6

The problem statement is "Let $F \in \mathbb{C}[t]$ have degree at most $D \geq 1$, and let $G \in \mathbb{C}[t]$ have degree $E \geq 1$.

Show that there is a $P \not = 0$ in $\mathbb{C}[X,Y]$ with degree at most $E$ in $X$ and $D$ in $Y$ such that $P(F,G) = 0$."

I've tried to work this out by supposing $$P = \sum_{i=0}^{E} \sum_{j=0}^{D}c_{ij}X^iY^j$$, and noticing that I have control over the choice of $(1+E)(1+D)$ coefficients. I was hoping to be able to use this information to create a homogeneous system of linear equations to give a non-trivial solution for the $c_{ij}$ that would force $P(F,G) = 0$.

Is this a viable approach? If so, what would my next step be? I asked my professor, and the hint he gave me was to think about the resultant of $F$ and $G$. I know how to construct the matrix whose determinant is the resultant of $F$ and $G$, and I know the resultant is $0$ if $F$ and $G$ share a common factor, but I don't know how that helps us with this problem

Thanks in advance for any comments, hints, or solutions!

user413766
  • 1,216
  • 8
  • 17
  • 1
    Your approach, I fear, only gives weaker bounds (degrees at most $2E$ and $2D$, not $E$ and $D$). – darij grinberg Sep 30 '18 at 21:52
  • 5
    See https://mathoverflow.net/a/189344/ for the proof using resultants. The idea is to take the resultant of the two polynomials $F\left(T\right) - X$ and $G\left(T\right) - Y$ in the indeterminate $T$ over the ring $\mathbb{C}\left[X,Y\right]$. This resultant is nonzero as a polynomial, but will become $0$ when $F\left(U\right)$ and $G\left(U\right)$ (with $U$ being yet another indeterminate) are substituted for $X$ and $Y$, since then the two polynomials will have the common root $U = T$. – darij grinberg Sep 30 '18 at 21:58
  • Would you like to post this as an answer so you can receive the bounty on the problem? This answered everything I had in mind. – user413766 Oct 08 '18 at 00:19

2 Answers2

9

Here is a proof using resultants, taken mostly from https://mathoverflow.net/questions/189181//189344#189344 . (For a short summary, see one of my comments to the OP.)

$\newcommand{\KK}{\mathbb{K}} \newcommand{\LL}{\mathbb{L}} \newcommand{\NN}{\mathbb{N}} \newcommand{\ww}{\mathbf{w}} \newcommand{\eps}{\varepsilon} \newcommand{\Res}{\operatorname{Res}} \newcommand{\Syl}{\operatorname{Syl}} \newcommand{\adj}{\operatorname{adj}} \newcommand{\id}{\operatorname{id}} \newcommand{\tilF}{\widetilde{F}} \newcommand{\tilG}{\widetilde{G}} \newcommand{\tilKK}{\widetilde{\KK}} \newcommand{\ive}[1]{\left[ #1 \right]} \newcommand{\tup}[1]{\left( #1 \right)} \newcommand{\zeroes}[1]{\underbrace{0,0,\ldots,0}_{#1 \text{ zeroes}}}$ We shall prove a more general statement:

Theorem 1. Let $\KK$ be a nontrivial commutative ring. Let $F$ and $G$ be two polynomials in the polynomial ring $\KK \ive{T}$. Let $d$ and $e$ be nonnegative integers such that $d+e > 0$ and $\deg F \leq d$ and $\deg G \leq e$. Then, there exists a nonzero polynomial $P\in\KK \ive{X, Y}$ in two indeterminates $X$ and $Y$ such that $\deg_X P\leq e$ and $\deg_Y P\leq d$ and $P\tup{F, G} =0$.

Here and in the following, we are using the following notations:

  • "Ring" always means "associative ring with unity".

  • A ring $R$ is said to be nontrivial if $0 \neq 1$ in $R$.

  • If $R$ is any polynomial in the polynomial ring $\KK \ive{X, Y}$, then $\deg_X R$ denotes the degree of $R$ with respect to the variable $X$ (that is, it denotes the degree of $R$ when $R$ is considered as a polynomial in $\tup{\KK \ive{Y}} \ive{X}$), whereas $\deg_Y R$ denotes the degree of the polynomial $R$ with respect to the variable $Y$.

To prove Theorem 1, we recall the notion of the resultant of two polynomials over a commutative ring:

