when did imo become easy
TRANSCRIPT
-
7/25/2019 When Did Imo Become Easy
1/29
Mean Geometry
Zachary R. Abel
July 9, 2007
-
7/25/2019 When Did Imo Become Easy
2/29
Contents
1 Introduction 3
1.1 Basic Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Point Averages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Figure Averages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.4 Finally! . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2 Proofs of MGT 5
2.1 Proof 1: Spiral Similarity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.2 Proof 2: Complex Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
3 Some Problems 7
3.1 Equilaterals Joined at the Hip . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
3.2 Napoleons Last Hurrah . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
3.3 Extending MGT? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
3.4 But Wait, Theres More! . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
3.5 The Asymmetric Propeller . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
4 Figure Addition 11
4.1 When Did the IMO Get So Easy? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
4.2 A Fresh Look at an Old Result . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
5 A Few Harder Problems 13
5.1 A Pretty(,) Busy Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
5.2 Hidden Circles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
5.3 Nagel Who? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
6 Additional Problems 19
7 Solutions to Additional Problems 21
1
-
7/25/2019 When Did Imo Become Easy
3/29
List of Figures
1 Direct and inverse similarity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2 Point averaging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
3 Figure averaging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
4 MGT for triangles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
5 Similar triangles AiOBi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
6 Similar triangles AiOCi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
7 Problem 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
8 Napoleons Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
9 Equilateral triangleJ KL . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
10 More Napoleon . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
11 Problem 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
12 Problem 5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
13 Problem 6 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
14 Point addition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
15 Problem 7 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
16 Pompeius Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1217 Problem 9 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
18 Studying quadrilateralAB CD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
19 Quadrilateral AB CD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
20 Problem 10 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
21 CirclesO1 andO2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
22 Three similar triangles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
23 Problem 11 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
24 ExcircleIa. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
25 Problem 11 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
26 Problem 12 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
27 Problem 13 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2128 Problem 14.c . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
29 Van Aubels Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
30 Problem 16.a . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
31 Problem 16.b . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
32 Problem 16.c . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
33 Problem 16.d . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
34 Proof 2 of perpendicularization lemma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
35 Problem 18.b . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
36 Problem 18.c . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
37 Generalization of problem 14 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
38 Problem 19 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
39 Problem 20 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
40 Problem 21 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2
-
7/25/2019 When Did Imo Become Easy
4/29
1 Introduction (for lack of a better section header)
There are generally two opposite approaches to Olympiad geometry. Some prefer to draw the diagram and simply
stare (labeling points only clutters the diagram!), waiting for the interactions between the problems various elements
to present themselves visually. Others toss the diagram onto the complex or coordinate plane and attempt to
establish the necessary connections through algebraic calculation rather than geometric insight.
This article discusses an interesting way to visualize and approach a variety of geometry problems by combining
these two common methods: synthetic and analytic. Well focus on a theorem known as The Fundamental Theorem
of Directly Similar Figures [4] or The Mean Geometry Theorem (abbreviated here as MGT). Although the result
is quite simple, it nevertheless encourages a powerful new point of view.
1.1 Basic Definitions
A figure has positive orientation if its vertices are listed in counterclockwise order (like pentagons ABCDE or
LMNOP in diagram 1), and otherwise it hasnegative orientation (likeV W X Y Z or even EDCBA). Two figures
aresimilar if. . . , well, were all familiar with similar figures; two similar figures aredirectly similar if they have the
same orientation, and otherwise theyre inversely similar. For example, pentagon ABCDE is directly similar to
pentagonLMNOP, but inversely similar to V W X Y Z .
AB
C
D
E
L
MN
O
P
V
W
X
YZ
Figure 1: Direct and inverse similarity
1.2 Point Averages
A
N= 2
3A+
1
3B
1
2A+
1
2B = M
Figure 2: Point averaging
Now for some non-standard notation. We define theaverage of two points
Aand B as the midpointMof segmentAB, and we write 12
A + 12
B= M.
We can also compute weighted averages with weights other than 12
and1
2, as long as the weights add to 1: the weighted average (1 k)A + (k)B
is the point Xon line AB so that AX/AB = k . (This is consistent with
the complex-number model of the plane, though we will still treat A and
B as points rather than complex values.) For example, N = 23
A + 13
B is
one-third of the way across from A to B (see Figure 2).
1.3 Figure Averages
We can use the concept of averaging points to define the (weighted) average of two figures, where a figure may
be a polygon, circle, or any other 2-dimensional path or region (this article will stick to polygons). To average two
figures, simply take the appropriate average of corresponding pairs of points. For polygons, we need only average
corresponding vertices:
3
-
7/25/2019 When Did Imo Become Easy
5/29
-
7/25/2019 When Did Imo Become Easy
6/29
2 Proofs of MGT
A1
A2
A3
B1
B2B3
C1
C2
C3
Figure 4: MGT for triangles
We mentioned in the introduction that this article incorporates some ideas
from both synthetic and analytic geometry. Here we offer two proofs of the
Theorem one based in each category to better illustrate the connection.
Although MGT holds for all types of figures, it suffices to prove result
for triangles (why?):
MGT for Triangles. Ifa = A1A2A3 andb = B1B2B3 are di-rectly similar triangles in the plane, then the weighted averagec =(1 k)A1A2A3 + (k)B1B2B3 = C1C2C3 is similar to botha andb(Figure 4).
2.1 Proof 1: Spiral Similarity
Ifb is simply a translation ofa, say by vectorV, then the transla-tion by vector k
V sendsa toc, so they are directly similar (in fact
congruent).
Otherwise, there exists a unique spiral similarity1
rO that takesatob (see [5, Theorem 4.82]). The three triangles AiOBi (fori = 1, 2, 3)have AiOBi= and OBi/OAi= r, so they are all similar (see Figure 5).
A1
A2
A3
B1
B2B3
C1
C2
C3
O
Figure 5: Similar triangles AiOBi
A1
A2
A3
B1
B2B3
C1
C2
C3
O
Figure 6: Similar triangles AiOCi
Since C1, C2, and C3 are in corresponding positions in these three similar triangles (since AiCi/AiBi = k for
i = 1, 2, 3), the three triangles AiOCi (i = 1, 2, 3) are similar to each other (Figure 6). Thus the spiral similarityr
O, where
=
A1
OC1
and r
=OC1
/OA1
, sends A1A2A3 to C1C2C3, so these triangles are directly similar,as desired.
QED
2.2 Proof 2: Complex Numbers
Well use capital letters for points and lower case for the corresponding complex numbers: point A1 corresponds to
the complex number a1, and so on.
