Browse · MathNet
PrintBulgarian Winter Tournament
Bulgaria counting and probability
Problem
Given a natural number . To find the smallest real number with the following property: If is a connected graph with vertices and edges, then it is always possible to delete no-more than edges so that vertices can be colored in two colors and every undeleted edge has multi-colored vertices. (Alexander Ivanov)
Solution
Lemma: Let be a connected graph with at least vertices. Then either there exist two vertices connected by an edge whose removal (along with the outgoing edges) leaves connected, or there exist two vertices of degree (i.e., "leaves"). Consider an arbitrary "covering tree" of and take as its "root" an arbitrary vertex that is not a "leaf". Let be the farthest vertex from the "root" and be its "ancestor". Let be the "successors" of . Clearly, they are all leaves on the tree.
Case 1. Among there are two vertices connected by an edge in . Then removing these two vertices leaves the tree (and therefore ) connected.
Case 2. Among there are two vertices that are leaves in . Then there are indeed at least two leaves in the output graph.
Case 3. Among there is at most one vertex that is a leaf in (b.o.o., let it be ). Then let us "connect" each of to an arbitrary vertex in other than (such vertices exist, and none of these "connecting" edges are part of the covering tree, due to the extreme choice of , i.e., each of them is part of a loop, all remaining edges of which are from the covering tree). Now we can remove and and we will have a spanning tree again, so remains connected.
This proves the lemma. With its help we can easily prove the following
Assertion: Let be a connected graph with vertices. Then we can color its vertices in two colors, so that if and are the number of "multicolored" and "single colored" edges, respectively, then .
Proof: For the statement is immediately verified. Let and be a connected graph with vertices. Let and be the two vertices from the Lemma. We "remove" and and color according to the induction hypothesis. Now, it is not difficult to see that we can color such that the difference under consideration increases by at least . Indeed, this is clear if and are leaves, and otherwise case, considering the parity of the number of neighbors of in , we see that there is always such a way. The statement is proved by induction.
Let us now consider an arbitrary connected graph with vertices and edges. We "color" it according to the Assertion: we have ; , therefore and deleting edges satisfies the condition. Thus, we got .
To show that let us consider the complete graph with vertices. A necessary and sufficient condition for having the coloring from the condition is that the graph obtained after deleting the edges is bipartite. Indeed, there must be no cycles of odd length in the graph, which is equivalent to the above.
Case 1. , . Then . To reach a bipartite graph, we need to delete all the edges in each of the two groups of vertices, and the number of deleted edges is minimal when the two groups are of equal power and contain vertices. So, we need to delete at least
Case 2. , . Similarly, here we need to delete at least edges and again $$ n_1^2 \le k \cdot \left( \frac{2n_1(2n_1+1)}{2} - n_1 \right) \implies k \ge \frac{1}{2}. \quad \square
Case 1. Among there are two vertices connected by an edge in . Then removing these two vertices leaves the tree (and therefore ) connected.
Case 2. Among there are two vertices that are leaves in . Then there are indeed at least two leaves in the output graph.
Case 3. Among there is at most one vertex that is a leaf in (b.o.o., let it be ). Then let us "connect" each of to an arbitrary vertex in other than (such vertices exist, and none of these "connecting" edges are part of the covering tree, due to the extreme choice of , i.e., each of them is part of a loop, all remaining edges of which are from the covering tree). Now we can remove and and we will have a spanning tree again, so remains connected.
This proves the lemma. With its help we can easily prove the following
Assertion: Let be a connected graph with vertices. Then we can color its vertices in two colors, so that if and are the number of "multicolored" and "single colored" edges, respectively, then .
Proof: For the statement is immediately verified. Let and be a connected graph with vertices. Let and be the two vertices from the Lemma. We "remove" and and color according to the induction hypothesis. Now, it is not difficult to see that we can color such that the difference under consideration increases by at least . Indeed, this is clear if and are leaves, and otherwise case, considering the parity of the number of neighbors of in , we see that there is always such a way. The statement is proved by induction.
Let us now consider an arbitrary connected graph with vertices and edges. We "color" it according to the Assertion: we have ; , therefore and deleting edges satisfies the condition. Thus, we got .
To show that let us consider the complete graph with vertices. A necessary and sufficient condition for having the coloring from the condition is that the graph obtained after deleting the edges is bipartite. Indeed, there must be no cycles of odd length in the graph, which is equivalent to the above.
Case 1. , . Then . To reach a bipartite graph, we need to delete all the edges in each of the two groups of vertices, and the number of deleted edges is minimal when the two groups are of equal power and contain vertices. So, we need to delete at least
Case 2. , . Similarly, here we need to delete at least edges and again $$ n_1^2 \le k \cdot \left( \frac{2n_1(2n_1+1)}{2} - n_1 \right) \implies k \ge \frac{1}{2}. \quad \square
Final answer
1/2
Techniques
Graph TheoryInduction / smoothingColoring schemes, extremal arguments