Writing down finitely many generators is only the beginning of describing a module. There may still be infinitely many independent relations between them. The Noetherian condition controls this second difficulty: it makes every submodule, including a module of relations, finitely generated. This chapter develops that foundational consequence before using it to prove that adjoining finitely many polynomial variables preserves the condition.
Use module actions and quotients, free modules and exact sequences, and the chain-condition discussion in composition series.
What finite generation does not control
For a ring with identity, a generating list gives a surjection
Its kernel consists of the relations satisfied by this list. Finding generators for and finding generators for are different tasks. The first task says every module element has a finite description in the selected symbols. The second says every equality between those descriptions follows from a fixed finite list of relations.
Unless a section explicitly restricts the scalars, rings may be noncommutative and modules are unital left modules. The polynomial-ring theorem below is stated for commutative rings.
Definition 1.2.1
(Noetherian modules and rings)
A module is Noetherian if every ascending sequence of submodules eventually becomes constant. A ring is left Noetherian if its regular left module is Noetherian; equivalently, its left ideals satisfy this condition. For a commutative ring we simply say Noetherian.
The useful equivalent formulation is that every submodule of is finitely generated. For completeness, an endlessly increasing chain can be built from a submodule that has no finite generating set: choose a new element outside the span of those already chosen at each step. Conversely, the union of an ascending chain is a submodule. If that union has finitely many generators, they all belong to one member of the chain, after which no further increase is possible. These are ordinary classical arguments, with the successive choices understood.
In particular, a Noetherian module is finitely generated. The converse depends on the ring. A cyclic module over a poorly behaved ring can contain a submodule needing infinitely many generators; we will construct one explicitly.
Finiteness through an exact sequence
Theorem 1.2.2
(Noetherian modules are closed under extensions)
In a short exact sequence of left -modules
is Noetherian if and only if both and are Noetherian. No commutativity or splitting hypothesis is required.
Proof
(Generators of the kernel and lifts of the image)
Identify with its image in . If is Noetherian, each submodule of is also a submodule of , so is finitely generated. For a submodule , its inverse image is finitely generated, and the images of those generators generate . This proves the forward direction.
For the converse, let be any submodule. The submodule has generators . The image has generators ; choose with . For any , write . Then
Express this difference using the . Thus generate . Because was arbitrary, is Noetherian. The proof does not construct a section of : the lifts of the selected generators need not satisfy all the relations they satisfy in .
Proof X-ray
Lifting generators leaves a controlled error
Generators of the image explain an element only modulo the kernel. Subtracting their lifts moves the unexplained part into the kernel, where a second finite list handles it. This is the same mechanism that proves an extension of two finitely generated modules is finitely generated. Applying that mechanism to every submodule proves the stronger Noetherian assertion.
Finite direct sums are consequently Noetherian: use the split exact sequence with a summand and the remaining summands, then induct. Infinite direct sums need not be Noetherian. For example, the subspaces spanned by the first standard vectors in form a strictly increasing chain, even though the scalar field is Noetherian.
From a Noetherian ring to finite relations
Theorem 1.2.3
(The ring controls all finitely generated modules)
For a unital ring , the following conditions are equivalent:
- is left Noetherian.
- Every finitely generated left -module is Noetherian.
When these conditions hold, every finitely generated left module has a finite presentation.
Proof
(Quotients of finite free modules)
Assume the first condition. Finite direct sums show that is Noetherian. A finitely generated module is a quotient of some , and the exact-sequence theorem shows its quotient is Noetherian. Conversely, apply the second condition to the cyclic left module itself.
For the final assertion, choose a finite generating surjection . Since is Noetherian, has a finite generating list . Sending the standard basis of to these vectors gives an exact sequence
This is a finite presentation: there are finitely many generators and finitely many generating relations. There is no assertion that is injective. Relations can themselves satisfy relations, a question addressed by free resolutions in later chapters.
For noncommutative , interpret as the specified left-linear map determined by its values on basis vectors. This avoids silently using a matrix convention that puts left-module coefficients on the wrong side. Over a commutative ring the usual column-matrix description applies directly.
Why the generating list does not decide finite presentation
Finite presentation is a property of , rather than of a lucky list of generators. We prove the necessary independence without assuming the ring Noetherian.
Proposition 1.2.4
(Changing a finite generating list preserves finite relations)
Suppose is surjective and is finitely generated. For every surjection with finite, is finitely generated.
Proof
(Transporting relations between two generating lists)
Lift the images of the finitely many basis vectors to obtain left-linear maps
Let generate . The vectors belong to . So do all , because . If , then , and
The first term is generated by the and the second by the finitely many . These vectors therefore generate . The proof permits zero generators and zero modules; an empty generating list denotes the zero submodule.
