Maths Olympiad Prep

Library / /348 of 377

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 s1uss^{-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 vτ\sqrt{v} \geq \tau.

Solution

Solution:

Call an element tSt \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 1im1 \leq i \leq m, let lil_{i} denote the length of the iith cycle, so that ss may be written as
(a1,1a1,2a1,l1)(a2,1a2,2a2,l2)(am,1am,2am,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+1jq\left(a_{i, j}\right)=a_{i, l_{i}+1-j} for all 1jli1 \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+2kr\left(a_{i, k}\right)=a_{i, l_{i}+2-k} for all 2kli2 \leq k \leq l_{i}, for all 1im1 \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,rSq, 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 τλfλn!\tau \leq \sum_{\lambda} \frac{f^{\lambda}}{n!}, where both sums range over all partitions λ\lambda of nn.

For all elements uSu \in S, define the conjugacy class of uu to be the set of elements that may be written in the form s1uss^{-1} u s for some sSs \in S. It is easy to see that for all u,uSu, u' \in S, the conjugacy classes of uu and uu' 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
(a1,1a1,2a1,l1)(a2,1a2,2a2,l2)(am,1am,2am,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 w1w2wmw_{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, s1uss^{-1} u s may be written in the form
s1(a1,1a1,2a1,l1)ss1(a2,1a2,2a2,l2)ss1(am,1am,2am,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
(b1,1b1,2b1,l1)(b2,1b2,2b2,l2)(bm,1bm,2bm,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 s1s^{-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 s1uss^{-1} u s for all sSs \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 vλ1n!v \geq \sum_{\lambda} \frac{1}{n!} by our above claim, for then
n!v=n!λ1=(λ(fλ)2)(λ1)λfλn!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 2,000 problems in a working order, from AMC 10 level to the IMO shortlist.

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