Affine Varieties and the Nullstellensatz

⚡ Key Takeaways

Hilbert's Nullstellensatz establishes a precise dictionary between geometry (algebraic subsets of affine space over an algebraically closed field) and algebra (radical ideals of the polynomial ring). One-line statement: over an algebraically closed field $k$ , $I(V(\mathfrak{a})) = \sqrt{\mathfrak{a}}$ for every ideal $\mathfrak{a} \subseteq k[x_1,\dots,x_n]$ . Applies throughout classical and modern algebraic geometry, commutative algebra, and elimination theory. Proved by David Hilbert in 1893.

💡 The Core Insight

Polynomials see zeros, but they cannot distinguish a function $f$ from its powers $f^m$ when only the set of zeros matters: $f$ and $f^m$ vanish at exactly the same points. Hence the only algebraic information recoverable from a zero set is the ideal up to radical. The Nullstellensatz says this is the only obstruction — over $\bar{k}$ , the radical is a complete invariant. Geometry remembers exactly the radical ideal, nothing more, nothing less.

DefinitionAffine space and the Zariski correspondence

Let $k$ be a field with algebraic closure $\bar k$ , and fix $n \geq 1$ . Affine $n$ -space is the set $\mathbb{A}^n_k = k^n$ of $n$ -tuples of elements of $k$ . Write $R = k[x_1,\dots,x_n]$ for the polynomial ring. For a subset $S \subseteq R$ , the vanishing set is

1

\quad V(S) = \{\, p \in \mathbb{A}^n_k : f(p) = 0 \text{ for all } f \in S \,\}.

Derivation hint: This is the simultaneous solution locus; since $V(S) = V(\langle S\rangle)$ , we lose nothing by taking $S$ to be an ideal. For a subset $X \subseteq \mathbb{A}^n_k$ , the vanishing ideal is

2

\quad I(X) = \{\, f \in R : f(p) = 0 \text{ for all } p \in X \,\}.

Derivation hint: Closure under sums and $R$ -multiples is immediate from the pointwise definition, so $I(X)$ is genuinely an ideal. A subset $X \subseteq \mathbb{A}^n_k$ is an (affine) algebraic set if $X = V(S)$ for some $S$ . An affine variety is (in the classical convention) an irreducible algebraic set; many authors use "variety" for any algebraic set, so we state results in terms of algebraic sets.

DefinitionRadical of an ideal

The radical of an ideal $\mathfrak{a} \subseteq R$ is

3

\quad \sqrt{\mathfrak{a}} = \{\, f \in R : f^m \in \mathfrak{a} \text{ for some } m \geq 1 \,\}.

Derivation hint: That $\sqrt{\mathfrak a}$ is itself an ideal follows from the binomial theorem applied to $(f+g)^{m+n}$ when $f^m, g^n \in \mathfrak a$ . An ideal is radical if $\mathfrak{a} = \sqrt{\mathfrak{a}}$ . Equivalently, $R/\mathfrak{a}$ is reduced (has no nonzero nilpotents).

TheoremHilbert's Nullstellensatz — Weak Form

Let $k$ be algebraically closed and $\mathfrak{a} \subseteq k[x_1,\dots,x_n]$ an ideal. Then

4

\quad V(\mathfrak{a}) = \varnothing \iff \mathfrak{a} = (1) = R.

Derivation hint: One direction is trivial ( $V(R)=\varnothing$ ); the content is that a proper ideal always has a common zero, equivalent to every maximal ideal of $R$ having residue field $k$ .

TheoremHilbert's Nullstellensatz — Strong Form

Let $k$ be algebraically closed. For every ideal $\mathfrak{a} \subseteq R$ ,

5

\quad I\big(V(\mathfrak{a})\big) = \sqrt{\mathfrak{a}}.

Derivation hint: $\sqrt{\mathfrak a} \subseteq I(V(\mathfrak a))$ is elementary; the reverse inclusion is extracted from the weak form via the Rabinowitsch trick. Consequently the maps $V$ and $I$ restrict to mutually inverse, inclusion-reversing bijections

6

\quad \{\text{radical ideals of } R\} \;\xleftrightarrow{\;\;I,\,V\;\;}\; \{\text{algebraic subsets of } \mathbb{A}^n_k\}.

