Motivation
A quotient can be easier to understand than the module from which it came. For a module over a local ring, reducing modulo the maximal ideal produces a vector space. A basis of that vector space suggests a small collection of generators of the original module. Why should those lifts generate? The missing ingredient is finite generation: Nakayama's lemma says that, under the right hypothesis, a finite module cannot be hidden entirely inside its own multiples by the maximal ideal.
Suppose the images of some elements span the quotient. For each , we can subtract a linear combination of those elements and leave a remainder in . This is useful, but the remainder has not vanished. Repeating the argument would only place successive errors in higher powers of ; it would not by itself produce a finite expression for . Nakayama supplies the missing finite argument. It avoids assuming that an infinite series converges or that the intersection of the powers of is zero.
There are two different questions to keep separate. First, does a chosen family generate? Second, are its expressions unique? The quotient test answers the first for finite modules. Projectivity, used later, gives an additional splitting that lets us answer the second in a particular case.
Definitions
Throughout this section is a commutative ring with identity and modules are unital.
Definition 1.4.1
(Jacobson radical)
The Jacobson radical is the intersection of all maximal ideals of .
An ideal satisfies precisely when is a unit for every : if is not a unit it lies in a maximal ideal, and if also lies there then that ideal contains ; the converse follows by choosing, for outside a maximal ideal , an element with ; then is not a unit although .
Theorem 1.4.2
(Nakayama's lemma)
Let be a finitely generated -module and let be an ideal. If , then . Consequently, if and , then .
The second assertion applies the first to the finitely generated quotient : the displayed equality says . It does not require to be finitely generated.
Proof: the determinant trick
Proof
(Inverting )
Choose generators of . Because each belongs to , there are coefficients such that
Let and write for the column of generators. Then . Multiplication by the adjugate matrix gives , where . Expanding the determinant shows for some : every nonconstant term contains a coefficient of . Since , is a unit. Hence every is zero, and so is .
Proof X-ray
Where finite generation and the radical enter
Why does finite generation enter exactly here? Each of finitely many generators has an expression involving finitely many elements of . Expressing those elements again in the fixed finite generating family produces a single finite square matrix . Without a finite family there is no finite determinant to invert; the later example over shows that this is a genuine obstruction, not a cosmetic feature of the proof.
The radical condition is also doing a precise job. The determinant is congruent to modulo , but congruence to modulo an arbitrary ideal does not imply invertibility. For instance, is congruent to modulo , yet is not a unit in . Containment in ensures that the particular determinant is a unit; it does not say that every element of is nilpotent.
Commutativity is used in the determinant and adjugate argument. The result has noncommutative versions, but one should not silently transfer this proof to them. The argument also gives a useful stronger conclusion before invoking the radical: if for a finitely generated module, some satisfies .
Lift a basis from the residue field
Let be a commutative local ring and its residue field. For a finitely generated module , the quotient is a finite-dimensional -vector space.
Theorem 1.4.3
(Minimal number of generators)
Elements generate if and only if their images span over . The smallest possible number of generators is therefore .
Proof
(Reduction and Nakayama)
The forward implication follows by reduction. Conversely, if the images span and , then every element of lies in . The local ring has , so the preceding consequence gives . Any generating family has spanning images and hence at least the vector-space dimension many members; lifting a basis gives a generating family of exactly that size.
For an explicit lifting procedure, choose a basis of the quotient and choose any representatives . For a given , write its class as and lift the scalars to . Then . Thus the quotient of by the submodule generated by the lifts is equal to its own -multiple. The lemma kills that quotient, all at once. Different choices of representatives may give different generating sets; no canonical lifting is asserted.
The word “minimal” here refers to the smallest possible number of generators. A generating set with redundant members can still have images spanning the quotient. Conversely, a basis of the quotient cannot be lifted to fewer than generators, because vectors are required to span an -dimensional vector space. The statement includes : a finite with is zero.
This counts generators, not relations. A lift of a residue-field basis need not be a basis of : it may satisfy nontrivial relations whose coefficients lie in .
Worked example 1.4.4
(Generators without freeness)
Take , with maximal ideal , and . Then , so the class of is a minimal generating set. But although in : this generator is not a free basis. Indeed has four elements, while a free rank-one -module has eight.
Finite projectives over local rings
Theorem 1.4.5
(Local projectives are free in the finite case)
If is a finitely generated projective module over a commutative local ring , then is free of rank .
