A module is meant to be the linear algebra of a ring, but the first surprise is that a module can be too large to describe by any finite list of vectors. A [vector space](/page/Vector%20Space) over a field is manageable when it has a finite basis: every vector is controlled by finitely many coordinates, and linear maps out of the space are determined by finitely many images. For a module over a general ring, bases may not exist, coordinates may not be unique, and spanning may be the only finite control available. The question behind finitely generated modules is therefore practical and structural at the same time: when can an $R$-module be built from finitely many elements using the scalars of $R$? This condition is weaker than being free of finite rank, but strong enough to support quotients, exact sequences, presentations, Nakayama-type arguments, and the basic finiteness language of commutative algebra. [example: The First Infinite Failure] Let $R$ be a nonzero ring, and consider the [direct sum](/page/Direct%20Sum) \begin{align*} M = \bigoplus_{i=1}^{\infty} R e_i. \end{align*} By definition of the direct sum, each element $m \in M$ has a unique expression \begin{align*} m = r_1e_1 + \cdots + r_de_d \end{align*} with only finitely many nonzero coefficients among the $r_i \in R$. We show that no finite list generates $M$. Let $m_1,\ldots,m_k \in M$. For each $j$, write \begin{align*} m_j = \sum_{i \in S_j} r_{ij} e_i \end{align*} where $S_j \subset \mathbb{N}$ is finite and $r_{ij} \in R$. The union \begin{align*} S = S_1 \cup \cdots \cup S_k \end{align*} is finite, so choose $N \notin S$. If $a_1,\ldots,a_k \in R$, then \begin{align*} a_1m_1 + \cdots + a_km_k = \sum_{j=1}^{k} \sum_{i \in S_j} a_j r_{ij} e_i. \end{align*} Every index appearing on the right lies in $S$, and $N \notin S$, so the $e_N$-component of this linear combination is $0$. But $e_N$ has $e_N$-component $1$, and $1 \ne 0$ because $R$ is nonzero. Hence $e_N$ is not in the submodule generated by $m_1,\ldots,m_k$. Thus every single element of $M$ has finite support, but no one finite stock of elements can generate all of $M$. [/example] This example isolates the main issue. Finiteness is not about each individual expression being finite; module operations already require finite sums. It is about whether one fixed finite stock of elements suffices for every expression in the module. ## Definition A [module](/page/Module) gives the ambient linear structure: addition inside $M$ and scalar multiplication by elements of $R$. The new condition answers a more economical question: can the entire module be reached from a fixed finite list, without asking for independence or unique coordinates? [definition: Finitely Generated Module] Let $R$ be a ring, and let $M$ be a left $R$-module. The module $M$ is finitely generated if there exist an integer $k \ge 0$ and elements $m_1, \ldots, m_k \in M$ such that \begin{align*} M = Rm_1 + \cdots + Rm_k = \{r_1m_1 + \cdots + r_km_k : r_1, \ldots, r_k \in R\}. \end{align*} The elements $m_1, \ldots, m_k$ are called a finite generating set for $M$. [/definition] When $k=0$, the displayed set is the empty sum construction, so the resulting module is the zero module. The definition does not require the list to be minimal. It also does not require the coefficients to be unique. To connect this spanning condition with maps and exact sequences, we need a formulation that replaces a list of generators by one canonical surjection from a finite free module. [quotetheorem:9966] This characterization is the bridge from spanning language to homological language. A finitely generated module is exactly a quotient of a finite free module. That viewpoint becomes important whenever a ring extension or algebra is being studied through its underlying module structure, because algebra generation and module generation answer different questions. Sometimes the object being studied is a ring $S$ equipped with a map from $R$, but the intended finiteness is still module-theoretic. The following definition prevents ambiguity by saying that multiplication in $S$ is not