Let $\mathbb R[x,y]_+$ denote the set of positive polynomials in two variables. My problem can be stated as follows:

Does there exist a countable set $M\subseteq \mathbb R[x,y]_+$ such that every $f\in \mathbb R[x,y]_+$ can be represented as $$ f=\sum_{i=1}^n p_i^2\cdot h_i,\ h_i\in M,\ p_i\in\mathbb R[x,y]\ ? $$ More concretely, can we take $M=\mathbb Q[x,y]_+\ ?$

Recall that a quadratic module in a commutative ring $A$ with $1\in A$ is a subset $Q\subseteq A$ such that

$(i)\ 1\in Q,\ -1\notin Q$, $(ii)\ Q+Q\subseteq Q$, $(iii)\ p^2\cdot Q\subseteq Q,\ \forall p\in A.$

The problem can be now stated as follows:

Is $\mathbb R[x,y]_+$ countably generated as a quadratic module?

The smallest quadratic module in $\mathbb R[x_1,\dots,x_d]$ is the set $\sum^2\mathbb R[x_1,\dots,x_d]$ of finite sums of squares. We have $\mathbb R[x_1,\dots,x_d]_+=\sum^2\mathbb R[x_1,\dots,x_d]$ iff $d=1$ (Hilberts theorem). It is known that $\mathbb R[x_1,\dots,x_d]_+$ is not finitely generated for $d\geq 2.$ One can also easily show that $\mathbb R[x_1,\dots,x_d]_+$ is not countably generated for $d\geq 3.$


Suppose that a countable set $M$ as required exists. (As you say, $M$ cannot be finite - in particular it cannot be empty.) Let $d$ be the minimal degree of all polynomials in $M$ and set $M_d \subset M$ be the subset of elements of minimal degree. Since $M$ is not empty, $d\neq 0$, and the zero-set of any polynomial in $M_d$ is a proper algebraic subset of $\mathbb R^2$.

Let $f \in \mathbb R[x,y]_+$ be any element of minimal degree and without loss of generality, assume that there exists $(x_0,y_0) \in \mathbb R^2$, such that $f(x_0,y_0)=0$. (This needs a little additional argument.)

Now, for any $(\lambda,\mu) \in \mathbb R^2$, set $$f_{\lambda,\mu}(x,y):= f(x-\lambda,y-\mu)$$ and note that $f_{\lambda,\mu} \in \mathbb R[x,y]_+$.

For any $(\lambda,\mu) \in \mathbb R^2$, the polynomial $f_{\lambda,\mu}$ must be (for degree reasons) a positive linear combination of elements in $M_d$. Thus, there is some $h \in M_d$, which vanishes at $(x_0+ \lambda,y_0 + \mu)$. We conclude that the union of zero-sets of polynomials in $M_d$ must be the whole of $\mathbb R^2$. This is a contradiction, since $M_d$ is countable and this union is a countable union of proper algebraic subsets.

| cite | improve this answer | |
  • $\begingroup$ I just realize that a positive polynomial does not need to realize its infimum, e.g., $f(x,y)=x^2 + (xy-1)^2$. Hence, the argument above is not complete in this form. $\endgroup$ – Andreas Thom Sep 22 '14 at 16:10

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.