Noether normalization lemma

In mathematics, the Noether normalization lemma is a result of commutative algebra, introduced by Emmy Noether in 1926.^[1] It states that for any field k, and any finitely generated commutative k-algebra A, there exist elements y₁, y₂, ..., y_d in A that are algebraically independent over k and such that A is a finitely generated module over the polynomial ring S = k [y₁, y₂, ..., y_d]. The integer d is equal to the Krull dimension of the ring A; and if A is an integral domain, d is also the transcendence degree of the field of fractions of A over k.

The theorem has a geometric interpretation. Suppose A is the coordinate ring of an affine variety X, and consider S as the coordinate ring of a d-dimensional affine space ${\displaystyle \mathbb {A} _{k}^{d}}$. Then the inclusion map ${\displaystyle S\hookrightarrow A}$ induces a surjective finite morphism of affine varieties ${\displaystyle X\to \mathbb {A} _{k}^{d}}$: that is, any affine variety is a branched covering of affine space. When k is infinite, such a branched covering map can be constructed by taking a general projection from an affine space containing X to a d-dimensional subspace.

More generally, in the language of schemes, the theorem can equivalently be stated as: every affine k-scheme (of finite type) X is finite over an affine n-dimensional space. The theorem can be refined to include a chain of ideals of R (equivalently, closed subsets of X) that are finite over the affine coordinate subspaces of the corresponding dimensions.^[2]

The Noether normalization lemma can be used as an important step in proving Hilbert's Nullstellensatz, one of the most fundamental results of classical algebraic geometry. The normalization theorem is also an important tool in establishing the notions of Krull dimension for k-algebras.

Proof

The following proof is due to Nagata, following Mumford's red book. A more geometric proof is given on page 127 of the red book.

The ring A in the lemma is generated as a k-algebra by some elements ${\displaystyle y_{1},...,y_{m}}$. We shall induct on m. Case ${\displaystyle m=0}$ is ${\displaystyle k=A}$ and there is nothing to prove. Assume ${\displaystyle m=1}$. Then ${\displaystyle A\cong k[y]/I}$ as k-algebras, where ${\displaystyle I\subset k[y]}$ is some ideal. Since ${\displaystyle k[y]}$ is a PID (it is a Euclidean domain), ${\displaystyle I=(f)}$. If ${\displaystyle f=0}$ we are done, so assume ${\displaystyle f\neq 0}$. Let e be the degree of f. Then A is generated, as a k-vector space, by ${\displaystyle 1,y,y^{2},\dots ,y^{e-1}}$. Thus A is finite over k. Assume now ${\displaystyle m\geq 2}$. It is enough to show that there is a k-subalgebra S of A that is generated by ${\displaystyle m-1}$ elements, such that A is finite over S. Indeed, by the inductive hypothesis, we can find algebraically independent elements ${\displaystyle x_{1},...,x_{d}}$ of S such that S is finite over ${\displaystyle k[x_{1},...,x_{d}]}$.

Since otherwise there would be nothing to prove, we can also assume that there is a nonzero polynomial f in m variables over k such that

${\displaystyle f(y_{1},\ldots ,y_{m})=0}$.

Given an integer r which is determined later, set

${\displaystyle z_{i}=y_{i}-y_{1}^{r^{i-1}},\quad 2\leq i\leq m.}$

Then the preceding reads:

${\displaystyle f(y_{1},z_{2}+y_{1}^{r},z_{3}+y_{1}^{r^{2}},\ldots ,z_{m}+y_{1}^{r^{m-1}})=0}$.

Now, if ${\displaystyle ay_{1}^{\alpha _{1}}\prod _{2}^{m}(z_{i}+y_{1}^{r^{i-1}})^{\alpha _{i}}}$ is a monomial appearing in the left-hand side of the above equation, with coefficient ${\displaystyle a\in k}$, the highest term in ${\displaystyle y_{1}}$ after expanding the product looks like