1i.e. a rotation through an angle followed by a dilation with ratio r, both centered at the point O
5
-
7/25/2019 When Did Imo Become Easy
7/29
Since the triangles are similar, A2A1A3 = B2B1B3 and A1A3/A1A2 = B1B3/B1B2. Ifz is the complex
number with argument A2A1A3 and magnitude A1A3/A1A2, we have a3a1a2a1
=z = b3b1b2b1 . Or, rearranged,
a3 = (1 z)a1+ (z)a2 and b3 = (1 z)b1+ (z)b2. ()
To prove the theorem, we need to show thatC1C2C3 behaves similarly (no pun intended), i.e. we need to showthatc3 = (1
z)c1+ (z)c2. But sinceci= (1
k)ai+ (k)bi fori = 1, 2, 3 (by definition of
c), this equality follows
directly from ():
c3 = (1 k)a3+ (k)b3= (1 k)(1 z)a1+ (z)a2+ (k)(1 z)b1+ (z)b2= (1 z)(1 k)a1+ (k)b1+ (z)(1 k)a2+ (k)b2= (1 z)c1+ (z)c2,
as desired. QED
6
-
7/25/2019 When Did Imo Become Easy
8/29
3 Some Problems
The theorem has been proved, but there might be some lingering doubts about the usefulness of such a seemingly
simple and specialized statement.2 In this section, well put the Theorem to work, and well learn to recognize when
and how MGT may be applied to a variety of problems.
3.1 Equilaterals Joined at the HipProblem 1 (Engel). OAB andOA1B1 are positively oriented regular triangles with a common vertexO. Show
that the midpoints ofOB, OA1, andAB1 are vertices of a regular triangle. (Remember: positively oriented means
the vertices are listed in counterclockwise order.)
A
B
A1
B1
O
D
EF
Figure 7: Problem 1
To show that DEF is equilateral, wed like to expressDEF as an average of two other directly similarequilateral triangles. From the diagram we see that
1
2 A +1
2 B1 = D, 1
2 B+1
2 O= E, and 1
2 O+1
2 A1 = F,
and putting these together yields
1
2ABO+
1
2B1OA1 = DEF.
So were done by MGT, right?
Lets make sure everything is in place. Are trianglesABO and B1OA1 directly similar? The problem tells us
that they are both positively oriented equilateral triangles, so yes. Are we indeed taking a weighted average of the
two triangles? In other words, do the weights add to 1? Of course! 12
+ 12
= 1.
Now we can confidently apply the Mean Geometry Theorem and conclude that triangleDEFmust be directly
similar toABO andB1OA1, i.e.DEF is equilateral. QED
3.2 Napoleons Last Hurrah
Problem 2 (Napoleons Theorem). If equilateral triangles BCP, CAQ, ABR are erected externally on the sides of
triangle ABC, their centers X, Y, Z form an equilateral triangle (Figure 8).
2Please pardon the unintentional alliteration.
7
-
7/25/2019 When Did Imo Become Easy
9/29
A
B
C
P
Q
R
X
Y
Z
Figure 8: Napoleons Theorem
A
B
C
P
Q
R
X
Y
Z
J
K
L
Figure 9: Equilateral triangle J KL
The setup looks a lot like the configuration in the first problem, where two equilaterals share a common vertex.
We can try to mimic this earlier configuration by considering equilateral triangle 12
P CB + 12
BAR = J KL. (Figure
9)
Now the way to reach the target points, X, Y, andZ, presents itself: Xis on median J C, and its 13 of the wayacross. This means that X= 2
3J+ 1
3C, and likewise for Y andZ. We can write
XY Z=2
3JKL +
1
3CQA,
and since bothJKL andCQA are negatively oriented equilateral triangles, were done! QED
3.3 Extending MGT?
Lets look closer at that solution. First we averagedP CB andBAR: 12
P CB + 12
BAR = JKL. Then, we averaged
this withC QA: 23
JKL + 13
CQA = XY Z.Momentarily indulging ourselves in some questionable manipulation, we
can substitute the first equation into the second and simplify to find
1
3P CB +
1
3BAR +
1
3CQA = XY Z. ()
This suggests that we may be able to generalize MGT to three figures, like so:
Extended MGT. Define theweighted average of 3 points 1P1+ 2P2+ 3P3 (where the weights1, 2, and
3 are real numbers and add to 1) just as we would in the complex plane.3 Then ifA1A2A3,B1B2B3, and
C1C2C3 are directly similar triangles, their weighted average
aA1A2A3+ bB1B2B3+ cC1C2C3
is also directly similar to them. This naturally extends to include figures other than triangles (as long as theyre all
similar to each other) or more than 3 similar figures (as long as the weights add to 1).
Can you prove this? Anyway, by this generalized version of MGT, equation () alone provides a succinct,one-line proof of Napoleons Theorem. (Cool, huh?)
3.4 But Wait, Theres More!
Problem 3. Napoleon isnt done with us yet: prove that trianglesABC andXY Zhave the same centroid.
3As a notable special case, 13P1+
1
3P2+
1
3P3 is the centroid of triangle P1P2P3.
8
-
7/25/2019 When Did Imo Become Easy
10/29
A
B
C
P
Q
R
X
Y
ZG
Figure 10: More Napoleon
Like Napoleons Theorem itself, this can also be done in one line. Heres the
first half:1
3BX C+
1
3CY A +
1
3AZB = ??? ()
Before you read ahead, try to figure out how and why this proves the problem.
(Jeopardy song plays. . . ) Ok, welcome back!
The three red triangles are isosceles 30
120
30 triangles, so they are all similar.
IfGandHdenote the centroids ofABCand XY Zrespectively, then the resultof the expression in equation () is triangle GHG, which (by the Theorem) mustalso be a 30 120 30 triangle. But two of its vertices are at the same place! Thetriangle has thus degenerated into a point, so all threeof its vertices are at the same
place, and G = H. QED
3.5 I Cant Take Any More Equilaterals! and the Asymmet-
ric Propeller
Lets do one more problem of a similar flavor before we move on.
Problem 4. Positively oriented equilateral trianglesXAB, XCD, andXEF share a vertexX. IfP, Q, andRare the midpoints ofB C, DE, andF A respectively, prove thatP QR is equilateral.
Problem 5 (Crux Mathematicorum). In quadrilateral ABCD, M is the midpoint of AB, and three equilateral
trianglesBC E, C DF, andDAG are constructed externally. IfNis the midpoint ofE F andP is the midpoint of
F G, prove thatM N Pis equilateral.
Problem 6. The four trianglesABC, AAbAc, BaBBc, andCaCbCare directly similar, andMa, Mb, andMc are
the midpoints ofBaCa, CbAb, andAcBc. Show thatMaMbMc is also similar to AB C.
