k shortest paths

Finds the k quickest routes from one vertex to another that don't repeat a vertex, quickest first, the way a navigation app offers a few routes.

Time complexity O(k·V·(V+E) log V), gograph function path.KShortest.

How it works

A navigation app rarely shows one route. It offers the quickest and two or three others that aren't much slower, so the driver can avoid a bridge or a busy street. Yen's algorithm finds the k quickest routes that don't repeat a vertex. It takes the quickest route, then at each vertex along it blocks the road that route takes from there, and the roads the routes found before take from the same start of the route, and finds the quickest way around. Each way around is a candidate, and the quickest candidate is the next route.

gograph runs it with Lawler's change: a route only branches from the vertex where it left the route it came from, so the vertices before that are never searched again and no route is found twice. It keeps at most k-1 candidates, since no more can be used, and skips a detour that can't beat the slowest of them.

Each detour search is A* guided by the exact time from each vertex to the goal, found once by one backward search at the start. Blocking roads only makes routes slower, so that guess never says more than the time left. Time is O(k·V·(V+E) log V) in the worst case, and the detours stay short in practice.

path.KShortest returns the routes without their costs, so the page adds up the minutes itself. Fewer than k routes is not an error, and k below 1 returns ErrInvalidK. On a street grid, many of the quickest routes differ by a single block, so the page shows how much of each route it shares with the first.

San Francisco streets

The main streets of San Francisco from the Embarcadero to Van Ness Avenue and the Mission, from OpenStreetMap, with the freeways that cross SoMa. Each intersection is named after its streets, one-way streets go one way, and each edge is the minutes to drive it at the speed limit, or at 22 to 40 km/h by road type where the map has no limit.

A directed, weighted graph with 248 vertices and 548 edges.

Map data © OpenStreetMap contributors.

What are the three quickest drives from The Embarcadero & Mission St to 16th St & S Van Ness Ave that don't pass the same place twice, and how much slower is each one than the first?

The three quickest drives from The Embarcadero & Mission St to 16th St & S Van Ness Ave take 6.42, 6.72 and 6.81 minutes. Route 2 is 0.3 minutes slower and shares 99% of route 1's time. Route 3 is 0.39 minutes slower and shares 83% of route 1's time.

Step by step

  1. First, one search backward from 16th St & S Van Ness Ave finds the exact time to it from each of the 248 vertices The Embarcadero & Mission St reaches: 248 of them can get there. Each detour search uses that time as its guess, and blocking roads only makes routes slower, so the guess never says more than the time left.
  2. An A* search from The Embarcadero & Mission St finds the quickest route, 6.42 minutes, after taking 25 vertices from its queue. It is the first candidate.
  3. Route 1 is the quickest candidate: 6.42 minutes.
  4. Detour from The Embarcadero & Mission St: block The Embarcadero & Mission St to The Embarcadero & Howard St, the road the route takes from here. An A* search from The Embarcadero & Mission St, at 0 minutes so far, takes 32 vertices from its queue and finds a way around: a candidate of 8.68 minutes.
  5. Detour from The Embarcadero & Howard St: block The Embarcadero & Howard St to Howard St & Fremont St, the road the route takes from here, and The Embarcadero & Mission St before The Embarcadero & Howard St so the detour can't loop back. An A* search from The Embarcadero & Howard St, at 0.25 minutes so far, takes 27 vertices from its queue and finds a way around: a candidate of 7.13 minutes.
  6. Detour from Howard St & Fremont St: block Howard St & Fremont St to 1st St & Howard St, the road the route takes from here, and The Embarcadero & Mission St and The Embarcadero & Howard St before Howard St & Fremont St so the detour can't loop back. An A* search from Howard St & Fremont St, at 0.98 minutes so far, takes 25 vertices from its queue and finds a way around: a candidate of 8.23 minutes. Only 2 candidates can still be used, so drop the slowest, 8.68 minutes.

32 more steps lead to the last one:

  1. Done. The three quickest routes take 6.42, 6.72 and 6.81 minutes.

Use it in Go

routes, err := path.KShortest(g, "The Embarcadero & Mission St", "16th St & S Van Ness Ave", 3, nil)
if err != nil {
	return err
}
for _, route := range routes {
	minutes := 0.0
	for i := 1; i < len(route); i++ {
		minutes += g.GetEdge(route[i-1], route[i]).Weight()
	}
	fmt.Println(minutes, len(route))
}

Example graphs

More in Shortest paths

  • Dijkstra: Finds the shortest distance from a start vertex to every other vertex when no edge weight is negative.
  • A*: Finds the quickest route from one vertex to another, looking first where a guess of the time left says the goal is, so it settles far fewer vertices than Dijkstra.
  • Bidirectional search: Finds the quickest route from one vertex to another by searching forward from the start and backward from the goal at once, until the two searches meet.
  • Beam search: Looks for a route from one vertex to another while keeping only the few best partial paths at each step, so its memory and time stay bounded, at the cost of sometimes missing the best route.
  • Multi-source Dijkstra: Finds the nearest of several sources for every vertex, and the distance from it, in one search.
  • k-center: Chooses k centers so that the vertex farthest from its nearest center is as near as it can make it.
  • Bellman-Ford: Finds the shortest distance from a start vertex to every other vertex, even with negative weights, and reports a negative cycle when the start can reach one.
  • Floyd-Warshall: Finds the shortest distance between every pair of vertices, even with negative weights, by letting paths stop at one more vertex in each round.

All algorithms