2. A closer look at induction
2.9. Two unexpected integralities
2.9.1. The first integrality
We shall illustrate strong induction on two further examples, which both have the form of an “unexpected integrality”: A sequence of rational numbers is defined recursively by an equation that involves fractions, but it turns out that all the en-tries of the sequence are nevertheless integers. There is by now a whole genre of such results (see [Gale98, Chapter 1] for an introduction67), and many of them are connected with recent research in the theory of cluster algebras (see [Lampe13] for an introduction).
The first of these examples is the following result:
Proposition 2.64. Define a sequence (t0,t1,t2, . . .) of positive rational numbers recursively by setting
t0=1, t1 =1, t2=1, and tn = 1+tn−1tn−2
tn−3 for eachn ≥3.
67See also [FomZel02] for a seminal research paper at a more advanced level.
(Thus,
t3 = 1+t2t1 t0
= 1+1·1 1 =2;
t4 = 1+t3t2 t1
= 1+2·1 1 =3;
t5 = 1+t4t3
t2
= 1+3·2 1 =7;
t6 = 1+t5t4
t3
= 1+7·3 2 =11, and so on.) Then:
(a)We have tn+2 =4tn−tn−2for each n∈ Z≥2. (b)We have tn ∈Nfor eachn ∈N.
Note that the sequence (t0,t1,t2, . . .) in Proposition 2.64 is clearly well-defined, because the expression 1+tn−1tn−2
tn−3
always yields a well-defined positive ratio-nal number when tn−1,tn−2,tn−3 are positive rational numbers. (In particular, the denominator tn−3 of this fraction is 6= 0 because it is positive.) In contrast, if we had set tn = 1−tn−1tn−2
tn−3 instead of tn = 1+tn−1tn−2
tn−3 , then the sequence would not be well-defined (because then, we would get t3 = 1−1·1
1 = 0 and
t6 = 1−t5t4
t3
= 1−t5t4
0 , which is undefined).
Remark 2.65. The sequence (t0,t1,t2, . . .) defined in Proposition 2.64 is the se-quence A005246 in the OEIS (Online Encyclopedia of Integer Sese-quences). Its first entries are
t0 =1, t1 =1, t2 =1, t3=2, t4=3, t5 =7, t6 =11, t7 =26, t8=41, t9=97.
Proposition 2.64 (b) is an instance of the Laurent phenomenon (see, e.g., [FomZel02, Example 3.2]).
Part(a)of Proposition 2.64 is proven by a (regular) induction; it is part(b)where strong induction comes handy:
Proof of Proposition 2.64. First, we notice that the recursive definition of the sequence (t0,t1,t2, . . .)yields
t3 = 1+t3−1t3−2
t3−3
= 1+t2t1
t0 = 1+1·1
1 (sincet0=1 andt1 =1 andt2=1)
=2.
Furthermore, the recursive definition of the sequence(t0,t1,t2, . . .)yields
(a)We shall prove Proposition 2.64(a)by induction onn starting at 2:
Induction base: We have already shown that t2+2 = 4t2−t2−2. In other words, Proposition 2.64(a)holds for n=2. This completes the induction base.
Induction step: Let m ∈ Z≥2. Assume that Proposition 2.64 (a)holds for n = m.
We must prove that Proposition 2.64(a)holds forn =m+1.
We have assumed that Proposition 2.64 (a)holds for n =m. In other words, we havetm+2 =4tm−tm−2. completes the induction step. Hence, Proposition 2.64(a)is proven by induction.
(b)We shall prove Proposition 2.64(b)by strong induction on nstarting at 0:
Induction step: Let m ∈ N. 68 Assume that Proposition 2.64 (b)holds for every n ∈ N satisfying n < m. We must now show that Proposition 2.64 (b) holds for n=m.
We have assumed that Proposition 2.64 (b) holds for every n ∈ N satisfying n<m. In other words, we have
tn ∈Nfor everyn ∈Nsatisfyingn <m. (99) We must now show that Proposition 2.64 (b) holds for n = m. In other words, we must show thattm ∈ N.
Recall that(t0,t1,t2, . . .) is a sequence of positive rational numbers. Thus,tm is a positive rational number.
We are in one of the following five cases:
Case 1: We havem=0.