Derivation hint: $V(I(X)) = X$ for algebraic $X$ (by definition of algebraic set plus monotonicity), and $I(V(\mathfrak a)) = \sqrt{\mathfrak a}$ is the strong form; restricting domains gives a bijection.

TheoremZariski's Lemma — the algebraic engine

Let $k$ be any field and $L$ a field that is finitely generated as a $k$ -algebra. Then $L$ is a finite (hence algebraic) field extension of $k$ .

7

\quad L = k[a_1,\dots,a_m] \text{ a field} \implies [L:k] < \infty.

Derivation hint: If some $a_i$ were transcendental, a Noether normalization argument shows $L$ could not be a finitely generated module over a polynomial subring while being a field; the integral closure forces algebraicity.

ProofWeak Nullstellensatz from Zariski's Lemma

We prove the weak form, then deduce the strong form.

Step 1 — Reduce to a maximal ideal. Suppose $\mathfrak{a} \subsetneq R$ is proper. Since $R$ is Noetherian (Hilbert Basis Theorem), $\mathfrak a$ is contained in some maximal ideal $\mathfrak{m}$ . It suffices to find a point $p$ with $\mathfrak m \subseteq I(\{p\})$ , i.e. a common zero of $\mathfrak m$ ; that point lies in $V(\mathfrak a) \supseteq V(\mathfrak m)$ too... more precisely $V(\mathfrak a) \supseteq V(\mathfrak m) \ni p$ .

Step 2 — Form the residue field. Let $L = R/\mathfrak{m}$ . This is a field, and it is generated as a $k$ -algebra by the images $\bar x_1,\dots,\bar x_n$ . Thus $L$ is a finitely generated $k$ -algebra that is a field.

Step 3 — Apply Zariski's Lemma. By [7], $L$ is a finite extension of $k$ . Since $k$ is algebraically closed, the only finite extension is $k$ itself: the canonical map $k \to L$ is an isomorphism.

Step 4 — Read off the point. Let $a_i \in k$ be the unique preimage of $\bar x_i$ under $k \xrightarrow{\sim} L$ . Set $p = (a_1,\dots,a_n) \in \mathbb{A}^n_k$ . For any $f \in \mathfrak m$ , the image of $f$ in $L$ is $f(a_1,\dots,a_n) = f(p)$ , which is $0$ since $f \in \mathfrak m$ . Hence $p \in V(\mathfrak m) \subseteq V(\mathfrak a)$ , so $V(\mathfrak a) \neq \varnothing$ . $\quad\blacksquare$ This also shows every maximal ideal of $R$ has the form $\mathfrak m_p = (x_1 - a_1, \dots, x_n - a_n)$ .

∎

ProofStrong Nullstellensatz via the Rabinowitsch trick

We show $I(V(\mathfrak{a})) = \sqrt{\mathfrak a}$ .

Step 1 — Easy inclusion. If $f \in \sqrt{\mathfrak a}$ , say $f^m \in \mathfrak a$ , then for any $p \in V(\mathfrak a)$ we have $f(p)^m = (f^m)(p) = 0$ , so $f(p) = 0$ . Thus $f \in I(V(\mathfrak a))$ , giving $\sqrt{\mathfrak a} \subseteq I(V(\mathfrak a))$ .

Step 2 — Set up the trick. Suppose $f \in I(V(\mathfrak a))$ . Since $R$ is Noetherian, $\mathfrak a = (g_1,\dots,g_r)$ . Introduce a new variable $y$ and work in $R' = k[x_1,\dots,x_n,y]$ . Consider the ideal

8

\quad \mathfrak{b} = \big(g_1,\dots,g_r,\; 1 - y\,f\big) \subseteq R'.

Derivation hint: The hyperplane $1 = yf$ excludes precisely the locus $f = 0$ in $\mathbb A^{n+1}$ , so it kills every common zero of $\mathfrak a$ .

Step 3 — $\mathfrak b$ has no zeros. Any common zero $(p, b) \in \mathbb{A}^{n+1}_k$ of $\mathfrak b$ would satisfy $g_i(p) = 0$ (so $p \in V(\mathfrak a)$ ) and $1 - b\,f(p) = 0$ . But $p \in V(\mathfrak a)$ forces $f(p) = 0$ (since $f \in I(V(\mathfrak a))$ ), making $1 - 0 = 0$ , a contradiction. So $V(\mathfrak b) = \varnothing$ .

