18

$\color{brown}{\textbf{The setup and observations}}$

For $d\in{\bf N}$ and $n\in{\bf N}^\star$, denote by $S_d(n)$ the sum of the $d$-powers of integers $1$ to $n$: $$S_d(n)=\sum_{k=1}^nk^d.$$

A simple use of the binomial theorem in developping the quantity $(k+1)^{d+2}-k^{d+2}$ and summing it yields the following recursive relation: $$S_{d+1}(n)=\frac{1}{d+1}\left\{(n+1)^{d+2}-1-\sum_{k=0}^d\binom{d+2}{k}S_k(n)\right\}.$$

This allowed me to implement a $\verb|Python|$ script to compute them inductively. I calculated the first 100 polynomials. Here are some examples:

  • $\small S_8(n)=\displaystyle\frac{n(n + 1)(2n + 1)(5n^6 + 15n^5 + 5n^4 - 15n^3 - n^2 + 9n - 3)}{90}$

  • $\small S_9(n)=\displaystyle\frac{n^2(n + 1)^2(n^2 + n - 1)(2n^4 + 4n^3 - n^2 - 3n + 3)}{20}$

  • $\small S_{10}(n)=\displaystyle\frac{n(n + 1)(2n + 1)(n^2 + n - 1)(3n^6 + 9n^5 + 2n^4 - 11n^3 + 3n^2 + 10n - 5)}{66}$

  • $\small S_{11}(n)=\displaystyle\frac{n^2(n + 1)^2(2n^8 + 8n^7 + 4n^6 - 16n^5 - 5n^4 + 26n^3 - 3n^2 - 20n + 10)}{24}$

Amongst all things, there are a few of interest:

  • Every $S_d(n)$ is a rational polynomial in $n$, where OEIS:A064538 appears.

  • The roots of the polynomials seem to follow some interesting pattern, as described on this thread.

  • For odd values of $d$, $S_d$ is divisible by $\left(\frac{n(n+1)}{2}\right)^2$. For even values of $d$, it is divisible by $\frac{n(n+1)(2n+1)}{6}$.

$\color{brown}{\textbf{Getting into polynomials}}$

For $\delta\in{\bf N}$, denote as $F_\delta(X)$ the Faulhaber polynomial: $$F_\delta(X)=\frac{1}{2^{2\delta+2}(\delta+1)}\sum_{q=0}^\delta\binom{2\delta+2}{2q}(1-q)B_{2q}\left[(8X+1)^{\delta+1-q}-1\right],$$ where $(B_n)_{n\in{\bf N}}$ are the Bernoulli numbers. Then, this polynomial has lowest degree term $X^2$, and Faulhaber showed that:

$$S_{2d+1}(n)=F_d\left(\frac{n(n+1)}{2}\right)\qquad\text{and}\qquad S_{2d}(n)=\frac{n+1/2}{2d+1}F_d'\left(\frac{n(n+1)}{2}\right).$$

The question now comes when one looks at the irreducible decomposition of $S_d$ as a polynomial over ${\bf Q}$. Having the previous properties in mind, set: $$T_d(n)=\begin{cases}\frac{4}{n^2(n+1)^2}S_d(n)&\text{if $d$ is odd,}\\\frac{6}{n(n+1)(2n+1)}S_d(n)&\text{otherwise.}\end{cases}$$

Then my script returned that $T_d$ was always irreducible, except for the values $d=9$ and $d=10$. Just so that we are on the same page, here are the corresponding examples from before:

  • $\small T_8(X)=\displaystyle\frac{5X^6 + 15X^5 + 5X^4 - 15X^3 - X^2 + 9X - 3}{15}$

  • $\small T_9(X)=\displaystyle\frac{(X^2 + X - 1)(2X^4 + 4X^3 - X^2 - 3X + 3)}{5}$

  • $\small T_{10}(X)=\displaystyle\frac{(X^2 + X - 1)(3X^6 + 9X^5 + 2X^4 - 11X^3 + 3X^2 + 10X - 5)}{11}$

  • $\small T_{11}(X)=\displaystyle\frac{2X^8 + 8X^7 + 4X^6 - 16X^5 - 5X^4 + 26X^3 - 3X^2 - 20X + 10}{6}$

$\color{brown}{\textbf{The questions}}$

Are there known results regarding this? One can ask the following questions:

  • Is it true that $T_d$ is always irreducible over ${\bf Q}$ for $d\neq 9,10$?

  • Is there an explanation behind those exceptional values $d=9,10$?

