The Bellman-Ford Algorithm 笔记

1. Single-Source Shortest Paths Revisited
   1.1 The Single-Source Shortest Path Problem.
   Input: Directed graph $G=(V,E)$, edge lengths $c_e$ for each $e \in E$, source vertex $s \in V$. [Can assume no parallel edges].
   1.2 Goal: For every destination $v \in V$, compute the length(sum of edge costs) pf a shortest s-v path.
2. On Dijkstra’s Algorithm
   2.1 Good news: $O(mlogn)$ running time using heaps.
   Bad news:
   (1) Not always correct with negative edge lengths.
   (2)Not very distributed (relevant for Internet routing)
   Solution: The Bellman-Ford algorithm.
   On Negative Cycles.
   Question:How to define shortest path when $G$ has a negative cycle?
   Soution #1: Compute the shortest s-v path, which cycles allowed.
   Problem: Undefined or $-\infty$.
   Solution #2: Computes shortest cycle-free s-v path.
   Problem: NP-hard (no polynomial algorithm, unless P=NP)
   Solution #3: (For now) Assume input graph has no negative cycles.
   Single-Source Shortest Path Problem, Revisited.
   Input: Directed graph $G=(V,E)$, edge lengths $c_e$,
   source vertex $s \in V$.
   Goal: either
   (A) For all destinations $v \in V$ ,compute the length of a shortest $s-v$ path. OR
   (B) Output a negative cycle.
   Optimal Substructure(Informal)
   Intuition: Exploit sequential nature of paths. Subpath of a shortest path should itself be shortest.
   Issue: Not clear how to define “smaller” & “larger” subproblems.
   Key idea: Artificially restrict the number of edges in a path.
   Subproblem size $\Leftrightarrow$ Number of permitted edges.
   Lemma: Let $G=(V,E)$ be a directed graph with edge lengths $c_e$ and source vertex s.
   [G might or might not have a negative cycle]
   For every $v \in V, i \in \{1,2...\}$ , let $P=$ shortest $s-v$ path with at most $i$ edges. (Cycles are permitted)
   Case 1: If P has $\leq (i-1)$ edges, it is a shortest $s-v$ path with $\leq (i-1)$ edges.
   Case 2: If P has $i$ edges with last hop($w$,$v$), then $P^{'}$ is a shortest $s-w$ path with $\leq (i-1)$ edges.
   The Basic Algorithm
   The Recurrence
   Notation: Let $L_{i,v} =$ minimum length of a s-v path with $\leq i$ edges.
   $-$ With cycles allowed
   $-$ Defined as $+ \infty$ if no $s-v$ paths with $\leq i$ edges.
   Recurrence: For every $v \in V, i \in \{ 1,2,...\}$

$$\begin{equation} L_{i,v} = min \begin{cases} L_{(i-1), v} \\ min_{(u,v)\in E} \{L_ {(i-1),w} + c_{wv} \}, \end{cases} \end{equation}$$
If No Negative Cycles
Now: Suppose input graph $G$ has negative cycles.
$\Rightarrow$ Shortest paths do not have cycles.
[removing a cycle only decrease length]
$\Rightarrow$ Have $\leq (n-1)$ egdes.
Point: If $G$ has negative cycle, only need to solve subproblems up to $i=n-1$.
Point: If $G$ has no negative cycle, only need to solve subproblems up to $i=n-1$.
Subproblems:
Compute $L_{i,v}$ for all $i \in \{0,1,...,n-1 \}$ and all $v \in V$.
The Bellman-Ford Algorithm
(1) Let $A = 2-D$ array(indexed by $i$ and $v$)
Base case:
$A[0,s] = 0; A[0,v] = + \infty$ for all $v \neq s$/
For $i = 1, 2 ,...,n-1$
For each $v \in V$
$$\begin{equation} L_{i,v} = min \begin{cases} L_{(i-1), v} \\ min_{(u,v)\in E} \{L_ {(i-1),w} + c_{wv} \}, \end{cases} \end{equation}$$
As discussed: IF G has no negative cycle, then algorithm is correct.
[with final answer = $A[n-1,v]'s$ ]

Whatis the running time of the Bellman-Ford algorithm?
Reason: Total work is $O(n \sum_{v \in V} in.deg(v))$ = $O(mn)$入度。
Checking for a Negative Cycle
Question: What if the input graph G has a negative cycle.
Claim:
G has a no negative-cost cycle $\Leftrightarrow$ In the extended Bellman-Ford algorithm,
$A[n-1,v] = A[n,v]$ for all $v \in V$.
Consequence: Can check for a negative cycle just by running Bellman-Ford for one extra iteration(running time still $O(mn)$).
Proof of Claim
($\Rightarrow$) Already proved in correctness of Bellman-Ford.
($\Leftarrow$) Assume $A[n-1,v] = A[n,v]$ for all
$v \in V$
Consequence: Can check for a negative cycle just by jsut running Bellman-Ford for one extra iteration.
(running time still $O(mn)$) .
Let $d(v)$ denote the common value of $A[n-1,v]$ and $A[n,v]$.
Recall algorithm:

$$\begin{equation} A[n,v]= min \begin{cases} A[i-1, v] \\ min_{(w,v)\in E} \{A[n-1,w] + c_{wv} \}, \end{cases} \end{equation}$$
其中 $A[n,v] \rightarrow d(v) , A[n-1,w] \rightarrow d(w)$
Thus: $d(v) \leq d(w) + c_{wv}$ for all edges $(w,v) \in E$
Equivalently : $d(v) -d(w) \leq c_{wv}$