allowed as a generating operation; only $R$-linear combinations are allowed. [definition: Module-Finite Ring Extension] Let $\varphi: R \to S$ be a ring homomorphism. Give $S$ the left $R$-module structure defined by \begin{align*} r \cdot s = \varphi(r)s \end{align*} for $r \in R$ and $s \in S$. The ring $S$ is module-finite over $R$, or finitely generated over $R$ as an $R$-module, if this left $R$-module is finitely generated. [/definition] In the usual commutative unital algebra setting, module-finite is stronger than being finitely generated as an $R$-algebra. For instance, $R[x]$ is generated as an $R$-algebra by $x$, but it is usually not finitely generated as an $R$-module, because the powers $1,x,x^2,\ldots$ cannot be captured by finitely many $R$-linear generators. [example: Polynomial Rings Separate Two Finiteness Notions] Let $R$ be a nonzero commutative ring. As an $R$-algebra, $R[x]$ is generated by $x$: if \begin{align*} f(x)=a_0+a_1x+\cdots+a_nx^n \end{align*} with $a_i \in R$, then each term $a_ix^i$ is obtained from the scalar $a_i \in R$ and the generator $x$ by multiplication, and the polynomial is obtained by adding these finitely many terms. We show that the same ring is not finitely generated as an $R$-module. Suppose, toward a contradiction, that $f_1,\ldots,f_k \in R[x]$ generate $R[x]$ as an $R$-module. For each $j$, write \begin{align*} f_j=a_{j0}+a_{j1}x+\cdots+a_{jd_j}x^{d_j} \end{align*} with $a_{ji}\in R$ and $d_j \ge 0$, allowing $f_j=0$ by taking all displayed coefficients equal to $0$. Let $d=\max(d_1,\ldots,d_k)$. If $r_1,\ldots,r_k \in R$, then \begin{align*} r_1f_1+\cdots+r_kf_k=\sum_{j=1}^{k}r_j(a_{j0}+a_{j1}x+\cdots+a_{jd}x^d) \end{align*} where $a_{ji}=0$ for $i>d_j$. Expanding by distributivity, \begin{align*} r_1f_1+\cdots+r_kf_k=(\sum_{j=1}^{k}r_ja_{j0})+(\sum_{j=1}^{k}r_ja_{j1})x+\cdots+(\sum_{j=1}^{k}r_ja_{jd})x^d. \end{align*} Thus every $R$-linear combination of $f_1,\ldots,f_k$ has zero coefficient of $x^{d+1}$. But the polynomial $x^{d+1}$ has coefficient $1$ on $x^{d+1}$, and $1\ne 0$ because $R$ is nonzero. Hence $x^{d+1}$ is not in the $R$-span of $f_1,\ldots,f_k$, contradicting the assumption that they generate $R[x]$. So $R[x]$ is generated by one element as an $R$-algebra, but as an $R$-module it needs the infinite family $1,x,x^2,\ldots$ rather than any finite list. [/example] The definition is deliberately modest. It asks for finite spanning data, and much of algebra is about understanding how far that data can be pushed. ## Generators and Relations Finite generation tells us that the module is a quotient of $R^k$, but the quotient may identify many different coordinate vectors. The next problem is to distinguish the generators from the relations among them. This is where module theory starts to depart from finite-dimensional vector spaces. ### Generated Submodules Before relations can be named, we need a construction that starts with an arbitrary subset and returns the smallest submodule forced to contain it. This answers the local question behind finite generation: what exactly is produced by a chosen stock of elements? [definition: Submodule Generated by a Set] Let $R$ be a ring, let $M$ be a left $R$-module, and let $S \subset M$. The submodule generated by $S$ is \begin{align*} RS = \left\{\sum_{i=1}^{n} r_i s_i : n \ge 0,\ r_i \in R,\ s_i \in S\right\}, \end{align*} where the case $n=0$ is the empty sum $0$. For a finite set $S = \{m_1, \ldots, m_k\}$, this submodule is denoted $Rm_1 + \cdots + Rm_k$. [/definition] The phrase "generated by" is now literal: it means taking all finite $R$-linear combinations. The smallest possible nonzero spanning pattern is generation by a single element, and that case deserves separate language because it turns modules into quotients of the ring itself. ### Cyclic Modules When a module has one generator, every element is a scalar multiple of a single element. This case is small enough to classify in terms of annihilators, and it supplies the local model for general finite generation. [definition: Cyclic Module] Let $R$ be a ring, and let $M$ be a