44

How can I prove, that a polynomial function $$f(x) = \sum_{0\le k \le n}a_k x^k\qquad n\in\mathbb N,\ a_k\in\mathbb C$$ is zero for at most $n$ different values of $x$, unless all $a_0,a_1,\ldots,a_n$ are zero?

FUZxxl
  • 9,499
  • 15
    Induction. If it has a root r, divide by (x - r). – Qiaochu Yuan Mar 08 '11 at 19:56
  • 2
    Isn't this the fundamental theorem of algebra? Wikipedia (http://en.wikipedia.org/wiki/Fundamental_theorem_of_algebra) gives several proofs. – Eivind Mar 08 '11 at 19:56
  • 2
    Hint: First show that $f(\alpha) = 0$ iff $x-\alpha$ divides $f(x)$ and then use induction on the degree of $f$ – kahen Mar 08 '11 at 19:57
  • Is there any background one needs to consider? For instance, is this some problem assigned in some course? If so, which course? – Aryabhata Mar 08 '11 at 20:01
  • @Moron: I'm not (yet) in university. This isn't homework. Just asking this as a part to proof my last question. I'm asking this because I didn't knew, that this is a fundamental theorem of algebra. – FUZxxl Mar 08 '11 at 20:08
  • @Fuz: Well technically what you ask in not fundamental theorem of algebra. What you ask is easier to prove. – Aryabhata Mar 08 '11 at 20:09
  • @Moron: It's a part of the fundamental theorem. Consider this question as answered. – FUZxxl Mar 08 '11 at 20:11
  • 4
    @Fuz: No, it is not "part" of fundamental theorem. It might be what you call a "Corollary", but it has other easier proofs. – Aryabhata Mar 08 '11 at 20:13
  • Also, I wasn't asking if this is homework. I was asking what the background is, which might help give a more relevant proof. – Aryabhata Mar 08 '11 at 20:19
  • @Moron: Okay. Thank you for this. – FUZxxl Mar 08 '11 at 20:24
  • Isn't this a duplicate? http://math.stackexchange.com/questions/7990/roots-of-a-polynomial-in-an-integral-domain/7994#7994 – Adrián Barquero Mar 08 '11 at 20:49
  • @Adrián: Not really sure, as perhaps one could give complex analysis proofs of this. The only reason I talked about fields was to try and point out the potential inaccuracy of the statement "It's a part of the fundamental theorem". – Aryabhata Mar 08 '11 at 20:54

7 Answers7

91

You don't need the fundamental theorem of algebra or the Vandermonde determinant, only the factor theorem.

Proposition: A polynomial of degree at most n with more than n roots vanishes identically.

Proof: By induction. The base case is $n=0$, which is obvious. Now take a polynomial f of degree at most n, and let $x_1,\ldots,x_{n+1}$ be distinct roots of f. By the factor theorem, we can write $$f(x) = (x-x_{n+1})g(x)$$ where g plainly has degree at most $n-1$. Now substitute $x = x_i$ for $i=1,\ldots,n$. For all these values of x the left hand side vanishes and the factor $(x_i-x_{n+1})$ is nonzero. Hence all these $x_i$ must be roots of g and by induction g is identically zero. QED

This same proof works over any field (or even integral domain).

Dan Petersen
  • 7,045
  • for polynomials $f,g\in K[x]$ there are $q,r\in K[x]$ such that $f=qg+r$ with $\text{deg}(r)<\text{deg}(g)$ or $r=0$ (polynomial division). if $f(\alpha)=0$, take $g(x)=x-\alpha$ to get $0=f(\alpha)=r$. Hence $r=0$ and $f(x)=q(x)(x-\alpha)$. – yoyo Mar 08 '11 at 21:18
  • How should we understand this when the field in question is finite? Say, if it has size $m$ (so that there can be at most $m$ distinct roots of any polynomial equation thereon), do we proceed by your inductive argument up to $n=m-1$, and then for $n = m,m+1,...$ the statement is true vacuously? – EE18 Sep 09 '23 at 15:42
  • Just to confirm whether I am correct: first we shall prove that if a deg $1$ polynomial has more two or more roots then it is identically $0$. After that we shall use this solution to complete the inductive step, correct? – prashant sharma Jul 16 '24 at 09:38
  • In the proof you used $n$ distinct roots but, in the statement, you didn't mention it. – prashant sharma Jul 16 '24 at 10:44
41

The following literature may be of use here:

Theorem. A polynomial $\text{f}$ of degree $\text{n}$ over a field $\text{F}$ has at most $\text{n}$ roots in $\text{F}$.*

Proof. The results is obviously true for polynomials of degree $0$ and degree $1$. We assume it to be true for polynomials of degree $n-1$. If $a$ is a root of $f$, $f=(x-a)q$ where $q$ has degree $n-1$. Since $f(b)=0$ if and only if $a=b$ or $q(b)=0$, it follows by our inductive assumption that $f$ has at most $n$ roots. $\Box$

Trancot
  • 4,127
  • 4
    Most elegant proof of the lot. – Noldorin Apr 30 '14 at 17:25
  • 1
    It's obvious that $f=(x-a)q$ if $a$ is a root, but how can that be demonstrated? – FUZxxl Feb 26 '16 at 16:57
  • 12
    Note that this proof uses the fact $F[x]$ is a UFD, so that the factorization $f(x) = (x-a)q(x)$ is unique. Without uniqueness the argument can fail. For instance, in $\mathbf{Z}/12\mathbf{Z}$ the polynomial $x^2 - 4$ has four roots $2, 4, 8, 10$ and two distinct factorizations $x^2 - 4 = (x - 8)(x - 4) = (x - 10)(x - 2)$. Note that the theorem holds if the field $F$ is replaced by any integral domain $R$ (see Dan Petersen's proof on this page). But I believe $R[x]$ need not be a UFD if $R$ is not a UFD. – Doug Jul 27 '16 at 00:32
  • @Doug This comment is confusing, Z/12Z is not a field as 12 is not prime, so what are you saying? – john Sep 25 '16 at 16:25
  • 4
    Hi john, the point I was making is that the proof by Trancot directly uses the fact $F$ is a field, where $F[x]$ then becomes a UFD. The result is false generally for rings that are not integral domains, e.g. $Z_{12}$. In this example, the proof by Trancot fails because $Z_{12}[x]$ is not a UFD. – Doug Sep 29 '16 at 19:43
  • 4
    @Doug: No, this proof does not use that $F\left[x\right]$ is a UFD. All roots of $f$ other than $a$ are roots of $q$ (as the proof points out), so induction yields that there are at most $n-1$ of them. We don't need to consider two different factorizations $f = \left(x-a\right) q = \left(x-b\right) r$ of $f$; after factoring out $x-a$ from $f$, we then factor out the next $x-b$ from $q$ (not from $f$). – darij grinberg Oct 25 '19 at 19:51
  • @symplectomorphic It is not correct that the degree of the zero polynomial is 0. The degree of the zero polynomial is either undefined, $-1$ or $-\infty$. The number of roots is the number of members of $\text{F}$. – Jeppe Stig Nielsen Apr 04 '20 at 09:35
  • 1
    @darijgrinberg To see what Doug means, take $F$ as the nonfield $F=Z/12Z$, consider the polynomial $f=x^2-4$ with coefficients in $Z/12Z$ and the root $a=2\in Z/12Z$. We get the factorization $$f=x^2-4=(x-2)(x-10)$$ in $Z/12Z$. In this case you cannot say that all roots of $f$ other than $a=2$ are roots of the quotient $q=x-10$. For example we have the root $4\in Z/12Z$, it is root in neither $x-a$ nor $q$. If we plug in $4$ in the factorization, we see the issue is the identity $2\cdot (-6) \equiv 0$, i.e. zero divisors in $F$. So proof uses nonexistence of nontrivial zero divisors instead. – Jeppe Stig Nielsen Apr 04 '20 at 09:53
  • @JeppeStigNielsen: Right. It's the "D" part of "UFD". But it's the easiest part to prove :) – darij grinberg Apr 04 '20 at 13:07
39

Using Linear Algebra,

If the $n+1$ distinct roots are $\alpha_i$, then we have that $x = [a_0, a_1, \dots, a_n]^{T}$ is a solution of $Ax = 0$ where $A$ is the Vandermonde matrix using the $\alpha_i$.

Since the Vandermonde matrix is invertible for distinct $\alpha_i$, it follows that $x = [0, 0, \dots, 0]$.

Thus if $a_j \neq 0$ for some $j$, then your polynomial can have at most $n$ different roots.

Note: This is basically saying that given a field $K$, any polynomial of degree $n$ in $K[x]$ has at most $n$ distinct roots.

Fundamental Theorem of Algebra is an assertion of the fact that $\mathbb{C}$ is algebraically closed, and the $K$ above need not be algebraically closed.