Case 2: We havem=1.
Case 3: We havem=2.
Case 4: We havem=3.
Case 5: We havem>3.
Let us first consider Case 1. In this case, we havem =0. Thus, tm =t0 =1∈ N.
Hence,tm ∈ Nis proven in Case 1.
Similarly, we can prove tm ∈ N in Case 2 (using t1 = 1) and in Case 3 (using t2 =1) and in Case 4 (usingt3 =2). It thus remains to provetm ∈Nin Case 5.
So let us consider Case 5. In this case, we havem >3. Thus,m ≥4 (sincemis an integer), so that m−2 ≥ 4−2 =2. Thus, m−2 is an integer that is ≥ 2. In other words, m−2 ∈ Z≥2. Hence, Proposition 2.64 (a) (applied to n = m−2) yields t(m−2)+2 =4tm−2−t(m−2)−2. In view of(m−2) +2=m and (m−2)−2=m−4, this rewrites astm =4tm−2−tm−4.
Butm ≥4, so thatm−4∈ N, andm−4<m. Hence, (99) (applied to n=m−4) yieldstm−4 ∈N⊆Z. Similarly,tm−2 ∈Z.
So we know thattm−2and tm−4are both integers (sincetm−2∈ Zandtm−4 ∈ Z).
Hence, 4tm−2−tm−4 is an integer as well. In other words,tm is an integer (because tm = 4tm−2−tm−4). Since tm is positive, we thus conclude that tm is a positive integer. Hence, tm ∈N. This shows thattm ∈Nin Case 5.
We now have proventm ∈ Nin each of the five Cases 1, 2, 3, 4 and 5. Since these five Cases cover all possibilities, we thus conclude that tm ∈ N always holds. In other words, Proposition 2.64 (b) holds for n = m. This completes the induction step. Thus, Proposition 2.64(b)is proven by strong induction.
2.9.2. The second integrality
Our next example of an “unexpected integrality” is the following fact:
68In order to match the notations used in Theorem 2.60, we should be saying “Letm∈Z≥0” here, rather than “Letm∈N”. But of course, this amounts to the same thing, sinceN=Z≥0.
Proposition 2.66. Fix a positive integer r. Define a sequence (b0,b1,b2, . . .) of positive rational numbers recursively by setting
b0=1, b1 =1, and bn = b
rn−1+1 bn−2
for eachn ≥2.
(Thus,
b2= b
1r +1 b0
= 1
r+1 1 =2;
b3= b
2r +1 b1
= 2
r+1
1 =2r+1;
b4= b
r 3+1
b2
= (2r+1)r+1
2 ,
and so on.) Then:
(a)We have bn ∈Nfor eachn ∈ N.
(b)If r≥2, thenbn | bn−2+bn+2 for eachn ∈Z≥2.
Remark 2.67. If r = 1, then the sequence (b0,b1,b2, . . .) defined in Proposition 2.66 is
(1, 1, 2, 3, 2, 1, 1, 2, 3, 2, 1, 1, 2, 3, 2, . . .)
(this is a periodic sequence, which consists of the five terms 1, 1, 2, 3, 2 repeated over and over); this can easily be proven by strong induction. Despite its sim-plicity, this sequence is the sequence A076839 in the OEIS.
Ifr =2, then the sequence(b0,b1,b2, . . .) defined in Proposition 2.66 is (1, f1, f3, f5, f7, . . .) = (1, 1, 2, 5, 13, 34, 89, 233, 610, 1597, . . .)
consisting of all Fibonacci numbers at odd positions (i.e., Fibonacci numbers of the form f2n−1 for n ∈ N) with an extra 1 at the front. This, again, can be proven by induction. Also, this sequence satisfies the recurrence relation bn =3bn−1−bn−2for all n ≥2. This is the sequence A001519 in the OEIS.
Ifr =3, then the sequence(b0,b1,b2, . . .) defined in Proposition 2.66 is (1, 1, 2, 9, 365, 5403014, 432130991537958813, . . .);
its entries grow so fast that the next entry would need a separate line. This is the sequence A003818 in the OEIS. Unlike the cases of r = 1 and r = 2, not much can be said about this sequence, other than what has been said in Proposition 2.66.
