Maths Olympiad Prep

Library / Harvard-MIT Mathematics Tournament /325 of 352

Harvard-MIT Mathematics Tournament

Algebra Difficulty 5.8 AIME, harder Prove it United States

Problem:

Denote {1,2,…,n}\{1,2, \ldots, n\} by [n][n], and let SS be the set of all permutations of [n][n]. Call a subset TT of SS good if every permutation σ\sigma in SS may be written as t1t2t_{1} t_{2} for elements t1t_{1} and t2t_{2} of TT, where the product of two permutations is defined to be their composition. Call a subset UU of SS extremely good if every permutation σ\sigma in SS may be written as s−1uss^{-1} u s for elements ss of SS and uu of UU. Let τ\tau be the smallest value of ∣T∣/∣S∣|T| /|S| for all good subsets TT, and let vv be the smallest value of ∣U∣/∣S∣|U| /|S| for all extremely good subsets UU. Prove that v≥τ\sqrt{v} \geq \tau.

Solution

Solution:

Call an element t∈St \in S an involution if and only if t2t^{2} is the identity permutation. We claim that the set of all involutions in SS constitutes a good subset of SS. The proof is simple. Let ss be an arbitrary permutation in SS. Note that ss may be decomposed into a product of disjoint cycles of length at least 22, and suppose that there are mm such cycles in its decomposition. For 1≤i≤m1 \leq i \leq m, let lil_{i} denote the length of the iith cycle, so that ss may be written as
(a1,1a1,2…a1,l1)(a2,1a2,2…a2,l2)…(am,1am,2…am,lm) \left(a_{1,1} a_{1,2} \ldots a_{1, l_{1}}\right)\left(a_{2,1} a_{2,2} \ldots a_{2, l_{2}}\right) \ldots\left(a_{m, 1} a_{m, 2} \ldots a_{m, l_{m}}\right)
for some pairwise distinct elements
a1,1,a1,2,…,a1,l1,a2,1,a2,2,…,a2,l2,…,am,1,am,2,…,am,lm a_{1,1}, a_{1,2}, \ldots, a_{1, l_{1}}, a_{2,1}, a_{2,2}, \ldots, a_{2, l_{2}}, \ldots, a_{m, 1}, a_{m, 2}, \ldots, a_{m, l_{m}}
of [n][n]. Consider the permutations qq and rr defined by q(ai,j)=ai,li+1−jq\left(a_{i, j}\right)=a_{i, l_{i}+1-j} for all 1≤j≤li1 \leq j \leq l_{i} and r(ai,1)=ai,1r\left(a_{i, 1}\right)=a_{i, 1} and r(ai,k)=ai,li+2−kr\left(a_{i, k}\right)=a_{i, l_{i}+2-k} for all 2≤k≤li2 \leq k \leq l_{i}, for all 1≤i≤m1 \leq i \leq m, and by q(x)=r(x)=xq(x)=r(x)=x for all x∈[n]x \in[n] otherwise. Since q,r∈Sq, r \in S, q2=r2=1q^{2}=r^{2}=1, and rq=sr q=s, it follows that the set of all involutions in SS is indeed good, as desired.

For all integer partitions λ\lambda of nn, let fλf^{\lambda} denote the number of standard Young tableaux of shape λ\lambda. By the Robinson-Schensted-Knuth correspondence, the permutations of [n][n] are in bijection with pairs of standard Young tableaux of the same shape in such a way that the involutions of [n][n] are in bijection with pairs of identical standard Young tableaux. In other words, the number of permutations of [n][n] is equal to the number of pairs of identically shaped standard Young tableaux whose shape is a partition of nn, and the number of involutions of [n][n] is equal to the number of standard Young tableaux whose shape is a partition of nn. Hence n!=∑λ(fλ)2n!=\sum_{\lambda}\left(f^{\lambda}\right)^{2} and τ≤∑λfλn!\tau \leq \sum_{\lambda} \frac{f^{\lambda}}{n!}, where both sums range over all partitions λ\lambda of nn.

