4
$\begingroup$

It is well known that if linear recurrences $u_n$ and $v_n$ have characteristic polynomials $K_u$ and $K_v$, then $u_n + v_n$ has characteristic polynomial $K_u K_v$. Is the converse true? If my characteristic polynomial factors as $K_u K_v$, then can I find $u_n$ and $v_n$ such that my recurrence can be written as $u_n + v_n$?

$\endgroup$
6
  • $\begingroup$ Not quite. The sequence $\left(1, 2, 3, 4, \ldots\right)$ is recurrent with characteristic polynomial $\left(X-1\right)^2$, but it is not the sum of two recurrent sequences with characteristic polynomials $X-1$ (that is, two constant sequences). But you aren't too far away from the truth. Unless I am mistaken, the converse is true if you assume $K_u$ and $K_v$ to be coprime. You should probably be thinking of this problem in terms of linear algebra, or, better, abstract algebra: The vector space of all sequences ... $\endgroup$ Apr 25, 2017 at 7:40
  • $\begingroup$ ... is an $F\left[X\right]$-module (where $F$ is the ground field), on which $X$ acts as the shift operator (i.e., $X \left(a_1, a_2, a_3, \ldots\right) = \left(a_2, a_3, a_4, \ldots\right)$). For a given monic polynomial $P \in F\left[X\right]$, the sequences with characteristic polynomial $P$ are precisely the elements of this module that are annihilated by $P$. So you want to prove that an element annihilated by $PQ$ must be a sum of an element annihilated by $P$ and an element annihilated by $Q$, provided that $P$ and $Q$ are coprime. This is an easy consequence of the fact that ... $\endgroup$ Apr 25, 2017 at 7:43
  • $\begingroup$ ... there exist polynomials $U$ and $V$ satisfying $PU + QV = 1$ (since $P$ and $Q$ are coprime in the principal ideal domain $F\left[X\right]$). $\endgroup$ Apr 25, 2017 at 7:44
  • $\begingroup$ @darijgrinberg Thanks! that helps. Is there a different way to decompose a sequence in case of repeated roots? Or it is not possible at all to do so uniquely? For example, if I use Jordan decomposition to get the exponential polynomial solution for this recurrence, then can I decompose my recurrence in to simpler recurrences by using this? $\endgroup$ Apr 25, 2017 at 8:01
  • $\begingroup$ Well, if $P$ is irreducible, then you cannot decompose a recurrent sequence with characteristic polynomial $P^k$ (for $k \geq 0$) into sums of recurrent sequences with smaller characteristic polynomials (unless it happens to not use the whole $P^k$). $\endgroup$ Apr 25, 2017 at 8:07

1 Answer 1

2
$\begingroup$

For the sake of closure, let me expand my comments into an answer.

The answer is "no"; but it becomes a "yes" if you assume $K_{u}$ and $K_{v}$ to be coprime. Before I start proving it, let me introduce my own notation, which I believe is clearer than yours (specifically, I will avoid the use of the nebulous notion of a "linear recurrence" and the easily misunderstood word "characteristic polynomial").

Fix a field $F$. Set $\mathbb{N}=\left\{ 0,1,2,\ldots\right\} $. Let $F^{\infty}$ be the set $\left\{ \left( a_{0},a_{1},a_{2},\ldots\right) \ \mid\ a_{i}\in F\text{ for each }i\in\mathbb{N}\right\} $ of all infinite sequences of elements of $F$. This set $F^{\infty}$ is an $F$-vector space (where the operations are entrywise: e.g., we have $\left( a_{0},a_{1} ,a_{2},\ldots\right) +\left( b_{0},b_{1},b_{2},\ldots\right) =\left( a_{0}+b_{0},a_{1}+b_{1},a_{2}+b_{2},\ldots\right) $).

Definition. Let $P\in F\left[ X\right] $ be a polynomial. Write $P$ in the form $P=\sum\limits_{i=0}^{k}p_{i}X^{i}$ for some $k\in\mathbb{N}$ and some $p_{0},p_{1},\ldots,p_{k}\in F$.