Proof
(Splitting off the kernel)
Choose lifts of a -basis of . They induce a surjection . Projectivity gives a section, so with . Reduction modulo makes an isomorphism, so . Let be the projection associated with the splitting. For , write in the standard basis , with ; then . Thus . The module is finitely generated, since it is the image of the finite free module under . Nakayama gives , and is an isomorphism.
Proof X-ray
Why the projection is needed
The inclusion alone is not enough to apply Nakayama: its conclusion concerns , not merely containment in a larger module. The projection is the bridge. Applying it preserves the coefficients in and moves the other factors into . This also explains where projectivity is used: without a section there need not be such a projection, and the kernel need not vanish.
Worked example 1.4.6
(An explicit relation matrix)
Let , the rationals with denominator not divisible by , and set . Modulo , the relation becomes , so has dimension one and needs one generator. In fact the relation gives . The map is a well-defined isomorphism , with inverse . The quotient computation predicts the number of generators; the explicit map additionally proves freeness.
Worked example 1.4.7
(A matrix whose columns generate over a local ring)
Let and consider , in . Their reductions are the standard basis of , so they generate . The actual calculation is instructive: the matrix has determinant , a unit in because its numerator is not divisible by . Hence
For example, . The local criterion predicts generation before we calculate the coefficients; the inverse matrix makes the result concrete. In contrast, over , the same matrix has determinant , which is not a unit, and its columns do not generate . For instance, solving forces , not an integer. Reduction modulo one prime alone cannot certify generation over a general nonlocal ring.
Reading a presentation modulo the maximal ideal
Suppose , where is a submodule of ; need not itself be finitely generated. Reduction gives a natural isomorphism
The right side is the quotient of by the image of . When a finite matrix presents , reduce its entries modulo and compute its rank by ordinary row reduction over . If this rank is , the module needs exactly generators. The number of rows of the presentation is not necessarily the number of independent relations after reduction: zero rows and dependent relations contribute no further rank. This calculation determines a minimal number of generators; it does not classify the module or prove that it is free.
For instance, let and let be generated by and in . The reduced relation matrix is zero, so and two generators are necessary and sufficient. Nevertheless , so this nonzero module is not free over . Compare the earlier relation : its reduction has rank one, and only one generator is needed. The crucial data are the reduced relations themselves, not merely the number of displayed relations.
Why the hypotheses matter
Counterexample mode
Dropping either hypothesis of Nakayama
If is not contained in , finite generation does not suffice. Take , and . Multiplication by two is surjective on , so , but .
Finite generation cannot be dropped either. Over the local ring , the nonzero module satisfies , even though is the maximal ideal. If were finitely generated, Nakayama would force it to vanish. In particular, does not detect this non-finite module.
Common mistakes
Common mistake
(A basis after reduction is already a basis upstairs)
The lifted family generates, but could have relations. The module over gives an explicit relation. To obtain freeness for , the proof used projectivity and showed that the kernel of a free presentation vanishes.
Common mistake
(An ideal inside the Jacobson radical must be nilpotent)
In , the maximal ideal lies in the radical but no power of it is zero. The determinant proof uses invertibility of , not nilpotence of .
Common mistake
(Local freeness says every projective module is globally free)
For , the summand is projective and finitely generated, but it is not free over : its two coordinate vector-space dimensions are and , whereas those of a free module are equal. The ring is not local, so the theorem does not apply globally.
Exercises with solutions
Quick check
(Over , let and . Determine and decide whether .)
Solution · Answer
In , multiplication by is multiplication by , so . The maximal ideals of are and , whose intersection is ; hence . Nakayama does not say .
Quick check
(Let be local and finite. Prove that an endomorphism is surjective if its induced map on is surjective.)
Solution · Answer
Surjectivity modulo gives , so Nakayama gives . The conclusion follows from finite generation of , without assuming that the image has a prescribed set of generators.
Quick check
(Let and . Find the minimal number of generators and exhibit a relation showing is not free.)
Solution · Answer
Reduction modulo gives , so the minimum is two. Every element of is killed by , while a nonzero free -module has an element not killed by . Thus is not free.
Quick check
(Let be local, finite and generate . Suppose the image of is a linear combination of the images of the other in . Prove that can be removed while the remaining elements still generate.)
Solution · Answer
Set . The given linear dependence shows that the images of still span , since the original images span. By the generator criterion, . Equivalently, and Nakayama applies to . Thus an inclusion-minimal finite generating set has linearly independent images; together with spanning, those images form a basis.