Sie sind auf Seite 1von 11

Stern Polynomials

Kathryn McCormick December 2, 2011


Abstract Sterns sequence has been studied in several forms since 1858 as Sterns diatomic array, the fusc function, the Calkin-Wilf tree, and as a type of binary partition function. Recently, Sterns sequence was generalized to a sequence of polynomials, which is equal to Sterns sequence when x = 1. Motivated by the work of Carlitz on Sterling numbers, Stern polynomials can be shown to be congruent modulo 2 to two other very interesting sequences of polynomials: Bell polynomials and Chebyshev polynomials.

1
1.1

Sterns Sequence
Introduction
0, 1, 1, 2, 1, 3, 2, 3, 1, 4, 3, 5, 2, 5, 3, 4, 1, . . . Sterns sequence begins:

where it follows the recursive denition a(0) = 0, a(1) = 1, and for n 1, a(2n) = a(n) a(2n + 1) = a(n) + a(n + 1) As can be seen, the sequence is fairly repetitive, partially because of the repetitions of the even terms, so that it is the odd-numbered terms that are driving the sequence to change. In particular, it can be noted that every term numbered a power of two is 1 (This is true for 20 and if we assume this up to the term numbered 2k , by the denition we have a(2k+1 ) = a(2k ) = 1). In section 1.3 we will go over the connection between Sterns sequence and binary numbers. If we look between each of these repeating 1s, we notice that these sequences inside appear symmetric. In fact, when M. A. Stern published the sequence in 1858 [11] it was in the form of an array, and generated much like Pascals triangle.

Sterns Diatomic Array 1 1 1 2 1 1 3 2 3 1 1 4 3 5 2 5 3 4

(For that reason Sterns sequence is often referred to as Sterns Diatomic array [8]) The next line in the array is generated by bringing down the last line and inserting between each pair of numbers their sum. Sterns sequence is very much related to Stern-Brocot trees (see [1],[5], or [11]). In the very same paper from 1858 Stern described this ordering of fractions and also how each positive rational number appears only once in this ordering.

1.2

Counting the rationals

So, lets construct a sequence of rational numbers from Sterns sequence, so we can then follow Stern in showing a one to-one correspondence between these fraction and the positive rational numbers (and so give one proof of the fact that the rational numbers are countable). We will do this by looking at the sequence: a(n) 1 1 2 1 3 2 3 , , , , , , ,..., ,... 1 2 1 3 2 3 1 a(n + 1) The following proof was done by Northshield [8], there is also a very nice proof by Calkin and Wilf in [1]. Proposition 1. The map n a(n)/a(n + 1) is a bijection between the natural numbers and the positive rationals. Proof. This will be proved once we show that every relatively prime pair appears once on the list L = [1, 1], [1, 2], [2, 1], [1, 3], [3, 2], [2, 3], [3, 1], . . . , [a(n), a(n + 1)], . . . (i.e. every reduced fraction can be written as a pair of relatively prime numbers, and every ordered pair of relatively prime numbers corresponds to a reduced fraction). Dene the Slow Euclidean Algorithm (SEA) on pairs of natural numbers [a, b]. Given [a, b], SEA subtracts the larger number from the smaller until they are equal. For example: [3, 7] [3, 4] [3, 1] [2, 1] [1, 1] We can see that the gcd of the two numbers is preserved at each step (if [a = da1 , b = db1 ], a > b, then [d(a1 b1 , db1 ] has the same gcd d), so this algorithm terminates on [gcd(a, b), gcd(a, b)]. Let Ln = [a(n), a(n + 1)] be a specic pair from the list. From the denition of Sterns sequence, we can see that for n > 1, at each step of the SEA we have SEA : L2n L L2n+1 L 2

since at every step we have either [a(2n), a(2n + 1)] = [a(n), a(n) + a(n + 1)] or [a(2n + 1), a(2n + 2)] = [a(n) + a(n + 1), a(n + 1)]. Furthermore, if SEA : [a, b] Ln then either [a, b] = L2n or [a, b] = L2n+1 by denition. Since when SEA is given an L2n or an L2n+1 it must eventually terminate at L1 = [1, 1], and so every Ln is a pair of relatively prime numbers. And, if there were a relatively prime pair not in L, then all of its successors, including the last [1, 1], couldnt be in L, which would be a contradiction. Hence, every relatively prime pair appears in L. If there were a pair that appeared more than once then there would exist a smallest n > 1 such that Ln = Lm for m > n. By applying one step of the SEA to Ln and Lm we must have n/2 = m/2 so that m = n + 1. But then a(n) = a(n + 1) = a(n + 2), which contradicts the assumption of the smallest n (which we did by using the denition of our sequence and looking at a( n/2 ) and a( n + 1/2 )). So, each relatively prime pair appears once and only once in L.

