IMC Simulation

June 15, 2009

Problem 1. For every 2 × 2 real matrix A. Show that it is possible to find a matrix B and a
symetrix matrix C for which A = B + C and det A = det B + det C, and det B ≥ 0 ≥ det C.
 
a b
Solution: Letting A = , we have
c d

1
(a + d)2 + (b − c)2 − (a − d)2 − (b + c)2


det A = ad − bc =
 a+d 4 b−c   a−d b+c 
= det c−b 2 2 + det b+c 2 2 .
a+d d−a
2 2 2 2

a+d b−c a−d b+c

   
Thus, we may set B = 2 2 and C = 2 2 .
c−b a+d b+c d−a
2 2 2 2

Problem 2. For n = 1, 2, · · · , determine the subset of (0, 1) on which

 n
d
{ln x · ln(1 − x)} < 0
dx

Solution: If f (x) = ln x · ln(1 − x) then

∞
ln(1 − x) ln x X (1 − x)m−1 − xm−1
f 0 (x) = − =
x 1 − x m=1 m

From this it is evident that for n even f (n) (x) is always negative because all terms are
negative, while for n odd f (n) (x) is negative if an only if 1 − x < x, or x > 12 .

Problem 3. Consider the real sequence (xn ) defined by the recurrence

1
xn+1 = n≥1
1 + xn
For which x0 does lim xn exist. And find the value of that limit.

1
Solution: First assume x0 is chosen so that the sequence is well defined, i.e., xn 6= −1 for
all n. A simple induction argument on n yields

x0 Fn−1 + Fn Fn−1 x0 + Fn /Fn−1

xn = = (1)
x0 Fn + Fn+1 Fn x0 + Fn+1 /Fn

where Fk is the kth Fibonacci number. If x0 = −Fn+1 /Fn , n ≥ 1, then xn is undefined,

so clearly the sequence does
√ not converge for these choices of x0 . It is well known that
1
lim Fn+1 /Fn = ϕ = 2 (1 + 5). When x0 6= ϕ the denominator in (1) has nonzero limit and
thus lim xn = 1/ϕ. If x = −ϕ then using identity 1/(1 − ϕ) = −ϕ, we have xn = −ϕ for all
n.

Problem 4. Let G be a finite group and H a subgroup of G. Suppose S is a nonempty

subset of G with the property that x ∈ S implies x2 6∈ H. Prove that there exists a set
T ⊂ S,
1
|T | ≥ |S|,
2
such that a, b ∈ T implies ab 6∈ H or ba 6∈ H.

Solution: Let K = {T ⊂ S|a, b ∈ T implies ab ∈ / H or ba ∈ / H}. We have K 6= ∅, since

each singleton T = {s} lies in K by hypothesis. Since G is finite, the set K has an element
T such that |T | is maximal.
Suppose that |T | < 21 |S|. Then |S\T | > |T |. Since |T | is maximal, we have S\T ∈ /
K so that there exists s1 , s2 ∈ S\T such that s1 s2 ∈ H and s2 s1 ∈ H. Since |T | is
maximal and s1 ∈ / T , we have T ∪ {s1 } ∈ / K so there exists t1 ∈ T such that s1 t1 ∈ H
and t1 s1 ∈ H. Similarly, there exists t2 ∈ T such that s2 t2 ∈ H and t2 s2 ∈ H. Thus
t1 t2 = (t1 s1 )(s−1 −1 −1 −1
1 s2 )(s2 t2 ) ∈ H and t2 t1 = (t2 s2 )(s2 s1 )(s1 t1 ) ∈ H. This contradicts t1 ∈ T
1
and t2 ∈ T . Therefore |T | ≥ 2 |S|

Problem 5. Let n be a positive integer. An n × n matrix A is a magic matrix of order m

if each entry
P is a non-negative
P integer and each row and column sum is m. (That is, for all
i and j, k aik = k akj = m.) Let A be a magic matrix of order m. Show that A can be
expressed as the sum of m magic matrices of order 1.