AB
C
D
E
F
X
P
Q
R
Figure 11: Problem 4
A
B
CD
E
F
G
M
N
P
Figure 12: Problem 5
A
Ab
Ac
Ba
B
Bc
Ca
Cb
C
Ma
Mb
Mc
Figure 13: Problem 6
Hm, didnt I just say onemore problem?Indeed, all we have to do is solve problem 6; the rest come free. The diagrams are drawn to illustrate illustrate
how the first two problems are special cases of problem 6.4 So, lets solve problem 6, known as the Asymmetric
Propeller [2].
The problem gives us a plethora of similar triangles to work with, so our first instinct should be to try to write
triangleMaMbMc as an average of these:
MaMbMc = ABC+ aAAbAc+ bBaBBc+ cCaCbC. ()4In problem 4, the middle triangle has degenerated into point X. In problem 5, one of the outside triangles degenerates into point
F.
9
-
7/25/2019 When Did Imo Become Easy
11/29
The three triangles AAbAc, BaBBc, and CaCbCplay identical roles in the problem, so we have every reason to
guess thata= b= c. Now, for () to work, we need
MA=1
2(Ba+ Ca) = (+ a)A + a(Ba+ Ca).
If we set a = 1
2 and = a= 12 , then this works out perfectly (and likewise for Mb andMc). So does
MaMbMc = 12
ABC+1
2AAbAc+
1
2BaBBc+
1
2CaCbC
finish the proof?5 The triangles are all similar and the weights add to 1 (= 12
+ 12
+ 12
+ 12
), so yes. Sweet, another
one liner. QED
5Remember that negative weights are allowed, as long as the weights add to 1.
10
-
7/25/2019 When Did Imo Become Easy
12/29
4 Figure Addition
I keep stressing that the weights in a weighted average must add to 1. What would happen if they didnt?
A
B
M
O
P
Figure 14: Point addition
In the complex-number proof of MGT from section 2.2, theres really no reason
the weights must be k and (1 k). They could beanyreal numbers, and the proofworks the same! There is one subtle difference, though. In the diagram to the right,
use weights of1 = 2 = 1, and well try to calculate A + B. Where is it?
To add vectors or complex numbers like this, we need to know where the origin is.
If the origin is at, say, M, the midpoint between A andB, then A + B represents
the sum of the blue vectors
MA+MB =
0 = M. But if we put the origin
at a different point O, then the sum A+ B is now the sum of the red vectorsOA +
OB=
OP= P. So, withpoint addition, the sum depends on the location
of the origin, i.e. we must first specify an origin. With this minor change, MGT
extends yet again:
MGT: Figure Addition. For real numbers1 and2, define thesum of points 1P1+ 2P2 as the endpoint of
the vector1OP1+ 2
OP2, whereO is a specified origin. Then ifA1A2A3 andB1B2B3 are directly similar,
the sum of figures aA1A2A3 + bB1B2B3 (formed by adding corresponding vertices) is directly similar to the
original two triangles. As before, this naturally extends to more complicated figures and to more than two figures.
There is another way to visualize figure addition through dilation. For two similar triangles A1B1C1 and
A2B2C2, the sum A1B1C1+ A2B2C2 may be constructed by averaging the two triangles, 1
2A1B1C1+
1
2A2B2C2,
and then dilating this with center O and ratio 2.
4.1 When Did the IMO Get So Easy? (answer: 1977)
Problem 7 (IMO jury 1977 [6]). OAB andOAB are regular triangles of the same orientation, S is the centroid
ofOAB, andM andNare the midpoints ofAB andAB , respectively. Show thatSM B SN A (Figure15).
O
A B
A
B
S
M
N
W
X
Y
Z
Figure 15: Problem 7
(In the diagram, points W, X, Y , and Zare midpoints.)First of all, the red triangles both look like 30 60 90 triangles,
and if we can prove that, were done. The two equilateral triangles give
us plenty of 30 60 90s to work with; all we have to do is find theright ones!
Points O, Z, N, and W form a parallelogram, so if O is the
origin, Z + W = N. With a little experimentation, we arrive at
N SA = ZOA + W SO, and since bothZOA andW SO arepositively oriented 3060 90 triangles, so isN SA. Similarly,M SB =Y OB+ XSO (again withO as origin) proves that MSB isa negatively oriented 30
60
90 triangle. Whew, that was quick!
QED
4.2 A Fresh Look at an Old Result (1936, to be
precise)
Problem 8 (Pompeius Theorem [1]). Given an equilateral triangleABCand a pointP that does not lie on the
circumcircle of ABC, one can construct a triangle of side lengths equal to P A, P B, and P C. If P lies on the
circumcircle, then one of these three lengths is equal to the sum of the other two.
Erect equilateral triangles P CY and BP Xwith the same orientation asABC. WithPas origin, considerequilateral triangle P CY +BP X = BC A. It must be equilateral with the same orientation as ABC, which
11
-
7/25/2019 When Did Imo Become Easy
13/29
A
B C
P
X
Y
Figure 16: Pompeius Theorem
means A = Y +X = A, i.e. P Y A X is a parallelogram. Notice thatAP Y has AP = AP, P Y = CP, andY A= P X= BP, so if it is not degenerate,AP Y is the triangle were looking for.
When is this triangle degenerate, i.e. when are A,PandYcollinear? This happens if and only if (using directed
angles modulo 180) CP A= CP Y = 60 = CB A, i.e. quadrilateral ABCP is cyclic. And certainly, ifAP Yis degenerate, then one of its sides equals the sum of the other two.
QED
12
-
7/25/2019 When Did Imo Become Easy
14/29
5 A Few Harder Problems
At this point weve become proficient in utilizing the various forms of the Theorem, and weve learned to look for a
MGT approach when the problem hands us loads of similar triangles with which to play. But MGT can be useful
in many other situations as well, even if its application may be far from obvious. In this section, well look at a
few harder problems to stress that the MGT may not solve every problem immediately, but its a valuable method
to keep in mind as you explore a problem. It may be that MGT is only one of many steps in your solution. Or,
ideas related to MGT may lead you to a solution that doesnt use it at all. The point is that looking at a diagram
from an MGT viewpoint may tell you things that you formerly wouldnt have noticed.
This brings up another point: in order to recognize uses for MGT, youmust have an accurate diagram or
two, or three to look at. This article is filled with diagrams for exactly that purpose. (Whether youre solving a
geometry problem using MGT or not, its usually a good idea to have a decent diagram handy!)
On that note, lets bring on the problems.
5.1 A Pretty(,) Busy Diagram
Problem 9 (IMO Shortlist 2000, G6). LetABCD be a convex quadrilateral withAB not parallel to CD, and let
Xbe a point insideABCD such thatADX= BCX < 90 andDAX= CBX < 90. IfY is the point of
intersection of the perpendicular bisectors ofAB andC D, prove thatAY B= 2ADX.