1.3

Hyperbinary representation

Not only does {a(n)/a(n+1)} count something interesting, but {a(n)} itself can be said to count something interesting: hyperbinary representations. Given a positive integer n, a(n+1) is the number of ways to write n = e0 20 + e1 21 + . . . + ek 2k where ei {0, 1, 2}. For example, 5 = 20 + 22 = 20 + 2 21 and so 5 has two hyperbinary representations; note that a(5 + 1) = 2. And, since a number in the form n = 2k 1 = 2k1 + (2k1 1) = 2k1 + 2k2 + . . . + 20 can only be written in its binary form, the number has one hyperbinary representation corresponding to the fact that a(2k 1 + 1) = 1 for all k > 0. We will prove this correspondence by looking at the generating function of a(n + 1) and showing it is the same as the generating function of the number of hyperbinary representations of n. This proof follows Northshield [8] and Carlitz [2],[3]. Proposition 2. For n 1, the number of hyperbinary representations of n is equal to a(n + 1). Proof. Consider the generating function A(x) =
n=1

a(n + 1)xn

Lemma. lim supx |a(n)|1/n is nite. Proof. First, we can see that a(k) 2n when 1 k n 2i < 2n+1 . By the i=0 denition of Sterns diatomic array, a term in the nth row of the array cant be more than 2n1 + 2n1 = 2n . Now, lim supn |2n |1/n = 2 and so by comparison, lim supx |a(n)|1/n is nite.

By the lemma and Cauchys root test, then, A(x) has a positive radius of convergence. For x in the interval of convergence, then,

A(x) =
n=0

a(2n + 1)x

2n

+
n=0

a(2n + 2)x2n+1
2n

=
n=0

[a(n) + a(n + 1)]x a(n + 1)x2n + x


n=0

+
n=0

a(n + 1)x2n+1

a(n + 1)x2n +
n=0 n=0

a(n)x2n

= (1 + x + x2 )A(x2 )

(since a(0) = 0)

Since A(x) is continuous at x = 0 and A(0) = 1, we have: A(x) = (1 + x + x2 )(1 + x2 + x4 )(1 + x4 + x8 ) = (x01 + x11 + x21 )(x02 + x12 + x22 )(x04 + x14 + x24 ) So a general term in A(x) will look like a(n + 1)xn = cxe0 +e1 2+e2 4+e3 8+... where ei {0, 1, 2}, and the coecient c will be the number of ways to write n as this kind of sumthat is, the number of hyperbinary representations of n. Sterns sequence has other properties such as relations to the Fibonacci sequence [8], continued fractions [8], and properties related to the diatomic array [7]. There are many other references in [10, A002487]. We, however, are going to move on to an extension of the sequence.

2
2.1

Stern Polynomials
An extension of Sterns sequence
a(2n; x) = a(n; x2 )a(2n + 1; x) = xa(n; x2 ) + a(n + 1; x2 ) Stern polynomials [5] are dened recursively as a(0; x) = 0, a(1; x) = 1, and for n > 1, (1)

We can see that the rst few Stern polynomials are 0, 1, 1, 1 + x, 1, 1 + x + x2 , 1 + x2 , 1 + x + x3 , 1, . . . Just like Sterns sequence, the 0th and 1st terms are 0 and 1, respectively, and all the terms numbered powers of two will be 1. To check that it makes sense to call this an extension of Sterns sequence, set x = 1. The recursion will be just the same as in Sterns sequence, 4

and looking at the example of the rst terms, we obviously recover the rst part of Sterns sequence. One additional recursion that we will be using comes directly from the denition, that a(2n + 1; x) = xa(2n; x) + a(2n + 2; x). (2)

We get this by substituting the even recursion into the odd recursion. It is useful to have a recursion without any x2 s. To explain why we are interested in Stern polynomials as an extension of Sterns sequence requires that we go back to some work of Carlitz.

