Scribd
Upload
3 views5 pages

Advanced Algorithms Lecture 10.0

The document discusses advanced algorithms for solving the Minimum Cost Circulation Problem, focusing on Klein's Cycle Canceling Algorithm and the Goldberg-Tarjan Algorithm. It outlines theorems related to minimum cost circulations and provides proofs for their correctness, as well as the complexities involved in finding optimal solutions. The document also compares different approaches to the Max-Flow Problem and their computational implications.

Uploaded by

mhmad240
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
3 views5 pages

Advanced Algorithms Lecture 10.0

The document discusses advanced algorithms for solving the Minimum Cost Circulation Problem, focusing on Klein's Cycle Canceling Algorithm and the Goldberg-Tarjan Algorithm. It outlines theorems related to minimum cost circulations and provides proofs for their correctness, as well as the complexities involved in finding optimal solutions. The document also compares different approaches to the Max-Flow Problem and their computational implications.

Uploaded by

mhmad240
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 5

MIT OpenCourseWare

http://ocw.mit.edu

6.854J / 18.415J Advanced Algorithms


Fall 2008
��

For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.
18.415/6.854 Advanced Algorithms October 17, 2001

Lecture 10
Lecturer: Michel X. Goemans Scribe: Ashish Koul

1 Minimum Cost Circulation Problem


Theorem 1 Let f be a circulation. The following are equivalent:
(i) f is of minimum cost.
(ii) Gf has no negative cost directed cycles.
(iii) 3p : cp(u,w) 20 V(u, w) E Ef, where cp(v,w) = c(u,w) + p(u) - p(w).

Proof: i j ii and iii +- i were proven last lecture. All that remains is the proof of ii +- iii:
Let G' be obtained from the residual graph Gf by adding a vertex s linked t o all other vertices by
edges of cost 0 (the costs of these edges do not matter). Let p(u) be the length of the shortest path
from s to v in G' with respect to the costs.
These quantities are well-defined since Gf does not contain any negative cost directed cycles, and
+
every vertex is reachable from s. By definition of the shortest paths, p(w) 5 p(u) c(u,w) V(u,w) E
>
Ef. This implies that cp(u,w) 0 V(u,w) E Ef.

2 Klein's Algorithm for MCCP


Klein's Cycle canceling algorithm:
1. Let f be any circulation.
2. While Gf contains a negative cycle r do
push 6 = min(,,,)Er vf (u,w) along I?.
Argument for Correctness:

If the algorithm terminates, then the circulation found must be optimum. Furthermore, if all ca-
pacities and costs are integers, then the algorithm will terminate.

Why?
f (v, w) is always an integer, thus 6 = min(,,,)Er uf(u,w) 2 1
If Ic(u,w) I 5 C and If (u,w) 1 5 U , then the absolute value of the cost of the optimal circulation
is at most mCU
Therefore, the algorithm terminates after 0 (mCU) iterations.

Remark 1 If the edge capacities i n the graph are irrational, then the algorithm is not correct.

The cycle canceling algorithm can be applied to the Max-Flow Problem by making appropriate
modifications to the graph G. Let G' be obtained by setting the cost of all edges within G to 0.
Furthermore, select two vertices s and t from within the graph, and add the directed edges (s, t )
and (t,s), where c(s, t ) = 1,c(t, s) = -1 and both edges have infinite capacity. Now, solving for
the maximum flow between s to t is equivalent to solving for the minimum cost circulation, which
contains s and t. In this circumstance, Klein's Algorithm reduces to the Ford-Fulkerson Algorithm.

Ford-Fulkerson Augmenting Path Algorithm:


1. Begin with zero flow: f = 0.
2. While Gf contains a directed path P from s to t do
push 6 = min(,,,),p u ( v ,w ) along P.
The running time given above for Klein's Cycle-Canceling Algorithm is not polynomial. The nega-
tive cost cycle in Klein's Algorithm (or the directed path in the Ford-Fulkerson Algorithm) must be
chosen appropriately to insure a polynomial running time.

