Maths Olympiad Prep

Library / /2 of 2

Combinatorics Difficulty 7.7 National olympiad, round 2 Prove it Romania

Fix an integer n3n \ge 3. Determine the smallest positive integer kk satisfying the following condition:
For any tree TT with vertices v1,v2,,vnv_1, v_2, \dots, v_n and any pairwise distinct complex numbers z1,z2,,znz_1, z_2, \dots, z_n, there is a polynomial P(X,Y)P(X, Y) with complex coefficients of total degree at most kk such that for all iji \ne j satisfying 1i,jn1 \le i, j \le n, we have P(zi,zj)=0P(z_i, z_j) = 0 if and only if there is an edge in TT joining viv_i to vjv_j.

*Note, for example, that the total degree of the polynomial 9X3Y4+XY5+X629X^3Y^4 + XY^5 + X^6 - 2 is 7 because 7=3+47 = 3 + 4.*

Solutions — 3

Solution 1

*First solution.* First we provide a proof that kn1k \ge n - 1. Let TT be the path where viv_i and vi+1v_{i+1} are adjacent for all 1in11 \le i \le n - 1. Let ω\omega be a primitive root of unity of order nn and let ai=ωia_i = \omega^i for all 1in1 \le i \le n.
If f(X)=P(X,ωX)f(X) = P(X, \omega X), then for all 1in11 \le i \le n-1 we have f(ωi)=P(ai,ai+1)=0f(\omega^i) = P(a_i, a_{i+1}) = 0. Since f(1)=P(an,a1)0f(1) = P(a_n, a_1) \ne 0, ff is non-zero and has at least n1n-1 roots. This means that degPdegfn1\deg P \ge \deg f \ge n-1, proving kn1k \ge n-1.

It remains to prove that k=n1k = n - 1 is sufficient i.e. for any tree TT and any a1,a2,,ana_1, a_2, \dots, a_n we can find a polynomial PP of degree at most n1n - 1. For brevity, we call a two-variable polynomial A(X,Y)A(X, Y) *symmetric* if A(X,Y)=A(Y,X)A(X, Y) = A(Y, X).
We begin with the following observation. Suppose that AA and BB are two variable polynomials of degree at most dd. Then we can find αC\alpha \in \mathbb{C} such that for any 1i,jn1 \le i, j \le n, A(ai,aj)+αB(ai,aj)=0A(a_i, a_j) + \alpha B(a_i, a_j) = 0 if and only if A(ai,aj)=B(ai,aj)=0A(a_i, a_j) = B(a_i, a_j) = 0. This means that we can "merge" two conditions of degree at most dd into a condition of degree at most dd (note that this produces a symmetric polynomial if the initial polynomials are symmetric).
For any integer t2t \ge 2, let a *star* of size tt be a collection of tt edges for which there is a vertex which belongs to all edges. We will prove the following claims.

Claim 1. Let GG be a graph with vertices v1,v2,,vnv_1, v_2, \dots, v_n and EE edges. Suppose that we can partition the edges of GG into a number of stars. Then for any distinct complex numbers a1,a2,,ana_1, a_2, \dots, a_n we can find a symmetric polynomial PP of degree at most EE such that for all 1i,jn,ij1 \le i, j \le n, i \ne j, P(ai,aj)=0P(a_i, a_j) = 0 if and only if there is an edge between viv_i and vjv_j in GG.
*Proof.* We will first prove the claim when GG consists of a star of size En1E \le n - 1 and some isolated vertices. Without loss of generality, let v1v2,v1v3,,v1vE+1v_1v_2, v_1v_3, \dots, v_1v_{E+1} be the edges of GG. Also let s1=a1+a2,s2=a1+a3,,sE=a1+aE+1s_1 = a_1 + a_2, s_2 = a_1 + a_3, \dots, s_E = a_1 + a_{E+1}.
Consider merging the polynomials (Xa1)(Ya1)(X - a_1)(Y - a_1) and (X+Ys1)(X+Ys2)(X+YsE)(X + Y - s_1)(X + Y - s_2) \dots (X + Y - s_E) into a polynomial of degree at most EE (which is of course symmetric). They both vanish at a pair ai,aja_i, a_j if and only if 1{i,j}1 \in \{i, j\} and ai+aj{s1,s2,,sE}a_i + a_j \in \{s_1, s_2, \dots, s_E\}. These two happen if and only if viv_i and vjv_j are adjacent, so this produces a valid polynomial.
For the general case, let S1S2SkS_1 \cup S_2 \cup \dots \cup S_k be the partition of the edges of GG. For each 1ik1 \le i \le k, we can find a two variable polynomial PiP_i of degree at most Si|S_i| which vanishes only at the edges of SiS_i. Then we can let P=P1P2PkP = P_1P_2\dots P_k, which satisfies the claim as degPS1+S2++Sk=E\deg P \le |S_1| + |S_2| + \dots + |S_k| = E, as desired.

