vertices at level k is k, since we assume that each edge adds 1 to the path from s to i. Now by adding all the vertices adjacent (and not visited yet) to the paths of level k we get paths with length k+1 which is again the shortest paths from s to level k+1.

Actually we can talk about a tree built out of the graph by staring at s (which is the root of the tree).

Code

OK, now we know that BFS will find us the shortest paths from s to all the reachable vertices from s. Here’s a simple PHP implementation, that makes use of the Standard PHP Library data structures. Of course, everyone can code and use his own implementation of lists in order to keep the information of the adjacency lists.

The important thing to note is that we keep an additional information in each vertex – the distance between it and s, which is initially infinite. First we go with the modification of BFS in order to find all the distances between s and the other vertices.

Here’s our graph:

class vertex
{
    public $key = null;
    public $color = 'white';
    public $distance = -1;  // infinite
    
    public function __construct($key) 
    {
        $this->key = $key;
    }
}

$v0 = new vertex(0);
$v1 = new vertex(1);
$v2 = new vertex(2);
$v3 = new vertex(3);
$v4 = new vertex(4);
$v5 = new vertex(5);

$list0 = new SplDoublyLinkedList();
$list0->push($v1);
$list0->push($v3);
$list0->rewind();

$list1 = new SplDoublyLinkedList();
$list1->push($v0);
$list1->push($v2);
$list1->rewind();

$list2 = new SplDoublyLinkedList();
$list2->push($v1);
$list2->push($v3);
$list2->push($v4);
$list2->rewind();

$list3 = new SplDoublyLinkedList();
$list3->push($v1);
$list3->push($v2);
$list3->rewind();

$list4 = new SplDoublyLinkedList();
$list4->push($v2);
$list4->push($v5);
$list4->rewind();

$list5 = new SplDoublyLinkedList();
$list5->push($v4);
$list5->rewind();

$adjacencyList = array(
    $list0,
    $list1,
    $list2,
    $list3,
    $list4,
    $list5,
);

function calcDistances(vertex $start, &$adjLists)
{
    // define an empty queue
    $q = array();
    
    // push the starting vertex into the queue
    array_push($q, $start);
    
    // color it gray
    $start->color = 'gray';
    
    // mark the distance to it 0
    $start->distance = 0;
    
    while ($q) {
        // 1. pop from the queue
        $t = array_pop($q);
        
        // 2. foreach poped item find it's adjacent white vertices
        $l = $adjLists[$t->key];
        while ($l->valid()) {
            // 3. mark them gray, increment their length with one from their parent
            if ($l->current()->color == 'white') {
                $l->current()->color = 'gray';
                $l->current()->distance = $t->distance + 1;
                // 4. push them to the queue
                array_push($q, $l->current());
            }
            
            $l->next();
        }
    }
}

calcDistances($v0, $adjacencyList);

print_r($adjacencyList);

Now we can modify the algorithm even more and we add the path property of each vertex. Now each vertex will keep the path from s.

class vertex
{
    public $key         = null;
    public $color       = 'white';
    public $distance    = -1;  // infinite
    public $path        = null;
    
    public function __construct($key) 
    {
        $this->key  = $key;
    }
}

$v0 = new vertex(0);
$v1 = new vertex(1);
$v2 = new vertex(2);
$v3 = new vertex(3);
$v4 = new vertex(4);
$v5 = new vertex(5);

$list0 = new SplDoublyLinkedList();
$list0->push($v1);
$list0->push($v3);
$list0->rewind();

$list1 = new SplDoublyLinkedList();
$list1->push($v0);
$list1->push($v2);
$list1->rewind();

$list2 = new SplDoublyLinkedList();
$list2->push($v1);
$list2->push($v3);
$list2->push($v4);
$list2->rewind();

$list3 = new SplDoublyLinkedList();
$list3->push($v1);
$list3->push($v2);
$list3->rewind();

$list4 = new SplDoublyLinkedList();
$list4->push($v2);
$list4->push($v5);
$list4->rewind();

$list5 = new SplDoublyLinkedList();
$list5->push($v4);
$list5->rewind();

$adjacencyList = array(
    $list0,
    $list1,
    $list2,
    $list3,
    $list4,
    $list5,
);

function calcShortestPaths(vertex $start, &$adjLists)
{
    // define an empty queue
    $q = array();
    
    // push the starting vertex into the queue
    array_push($q, $start);
    
    // color it gray
    $start->color = 'gray';
    
    // mark the distance to it 0
    $start->distance = 0;
    
    // the path to the starting vertex
    $start->path = new SplDoublyLinkedList();
    $start->path->push($start->key);
    
    while ($q) {
        // 1. pop from the queue
        $t = array_pop($q);
        
        // 2. foreach poped item find it's adjacent white vertices
        $l = $adjLists[$t->key];
        while ($l->valid()) {
            // 3. mark them gray, increment their length with one from their parent
            if ($l->current()->color == 'white') {
                $l->current()->color = 'gray';
                $l->current()->distance = $t->distance + 1;
                $l->current()->path = clone $t->path;
                $l->current()->path->push($l->current()->key);
                
                // 4. push them to the queue
                array_push($q, $l->current());
            }
            
            $l->next();
        }
    }
}

calcShortestPaths($v0, $adjacencyList);

print_r($adjacencyList);

Complexity

Clearly the complexity of enqueue and dequeue is O(V), while searching for adjacent vertices is O(E), thus the complexity of this algorithm is O(V + E)!