Step 4 — Apply the weak form. By the weak Nullstellensatz, $\mathfrak b = R'$ . Hence there exist $h_i, h \in R'$ with

9

\quad 1 = \sum_{i=1}^r h_i\,g_i + h\,(1 - y f).

Derivation hint: This is exactly the statement $1 \in \mathfrak b$ .

Step 5 — Substitute $y = 1/f$ . Pass to the field of fractions and set $y = 1/f$ (valid where $f \neq 0$ ). The last term vanishes, leaving an identity in $k(x_1,\dots,x_n)$ :

10

\quad 1 = \sum_{i=1}^r h_i\!\left(x, \tfrac{1}{f}\right) g_i(x).

Derivation hint: Substitution is a ring homomorphism $R' \to k(x)$ sending $1-yf \mapsto 0$ .

Step 6 — Clear denominators. Each $h_i(x, 1/f)$ is a polynomial in the $x_j$ divided by some power of $f$ . Let $m$ be the largest such power. Multiplying [10] through by $f^m$ gives

11

\quad f^m = \sum_{i=1}^r \tilde h_i(x)\, g_i(x) \in \mathfrak a,

Derivation hint: $f^m \cdot h_i(x,1/f)$ becomes an honest polynomial $\tilde h_i \in R$ once $m \geq \deg_y$ -contribution. with $\tilde h_i \in R$ . Therefore $f^m \in \mathfrak a$ , i.e. $f \in \sqrt{\mathfrak a}$ . This proves $I(V(\mathfrak a)) \subseteq \sqrt{\mathfrak a}$ , completing the equality. $\quad\blacksquare$

∎

mathematician · David Hilbert | 1862–1943

Hilbert proved the Nullstellensatz in his foundational 1893 paper Über die vollen Invariantensysteme, where it appeared as a tool in invariant theory alongside the Basis Theorem and the Syzygy Theorem. His revolutionary move was to abandon explicit computation of invariants in favor of existence arguments — provoking Paul Gordan's famous quip "Das ist nicht Mathematik, das ist Theologie." The Nullstellensatz, the Basis Theorem, and Hilbert's later work on number fields and logic reshaped the foundations of mathematics; the algebra–geometry dictionary it launched matured fully only with Zariski, Weil, and Grothendieck.

🧠 Intuition

Think of $V$ and $I$ as a "spelling" and "reading" pair. The geometry (a set of points) is what you see; the ideal is the spelling of polynomials that describe it. Many different ideals can describe the same set — for instance $(x)$ and $(x^2)$ both have zero locus $\{x = 0\}$ on the line. The radical operation is a normalization: it strips away the spurious multiplicity ( $x^2 \rightsquigarrow x$ ) that the geometry simply cannot detect. The Nullstellensatz guarantees this is the whole discrepancy: once you pass to radicals, ideals and geometry are in perfect bijection. The weak form is the load-bearing wall. It says "a system of polynomial equations with no algebraic obstruction (i.e. you cannot derive $1 = 0$ from them) actually has a solution" — provided you allow coordinates in an algebraically closed field. This is the multivariate, finite-degree analogue of the Fundamental Theorem of Algebra, which is exactly the case $n = 1$ .

RT

Reading the dictionary physically

A polynomial system is a set of constraints; the variety is the configuration space satisfying them (think kinematic constraints of a linkage, or conserved-quantity level sets). The weak form is a consistency theorem: either the constraints are jointly satisfiable, or you can algebraically combine them to produce the contradiction $1=0$ — there is no third option over $\bar k$ . The radical captures the idea that physical observables only "see" the actual solution set, not the algebraic redundancy in how you wrote the equations.

Figure

The radical collapses multiplicity:

V(x^2)=V(x)

as point sets, but the ideals differ

ExampleRecovering a radical from a generated ideal

Given: $k = \mathbb{C}$ , $R = \mathbb{C}[x,y]$ , and the ideal $\mathfrak a = (x^2,\, xy)$ . Find $V(\mathfrak a)$ , then $I(V(\mathfrak a))$ , and verify the Nullstellensatz.

Step 1 — Compute $V(\mathfrak a)$ . A point $(a,b)$ lies in $V(\mathfrak a)$ iff $a^2 = 0$ and $ab = 0$ , i.e. $a = 0$ (and $b$ arbitrary). So

12

\quad V(\mathfrak a) = \{(0,b) : b \in \mathbb{C}\} = V(x) = \text{the } y\text{-axis}.