Claim 2. Any tree Γ\Gamma with odd number of vertices can be partitioned into stars of size 2.
*Proof.* We prove this by induction on the number of the vertices of Γ\Gamma. The base case is clear, since Γ\Gamma is a star of size 2 when Γ\Gamma has three vertices.
For the inductive step, let Γ\Gamma be a tree with 2m+12m + 1 vertices, where m2m \ge 2. Let u1u2utu_1u_2\dots u_t be a path of maximal length in Γ\Gamma (of course, t3t \ge 3). Then any neighbour of u2u_2 except for maybe u3u_3 must have degree 1, otherwise we can delete u1u_1 and insert two edges, contradicting the maximality of tt. If degu2=2\deg u_2 = 2, we can form the star u1u2,u2u3u_1u_2, u_2u_3 and apply the inductive hypothesis on Γ{u1,u2}\Gamma \setminus \{u_1, u_2\}. If degu23\deg u_2 \ge 3, let uu1,u3u \neq u_1, u_3 be a neighbour of u2u_2. Then create the star u1u2,uu2u_1u_2, uu_2 and apply the inductive hypothesis on Γ{u,u1}\Gamma \setminus \{u, u_1\}. This proves the claim.

The case where nn is odd becomes trivial, since TT has n1n-1 edges. Assume that nn is even. Without loss of generality, let degvn=1\deg v_n = 1 and let vnv_n be adjacent to vn1v_{n-1}.
Consider the graph TT' formed by deleting the edge vn1vnv_{n-1}v_n from TT (but not permuting the nn vertices). Clearly, TT' consists of vnv_n and a tree on v1,v2,,vn1v_1, v_2, \dots, v_{n-1}. As n1n-1 is odd, the edges of TT' can be partitioned into stars of size 2 from Claim 2. From Claim 1 it follows that there is a symmetric polynomial Q(X,Y)Q(X, Y) of degree at most n2n-2 such that Q(ai,aj)=0Q(a_i, a_j) = 0 if and only if i,jn1i, j \le n-1 and vi,vjv_i, v_j are adjacent in TT.
Let g(X)g(X) be the polynomial of degree n1n-1 such that g(a1)=g(a2)==g(an1)=0g(a_1) = g(a_2) = \dots = g(a_{n-1}) = 0 and g(an)=Q(an1,an)=Q(an,an1)g(a_n) = -Q(a_{n-1}, a_n) = -Q(a_n, a_{n-1}). Let PP be the polynomial of degree at most n1n-1 obtained by merging F1(X,Y)=Q(X,Y)+g(X)+g(Y)F_1(X, Y) = Q(X, Y) + g(X) + g(Y) and F2(X,Y)=Q(X,Y)(X+Ys)F_2(X, Y) = Q(X, Y)(X+Y-s), where s=an+an1s = a_n + a_{n-1}. It is easy to see that PP vanishes at each (ai,aj)(a_i, a_j) for which vi,vjv_i, v_j are adjacent in TT.
Suppose that viv_i and vjv_j are not adjacent in TT. If i,jn1i, j \le n-1, then F1(ai,aj)=Q(ai,aj)0F_1(a_i, a_j) = Q(a_i, a_j) \neq 0. If n{i,j}n \in \{i, j\}, then ai+ajsa_i + a_j \neq s and Q(ai,aj)0Q(a_i, a_j) \neq 0, so F2(ai,aj)0F_2(a_i, a_j) \neq 0. Hence P(ai,aj)0P(a_i, a_j) \neq 0. This proves that PP satisfies the required conditions, as desired.

Solution 2

*Second solution.* Establish the lower bound as in Solution 1. We now address the upper bound differently. Let GG be a graph on vertices v1,,vnv_1, \dots, v_n. Say that a polynomial P(X,Y)P(X, Y) is GG-good if it satisfies the conditions in the statement of the problem. We prove the more general fact below:

Claim. Let did_i be the degree of viv_i. Assume that these degrees satisfy dinid_i \le n-i for all in1i \le n-1, and dn=1d_n = 1. Then there is a GG-admissible polynomial of degree at most n1n-1.
Notice here that, if GG is a tree with vertices ordered so that d1d2dnd_1 \ge d_2 \ge \dots \ge d_n, then it satisfies the conditions in the Claim. Indeed, we have dn=1d_n = 1, and if di>nid_i > n-i for some in1i \le n-1, then we have 2n2j=1ndjj=1i(ni+1)+j=i+1n1=i(ni+1)+(ni)=(i+1)(ni+1)12n12n-2 \sum_{j=1}^n d_j \ge \sum_{j=1}^i (n-i+1) + \sum_{j=i+1}^n 1 = i(n-i+1) + (n-i) = (i+1)(n-i+1) - 1 \ge 2n-1, which is a contradiction. So, it suffices to prove the Claim.