For all elements u∈Su \in S, define the conjugacy class of uu to be the set of elements that may be written in the form s−1uss^{-1} u s for some s∈Ss \in S. It is easy to see that for all u,u′∈Su, u' \in S, the conjugacy classes of uu and u′u' are either identical or disjoint. It follows that SS may be partitioned into disjoint conjugacy classes and that any extremely good subset of SS must contain at least one element from each distinct conjugacy class. We claim that the number of distinct conjugacy classes of SS is at least the number of integer partitions of nn. It turns out that the two numbers are in fact equal, but such a result is not necessary for the purposes of this problem. Let uu be an arbitrary permutation in SS. Recall that uu may be written as
(a1,1a1,2…a1,l1)(a2,1a2,2…a2,l2)…(am,1am,2…am,lm) \left(a_{1,1} a_{1,2} \ldots a_{1, l_{1}}\right)\left(a_{2,1} a_{2,2} \ldots a_{2, l_{2}}\right) \ldots\left(a_{m, 1} a_{m, 2} \ldots a_{m, l_{m}}\right)
for some pairwise distinct elements
a1,1,a1,2,…,a1,l1,a2,1,a2,2,…,a2,l2,…,am,1,am,2,…,am,lm a_{1,1}, a_{1,2}, \ldots, a_{1, l_{1}}, a_{2,1}, a_{2,2}, \ldots, a_{2, l_{2}}, \ldots, a_{m, 1}, a_{m, 2}, \ldots, a_{m, l_{m}}
of [n][n]. Associate to the conjugacy class of uu the partition n=lw1+lw2+…+lwm+1+…+1n=l_{w_{1}}+l_{w_{2}}+\ldots+l_{w_{m}}+1+\ldots+1, where w1,w2,…,wmw_{1}, w_{2}, \ldots, w_{m} is a partition of 1,2,…,m1,2, \ldots, m such that w1≥w2≥…≥wmw_{1} \geq w_{2} \geq \ldots \geq w_{m} and the 11's represent all the fixed points of ss. Now note that if ss is any permutation in SS, s−1uss^{-1} u s may be written in the form
s−1(a1,1a1,2…a1,l1)ss−1(a2,1a2,2…a2,l2)s…s−1(am,1am,2…am,lm)s s^{-1}\left(a_{1,1} a_{1,2} \ldots a_{1, l_{1}}\right) s s^{-1}\left(a_{2,1} a_{2,2} \ldots a_{2, l_{2}}\right) s \ldots s^{-1}\left(a_{m, 1} a_{m, 2} \ldots a_{m, l_{m}}\right) s
which may be written as
(b1,1b1,2…b1,l1)(b2,1b2,2…b2,l2)…(bm,1bm,2…bm,lm) \left(b_{1,1} b_{1,2} \ldots b_{1, l_{1}}\right)\left(b_{2,1} b_{2,2} \ldots b_{2, l_{2}}\right) \ldots\left(b_{m, 1} b_{m, 2} \ldots b_{m, l_{m}}\right)
for some pairwise distinct elements
b1,1,b1,2,…,b1,l1,b2,1,b2,2,…,b2,l2,…,bm,1,bm,2,…,bm,lm b_{1,1}, b_{1,2}, \ldots, b_{1, l_{1}}, b_{2,1}, b_{2,2}, \ldots, b_{2, l_{2}}, \ldots, b_{m, 1}, b_{m, 2}, \ldots, b_{m, l_{m}}
of [n][n] because multiplying by s−1s^{-1} on the left and by ss on the right is equivalent to re-indexing the letters 1,2,…,n1,2, \ldots, n. Hence if partitions of nn are associated to all conjugacy classes of SS analogously, the same partition that is associated to uu is associated to s−1uss^{-1} u s for all s∈Ss \in S. Since exactly one partition is associated to each conjugacy class, it follows that the number of conjugacy classes cannot exceed the number of partitions, as desired.

To conclude, we need only observe that v≥∑λ1n!v \geq \sum_{\lambda} \frac{1}{n!} by our above claim, for then
n!v=n!∑λ1=(∑λ(fλ)2)(∑λ1)≥∑λfλ≥n!τ n!\sqrt{v}=\sqrt{n!\sum_{\lambda} 1}=\sqrt{\left(\sum_{\lambda}\left(f^{\lambda}\right)^{2}\right)\left(\sum_{\lambda} 1\right)} \geq \sum_{\lambda} f^{\lambda} \geq n!\tau
by the Cauchy-Schwarz inequality, and this completes the proof.

Want a route through all this instead of an archive? The track puts 2,604 problems in a working order, from Junior Challenge level to the IMO shortlist.

Source: MathNet, licensed CC-BY-4.0. Statement reproduced verbatim; metadata (topic, difficulty) added by this project.