This result is useful when a simple presentation has an obviously infinite relation kernel. It rules out the possibility that another finite generating list could somehow hide that obstruction.
Worked example 1.2.5
(A redundant generator creates a new relation)
For , the quotient map , , has kernel . Now choose the generating list , giving , . In the preceding proof we may take and . The old relation gives ; the correction gives zero on and on . Thus
Directly, if , then . The new relation says that the second chosen generator is twice the first. Adding a redundant generator has changed the relation list, while preserving the underlying module and finite presentation.
A polynomial ideal with one generating relation
Worked example 1.2.6
(The relation between x and y)
Let be a field, , and . The surjection sends to . Its kernel is exactly .
Indeed, if , reduce modulo . In , the resulting equation is . Since is a domain and , , so for some . Substituting gives in the domain , whence . Conversely, every maps to zero. Moreover, is injective because forces . We have the exact sequence
The ideal is not cyclic. If , then divides both and . Additivity of total degree for nonzero products over a field implies either is a nonzero constant, making the ideal all of , or has degree one and both and are constant multiples of it. The first possibility contradicts evaluation at , and the second contradicts linear independence of and over .
Thus a finite presentation is a description by generators and equations; it does not turn the chosen generators into a basis. Here the relation is visible before any general finiteness theorem is invoked. The next theorem guarantees analogous finite descriptions for arbitrary finitely generated modules over polynomial rings in finitely many variables, though it does not automatically compute them.
Hilbert basis theorem: controlling the highest degree
For this section is commutative with identity. Its elements may include zero divisors. If is the zero ring, all the assertions are immediate, so the proof may assume otherwise.
Theorem 1.2.7
(Hilbert basis theorem)
If is Noetherian, then is Noetherian. Consequently is Noetherian for every finite , and every commutative -algebra generated by finitely many elements is Noetherian.
Proof
(Canceling high-degree leading terms)
Let be an ideal. We must find finitely many polynomials generating it. Define to consist of zero and the leading coefficients of the nonzero polynomials in .
First, is an ideal. Given leading coefficients , multiply the corresponding polynomials by suitable powers of to make their degrees equal. Their sum has leading coefficient if this is nonzero, while zero already belongs to . Negatives are treated by negating a polynomial. If , multiplying a polynomial of leading coefficient by has leading coefficient whenever ; if there is nothing to prove. This argument allows zero divisors.
If , then . Otherwise choose nonzero generators of and polynomials with leading coefficient . Put and . Whenever has degree , write its leading coefficient as . The polynomial
belongs to and either is zero or has degree strictly below . Every exponent used here is nonnegative. We have canceled a leading coefficient by expressing it as a combination, without dividing by any coefficient.
Repeated reduction therefore stops after finitely many steps. It expresses as a polynomial combination of the , plus an element of of degree less than . But those remaining low-degree elements still need to be accounted for.
Hilbert basis theorem: retaining the low-degree remainder
Proof
(Generating the bounded-degree remainder)
The polynomials of degree less than , together with zero, form a free -module with basis . This is a module, not usually an ideal. When , it means the zero module. The intersection
is an -submodule of that finite free module. By the Noetherian module theorem, has finite -module generators . Combining them with the reductions proves
Both containments deserve attention. All listed polynomials lie in , giving one containment. For the other, degree reduction leaves an -linear combination of the , which is also an -linear combination. Thus every element of is in the displayed ideal. This proves Noetherian.
Iterating gives the finite-variable assertion, starting with variables, which is just . A commutative algebra generated over by is the image of under evaluation . A quotient of a Noetherian ring is Noetherian, because ascending chains of quotient ideals lift to chains of ideals containing the kernel. This proves the last assertion.
Proof X-ray
A degree reduction is not yet a whole proof
Canceling every sufficiently high leading term proves that a finite list handles high degrees. It says nothing by itself about smaller degrees. The finite free coefficient module handles precisely that remainder. The argument needs both the leading-coefficient ideal and Noetherian control of the bounded-degree part; it never assumes a division algorithm with invertible leading coefficients.
Worked example 1.2.8
(A remainder that must not disappear from the proof)
In , let . Its leading-coefficient ideal is all of , since has leading coefficient one. We can choose , giving . However, does not contain . Direct expansion of shows that . Its module generator is the missing low-degree generator. The proof correctly recovers both generators instead of discarding the remainder.
Where finite generation fails to imply finite relations
Counterexample mode
A cyclic module with infinitely many necessary relations
Let be the polynomial ring in countably many variables over a field, where each individual polynomial involves finitely many variables. Let .
