Maths Olympiad Prep

Library / /28 of 55

, 2006

Geometry Difficulty 8.7 Shortlist Prove it IMO

To each side aa of a convex polygon we assign the maximum area of a triangle contained in the polygon and having aa as one of its sides. Show that the sum of the areas assigned to all sides of the polygon is not less than twice the area of the polygon.
(Serbia)

Solutions — 2

Solution 1

Lemma. Every convex (2n)(2n)-gon, of area SS, has a side and a vertex that jointly span a triangle of area not less than S/nS / n.

Proof. By main diagonals of the (2n)(2n)-gon we shall mean those which partition the (2n)(2n)-gon into two polygons with equally many sides. For any side bb of the (2n)(2n)-gon denote by Δb\Delta_{b} the triangle ABPABP where A,BA, B are the endpoints of bb and PP is the intersection point of the main diagonals AA,BBAA', BB'. We claim that the union of triangles Δb\Delta_{b}, taken over all sides, covers the whole polygon.

To show this, choose any side ABAB and consider the main diagonal AAAA' as a directed segment. Let XX be any point in the polygon, not on any main diagonal. For definiteness, let XX lie on the left side of the ray AAAA'. Consider the sequence of main diagonals AA,BB,CC,AA', BB', CC', \ldots, where A,B,C,A, B, C, \ldots are consecutive vertices, situated right to AAAA'. The nn-th item in this sequence is the diagonal AAA'A (i.e. AAAA' reversed), having XX on its right side. So there are two successive vertices K,LK, L in the sequence A,B,C,A, B, C, \ldots before AA' such that XX still lies to the left of KKKK' but to the right of LLLL'. And this means that XX is in the triangle Δ\Delta_{\ell'}, =KL\ell' = K' L'. Analogous reasoning applies to points XX on the right of AAAA' (points lying on main diagonals can be safely ignored). Thus indeed the triangles Δb\Delta_{b} jointly cover the whole polygon.

The sum of their areas is no less than SS. So we can find two opposite sides, say b=ABb = AB and b=ABb' = A'B' (with AA,BBAA', BB' main diagonals) such that [Δb]+[Δb]S/n[\Delta_{b}] + [\Delta_{b'}] \geq S / n, where [][\cdots] stands for the area of a region. Let AA,BBAA', BB' intersect at PP; assume without loss of generality that PBPBPB \geq PB'. Then
[ABA]=[ABP]+[PBA][ABP]+[PAB]=[Δb]+[Δb]S/n, [ABA'] = [ABP] + [PBA'] \geq [ABP] + [PA'B'] = [\Delta_{b}] + [\Delta_{b'}] \geq S / n,
proving the lemma. \square

Now, let P\mathcal{P} be any convex polygon, of area SS, with mm sides a1,,ama_{1}, \ldots, a_{m}. Let SiS_{i} be the area of the greatest triangle in P\mathcal{P} with side aia_{i}. Suppose, contrary to the assertion, that
i=1mSiS<2. \sum_{i=1}^{m} \frac{S_{i}}{S} < 2.
Then there exist rational numbers q1,,qmq_{1}, \ldots, q_{m} such that qi=2\sum q_{i} = 2 and qi>Si/Sq_{i} > S_{i} / S for each ii.

Let nn be a common denominator of the mm fractions q1,,qmq_{1}, \ldots, q_{m}. Write qi=ki/nq_{i} = k_{i} / n; so ki=2n\sum k_{i} = 2n. Partition each side aia_{i} of P\mathcal{P} into kik_{i} equal segments, creating a convex (2n)(2n)-gon of area SS (with some angles of size 180180^{\circ}), to which we apply the lemma. Accordingly, this refined polygon has a side bb and a vertex HH spanning a triangle TT of area [T]S/n[T] \geq S / n. If bb is a piece of a side aia_{i} of P\mathcal{P}, then the triangle WW with base aia_{i} and summit HH has area
[W]=ki[T]kiS/n=qiS>Si, [W] = k_{i} \cdot [T] \geq k_{i} \cdot S / n = q_{i} \cdot S > S_{i},
in contradiction with the definition of SiS_{i}. This ends the proof.

Solution 2

As in the first solution, we allow again angles of size 180180^{\circ} at some vertices of the convex polygons considered.