Definition. Let $\KK$ be a commutative ring. Let $P\in \KK \ive{T}$ and $Q\in\KK \ive{T}$ be two polynomials in the polynomial ring $\KK \ive{T}$. Let $d\in\NN$ and $e\in\NN$ be such that $\deg P\leq d$ and $\deg Q\leq e$. Thus, write the polynomials $P$ and $Q$ in the forms \begin{align*} P & =p_0 +p_1 T+p_2 T^2 +\cdots+p_d T^d \qquad\text{and}\\ Q & =q_0 +q_1 T+q_2 T^2 +\cdots+q_e T^e , \end{align*} where $p_0 ,p_1 ,\ldots,p_d ,q_0 ,q_1 ,\ldots,q_e $ belong to $\KK$. Then, we let $\Syl_{d,e} \tup{P, Q}$ be the matrix \begin{align} \left( \begin{array}[c]{c} \begin{array}[c]{ccccccccc} p_0 & 0 & 0 & \cdots & 0 & q_0 & 0 & \cdots & 0\\ p_1 & p_0 & 0 & \cdots & 0 & q_1 & q_0 & \cdots & 0\\ \vdots & p_1 & p_0 & \cdots & 0 & \vdots & q_1 & \ddots & \vdots\\ \vdots & \vdots & p_1 & \ddots & \vdots & \vdots & \vdots & \ddots & q_0 \\ p_d & \vdots & \vdots & \ddots & p_0 & \vdots & \vdots & \ddots & q_1 \\ 0 & p_d & \vdots & \ddots & p_1 & q_e & \vdots & \ddots & \vdots\\ \vdots & \vdots & \ddots & \ddots & \vdots & 0 & q_e & \ddots & \vdots\\ 0 & 0 & 0 & \ddots & \vdots & \vdots & \vdots & \ddots & \vdots\\ 0 & 0 & 0 & \cdots & p_d & 0 & 0 & \cdots & q_e \end{array} \\ \underbrace{\ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ }_{e\text{ columns}} \underbrace{\ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ }_{d\text{ columns}} \end{array} \right) \in\KK^{\tup{d+e} \times\tup{d+e}}; \end{align} this is the $\tup{d+e} \times\tup{d+e}$-matrix whose first $e$ columns have the form \begin{align} \left( \zeroes{k},p_0 ,p_1 ,\ldots ,p_d ,\zeroes{e-1-k}\right) ^{T} \qquad\text{for }k\in\left\{ 0,1,\ldots,e-1\right\} , \end{align} and whose last $d$ columns have the form \begin{align} \left( \zeroes{\ell},q_0 ,q_1 ,\ldots,q_e ,\zeroes{d-1-\ell}\right) ^{T} \qquad\text{for }\ell\in\left\{ 0,1,\ldots,d-1\right\} . \end{align} Furthermore, we define $\Res_{d,e}\tup{P, Q}$ to be the element \begin{align} \det \tup{ \Syl_{d,e}\tup{P, Q} } \in \KK . \end{align} The matrix $\Syl_{d,e}\tup{P, Q}$ is called the Sylvester matrix of $P$ and $Q$ in degrees $d$ and $e$. Its determinant $\Res_{d,e}\tup{P, Q}$ is called the resultant of $P$ and $Q$ in degrees $d$ and $e$.

It is common to apply this definition to the case when $d=\deg P$ and $e=\deg Q$; in this case, we simply call $\Res_{d,e}\tup{P, Q}$ the resultant of $P$ and $Q$, and denote it by $\Res \tup{P, Q}$.

Here, we take $\NN$ to mean the set $\left\{0,1,2,\ldots\right\}$ of all nonnegative integers.

One of the main properties of resultants is the following:

Theorem 2. Let $\KK$ be a commutative ring. Let $P\in \KK \ive{T}$ and $Q\in\KK \ive{T}$ be two polynomials in the polynomial ring $\KK \ive{T}$. Let $d\in\NN$ and $e\in\NN$ be such that $d+e > 0$ and $\deg P\leq d$ and $\deg Q\leq e$. Let $\LL$ be a commutative $\KK$-algebra, and let $w\in\LL$ satisfy $P\tup{w} =0$ and $Q\tup{w} = 0$. Then, $\Res_{d,e}\tup{P, Q} =0$ in $\LL$.

Proof of Theorem 2 (sketched). Recall that $\Res_{d,e}\tup{P, Q} =\det \tup{ \Syl_{d,e}\tup{P, Q} }$ (by the definition of $\Res_{d,e}\tup{P, Q}$).

Write the polynomials $P$ and $Q$ in the forms \begin{align*} P & =p_0 +p_1 T+p_2 T^2 +\cdots+p_d T^d \qquad\text{and}\\ Q & =q_0 +q_1 T+q_2 T^2 +\cdots+q_e T^e , \end{align*} where $p_0 ,p_1 ,\ldots,p_d ,q_0 ,q_1 ,\ldots,q_e $ belong to $\KK$. (We can do this, since $\deg P \leq d$ and $\deg Q \leq e$.) From $p_0 +p_1 T+p_2 T^2 +\cdots+p_d T^d = P$, we obtain $p_0 + p_1 w + p_2 w^2 + \cdots + p_d w^d = P\left(w\right) = 0$. Similarly, $q_0 + q_1 w + q_2 w^2 + \cdots + q_e w^e = 0$.

Let $A$ be the matrix $\Syl_{d,e}\tup{P, Q} \in\KK^{\tup{d+e} \times\tup{d+e} }$, regarded as a matrix in $\LL ^{\tup{d+e} \times\tup{d+e} }$ (by applying the canonical $\KK$-algebra homomorphism $\KK \rightarrow\LL$ to all its entries).

Let $\ww$ be the row vector $\left( w^{0},w^{1},\ldots,w^{d+e-1} \right) \in\LL ^{1\times\tup{d+e} }$. Let $\mathbf{0}$ denote the zero vector in $\LL ^{1\times\tup{d+e} }$.

