Homework #2 Solutions:

1) Define the multicolor Ramsey number Rk (3) to be the minimum n such that no matter
how the edges of Kn are colored with k colors, there is a monochromatic copy of K3 . Prove
that
Rk (3) − 1 ≤ 1 + k(Rk−1 (3) − 1).

Use this to prove the upper bound Rk (3) < 1 + ek!, where e is the (usual) base of the
natural logarithm, i.e., e = 2.718.. .
Sol: Let N = 2+k(Rk−1 (3)−1), and suppose we have a k-coloring of E(Kn ). Pick a vertex
v. By the pigeonhole principle, there is a color i and a set S of size Rk−1 (3) such that
all v, S edges have color i. If color i appears anywhere within S, then we have a triangle
in color i. Otherwise the edges within S are (k − 1)-colored, and since |S| ≥ Rk−1 (3), we
conclude that S contains a monochromatic triangle inPsome other color.
k
We show by induction on k that Rk (3) ≤ 1 + k!( i=0 1/i!) < 1 + k!e. The base case
follows by R(3, 3) ≤ 6. For the induction step,
à k−1
! k
X X
Rk (3) ≤ 2 + k(Rk−1 (3) − 1) ≤ 2 + k (k − 1)! 1/i! = 1 + k! 1/i!.
i=0 i=0

2) Let k = 50, L = {0, 26, 27}, and F ⊆ 2[n] be an¡L-intersecting

¢ k-uniform family. Prove,
n
using the uniform RW Theorem, that m = |F| ≤ 2 .
Hint: For A, B ∈ F, let A ∼ B if A ∩ PB 6= ∅. Prove that
P this
¡ni ¢ is an
¡n¢equivalence relation.
You may also use the easy fact that if i ni = n, then i 2 ≤ 2 .
Sol: If A ∼ B and B ∼ C, then both |A ∩ B| and |B ∩ C| are at least 26 > |B|/2,
so |A ∩ C| > 0. This implies that A ∼ C. This relation divides the sets in F into
equivalence classes. Each equivalence class is a subfamily¡ F¢i that is L0 -intersecting, where
L0 = {26, 27}. By the uniform RW-inequality, |Fi | ≤ ¡n2i ¢, where ¡n¢ ni = | ∪A∈Fi A|. By
P ni
definition of ∼, (∪A∈Fi A) ∩ (∪B∈Fj B) = ∅, so |F| ≤ i 2 ≤ 2 .

3) Prove that in Nagy’s coloring given in class, if t ≡ 2 or 3 (mod 4), then there is no blue
Kr for r > t − 2.
¡ ¢
Recall that in the coloring, the vertex set of Kn is [t]
3 , and an edge is blue iff the endpoints
intersect in a set of size zero or two.
Sol: Let C1 , . . . , Cr be the vertex of a blue Kr . Put Ci ∼ Cj if |Ci ∩ Cj | = 2. Then it is
easy to see that this defines an equivalence relation, since no two of these sets intersect in
exactly one element. Each equivalence class is a subfamily Fi that is 2-intersecting. Let
A1 , . . . , Ami be the sets in Fi . We will show that either mi = 1, Aj ∩ Ak is the same set (of
size two) for all j, k (this is called a sunflower), or Fi consists of (at most four) 3-subsets
of a four element set.

1
 Suppose that mi > 1 and that Fi is not a sunflower. Let A, B, C ∈ Fi with |A∩B| = 2,
and C ∩A 6= A∩B. We may assume by symmetry that A = 123 and B = 234 and C = 124.
Then it is easy to see that the only choice for a fourth set D ∈ Fi is D = 134.
Let mi = |Fi |, and Xi = ∪A∈Fi A with ti = |Xi |. By definition, Xi ∩ Xj = ∅ for
i 6= j. Clearly mi ≤ ti − 2 if Fi is a sunflower, and mi ≤ ti otherwise. By the choice of
t, there
P are at Pleast two points that are in an Xi for which Fi is a sunflower. This gives
m= mi ≤ ( ti ) − 2 = t − 2.