Aryabhata
  • 83,877
  • 1
    I just hope that Vandermonde determinant formula in itself does not use the theorem asked in question. – Pranav Bisht Apr 04 '17 at 17:09
  • 2
    @PranavBisht: No. There is an explicit formula for the determinant of the Vandermonde matrix. It would actually be an interesting proof that uses the theorem in question to prove that the Vandermonde matrix is invertible... – Aryabhata Apr 04 '17 at 20:34
  • Powerful proof... Works over integral domains as well. For Vandermonde determinant, see: https://qr.ae/psG9CS – Nothing special Sep 28 '24 at 11:57
5

Just a clarification here. The Fundamental Theorem of Algebra says that a polynomial of degree n will have exactly n roots (counting multiplicity). This is not the same as saying it has at most n roots. To get from "at most" to "exactly" you need a way to show that a polynomial of degree n has at least one root. Then you can proceed by induction.

There are lots of different kinds of proofs that a polynomial must have at least one root. None of them are totally trivial.

Betty Mock
  • 3,606
1

Suppose that the polynomial is $$f=\sum_{i=0}^na_iX^i \in \mathbb{C}\left[X\right] .$$ Assume that the polynomial function defined by the above polynomial has $n+1$ distinct roots, i.e., $f(x)=0$ for $n+1$ distinct values of x $\in \Bbb C$.
I shall use the theorem:$$\prod_{i=1}^k\left(X-x_i\right)q=f \Longleftrightarrow x_1,...,x_k \text{ are distinct roots of }f\text{, where }q\in\Bbb C[X] $$ Let $n+1$ distinct roots of $f$ are $x_1,...,x_{n+1}$. Hence, by above theorem, we can say that $ \prod_{i=1}^{n+1}\left(X-x_i\right)q=f$. Since the coefficient lie in the field of complex numbers, hence left hand side has degree $n+1$, while $f$ is of degree n. This is impossible, therefore $f$ can have at most n distinct roots.

1

Let $\alpha_1,\ldots,\alpha_{n+1}$ be distinct roots for $f(x)$ with ${\sf deg}(f)\leq n$. W.l.o.g, we can assume the leading coefficient of $f(x)$ is always $1$. This can be done by enumerating all the cases:

If ${\sf deg}(f)=n$, then $\alpha_1,\ldots,\alpha_n$ are roots implies $f(x)=(x-\alpha_1)\cdots(x-\alpha_n)$. But $f(\alpha_{n+1})\neq0$, so this does not work.

If ${\sf deg}(f)=n-1$, then $\alpha_1,\ldots,\alpha_{n-1}$ are roots implies $f(x)=(x-\alpha_1)\cdots(x-\alpha_{n-1})$. But $f(\alpha_n)\neq0$, so this does not work either.

$\vdots$

If ${\sf deg}(f)=1$, then $\alpha_1$ is a root implies $f(x)=x-\alpha_1$. But $f(\alpha_2)\neq0$, so this does not work either.

Hence this only case that would work is ${\sf deg}(f)=0$, and so $f(x)=0$.

Easy
  • 4,515
  • How do you prove that $f(x)=(x-\alpha_1)\cdots(x-\alpha_n)?$ – FUZxxl Nov 28 '16 at 15:55
  • @FUZxxl This is a consequence of factor theorem: $\alpha$ is a root of $f(x)$ iff $f(x)=(x-\alpha)g(x)$ where ${\sf deg}(g)={\sf deg}(f)-1$. And note that $\alpha_2$ is a root of $(x-\alpha_1)g(x)$ implies $\alpha_2$ is a root of $g(x)$... – Easy Nov 29 '16 at 01:58
0

Try Rolle's Theorem with induction.

For base case it works apparently.

Let set $S = \{x : f(x)=0\}$. For induction step, suppose that $|S|\leq n$ for $f(t)$ of degree $n$. Then we want to prove that $|S|\leq n+1$ for degree of $n+1$.

Suppose $|S| > n+1$ for degree of $n+1$, let $x_1 < x_2 <... < x_{n+1} < x_{n+2}$, where $x_i \in S$.

$p(x_1)=...=p(x_{n+2})=0$ By Rolle's Theorem we can find $c_i \in (x_i, x_{i+1})$ s.t. $f'(c_i)=0 \forall i$, where its derivative is just the case for degree $n$.

This implies that $|S|>n$ for the case of degree of $n$, which contradicts with our assumption.