2.2

Why extend? Some history

In the 1960s Carlitz was investigating a sequence of polynomials called single variable Bell polynomials [3], which are dened as
n

An (x) =
k=0

S(n, k)xk

where S(n, k) is a Sterling number of the second kind. These polynomails are of interest in combinatorics, S(n, k) being the number of ways to partition n objects into k nonempty subsets. S(n, k) also satises the triangular relation S(n, k) = S(n 1, k 1) + kS(n 1, k). The sequence of these Bell polynomials begins x, x + x2 , x + 3x2 + x3 , x + 7x2 + 6x3 + x4 , x + 15x2 + 25x3 + 10x4 + x5 , . . . Carlitz was interested in factorizing these polynomials (see [2],[3],[4]) He began his work by looking at the factorization of the An (x) with the coecients taken modulo 2. Every An (x) has a factor of x since S(n, 0) = 0, but after factoring this out he looked to see if the remaining part was irreducible. If he could determine that the remainder was irreducible mod 2, then the remainder would be irreducible in the rationals. This is due to the fact that if a polynomials is reducible in the integers, it is reducible modulo a prime, so a polynomial irreducible mod 2 cannot be reducible originally in the rationals [6, pg.310]. For example, he noted that since x1 A4 (x) = 1 + 7x + 6x2 + x3 1 + x + x3 (mod 2) and 1 + x + x3 is irreducible mod 2, x1 A4 (x) is irreducible over the rationals [3]. As Carlitz was working on this he became interested in nding the number of odd coefcients in the An (x) and he obtained Sterns sequence (shifted by one) as one of his results (which he denoted by a recursion formula and in connection with hyperbinary partitions, but not by name). In 2001 Urbiha noted the connection between Carlitzs work and Sterns sequence [12]. In 2007 Dilcher and Stolarsky went on to set up an explicit connection between single variable Bell polynomials and Sterns sequence by constructing Stern polynomials (see 1 [5]). That is, they proved that the polynomials an (x) = x An (x) are congruent to the Stern polynomials a(2n 1; x), modulo 2. 5

2.3

Two congruences modulo 2

To prove our main theorem we rst have to nd recurrences of Stern polynomials modulo 2, so that we can easily dene them modulo 2. These proofs naturally originate in [5]. Lemma 1. For all n 1, a(n + 1; x) a(n; x) + xa(n 1; x) (mod 2) Proof. When n is odd, (4) follows from taking (3) modulo 2 and rearranging, that is, a(n; x) a(n + 1; x) + xa(n 1; x) a(n + 1; x) a(n; x) + xa(n 1; x) (mod 2) Now let n be even so n = 2k and we will prove (4) by induction on k. When k = 1 we have x + 1 = 1 + 1 x which is true. Suppose that (4) holds up to a specic k 1 (and so, up to n = 2k 2). Then either by the induction hypothesis or by the already-proven odd case we have: a(k + 1; x) a(k; x) + xa(k 1; x) (mod 2) 2 2 2 2 a(k + 1; x ) a(k; x ) + x a(k 1; x ) (mod 2) 2 By (1), we get a(2k + 2; x) a(2k; x) + x a(2k 2; x) (mod 2) On the other hand, from (4) we have both a(2k + 2; x) = a(2k + 1) xa(2k; x) and a(2k 2; x) = a(2k 1; x) a(2k; x) so that when we substitute these equations into (5) we arrive at a(2k + 1; x) xa(2k; x) a(2k; x) + xa(2k 1; x) xa(2k; x) which gives (4) for n = 2k and completes the proof by induction. Lemma 2. For all n 2, a(n + 2; x) a(n; x) + x2 a(n 2; x) (mod 2) (5) (mod 2) (3)

(4)

Proof. First, take (4), which we just proved, and replace n by n + 1, and then afterwards substitute in (4) with n replaced by n 1 for the last term to get a(n + 2) a(n + 1; x) + x[a(n 1; x) + xa(n 2; x)] Again, by (4) we have a(n + 1; x) + xa(n 1; x) a(n; x) and combining this with (7) we prove (6). Now that we have this second recursion formula, we will look at the Bell polynomials to prove they have the same recursion. Note that since this recursion is only dealing with all odd or all even terms, we will be comparing the modied Bell polynomials {an (x)} to a subsequence of Stern polynomials, in this case the subsequence of odd terms. It makes sense that it would be the odd subsequence since, just like Sterns sequence, the odd terms are the ones that are most drastically changing. 6 (mod 2) (mod 2) (6)