Now, it is easy to see that $\ww A=\mathbf{0}$. (Indeed, for each $k\in\left\{ 1,2,\ldots,d+e\right\} $, we have \begin{align*} & \ww\left( \text{the }k\text{-th column of }A\right) \\ & = \begin{cases} p_0 w^{k-1}+p_1 w^k +p_2 w^{k+1}+\cdots+p_d w^{k-1+d}, & \text{if }k\leq e;\\ q_0 w^{k-e-1}+q_1 w^{k-e}+q_2 w^{k-e+1}+\cdots+q_e w^{k-1}, & \text{if }k>e \end{cases} \\ & = \begin{cases} w^{k-1}\left( p_0 +p_1 w+p_2 w^2 +\cdots+p_d w^d \right) , & \text{if }k\leq e;\\ w^{k-e-1}\left( q_0 +q_1 w+q_2 w^2 +\cdots+q_e w^e\right) , & \text{if }k>e \end{cases} \\ & = \begin{cases} w^{k-1}0, & \text{if }k\leq e;\\ w^{k-e-1}0, & \text{if }k>e \end{cases} \\ & \qquad\left( \begin{array}[c]{c} \text{since }p_0 +p_1 w+p_2 w^2 +\cdots+p_d w^d =0\\ \text{and }q_0 +q_1 w+q_2 w^2 +\cdots+q_e w^e =0 \end{array} \right) \\ & =0. \end{align*} But this means precisely that $\ww A=\mathbf{0}$.)

But $A$ is a square matrix over a commutative ring; thus, the adjugate $\adj A$ of $A$ satisfies $A\cdot\adj A=\det A\cdot I_{d+e}$ (where $I_{d+e}$ denotes the identity matrix of size $d+e$). Hence, $\ww\underbrace{A\cdot\adj A}_{=\det A\cdot I_{d+e}}=\ww\det A\cdot I_{d+e}=\det A\cdot\ww$. Comparing this with $\underbrace{\ww A}_{=\mathbf{0}}\cdot\adj A =\mathbf{0}\cdot\adj A=\mathbf{0}$, we obtain $\det A\cdot\ww=\mathbf{0}$.

But $d+e > 0$; thus, the row vector $\ww$ has a well-defined first entry. This first entry is $w^0 = 1$. Hence, the first entry of the row vector $\det A\cdot\ww$ is $\det A \cdot 1 = \det A$. Hence, from $\det A\cdot\ww=\mathbf{0}$, we conclude that $\det A=0$. Comparing this with \begin{align} \det\underbrace{A}_{=\Syl_{d,e}\tup{P, Q}} =\det \tup{ \Syl_{d,e}\tup{P, Q} } =\Res_{d,e}\tup{P, Q} , \end{align} we obtain $\Res_{d,e}\tup{P, Q} =0$ (in $\LL$). This proves Theorem 2. $\blacksquare$

Theorem 2 (which I have proven in detail to stress how the proof uses nothing about $\LL$ other than its commutativity) was just the meek tip of the resultant iceberg. Here are some further sources with deeper results:

Some of these sources use the matrix $\tup{\Syl_{d,e}\tup{P, Q}}^T$ instead of our $\Syl_{d,e}\tup{P, Q}$, but of course this matrix has the same determinant as $\Syl_{d,e}\tup{P, Q}$, so that their definition of a resultant is the same as mine. Some other among these sources use a different matrix, in which the coefficients $p_0, p_1, \ldots, p_d$ occur in the opposite order (i.e., from bottom to top rather than from top to bottom) and similarly for the coefficients $q_0, q_1, \ldots, q_e$; this leads to a notion of resultant that differs from mine in sign (because their matrix is obtained from mine by a permutation of rows and columns).

We are not yet ready to prove Theorem 1 directly. Instead, let us prove a weaker version of Theorem 1:

Lemma 3. Let $\KK$ be a commutative ring. Let $F$ and $G$ be two polynomials in the polynomial ring $\KK \ive{T}$. Let $d$ and $e$ be nonnegative integers such that $d+e > 0$ and $\deg F \leq d$ and $\deg G \leq e$. Write the polynomial $G$ in the form $G=g_0 +g_1 T+g_2 T^2 +\cdots +g_e T^e $, where $g_0 ,g_1 ,\ldots,g_e \in\KK$. Assume that $g_e^d \neq 0$. Then, there exists a nonzero polynomial $P\in\KK \ive{X, Y}$ in two indeterminates $X$ and $Y$ such that $\deg_X P\leq e$ and $\deg_Y P\leq d$ and $P\tup{F, G} =0$.

Proof of Lemma 3 (sketched). Let $\tilKK$ be the commutative ring $\KK \ive{X, Y}$. Define two polynomials $\tilF \in\tilKK\ive{T}$ and $\tilG \in\tilKK\ive{T}$ by \begin{align} \tilF = F-X = F\tup{T}-X \qquad\text{and}\qquad \tilG = G-Y = G\tup{T}-Y. \end{align} Note that $X$ and $Y$ have degree $0$ when considered as polynomials in $\tilKK\ive{T}$ (since $X$ and $Y$ belong to the ring $\tilKK$). Thus, these new polynomials $\tilF = F - X$ and $\tilG = G - Y$ have degrees $\deg\tilF \leq d$ (because $\deg X = 0 \leq d$ and $\deg F \leq d$) and $\deg\tilG \leq e$ (similarly). Hence, the resultant $\Res_{d,e}\tup{\tilF, \tilG} \in \tilKK$ of these polynomials $\tilF$ and $\tilG$ in degrees $d$ and $e$ is well-defined. Let us denote this resultant $\Res_{d,e}\tup{\tilF, \tilG}$ by $P$. Hence, \begin{align} P = \Res_{d,e}{\tilF, \tilG} \in \tilKK =\KK \ive{X, Y} . \end{align}

Our next goal is to show that $P$ is a nonzero polynomial and satisfies $\deg_X P\leq e$ and $\deg_Y P\leq d$ and $P\tup{F, G} =0$. Once this is shown, Lemma 3 will obviously follow.