  • Or if the first question turns out to be false, what are those values of $d$ for which $T_d$ is reducible? Do they follow some pattern? (e.g. would it happen that an odd value would be followed by an even value, both sharing the property that $T_d$ is reducible?)

$\color{brown}{\textbf{My (failing) ideas}}$

What I thought about doing first is proving that if $P(X)$ is irreducible over ${\bf Q}$, then $P\left(\frac{X(X+1)}{2}\right)$ is irreducible as well. But it turns out it isn't true, for if simply $P(X)=X$, then the statement doesn't hold. This means that studying the Faulhaber polynomial just won't work, and one really has to work with the $T_d$.

Now, working for instance on the odd case scenarios, we seek to prove irreducibility of $$\small T_{2d+1}(X)=\frac{1}{2^{2d}(d+1)X^2(X+1)^2}\sum_{q=0}^d\binom{2d+2}{2q}(1-q)B_{2q}\left[(4X(X+1)+1)^{d+1-q}-1\right]$$ for all $d\neq4$.

I'm lost. I have no clue of where to even begin, the definition seems to rely on the values of the numerators and denominators of the Benoulli numbers, which I know are some very tough problems from number theory. Thus, applying either the Eisenstein criterion, or reducing modulo some prime number, seems like a dead end.

I have made a related question involving an expression I have derived for the odd case: $$\small T_{2d+1}(X)=\frac{1}{2(d+1)}\sum_{\ell=0}^{2d-2}\left\{(-1)^\ell\sum_{k=\ell+2}^{2d}\binom{k}{\ell+2}\binom{2d+2}{k+2}B_{2d-k}\right\}X^\ell.$$

Now, I have no clue whether this is any useful into proving that it is irreducible. Indeed, assuming the inner sum cannot be simplified further (which was the point of the related question), proving irreducibility of $T_{2d+1}$ therefore highly relies on the numerators and denominators of the Bernoulli numbers, which are known to be hard questions...

Anthony
  • 1,389

2 Answers2

7

This is a very partial answer, but I hope it will give some insight.

  • For $d\ge 4$ even, $T_d(X)$ does not have any rational roots.

  • For $d\ge 5$ odd, $0,-1$ are not the roots of $T_d(X)$.


I started with $$S_d(n)=\frac{B_{d+1}(n+1)-B_{d+1}(0)}{d+1}$$ (see here) where $B_n(x)$ is the Bernoulli polynomials.

Using $B_{d+1}(n+1)=(-1)^{d+1}B_{d+1}(-n)$ (see here), we have $$S_d(n)=\frac{(-1)^{d+1}B_{d+1}(-n)-B_{d+1}(0)}{d+1}$$

Here, let us separate it into two cases.


Case 1 : For $d\ge 4$ even, since $B_{d+1}(0)=0$, we have $$S_d(n)=\frac{-B_{d+1}(-n)}{d+1}$$

By google search of "zeros of the Bernoulli polynomial", I found this paper which says in page 125 that "K. Inkeri [8] showed that $0,\frac 12$, and $1$ are the only rational zeros of $B_{2n+1}(x), n\geqslant 1$".

[8] K. INKERI, The real roots of Bernoulli polynomials, Ann. Univ. Turku. Ser. A I 37 (1959), 1-20.

So, it follows that $0,-\frac 12$ and $-1$ are the only rational roots of $S_d(x)$.

(I found this MO question. See this answer.)

Also, $S_d(n)$ dose not have factors $n^2, (n+1)^2, (2n+1)^2$.

This is because we have $S_d'(x)=B_{d}(-x)$ (see here), and $B_d(x)$ has no rational roots. (see Theorem 4 of this paper)

Therefore, we can say that $T_d(X)$ does not have any rational roots.


Case 2 : For $d\ge 5$ odd, we have $$S_d(n)=\frac{B_{d+1}(-n)-B_{d+1}(0)}{d+1}$$ However, it might be difficult to use this because $B_{d+1}(0)$ is not equal to $0$.

We see that $S_d(n)$ does not have factors $n^3,(n+1)^3$ since we have $S_d''(x)=-dB_{d-1}(-x)$.

Therefore, we can say that $0,-1$ are not the roots of $T_d(X)$.

mathlove
  • 155,494
1

This is also a very partial answer.but 0,−1/2,−1 are the only rational roots of the Faulhaber polynomial. A high school-level proof is possible without using Bernoulli numbers. When a and b are coprime integers (b > 0) satisfying S_m(a/b) = 0, we first show the following two points:
・When m is odd, a/b = -1 or 0
・When m is even, if a/b ≠ -1 or 0, then b is even
Finally, I provided a complete proof for the case where m is even. Hereafter, we assume $m \geq 1$.