2.4

Bell polynomials and Stern polynomials

To show a modulo-2 recurrence for the Bell polynomials, we will be following a special case of a result of Touchard that Dilcher and Stolarsky gave in [5]. Before starting the proof it is useful to remember the denition of Sterling numbers (see section 2.2). Lemma 3. For n 1, an+1 (x) an (x) + x2 an1 (x) (mod 2) (7)

Proof. Given a polynomial p, let E(p) be the even part, so that


n n/2

E(
i=0

ai x ) =
i=1

a2i x2i .

Remembering that the Sterling numbers S(n, k) satisfy the triangular relation S(n, k) = S(n 1, i 1) + iS(n 1, i), and noting
n+1 n

an (x) =
i=0

S(n + 1, i)x

i1

=
i=0

S(n + 1, i + 1)xi

(Since S(n + 1, 0) = 0)

we nd
n+1

an+1 (x) =
i=0 n+1

S(n + 2, j + 1)xi [S(n + 1, j) + iS(n + 1, j + 1)]xi


i=0 n+1

= xan (x) +
i=0

iS(n + 1, j + 1)]xi (mod 2)

xan (x) + E(an (x))

so that an (x) satisfy the recurrence relation a0 (x) = 1, a1 (x) = 1 + x, an+1 (x) xan (x) + E(an (x)). (mod 2) (8)

Now consider the polynomials bn (x) dened by b0 (x) = 1, b1 (x) = 1 + x, and bn+1 (x) = bn (x) + x2 bn1 (x). (9)

We can show that both E(bn (x)) and x2 bn1 (x) (x 1)bn (x) (= bn+1 (x) bn (x)) satisfy the same recursion (9), and by checking initial conditions (n = 1, 2) that they are equal, therefore we have E(bn (x)) + xbn (x) = x2 bn1 (x) (x 1)bn (x) + xbn (x) = bn+1 (x). 7 (10)

Finally, using induction we will prove that an (x) bn (x) (mod 2), which once proven we can combine with (9) to establish this lemma. Checking initial conditions (n = 0, 1) we see that they agree. Suppose the congruence held for all k n. Then with (8) and (10) we would have an+1 (x) xan (x) + E(an (x)) xbn (x) + E(bn (x)) bn+1 (x) which completes the induction and proves the lemma. Now that we have our last recurrence modulo 2 done, the congruence theorem is simple to state. Proposition 3. The polynomials {an (x)} are congruent to the polynomials {a(2n + 1; x)} modulo 2 for n 0. Proof. Remembering (6) we nd that for the odd Stern polynomials {a(2n 1; x)} we have the recurrence relation a(2(n + 1) 1) a(2(n) 1; x) + x2 a(2(n 1) 1; x) (mod 2) (mod 2)

and that a(1; x) = 1 and a(3; x) = 1 + x. We just showed in Lemma 3 that for the scaled Bell polynomials {an (x)} we have the recurrence relation an+1 (x) an (x) + x2 an1 (x) (mod 2) and looking at the denition of Bell polynomials we see 1 1 a0 (x) = A1 (x) = x x and a1 (x) = 1 1 A2 (x) = x x
1

S(1, i)xi =
i=0 2

1 (1x) x

=1

S(2, i)xi =
i=0

1 (1x + 1x2 ) = 1 + x x

and so by their recursion formulas these sequences of polynomials must be equal.