To each convex nn-gon P=A1A2An\mathcal{P} = A_{1} A_{2} \ldots A_{n} we assign a centrally symmetric convex (2n)(2n)-gon Q\mathcal{Q} with side vectors ±AiAi+1,1in\pm \overrightarrow{A_{i} A_{i+1}}, 1 \leq i \leq n. The construction is as follows. Attach the 2n2n vectors ±AiAi+1\pm \overrightarrow{A_{i} A_{i+1}} at a common origin and label them b1,b2,,b2n\overrightarrow{\mathbf{b}_{1}}, \overrightarrow{\mathbf{b}_{2}}, \ldots, \overrightarrow{\mathbf{b}_{2n}} in counterclockwise direction; the choice of the first vector b1\overrightarrow{\mathbf{b}_{1}} is irrelevant. The order of labelling is well-defined if P\mathcal{P} has neither parallel sides nor angles equal to 180180^{\circ}. Otherwise several collinear vectors with the same direction are labelled consecutively bj,bj+1,,bj+r\overrightarrow{\mathbf{b}_{j}}, \overrightarrow{\mathbf{b}_{j+1}}, \ldots, \overrightarrow{\mathbf{b}_{j+r}}. One can assume that in such cases the respective opposite vectors occur in the order bj,bj+1,,bj+r-\overrightarrow{\mathbf{b}_{j}}', -\overrightarrow{\mathbf{b}_{j+1}}, \ldots, -\overrightarrow{\mathbf{b}_{j+r}}, ensuring that bj+n=bj\overrightarrow{\mathbf{b}_{j+n}} = -\overrightarrow{\mathbf{b}_{j}} for j=1,,2nj = 1, \ldots, 2n. Indices are taken cyclically here and in similar situations below.

Choose points B1,B2,,B2nB_{1}, B_{2}, \ldots, B_{2n} satisfying BjBj+1=bj\overrightarrow{B_{j} B_{j+1}} = \overrightarrow{\mathbf{b}_{j}} for j=1,,2nj = 1, \ldots, 2n. The polygonal line Q=B1B2B2n\mathcal{Q} = B_{1} B_{2} \ldots B_{2n} is closed, since j=12nbj=0\sum_{j=1}^{2n} \overrightarrow{\mathbf{b}_{j}} = \overrightarrow{0}. Moreover, Q\mathcal{Q} is a convex (2n)(2n)-gon due to the arrangement of the vectors bj\overrightarrow{\mathbf{b}_{j}}, possibly with 180180^{\circ}-angles. The side vectors of Q\mathcal{Q} are ±AiAi+1\pm \overrightarrow{A_{i} A_{i+1}}, 1in1 \leq i \leq n. So in particular Q\mathcal{Q} is centrally symmetric, because it contains as side vectors AiAi+1\overrightarrow{A_{i} A_{i+1}} and AiAi+1-\overrightarrow{A_{i} A_{i+1}} for each i=1,,ni = 1, \ldots, n. Note that BjBj+1B_{j} B_{j+1} and Bj+nBj+n+1B_{j+n} B_{j+n+1} are opposite sides of Q\mathcal{Q}, 1jn1 \leq j \leq n. We call Q\mathcal{Q} the associate of P\mathcal{P}.

Let SiS_{i} be the maximum area of a triangle with side AiAi+1A_{i} A_{i+1} in P\mathcal{P}, 1in1 \leq i \leq n. We prove that
[B1B2B2n]=2i=1nSi(1) \left[ B_{1} B_{2} \ldots B_{2n} \right] = 2 \sum_{i=1}^{n} S_{i} \tag{1}
and
[B1B2B2n]4[A1A2An](2) \left[ B_{1} B_{2} \ldots B_{2n} \right] \geq 4 \left[ A_{1} A_{2} \ldots A_{n} \right] \tag{2}
It is clear that (1) and (2) imply the conclusion of the original problem.

Lemma. For a side AiAi+1A_{i} A_{i+1} of P\mathcal{P}, let hih_{i} be the maximum distance from a point of P\mathcal{P} to line AiAi+1A_{i} A_{i+1}, i=1,,ni = 1, \ldots, n. Denote by BjBj+1B_{j} B_{j+1} the side of Q\mathcal{Q} such that AiAi+1=BjBj+1\overrightarrow{A_{i} A_{i+1}} = \overrightarrow{B_{j} B_{j+1}}. Then the distance between BjBj+1B_{j} B_{j+1} and its opposite side in Q\mathcal{Q} is equal to 2hi2 h_{i}.