A
B
C
D
X
Y
Figure 17: Problem 9
First thing to notice: by simply relabeling the diagram, it must also be true that DY C = 2DAX. So,
designate ADX= and DAX=. Next, notice that triangles AX D and B XCare similar. Ooh, that means
we should form another similar triangle 12
AXD + 12
BX C!Thats a decent thought, but unfortunately, even though
AXD and B XCare similar, theyre notdirectly similar.
It may not clear where to go from here, but since its necessary to start somewhere, well begin with the only
tangible fact we have: similar triangles AX Dand BXC. Its interesting that no matter how these two triangles are
hinged or scaled aroundX, pointYstill has its curious property. This suggests a possible direction for exploration:
scale the triangles and see what happens to point Y.
Lets leave triangleAXDfixed while we enlarge and shrink BX C. What size triangle would be easier to analyze?
How about zero! Consider quadrilateral ABCD, where BXC has shrunken to the degenerate triangle at point X(Figure 18). The corresponding Y is the intersection of the perpendicular bisectors ofAB =AXandDC =DX,
i.e. the circumcenterO of triangle AXD . Is it true that 2ADX = AOX? Yes, since arcAXof circle O hasmeasure 2. So when triangle B XC shrinks to zero, everything works out as expected.
What other size for triangle BX C might work well? Lets look at quadrilateralABCD, where BXC is
congruent to AXD as shown in Figure 19. To locateY, we should be looking for the perpendicular bisectors
13
-
7/25/2019 When Did Imo Become Easy
15/29
A
B
C
D
X = B = C
OY
Figure 18: Studying quadrilateral AB CD
ofAB and CD. But these are the same lineMN, so where is Y? We can instead locateY by finding the
point on this line so that AYB = 2, since this is another property our Ys should have. This means that
AYX= = ADX, so AX YD is cyclic, i.e. Y is the second intersection of circle O with line M N. (If the
circle happens to be tangent to line M N, second intersection simply means tangency point.)
A
B
C
D
B
C
X
M
N
O
YY
Figure 19: Quadrilateral AB CD
It seems that weve lost sight of our original problem. Weve studied quadrilaterals AXXD and ABCD,
but not the original ABCD. Luckily, a perusal of diagram 19 reveals the next step: trianglesAOX, AY B, and
AY B are similar since theyre all isosceles with vertex angle 2, so we should be able to average them. Indeed, if
XB/XB =k, then we should have (1
k)AOX+ (k)AYB =AY B. This is (almost) the last step to a complete
solution! A full, self-contained solution is given below.
Full Solution. Let XB/XA = k , and dilate B and Caround Xwith ratio 1/k to points B and C respectively,
so that ABCD is an isosceles trapezoid. M and N are the midpoints of AB and CD. Define O as the
circumcenter of triangle AX D, and let Y be the second intersection of circle ADXwith lineM N. IfADX=
and DAX= , it follows that AOX= 2, AYX= which implies AYB = 2, and likewise,DOX= 2
and DYC = 2. Therefore, the isosceles trianglesAOX and AYB are similar, as are triangles DOX and
DYC.
Define Y1 = (1 k)O+ (k)Y, and notice that triangles AY1B = (1 k)AOX+ (k)AYB and DY1C =
14
-
7/25/2019 When Did Imo Become Easy
16/29
(1 k)DOX+ (k)DYC are isosceles by MGT. Thus, Y1 is the intersection of the perpendicular bisectors ofABandC D, i.e. Y1= Y. Furthermore, by our application of MGT, AY B= AOX= 2, QED.
QED
5.2 Hidden Circles
Problem 10(USA TST 2005 #6). LetABCbe an acute scalene triangle withO as its circumcenter. PointP lies
inside triangleAB CwithP AB= P BCandP AC= P CB . PointQ lies on lineB CwithQA = QP. Prove
thatAQP= 2OQB.
A
BC
Q
O
M
P
X
Figure 20: Problem 10
Most of the diagram is straightforward: Ois the circumcenter,Qis the intersection ofBCwith the perpendicular
bisector ofAP, and Ive added M, the intersection ofAP with BC. Everything is simple to navigate exceptP
itself, so thats where well start investigating.
The first strange angle equality, i.e. P AB = P BC, shows thatM AB M BP. Thus, MA/MB =MB/MP, or (MA)(MP) = (MB)2. This shows, by power of a point, that the circle through A, P, and B is
tangent to line M B at B.6 Likewise, the circle throughA,P, andCis tangent to M CatC. (Call these two circles
[and their centers]O1 and O2 respectively.) Finally, (MB)2 = (MA)(MP) = (MC)2, soM is the midpoint ofBC.
A
BC
Q
OO1
O2
M
P
X
Figure 21: Circles O1 andO2
The diagram becomes much clearer from the viewpoint of the two circles. The circles intersect at A and P, and
line BC is their common tangent. PointQ, being the intersection of their axis of symmetry (line O1O2) with a
common tangent (line B C), must be the center of homothecy between circles O1 andO2.
LetXbe the midpoint ofAP. Were asked to prove that AQP= 2OQB, or equivalently, AQX= OQM.
But triangles AQX and OQM are both right triangles, so we need to prove that they are similar. Heres where
6If this circle intersected line MB at some other point B, then power of a point would show that (MB )2 = (MA)(MP) =(MB )(MB ), so MB = MB and B= B.
15
-
7/25/2019 When Did Imo Become Easy
17/29
Mean Geometry might come in handy: to prove that these triangles are similar, we might be able to find a third
triangle similar to, sayAQX, and then express triangle OQM as a weighted average of those two. So now wehunt for a third similar triangle. In order for this to work, this triangle should have one vertex at Q, one vertex
along line X M, and one vertex along line AO. What key points are on these two lines? JustA and P. And since
A is already being used, lets think about P. If trianglesAQX and OQMwere similar to triangle ZQP, where
would this mystery point Zhave to be? We already want it on line AO, and since AQX= XQP = ZQP, we
would also need Zto be on line QX. So define Z= AO QX, and lets see ifZQP is indeed the triangle werelooking for. First of all, is it a right triangle?
Angle QP Z is a right angle if and only ifQAZ is right, so we need QA OA, i.e. QA should be tangentto circle O at A. It turns out to be true, as follows. LetTbe the second intersection ofQA with circle O1. IfrQ (where r = QO2/QO1) is the homothecy taking circle O1 to circle O2, we have rQ(T) = A and rQ(B) = C,
so ACB = T BQ = T AB. Thus, trianglesQBA and QAC are similar, so (QB)(QC) = (QA)2, and QA is
tangent to circle O. So (by tracing backwards through a few lines of reasoning above), triangle Z QPactuallydoes
have a right angle at P. This means it is similar to triangle AQX.
A
BC
Q
OO1
O2
M
Z
T
P
X
Figure 22: Three similar triangles
The next part of our initial plan was to find a weighted average of triangles Z QP andAQXthat would produce
triangleOQM, thus completing the proof. So, all we need to show is that AO/ZO= XM/P M. Good luck.