Let $\mathbf{a}=\left( a_{0},a_{1},a_{2},\ldots\right) \in F^{\infty}$ be a sequence. Then, we say that the sequence $\mathbf{a}$ is $P$-linearly recursive if and only if each $n\in\mathbb{N}$ satisfies $\sum\limits_{i=0} ^{k}p_{i}a_{n+i}=0$. (Notice that this definition does not depend on how precisely we represent $P$ in the form $\sum\limits_{i=0}^{k}p_{i}X^{i}$; indeed, any two such representations differ only in zero addends, and these zero addends do not change the sum $\sum\limits_{i=0}^{k}p_{i}a_{n+i}$.)

(My notion of a "$P$-linearly recursive sequence" roughly corresponds to your "recursive sequence with characteristic polynomial $P$", but keep in mind that $P$ is not uniquely determined by the sequence. For example, each $X$-linearly recursive sequence is also $X^{2}$-linearly recursive.)

Theorem 1. Let $P$ and $Q$ be two coprime polynomials in the principal ideal domain $F\left[ X\right] $. Let $\mathbf{a}\in F^{\infty}$ be a $PQ$-linearly recursive sequence. Then, $\mathbf{a}$ is the sum of a $P$-linearly recursive sequence with a $Q$-linearly recursive sequence.

Before we prove Theorem 1, we introduce a helpful $F\left[ X\right] $-module structure on $F^{\infty}$. Namely, let $S$ be the $F$-linear map

$F^{\infty}\rightarrow F^{\infty},\ \left( a_{0},a_{1},a_{2},\ldots\right) \mapsto\left( a_{1},a_{2},a_{3},\ldots\right) $.

This map $S$ is called the shift operator (since it shifts a sequence by $1$ forward, dropping the very first entry). It is easy to see that

\begin{equation} S^{i}\left( a_{0},a_{1},a_{2},\ldots\right) =\left( a_{0+i} ,a_{1+i},a_{2+i},\ldots\right) \label{1} \tag{1} \end{equation}

for each $i\in\mathbb{N}$ and each $\left( a_{0},a_{1},a_{2},\ldots\right) \in F^{\infty}$.

Let $\operatorname*{End}\left( F^{\infty}\right) $ denote the $F$-algebra of all endomorphisms of the $F$-vector space $F^{\infty}$. By the universal property of the polynomial ring $F\left[ X\right] $, there exists a unique $F$-algebra homomorphism $\phi:F\left[ X\right] \rightarrow \operatorname*{End}\left( F^{\infty}\right) $ satisfying $\phi\left( X\right) =S$. Consider this $\phi$, and use it to make $F^{\infty}$ into an $F\left[ X\right] $-module. Thus, this $F\left[ X\right] $-module structure on $F^{\infty}$ satisfies $X\mathbf{a}=\underbrace{\phi\left( X\right) }_{=S}\mathbf{a}=S\mathbf{a}$ for each $\mathbf{a}\in F^{\infty}$. We can now easily describe how any polynomial acts on $F^{\infty}$:

Proposition 2. Let $P\in F\left[ X\right] $ be a polynomial. Write $P$ in the form $P=\sum\limits_{i=0}^{k}p_{i}X^{i}$ for some $k\in\mathbb{N}$ and some $p_{0},p_{1},\ldots,p_{k}\in F$.

Let $\mathbf{a}=\left( a_{0},a_{1},a_{2},\ldots\right) \in F^{\infty}$ be a sequence. Then, $P\mathbf{a}=\left( \sum\limits_{i=0}^{k}p_{i}a_{0+i} ,\sum\limits_{i=0}^{k}p_{i}a_{1+i},\sum\limits_{i=0}^{k}p_{i}a_{2+i} ,\ldots\right) $.

Proof of Proposition 2. We have $\mathbf{a}=\left( a_{0},a_{1},a_{2} ,\ldots\right) $. Hence, each $i\in\mathbb{N}$ satisfies \begin{equation} S^{i}\mathbf{a}=S^{i}\left( a_{0},a_{1},a_{2},\ldots\right) =\left( a_{0+i},a_{1+i},a_{2+i},\ldots\right) \label{2} \tag{2} \end{equation} (by \eqref{1}). On the other hand, applying the map $\phi$ to the equality $P=\sum\limits_{i=0}^{k}p_{i}X^{i}$, we obtain

$\phi\left( P\right) =\phi\left( \sum\limits_{i=0}^{k}p_{i}X^{i}\right) =\sum\limits_{i=0}^{k}p_{i}\phi\left( X\right) ^{i}$ (since $\phi$ is an $F$-algebra homomorphism)

$=\sum\limits_{i=0}^{k}p_{i}S^{i}$ (since $\phi\left( X\right) =S$).