Proof. Choose a vertex AkA_{k} of P\mathcal{P} at distance hih_{i} from line AiAi+1A_{i} A_{i+1}. Let u\mathbf{u} be the unit vector perpendicular to AiAi+1A_{i} A_{i+1} and pointing inside P\mathcal{P}. Denoting by xy\mathbf{x} \cdot \mathbf{y} the dot product of vectors x\mathbf{x} and y\mathbf{y}, we have
h=uAiAk=u(AiAi+1++Ak1Ak)=u(AiAi1++Ak+1Ak). h = \mathbf{u} \cdot \overrightarrow{A_{i} A_{k}} = \mathbf{u} \cdot \left( \overrightarrow{A_{i} A_{i+1}} + \cdots + \overrightarrow{A_{k-1} A_{k}} \right ) = \mathbf{u} \cdot \left( \overrightarrow{A_{i} A_{i-1}} + \cdots + \overrightarrow{A_{k+1} A_{k}} \right ).
In Q\mathcal{Q}, the distance HiH_{i} between the opposite sides BjBj+1B_{j} B_{j+1} and Bj+nBj+n+1B_{j+n} B_{j+n+1} is given by
Hi=u(BjBj+1++Bj+n1Bj+n)=u(bj+bj+1++bj+n1). H_{i} = \mathbf{u} \cdot \left( \overrightarrow{B_{j} B_{j+1}} + \cdots + \overrightarrow{B_{j+n-1} B_{j+n}} \right ) = \mathbf{u} \cdot \left( \overrightarrow{\mathbf{b}_{j}} + \overrightarrow{\mathbf{b}_{j+1}} + \cdots + \overrightarrow{\mathbf{b}_{j+n-1}} \right ).
The choice of vertex AkA_{k} implies that the nn consecutive vectors bj,bj+1,,bj+n1\overrightarrow{\mathbf{b}_{j}}, \overrightarrow{\mathbf{b}_{j+1}}, \ldots, \overrightarrow{\mathbf{b}_{j+n-1}} are precisely AiAi+1,,Ak1Ak\overrightarrow{A_{i} A_{i+1}}, \ldots, \overrightarrow{A_{k-1} A_{k}} and AiAi1,,Ak+1Ak\overrightarrow{A_{i} A_{i-1}}, \ldots, \overrightarrow{A_{k+1} A_{k}}, taken in some order. This implies Hi=2hiH_{i} = 2 h_{i}.

For a proof of (1), apply the lemma to each side of P\mathcal{P}. If OO is the centre of Q\mathcal{Q} then, using the notation of the lemma,
[BjBj+1O]=[Bj+nBj+n+1O]=[AiAi+1Ak]=Si [ B_{j} B_{j+1} O ] = [ B_{j+n} B_{j+n+1} O ] = [ A_{i} A_{i+1} A_{k} ] = S_{i}
Summation over all sides of P\mathcal{P} yields (1).

Set d(P)=[Q]4[P]d(\mathcal{P}) = [\mathcal{Q}] - 4 [\mathcal{P}] for a convex polygon P\mathcal{P} with associate Q\mathcal{Q}. Inequality (2) means that d(P)0d(\mathcal{P}) \geq 0 for each convex polygon P\mathcal{P}. The last inequality will be proved by induction on the number \ell of side directions of P\mathcal{P}, i.e. the number of pairwise nonparallel lines each containing a side of P\mathcal{P}.

We choose to start the induction with =1\ell = 1 as a base case, meaning that certain degenerate polygons are allowed. More exactly, we regard as degenerate convex polygons all closed polygonal lines of the form X1X2XkY1Y2YmX1X_{1} X_{2} \ldots X_{k} Y_{1} Y_{2} \ldots Y_{m} X_{1}, where X1,X2,,XkX_{1}, X_{2}, \ldots, X_{k} are points in this order on a line segment X1Y1X_{1} Y_{1}, and so are Ym,Ym1,,Y1Y_{m}, Y_{m-1}, \ldots, Y_{1}. The initial construction applies to degenerate polygons; their associates are also degenerate, and the value of dd is zero.

For the inductive step, consider a convex polygon P\mathcal{P} which determines \ell side directions, assuming that d(P)0d(\mathcal{P}) \geq 0 for polygons with smaller values of \ell.