Those segment lengths arent easy to calculate, even with plenty of paper and tons of time. For the first time
in this article, MGT fails to miraculously save the day. But weve come far enough with the MGT idea so that the
proof is moments away. Let rQ be the spiral dilation centered at Q that sendsX P toAZ. SinceMis on lineX P,rQ(M) = M
is on lineZ A. Also, since triangleMQMis similar to triangleAQX, we have QMM = 90, soM
is also on the perpendicular bisector ofB C. This means rQ(M) =M = O. Thus,OQM = AQX, as desired.
(Notice that, even though MGT was a driving force for most of the solution, not a single mention of it is necessary
in the final writeup.) QED
5.3 Nagel Who?
Problem 11 (USAMO 2001 #2). Let ABC be a triangle and let be its incircle. Denote by D1 and E1 the
points where is tangent to sides BC and AC, respectively. Denote by D2 andE2 the points on sides BC and
AC, respectively, such thatCD2 = B D1 andCE2 = AE1, and denote byP the point of intersection of segments
AD2 andB E2. Circle intersects segmentAD2 at two points, the closer of which to the vertexA is denoted byQ.
Prove thatAQ= D2P.
In the diagram Ive added F1 and F2 to preserve symmetry, and I included the midpoints of the sides of the
triangle.
16
-
7/25/2019 When Did Imo Become Easy
18/29
A
B CD
EF
D1
E1
F1
D2
E2
F2
Q
R
SIP
Figure 23: Problem 11
A
BC
D1
E1
F1
D2
I
Ia
Q
Figure 24: Excircle Ia
Using the standard notations a = BC, b= C A, c = AB, and s = 1
2 (a + b + c), it is relatively well known thatBD1 = s b. Indeed, ifAE1 = AF1 = x, BF1 = BD1 = y, C D1 = CE1 = z, then the system y + z = a, z + x= b,x + y= c can be solved to give B D1 = y = (c + a b)/2 = s b. Likewise, if the excircle opposite A is tangent toBCatV, a similar calculation shows that C V =s b= BD1, i.e. V =D2. SoD2, E2,F2 are the tangency pointsofB C, C A, AB with the triangles three excircles.
Another relatively well-known point in the diagram is P, commonly referred to as the Nagel point of triangle
ABC. Its simply the intersection of the three ceviansAD2,BE2, andC F2, which must concur by Cevas theorem:
BD2D2C
C E2E2A
AF2F2B
=s cs b
s as c
s bs a = 1.
A useful property of the Nagel point,7 other than its mere existence, is how its defining cevians interact with the
incircle: lineAD2intersects the incircle at point Qdiametrically opposite fromD1. To prove this, letr = AI/AIa(in
Figure 24), so that rAis the homothecy centered atA taking excircle IA to incircle I. SinceD2IA is perpendicular
to B C, its image through rA, namely QI, is also perpendicular to BC. Therefore, QD1 is a diameter of circle I.
The same goes for diameters E1R and F1S.
Now to the problem. The midpoints D , E, andF inspire us to consider
1
2D1E1F1+1
2D2E2F2 = DEF. ()
TrianglesD1E1F1and D2E2F2arent similar, so MGT doesnt tell us anything directly. But its still worth noticing.
If its true that AQ = D2P, it must also happen that B R= E2P and C S= F2P. This means that, with P as
origin, wed like to be able to show that
ABC+ D2E2F2 = QRS.
Again, these triangles are not similar, so MGT isnt applicable. But this equation has striking similarities with
(). These similarities become more pronounced if we rewrite equation () as
2DEF+ D2E2F2 = D1E1F1 ()7The Nagel point is also (and less commonly) known as the bisected perimeter point [8] or the splitting center [7], since the cevians
AD2, etc., bisect the perimeter of the triangle, i.e. AB + BD2 = AC+ CD2 = s.
17
-
7/25/2019 When Did Imo Become Easy
19/29
A
B CD
EF
D1
E1
F1
D2
E2
F2
Q
R
SIP
Figure 25: Problem 11
Triangle 2DEFis simply a translation ofABC, and triangle D1E1F1is a translation ofQRS. The two equationsare almost identical! This inspires the following observation:
Non-similar Figure Addition. If three triangles (figures) satisfy11+ 22 = 3, and if1 is a translatedversion of1, then11+ 22 is a translation of3, regardless of the location of the origin.
Indeed, if1 = 1+V,8 then
11+ 22 = 11+ 22+ 1V = 3+ 1V.
Now we can finish the problem. Beginning with equation (), translate2DEF to coincide with AB C. Theabove observation (with our origin still at P) proves thatQRS =ABC+ D2E2F2 must be a translation ofQRS.
But sinceA + D2 = Q lies onAD2, and likewise forR andS, trianglesQRSand QRS are also homothetic with
center P. So the two triangles must be identical, and in particular, Q = Q =A + D2. SoAQ=
PD2, proving the
desired result. QED
8i.e. a translation by vector
V
18
-
7/25/2019 When Did Imo Become Easy
20/29
6 Additional Problems
Problem 12. Recall problem 4: Positively oriented equilateral trianglesXAB, XC D, andXEFshare a vertex
X. IfP, Q, andR are the midpoints ofB C, DE, andF A respectively, prove thatP QR is equilateral. Prove this
problem using Napoleons Theorem.
Problem 13 (Crux Mathematicorum). A line parallel to the side AC of equilateralABC intersects B C at Mand AB at P, thus making BM P equilateral as well. D is the center ofBM P and E is the midpoint of CP.Determine the angles ofADE.
Problem 14. The following theorem appears in Geometry Revisited[5] as a special case of a theorem of Petersen
and Schoute: IfABCandABC are two directly similar triangles, whileAAA,BBB,C CC are three directly
similar triangles, thenABC is directly similar toABC.
a. Show that this theorem generalizes the triangle version of MGT.
b. What minor adjustment can be made to the statement of MGT to account for this generalization (and its
proof)?
c. Show that Napoleons theorem is a special case of this theorem.
Problem 15 (Van Aubels Theorem). Given an arbitrary planar quadrilateral, place a square outwardly on each
side, and connect the centers of opposite squares. Then these two lines perpendicular and of equal length.
Problem 16. Equilateral triangles AEB, BF C, CGD, DH A are erected outwardly on the sides of a plane
quadrilateralABCD.
a. Let M, N, O, and Pbe the midpoints of segments EG, HF, AC, and BD respectively. What is the shape
ofPMON?
b. Md andMa are the centroids ofDAH andAEB, and equilateral triangle MdT Ma is oppositely orientedwith respect to ABCD. Find the angles of triangleF T G.
c. Ma and Mc are the centroids ofAEB and CGD. Prove that segmentsMaMc and F Hare perpendicular,and, in addition,|F H| = 3 |MaMc|.
d. Equilateral triangles EW F, F XG, GY H, and HZ Eare oppositely oriented with respect to ABCD. Prove
that quadrilaterals ABCD and W XY Z have the same area.
