$p(x)$ is a polynomial. Assume $p(x)\ne0$. How to prove that $(1+x)^np(x)$ has at least $n+1$ terms?
-
You mean $(1+x)^np(x)=\sum a_kx^k$ where at least $n+1$ of the $a_k$ are nonzero? – Hagen von Eitzen Oct 05 '15 at 22:21
-
Yes, this is exactly what I ask! – Will Turner Oct 05 '15 at 22:25
-
2Aside: This is only true over fields of characteristic $0$. Presumably, we are talking about polynomials with real or complex coefficients, so that is okay – Thomas Andrews Oct 05 '15 at 22:43
4 Answers
Hint: Induction with differentiation to get to smaller step.
- 177,126
-
-
-
-
2Which is $(np(x)+p'(x)(x+1))(x+1)^{n-1}$. So how many terms must it have, if we assume true for $n-1$? – Thomas Andrews Oct 05 '15 at 22:49
For a polynomial $f\in\mathbb C[X]$, $f(x)=\sum_ka_kX^k$ let $$w(f)=|\{\,k\in\Bbb N:a_k\ne 0\,\}|$$ denote the number of nonzero terms occurring in $f$ and let $$v(f)=\begin{cases}\max\{\,k\in\Bbb N:(1+X)^k\mid f\,\}&\text{if $f\ne 0$}\\\infty&\text{if $f=0$}\end{cases}$$ denote the multiplicity of $-1$ as a root of $f$. Observe that $$\tag1w(Xf)=w(f)$$ and that $$\tag2w(f')=\begin{cases} w(f)&\text{if $X\mid f$}\\w(f)-1&\text{otherwise.}\end{cases}$$ Also, $$\tag3v(Xf)=v(f) $$ and as the derivative of $(1+X)^kq(X)$ with $q(-1)\ne 0$ is $(1+X)^{k-1}r(X)$ with $r(X):=kq(X)+(1+X)q'(X)$ and $r(-1)=kq(-1)$ we have $$\tag4v(f')=v(f)-1\qquad\text{if $0<v(f)<\infty$} $$
The problem statement is equivalent to saying that $w(f)>v(f)$ for all nonzero $f$. This follows readily by induction, the base step being constant polynomials where $w(f)=1>0=v(f)$, and the induction step using $(1)$ and $(3)$ if $X\mid f$ and $(2)$ and $(4)$ if $X\nmid f$.
- 374,180
Let's notice first an equivalent statement: for $\alpha \ne 0$ and $p\not \equiv 0$, the polynomial
$$q(x)\colon = (x-\alpha)^n p(x)$$
has at least $n+1$ non-zero coefficients. It is more convenient to consider $\alpha = 1$.
Now, the polynomial $q(x)$ above has $1$ root with multiplicity at least $n$, so we have
$$q(1) = q'(1) = \cdots = q^{(n-1)}(1)=0$$
Let $q(x) = c_0 x^0+ c_1 x + \cdots c_k x^k$. The above system of equalities is equivalent to \begin{eqnarray} &c_0 + c_1 + c_2 + \cdots + c_N &= &0 \\ &0^1 c_0 + 1^1 c_1 + 2^1 c_2 + \cdots + k c_k&= &0 \\ \ldots \ldots \ldots \\ &0^{n-1} c_0 + 1^{n-1} c_1 + \cdots + k^{n-1} c_k &=& 0 \end{eqnarray}
Now let's assume that the non-zero coefficients of $q(x)$ are at position $m_1$, $m_2$, $\ldots$, $m_n$. From the above we get a system of linear equations for $c_{m_1}$, $\ldots$, $c_{m_n}$. The matrix of coefficients is a Vandermonde matrix corresponding to $(m_1, \ldots, m_n)$, so of non-zero determinant. But that implies that $m_1 = \ldots = m_n = 0$, contradiction.
- 53,909
If p(x) is non-zero then it contains at least one term in it. How many terms are in $(1+x)^n$? The binomial expansion will help here.
- 286
-
-
Yes that's true! But there may be cancellations for different p(x)s. – Will Turner Oct 05 '15 at 22:21
-
-
Yes! But what we concern here is the number of terms not the degree of the polynomial. – Will Turner Oct 05 '15 at 22:23
-
Poor wording on my follow up comment. If a term x^(n-j) from the binomial expansion multiplies with a term x^j from p(x) it may be able to cancel out x^n. But x^n is also going to be multiplied by x^j yielding an x^n+j term. The proof has to be done inductively, as stated below, but you don't need differentiation to do so. – Mark Oct 05 '15 at 22:26
-
I don't quite get what you mean. Is it possible for you to show me the proof? Thanks! – Will Turner Oct 05 '15 at 22:31
-
Is it generally true, then, that $p(x)q(x)$ has at least as many terms as $q(x)$ has? – Thomas Andrews Oct 05 '15 at 22:33
-
(Hint: My last question was rhetorical. It is not true - take $q(x)=1+x+x^2$ and $p(x)=(x-1)$. – Thomas Andrews Oct 05 '15 at 22:47