Hilbert's Nullstellensatz: The Dictionary Between Ideals and Varieties

What the theorem claims

Fix an algebraically closed field k (think ℂ) and the polynomial ring R = k[x₁,…,xₙ]. For an ideal I ⊆ R, its vanishing set is V(I) = { a ∈ kⁿ : f(a) = 0 for all f ∈ I }, an affine algebraic variety. For a subset X ⊆ kⁿ, the ideal of X is I(X) = { f ∈ R : f(a) = 0 for all a ∈ X }.

The weak Nullstellensatz: if I is a proper ideal (I ≠ R), then V(I) ≠ ∅. Equivalently, the equations f₁ = ⋯ = fₘ = 0 have no common solution in kⁿ iff 1 ∈ (f₁,…,fₘ), i.e. iff there exist g₁,…,gₘ ∈ R with ∑ gᵢfᵢ = 1.

The strong Nullstellensatz upgrades this to an exact formula: for every ideal I,

I(V(I)) = √I = { f ∈ R : fᵏ ∈ I for some k ≥ 1 }.

So a polynomial vanishes on V(I) exactly when a power of it lies in I. Both halves fail badly without algebraic closure.

The picture: a dictionary between algebra and geometry

Restricting the maps V and I to radical ideals (ideals with √I = I) and to varieties makes them mutually inverse, order-reversing bijections. That single fact is the reason algebraic geometry exists: every geometric notion has an algebraic mirror image.

Points a = (a₁,…,aₙ) of kⁿ correspond to maximal ideals mₐ = (x₁−a₁,…,xₙ−aₙ). Irreducible varieties correspond to prime ideals. The empty variety corresponds to the whole ring R (the only 'ideal' with no zeros, because 1 ∈ R). Union of varieties matches intersection of ideals; intersection matches (the radical of) the sum. A variety sits inside another, X ⊆ Y, exactly when the ideals reverse, I(Y) ⊆ I(X).

The picture to hold in your head: kⁿ is a space whose 'coordinate functions' live in R, and the theorem certifies that R sees every point — no point is invisible to the algebra, and no phantom relations exist. The coordinate ring R/I(X) is precisely the ring of polynomial functions on X.

Key idea of the proof

The engine is Zariski's lemma (1947): if a field K is finitely generated as an algebra over a field k (i.e. K = k[y₁,…,yₙ] is a field), then K is a finite algebraic extension of k. Two standard routes prove it — Noether normalization (K is finite over a polynomial subring, which must then be k itself) or Artin–Tate.

From Zariski's lemma the weak form follows: take a maximal ideal m ⊆ R. Then R/m is a field, finitely generated over k, hence (Zariski) a finite extension; since k is algebraically closed, R/m ≅ k. The images of x₁,…,xₙ give a point a with m = mₐ, so m has a zero — every proper ideal, contained in some maximal ideal, therefore has a zero.

The strong form follows from the weak one by the elegant Rabinowitsch trick: to show f vanishes on V(I) ⟹ fᵏ ∈ I, introduce an extra variable t and work in k[x₁,…,xₙ,t] with the ideal (I, tf − 1). It has empty zero set, so by the weak form 1 lies in it; clearing denominators (substitute t = 1/f) yields fᵏ ∈ I.

A worked example

Take n = 1, k = ℂ, and I = (x²) ⊆ ℂ[x]. Then V(I) = { 0 }, and I(V(I)) is the ideal of all polynomials vanishing at 0, which is (x). Indeed √(x²) = (x): the element x satisfies x² ∈ (x²), so x ∈ √I, and (x) is already radical. The strong Nullstellensatz predicts I(V(I)) = √(x²) = (x) — exactly what we found. Note I(V(I)) = (x) is strictly bigger than I = (x²): passing to zero sets forgets multiplicity, and the radical is what records that loss.

A two-variable case: I = (x² + y², x) in ℂ[x,y]. From x = 0 the first generator gives y² = 0, so V(I) = { (0,0) }, hence I(V(I)) = (x,y). And indeed (x,y) = √I, since y ∈ √I because y² = (x²+y²) − x·x ∈ I. Over ℝ the story would differ (see below): x²+y² = 0 already forces the single real point, but the radical computation still needs the algebra.

Why algebraic closure is essential

Drop 'algebraically closed' and everything collapses. Over ℝ take I = (x² + 1) ⊆ ℝ[x]. It is a proper ideal (in fact maximal, since ℝ[x]/(x²+1) ≅ ℂ is a field), yet V(I) = ∅ in ℝ: there is no real root. The weak Nullstellensatz fails — a perfectly consistent-looking equation has no solution. Correspondingly √I = I = (x²+1) but I(V(I)) = I(∅) = ℝ[x], so the strong form fails too.

The deeper reason: over a non-closed field, maximal ideals need not be points — R/m can be a proper finite extension of k (like ℂ over ℝ), so 'solutions' live in an extension, not in kⁿ. Algebraic closure guarantees R/m ≅ k, pinning every maximal ideal to an honest point. Finiteness of the generating set is also implicit through R being Noetherian (Hilbert's basis theorem), which keeps ideals finitely generated. These hypotheses connect the Nullstellensatz to field-extension theory, Galois theory, and, through primes and radicals, to ring theory.

Why it matters and what it unlocks

The Nullstellensatz is the foundation stone of classical and modern algebraic geometry. It legitimizes the coordinate-ring dictionary that lets one study varieties X entirely through their rings R/I(X), and it identifies points of a variety with maximal ideals — the seed of Grothendieck's Spec and scheme theory, where all primes become points and the theorem's punchline (max ideals over an algebraically closed field ↔ points) becomes a special case.

Practically, the weak form is a solvability certificate: a polynomial system is inconsistent iff a Gröbner basis of the ideal contains 1, which is exactly how computer algebra systems decide emptiness of a variety. Effective versions (Kollár, Brownawell) bound the degrees needed in 1 = ∑ gᵢfᵢ, linking to complexity theory. It also generalizes the Fundamental Theorem of Algebra (the n = 1 case over ℂ), underpins dimension theory, Bézout's theorem, and elimination theory, and its 'real' analogue (the Positivstellensatz / Real Nullstellensatz of Krivine–Stengle) drives modern polynomial optimization and sum-of-squares methods.

The weak and strong Nullstellensatz, plus what they specialize to and require.

Version	Statement (k alg. closed)	Geometric meaning
Weak	If I ⊊ k[x₁,…,xₙ] is a proper ideal, then V(I) ≠ ∅	A consistent-looking system (1 ∉ I) really has a common solution
Maximal ideals	Every maximal ideal is (x₁−a₁,…,xₙ−aₙ) for some point a ∈ kⁿ	Points of kⁿ ↔ maximal ideals of k[x₁,…,xₙ]
Strong	I(V(I)) = √I for every ideal I	The functions vanishing on V(I) are exactly those with a power in I
Correspondence	V and I are inverse, inclusion-reversing bijections	Radical ideals ↔ affine varieties (the 'dictionary')
n = 1, k = ℂ	Every nonconstant f ∈ ℂ[x] has a root	Recovers the Fundamental Theorem of Algebra