Application

Finding the shortest path between two nodes is obviousely a very handy algorithm. Applied almost everywhere graphs exists this algorithm is widely used. However there’s one very reasonable question. We’re searching for the shortest path between two vertices and we end with the shortest paths between a starting node an all other vertices? Why we need this “useless” information? Acutally the question should be: is there a faster and more efficient algorithm compared to this one. Well, we’ll see that!

Let’s assume we have a list of tasks to accomplish. Some of the tasks depend on others, so we must be very careful with the order of their execution. If the relationship between these tasks were simple enough we could represent them as a linked list, which would be great, and we would know the exact order of their execution. The problem is that sometimes the relations between the different tasks are more complex and some tasks depend on two or more other tasks, which in their turn depend on one or more tasks, etc.

Thus we can’t model this problem using linked lists or trees. The only rational solution is to model the problem using a graph. What kind of graph do we need? Well, we definitely need a directed graph, to desribe the relations, and this graph shouldn’t have cycles. So we need the so called directed acyclic graph (DAG).

Why we don’t what a cycle in the graph? The answer of this question is simple and obvious. In case of cyclic graph, we wouldn’t be able to determine the priority of task execution, thus we won’t be able to sort the tasks properly.

Now the solution we want is to sort the vertices of the graph in some order so for each edge (u, v) u will precede v. Then we’ll have a linear order of all tasks and by starting their execution we’ll know that everything will be OK.

This kind of sort is also known as “topological” sort (or topsort) and it is one of the very basic graph algorithms.

Overview

OK, so we have an acyclic directed graph, how do we proceed to get a linked list with all the vertices sorted? Since it’s an acyclic graph we know that there is at least one vertex without predecessor. Thus at first place, we can put all the vertices without predecessors into our linked list.

This approach answers the question – is there a possibility to have more than one valid topological sort of a graph? Indeed, the only thing we’d like to do is to put all the vertices in the correct order, but since there might be vertices with no predecessors any combination of them will be a valid topological sort for a graph.

As we can see from the picture above even for vertices with predecessors our topological sort can vary. Thus [9, 6, 2, 7, 4, 1] is a valid topological sorted graph, but [6, 9, 2, 7, 4, 1] is also a valid topological sort out of the same graph!

Now we can generalize the algorithm in some basic steps.

1. Make an empty list L and an empty list S;
2. Put all the vertices with no predecessors in L;
3. While L has items in it;
3.1. Pop an item from L – n, and push it to S;
3.2. For each vertex m adjacent to n;
3.2.1. Remove (n, m);
3.2.2. If m has no predecessors – push it to L;

Code

Here’s the very basic PHP implementation. As you can see the short implementation shows us how easy this algorithm is. However its importance to computer science and programming is enormous.

class G
{
    protected $_g = array(
        array(0, 1, 1, 0, 0, 0, 0),
        array(0, 0, 0, 1, 0, 0, 0),
        array(0, 0, 0, 0, 1, 0, 0),
        array(0, 0, 0, 0, 1, 0, 0),
        array(0, 0, 0, 0, 0, 0, 1),
        array(0, 0, 0, 0, 0, 0, 1),
        array(0, 0, 0, 0, 0, 0, 0),
    );
    protected $_list = array();
    protected $_ts   = array();
    protected $_len  = null;
    
    public function __construct()
    {
        $this->_len = count($this->_g);
        
        // finds the vertices with no predecessors
        $sum = 0;
        for ($i = 0; $i < $this->_len; $i++) {
            for ($j = 0; $j < $this->_len; $j++) {
                $sum += $this->_g[$j][$i];
            }
            
            if (!$sum) {
                // append to list
                array_push($this->_list, $i);
            }
            $sum = 0;
        }
    }
    
    public function topologicalSort() 
    {
        while ($this->_list) {
            $t = array_shift($this->_list);
            array_push($this->_ts, $t);
            
            foreach ($this->_g[$t] as $key => $vertex) {
                if ($vertex == 1) {
                    $this->_g[$t][$key] = 0;
                    
                    $sum = 0;
                    for ($i = 0; $i < $this->_len; $i++) {
                        $sum += $this->_g[$i][$key];
                    }
                    
                    if (!$sum) {
                        array_push($this->_list, $key);
                    }
                }
                $sum = 0;
            }
        }
        
        print_r($this->_ts);
    }
}

$g = new G();
/*
Array
(
    [0] => 0
    [1] => 5
    [2] => 1
    [3] => 2
    [4] => 3
    [5] => 4
    [6] => 6
)*/
$g->topologicalSort();

Application

As I already mentioned above this algorithm is practically used to sort the execution of different tasks that depend on each other. However this isn’t its only use. Actually any kind of objects that depend on each other can be modeled with a graph. Indeed sometimes these graphs may be a trees, but most of the cases that isn’t true.