Derivation hint: $x^2=0 \Rightarrow x=0$ in a field; then $xy=0$ is automatic.

Step 2 — Compute $I(V(\mathfrak a))$ . The vanishing ideal of the $y$ -axis is exactly $(x)$ , since a polynomial vanishes on all $(0,b)$ iff it has no term free of $x$ , iff it is divisible by $x$ .

Step 3 — Compute $\sqrt{\mathfrak a}$ directly. Clearly $x \in \sqrt{\mathfrak a}$ since $x^2 \in \mathfrak a$ . And $\sqrt{\mathfrak a} \subseteq \sqrt{(x)} = (x)$ since $\mathfrak a \subseteq (x)$ . Hence $\sqrt{\mathfrak a} = (x)$ .

Step 4 — Verify. $I(V(\mathfrak a)) = (x) = \sqrt{\mathfrak a}$ . ✓

\boxed{V(\mathfrak a) = y\text{-axis},\qquad I(V(\mathfrak a)) = \sqrt{(x^2,xy)} = (x).}

Note $\mathfrak a \neq \sqrt{\mathfrak a}$ : the original ideal is not radical, and the geometry only recovered its radical.

ExampleA system with no solutions — weak form in action

Given: $k = \mathbb{C}$ , the system $x^2 + y^2 = 1$ , $x^2 + y^2 = 4$ in $\mathbb{C}[x,y]$ . Decide solvability and exhibit the algebraic certificate.

Step 1 — Form the ideal. $\mathfrak a = (x^2 + y^2 - 1,\; x^2 + y^2 - 4)$ .

Step 2 — Detect inconsistency. Subtract the generators:

13

\quad (x^2 + y^2 - 1) - (x^2 + y^2 - 4) = 3 \in \mathfrak a.

Derivation hint: An ideal is closed under subtraction; the variables cancel exactly.

Step 3 — Conclude $\mathfrak a = R$ . Since $3 \in \mathfrak a$ and $3$ is a unit in $\mathbb{C}$ , we get $1 = \tfrac13\cdot 3 \in \mathfrak a$ , so $\mathfrak a = (1)$ .

Step 4 — Apply the weak Nullstellensatz. By [4], $\mathfrak a = (1) \iff V(\mathfrak a) = \varnothing$ .

\boxed{\text{No common solution exists; certificate: } 1 = \tfrac13(g_1 - g_2),\ g_i \text{ the generators.}}

The two "concentric circles of different radii don't meet" is, algebraically, the derivation of $1 = 0$ .

⚠️ Common mistake

The hypothesis $k = \bar k$ is essential — do not drop it. Over a non-algebraically-closed field both forms can fail spectacularly. The map "maximal ideals ↔ points" breaks: there are maximal ideals with residue field a proper extension of $k$ . The weak form fails because proper ideals can have empty zero locus in $k^n$ while still being proper. Always check the field is algebraically closed before invoking [4] or [5]. A second trap: the Nullstellensatz says $I(V(\mathfrak a)) = \sqrt{\mathfrak a}$ , not $I(V(\mathfrak a)) = \mathfrak a$ . Forgetting the radical leads to false claims about ideal equality.

CounterexampleFailure over

\mathbb{R}

Take $k = \mathbb{R}$ , $R = \mathbb{R}[x]$ , $\mathfrak a = (x^2 + 1)$ . Then

14

\quad V(\mathfrak a) = \{a \in \mathbb{R} : a^2 + 1 = 0\} = \varnothing,

Derivation hint: $x^2+1$ has no real root; its complex roots $\pm i \notin \mathbb R$ . yet $\mathfrak a = (x^2+1) \neq (1)$ is a proper (indeed maximal!) ideal. The weak form [4] is violated. Correspondingly, $I(V(\mathfrak a)) = I(\varnothing) = (1) \neq \sqrt{x^2+1} = (x^2+1)$ , so the strong form fails too. The fix is to pass to $\mathbb{C}$ , where $V(x^2+1) = \{i, -i\} \neq \varnothing$ .