Problem 17. Let (P,QR) denote the line through point Pperpendicular to line QR. Say thatXY Zperpen-dicularizesABC if (X,BC), (Y, CA), and (Z,AB) concur at a point. IfXY Z perpendicularizesABCandDEF, prove thatXY Zalso perpendicularizes any linear combination ofABCandDEF.
Problem 18 (Alex Zhai).
a. In triangle ABC, AD, BE, and CF are altitudes. Dc and Db are the projections ofD onto AB and AC,
respectively, and points Ea,Ec,Fa, and Fbare defined similarly. Prove that quadrilaterals BDcDbC,CEaEcA,
andAFbFaB are cyclic.
b. Let OA be the center of circle BDcDbC, and let Ta be the midpoint of altitude AD. Similarly define Ob,
Oc, Tb, and Tc. IfO is the circumcenter of triangle ABC, show that AOOaTa is a parallelogram, as well as
BOObTb andC OOcTc.
c. Prove that lines OaTa, ObTb, and OcTc are concurrent.
19
-
7/25/2019 When Did Imo Become Easy
21/29
Problem 19. LetP be a convex polygon in the plane. A real number is assigned to each point in the plane sothat the sum of the number assigned to the vertices of any polygon similar toP is equal to 0. Prove that all theassigned numbers are equal to 0.
Problem 20. Show that I, G, and N(the incenter, centroid, and Nagel point of a triangle) are collinear in that
order with 2 IG = GN. Hint: see problem 11.
Problem 21. Given a convex quadrilateral ABCD, construct (with ruler and compass) a square of the sameorientation with one vertex on each side ofABCD.
20
-
7/25/2019 When Did Imo Become Easy
22/29
7 Solutions to Additional Problems
Problem 12. Recall problem 4: Positively oriented equilateral trianglesXAB, XC D, andXEFshare a vertex
X. IfP, Q, andR are the midpoints ofB C, DE, andF A respectively, prove thatP QR is equilateral. Prove this
problem using Napoleons Theorem.
Solution. In diagram 26, trianglesBCJ,DE K,F ALare equilateral, andG,H,I,M,N,O are the centers of their
respective triangles. By napoleons theorem thrice,HGM,IH N, andGIO are equilateral, and their centersT, U, Vare also the centroids ofXBC,XDE, andXF A respectively (by problem 3). Again by napoleonstheorem (on the blue triangles),T U V is equilateral, and the dilation 3/2XcarriesT U V toP QR.
AB
C
D
E
F
X
G
H
I
J
K
L
M
N
OT
U
V
P
Q
R
Figure 26: Problem 12
AB
C
TS
X
P
E
M
D
Figure 27: Problem 13
A
B
A = C
B
C A
B
C
Figure 28: Problem 14.c
QED
Problem 13 (Crux Mathematicorum). A line parallel to the side AC of equilateralABC intersects B C at Mand AB at P, thus making BM P equilateral as well. D is the center ofBM P and E is the midpoint of CP.
Determine the angles ofADE.
Solution. S and Tare the midpoints ofCB and CA, and X is the center ofABC. Let BP/BA = k . BecauseE must lie on line ST, and sinceCS T CB A, SE/ST = BP/BA = k. Also, sinceBP D BAX,BD/BX=BP/BA= k. Thus, (1 k)ABS+ (k)AXT =ADE, soADEis a 30 60 90 triangle. QED
Problem 14. The following theorem appears in Geometry Revisited[5] as a special case of a theorem of Petersen
and Schoute: IfABCandABC are two directly similar triangles, whileAAA,BBB,C CC are three directly
similar triangles, then
ABC is directly similar to
ABC.
a. Show that this theorem generalizes the triangle version of MGT.
b. What minor adjustment can be made to the statement of MGT to account for this generalization (and its
proof)?
c. Show that Napoleons theorem is a special case of this theorem.
Solution. a. In the special case where A is on line AA, the similarity of degenerate triangles AAA, BB B,
andC CC simply means that AA/AA =BB/BB =C C/CC =k. Now, the fact that ABC is similar to
21
-
7/25/2019 When Did Imo Become Easy
23/29
A
B
C
D
E
F
G
H
I
J
K
L
M
N
O
P Q
RS
T
Figure 29: Van Aubels Theorem
ABC is exactly the statement of MGT, since
(1 k)ABC+ (k)ABC =ABC.
b. In the complex number proof of MGT (section 2.2), we required the two weights k and 1 k to be real numbersthat add to 1. If we allow the weights to becomplex, we obtain this generalization. The proof in section 2.2 remains
unchanged.
c. The napoleon diagram has been relabeled in figure 28 to show the correspondence. QED
Problem 15 (Van Aubels Theorem). Given an arbitrary planar quadrilateral, place a square outwardly on each
side, and connect the centers of opposite squares. Then these two lines perpendicular and of equal length.
Solution. Q, R, S, and T are the midpoints of the sides of quadrilateral MNOP. Since 14
(A+ F +G + C) =
12 (M+ N) = Q, and likewise around, averaging the four given squares proves that QRSTis itself a square:
1
4FBAE+
1
4GHCB+
1
4CIJD+
1
4ADKL= QRST.
SinceMO= 2
TS and
PN= 2
TQ, and since T S and T Q are equal in length and perpendicular, diagonals M O
andN Pmust share this property, as desired.
QED
Problem 16. Equilateral triangles AEB, BF C, CGD, DH A are erected outwardly on the sides of a plane
quadrilateralABCD.
a. Let M, N, O, and Pbe the midpoints of segments EG, HF, AC, and BD respectively. What is the shapeofPMON?
b. Md andMa are the centroids ofDAH andAEB, and equilateral triangle MdT Ma is oppositely orientedwith respect to ABCD. Find the angles of triangleF T G.
c. Ma and Mc are the centroids ofAEB and CGD. Prove that segmentsMaMc and F Hare perpendicular,and, in addition,|F H| = 3 |MaMc|.
d. Equilateral triangles EW F, F XG, GY H, and HZ Eare oppositely oriented with respect to ABCD. Prove
that quadrilaterals ABCD and W XY Z have the same area.
