Havel Hakimi

Havel-Hakimi Algorithm
July 28, 2010

Abstract We study the problem of existence of a graph whosse degree sequence is given. The obvious condition that the sum of all the degrees must be even (Hand-Shaking Lemma) turns out not to be sucient. A recursive algorithm that produces a graph that corresponds to the given degree sequence is discussed. The same technique is used to prove that two graphs with the save vertex set can be transformed into each other by a series of 2-switches.
Denition. The degree sequence of a graph is the list of vertex degrees written in nonincreasing order, as d1 dn . Denition. A graphic sequence is a list on nonnegative numbers that is the degree sequence of some simple graph. A simple graph with degree sequence d realizes d. Theorem. For n > 1, an integer list d = ( = d1 , . . . dn ) of size n is graphic if and only if d = (d2 1, . . . , d+1 1, d+2 , . . . , dn ) is graphic. Proof of suciency. Given a simple graph G whose degree sequence is d , we add a new vertex adjacent to vertices in G with degrees d2 1, . . . , d+1 1. Proof of necessity. Let G be a simple graph realizing d. Let w be a vertex of degree in G. Let S be a set of vertices in G having degrees d2 ,. . . ,d+1 . In N (w) = S, then we delete w to obtain G that realizes d . Otherwise, we claim that some vertex of S is missing from N (w)that is, there is a vertex in S that is not in N (w): if not, then any vertex of S in N (w), that is, S N (W ). Since |S| = |N (w)| = , we have S = N (w), contradiction. In this case, we can modify G to increase |N (w) S| N (w) = without changing any vertex degree. Repeating this converts G into another G that realizes d and has S as the neighborhood of w. From G we then delete w to obtain the desired graph G realizing d . To nd the modication when N (w) = S (which means N (w) S and S N (w)) we can choose x S \ N (w) and z N (w) \ S so that z w, x w. Since x S and z S, we have d(x) d(z). We claim that / there exists a vertex y adjacent to x but not to z: for if not, every vertex y adjacent to x is also adjacent to z. That is, N (x) N (z). Since w N (z) and w N (x), we have / d(z) = |N (z)| |N (x) {w}| = |N (x)| + |{w}| = d(x) + 1
which contradicts d(z) d(x) obtained above. Now we delete {wz, xy} and add {wx, wz} to increase |N (x) S|. Denition. A 2-switch is the replacement of a pair of edges xy and zw in a simple graph by the edges yz and wx, given that yz and wx did not appear in the graph originally. Note that xz and yw would be a 2-switch, too, provided that the new edges did not appear in the graph originally. Clearly, a 2-switch preserves all vertex degrees. Also if some 2-switch turns H into H , then a 2-switch on the same four vertices turn H H.
Theorem. If G and H are two simple graphs with vertex set V , then dG (v) = dH (V ) (that is, the two graphs have the same sequence degree) for every v V if and only if there is a sequence of 2-switches that transforms G into H. Proof. Suciency is obvious since every 2-switch preserves vertex degrees. Suppose, conversely, that dG (v) = dH (v) for all v V . If n 3, then for any d1 , . . . , dn there is at most one simple graph with d(vi ) = di (for n = 3 we have 211 and 222). We mean to use induction and n = 3 can be used as the basis step. Consider n 4, and let w be a vertex of maximum degree, . Let S = {v1 , . . . , v } be a xed set pf vertices with the highest degrees other than w. As we saw above, some sequence of 2-switches transforms G to a graph G such that NG (w) = S, and H is transformed into H with NH (w) = S. Now since NH (w) = S = NG (w), deleting w leaves simple graphs G = G w and H = H w with equal degree sequences: dG (v) = dH (v) for every vertex v. By the induction hypothesis, some sequence of 2switches transforms G to H . These transformations do not involve, and we know that w has the same neighbors in G and H applying these transformation to G and H (rather than to G and H ) transforms G to H . Since 2-switches are reversible, G G H H.
Denition. A partition of r is represented by a sequence = (1 , . . . , n ), where 1 + + n = r and 1 n 0. This convention is denoted by r Denition. The partition 2m is graphic if there is a graph G = (V, E), |E| = 2m, whose degree sequence d(G) = . Suppose = (1 , 2 , . . . , n ) 2m. Let = 1 and denote by 1 the sequence obtained by rearranging (if necessary) the integers 2 1, 3 1,. . . ,+1 1, +2 ,. . . ,n into nonincreasing order. If +1 1, then 1 2(m ): in fact, if +1 1, then 1 1,. . . , 1, and (2 1) + (3 1) + + ( 1) + (+1 1) + +2 + + n = 2m 1 = 2m 2 = 2(m ). Theorem (Havel-Hakimi). Suppose = (1 , . . . , n ) 2m. Then is graphic if and only if 1 is graphic. Proof. Suppose 1 is graphic. That is, there exists a graph H such that d(H) = 1 . We may assume that V (H) = {v2 , . . . , vn }, where dH (vi ) = i 1, 2 i +1 1 = 1 + 1, and dH (vi ) = , = 1 + 2 = +2 i n. Let G be the graph obtained from H by a new vertex v1 , and new edges v1 vi , 2 i + 1. Then d(G) = , proving that the condition is sucient. Conversely, suppose d(G) = , so that = 1 = (G). Let u V (G) be a vertex of degree . Let W be a set of vertices having degrees 2 ,. . . ,+1 . If NG (u) = W , i.e. if uw E(G) for all w W , then the graph H = G u satises 1 = d(H), and the proof is nished. Otherwise, there exist vertices w W and v W such that uv E(G) / and uw E(G). Suppose not: that is, for all z V (G), if wz E(G) then / vz E(G) which yields N (w) N (v). But since u N (v) and u N (u) / we have that d(w) = |N (w)| |N (v)| + 1 = d(w) + 1 which contradicts to d(v) d(w). Let G1 be the graph obtained from G by deleting edges uv and wz, and adding new edges vz and uw. Then V (G1 ) = V (G) and, because each vertex has the same in G1 as in G, d(G1 ) = d(G) = . The important dierence is that |W NG1 (u) | > |W NG (u)| which follows from the fact that W has not changed, and one vertex (w) that was not in N (u) from W was inserted into N (u). If NG1 (u) = W , then H = G1 u is a graph whose degree sentence is 1 . Otherwise another switch will yield a graph G2 with |W NG2 (u) | > |W NG1 (u)|. This process cannot be innite: vertices from N (u) that are not in W are being removed, and vertices from W that are not in N (u) are being inserted into W . At each step the vertex step of a new graph and its degree sequence do not change. Eventually we will get a graph Gk such that = d(Gk ), and NGk (u) = W . At this point the graph H = Gk u will be a graph with degree sequence d(H) = 1 .

Havel Hakimi

Uploaded by

Copyright:

Available Formats

Havel Hakimi

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Havel Hakimi

Uploaded by

Copyright:

Available Formats

What is the Havel-Hakimi algorithm used for?

What is a 2-switch and how does it relate to graphs?

Havel-Hakimi Algorithm

July 28, 2010

You might also like