Solution: Consider the magic matrix A as the adjacency matrix of a weighted bipartite
graph G between two sets (”‘left” and ”‘right”) of n vertices: If the (i, j)th entry of A is
greater than 0, place an edge in G between the ith vertex on the left and the j th vertex on
the right and give the edge a weight equal to the (i, j)th entry of A. If the (i, j)th entry of
the A is 0, do not place an edge at all. The condition that the matrix A is a magic matrix
implies that total weight of all the edges emanating from any single vertex of G, left or right,
is equal to m.
Given a subset S of the left-hand vertices of G, let us compute the size of its neighbour-
hood (the collection of all vertices on the right which are joined by an edge to a vertex in
S). Remove from G all of the edges whose left-hand vertices are not in S. Then the total
weight of the remaining edges is exactly m|S|, and the neighbourhood of S is exactly the set
of right-hand vertices whose weight is still non-zero. (By the weight of a vertex, we mean the
weight of all the edges touching that vertex.) But each right-hand vertex has weight at most
m, so by the Pigeonhole Principle the number of right-hand vertices with non-zero weight
must be at least m|S|/m = |S|. That is, every subset on the left has a neighbourhood on the
right which is at least as big. Thus, the conditions of the Marriage Theorem are satisfied,
with the left-hand vertices as boys, the right-hand vertices as girls, and edges as acceptable
marriages.
Theorem (Hall’s Marriage Theorem). Suppose there are n boys and n girls, and that
each boy knows precisely which (possibly more than one) of the girls he is willing to marry.
Suppose further that given any set S of boys, the total number of different girls that boys in
S are willing to marry is at least S. Then there exists a way of pairing all the boys with the
girls in such a way that each boy is willing to marry the girl to whom he is paired.
In our case, if the pairing obtained from the Marriage Theorem pairs the vertex i on the
left with the vertex σi on the right, then we know that the (i, σi )th entry of A is positive. Let
A0 be the matrix whose (i, σi )th entry is 1 for all i and whose other entries are all 0. Then
A0 is a magic matrix of order 1 and A − A0 is a magic matrix of order m − 1, and the result
follows by induction.

Problem 6. Let f be a function from the plane R2 to the reals. Given a polygon P in the
plane, let f (P) denote the sum of the values of f at each of the vertices of P. Suppose there
exists a convex polygon Q in the plane such that for every polygon P similar to Q, we find
that f (P) = 0. Show that f is identically zero.

Solution: Let v1 , · · · , vk denote the vertices of Q. Given any x ∈ R2 and integer n, the
polygon with vertices x + n · v1 , · · · , x + n · vk is similar to Q, and so by hypothesis we have
k
X
f (x + n · vi ) = 0.
i=1

Let Ti be thr translation-by-vi operator on functions on the plane; that P

is, given a function
g on the plane, we have (Ti g)(x) = g(x+vi ). Another way of writing that ki=1 f (x+n·vi ) =
0, then, is to say that the function
X k
Tin f
i=1

is identically zero. Thus, f is in the kernel of each of the operators Pn = ki=1 Tin . (Recall
P
that the kernel of an operator is just another name for the things which are mapped to zero
by the opreator.)
Define S1 = T1 + · · · + Tk . Next S2 = T1 · T2 + T1 · T3 + · · · + Tk−1 · Tk , so that S2 is the
sum of the composition of pairs of distinct Ti s. (Note that the T − i’s all commute with one
anither, so this definition makes sense.) Continuing, for n ≤ k, let Sn be the operator which
is the sum of the composition of all n−tuples of distinct Ti s; inother words,

Sn = T1 · · · Tn−1 Tn + T1 · · · Tn−1 Tn+1 + · · · Tk−n+1 · · · Tk ,

and, in particular,
Sk = T1 · · · Tk .
 It is standard result that the Sn ’s may all be written as polynomials in terms of the Pn0 s.
For example, S1 = P1 , and 2S2 = P12 − P2 . To prove this result, for example, observe that
the result follows immediately by induction from the identity

(n + 1)Sn+1 = P1 Sn − P2 Sn−1 + P3 Sn−2 + · · · + (−1)n Pn+1 S0 .

(Here, S0 denotes the identity operator.) The identity follows by careful bookkeeping after
sbstituting inthe definitions, in terms of the Ti ’s, for the Pi Si ’s.
Since our function f is in the kernel of each Pn , it is in the kernel of any polynomial in
the Pn ’s, and specifically f is mapped to 0 by the Sn ’s. In particular, for any x,

0 = (Sk f )(x) = (T1 · · · Tk f )(x) = f (x + v1 + · · · + vk ).

it follows immediately that f is identically zero.