But the definition of the $F\left[ X\right] $-module structure on $F^{\infty}$ shows that

$P\mathbf{a}=\underbrace{\phi\left( P\right) }_{=\sum\limits_{i=0}^{k} p_{i}S^{i}}\mathbf{a}=\sum\limits_{i=0}^{k}p_{i}\underbrace{S^{i}\mathbf{a} }_{\substack{=\left( a_{0+i},a_{1+i},a_{2+i},\ldots\right) \\\text{(by \eqref{2})}}}$

$=\sum\limits_{i=0}^{k}p_{i}\left( a_{0+i},a_{1+i},a_{2+i},\ldots\right) =\left( \sum\limits_{i=0}^{k}p_{i}a_{0+i},\sum\limits_{i=0}^{k}p_{i} a_{1+i},\sum\limits_{i=0}^{k}p_{i}a_{2+i},\ldots\right) $.

This proves Proposition 2.

Corollary 3. Let $P\in F\left[ X\right] $ be a polynomial. Let $\mathbf{a}\in F^{\infty}$ be a sequence. Then, $\mathbf{a}$ is $P$-linearly recursive if and only if $P\mathbf{a}=0$.

Proof of Corollary 3. Write $P$ in the form $P=\sum\limits_{i=0}^{k} p_{i}X^{i}$ for some $k\in\mathbb{N}$ and some $p_{0},p_{1},\ldots,p_{k}\in F$. Write the sequence $\mathbf{a}\in F^{\infty}$ in the form $\mathbf{a} =\left( a_{0},a_{1},a_{2},\ldots\right) $. Thus, Proposition 2 yields $P\mathbf{a}=\left( \sum\limits_{i=0}^{k}p_{i}a_{0+i},\sum\limits_{i=0} ^{k}p_{i}a_{1+i},\sum\limits_{i=0}^{k}p_{i}a_{2+i},\ldots\right) $. Hence, we have the following chain of equivalences:

$\left( P\mathbf{a}=0\right) $

$\Longleftrightarrow\ \left( \left( \sum\limits_{i=0}^{k}p_{i}a_{0+i} ,\sum\limits_{i=0}^{k}p_{i}a_{1+i},\sum\limits_{i=0}^{k}p_{i}a_{2+i} ,\ldots\right) =0\right) $

$\Longleftrightarrow\ \left( \text{each }n\in\mathbb{N}\text{ satisfies } \sum\limits_{i=0}^{k}p_{i}a_{n+i}=0\right) $

$\Longleftrightarrow\ \left( \mathbf{a}\text{ is }P\text{-linearly recursive}\right) $

(by the definition of "$P$-linearly recursive"). This proves Corollary 3.

Proof of Theorem 1. Since $F\left[ X\right] $ is a principal ideal domain, Bezout's identity shows that there are two polynomials $U\in F\left[ X\right] $ and $V\in F\left[ X\right] $ such that $PU+QV=\gcd\left( P,Q\right) $. Consider these $U$ and $V$. Thus, $PU+QV=\gcd\left( P,Q\right) =1$ (since $P$ and $Q$ are coprime).

Corollary 3 (applied to $PQ$ instead of $P$) shows that $\mathbf{a}$ is $PQ$-linearly recursive if and only if $PQ\mathbf{a}=0$. Hence, $PQ\mathbf{a} =0$ (since $\mathbf{a}$ is $PQ$-linearly recursive).

The sequence $PU\mathbf{a}\in F^{\infty}$ satisfies $Q\left( PU\mathbf{a} \right) =\underbrace{QPU}_{=UPQ}\mathbf{a}=U\underbrace{PQ\mathbf{a}}_{=0} =0$. But Corollary 3 (applied to $Q$ and $PU\mathbf{a}$ instead of $P$ and $\mathbf{a}$) shows that $PU\mathbf{a}$ is $Q$-linearly recursive if and only if $Q\left( PU\mathbf{a}\right) =0$. Thus, $PU\mathbf{a}$ is $Q$-linearly recursive (since $Q\left( PU\mathbf{a}\right) =0$).