We have \begin{align} P = \Res_{d,e} \tup{\tilF, \tilG} = \det\tup{\Syl_{d,e}\tup{\tilF, \tilG}} \label{darij1.pf.t1.P=det} \tag{1} \end{align} (by the definition of $\Res_{d,e}\tup{\tilF, \tilG}$).

Write the polynomial $F$ in the form $F=f_0 +f_1 T+f_2 T^2 +\cdots +f_d T^d $, where $f_0 ,f_1 ,\ldots,f_d \in\KK$. (This can be done, since $\deg F \leq d$.)

Recall that $g_e ^d \neq 0$. Thus, $\left( -1\right) ^e g_e^d \neq 0$.

For each $p\in\NN$, we let $S_{p}$ be the group of all permutations of the set $\left\{ 1,2,\ldots,p\right\} $.

Now, \begin{align*} \tilF & =F-X=\left( f_0 +f_1 T+f_2 T^2 +\cdots+f_d T^d \right) -X\\ & \qquad\left( \text{since }F=f_0 +f_1 T+f_2 T^2 +\cdots+f_d T^d \right) \\ & =\tup{f_0 - X} +f_1 T+f_2 T^2 +\cdots+f_d T^d . \end{align*} Thus, $f_0 -X,f_1 ,f_2 ,\ldots,f_d $ are the coefficients of the polynomial $\tilF \in\tilKK\ive{T}$ (since $f_0 -X\in\tilKK$). Similarly, $g_0 -Y,g_1 ,g_2 ,\ldots,g_e $ are the coefficients of the polynomial $\tilG \in\tilKK\ive{T}$. Hence, the definition of the matrix $\Syl_{d,e}\left( \tilF ,\tilG \right)$ yields \begin{align} &\Syl_{d,e}\tup{\tilF, \tilG} \\ &=\left( \begin{array}[c]{c} \begin{array}[c]{ccccccccc} f_0 -X & 0 & 0 & \cdots & 0 & g_0 -Y & 0 & \cdots & 0\\ f_1 & f_0 -X & 0 & \cdots & 0 & g_1 & g_0 -Y & \cdots & 0\\ \vdots & f_1 & f_0-X & \cdots & 0 & \vdots & g_1 & \ddots & \vdots\\ \vdots & \vdots & f_1 & \ddots & \vdots & \vdots & \vdots & \ddots & g_0 -Y\\ f_d & \vdots & \vdots & \ddots & f_0 -X & \vdots & \vdots & \ddots & g_1 \\ 0 & f_d & \vdots & \ddots & f_1 & g_e & \vdots & \ddots & \vdots\\ \vdots & \vdots & \ddots & \ddots & \vdots & 0 & g_e & \ddots & \vdots\\ 0 & 0 & 0 & \ddots & \vdots & \vdots & \vdots & \ddots & \vdots\\ 0 & 0 & 0 & \cdots & f_d & 0 & 0 & \cdots & g_e \end{array} \\ \underbrace{\qquad \qquad \qquad \qquad \qquad \qquad \qquad \qquad \qquad \qquad} _{e\text{ columns}} \underbrace{\qquad \qquad \qquad \qquad \qquad \qquad \qquad} _{d\text{ columns}} \end{array} \right) \\ &\in\tilKK^{\tup{d+e} \times\left( d+e\right) }. \end{align} Now, let us use this explicit form of $\Syl_{d,e} \tup{\tilF, \tilG}$ to compute $\det\left( \Syl_{d,e}\tup{\tilF, \tilG} \right)$ using the Leibniz formula. The Leibniz formula yields \begin{align} \det\left( \Syl_{d,e} \tup{ \tilF, \tilG } \right) =\sum_{\sigma\in S_{d+e}}a_{\sigma}, \label{darij1.pf.t1.det=sum} \tag{2} \end{align} where for each permutation $\sigma\in S_{d+e}$, the addend $a_{\sigma}$ is a product of entries of $\Syl_{d,e} \tup{\tilF, \tilG}$, possibly with a minus sign. More precisely, \begin{align} a_{\sigma}=\tup{-1}^{\sigma}\prod_{i=1}^{d+e}\left( \text{the } \left( i,\sigma\left( i\right) \right) \text{-th entry of } \Syl_{d,e}\tup{\tilF, \tilG} \right) \end{align} for each $\sigma\in S_{d+e}$ (where $\tup{-1}^{\sigma}$ denotes the sign of the permutation $\sigma$).

Now, \eqref{darij1.pf.t1.P=det} becomes \begin{align} P=\det\left( \Syl_{d,e}\tup{\tilF, \tilG} \right) =\sum_{\sigma\in S_{d+e}}a_{\sigma} \label{darij1.pf.t1.P=sum} \tag{3} \end{align} (by \eqref{darij1.pf.t1.det=sum}).

