Maths Olympiad Prep

Library / /70 of 70

Combinatorics Difficulty 9.1 IMO level Prove it Romania

Given a positive integer nn, consider a triangular array with entries aija_{ij} where ii ranges from 11 to nn and jj ranges from 11 to ni+1n - i + 1. The entries of the array are all either 00 or 11, and, for all i>1i > 1 and any associated jj, aija_{ij} is 00 if ai1,j=ai1,j+1a_{i-1,j} = a_{i-1,j+1}, and aija_{ij} is 11 otherwise.
Let SS denote the set of binary sequences of length nn, and define a map f:SSf: S \to S via f:(a11,a12,,a1n)(an1,an1,2,,a1n)f: (a_{11}, a_{12}, \dots, a_{1n}) \mapsto (a_{n1}, a_{n-1,2}, \dots, a_{1n}). Determine the number of fixed points of ff.
Geoffrey Smith, United Kingdom, RMM 2013 Short List

Solutions — 2

Solution 1

(a11,a12,,a1n)(a11,a21,,an1)(a_{11}, a_{12}, \dots, a_{1n}) \mapsto (a_{11}, a_{21}, \dots, a_{n1})
is bijective, on one hand, and that (a11,a12,,a1n)(a_{11}, a_{12}, \dots, a_{1n}) is fixed by ff if and only if (a11,a21,,an1)(a_{11}, a_{21}, \dots, a_{n1}) is a palindrome (ai1=ani+1,1,i=1,,n)(a_{i1} = a_{n-i+1,1}, i = 1, \dots, n), on the other.
To begin with, notice that the definition of the aija_{ij} is equivalent to the Pascal-like relation in Z2\mathbb{Z}_2:
aij+ai1,j+ai1,j+1=0.(1) a_{ij} + a_{i-1,j} + a_{i-1,j+1} = 0. \quad (1)
Henceforth, such a triangular array will be called a Pascal binary (or dyadic) array.
Clearly, each binary string a in S yields a unique Pascal binary array a^\hat{\mathbf{a}}.
For more convenience, view a triangular array as a function on the standard lattice triangle
Δn={(i,j):i,j1 and 2i+jn+1}, \Delta_n = \{(i, j) : i, j \ge 1 \text{ and } 2 \le i + j \le n + 1\},
situated in the first quadrant; thus, the first index runs horizontally and corresponds to columns, and the second index runs vertically and corresponds to rows.
Further, use the symbols \to, \leftarrow, \uparrow, \downarrow, \searrow and \nwarrow to denote the oriented sides of Δn\Delta_n; explicitly,
={(i,1):i=1,,n},={(ni+1,1):i=1,,n},={(1,j):j=1,,n},={(1,nj+1):j=1,,n},={(i,ni+1):i=1,,n},={(nj+1,j):j=1,,n}. \begin{align*} \to &= \{(i, 1) : i = 1, \dots, n\}, & \leftarrow &= \{(n - i + 1, 1) : i = 1, \dots, n\}, \\ \uparrow &= \{(1, j) : j = 1, \dots, n\}, & \downarrow &= \{(1, n - j + 1) : j = 1, \dots, n\}, \\ \searrow &= \{(i, n - i + 1) : i = 1, \dots, n\}, & \nwarrow &= \{(n - j + 1, j) : j = 1, \dots, n\}. \end{align*}
To establish a bijection between the fixed points of ff and the binary palindromes of length nn, consider the transformations ϱ\varrho and σ\sigma of Δn\Delta_n defined by
ϱ(i,j)=(nij+2,i)andσ(i,j)=(nij+2,j). \varrho(i, j) = (n - i - j + 2, i) \quad \text{and} \quad \sigma(i, j) = (n - i - j + 2, j).
The former is a permutation of order 33 (ϱ3\varrho^3 is the identity), and the latter is an involution (σ2\sigma^2 is the identity). It is easily seen that
ϱ()=,ϱ()=,ϱ()=andσ()=,σ()=,σ()=, \varrho(\rightarrow) = \nwarrow, \varrho(\uparrow) = \leftarrow, \varrho(\nwarrow) = \downarrow \quad \text{and} \quad \sigma(\rightarrow) = \leftarrow, \sigma(\uparrow) = \nwarrow, \sigma(\nwarrow) = \uparrow,
so σϱ()=\sigma\varrho(\rightarrow) = \uparrow, σϱ()=\sigma\varrho(\uparrow) = \rightarrow, and σϱ()=\sigma\varrho(\nwarrow) = \nwarrow. It is also readily checked by (1) that, if a^=(aij)\hat{\mathbf{a}} = (a_{ij}) is a Pascal binary array, then so are both
ϱa^=(aϱ(i,j))andσa^=(aσ(i,j)). \varrho\hat{\mathbf{a}} = (a_{\varrho(i,j)}) \quad \text{and} \quad \sigma\hat{\mathbf{a}} = (a_{\sigma(i,j)}).
We are now in a position to prove the desired results.
Since σϱ\sigma\varrho exchanges \rightarrow and \uparrow, the assignment a^a^\hat{\mathbf{a}}_{\uparrow} \mapsto \hat{\mathbf{a}}_{\downarrow} is bijective.
Since σ\sigma exchanges \uparrow and \nwarrow, and reverses orientation on the bottom row of Δn\Delta_n, if a^\hat{\mathbf{a}}_{\uparrow} is fixed by ff, then a^=f(a^)=a^=(σa^)\hat{\mathbf{a}}_{\uparrow} = f(\hat{\mathbf{a}}_{\uparrow}) = \hat{\mathbf{a}}_{\nwarrow} = (\sigma\hat{\mathbf{a}})_{\uparrow}, so a^=σa^\hat{\mathbf{a}} = \sigma\hat{\mathbf{a}} and consequently a^=(σa^)=a^\hat{\mathbf{a}}_{\rightarrow} = (\sigma\hat{\mathbf{a}})_{\rightarrow} = \hat{\mathbf{a}}_{\leftarrow}; that is, the bottom row of a^\hat{\mathbf{a}} is a palindrome.
Conversely, if the bottom row of a^\hat{\mathbf{a}} is a palindrome, a^=a^\hat{\mathbf{a}}_{\rightarrow} = \hat{\mathbf{a}}_{\leftarrow}, then (σϱa^)=(ϱa^)(\sigma\varrho\hat{\mathbf{a}})_{\uparrow} = (\varrho\hat{\mathbf{a}})_{\uparrow}, so σϱa^=ϱa^\sigma\varrho\hat{\mathbf{a}} = \varrho\hat{\mathbf{a}} and consequently, f(a^)=a^=(σϱa^)=(ϱa^)=a^f(\hat{\mathbf{a}}_{\uparrow}) = \hat{\mathbf{a}}_{\nwarrow} = (\sigma\varrho\hat{\mathbf{a}})_{\downarrow} = (\varrho\hat{\mathbf{a}})_{\downarrow} = \hat{\mathbf{a}}_{\uparrow}; that is, a=a^\mathbf{a} = \hat{\mathbf{a}}_{\uparrow} is fixed by ff. This ends the proof.