22
-
7/25/2019 When Did Imo Become Easy
24/29
A
B
CD
E
F
G
H
M
N
O
P
Figure 30: Problem 16.a
A
B
CD
E
F
G
HT
Ma
Md
Figure 31: Problem 16.b
A
B
CD
E
F
G
H
Ma
Mc
I
J
K
LMNO
P
Figure 32: Problem 16.c
A
B
CD
E
F
G
H
P
Q
R
S
W
X
Y
Z
Figure 33: Problem 16.d
Solution. a. Since 12
ABE+ 12
CDG = OP M and 12
BC F+ 12
DAH = P ON, trianglesOP M andP ON areequilateral, so quadrilateral MONPis a rhombus with vertex angles of 60 and 120.
b. By Napoleons theorem on triangle DAB, Tmust be the center of the equilateral triangle erected on sideDB, i.e. DT B is a 30 120 30 triangle. Now apply problem 9 to quadrilateral DBFG. Point C hasGDC = F BC = 60 and DGC = BF C = 60, so problem 9 guarantees the existence of a point Y on
the perpendicular bisectors ofDB andF Gso that DY B = GY F= 120. Tis on the perpendicular bisector of
DB and has DT B= 120, soT is Y, and thusGT F is a 30 120 30 triangle.c. This solution mimics the proof of Van Aubels Theorem (problem 15). Erect rectangles JBAI,KLCB,CMND,
andADOB with
3 : 1 side ratios as shown. The average of these four rectangles (with vertices in the listed order)
is rectangle W XY Z (not shown) which connects the midpoints of the sides of quadrilateral MaF McHand which
must be similar to the red rectangles. And as explained in problem 15, since the Varignon Parallelogram [5,
Theorem 3.11] of quadrilateral MaF McHhas sides that are perpendicular with a ratio of
3/1, the diagonalsF H
andMaMc must have this property too.
d. Recall from problem 8 that, with C as center,DGC+ BC F =XGF, proving that D+ B = X andso XDCB is a parallelogram. Likewise, ADCY is a parallelogram. Segments XB and AYare both parallel and
congruent to DC, so XAY B is a parallelogram, and AB and XY share the same midpoint P. The same goes
for Q, R, and S. Thus, quadrilaterals ABCD and W XY Z have the same Varignon Parallelogram PQRS, so
area(ABCD) = 2 area(PQRS) = area(W X Y Z ). QED
Problem 17. Let (P,QR) denote the line through point Pperpendicular to line QR. Say thatXY Zperpen-dicularizesABC if (X,BC), (Y, CA), and (Z,AB) concur at a point. IfXY Z perpendicularizesABCandDEF, prove thatXY Zalso perpendicularizes any linear combination ofABCandDEF.
23
-
7/25/2019 When Did Imo Become Easy
25/29
Solution. The following lemma is key: XY Z perpendicularizesABC if and only ifABC perpendicularizesXY Z. To finish problem 17 with this property, note that ifX Y Zperpendicularizes ABC andDEF, then bothABC and DEF perpendicularize XY Z, say at points P and Q respectively. Define GHI = 1ABC+ 2DEF
and R = 1P +2Q. Segment GR as a linear combination of AP and DQ is perpendicular to Y Z, and
likewise for H Rand I R. SoGH Iperpendicularizes X Y ZatR, and the above lemma guarantees that X Y Zmust
perpendicularizeGH I.
We offer two very different proofs of the lemma.
Proof 1. Well make use of the following property: lines P Qand RSare perpendicular if and only ifP R2 + QS2 =
P S2 + QR2. Indeed, using as the vector dot product,
(P R) (P R) + (Q S ) (Q S ) = (P S ) (P S ) + (Q R) (Q R)
is equivalent to (P Q) (R S ) = 0 after expanding and simplifying.
For two trianglesABCandX Y Zin the plane, let(Y,AC) and(Z,AB) meet atQ.XY ZperpendicularizesABCif and only ifX Q BC, i.e. if and only if
0 = (CQ2
QB2) + (BX2
XC2)
= (CQ2 QA2) + (BX2 XC2) + (AQ2 QB2)= (CY2 Y A2) + (BX2 XC2) + (AZ2 ZB2).
But this last expression makes no distinction between triangles ABCand X Y Z, so if one triangle perpendicularizes
another, the other must perpendicularize the first. QED
A
B C
Q
X
Y
Z
L
M
N
Figure 34: Proof 2 of perpendicularization lemma
Proof 2. AssumeXY Z perpendicularizesABC at point Q. Draw three lines: line a through X parallel toBC, lineb throughYparallel toC A, and linec throughZparallel toAB. These lines determine triangle A
BC
homothetic toABC, and so ABCperpendicularizes X Y Zif and only ifABC does.Let L, M, Nbe the projections ofA
, B
, C
onto Y Z, ZX, X Y respectively. QuadrilateralA
ZQY is cyclic,so we may calculate
CAL= 90 ZY A = 90 ZQA = QAB,
and likewise for B M andCN. Thus,
sinCAL
sinLAB sinA
BM
sinMBC sin B
CN
sinN CA =
sinQAB
sinCAQ sin QB
C
sinABQ sinQC
A
sinBCQ= 1
i.e. AL, BM, and CN do in fact concur by the trigonometric form of Cevas theorem. (They concur at the
isogonal conjugate ofQ with respect to triangle ABC.) QED
24
-
7/25/2019 When Did Imo Become Easy
26/29
Problem 18 (Alex Zhai).
a. In triangle ABC, AD, BE, and CF are altitudes. Dc and Db are the projections ofD onto AB and AC,
respectively, and points Ea,Ec,Fa, and Fbare defined similarly. Prove that quadrilaterals BDcDbC,CEaEcA,
andAFbFaB are cyclic.
b. Let OA be the center of circle BDcDbC, and let Ta be the midpoint of altitude AD. Similarly define Ob,
Oc, Tb, and Tc. IfO is the circumcenter of triangle ABC, show that AOOaTa is a parallelogram, as well as
BOObTb andC OOcTc.
c. Prove that lines OaTa, ObTb, and OcTc are concurrent.
Solution. a. By similar triangles ADcD and ADB, ADc/AD = AD/AB, i.e. (ADc)(AB) = (AD)2. Likewise,
(ADb)(AC) = (AD)2 = (ADc)(AB), so B DcDbC is cyclic by the converse of power-of-a-point. Similar arguments
work for the other two quadrilaterals.
A
B CD
F
Mc
Oa
H
O
Ta
Ta
Dc
Oa
Figure 35: Problem 18.b
A
B CD
E
F
Ma
MbMc
Oa
Ob
Oc
HO
Ta
Tb
Tc
Figure 36: Problem 18.c
b. ATaand OOaare parallel because they are both perpendicular to BC, so it is only necessary to show they have the
same length. ProjectTa, Oa, andO ontoAB at pointsTa,Oa, andMc, which must be the midpoints ofADc,AB,
and DcB respectively. We have TaA= DcA/2 and OaMc = BMc BO a = (BDc+ DcA)/2 (BDc)/2 = DcA/2.
