Maths Olympiad Prep

Library / /327 of 397

Algebra Difficulty 6.8 National Olympiad Prove it Taiwan

Let a0,a1,a2,a_0, a_1, a_2, \dots be a sequence of integers and b0,b1,b2,b_0, b_1, b_2, \dots be a sequence of positive integers such that a0=0,a1=1a_0 = 0, a_1 = 1, and for all n=1,2,n = 1, 2, \dots, we have
an+1={anbn+an1if bn1=1;anbnan1if bn1>1. a_{n+1} = \begin{cases} a_n b_n + a_{n-1} & \text{if } b_{n-1} = 1; \\ a_n b_n - a_{n-1} & \text{if } b_{n-1} > 1. \end{cases}
Prove that at least one of the two numbers a2017a_{2017} and a2018a_{2018} is greater than or equal to 2017.

Solutions — 2

Solution 1

The value of b0b_0 is irrelevant since a0=0a_0 = 0, so we may assume that b0=1b_0 = 1.

Lemma. We have an1a_n \ge 1 for all n1n \ge 1.

Proof. Let us suppose otherwise in order to obtain a contradiction. Let
n1 be the smallest integer with an0.(1) n \ge 1 \text{ be the smallest integer with } a_n \le 0. \quad (1)
Note that n2n \ge 2. It follows that an11a_{n-1} \ge 1 and an20a_{n-2} \ge 0. Thus we cannot have
an=an1bn1+an2, a_n = a_{n-1}b_{n-1} + a_{n-2},
so we must have
an=an1bn1an2. a_n = a_{n-1}b_{n-1} - a_{n-2}.
Since an0a_n \le 0, we have an1an2a_{n-1} \le a_{n-2}. Thus we have
an2an1an. a_{n-2} \ge a_{n-1} \ge a_n.
Let
r be the smallest index with arar+1ar+2.(2) r \text{ be the smallest index with } a_r \ge a_{r+1} \ge a_{r+2}. \quad (2)
Then rn2r \le n - 2 by the above, but also r2r \ge 2: if b1=1b_1 = 1, then a2=a1=1a_2 = a_1 = 1 and a3=a2b2+a1>a2a_3 = a_2b_2 + a_1 > a_2; if b1>1b_1 > 1, then a2=b1>1=a1a_2 = b_1 > 1 = a_1.
By the minimal choice Eq. (2) of rr, it follows that ar1<ara_{r-1} < a_r. And since 2rn22 \le r \le n-2, by the minimal choice Eq. (1) of nn we have
ar1,ar,ar+1>0. a_{r-1}, a_r, a_{r+1} > 0.
In order to have ar+1ar+2a_{r+1} \ge a_{r+2}, we must have
ar+2=ar+1br+1ar a_{r+2} = a_{r+1}b_{r+1} - a_r
so that br2b_r \ge 2. Putting everything together, we conclude that
ar+1=arbr±ar12arar1=ar+(arar1)>ar, a_{r+1} = a_r b_r \pm a_{r-1} \ge 2a_r - a_{r-1} = a_r + (a_r - a_{r-1}) > a_r,
which contradicts Eq. (2).

To complete the problem, we prove that max{an,an+1}n\max\{a_n, a_{n+1}\} \ge n by induction. The case n=0,1n = 0, 1 are given. Assume it is true for all non-negative integers strictly less than nn, where n2n \ge 2. There are two cases:

Case 1: bn1=1b_{n-1} = 1.

Then an+1=anbn+an1a_{n+1} = a_n b_n + a_{n-1}. By the inductive assumption one of an1,ana_{n-1}, a_n is at least n1n-1 and the other, by lemma, is at least 1. Hence
an+1=anbn+an1an+an1(n1)+1=n. a_{n+1} = a_n b_n + a_{n-1} \ge a_n + a_{n-1} \ge (n-1) + 1 = n.
Thus max{an,an+1}n\max\{a_n, a_{n+1}\} \ge n, as desired.

Case 2: bn1>1b_{n-1} > 1.

Since we defined b0=1b_0 = 1 there is an index rr with 1rn11 \le r \le n-1 such that
bn1,bn2,,br2 and br1=1. b_{n-1}, b_{n-2}, \dots, b_r \ge 2 \text{ and } b_{r-1} = 1.
We have ar+1=arbr+ar12ar+ar1a_{r+1} = a_r b_r + a_{r-1} \ge 2a_r + a_{r-1}. Thus
ar+1arar+ar1. a_{r+1} - a_r \ge a_r + a_{r-1}.
Now we claim that ar+ar1ra_r + a_{r-1} \ge r. Indeed, this holds by inspection for r=1r = 1; for r2r \ge 2, one of ar,ar1a_r, a_{r-1} is at least r1r-1 by the inductive assumption, while the other, by the lemma, is at least 1. Hence ar+ar1ra_r+a_{r-1} \ge r, as claimed, and therefore ar+1arra_{r+1} - a_r \ge r by the last inequality in the previous paragraph.
Since r1r \ge 1 and, by the lemma, ar1a_r \ge 1, from ar+1arra_{r+1} - a_r \ge r we get the following two inequalities:
ar+1r+1 and ar+1>ar. a_{r+1} \ge r+1 \text{ and } a_{r+1} > a_r.
Now observe that
am>am1am+1>am for m=r+1,r+2,,n1, a_m > a_{m-1} \Rightarrow a_{m+1} > a_m \text{ for } m = r+1, r+2, \dots, n-1,
since
am+1=ambmam12amam1=am+(amam1)>am. a_{m+1} = a_m b_m - a_{m-1} \ge 2a_m - a_{m-1} = a_m + (a_m - a_{m-1}) > a_m.
Thus
an>an1>>ar+1r+1ann. a_n > a_{n-1} > \dots > a_{r+1} \ge r+1 \Rightarrow a_n \ge n.
So max{an,an+1}n\max\{a_n, a_{n+1}\} \ge n, as derived.