CounterexampleWhy the radical, not the ideal

Over $\mathbb{C}$ , take $\mathfrak a = (x^2) \subseteq \mathbb{C}[x]$ . Then $V(\mathfrak a) = \{0\}$ and $I(V(\mathfrak a)) = I(\{0\}) = (x)$ . Here $I(V(\mathfrak a)) = (x) = \sqrt{(x^2)} \supsetneq (x^2) = \mathfrak a$ . The double point's scheme structure (the nilpotent $\bar x$ in $\mathbb{C}[x]/(x^2)$ ) is invisible to the variety; only scheme theory recovers it. This is precisely why classical varieties lose information that schemes retain.

connection

The Nullstellensatz is the gateway to the entire algebra–geometry dictionary. Combined with Prime Ideals and Irreducible Varieties, it shows irreducible algebraic sets correspond to prime ideals, and points to maximal ideals. It is the classical shadow of Spectrum of a Ring and Affine Schemes: $\operatorname{Spec} R$ replaces the maximal-ideal point set with all prime ideals, eliminating the need for $k = \bar k$ and recovering the nilpotent information lost in Reduced Rings and Radicals. The bijection [6] is the affine case of the equivalence between Reduced Finite-Type k-Algebras and affine varieties, an anti-equivalence of categories. Zariski's Lemma [7] connects to Noether Normalization and Integral Extensions, and the Hilbert Basis Theorem (used for Noetherianity) ties to Noetherian Rings. The single-variable case is exactly the Fundamental Theorem of Algebra.

🔬 Where this appears

⏳ History

- 1799 / 1815 — Gauss proves the Fundamental Theorem of Algebra, the one-variable seed of the Nullstellensatz.

- 1882 — Kronecker and Dedekind–Weber develop algebraic approaches to curves and the arithmetic of ideals, framing the question of zero loci ideal-theoretically.

- 1890 — Hilbert proves the Basis Theorem: every ideal in a polynomial ring is finitely generated, supplying the Noetherian foundation.

- 1893 — Hilbert proves the Nullstellensatz in Über die vollen Invariantensysteme (Math. Ann. 42), as part of his nonconstructive revolution in invariant theory.

- 1926 — Emmy Noether and her school recast the theorem in fully abstract ideal-theoretic language; Noether normalization (1926) gives the modern algebraic engine.

- 1947 — Oscar Zariski systematizes the topology and the algebra–geometry correspondence; Zariski's Lemma becomes the standard route to the weak form.

- 1953 — J. L. Rabinowitsch's one-line "trick" (introducing the variable

y

) for deriving the strong form from the weak form becomes textbook standard.

- 1960s — Grothendieck's Éléments de géométrie algébrique subsumes the Nullstellensatz into scheme theory via

\operatorname{Spec}

, removing the

k = \bar k

hypothesis and recovering nilpotents.

- 1974 — Brownawell, Kollár, and others (through the 1980s–90s) establish effective Nullstellensatz degree bounds:

f^m = \sum h_i g_i

with

m

and

\deg h_i

bounded explicitly in terms of degrees and number of variables.

open-problem

Effective bounds. Kollár (1988) proved that if $f \in \sqrt{(g_1,\dots,g_r)}$ with $\deg g_i \le d$ and $r,n \ge 2$ , one may take $m \le d^n$ in $f^m \in \mathfrak a$ , with matching examples showing $d^n$ is essentially sharp. Determining optimal constants and degree bounds for the Bézout-type identity [11], and the analogous arithmetic Nullstellensatz controlling the heights (sizes of coefficients) of the multipliers $\tilde h_i$ , remains an active area (Krick–Pardo–Sombra and successors).

Real and positive analogues. Over $\mathbb{R}$ the right framework is the Positivstellensatz (Krivine–Stengle) and the Real Nullstellensatz, where the radical is replaced by the real radical $\sqrt[\mathbb{R}]{\mathfrak a}$ . Sharp degree bounds for these sum-of-squares certificates, central to polynomial optimization, are far less understood than the complex case.

Differential and difference settings. The Ritt–Raudenbush theory provides Nullstellensätze for differential polynomial rings, but effective and constructive versions remain delicate, with connections to decidability questions in model theory.

Affine Varieties and the Nullstellensatz

Formal Statement

Why It Holds / Derivation

Building Intuition

Worked Examples

Common Misconceptions

Connections

Real-World Applications

Historical Development

Subtleties & Limits

See Also