Suppose first that P\mathcal{P} has a pair of parallel sides, i.e. sides on distinct parallel lines. Let AiAi+1A_{i} A_{i+1} and AjAj+1A_{j} A_{j+1} be such a pair, and let AiAi+1AjAj+1A_{i} A_{i+1} \leq A_{j} A_{j+1}. Remove from P\mathcal{P} the parallelogram RR determined by vectors AiAi+1\overrightarrow{A_{i} A_{i+1}} and AiAj+1\overrightarrow{A_{i} A_{j+1}}. Two polygons are obtained in this way. Translating one of them by vector AiAi+1\overrightarrow{A_{i} A_{i+1}} yields a new convex polygon P\mathcal{P}', of area [P][R][\mathcal{P}] - [R] and with value of \ell not exceeding the one of P\mathcal{P}. The construction just described will be called operation A.

Figure 1
Figure 2

The associate of P\mathcal{P}' is obtained from Q\mathcal{Q} upon decreasing the lengths of two opposite sides by an amount of 2AiAi+12 A_{i} A_{i+1}. By the lemma, the distance between these opposite sides is twice the distance between AiAi+1A_{i} A_{i+1} and AjAj+1A_{j} A_{j+1}. Thus operation A\mathbf{A} decreases [Q][\mathcal{Q}] by the area of a parallelogram with base and respective altitude twice the ones of RR, i.e. by 4[R]4[R]. Hence A\mathbf{A} leaves the difference d(P)=[Q]4[P]d(\mathcal{P}) = [\mathcal{Q}] - 4[\mathcal{P}] unchanged.

Now, if P\mathcal{P}' also has a pair of parallel sides, apply operation A\mathbf{A} to it. Keep doing so with the subsequent polygons obtained for as long as possible. Now, A decreases the number pp of pairs of parallel sides in P\mathcal{P}. Hence its repeated applications gradually reduce pp to 00, and further applications of A\mathbf{A} will be impossible after several steps. For clarity, let us denote by P\mathcal{P} again the polygon obtained at that stage.

The inductive step is complete if P\mathcal{P} is degenerate. Otherwise >1\ell > 1 and p=0p = 0, i.e. there are no parallel sides in P\mathcal{P}. Observe that then 3\ell \geq 3. Indeed, =2\ell = 2 means that the vertices of P\mathcal{P} all lie on the boundary of a parallelogram, implying p>0p > 0.

Furthermore, since P\mathcal{P} has no parallel sides, consecutive collinear vectors in the sequence (bk)\left( \overrightarrow{\mathbf{b}_{k}} \right) (if any) correspond to consecutive 180180^{\circ}-angles in P\mathcal{P}. Removing the vertices of such angles, we obtain a convex polygon with the same value of d(P)d(\mathcal{P}).

In summary, if operation A\mathbf{A} is impossible for a nondegenerate polygon P\mathcal{P}, then 3\ell \geq 3. In addition, one may assume that P\mathcal{P} has no angles of size 180180^{\circ}.

The last two conditions then also hold for the associate Q\mathcal{Q} of P\mathcal{P}, and we perform the following construction. Since 3\ell \geq 3, there is a side BjBj+1B_{j} B_{j+1} of Q\mathcal{Q} such that the sum of the angles at BjB_{j} and Bj+1B_{j+1} is greater than 180180^{\circ}. (Such a side exists in each convex kk-gon for k>4k > 4.) Naturally, Bj+nBj+n+1B_{j+n} B_{j+n+1} is a side with the same property. Extend the pairs of sides Bj1Bj,Bj+1Bj+2B_{j-1} B_{j}, B_{j+1} B_{j+2} and Bj+n1Bj+n,Bj+n+1Bj+n+2B_{j+n-1} B_{j+n}, B_{j+n+1} B_{j+n+2} to meet at UU and VV, respectively. Let Q\mathcal{Q}' be the centrally symmetric convex 2(n+1)2(n+1)-gon obtained from Q\mathcal{Q} by inserting UU and VV into the sequence B1,,B2nB_{1}, \ldots, B_{2n} as new vertices between Bj,Bj+1B_{j}, B_{j+1} and Bj+n,Bj+n+1B_{j+n}, B_{j+n+1}, respectively. Informally, we adjoin to Q\mathcal{Q} the congruent triangles BjBj+1UB_{j} B_{j+1} U and Bj+nBj+n+1VB_{j+n} B_{j+n+1} V. Note that Bj,Bj+1,Bj+nB_{j}, B_{j+1}, B_{j+n} and Bj+n+1B_{j+n+1} are kept as vertices of Q\mathcal{Q}', although BjBj+1B_{j} B_{j+1} and Bj+nBj+n+1B_{j+n} B_{j+n+1} are no longer its sides.