The required number is 2(n+1)/22^{\lfloor(n+1)/2\rfloor}.

Solution 2

For convenience, we denote bk=a1,nkb_k = a_{1,n-k} and ck=ak+1,nkc_k = a_{k+1,n-k} for every k=0,1,,n1k = 0, 1, \dots, n-1. Our aim is to find the set of relations for (bk)(b_k) which are equivalent to the relation (bk)=(ck)(b_k) = (c_k). All the calculations will be made in Z2\mathbb{Z}_2.
The definition of aija_{ij} is equivalent to
aij=ai1,j+ai1,j+1. a_{ij} = a_{i-1,j} + a_{i-1,j+1}.
A straightforward check shows then that
aij==0i1(i1)a1,+j. a_{ij} = \sum_{\ell=0}^{i-1} \binom{i-1}{\ell} a_{1,\ell+j}.
For two nonnegative integers kk and \ell, we will write k\ell \preceq k if the binary representation of \ell can be obtained from that of kk by replacing some ones by zeroes (the leading zeroes are allowed; thus 0k0 \preceq k and kkk \preceq k for every kk). We write k\ell \prec k if <k\ell < k and k\ell \preceq k. Recall that by Lucas' theorem, (k)\binom{k}{\ell} is odd if and only if k\ell \preceq k. Thus,
ck=ak+1,nk==0k(k)a1,+nk==0k(k)bk==0k(k)b=kb. c_k = a_{k+1,n-k} = \sum_{\ell=0}^{k} \binom{k}{\ell} a_{1,\ell+n-k} = \sum_{\ell=0}^{k} \binom{k}{\ell} b_{k-\ell} = \sum_{\ell=0}^{k} \binom{k}{\ell} b_{\ell} = \sum_{\ell \preceq k} b_{\ell}.
Now, the conditions (bk)=(ck)(b_k) = (c_k) rewrite as the set of equations
0=kb()k 0 = \sum_{\ell \prec k} b_{\ell} \qquad (*)_{k}
for all k=0,1,,n1k = 0, 1, \dots, n-1.
Denote by TT the set of all strings (bk)(b_k) such that ()k(*)_k are satisfied for all odd kn1k \le n-1. Each string in this set is determined uniquely by the values of b2i1b_{2i-1} (2i<n2i < n) and bn1b_{n-1}: the values of b2ib_{2i} (for 2i<n12i < n-1) are found inductively from ()2i+1(*)_{2i+1}. Thus
T=2n/2=2(n+1)/2. |T| = 2^{\lceil n/2 \rceil} = 2^{\lfloor (n+1)/2 \rfloor}.
Now we claim that TT is exactly the desired set of fixed points; in fact, we will prove that all the relations ()d(*)_d for even dd follow from the relations ()k(*)_k for odd kk.
Consider any even dn1d \le n-1. To establish (d)(*_d), we add up all the relations (k)(*_k), where kdk \prec d, obtaining a sum
0=kdk+1b=kd(kb+kb+1)=db{k:kd}+db+1{k:kd}. \begin{aligned} 0 &= \sum_{k \prec d} \sum_{\ell \prec k+1} b_{\ell} = \sum_{k \prec d} \left( \sum_{\ell \preceq k} b_{\ell} + \sum_{\ell \prec k} b_{\ell+1} \right) \\ &= \sum_{\ell \prec d} b_{\ell} \cdot \left| \{k: \ell \preceq k \prec d\} \right| + \sum_{\ell \prec d} b_{\ell+1} \cdot \left| \{k: \ell \prec k \prec d\} \right|. \end{aligned}
But one can easily check that {k:kd}|\{k: \ell \preceq k \prec d\}| is odd, and {k:kd}|\{k: \ell \prec k \prec d\}| is even for all d\ell \prec d. Thus our equality rewrites exactly as (d)(*_d).

Therefore, the number of fixed points is 2(n+1)/22^{\lfloor(n+1)/2\rfloor}.

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.