${\displaystyle ay_{1}^{\alpha _{1}+r\alpha _{2}+\cdots +\alpha _{m}r^{m-1}}.}$

Whenever the above exponent agrees with the highest ${\displaystyle y_{1}}$ exponent produced by some other monomial, it is possible that the highest term in ${\displaystyle y_{1}}$ of ${\displaystyle f(y_{1},z_{2}+y_{1}^{r},z_{3}+y_{1}^{r^{2}},...,z_{m}+y_{1}^{r^{m-1}})}$ will not be of the above form, because it may be affected by cancellation. However, if r is larger than any exponent appearing in f, then each ${\displaystyle \alpha _{1}+r\alpha _{2}+\cdots +\alpha _{m}r^{m-1}}$ encodes a unique base r number, so this does not occur. For such an r, let ${\displaystyle c\in k}$ be the coefficient of the unique monomial of f of multidegree ${\displaystyle (\alpha _{1},\dots ,\alpha _{m})}$ for which the quantity ${\displaystyle \alpha _{1}+r\alpha _{2}+\cdots +\alpha _{m}r^{m-1}}$ is maximal. Multiplication of the last identity by ${\displaystyle 1/c}$ gives an integral dependence equation of ${\displaystyle y_{1}}$ over ${\displaystyle S=k[z_{2},...,z_{m}]}$, i.e., ${\displaystyle y_{1}}$ is integral over S. Since ${\displaystyle y_{i}=z_{i}+y_{1}^{r^{i-1}}}$ are also integral over that ring, A is integral over S. It follows A is finite over S, and since S is generated by m-1 elements, by the inductive hypothesis we are done.

If A is an integral domain, then d is the transcendence degree of its field of fractions. Indeed, A and ${\displaystyle S=k[y_{1},...,y_{d}]}$ have the same transcendence degree (i.e., the degree of the field of fractions) since the field of fractions of A is algebraic over that of S (as A is integral over S) and S has transcendence degree d. Thus, it remains to show the Krull dimension of the polynomial ring S is d. (This is also a consequence of dimension theory.) We induct on d, with the case ${\displaystyle d=0}$ being trivial. Since ${\displaystyle 0\subsetneq (y_{1})\subsetneq (y_{1},y_{2})\subsetneq \cdots \subsetneq (y_{1},\dots ,y_{d})}$ is a chain of prime ideals, the dimension is at least d. To get the reverse estimate, let ${\displaystyle 0\subsetneq {\mathfrak {p}}_{1}\subsetneq \cdots \subsetneq {\mathfrak {p}}_{m}}$ be a chain of prime ideals. Let ${\displaystyle 0\neq u\in {\mathfrak {p}}_{1}}$. We apply the noether normalization and get ${\displaystyle T=k[u,z_{2},\dots ,z_{d}]}$ (in the normalization process, we're free to choose the first variable) such that S is integral over T. By the inductive hypothesis, ${\displaystyle T/(u)}$ has dimension d - 1. By incomparability, ${\displaystyle {\mathfrak {p}}_{i}\cap T}$ is a chain of length ${\displaystyle m}$ and then, in ${\displaystyle T/({\mathfrak {p}}_{1}\cap T)}$, it becomes a chain of length ${\displaystyle m-1}$. Since ${\displaystyle \operatorname {dim} T/({\mathfrak {p}}_{1}\cap T)\leq \operatorname {dim} T/(u)}$, we have ${\displaystyle m-1\leq d-1}$. Hence, ${\displaystyle \dim S\leq d}$.

Refinement

The following refinement appears in Eisenbud's book, which builds on Nagata's idea:^[2]

Geometrically speaking, the last part of the theorem says that for ${\displaystyle X=\operatorname {Spec} A\subset \mathbf {A} ^{m}}$ any general linear projection ${\displaystyle \mathbf {A} ^{m}\to \mathbf {A} ^{d}}$ induces a finite morphism ${\displaystyle X\to \mathbf {A} ^{d}}$ (cf. the lede); besides Eisenbud, see also [1].

Illustrative application: generic freeness

A typical nontrivial application of the normalization lemma is the generic freeness theorem: Let ${\displaystyle A,B}$ be rings such that ${\displaystyle A}$ is a Noetherian integral domain and suppose there is a ring homomorphism ${\displaystyle A\to B}$ that exhibits ${\displaystyle B}$ as a finitely generated algebra over ${\displaystyle A}$. Then there is some ${\displaystyle 0\neq g\in A}$ such that ${\displaystyle B[g^{-1}]}$ is a free ${\displaystyle A[g^{-1}]}$-module.

