An Explicit Example of the Proof of the Nullstellensatz

11 Jan 2022 - Tags: sage

I’m in an algebraic geometry class right now, and a friend was struggling conceptually with the proof of the strong nullstellensatz. I thought it might be helpful to see a concrete example of the idea, since the proof is actually quite constructive! Which brings us to this post:

Formally, we’re going to assume the weak nullstellensatz, and use it to show the strong nullstellensatz. That is, we’ll assume

and we’ll show

Both of these are theorems of the form “the obvious issue is the only one”. Obviously if $1=p_1{f}_1+p_2{f}_2+\dots +p_r{f}_r$, then the $f_i$ cannot all be $0$ simultanously. Indeed, if $f_i(x^{\ast })=0$ for all $i$, then evaluating both sides of the above at $x^{\ast }$ gives $1=0$, which is a problem. The weak nullstellensatz says that this is the only reason a family of polynomials won’t have a common root.

Similarly, it’s obvious that if $g^n=p_1{f}_1+\dots +p_r{f}_r$, then at any common zero of the $f_i$, $g=0$ too. Again, if we evaluate both sides at $x^{\ast }$ we find $g(x^{\ast }{)}^n=0$, and so $g(x^{\ast })=0$ too (since a field has no nontrivial nilpotents). The strong nullstellensatz says that this is the only way for $g$ to share a zero with the $f_i$.

---

Now, it turns out the weak nullstellensatz has computational content. That is, if $f_1,\dots ,f_r$ don’t have a common zero, there’s a computer program¹ that will actually find the $p_i$ so that $1=p_1{f}_1+\dots +p_r{f}_r$.

For instance, let’s take a simple example:

$$f_1=xy-1f_2=x+y{f}_3=x{y}^3{f}_4=y{x}^3$$

First, let’s check that these polynomials really don’t have any points in common:

Next we can see that they generate the ideal $(1)$.

Of course, this means we should be able to write $1$ as a linear combination of the $f_i$:

and indeed, these are the coefficients to get $1$

---

So now what about the strong nullstellensatz?

Let’s take $g=yx+x+1$, which vanishes at every point of the variety defined by $x^2+2x+1$ and $y$ (do you see why?).

Then we expect $g^n\in (x^2+2x+1,y)$ for some $n$, and we’ll get there by the Rabinowitsch trick:

We’ll add a variable $z$ to the mix, and notice that $x^2+2x+1$, $y$, and $(yx+x+1)z-1$ don’t have any common zeroes.

Indeed, if $x^2+2x+1$ or $y$ is nonzero at some point, then we’re done. But if they’re both zero, then we know $g=yx+x+1$ is zero as well. Then $(yx+x+1)z-1=gz-1=0z-1=-1\ne 0$.

But then, by the weak nullstellensatz, that means these three polynomials must generate the ideal $(1)$ in $k[x,y,z]$!

Indeed,

So we know that

$$1=p_1{f}_1+p_2{f}_2+p_3(gz-1)$$

or

$$1=z^2(x^2+2x+1)+(x^2{z}^2+x{z}^2+xz)y+(-xz-z-1)(gz-1)$$

Now for the slick trick! We’re working in an ideal containing $zg-1$, which means that $z=\frac{1}{g}$ in all of our computations²! So let’s take this expression and plug in $z=\frac{1}{g}$ to get

$$1=\frac{1}{g^2}{f}_1+(\frac{x^2}{g^2}+\frac{x}{g^2}+\frac{x}{g})f_2+(-\frac{x}{g}-\frac{1}{g}-1)0$$

Of course, we can clear the denominators by multiplying through by $g^2$ to see

$$g^2=f_1+(x^2+x+xg)f_2\in (f_1,f_2)$$

So we found that, for some $n$, $g^n\in (f_1,f_2)$. As desired.

---

It turns out that this is exactly how the proof goes in general!

Say you give me polynomials $f_1,\dots ,f_r,g\in k[x_1,\dots ,x_m]$ so that $g$ vanishes whenever all the $f_i$ do.

Then we look at the ideal (in $k[x_1,\dots x_m,z]$)

$$(f_1,\dots ,f_r,zg-1)$$

which must equal $(1)$ by the weak nullstellensatz.

Then a computation, which sage will happily do for us, gives us polynomials $p_1,\dots ,p_{r+1}\in k[x_1,\dots ,x_m,z]$ so that

$$1=p_1{f}_1+\dots +p_r{f}_r+p_{r+1}(zg-1)$$

Then we plug in $\frac{1}{g}$ for $z$ to get a new expression

$$1=p_1(\overrightarrow{x},\frac{1}{g})f_1(\overrightarrow{x})+\dots +p_r(\overrightarrow{x},\frac{1}{g})f_r(\overrightarrow{x})$$

This is a polynomial with $g$s in the denominator. So we multiply both sides by some $g^n$ to clear denominators, and we find

$$g^n=g^n{p}_1(\overrightarrow{x},1)f_1+\dots +g^n{p}_r(\overrightarrow{x},1)f_r$$

Notice that the $p_i(\overrightarrow{x},1)$ are polynomials in $x_1,\dots ,x_m$, since we’ve plugged in $1$ for $z$ everywhere. This means we’ve shown $g^n$ is a linear combination of the $f_i$, so $g^n\in (f_1,\dots ,f_r)$, as desired.

---

Another quick post today! Hopefully other people find this helpful too ^_^.

I do have a few bigger ones in the pipeline, but I won’t say exactly what. I’ve learned that saying what post you’re planning to write next is a guaranteed way to not actually write it, haha.

See you soon!

---