The sequence $QV\mathbf{a}\in F^{\infty}$ satisfies $P\left( QV\mathbf{a} \right) =\underbrace{PQV}_{=VPQ}\mathbf{a}=V\underbrace{PQ\mathbf{a}}_{=0} =0$. But Corollary 3 (applied to $QV\mathbf{a}$ instead of $\mathbf{a}$) shows that $QV\mathbf{a}$ is $P$-linearly recursive if and only if $P\left( QV\mathbf{a}\right) =0$. Thus, $QV\mathbf{a}$ is $P$-linearly recursive (since $P\left( QV\mathbf{a}\right) =0$).

Now, $PU\mathbf{a}+QV\mathbf{a}=\underbrace{\left( PU+QV\right) } _{=1}\mathbf{a}=\mathbf{a}$. Hence, $\mathbf{a}=PU\mathbf{a}+QV\mathbf{a} =QV\mathbf{a}+PU\mathbf{a}$ is the sum of a $P$-linearly recursive sequence (namely, $QV\mathbf{a}$) with a $Q$-linearly recursive sequence (namely, $PU\mathbf{a}$). This proves Theorem 1.

Notice that Theorem 1 gives an alternative approach to explicit formulas for linearly recursive sequences (like the Binet formula for Fibonacci numbers). Indeed, repeated application of Theorem 1 yields the following corollary:

Corollary 4. Let $P\in F\left[ X\right] $ be a monic polynomial. Let $P=\prod\limits_{i=1}^{k}P_{i}^{m_{i}}$ be the factorization of $P$ into monic irreducible polynomials (with $P_{1},P_{2},\ldots,P_{k}$ being distinct monic irreducible polynomials, and $m_{1},m_{2},\ldots,m_{k}$ being nonnegative integers). Let $\mathbf{a}\in F^{\infty}$ be a $P$-linearly recursive sequence. Then, $\mathbf{a}$ can be written in the form $\mathbf{a} =\mathbf{a}_{1}+\mathbf{a}_{2}+\cdots+\mathbf{a}_{k}$, where each $\mathbf{a}_{i}$ is a $P_{i}^{m_{i}}$-linearly recursive sequence.

For example, if $F=\mathbb{C}$ and $P=X^{2}-X-1$, then we can apply Corollary 4 to any $P$-linearly recursive sequence (setting $k=2$, $P_{1}=X-\dfrac {1+\sqrt{5}}{2}$, $m_{1}=1$, $P_{2}=X-\dfrac{1-\sqrt{5}}{2}$, and $m_{2}=1$). We thus conclude that each $\left( X^{2}-X-1\right) $-linearly recursive sequence can be written as the sum of an $\left( X-\dfrac{1+\sqrt{5}} {2}\right) $-linearly recursive sequence (i.e., a geometric sequence with ratio $\dfrac{1+\sqrt{5}}{2}$) with an $\left( X-\dfrac{1-\sqrt{5}} {2}\right) $-linearly recursive sequence (i.e., a geometric sequence with ratio $\dfrac{1-\sqrt{5}}{2}$). This lets you easily prove the Binet formula for Fibonacci sequence. In general, you might have to work harder (particularly if $P$ is not squarefree, and so some of the $m_{i}$ are $>1$).

Notice also that Theorem 1 does not hold if we forget to require that $P$ and $Q$ be coprime. For example, the sequence $\left( 0,1,2,\ldots\right) $ is always $\left( X-1\right) ^{2}$-linearly recursive, but cannot be written as a sum of two $\left( X-1\right) $-linearly recursive sequences. (In fact, the $\left( X-1\right) ^{2}$-linearly recursive sequences are the arithmetic sequences, while the $\left( X-1\right) $-linearly recursive sequences are the constant sequences.) There is a counterexample for every pair $\left( P,Q\right) $ of non-coprime polynomials $P$ and $Q$ (over any field $F$).

$\endgroup$
1
  • $\begingroup$ Thanks for an amazing detailed answer! $\endgroup$ May 2, 2017 at 8:22

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .