Maths Olympiad Prep

Library / /389 of 397

, 2023

Geometry Difficulty 7.4 National Olympiad, round 2 Prove it Taiwan

There is an equilateral triangle ABCABC on the plane. Three straight lines pass through AA, BB and CC, respectively, such that the intersections of these lines form an equilateral triangle inside ABCABC. On each turn, Ming chooses a two-line intersection inside ABCABC, and draw the straight line determined by the intersection and one of AA, BB and CC of his choice. Find the maximum possible number of three-line intersections within ABCABC after 300300 turns.

Solution

The answer is 76517651; in general, for the 6n6n-th turn, the maximum possible value is 3n2+3n+13n^2 + 3n + 1.

Let us coordinatize these lines: for a line \ell passing through point AA, label it as (1,r)(1, r), where
r=sin(AB,)sin(,AC). r = \frac{\sin \angle(AB, \ell)}{\sin \angle(\ell, AC)}.
Define (2,r)(2, r) and (3,r)(3, r) similarly. Note that this means:
- there exists t>0t > 0 such that the initial three lines are (1,t)(1, t), (2,t)(2, t), and (3,t)(3, t);
- by the trigonometric form of Ceva's theorem, the three lines (1,r1)(1, r_1), (2,r2)(2, r_2), and (3,r3)(3, r_3) are concurrent if and only if r1r2r3=1r_1 r_2 r_3 = 1, i.e., logtr1+logtr2+logtr3=0\log_t r_1 + \log_t r_2 + \log_t r_3 = 0;
- each time Ming selects the intersection point of (i,ri)(i, r_i) and (j,rj)(j, r_j), the new line he can add is (k,rk)(k, r_k), where k{1,2,3}{i,j}k \in \{1, 2, 3\} - \{i, j\} and rk=(rirj)1r_k = (r_i r_j)^{-1}, i.e., logtrk=(logtri+logtrj)\log_t r_k = -(\log_t r_i + \log_t r_j).

Combining the above, by considering the logt\log_t version, we can reformulate the problem as the following equivalent combinatorial problem:
Let S1=S2=S3={1}S_1 = S_2 = S_3 = \{1\}. Each time, Ming takes niSin_i \in S_i and njSjn_j \in S_j, and adds (ni+nj)-(n_i + n_j) to the third set. Let S={(s1,s2,s3):siSi,s1+s2+s3=0}S = \{(s_1, s_2, s_3) : s_i \in S_i, s_1 + s_2 + s_3 = 0\}. Prove that after 6n6n operations, the maximum possible value of S|S| is 3n2+3n+13n^2 + 3n + 1.

Estimate: This is equivalent to proving that when Si=ni|S_i| = n_i and n1+n2+n3=6n+3n_1 + n_2 + n_3 = 6n + 3 (since initially Si=1|S_i| = 1), we have S3n2+3n+1|S| \le 3n^2 + 3n + 1. Let the elements of S1S_1 be a1<a2<<apa_1 < a_2 < \dots < a_p, the elements of S2S_2 be b1<b2<<bqb_1 < b_2 < \dots < b_q, and the elements of S3S_3 be c1<c2<<crc_1 < c_2 < \dots < c_r, and without loss of generality assume pqrp \le q \le r. Consider the sets
T1={a1+b1,a2+b1,,ap+b1,ap+b2,,ap+bq},T2={a1+b2,a2+b2,,ap1+b2,ap1+b3,,ap1+bq},Ti={a1+bi,a2+bi,,ap+1i+bi,ap+1i+bi+1,,ap+1i+bq}, \begin{align*} T_1 &= \{a_1 + b_1, a_2 + b_1, \dots, a_p + b_1, a_p + b_2, \dots, a_p + b_q\}, \\ T_2 &= \{a_1 + b_2, a_2 + b_2, \dots, a_{p-1} + b_2, a_{p-1} + b_3, \dots, a_{p-1} + b_q\}, \\ \vdots \\ T_i &= \{a_1 + b_i, a_2 + b_i, \dots, a_{p+1-i} + b_i, a_{p+1-i} + b_{i+1}, \dots, a_{p+1-i} + b_q\}, \\ \vdots \end{align*}

Then it is easy to see that i=1pTi=S1+S2\cup_{i=1}^p T_i = S_1 + S_2, and any ak+ba_k + b_\ell appears in exactly one TiT_i. Also,
Ti(S3)S3=randTi(S3)Ti=p+q2i+1, |T_i \cap (-S_3)| \le |S_3| = r \quad \text{and} \quad |T_i \cap (-S_3)| \le |T_i| = p + q - 2i + 1,
therefore
S=i=1pTi(S3)i=1pmin(r,p+q2i+1):=f(p,q,r), |S| = \sum_{i=1}^{p} |T_i \cap (-S_3)| \le \sum_{i=1}^{p} \min(r, p+q-2i+1) := f(p, q, r),
a direct computation (see Remark 1) gives
f(p,q,r)=14(p2q2r2+pq+pr+qr)+14(p+q+r)212+14=3n2+3n+1(1) \begin{aligned} f(p, q, r) &= \frac{1}{4}(-p^2 - q^2 - r^2 + pq + pr + qr) + \frac{1}{4} \\ &\le \frac{(p+q+r)^2}{12} + \frac{1}{4} = 3n^2 + 3n + 1 \end{aligned} \quad (1)

Remark 1: Derivation of the left-hand side of (1). Note that
- if r<qp+1r < q-p+1, then f(p,q,r)=f(p,q1,r+1)f(p, q, r) = f(p, q-1, r+1);
- if r>q+p1r > q+p-1, then f(p,q,r)=f(p,q+1,r1)f(p, q, r) = f(p, q+1, r-1).
Hence we may without loss of generality set r[qp+1,q+p1]r \in [q-p+1, q+p-1], and in this case a direct computation gives
f(p,q,r)=p+q2i+1rr+p+q2i+1>r(p+q2i+1)=tr+(pt)(qt)=t(p+qrt)+pq(3) \begin{aligned} f(p, q, r) &= \sum_{p+q-2i+1 \le r} r + \sum_{p+q-2i+1 > r} (p+q-2i+1) \\ &= tr + (p-t)(q-t) = -t(p+q-r-t) + pq \end{aligned} \quad (3)

where t=(p+q+1r)/2t = \lfloor (p+q+1-r)/2 \rfloor. But note that
t(p+qrt)=maxxZx(p+qrx)=(p+qr)214(4) t(p+q-r-t) = \max_{x \in \mathbb{Z}} x(p+q-r-x) = \frac{(p+q-r)^2-1}{4} \quad (4)
Substituting (4) back into (3) gives the left-hand side of (1).

Remark 2: Derivation of the left-hand side of (2). Since changing S1S_1 to (S1+2)/3(S_1+2)/3, S2S_2 to (S21)/3(S_2-1)/3, and S3S_3 to (S31)/3(S_3-1)/3 leaves the number of tuples satisfying s1+s2+s3=0s_1 + s_2 + s_3 = 0 unchanged, we may instead consider S1=S2=S3=S_1 = S_2 = S_3 =
{n,n+1,,n1,n}\{-n, -n+1, \cdots, n-1, n\}, in which case
S=k=nn{s2+s3=k:siSi}=(2) left-hand side.(5) |S| = \sum_{k=-n}^{n} |\{s_2 + s_3 = -k : s_i \in S_i\}| = (2) \text{ left-hand side.} \quad (5)

S=k=nn(2n+1k)=(2n+1)+2k=1n(2n+1k)=(2n+1)+n×(3n+1)=3n2+3n+1.(2) \begin{aligned} |S| &= \sum_{k=-n}^{n} (2n+1-|k|) = (2n+1) + 2 \sum_{k=1}^{n} (2n+1-k) \\ &= (2n+1) + n \times (3n+1) = 3n^2 + 3n + 1. \end{aligned} \quad (2)

Construction: Consider the following strategy: we first present Ming's optimal strategy:
1. Ming first adds (1+1)=2-(1+1) = -2 to all three sets;

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 translated into English from zh; metadata (topic, difficulty) added by this project.