Since the projections ofTaA and OaO ontoAB are equal in length, and sinceB A and AD are not perpendicular,
TaA= OaO, as desired.
c. It can be calculated that OAC+ AEF = (90 CB A) + (CB A) = 90, so AO EF, i.e. OaTa E F.Thus, we wish to prove that OaObOc perpendicularizes DEF (see problem 17).
Let the midpoints ofB C, CA, AB be Ma, Mb, Mc, and let Hbe the orthocenter ofABC. SinceMaMbMc is
similar toABCwith half its size, and since O is the orthocenter ofMaMbMc,OMa= 1
2AH. Since we have already
proved that OOa = 1
2AD, it follows that MaOa =
1
2HD. So with Has origin, OaObOc = MaMbMc +
1
2DEF,
i.e. DEF = 2OaObOc 2MaMbMc. And since OaObOc perpendicularizes MaMbMc (at O) and OaObOc (at itsown orthocenter), triangle OaObOc must perpendicularize the linear combination 2OaObOc 2MaMbMc = DEFby problem 17, as desired.
QED
Problem 19. LetP be a convex polygon in the plane. A real number is assigned to each point in the plane sothat the sum of the number assigned to the vertices of any polygon similar toP is equal to 0. Prove that all theassigned numbers are equal to 0.
25
-
7/25/2019 When Did Imo Become Easy
27/29
Solution. Letf(T) be the number associated with point Tin the plane.
The idea is to use an extended version of problem 14 (illustrated for n = 4 in diagram 37), and then to let one
of the polygons shrink to a single point X (figure 38).
A1,1
A2,1
A3,1
A4,1
A1,2
2,2
A3,2
A4,2
A1,3
A2,3
A3,3
A4,3
A1,4
A2,4
A3,4
A4,4
Figure 37: Generalization of problem 14
A1,1
A2,1
A3,1
A1,2
2,2
A3,2
A1,3
A2,3
A3,3
A1,4
A2,4
A3,4
X
Figure 38: Problem 19
An arbitrary point Xis chosen, polygonR1 = A1,1A1,2 A1,n similar toP is drawn with no vertices at X,and then polygonsBi = A1,iA2,i An1,iX (1in) are drawn all similar toP. It follows that each polygonRi = Ai,1Ai,2 Ai,n (1 i n 1) is also similar toP (see problem 14). Thus, since
bBi
f(b) = 0 andrRj
f(r) = 0 for any 1 i n and 1 j n 1, we can add over polygonsBi and over polygonsRi to cancelout most of the terms:
0 =ni=1
bBi
f(b) = n f(X) +ni=1
n1j=1
f(Aj,i) = n f(X) +n1j=1
rRj
f(r) = n f(X) + 0.
This means f(X) = 0, and since point Xwas arbitrary, this holds for all points in the plane. QED
Problem 20. Show that I, G, and N(the incenter, centroid, and Nagel point of a triangle) are collinear in that
order with 2 IG = GN. Hint: see problem 11.
Solution. Refer to figure 25 in problem 11. With originP, define DEF =2DEF and D1
E1
F1
=D1E1F1.These are the triangles used but never drawn in equation () on page 17, and they are translations ofABC
and QRS respectively. Also set X = 2D. SinceAQ =
PD2 =
D1X =
DQ, it follows that triangles DEF
and QRS are in the same relative position as AB C and QRS, i.e.AD =
QD
1. And sinceAD = 3
GP (since
ADD GDP) and QD1
= 2IP(since QD1D1 ID1P), the conclusion follows. QED
Problem 21. Given a convex quadrilateral ABCD, construct (with ruler and compass) a square of the same
orientation with one vertex on each side ofABCD.
Solution. First, we prove that it is impossible for AB CD andAD BC to occur simultaneously. Assume thisdoes happen, label AB CD = X and AD BC=Y, and assume (without loss of generality) that C is betweenXand D and between Y andB . Then
360 > B+ C+ D= 540 C >360,
contradiction. So it is safe to assume B Cis not perpendicular to DA.
Given a pointPi AB, construct square i= PiQiRiSi as follows: rotate line DA clockwise by 90 aroundPito intersect B C at Qi (this intersection exists uniquely since BC DA), and complete positively oriented square
26
-
7/25/2019 When Did Imo Become Easy
28/29
A
B CD
EF
D1
E1
F1
D2
E2
F2
Q
R
SIP
D
E F
D1
E1
F1
G
X
Figure 39: Problem 20
AB
C
D
P
Q
R
S
P0
Q0
R0
S0
P1
Q1
R1
S1
Figure 40: Problem 21
PiQiRiSi. This square with vertexPihas the properties thatQi BCand Si DA(by rotation), and furthermore,the construction proves that such a square is unique. Now it is only necessary to find the rightP AB so that thecorrespondingR lies onC D.
Choose two distinct points P0, P1 AB and draw squares 0 and 1 as above. Choose any other pointPt = (1 t)P0+ (t)P1 on AB, and consider t = PtQtRtSt = (1 t)0+ (t)1. Since Qt Q0Q1 BC andSt S0S1 DA, square t satisfies the defining conditions for t, i.e. t = t. In particular, Rt = Rt =(1 t)R0+ (t)R1, meaning Rt must lie on line R0R1. And sinceRt covers all of this line as t varies, the locus ofsuch points is exactly this line, i.e. R = C D
R0R1. The rest of the square can no be constructed now that the
correct ratiot = R0R/R0R1 is known.There are a few exceptional cases to consider. If lineR0R1 and line CD are identical, any point Pt AB will
produce a viable square t. If the two lines are parallel but not equal, there is no square with the desired properties.
And finally, ifR0 = R1, thenRt = R0 for all t, so all Pt work or no Pt work depending on whether or not R0 is on
lineC D. QED
27
-
7/25/2019 When Did Imo Become Easy
29/29
References
[1] Andreescu, T. and Gelca, R. Mathematical Olympiad Challenges.Boston, MA: Birkhauser, 2002.
[2] Bogomolny, A. Asymmetric Propeller. http://www.cut-the-knot-org/Curriculum/Geometry/
AsymmetricPropeller.shtml
[3] Bogomolny, A. A Problem in an Equilateral Triangle. http://www.cut-the-knot.org/Curriculum/
Geometry/RightInEquilateral.shtml
[4] Bogomolny, A. Three Similar Triangles. http://www.cut-the-knot.org/Curriculum/Geometry/
SimilarTriangles.shtml
[5] Coxeter, H.S.M and Greitzer, S.L. Geometry Revisited.Washington, DC: Math. Assoc. Amer., 1967.
[6] Engel, A.Problem Solving Strategies.New York: Springer, 1998.
[7] Honsberger, R. Episodes in Nineteenth and Twentieth Century Euclidean Geometry.Washington, DC: Math.
Assoc. Amer., 1995.
[8] Weisstein, Eric W. Nagel Point. FromMathworldA Wolfram Web Resource. http://mathworld.wolfram.com/NagelPoint.html