Suppose . Choose large enough that these finitely many polynomials involve only . Each has zero constant term. The homomorphism setting those first variables to zero therefore kills every and every element of their ideal, but it does not kill . This is a contradiction. Hence is not finitely generated.
The -module is cyclic but is not Noetherian. The cyclic module is not finitely presented: its natural one-generator surjection has kernel , and generating-list independence excludes every other finite presentation. The Hilbert basis theorem concerns finitely many adjoined variables, and does not apply to this ring.
The connection with finite-group representations
Let be finite and any field. The group algebra is finite-dimensional over . Each left ideal is a vector subspace, so a finite vector-space basis also generates it as a left ideal. Thus is left Noetherian, regardless of the characteristic of .
An -module generated by elements is a quotient of and consequently has -dimension at most . Conversely, a finite vector-space basis generates it as an -module. Finite-dimensional representations therefore have finite presentations over , and all their subrepresentations are finitely generated.
This conclusion supplies finiteness, not invariant complements. In characteristic dividing , a finite-dimensional representation can still be a nonsplit extension. Noetherian arguments ensure that the generators and relations terminate; Maschke's theorem asks an additional and different question about splitting those relations into invariant summands.
Common mistakes
Common mistake
(Noetherian does not mean finite-dimensional)
A Noetherian module need not be finite as a set or finite-dimensional over a field. The module over itself is cyclic and Noetherian, but its underlying -vector space has the infinite independent list . Its descending chain also shows that it is not Artinian and has no finite composition length.
Common mistake
(Counting generators minus relations)
A finite presentation need not be minimal, and the map giving its relations need not be injective. Do not conclude that counting written generators minus written relations computes a rank over a general ring. Even a redundant zero relation changes that count without changing the module.
Common mistake
(Existence of generators is not an algorithm)
The Hilbert basis theorem asserts existence of finite generators. The proof here does not supply a uniform algorithm for arbitrary coefficient rings, an effective ideal-membership test, or a bound on the number or degrees of generators. Those require additional hypotheses and arguments.
Exercises
Quick check
(Let be surjective, let , and suppose have finite generating lists. Prove that is finitely generated, without assuming any Noetherian hypothesis. Explain why this does not prove every submodule of is finitely generated.)
Solution
Lift each chosen generator of to , and combine these lifts with the generators of . Subtracting an appropriate combination of the lifts puts any element of in , proving generation. For an arbitrary submodule , the corresponding intersection and image need not be finitely generated merely because are. The extension theorem requires their Noetherian property to handle every such .
Quick check
(Find the kernel of , , and give an exact sequence presenting using this map. Is its relation map injective?)
Solution
If , reducing modulo two shows for some integer . Substitution gives . Thus the kernel is . Since , the map is surjective. The sequence is exact, and its relation map is injective because in implies .
Quick check
(Let be commutative Noetherian and let be finitely generated. Prove that is finitely generated as an -module. Specify how commutativity is used.)
Solution
Choose generators of . Evaluation defines an injective -linear map , : a homomorphism vanishing on all generators vanishes everywhere. The module is Noetherian, so its submodule, the image of evaluation, is finitely generated. Commutativity ensures that is again -linear, since . Over a noncommutative ring this Hom group is not generally a left -module with that action.
Quick check
(Let . Prove that is Noetherian but has infinite dimension over . Explain why being generated by finitely many elements as a -algebra does not imply finite generation as a -module.)
Solution
A field is Noetherian, so Hilbert's theorem makes Noetherian and its quotient Noetherian. The map setting is well defined and surjective. The elements in are linearly independent, because their images in are. Thus has infinite -dimension. Algebra generation allows products and arbitrarily high powers of its generators; module generation permits only scalar linear combinations of a fixed finite list.
Quick check
(Let . Prove that the sequence , with injection and quotient given by the constant coefficient, is exact but does not split as -modules. Give the action through .)
Solution
The injection has image , exactly the kernel of the constant-coefficient map, and it is injective. Both maps respect the specified action. If an -linear section existed, then for some . But , a contradiction. The algebra is finite-dimensional and Noetherian; neither fact supplies a splitting.
Quick check
(Show that is not a Noetherian -module. Identify the hypothesis missing from the theorem about finitely generated modules over a Noetherian ring.)
Solution
If finitely many rationals generated as an abelian group, take a positive common denominator . Their integer span is contained in , which does not contain . Thus is not finitely generated and hence not Noetherian. The scalar ring is Noetherian, but the theorem only applies to its finitely generated modules. Equivalently, the subgroups give a nonstabilizing ascending chain in .