Proposition 2.66(a)is an instance of the Laurent phenomenon for cluster algebras (see, e.g., [FomZel01, Example 2.5]; also, see [Marsh13] and [FoWiZe16] for ex-positions). See also [MusPro07] for a study of the specific recurrence equation from Proposition 2.66 (actually, a slightly more general equation).
Before we prove Proposition 2.66, let us state an auxiliary fact:
Proof of Proposition 2.66. First, we notice that the recursive definition of the sequence (b0,b1,b2, . . .) yields
Let us first prove the following observation:
Observation 1: Each integern ≥2 satisfiesbn+2 =bn−2brn+1−brn−1H(bnr).
[Proof of Observation 1: Let n ≥ 2 be an integer. Thus, n ≥ 2 ≥ 1. Thus, (100) rational numbers). Thus, brn is also a positive rational number. Hence, brn ∈ Q is nonzero. The definition of H(bnr) yieldsH(bnr) = (brn+1)r−1
(a)We shall prove Proposition 2.66(a)by strong induction on n:
Induction step: Let m ∈ N. Assume that Proposition 2.66 (a) holds for every n ∈ N satisfying n < m. We must now prove that Proposition 2.66 (a) holds for n=m.
We have assumed that Proposition 2.66 (a) holds for every n ∈ N satisfying n<m. In other words, we have
bn ∈Nfor everyn ∈Nsatisfyingn <m. (104) We must now show that Proposition 2.66 (a) holds for n = m. In other words, we must show thatbm ∈ N.
Recall that(b0,b1,b2, . . .) is a sequence of positive rational numbers. Thus, bm is a positive rational number.
We are in one of the following five cases:
Case 1: We havem=0.
Case 2: We havem=1.
Case 3: We havem=2.
Case 4: We havem=3.
Case 5: We havem>3.
Let us first consider Case 1. In this case, we havem=0. Thus,bm =b0=1∈ N.
Hence,bm ∈ Nis proven in Case 1.
Similarly, we can prove bm ∈ N in Case 2 (using b1 = 1) and in Case 3 (using b2 =2) and in Case 4 (usingb3 =2r+1). It thus remains to provebm ∈ Nin Case 5.
So let us consider Case 5. In this case, we havem >3. Thus,m ≥4 (sincemis an integer), so that m−2 ≥ 4−2 = 2. Hence, Observation 1 (applied to n = m−2) yields b(m−2)+2 = b(m−2)−2br(m−2)+1−brm−−12H brm−2
. In view of (m−2) +2 = m and (m−2)−2=m−4 and(m−2) +1=m−1, this rewrites as
bm =bm−4brm−1−brm−−12H brm−2
. (105)
But m−2 ∈ N (since m ≥ 4 ≥ 2) and m−2 < m. Hence, (104) (applied to n = m−2) yields bm−2 ∈ N⊆ Z. Also, bm−2 is a positive rational number (since (b0,b1,b2, . . .)is a sequence of positive rational numbers) and thus a positive integer (sincebm−2 ∈N), hence a nonzero integer. Thus,brm−2is a nonzero integer as well.
Therefore, Lemma 2.68 (applied to x = brm−2) shows that H bmr−2
∈ Z. In other words, H brm−2
is an integer. Also, r−1 ≥ 0 (since r ≥ 1), and thus r−1 ∈ N.
Hence,brm−−12is an integer (since bm−2 is an integer).
Also, m−4 ∈ N (since m ≥ 4) and m−4 < m. Hence, (104) (applied to n = m−4) yieldsbm−4∈ N⊆Z. In other words,bm−4is an integer.
Similarly,bm−1 is an integer. Thus, brm−1 is an integer.
We now know that the four numbersbm−4,bmr−1,bmr−−12and H brm−2
are integers.
Thus, the numberbm−4brm−1−brm−−12H brm−2
also is an integer (since it is obtained from these four numbers by multiplication and subtraction). In view of (105), this
rewrites as follows: The number bm is an integer. Since bm is positive, we thus conclude thatbm is a positive integer. Hence, bm ∈ N. This shows that bm ∈ Nin Case 5.