Theorem 1 (1.1). The identity $S_m(-x) = (-1)^{m+1} S_m(x-1) \quad \cdots(1.1)$ holds.

Proof. It suffices to show that (1.1) has infinitely many solutions. Since $S_m(0) = (-1)^{m+1} S_m(-1) = 0$, (1.1) holds when $x = 0$. Assume (1.1) holds when $x = k$. Then $$\begin{align*} S_m(-k-1) &= S_m(-k) - (-k)^m \\ &= (-1)^{m+1} S_m(k-1) + (-1)^{m+1} k^m \\ &= (-1)^{m+1} S_m(k) \end{align*}$$ Therefore, (1.1) also holds when $x = k+1$. □


Let $k$, $l$, $N$ be integers, and let $k$ be a multiple of $N$ (where $N > 0$).

Theorem 2 (1.2). The congruence $S_m(k+l) \equiv S_m(k) + S_m(l) \pmod{N} \quad \cdots(1.2)$ holds.

Proof. The case $l = 0$ is trivial. Let $l \geq 1$. $$\begin{align*} S_m(k+l) - S_m(k) &= (k+1)^m + (k+2)^m + \cdots + (k+l)^m \\ &\equiv 1^m + 2^m + \cdots + l^m \pmod{N} \\ &\equiv S_m(l) \pmod{N} \end{align*}$$

Also, from (1.1), $$\begin{align*} S_m(k-l) - S_m(k) &= -(k-l+1)^m - (k-l+2)^m - \cdots - (k-l+l)^m \\ &\equiv (-1)^{m+1} \{(l-1)^m + (l-2)^m + \cdots + 0^m\} \pmod{N} \\ &\equiv (-1)^{m+1} S_m(l-1) \pmod{N} \\ &\equiv S_m(-l) \pmod{N} \end{align*}$$ □


Theorem 3 (1.3). The congruence $S_m(kl) \equiv l \cdot S_m(k) \pmod{N} \quad \cdots(1.3)$ holds.

Proof. First, we show the case when $k = N$. The case $l = 0$ is trivial. When $l \geq 1$, from (1.2), $$\begin{align*} S_m(Nl) &= S_m(0) + \sum_{j=0}^{l-1} \{S_m(N(j+1)) - S_m(Nj)\} \\ &\equiv \sum_{j=0}^{l-1} S_m(N) \pmod{N} \\ &\equiv l \cdot S_m(N) \pmod{N} \end{align*}$$

From this and (1.1), $$\begin{align*} S_m(N(-l)) - (-l) \cdot S_m(N) &= (-1)^{m+1} S_m(Nl-1) + l \cdot S_m(N) \\ &\equiv (-1)^{m+1} S_m(Nl) + l \cdot S_m(N) \pmod{N} \\ &\equiv \{(-1)^{m+1} + 1\} \cdot l \cdot S_m(N) \pmod{N} \end{align*}$$

From this, when $m$ is even, (1.3) holds.

Let $m$ be odd. Then, if $N$ is odd, $$\begin{align*} S_m(N) &= 0^m + 1^m + 2^m + \cdots + \left(\frac{N-1}{2}\right)^m + \left(N - \frac{N-1}{2}\right)^m + \cdots + (N-1)^m + (N-0)^m \\ &\equiv 0^m + 1^m + 2^m + \cdots + \left(\frac{N-1}{2}\right)^m - \left(\frac{N-1}{2}\right)^m - \cdots - 1^m - 0^m \pmod{N} \\ &\equiv 0 \pmod{N} \end{align*}$$

Therefore, $S_m(N) \equiv 0 \pmod{N} \quad \cdots(1.3.1)$, so when $N$ is odd, (1.3) holds.

When $N$ is even, the center term remains and $S_m(N) \equiv (N/2)^m \pmod{N}$, so $$\begin{align*} S_m(N(-l)) - (-l) \cdot S_m(N) &= \{(-1)^{m+1} + 1\} \cdot l \cdot S_m(N) \\ &\equiv 2l \cdot (N/2)^m \pmod{N} \\ &\equiv l \cdot N \cdot (N/2)^{m-1} \pmod{N} \\ &\equiv 0 \pmod{N} \end{align*}$$

Therefore, (1.3) also holds when $N$ is even.