All entries of the matrix $\Syl_{d,e}\tup{\tilF, \tilG}$ are polynomials in the two indeterminates $X$ and $Y$; but only $d+e$ of these entries are non-constant polynomials (since all of $f_0 ,f_1 ,\ldots,f_d ,g_0 ,g_1 ,\ldots,g_e $ belong to $\KK$). More precisely, only $e$ entries of $\Syl_{d,e}\tup{\tilF, \tilG}$ have non-zero degree with respect to the variable $X$ (namely, the first $e$ entries of the diagonal of $\Syl_{d,e}\tup{\tilF, \tilG}$), and these $e$ entries have degree $1$ with respect to this variable. Thus, for each $\sigma\in S_{d+e}$, the product $a_{\sigma}$ contains at most $e$ many factors that have degree $1$ with respect to the variable $X$, while all its remaining factors have degree $0$ with respect to this variable. Therefore, for each $\sigma\in S_{d+e}$, the product $a_{\sigma}$ has degree $\leq e\cdot1=e$ with respect to the variable $X$. Hence, the sum $\sum_{\sigma\in S_{d+e}}a_{\sigma}$ of all these products $a_{\sigma}$ also has degree $\leq e$ with respect to the variable $X$. In other words, $\deg_X \left( \sum_{\sigma\in S_{d+e}}a_{\sigma}\right) \leq e$. In view of \eqref{darij1.pf.t1.P=sum}, this rewrites as $\deg_X P\leq e$. Similarly, $\deg_Y P\leq d$ (since only $d$ entries of the matrix $\Syl_{d,e}\tup{\tilF, \tilG} $ have non-zero degree with respect to the variable $Y$, and these $d$ entries have degree $1$ with respect to this variable).

Next, we shall show that the polynomial $P$ is nonzero. Indeed, let us consider all elements of $\tilKK$ as polynomials in the variable $X$ over the ring $\KK \ive{Y} $. For each permutation $\sigma\in S_{d+e}$, the product $a_{\sigma}$ (thus considered) has degree $\leq e$ (as we have previously shown). Let us now compute the coefficient of $X^e $ in this product $a_{\sigma}$. There are three possible cases:

  • Case 1: The permutation $\sigma\in S_{d+e}$ does not satisfy $\left( \sigma\left( i\right) =i\text{ for each }i\in\left\{ 1,2,\ldots,e\right\} \right)$. Thus, the product $a_{\sigma}$ has strictly fewer than $e$ factors that have degree $1$ with respect to the variable $X$, while all its remaining factors have degree $0$ with respect to this variable. Thus, the whole product $a_{\sigma}$ has degree $<e$ with respect to the variable $X$. Hence, the coefficient of $X^e $ in this product $a_{\sigma}$ is $0$.

  • Case 2: The permutation $\sigma\in S_{d+e}$ satisfies $\left( \sigma\left( i\right) =i\text{ for each }i\in\left\{ 1,2,\ldots,e\right\} \right)$, but is not the identity map $\id\in S_{d+e}$. Thus, there must exist at least one $i\in\left\{ 1,2,\ldots,d+e\right\} $ such that $\sigma\left( i\right) <i$. Consider such an $i$, and notice that it must satisfy $i>e$ and $\sigma\left( i\right) >e$; hence, the $\left( i,\sigma\left( i\right) \right)$-th entry of $\Syl_{d,e}\tup{\tilF, \tilG}$ is $0$. Thus, the whole product $a_{\sigma}$ is $0$ (since the latter entry is a factor in this product). Thus, the coefficient of $X^e $ in this product $a_{\sigma}$ is $0$.

  • Case 3: The permutation $\sigma\in S_{d+e}$ is the identity map $\id\in S_{d+e}$. Thus, the product $a_{\sigma}$ is $\left( f_0 -X\right) ^e g_e^d $ (since $\tup{-1}^{\id }=1$). Hence, the coefficient of $X^e $ in this product $a_{\sigma}$ is $\tup{-1}^e g_e^d $.

Summarizing, we thus conclude that the coefficient of $X^e $ in the product $a_{\sigma}$ is $0$ unless $\sigma=\id$, in which case it is $\tup{-1}^e g_e^d $. Hence, the coefficient of $X^e $ in the sum $\sum_{\sigma\in S_{d+e}}a_{\sigma}$ is $\tup{-1}^e g_e^d \neq0$. Therefore, $\sum_{\sigma\in S_{d+e}}a_{\sigma}\neq0$. In view of \eqref{darij1.pf.t1.P=sum}, this rewrites as $P\neq0$. In other words, the polynomial $P$ is nonzero.

Finally, it remains to prove that $P\tup{F, G} =0$. In order to do this, we let $\LL$ be the polynomial ring $\KK \ive{U} $ in a new indeterminate $U$. We let $\varphi:\KK \ive{X, Y} \rightarrow\LL$ be the unique $\KK$-algebra homomorphism that sends $X$ to $F\tup{U}$ and sends $Y$ to $G\tup{U}$. (This is well-defined by the universal property of the polynomial ring $\KK \ive{X, Y}$.) Note that $\varphi$ is a $\KK$-algebra homomorphism from $\tilKK$ to $\LL$ (since $\KK \ive{X, Y} = \tilKK$). Thus, $\LL$ becomes a $\tilKK$-algebra via this homomorphism $\varphi$.