4) We gave superpolynomial lower bounds in class for the Ramsey number R(t, t) for
infinitely many t. Prove the same lower bound for all t, namely, for any fixed ² > 0, there
is a t0 such that for t > t0 we have R(t, t) > t(1−²)ω(t) , where ω(t) = ln t/(4 ln ln t).
¡ n ¢
Hint: As in class, let n = p3 . Now let p be the largest prime such that 2 p−1 < t. You
may use the following consequence of the Prime Number Theorem: for any δ > 0, there is
a q0 such that, if q > q0 is a prime, then the next largest prime q 0 > q has the property
that q 0 < (1 + δ)q. Use this to prove that for any δ 0 > 0,

(1 − δ 0 ) ln t (1 + δ 0 ) ln t
<p<
2 ln ln t 2 ln ln t
for sufficiently large t. Then use the estimates for binomial coefficients we have proved to
complete the proof.
Sol: Let q be the next largest prime after p. Then
µ 3 ¶ µ 3 ¶
p q
2 <t<2 .
p−1 q−1

Using standard estimates for binomial coefficients (see the similar calculations given in
class), and the Prime Number Theorem, this yields

(2 − o(1))p log p < log t < (2 + o(1))q log q < (2 + o(1))p log p.

This implies that log t ∼ 2p log p and also that log log t ∼ log p. This yields

p = (1 + o(1))(log t)/(2 log p) = (1 + o(1))(log t)/(2 log log t).

¡ 3 ¢
To prove the lower bound on R(t, t), we construct a graph on n = p2p−1 vertices,
¡ p3 ¢
with no clique or independent set of size 2 p−1 < t. This was done in class. We only need
(1−²)ω(t)
to show that n > t . Lower bounding n and taking logs, this amounts to showing
2 2
that p log p > (1 − ²) log t/(4 log log t). The bounds for p in terms of t derived above
imply precisely this.

5) Let K = {k1 , k2 } and L = {l1 , . . . , ls } be two sets of nonnegative integers with ki > s−2
for i = 1, 2. Let F ⊆ 2[n] be an L-intersecting family with |S| ∈ K for each S ∈ F . Prove
that µ ¶ µ ¶
n n
m = |F| ≤ + .
s s−1

2
Hint: ProceedP as in the proof of the uniform
P RW Theorem
P presented in class. Instead of
the function ( i xi − k), use the function ( i xi − k1 )( i xi − k2 ), and instead of letting
|I| ≤ s − 1, let |I| ≤ s − 2.
Remark: This can be easily generalized to K = {k1 , . . . , kr } (no need to do it), and then
it provides a common proof of both the uniform and nonuniform RW Theorems (Alon-
Babai-Suzuki 1991).
Sol: We prove the more general version in the Remark. Recall the following Lemma proved
in class
Lemma: Let f : Ω → R. Assume that f (I) 6= 0 for any |I| ≤ r. Then the set of functions
{xI f : |I| ≤ r} is linearly independent.
Proof: Order the subsets of [n] such that I < J implies that |I| ≤ |J|. Then for I, J ⊆ [n]
with |I|, |J| ≤ r we have xI (J)f (J) 6= 0 if I = J and = 0 if J < I. By the triangular
criterion proved in class, we conclude that these functions are linearly independent.
Let F = {A1 , . . . , Am }, where |Ai | ≤ |Ai+1 | for all i, and define the polynomials
Y
fi (x) = (vi · x − lk ), (x ∈ Rn ),
k: lk <|Ai |

where vi is the incidence vector of Ai . Set

 
Yr n
X
f=  xj − ki  .
i=1 j=1

Then the Lemma implies that the set of functions {xI f : |I| ≤ r} is linearly independent.
We next show that this set of functions together with the set {fi : 1 ≤ i ≤ m} is linearly
independent. To prove this, suppose that
m
X X
λi f i + µI xI f = 0.
i=1 |I|≤s−r

We first argue that each λi = 0. If not, suppose that i0 is the smallest i for which λi0 6= 0.
Substituting Ai0 above yields the contradiction λi0 = 0. Now it follows the all the µI are
zero by the Lemma. Ps−r ¡ ¢
We may assume that all these function are multilinear. Thus we have m + i=0 ni
linearly independent functions,Peach of which
¡n¢ can be represented by polynomials of degree
s
at most s. Consequently m ≤ i=s−r+1 i .