For the general case of $k$, $k$ can be expressed as $k = Nk'$ for some integer $k'$, so $$S_m(kl) \equiv S_m((Nk')l) \equiv k' \cdot l \cdot S_m(N) \equiv l \cdot S_m(Nk') \equiv l \cdot S_m(k) \pmod{N}$$

Therefore, (1.3) holds. □


Hereafter, (1.1), (1.2), (1.3) are used without notice.

Let integers $a$, $b$, $c$ ($b > 0$) satisfy $S_m(a/b) = 0$, and let $a$ and $b$ be coprime. Also, let integer $N$ be a divisor of $a - bc$.

Lemma 1 (2.1). There exist integers $s$, $t$ such that $\displaystyle\frac{a}{b} = \frac{Ns - c}{N^2t - 1} \quad \cdots(2.1)$.

Proof. Since $a$ and $b$ are coprime, $a - bc$ and $b$ are coprime. Therefore, $aN$ and $b$ are coprime. Furthermore, since $(a - bc)/N$ is an integer, there exist integers $s$, $t$ satisfying the Diophantine equation $aNt - bs = (a - bc)/N \quad \cdots(2.1.1)$. Transforming (2.1.1) yields (2.1). □


Lemma 2 (2.2). The following identity holds (where $y \neq 0$): $$y^{2m+1} S_m(x/y) = y^m \{S_m(x+y) - S_m(y)\} - \sum_{j=0}^{m-1} \binom{m}{j} y^{m-1-j} S_{m-j}(y) \{y^{2j+1} S_j(x/y)\} \quad \cdots(2.2)$$

Proof. When $n$, $y > 0$ are integers, $$\begin{align*} S_m((n+1)y) - S_m(y) &= \sum_{l=1}^{n} \{S_m((l+1)y) - S_m(ly)\} \\ &= \sum_{l=1}^{n} \sum_{k=1}^{y} (ly + k)^m \\ &= \sum_{l=1}^{n} \sum_{k=1}^{y} \sum_{j=0}^{m} \binom{m}{j} l^j y^j k^{m-j} \\ &= \sum_{l=1}^{n} \sum_{j=0}^{m} \binom{m}{j} l^j y^j S_{m-j}(y) \\ &= \sum_{j=0}^{m} \binom{m}{j} y^j S_{m-j}(y) S_j(n) \end{align*}$$

Since there are infinitely many $n$, $y$ satisfying this equality, it is an identity. Therefore, when $y \neq 0$, we can substitute $n = x/y$: $$\begin{align*} S_m(x+y) - S_m(y) &= \sum_{j=0}^{m} \binom{m}{j} y^j S_{m-j}(y) S_j(x/y) \\ &= y^{m+1} S_m(x/y) + \sum_{j=0}^{m-1} \binom{m}{j} y^j S_{m-j}(y) S_j(x/y) \\ &= y^{m+1} S_m(x/y) + \sum_{j=0}^{m-1} \binom{m}{j} y^{-1-j} S_{m-j}(y) \{y^{2j+1} S_j(x/y)\} \end{align*}$$

Multiplying both sides by $y^m$ yields (2.2). □


Theorem 4 (2.3). The congruence $S_m(a - bc) \equiv -b \cdot S_m(c) \pmod{N} \quad \cdots(2.3)$ holds.

Proof. For the integers $s$, $t$ in (2.1), let $x = Ns - c$, $y = N^2t - 1$. Then, since $x/y = a/b$, $y^{2 \cdot 0 + 1} S_0(x/y) = x = Ns - c$ is an integer. Also, from (2.2), when $y^{2j+1} S_j(x/y)$ is an integer for $j = 0, 1, \ldots, m-1$, $y^{2m+1} S_m(x/y)$ is also an integer. Therefore, both the left side and the terms inside $\{\,\}$ on the right side of (2.2) can be integers.