Now, recall that the polynomial $\tilF \in\tilKK\ive{T}$ was defined by $\tilF =F-X$. Hence, $\tilF \tup{U} = F\tup{U} - \varphi\tup{X}$. (Indeed, when we regard $X$ as an element of $\tilKK\ive{T}$, the polynomial $X$ is simply a constant, and thus evaluating it at $U$ yields the canonical image of $X$ in $\LL$, which is $\varphi\tup{X}$.) But $\varphi\tup{X} = F\tup{U}$ (by the definition of $\varphi$). Hence, $\tilF \tup{U} = F\tup{U} - \varphi\tup{X} = 0$ (since $\varphi\tup{X} = F\tup{U}$). Similarly, $\tilG \tup{U} = 0$.

Thus, the element $U\in\LL$ satisfies $\tilF \tup{U} =0$ and $\tilG \tup{U} = 0$. Hence, Theorem 2 (applied to $\tilKK$, $\tilF$, $\tilG$ and $U$ instead of $\KK$, $P$, $Q$ and $w$) yields that $\Res_{d,e}\tup{\tilF, \tilG} = 0$ in $\LL$. In other words, $\varphi\tup{ \Res_{d,e}\tup{\tilF, \tilG} } =0$. In view of $\Res_{d,e}\tup{\tilF, \tilG} =P$, this rewrites as $\varphi\tup{P} =0$.

But recall that $\varphi$ is the $\KK$-algebra homomorphism that sends $X$ to $F\tup{U}$ and sends $Y$ to $G\tup{U}$. Hence, it sends any polynomial $Q\in\KK \ive{X, Y}$ to $Q \tup{ F\tup{U}, G\tup{U} }$. Applying this to $Q=P$, we conclude that it sends $P$ to $P \tup{ F\tup{U}, G\tup{U} }$. In other words, $\varphi\tup{P} =P \tup{ F\tup{U}, G\tup{U} }$; hence, $P \tup{ F\tup{U}, G\tup{U} } =\varphi\tup{P} =0$.

Now, $F\tup{U}$ and $G\tup{U}$ are polynomials in the indeterminate $U$ over $\KK$. If we rename the indeterminate $U$ as $T$, then these polynomials $F\tup{U}$ and $G\tup{U}$ become $F\tup{T}$ and $G\tup{T}$, and therefore the polynomial $P\left( F\tup{U} ,G\tup{U} \right)$ becomes $P\tup{ F\tup{T}, G\tup{T} }$. Hence, $P\tup{ F\tup{T}, G\tup{T} } =0$ (since $P\tup{ F\tup{U}, G\tup{U} } =0$). In other words, $P \tup{F, G} =0$ (since $F\tup{T} =F$ and $G\tup{T} =G$). This completes the proof of Lemma 3. $\blacksquare$

Lemma 4. (a) Theorem 1 holds when $d = 0$.

(b) Theorem 1 holds when $e = 0$.

Proof of Lemma 4. (a) Assume that $d = 0$. Thus, $d + e > 0$ rewrites as $e > 0$. Hence, $e \geq 1$. But the polynomial $F$ is constant (since $\deg F \leq d = 0$). In other words, $F = f$ for some $f \in \KK$. Consider this $f$. Now, let $Q$ be the polynomial $X - f \in \KK\ive{X, Y}$. Then, $Q$ is nonzero and satisfies $\deg_X Q = 1 \leq e$ (since $e \geq 1$) and $\deg_Y Q = 0 \leq d$ and $Q\left(F, G\right) = F - f = 0$ (since $F = f$). Hence, there exists a nonzero polynomial $P\in\KK \ive{X, Y}$ in two indeterminates $X$ and $Y$ such that $\deg_X P\leq e$ and $\deg_Y P\leq d$ and $P\tup{F, G} =0$ (namely, $P = Q$). In other words, Theorem 1 holds (under our assumption that $d = 0$). This proves Lemma 4 (a).

(b) The proof of Lemma 4 (b) is analogous to our above proof of Lemma 4 (a). $\blacksquare$

Now, we can prove Theorem 1 at last:

Proof of Theorem 1. We shall prove Theorem 1 by induction on $e$.

The induction base is the case when $e = 0$; this case follows from Lemma 4 (b).

For the induction step, we fix a positive integer $\eps$. Assume (as the induction hypothesis) that Theorem 1 holds for $e = \eps - 1$. We must now prove that Theorem 1 holds for $e = \eps$.

Let $\KK$ be a commutative ring. Let $F$ and $G$ be two polynomials in the polynomial ring $\KK \ive{T}$. Let $d$ be a nonnegative integer such that $d+\eps > 0$ and $\deg F \leq d$ and $\deg G \leq \eps$. Our goal is now to prove that the claim of Theorem 1 holds for $e = \eps$. In other words, our goal is to prove that there exists a nonzero polynomial $P\in\KK \ive{X, Y}$ in two indeterminates $X$ and $Y$ such that $\deg_X P\leq \eps$ and $\deg_Y P\leq d$ and $P\tup{F, G} =0$.

