Using model theory to prove a Real Nullstellensatz?

631 Views Asked by At

Let $F$ be a real closed field i.e. $F$ has no proper real algebraic extension i.e. $F[X]/(X^2+1)$ is an algebraically closed field. For $I$ an ideal of $F[X_1,...,X_n]$ , let $V_F(I) :=\{ \bar a \in F^n : f(\bar a)=0, ,\forall f \in I \}$.

Then how to show , using model theory (like completeness or model completeness of the first order theory of real closed fields which includes ordering in the language) , that $V_F(I)$ is non-empty iff whenever $\sum_i p_i^2 \in I$ then all $p_i \in I$ ?

1

There are 1 best solutions below

0
On BEST ANSWER

As pointed out by Dap in the comments, the statement of the exercise (Marker's Model Theory, Exercise 3.4.28) is wrong. In fact, neither direction holds. We say that an ideal $I\subseteq F[X_1,\dots,X_n]$ is real if $\sum_{i=1}^m p_i^2\in I$ implies $p_i\in I$ for all $1\leq i \leq m$. If $I = (X^2 + Y^2)$, then $V_F(I) = \{(0,0)\}$ is nonempty, but neither $X$ nor $Y$ are in $I$, so $I$ is not real. Conversely, the unit ideal $(1)$ is real, but $V_F((1))$ is empty.

The exercise is addressed in Marker's Errata (which is an important reference for anyone working through the book). There, Marker simply writes "Assume $I$ is prime". This fixes one direction: it's true that if $I$ is a real prime ideal, then $V_F(I)$ is non-empty (and note that the unit ideal is real but not prime). But it does not fix the other direction, since $(X^2+Y^2)$ is prime in $\mathbb{R}[X,Y]$.

There are several ways to state the real nullstellensatz (or positivestellensatz). Here is one that mirrors Marker's statement of the ordinary nullstellensatz (Theorem 3.2.11).

For any field $F$, the map $V\colon J\mapsto \{\overline{a}\in F^n\mid p(\overline{a}) = 0\text{ for all }p\in J\}$ is an inclusion-reversing map from ideals in $F[X_1,\dots,X_n]$ to algebraic sets in $F^n$. And the map $I\colon X\mapsto \{p\in F[X_1,\dots,X_n]\mid p(\overline{a}) = 0\text{ for all }\overline{a}\in X\}$ is an inclusion-reversing map from subsets of $F^n$ to ideals in $F[X_1,\dots,X_n]$. If $F$ is a formally real field, then $I(X)$ is always a real ideal.

For $X\subseteq F^n$, write $\overline{X}$ for the Zariski-closure of $X$, the smallest algebraic set containing $X$. It is easy to see that $V(I(X)) = \overline{X}$.

For an ideal $J\subseteq F[X_1,\dots,X_n]$, write $\sqrt[R]{J}$ for the real radical of $J$, the smallest real ideal containing $J$.

Theorem: Suppose $F$ is real closed. Then for any ideal $J\subseteq F[X_1,\dots,X_n]$, $I(V(J)) = \sqrt[R]{J}$. Thus the maps $V$ and $I$ give an inclusion-reversing bijection between the set of real ideals in $F[X_1,\dots,X_n]$ and the set of algebraic sets in $F^n$.

As a corollary, we get a correct version of the statement of the exercise.

Corollary: Suppose $F$ is real closed and $J\subseteq F[X_1,\dots,X_n]$ is an ideal. Then $V(J)$ is nonempty if and only if the real radical of $J$ is proper.

Proof: $V(J) = \emptyset = V((1))$ if and only if $\sqrt[R]{J} = I(V(J)) = I(V((1)) = (1)$.

You can give a model-theoretic proof of the real nullstellensatz in a way very similar to the way Marker proves the usual nullstellensatz for algebraically closed fields (Theorem 3.2.11). But you need the following three algebraic components. I think only the third is proven in Marker, but the first two are fairly straightforward exercises.

  • The real radical of an ideal $J$ is the intersection of all the prime real ideals containing $J$ (or it is the unit ideal, if there are no prime real ideals containing $J$).
  • If $J$ is a real prime ideal in a ring $A$, then $\text{Frac}(A/J)$ is a formally real field.
  • Every formally real field embeds in a real closed field.

To illustrate the main idea, here's a direct proof of the corollary (without going through the proof of the general real nullstellensatz).

Proof: Suppose $F$ is real closed and $J$ is an ideal in $F[X_1,\dots,X_n]$. If $V(J)$ is nonempty, then it's easy to check that $I(V(J))$ is a proper real ideal containing $J$, and hence containing $\sqrt[R]{J}$, so $\sqrt[R]{J}$ is proper. Conversely, suppose $\sqrt[R]{J}$ is proper. Then by the characterization of the real radical above, there is some prime real ideal $P$ containing $J$. Then $F$ embeds in the formally real field $K = \text{Frac}(F[X_1,\dots,X_n]/P)$, and hence into a real closure $R$ of $K$. Let $q_1,\dots,q_m$ be generators for $J$ (using Noetherianity of $F[X_1,\dots,X_n])$). Then in $K$, and hence in $R$, the image of the variables $(X_1,\dots,X_n)$ satisfy the polynomials $q_1,\dots,q_m$, so $R\models \exists \overline{x}\, \bigwedge_{i=1}^m q_i(\overline{x}) = 0$. The theory of real closed fields is model complete (in the field language), so $F$ also satisfies this formula, and hence $V(J)\neq \emptyset$.