Solution 2

We say that an index n>1n > 1 is bad if bn1=1b_{n-1} = 1 and bn2>1b_{n-2} > 1; otherwise nn is good. The value of b0b_0 is irrelevant to definition of (an)(a_n) since a0=0a_0 = 0; so we assume that b0>1b_0 > 1.

Lemma 1. (a) an1a_n \ge 1 for all n>0n > 0.
(b) If n>1n > 1 is good, then an>an1a_n > a_{n-1}.

Proof. Induction on nn. In the base cases n=1,2n = 1, 2 we have
a1=11,a2=b1a11, a_1 = 1 \ge 1, a_2 = b_1 a_1 \ge 1,
and finally a2>a1a_2 > a_1 if 2 is good, since in this case b1>1b_1 > 1.
Now we assume that the lemma statement is proved for n=1,2,,kn = 1, 2, \dots, k with k2k \ge 2, and prove it for n=k+1n = k + 1. Recall that aka_k and ak1a_{k-1} are positive by the induction hypothesis.

**Case 1: kk is bad.**
We have bk1=1b_{k-1} = 1, so
ak+1=bkak+ak1ak+ak1>ak1, a_{k+1} = b_k a_k + a_{k-1} \ge a_k + a_{k-1} > a_k \ge 1,
as required.

**Case 2: kk is good.**
We already have ak>ak11a_k > a_{k-1} \ge 1 by induction hypothesis. We consider three easy subcases.

**Subcase 2.1: bk>1b_k > 1.**
Then
ak+1bkakak1ak+(akak1)>ak1. a_{k+1} \ge b_k a_k - a_{k-1} \ge a_k + (a_k - a_{k-1}) > a_k \ge 1.
**Subcase 2.2: bk=bk1=1b_k = b_{k-1} = 1.**
Then
ak+1=ak+(akak1)>ak1. a_{k+1} = a_k + (a_k - a_{k-1}) > a_k \ge 1.
**Subcase 2.3: bk=1b_k = 1 but bk1>1b_{k-1} > 1.**
Then k+1k + 1 is bad, and we need to prove only (a), which is trivial:
ak+1=akak11. a_{k+1} = a_k - a_{k-1} \ge 1.
So, in all three subcases we have verified the required relations.

Lemma 2. Assume that n>1n > 1 is bad. Then there exists a j{1,2,3}j \in \{1, 2, 3\} such that
an+jan1+j+1, and an+ian1+i for all 1i<j. a_{n+j} \geq a_{n-1} + j + 1, \text{ and } a_{n+i} \geq a_{n-1} + i \text{ for all } 1 \leq i < j.
Proof. Recall that bn1=1b_{n-1} = 1. Set
m=inf{i>0:bn+i1>1} m = \inf\{i > 0 : b_{n+i-1} > 1\}
(possibly m=+m = +\infty). We claim that j=min{m,3}j = \min\{m, 3\} works. Again, we distinguish several cases, according to the value of mm; in each of them we use Lemma 1 without reference.

Case 1: m=1m = 1, so bn>1b_n > 1.
Then an+12an+an1an1+2a_{n+1} \geq 2a_n + a_{n-1} \geq a_{n-1} + 2, as required.

Case 2: m=2m = 2, so bn=1b_n = 1 and bn+1>1b_{n+1} > 1.
Then we successively get
an+1=an+an1an1+1,an+22an+1+an2(an1+1)+an=an1+(an1+an+2)an1+4, \begin{align*} a_{n+1} &= a_n + a_{n-1} \geq a_{n-1} + 1, \\ a_{n+2} &\geq 2a_{n+1} + a_n \geq 2(a_{n-1} + 1) + a_n \\ &= a_{n-1} + (a_{n-1} + a_n + 2) \geq a_{n-1} + 4, \end{align*}
which is even better than we need.

Case 3: m>2m > 2, so bn=bn+1=1b_n = b_{n+1} = 1.
Then we successively get
an+1=an+an1an1+1,an+2=an+1+anan1+1+anan1+2,an+3an+2+an+1(an1+1)+(an1+2)an1+4, \begin{align*} a_{n+1} &= a_n + a_{n-1} \geq a_{n-1} + 1, & a_{n+2} &= a_{n+1} + a_n \\ &\geq a_{n-1} + 1 + a_n \geq a_{n-1} + 2, \\ a_{n+3} &\geq a_{n+2} + a_{n+1} \geq (a_{n-1} + 1) + (a_{n-1} + 2) \geq a_{n-1} + 4, \end{align*}
as required.

Lemmas 1(b) and 2 provide enough information to prove that
max{an,an+1}n for all n \max\{a_n, a_{n+1}\} \geq n \text{ for all } n
and, moreover, that anna_n \geq n often enough. Indeed, assume that we have found some nn with an1n1a_{n-1} \geq n - 1. If nn is good, then Lemma 1(b) we have anna_n \geq n as well. If nn is bad, then Lemma 2 yields
max{an+i,an+i+1}an1+i+1n+i for all 0i<j \max\{a_{n+i}, a_{n+i+1}\} \geq a_{n-1} + i + 1 \geq n + i \text{ for all } 0 \leq i < j
and
an+jan1+j+1n+j; a_{n+j} \geq a_{n-1} + j + 1 \geq n + j;
so n+jn + j is the next index to start with.

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 translated into English from zh; metadata (topic, difficulty) added by this project.