Furthermore, for $S_m(y)$, $S_{m-j}(y)$ on the right side, letting $j' = 0, 1, \ldots, m$, $$S_{j'}(y) = S_{j'}(N^2t - 1) \equiv Nt \cdot S_{j'}(N) + S_{j'}(-1) \equiv 0 \pmod{N}$$

Therefore, taking mod $N$ on both sides of (2.2), $$\begin{align*} (N^2t - 1)^{2m+1} S_m(a/b) &\equiv (N^2t - 1)^m S_m(Ns - c + N^2t - 1) \pmod{N} \\ &\equiv (-1)^m S_m(N^2t + Ns - c - 1) \pmod{N} \\ &\equiv -S_m(-N^2t - Ns + c) \pmod{N} \end{align*}$$

Since $S_m(a/b) = 0$, we have $S_m(-N^2t - Ns + c) \equiv 0 \pmod{N}$. Also, $$\begin{align*} S_m(-N^2t - Ns + c) &\equiv -Nt \cdot S_m(N) - s \cdot S_m(N) + S_m(c) \pmod{N} \\ &\equiv -s \cdot S_m(N) + S_m(c) \pmod{N} \end{align*}$$

Therefore, $-bs \cdot S_m(N) \equiv -b \cdot S_m(c) \pmod{N}$. Furthermore, $$\begin{align*} S_m(a - bc) &= S_m((aNt - bs)N) \\ &\equiv aNt \cdot S_m(N) - bs \cdot S_m(N) \pmod{N} \\ &\equiv -bs \cdot S_m(N) \pmod{N} \end{align*}$$

From the above, $S_m(a - bc) \equiv -b \cdot S_m(c) \pmod{N}$. □


Theorem 5 (2.4). When $m$ is odd, $S_m(x)$ has no linear factors with rational coefficients other than $x$ and $x+1$.

Proof. When $m$ is odd, it suffices to show that $a/b = 0, -1$. Substituting $c = 1, -2$ into (2.3),

• When $c = 1$: $S_m(a - b) \equiv -b \pmod{N}$

• When $c = -2$: $S_m(a + 2b) \equiv -b \pmod{N}$

Since $-b$ and the divisor $N$ of $a - bc$ are coprime, letting $d \in \{a - b, a + 2b\}$, $d$ and $S_m(d)$ are coprime.

Here, when $d$ can be expressed as $d = d' p^2$ for some integer $d'$ and prime $p$, $$S_m(d) = S_m(d' p^2) \equiv d' p \cdot S_m(p) \equiv 0 \pmod{p}$$ so $d$ and $S_m(d)$ are not coprime.

Also, when $d$ can be expressed as $d = d' p$ for some integer $d'$ and odd prime $p$, from (1.3.1), $$S_m(d) = S_m(d' p) \equiv d' \cdot S_m(p) \equiv 0 \pmod{p}$$ so $d$ and $S_m(d)$ are not coprime.

Therefore, $d \in \{\pm 1, \pm 2\}$, and the $a$, $b$ satisfying this are only $(a, b) = (-1, 1), (0, 1)$. In this case, $a/b = 0, -1$, which indeed satisfy $S_m(a/b) = 0$. □


Theorem 6 (2.5). When $m$ is even, if $a/b \neq 0, -1$, then $b$ is even.

Proof. Substituting $c = 0, -1$ into (2.3),

• When $c = 0$: $S_m(a) \equiv 0 \pmod{N}$

• When $c = -1$: $S_m(a + b) \equiv 0 \pmod{N}$

Assume $b$ is odd. Then one of $a$, $a+b$ is even. Let the even one be $d = 2^j d'$ ($j \geq 1$, $d'$ is odd). Then, since $a/b \neq 0, -1$, we have $d, d' \neq 0$.

Since $S_m(d)$ is a multiple of $N$, $S_m(2^j d') \equiv 0 \pmod{2^j}$. Also, since $S_m(2^j d') \equiv d' S_m(2^j) \pmod{2^j}$, we have $S_m(2^j) \equiv 0 \pmod{2^j}$.

Here, $S_m(2^1) \equiv 1^m + 2^m \equiv 1 \pmod{2^1}$. Also, when $j \geq 2$, assume $S_m(2^{j-1}) \equiv 2^{j-2} \pmod{2^{j-1}}$.

Since $m$ is even, for integer $k$, $(k + 2^{j-1})^m \equiv k^m + m \cdot 2^{j-1} k^{m-1} \equiv k^m \pmod{2^j}$.