Write the polynomial $G$ in the form $G=g_0 +g_1 T+g_2 T^2 +\cdots +g_{\eps}T^{\eps}$, where $g_0 ,g_1 ,\ldots,g_{\eps}\in\KK$. (This can be done, since $\deg G \leq \eps$.) If $g_{\eps}^d \neq 0$, then our goal follows immediately by applying Lemma 3 to $e = \eps$. Thus, for the rest of this induction step, we WLOG assume that $g_{\eps}^d = 0$. Hence, there exists a positive integer $m$ such that $g_{\eps}^m = 0$ (namely, $m = \eps$). Thus, there exists a smallest such $m$. Consider this smallest $m$. Then, $g_{\eps}^m = 0$, but \begin{align} \text{every positive integer $\ell < m$ satisfies $g_{\eps}^{\ell} \neq 0$.} \label{darij1.pf.t1.epsilon-ell} \tag{4} \end{align}

We claim that $g_{\eps}^{m-1} \neq 0$. Indeed, if $m-1$ is a positive integer, then this follows from \eqref{darij1.pf.t1.epsilon-ell} (applied to $\ell = m-1$); otherwise, it follows from the fact that $g_{\eps}^0 = 1 \neq 0$ (since the ring $\KK$ is nontrivial).

Now recall again that our goal is to prove that the claim of Theorem 1 holds for $e = \eps$. If $d = 0$, then this goal follows from Lemma 4 (a). Hence, for the rest of this induction step, we WLOG assume that $d \neq 0$. Hence, $d > 0$ (since $d$ is a nonnegative integer).

We have $e \geq 1$ (since $e$ is a positive integer), thus $e - 1 \geq 0$. Hence, $d + \left(e-1\right) \geq d > 0$.

Let $I$ be the subset $\left\{x \in \KK \mid g_{\eps}^{m-1} x = 0 \right\}$ of $\KK$. Then, $I$ is an ideal of $\KK$ (namely, it is the annihilator of the subset $\left\{g_{\eps}^{m-1}\right\}$ of $\KK$); thus, $\KK / I$ is a commutative $\KK$-algebra. Denote this commutative $\KK$-algebra $\KK / I$ by $\LL$. Let $\pi$ be the canonical projection $\KK \to \LL$. Of course, $\pi$ is a surjective $\KK$-algebra homomorphism.

For any $a \in \KK$, we will denote the image of $a$ under $\pi$ by $\overline{a}$.

The $\KK$-algebra homomorphism $\pi : \KK \to \LL$ induces a canonical $\KK$-algebra homomorphism $\KK\ive{T} \to \LL\ive{T}$ (sending $T$ to $T$). For any $a \in \KK\ive{T}$, we will denote the image of $a$ under the latter homomorphism by $\overline{a}$.

The $\KK$-algebra homomorphism $\pi : \KK \to \LL$ induces a canonical $\KK$-algebra homomorphism $\KK\ive{X, Y} \to \LL\ive{X, Y}$ (sending $X$ and $Y$ to $X$ and $Y$). For any $a \in \KK\ive{X, Y}$, we will denote the image of $a$ under the latter homomorphism by $\overline{a}$.

We have $g_{\eps}^{m-1} g_{\eps} = g_{\eps}^m = 0$, so that $g_{\eps} \in I$ (by the definition of $I$); hence, the residue class $\overline{g_{\eps}}$ of $g_{\eps}$ modulo the ideal $I$ is $0$.

We have $g_{\eps}^{m-1} \cdot 1 = g_{\eps}^{m-1} \neq 0$ in $\KK$, and thus $1 \notin I$ (by the definition of $I$). Hence, the ideal $I$ is not the whole ring $\KK$. Thus, the quotient ring $\KK / I = \LL$ is nontrivial.

But $G=g_0 +g_1 T+g_2 T^2 +\cdots +g_{\eps}T^{\eps}$ and thus \begin{align} \overline{G} &= \overline{g_0} + \overline{g_1} T + \overline{g_2} T^2 + \cdots + \overline{g_{\eps}} T^{\eps} \\ &= \left( \overline{g_0} + \overline{g_1} T + \overline{g_2} T^2 + \cdots + \overline{g_{\eps-1}} T^{\eps-1} \right) + \underbrace{\overline{g_{\eps}}}_{= 0} T^{\eps} \\ &= \overline{g_0} + \overline{g_1} T + \overline{g_2} T^2 + \cdots + \overline{g_{\eps-1}} T^{\eps-1} , \end{align} so that $\deg \overline{G} \leq e-1$. Also, $\deg \overline{F} \leq \deg F \leq d$. But the induction hypothesis tells us that Theorem 1 holds for $e = \eps - 1$. Hence, we can apply Theorem 1 to $\LL$, $\overline{F}$, $\overline{G}$ and $\eps - 1$ instead of $\KK$, $F$, $G$ and $e$. We thus conclude that there exists a nonzero polynomial $P\in \LL\ive{X, Y}$ in two indeterminates $X$ and $Y$ such that $\deg_X P \leq \eps - 1$ and $\deg_Y P \leq d$ and $P\left( \overline{F}, \overline{G} \right) =0$. Consider this polynomial $P$, and denote it by $R$. Thus, $R \in \LL \ive{X, Y}$ is a nonzero polynomial in two indeterminates $X$ and $Y$ and satisfies $\deg_X R \leq \eps - 1$ and $\deg_Y R \leq d$ and $R \left( \overline{F}, \overline{G} \right) =0$.

