5

Define two polynomial sequences by $$ \cases{P_0(x)=x\\ Q_0(x)=1}\\ \cases{P_{n+1}(x)=P_n(x-1)Q_n(x+1)\\ Q_{n+1}(x)=P_n(x+1)Q_n(x-1)} $$ Is it true that the leading term of $ Q_n(x) - P_n(x) $ is $ 2^n \cdot (n-1)! \cdot x^{2^{n-1}-n} $ for all $ n \geq 1 $?

(From this post“the leading term of $g_n$ is $2^{n-1}(n-1)!x^{2^{n-1}-n}$.” I set $x_1=\dots=x_n=1$ to simplify the setting.)


Verification for $ n = 1 $ to $ 10 $.

var('x')
# P_0 and Q_0
P = [x]
Q = [1]

Compute P_n and Q_n for n=1 to 10

for n in range(10): P_next = P[n].subs(x=x-1) * Q[n].subs(x=x+1) Q_next = P[n].subs(x=x+1) * Q[n].subs(x=x-1) P.append(P_next) Q.append(Q_next)

for n in range(1, 11): diff = (Q[n] - P[n]).expand() LHS = diff.leading_coefficient(x) * x^diff.degree(x) RHS = 2^n * factorial(n-1) * x^(2^(n-1) - n) print(f"n={n}: {bool(LHS == RHS)}")

hbghlyj
  • 5,361
  • 3
    The roots of the polynomials have a strange looking pattern: https://i.sstatic.net/yrBFg1U0.png (I only calculated up to $n=9$ then got error, perhaps because the coefficinents become too large). – ploosu2 Jan 02 '25 at 15:10
  • @ploosu2 What about the roots of $P_n - Q_n$ ? – Gribouillis Jan 02 '25 at 15:19
  • They're similar. And appear to be symmetric w.r.t the imaginary axis. I use this code to draw the picture: R.<x> = QQ[] P, Q = x, 1 for n in range(1, 10+1): P, Q = P.subs(x=x-1)*Q.subs(x=x+1), P.subs(x=x+1)*Q.subs(x=x-1) show(points([(z.real(), z.imag()) for z,_ in (Q-P).roots(ring=CDF)])) – ploosu2 Jan 02 '25 at 15:52
  • I don't believe you have the degree of the leading term correct. From printing out (Q[n]-P[n]).expand() for $\ n=1\ $ to $\ 11\ $ in the final loop of your code, I'd surmise that the leading term is $\ 2^n\cdot(n-1)!,x^{2^n-n-1}\ .$ – lonza leggiera Jan 03 '25 at 01:26
  • @lonzaleggiera Corrected. Thanks! – hbghlyj Jan 03 '25 at 01:28
  • $Q_n - P_n$ looks like a "discrete" Wronskian, but I'm not sure how that helps. – Nolord Jan 03 '25 at 10:33

1 Answers1

4

By induction, we can write

$$P_n(x) = \prod_{i=0}^{\lfloor n/2\rfloor}(x - n + 4i)^{\binom{n}{2i}}, \quad Q_n(x) = \prod_{i=0}^{\lfloor(n-1)/2\rfloor} (x - n + 4i + 2)^{\binom{n}{2i + 1}}.$$

So for $n ≥ 1$, as $x → ∞$, we have

\begin{align*} Q_n(x) - P_n(x) &= \left(\frac{Q_n(x)}{P_n(x)} - 1\right)P_n(x) \\ &= \left(\exp\left(\ln Q_n(x) - \ln P_n(x)\right) - 1\right)P_n(x) \\ &= \left(\exp\left(-\sum_{i=0}^n (-1)^i \binom ni \ln (x-n+2i)\right) - 1\right)P_n(x) \\ &= \left(\exp\left(-\sum_{i=0}^n (-1)^i \binom ni \ln \left(1 + \left(\frac n2-i\right)⋅-\frac2x\right)\right) - 1\right)P_n(x) \\ &\sim -\ln^{(n)} 1⋅\left(-\frac2x\right)^n x^{2^{n-1}} \\ &= 2^n(n-1)!x^{2^{n-1} - n}. \end{align*}

Here we have used the facts that

$$\sum_{i=0}^n (-1)^i \binom ni \ln x = 0, \\ \sum_{i=0}^n (-1)^i \binom ni f(a + (b - i)h) \sim f^{(n)}(a)h^n \quad \text{as $h → 0$}, \\ \exp u - 1 \sim u \quad \text{as $u → 0$}, \\ P_n(x) → x^{2^{n-1}}. $$

  • Great answer! So the roots of $P$ and $Q$ are real (I had numerical errors calculating them). One thing, could you please explain how we go from line 3 to line 4 (how $x-n+2i$ changes to that, should there be $x$ multiplying the whole thing, and that gives then $- \ln x$) and then to line 5 (the asymptotics). – ploosu2 Jan 04 '25 at 07:13
  • 2
    @ploosu2 I’ve elaborated on that. – Anders Kaseorg Jan 04 '25 at 07:28