Candidates for Cycles in Klein's Algorithm:


1. The most negative cost cycle in Gf ?
Finding this cycle is an NP-Hard problem, so it would not be a viable choice.
2. The negative cycle in Gf which would yield the maximum cost improvement?
Finding this cycle is again an NP-Hard problem.
However for the Max-Flow Problem, this choice reduces to finding the st-path with maximum
residual capacity. Such a path can be found in O(m) time, m = IEl. The resulting Max-
Flow algorithm is known as the '<fattestvpath algorithm (Edmonds-Karp '72). The number
of iterations necessary is O(m log U), thus the running of the algorithm is O(m2log U).
3. Minimum Mean-Cost Cycle?
Define the mean cost of a cycle r to be:

where Irl denotes the number of edges in r. The minimum mean cost of all cycles of the
residual graph Gf would thus be:

min c(r)
-
'(') = cycles T. in G~ lr 1
The minimum mean-cost cycle can be determined in strongly polynomial time by using a
modified version of the Bellman-Ford Algorithm. More precisely, the minimum mean cost
cycle can be found in O(mn) time. Using this method to solve the Min-Cost Circulation
Problem yields the Goldberg-Tarjan Algorithm, which runs in polynomial time. Using this
method to solve the Max-Flow Problem yields what is known as the "shortest" augmenting
path algorithm (Edmonds-Karp). This Max-Flow Algorithm is able to find the augmenting
path in O(m) time, and requires O(mn) iterations to arrive at the solution. Thus, the total
running time is 0(m2n) .
3 The Goldberg-Tarjan Algorithm
Goldberg-Tarjan Algorithm:
1. Begin with zero flow: f = 0.
2. While p(f) < 0 do
push 6 = min(,,,),I- uf (u, w) along a minimum mean cost cycle r of Gf .
Analysis of Goldberg-Tarjan Algorithm:

In order to analyze this algorithm, it is necessary to define the concept of proximity measure for a
circulation f .

Definition 1 A circulation f is E-optimali f there exists p such that cp(u,w) 2 -E V(u,W)E Ef .

Definition 2 ~ ( f =
) minimum E such that f is E-optimal.

The following theorem states that the minimum mean cost p( f ) of all cycles in Gf is equal to -E( f ) ,
as defined above.

Theorem 2 For any circulation f, p( f ) = -E( f ) .

Proof:

~ ( f 2) - 4 f )
>
By definition, there exists p such that cp(u,w) -E( f ) V(v,w) E Ef . Thus, it is implied that
cP(r) 2 -~(f)lI'l for any directed cycle r E G f . But for any I' E G f , c ( r ) = cp(r). Thus,
dividing both sides by Irl yields that the mean cost of any directed cycle I' E Gf is at least
- ~ ( f ) .Therefore, p(f) 2 - ~ ( f ) .

4 61 I-4f 1
Consider p(f). For every cycle r E G f , it is the case that >
p(f). Let cl(v,w) =
C(U,W) - p(f ) V(u, W) E Ef . With respect to this new cost function c' every cycle I' E Gf
will have nonnegative cost. Now, let G' be obtained by adding a new node s to Gf and
adding directed edges from s to u Vu E V, all with zero cost. Let p(u) be the cost with
respect to c' of the shortest path from s to u in the new graph G'. For all edges (u,w),
>
p(w) Ip(u)+c'(u,w) =p(u)+c(u,w)-p(f). This implies that cp(v,w) p(f) V(u,w) E Ef.
Therefore, ~ ( f 5
) -p(f).

Remark 2 Along the minimum mean cost cycle I', cp(v,w) = -E( f ) V(v,w) E I'.
Having completed the necessary definitions and proofs, we may now proceed with the analysis of the
Goldberg-Tarjan Algorithm. The following theorem considers only one iteration of the algorithm.

Theorem 3 Let f be a circulation and let f ' be the circulation obtained b y canceling the minimum
<
mean cost cycle I' of Gf . Then, E(f ') E(f ) .

>
Proof: By definition, there existspsuch that cp(u,w) - ~ ( f )V(u,w) E E f . In the caseof the
minimum mean cost cycle I' of Gf , cp(u,w) = -E( f ) V(u,w) E I?. After performing the one cycle-
>
canceling step, we obtain the new residual graph Gft . We claim that cp(u,w) - ~ ( f )V(v,w) E Ef,.
n
In the case of all edges (u,w) E Eft Ef, the claim is certainly true. In the case of all edges
(u,w) E Ef/\Ef,it must be true that (w,u) E I?. For all (w,v) E I?, cp(w,u) = - ~ ( f ) ,and thus
cP(v,w) = ~ ( f 2
) - ~ ( f ) . Therefore, c,(u,w) 2 - ~ ( f )holds true for all (v,w) E E f , .

The above theorem shows that by completing a single iteration of the Goldberg-Tarjan algorithm,
it is impossible to generate a new flow which is farther from optimality than the original.

You might also like

pFad - Phonifier reborn

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.


Alternative Proxies:

Alternative Proxy

pFad Proxy

pFad v3 Proxy

pFad v4 Proxy