We now have proven bm ∈ N in each of the five Cases 1, 2, 3, 4 and 5. Thus, bm ∈ N always holds. In other words, Proposition 2.66 (a) holds for n = m.
This completes the induction step. Thus, Proposition 2.66 (a) is proven by strong induction.
(b)Assume thatr≥2. We must prove thatbn | bn−2+bn+2 for eachn ∈Z≥2. So let n∈ Z≥2. We must show that bn |bn−2+bn+2.
Fromn∈ Z≥2, we obtain n≥2, so that n−2∈ N.
Proposition 2.66 (a) (applied to n−2 instead of n) yields bn−2 ∈ N. Similarly, bn ∈ N and bn+1 ∈ N and bn+2 ∈ N. Thus, all of bn−2, bn+1, bn are bn+2 are integers.
Butbn is a positive rational number (since(b0,b1,b2, . . .)is a sequence of positive rational numbers), and therefore a positive integer (sincebn is an integer). Hence, brnis a positive integer, and thus a nonzero integer. Therefore, Lemma 2.68 (applied tox =brn) shows that H(brn)∈ Z. In other words, H(brn) is an integer.
We haven+2≥2. Hence, the recursive definition of the sequence (b0,b1,b2, . . .) yieldsbn+2 = b
r
(n+2)−1+1 b(n+2)−2
= b
r
n+1+1
bn . Multiplying this equality by bn, we obtain bnbn+2 =brn+1+1. (106) We haver−2∈ N(sincer≥2) andbn ∈ N. Hence, brn−2 is an integer.
Observation 1 yields bn+2 =bn−2brn+1−brn−1H(brn). Thus, bn+2
| {z }
=bn−2brn+1−br−1n H(brn)
+bn−2
=bn−2brn+1−brn−1H(brn) +bn−2 =bn−2brn+1+bn−2
| {z }
=bn−2(brn+1+1)
−brn−1H(brn)
=bn−2 brn+1+1
| {z }
=bnbn+2 (by (106))
− brn−1
|{z}
=bnbr−2n
H(brn) = bn−2bnbn+2−bnbrn−2H(brn)
=bn
bn−2bn+2−brn−2H(brn). (107) But bn−2bn+2−bnr−2H(brn) is an integer (because bn−2, bn+2, brn−2 and H(brn) are integers). Denote this integer byz. Thus, z = bn−2bn+2−brn−2H(bnr). Since bn and
zare integers, we have bn | bn z
|{z}
=bn−2bn+2−br−2n H(brn)
=bn
bn−2bn+2−brn−2H(brn)=bn+2+bn−2 (by (107))
=bn−2+bn+2. This proves Proposition 2.66(b).
For a (slightly) different proof of Proposition 2.66, seehttp://artofproblemsolving.
com/community/c6h428645p3705719.
Remark 2.69. You might wonder what happens if we replace “brn−1 +1” by
“brn−1+q” in Proposition 2.66, where q is some fixed nonnegative integer. The answer turns out to be somewhat disappointing in general: For example, if we set r=3 andq =2, then our sequence (b0,b1,b2, . . .) begins with
b0=1, b1 =1, b2= 1
3+2 1 =3, b3= 3
3+2
1 =29, b4 = 29
3+2
3 = 24 391 3 ,
at which point it becomes clear that bn ∈ Nno longer holds for all n ∈ N. The same happens for r = 4, 5, . . . , 11 (though the first n that violates bn ∈ Nis not always 4); this makes me suspect that it also happens for allr >2.
However, bn ∈ N still holds for all n ∈ N when r = 2. This follows from Exercise 2.1 below.
Exercise 2.1. Fix a nonnegative integer q. Define a sequence (b0,b1,b2, . . .) of positive rational numbers recursively by setting
b0 =1, b1 =1, and bn = b
2n−1+q bn−2
for eachn ≥2.
(Thus,
b2= b
12+q b0 = 1
2+q
1 =q+1;
b3= b
22+q b1
= (q+1)2+q
1 =q2+3q+1;
b4= b
32+q b2 = q
2+3q+12
+q
q+1 =q3+5q2+6q+1;
and so on.) Prove that:
(a)We have bn = (q+2)bn−1−bn−2 for eachn∈ Z≥2. (b)We have bn ∈Nfor eachn ∈ N.