Therefore, $$\begin{align*} S_m(2^j) &= \sum_{k=1}^{2^{j-1}} k^m + \sum_{k=1}^{2^{j-1}} (k + 2^{j-1})^m \\ &\equiv \sum_{k=1}^{2^{j-1}} k^m + \sum_{k=1}^{2^{j-1}} k^m \\ &\equiv 2 S_m(2^{j-1}) \equiv 2^{j-1} \pmod{2^j} \end{align*}$$

Therefore, $S_m(2^j) \not\equiv 0 \pmod{2^j}$, which is a contradiction. □


Lemma 3 (3.1). For integer $l \geq 2$, there exist integers $a_1, a_2, \ldots, a_{l-1}$ such that $$S_{2l}(x) = \frac{1}{2l+1} \left\{ \frac{x^l(x+1)^l(2x+1)}{2} - \sum_{j=1}^{l-1} a_j S_{2j}(x) \right\} \quad \cdots(3.1)$$

Proof. $f(x) = x^l(x+1)^l(2x+1)$ can be expressed as follows for some integers $a'_1, a'_2, \ldots, a'_{2l-1}$: $$f(x) = 2x^{2l+1} + (2l+1)x^{2l} + \sum_{j=1}^{2l-1} a'_j x^j$$

Also, from $f(-x) = (-x)^l(-x+1)^l(-2x+1) = (-1)^{2l+1} x^l(x-1)^l(2x-1) = -f(x-1)$, $$\frac{f(x) - f(x-1)}{2} = \frac{f(x) + f(-x)}{2} = (2l+1)x^{2l} + \sum_{j=1}^{l-1} a'_{2j} x^{2j}$$

Taking the sum of both sides for $x = 1, 2, \ldots, n$, $$\frac{f(n)}{2} = (2l+1) S_{2l}(n) + \sum_{j=1}^{l-1} a'_{2j} S_{2j}(n)$$

Replacing $n$ with $x$ and setting $a_j = a'_{2j}$ and rearranging, $$S_{2l}(x) = \frac{1}{2l+1} \left\{ \frac{x^l(x+1)^l(2x+1)}{2} - \sum_{j=1}^{l-1} a_j S_{2j}(x) \right\}$$ □


Here, for $l \geq 1$, let integer $A_l > 0$ and integer coefficient polynomial $T_l(x) = \sum_{j=0}^{2l-2} C_{l,j} x^j$ (where $C_{l,j}$ are integers) satisfy $$S_{2l}(x) = \frac{x(x+1)(2x+1) T_l(x)}{A_l}$$

Here, $A_l$ is coprime to the greatest common divisor of each $C_{l,j}$.

Lemma 4 (3.3). For $l \geq 1$, $C_{l, 2l-2}$ is odd.

Proof. When a positive integer $N$ is divisible by 2 at most $k$ times, we denote $V_2(N) = k$.

When $l = 1$, $C_{1,0} = 1$ is odd. Also, $A_1 = 6$, so $V_2(A_1) = 1$.

When $l \geq 2$, assume $V_2(A_1), V_2(A_2), \ldots, V_2(A_{l-1}) \leq 1$. From (3.1), $$S_{2l}(x) = \frac{x(x+1)(2x+1)}{2l+1} \left\{ \frac{x^{l-1}(x+1)^{l-1}}{2} - \sum_{j=1}^{l-1} \frac{a_j T_j(x)}{A_j} \right\} \quad \cdots(3.1.1)$$

From (3.1.1), $A_l$ is a divisor of $(2l+1) \mathrm{lcm}(2, A_1, A_2, \ldots, A_{l-1})$, so $$V_2(A_l) \leq V_2(2l+1) + V_2(\mathrm{lcm}(2, A_1, A_2, \ldots, A_{l-1})) = 0 + 1 = 1$$

Therefore, $V_2(A_l) \leq 1$.

Also, from (3.1.1), the coefficient of the highest degree term of $S_{2l}(x)$ is $1/(2l+1)$, so $$\frac{1}{2l+1} = \frac{2C_{l,2l-2}}{A_l} \Longleftrightarrow A_l = 2(2l+1)C_{l,2l-2}$$

From this, $1 \leq V_2(A_l)$, so $$1 \leq V_2(A_l) = V_2(2(2l+1)C_{l,2l-2}) \leq 1 + 0 + V_2(C_{l,2l-2}) \leq 1 \Longrightarrow V_2(C_{l,2l-2}) = 0$$

Therefore, $C_{l,2l-2}$ is odd. □


Theorem 7 (3.4). For $l \geq 1$, $S_{2l}(x)$ has no linear factors with rational coefficients other than $x$, $x+1$, and $2x+1$.

Proof. We show that $a/b = 0, -1, -1/2$.

If $a/b \neq 0, -1, -1/2$, then $a/b$ is a root of $T_l(x)$. In this case, from Theorem 6 (2.5), $b$ is even, so the coefficient of the highest degree term of $T_l(x)$, $C_{l,2l-2}$, must be even. However, this contradicts Lemma 4 (3.3). □