Clearly, there exists a polynomial $Q \in \KK\ive{X, Y}$ in two indeterminates $X$ and $Y$ that satisfies $\deg_X Q = \deg_X R$ and $\deg_Y Q = \deg_Y R$ and $\overline{Q} = R$. (Indeed, we can construct such a $Q$ as follows: Write $R$ in the form $R = \sum\limits_{i = 0}^{\deg_X R} \sum\limits_{j = 0}^{\deg_Y R} r_{i, j} X^i Y^j$ for some coefficients $r_{i, j} \in \LL$. For each pair $\left(i, j\right)$, pick some $p_{i, j} \in \KK$ such that $\overline{p_{i, j}} = r_{i, j}$ (this can be done, since the homomorphism $\pi : \KK \to \LL$ is surjective). Then, set $Q = \sum\limits_{i = 0}^{\deg_X R} \sum\limits_{j = 0}^{\deg_Y R} p_{i, j} X^i Y^j$. It is clear that this polynomial $Q$ satisfies $\deg_X Q = \deg_X R$ and $\deg_Y Q = \deg_Y R$ and $\overline{Q} = R$.)

We have $\overline{Q \left(F, G\right)} = \underbrace{\overline{Q}}_{=R} \left( \overline{F}, \overline{G} \right) = R \left( \overline{F}, \overline{G} \right) = 0$. In other words, the polynomial $Q \left(F, G\right) \in \KK\ive{T}$ lies in the kernel of the canonical $\KK$-algebra homomorphism $\KK\ive{T} \to \LL\ive{T}$. This means that each coefficient of this polynomial $Q \left(F, G\right) \in \KK\ive{T}$ lies in the kernel of the $\KK$-algebra homomorphism $\pi : \KK \to \LL$. In other words, each coefficient of this polynomial $Q \left(F, G\right) \in \KK\ive{T}$ lies in $I$ (since the kernel of the $\KK$-algebra homomorphism $\pi : \KK \to \LL$ is $I$). Hence, each coefficient $c$ of this polynomial $Q \left(F, G\right) \in \KK\ive{T}$ satisfies $g_{\eps}^{m-1} c = 0$ (by the definition of $I$). Therefore, $g_{\eps}^{m-1} Q \left(F, G\right) = 0$.

On the other hand, $\overline{Q} = R$ is nonzero. In other words, the polynomial $Q \in \KK\ive{X, Y}$ does not lie in the kernel of the canonical $\KK$-algebra homomorphism $\KK\ive{X, Y} \to \LL\ive{X, Y}$. This means that not every coefficient of this polynomial $Q \in \KK\ive{X, Y}$ lies in the kernel of the $\KK$-algebra homomorphism $\pi : \KK \to \LL$. In other words, not every coefficient of this polynomial $Q \in \KK\ive{X, Y}$ lies in $I$ (since the kernel of the $\KK$-algebra homomorphism $\pi : \KK \to \LL$ is $I$). Hence, not every coefficient $c$ of this polynomial $Q \in \KK\ive{X, Y}$ satisfies $g_{\eps}^{m-1} c = 0$ (by the definition of $I$). Therefore, $g_{\eps}^{m-1} Q \neq 0$. So $g_{\eps}^{m-1} Q \in \KK\ive{X, Y}$ is a nonzero polynomial in two indeterminates $X$ and $Y$ and satisfies $\deg_X \left( g_{\eps}^{m-1} Q \right) \leq \deg_X Q = \deg_X R \leq \eps - 1 \leq \eps$ and $\deg_Y \left( g_{\eps}^{m-1} Q \right) \leq \deg_Y Q = \deg_Y R \leq d$ and $\left(g_{\eps}^{m-1} Q \right) \left(F, G\right) = g_{\eps}^{m-1} Q \left(F, G\right) = 0$. Hence, there exists a nonzero polynomial $P\in\KK \ive{X, Y}$ in two indeterminates $X$ and $Y$ such that $\deg_X P\leq \eps$ and $\deg_Y P\leq d$ and $P\tup{F, G} =0$ (namely, $P = g_{\eps}^{m-1} Q$). We have thus reached our goal.

So we have proven that Theorem 1 holds for $e = \eps$. This completes the induction step. Thus, Theorem 1 is proven by induction. $\blacksquare$

1

Let $A\subset B$ commutative rings. An element $b \in B$ is called integral over $A$ if $$p(b) = b^n+ a_{n-1} b^{n-1} + \cdots + a_0 = 0$$ for some $n\ge 1$, and $a_0$, $\ldots$, $a_{n-1} \in A$.

Now, if $F \in A[t]$, and $b\in B$ is integral over $A$, then $F(b)$ is again integral over $A$. Moreover, there is an explicit way to find the equation for $F(b)$. Indeed, let $F(t) = t^m + \alpha_{m-1} t^{m-1} + \cdots + \alpha_0$. Consider the companion matrix $$C = C(p)$$ ( where $p\in A[x]$ is the polynomial of degree $n$ from above), and the matrix $$D = F(C)$$ Then $F(b)$ satisfies the equation of degree $n$

$$Q(F(b) ) = 0$$ where $Q$ is the characteristic polynomial of $F(C)$.

Now consider our problem: we are given polynomials $F(t)$, $G(t)$ with $G$ monic. Now $t$ satisfies an integral equation "over G(t)".

$$t^n + g_{n-1} t^{n-1} + \cdots (g_0 - G)= 0$$

From this, as above, get an equation in $F(t)$, and $G(t)$.

orangeskid
  • 56,630