Let AiAi+1A_{i} A_{i+1} be the side of P\mathcal{P} such that AiAi+1=BjBj+1=bj\overrightarrow{A_{i} A_{i+1}} = \overrightarrow{B_{j} B_{j+1}} = \overrightarrow{\mathbf{b}_{j}}. Consider the point WW such that triangle AiAi+1WA_{i} A_{i+1} W is congruent to triangle BjBj+1UB_{j} B_{j+1} U and exterior to P\mathcal{P}. Insert WW into the sequence A1,A2,,AnA_{1}, A_{2}, \ldots, A_{n} as a new vertex between AiA_{i} and Ai+1A_{i+1} to obtain an (n+1)(n+1)-gon P\mathcal{P}'. We claim that P\mathcal{P}' is convex and its associate is Q\mathcal{Q}'.

Figure 3
Figure 4

Vectors AiW\overrightarrow{A_{i} W} and bj1\overrightarrow{\mathbf{b}_{j-1}} are collinear and have the same direction, as well as vectors WAi+1\overrightarrow{W A_{i+1}} and bj+1\overrightarrow{\mathbf{b}_{j+1}}. Since bj1,bj,bj+1\overrightarrow{\mathbf{b}_{j-1}}, \overrightarrow{\mathbf{b}_{j}}, \overrightarrow{\mathbf{b}_{j+1}} are consecutive terms in the sequence (bk)\left( \overrightarrow{\mathbf{b}_{k}} \right), the angle inequalities (bj1,bj)(Ai1Ai,bj)\angle( \overrightarrow{\mathbf{b}_{j-1}}, \overrightarrow{\mathbf{b}_{j}} ) \leq \angle( \overrightarrow{A_{i-1} A_{i}}, \overrightarrow{\mathbf{b}_{j}} ) and (bj,bj+1)(bj,Ai+1Ai+2)\angle( \overrightarrow{\mathbf{b}_{j}}, \overrightarrow{\mathbf{b}_{j+1}} ) \leq \angle( \overrightarrow{\mathbf{b}_{j}}, \overrightarrow{A_{i+1} A_{i+2}} ) hold true. They show that P\mathcal{P}' is a convex polygon. To construct its associate, vectors ±AiAi+1=±bj\pm \overrightarrow{A_{i} A_{i+1}} = \pm \overrightarrow{\mathbf{b}_{j}} must be deleted from the defining sequence (bk)\left( \overrightarrow{\mathbf{b}_{k}} \right) of Q\mathcal{Q}, and the vectors ±AiW,±WAi+1\pm \overrightarrow{A_{i} W}, \pm \overrightarrow{W A_{i+1}} must be inserted appropriately into it. The latter can be done as follows:
,bj1,AiW,WAi+1,bj+1,,bj1,AiW,WAi+1,bj+1,. \ldots, \overrightarrow{\mathbf{b}_{j-1}}, \overrightarrow{A_{i} W}, \overrightarrow{W A_{i+1}}, \overrightarrow{\mathbf{b}_{j+1}}, \ldots, -\overrightarrow{\mathbf{b}_{j-1}}, -\overrightarrow{A_{i} W}, -\overrightarrow{W A_{i+1}}, -\overrightarrow{\mathbf{b}_{j+1}}, \ldots.
This updated sequence produces Q\mathcal{Q}' as the associate of P\mathcal{P}'.

It follows from the construction that [P]=[P]+[AiAi+1W][\mathcal{P}'] = [\mathcal{P}] + [A_{i} A_{i+1} W] and [Q]=[Q]+2[AiAi+1W][\mathcal{Q}'] = [\mathcal{Q}] + 2 [A_{i} A_{i+1} W]. Therefore d(P)=d(P)2[AiAi+1W]<d(P)d(\mathcal{P}') = d(\mathcal{P}) - 2 [A_{i} A_{i+1} W] < d(\mathcal{P}).

To finish the induction, it remains to notice that the value of \ell for P\mathcal{P}' is less than the one for P\mathcal{P}. This is because side AiAi+1A_{i} A_{i+1} was removed. The newly added sides AiWA_{i} W and WAi+1W A_{i+1} do not introduce new side directions. Each one of them is either parallel to a side of P\mathcal{P} or lies on the line determined by such a side. The proof is complete.

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.