To prove this, let ${\displaystyle F}$ be the fraction field of ${\displaystyle A}$. We argue by induction on the Krull dimension of ${\displaystyle F\otimes _{A}B}$. The base case is when the Krull dimension is ${\displaystyle -\infty }$; i.e., ${\displaystyle F\otimes _{A}B=0}$; that is, when there is some ${\displaystyle 0\neq g\in A}$ such that ${\displaystyle gB=0}$ , so that ${\displaystyle B[g^{-1}]}$ is free as an ${\displaystyle A[g^{-1}]}$-module. For the inductive step, note that ${\displaystyle F\otimes _{A}B}$ is a finitely generated ${\displaystyle F}$-algebra. Hence by the Noether normalization lemma, ${\displaystyle F\otimes _{A}B}$ contains algebraically independent elements ${\displaystyle x_{1},\dots ,x_{d}}$ such that ${\displaystyle F\otimes _{A}B}$ is finite over the polynomial ring ${\displaystyle F[x_{1},\dots ,x_{d}]}$. Multiplying each ${\displaystyle x_{i}}$ by elements of ${\displaystyle A}$, we can assume ${\displaystyle x_{i}}$ are in ${\displaystyle B}$. We now consider:

${\displaystyle A':=A[x_{1},\dots ,x_{d}]\to B.}$

Now ${\displaystyle B}$ may not be finite over ${\displaystyle A'}$, but it will become finite after inverting a single element as follows. If ${\displaystyle b}$ is an element of ${\displaystyle B}$, then, as an element of ${\displaystyle F\otimes _{A}B}$, it is integral over ${\displaystyle F[x_{1},\dots ,x_{d}]}$; i.e., ${\displaystyle b^{n}+a_{1}b^{n-1}+\dots +a_{n}=0}$ for some ${\displaystyle a_{i}}$ in ${\displaystyle F[x_{1},\dots ,x_{d}]}$. Thus, some ${\displaystyle 0\neq g\in A}$ kills all the denominators of the coefficients of ${\displaystyle a_{i}}$ and so ${\displaystyle b}$ is integral over ${\displaystyle A'[g^{-1}]}$. Choosing some finitely many generators of ${\displaystyle B}$ as an ${\displaystyle A'}$-algebra and applying this observation to each generator, we find some ${\displaystyle 0\neq g\in A}$ such that ${\displaystyle B[g^{-1}]}$ is integral (thus finite) over ${\displaystyle A'[g^{-1}]}$. Replace ${\displaystyle B,A}$ by ${\displaystyle B[g^{-1}],A[g^{-1}]}$ and then we can assume ${\displaystyle B}$ is finite over ${\displaystyle A':=A[x_{1},\dots ,x_{d}]}$. To finish, consider a finite filtration ${\displaystyle B=B_{0}\supset B_{1}\supset B_{2}\supset \cdots \supset B_{r}}$ by ${\displaystyle A'}$-submodules such that ${\displaystyle B_{i}/B_{i+1}\simeq A'/{\mathfrak {p}}_{i}}$ for prime ideals ${\displaystyle {\mathfrak {p}}_{i}}$ (such a filtration exists by the theory of associated primes). For each i, if ${\displaystyle {\mathfrak {p}}_{i}\neq 0}$, by inductive hypothesis, we can choose some ${\displaystyle g_{i}\neq 0}$ in ${\displaystyle A}$ such that ${\displaystyle A'/{\mathfrak {p}}_{i}[g_{i}^{-1}]}$ is free as an ${\displaystyle A[g_{i}^{-1}]}$-module, while ${\displaystyle A'}$ is a polynomial ring and thus free. Hence, with ${\displaystyle g=g_{0}\cdots g_{r}}$, ${\displaystyle B[g^{-1}]}$ is a free module over ${\displaystyle A[g^{-1}]}$. ${\displaystyle \square }$

Notes

References

• Eisenbud, David (1995), Commutative algebra. With a view toward algebraic geometry, Graduate Texts in Mathematics, vol. 150, Berlin, New York: Springer-Verlag, ISBN 3-540-94268-8, MR 1322960, Zbl 0819.13001
• "Noether theorem", Encyclopedia of Mathematics, EMS Press, 2001 [1994]. NB the lemma is in the updating comments.
• Noether, Emmy (1926), "Der Endlichkeitsatz der Invarianten endlicher linearer Gruppen der Charakteristik p", Nachrichten von der Gesellschaft der Wissenschaften zu Göttingen: 28–35, archived from the original on March 8, 2013

Further reading

• Robertz, D.: Noether normalization guided by monomial cone decompositions. Journal of Symbolic Computation 44(10), 1359–1373 (2009)