2.5

Chebyshev polynomials and Stern polynomials

In the same work Carlitz did on Bell polynomials [3] he noted a (modulo-2) congruence between the Bell polynomials and scaled Chebyshev polynomials. There are two kinds of Chebyshev polynomials, the rst kind being dened as Tn (x) = cos n, and the second kind as Un (x) = sin (n + 1) , sin 8 x = cos (11)

x = cos

(12)

where we can see Un1 (x) = kind are:

1 T (x). n n

[9] The rst few Chebyshev polynomials of the rst

1, x, 2x2 1, 4x3 3x, 8x4 8x2 + 1, 16x5 20x3 + 5x, 32x6 48x4 + 182 1, . . . It would make sense that if there was a (modulo-2) connection between these and the Bell polynomials, there would also be a connection with Stern polynomials. In fact, Stern polynomials have, in a way, even a slightly deeper connection to Chebyshev polynomials because we can match up every one of them with a Chebyshev polynomial, and not just the odd subsequence. Before we go into showing the Stern polynomials congruence with the Chebyshev polynomials it will be useful to prove the following lemmas. Lemma 4. The Chebyshev polynomials of the rst kind {Tn (x)} have the recurrence relation T0 (x) = 1, T1 (x) = x, and Tn (x) = 2xT n 1(x) Tn2 (x) Proof. T0 (x) = cos 0 = 1 and T1 (x) = cos cos1 x = x come directly from the denition. Consider 2xTn1 (x) Tn2 (x) = 2 cos (n 1) cos (n 2) By using the addition formula for cosine we can expand this and cancel terms to get = cos2 cos n + sin2 cos n = cos n. (13)

In a similar manner we can nd Lemma 5. The Chebyshev polynomials of the second kind can be dened by the recurrence relation U0 (x) = 1, U1 (x) = 2x, and Un (x) = 2xUn1 (x) Un2 (x) (14)

Obviously, if we take these recurrence relations as they are modulo 2 they will not help us, so we will scale the Chebyshev polynomials to get the same recurrence relation as that of the Stern polynomials. This follows the proof given by Dilcher and Stolarsky [5]. Proposition 4. For all n 0, 2xn/2 Tn and xn/2 Un 1 2x1/2 1 2x1/2 a(n; x) a(n + 1; x) (mod 2) (mod 2) (15) (16)

Proof. Denote the left-hand sides of (15) and (16) as Tn (x) and Un (x), respectively. 1 1/2 Substitute x = 2 y into the recurrence (13) to get Tn 1 1/2 y 2 = y 1/2 Tn1 1 1/2 y Tn2 2 1 1/2 y 2

Then multiply each side by 2y n/2 to get Tn (y) = Tn1 (y) y Tn2 (y) Tn1 (y) + y Tn2 (y)

(mod 2).

With a similar process we can get the came recurrence for Un (x). This is the same recurrence formula as (3). Now we note T0 (x) 0 = a(0; x) T1 (x) 1 = a(1; x) and U0 (x) 1 = a(1; x) U1 (x) 1 = a(2; x) (mod 2) (mod 2) (mod 2) (mod 2).

Thus by having the same recursion formula these sequences of polynomials are the same.

References

[1] N. Calkin and H.S. Wilf, Recounting the rationals, Amer. Math. Monthly 107 (2000) 360-367. [2] L. Carlitz, A problem in partitions related to the Sterling numbers, Bull. Amer. Math. Soc. 70 (1964) 275-278. [3] L. Carlitz, Single variable Bell polynomials, collect. Math. 14 (1960) 13-25. [4] L. Carlitz, Some partition problems related to the Sterling numbers of the second kind, Acta Arith. 10 (1965) 409-422. [5] K. Dilcher and K. Stolarsky, A polynomial analogue to the Stern sequence, International J. of Number Theory 3 (2007) 85-103. [6] D.S. Dummit and R.M Foote, Abstract Algebra, 2nd ed., John Wiley and Sons, New York, 1999. [7] D.H. Lehmer, On Sterns diatomic series, Duke Math. J. 36 (1969) 55-60. [8] S. Northshield, Sterns diatomic sequence 0,1,1,2,1,3,2,3,1,4,..., Amer. Math. Monthly 117 (2010) 581-598. [9] T.J. Rivlin, The Chebyshev polynomials, Pure and Applied Mathematics, Wiley Interscience, New York, 1974. [10] N.J.A. Sloane, On-Line Encyclopedia of Integer Sequences, http://www.research.att.com/ njas/sequences/. [11] M.A. Stern, Uber eine zahlentheortische Funktion, J. Reine Agnew. Math. 55 (1858) 193-220. 10

[12] I. Urbiha, Some properties of a function studied by de Rham, Carlitz and Dijkstra and its relation to the (Eisenstein-)Sterns diatomic sequence, Math. Commun. 6 (2001) 181-198.

11

Das könnte Ihnen auch gefallen