*Proof of the Claim.* Let
P(X,Y)=j=0n1Rj(Y)Xj P(X, Y) = \sum_{j=0}^{n-1} R_j(Y)X^j
be the sought polynomial; set Rn1(X)=1R_{n-1}(X) = 1. Denote
Qi(X)=P(X,ai)=Xn1+j=0n2qijXj,where qij=Rj(ai). Q_i(X) = P(X, a_i) = X^{n-1} + \sum_{j=0}^{n-2} q_{ij}X^j, \quad \text{where } q_{ij} = R_j(a_i).
So, we will seek for the sequences Cj=(q1j,q2j,,qnj)C_j = (q_{1j}, q_{2j}, \dots, q_{nj}) such that there exists a polynomial RjR_j with degRjnj1\deg R_j \le n - j - 1 such that qij=Rj(ai)q_{ij} = R_j(a_i). Notice that the first njn-j terms of such a sequence determine it uniquely; in particular, there are no restrictions on the sequence C0C_0.
We know that the polynomial QiQ_i has did_i prescribed roots. For every in1i \le n-1, augment this list by some numbers not from the set A={a1,a2,,an}A = \{a_1, a_2, \dots, a_n\} to the list bi1,bi2,,bi,nib_{i1}, b_{i2}, \dots, b_{i,n-i}. Also, denote by bn1b_{n1} the unique prescribed root of QnQ_n. Thus, we should have
Qi(X)=Si(X)j=1max(ni,1)(Xbij),(1) Q_i(X) = S_i(X) \prod_{j=1}^{\max(n-i, 1)} (X - b_{ij}), \quad (1)
where SiS_i is a monic polynomial with degSi=i1\deg S_i = i-1 for in1i \le n-1 and degSn=n2\deg S_n = n-2. The only extra conditions we have are that each SiS_i should achieve non-zero values at the prescribed finite subset AiA_i of AA (containing no prescribed roots of QiQ_i).
The polynomial Q1Q_1 is uniquely determined by (1).
Assume that the polynomials Q1,,Qi1Q_1, \dots, Q_{i-1} have already been determined, for some in2i \le n-2. This means that the sequences S1,,Si1S_1, \dots, S_{i-1} are also determined. This determines the polynomial SiS_i up to the constant term. So we may choose this constant term so that SiS_i has no prohibited roots, thus defining QiQ_i.

Solution 3

*Third solution.* The answer is n1n-1. Use the same argument as Solution 1 for the lower bound.
We can construct such a polynomial via a method that uses no graph theory other than the fact that TT has n1n-1 edges.

Lemma 1. Let kk be a positive integer and let A,BC2A, B \subseteq \mathbb{C}^2 be two disjoint finite sets. Suppose A=2k|A| = 2k and no line intersects ABA \cup B in more than k+1k+1 points. Then there exists a polynomial of degree kk that vanishes on AA and is non-zero on BB.
*Proof.* We first argue that we may reduce to the case when B=1|B| = 1. This is since if PP and PP' are polynomials that are zero on AA and nonzero on BB and BB', then a generic linear combination of PP and PP' is nonzero on AA and nonzero on BBB \cup B'. Now suppose B={b}B = \{b\}.
For a,aAa, a' \in A, write aaa \sim a' if a,a,ba, a', b are collinear. This is an equivalence relation, and each equivalence class has at most kk elements. Thus we may pair up the elements of AA such that no two paired elements are collinear with bb. Now let PP be the polynomial vanishing on the union of the kk lines determined by the pairs, which is nonzero at bb by construction.

Lemma 2. Let a1,a2,,ana_1, a_2, \dots, a_n be distinct complex numbers. Then a line can intersect at most nn points of the form (ai,aj)(a_i, a_j).
*Proof.* If not, then by the pigeonhole principle such a line must contain two points with the same xx-coordinate. But then it is vertical and thus can only contain nn points.
We are done by applying Lemma 1 with
A={(ai,aj):vivjE(T)}andB={(ai,aj):ij,vivjE(T)}. A = \{(a_i, a_j) : v_i v_j \in E(T)\} \quad \text{and} \quad B = \{(a_i, a_j) : i \neq j, v_i v_j \notin E(T)\}.

Want a route through all this instead of an archive? The track puts 2,000 problems in a working order, from AMC 10 level to the IMO shortlist.

Source: MathNet, licensed CC-BY-4.0. Statement and solution reproduced as published; topic and difficulty added by this site.