left $R$-module. The module $M$ is cyclic if there exists $m \in M$ such that \begin{align*} M = Rm. \end{align*} [/definition] Cyclic modules are the module-theoretic analogue of one-generator spanning spaces, but they need not behave like one-dimensional vector spaces. To understand what a cyclic generator loses, we need to record which scalars kill it. [definition: Annihilator of an Element] Let $R$ be a ring, let $M$ be a left $R$-module, and let $m \in M$. The annihilator of $m$ is \begin{align*} \operatorname{Ann}_R(m) = \{r \in R : rm = 0\}. \end{align*} [/definition] The annihilator is a left ideal of $R$, and it measures the failure of the map $R \to Rm$ given by $r \mapsto rm$ to be injective. Once this kernel has been named, the natural question is whether every cyclic module is exactly the corresponding quotient of $R$. [quotetheorem:9967] For several generators, the same idea persists: start from a finite free module and quotient by the kernel. The new object to name is that kernel, because it records all coefficient choices that collapse to zero in $M$. ### Relations and Presentations A list of generators gives a specific surjection from a finite free module. The elements of the kernel record all relations among the chosen generators, so naming this kernel separates the question of reaching elements from the question of representing them uniquely. [definition: Module of Relations] Let $R$ be a ring, let $M$ be a left $R$-module, and let $m_1, \ldots, m_k \in M$. The module of relations among $m_1, \ldots, m_k$ is the submodule \begin{align*} \operatorname{Rel}(m_1, \ldots, m_k) = \ker(\pi) \subset R^k, \end{align*} where $\pi: R^k \to M$ is the $R$-[module homomorphism](/page/Module%20Homomorphism) defined by \begin{align*} \pi(r_1, \ldots, r_k) = r_1m_1 + \cdots + r_km_k. \end{align*} [/definition] Relations explain why finite generation is weaker than finite rank freeness. Even if every element is reachable from finitely many generators, there may be many coefficient tuples representing the same element. [example: A Generated Module with a Relation] As a $\mathbb{Z}$-module, $\mathbb{Z}/6\mathbb{Z}$ is generated by $\bar{1}$ because every residue class has the form \begin{align*} \bar{n}=n\bar{1}. \end{align*} Define $\pi:\mathbb{Z}\to \mathbb{Z}/6\mathbb{Z}$ by $\pi(n)=n\bar{1}=\bar{n}$. This map is surjective by the displayed formula. Its kernel is \begin{align*} \ker(\pi)=\{n\in\mathbb{Z}:\bar{n}=\bar{0}\}. \end{align*} By the definition of congruence modulo $6$, the condition $\bar{n}=\bar{0}$ is equivalent to $6\mid n$, so \begin{align*} \ker(\pi)=\{6q:q\in\mathbb{Z}\}=6\mathbb{Z}. \end{align*} Thus the generator $\bar{1}$ satisfies the relation \begin{align*} 6\bar{1}=\bar{6}=\bar{0}. \end{align*} This module is finitely generated and cyclic. It is not a free $\mathbb{Z}$-module: if a free [abelian group](/page/Abelian%20Group) has basis $B$ and an element $x=\sum_{b\in F} a_b b$ with finite $F\subset B$ satisfies $6x=0$, then \begin{align*} 6x=\sum_{b\in F} 6a_b b=0 \end{align*} forces $6a_b=0$ for every $b\in F$ by uniqueness of coordinates in a basis, hence $a_b=0$ for every $b$ and $x=0$. But $\bar{1}\ne \bar{0}$ while $6\bar{1}=\bar{0}$, so $\mathbb{Z}/6\mathbb{Z}$ has nonzero torsion and cannot be free over $\mathbb{Z}$. [/example] Once relations are made explicit, a new obstruction appears: a module can be generated by finitely many elements while the relations among those generators require infinitely many independent pieces of data. To rule out that pathology, one asks for a finite free module of relations mapping into a finite free module of generators, with $M$ obtained as the cokernel. [definition: Finitely Presented Module] Let $R$ be a ring, and let $M$ be a left $R$-module. The module $M$ is finitely presented if there exist integers $m,n \ge 0$ and an exact sequence of left $R$-modules \begin{align*} R^m \to R^n \to M \to 0. \end{align*} [/definition] Finite presentation says that both generators and relations are finite. Since a finite presentation ends in a surjection from a finite free module, the immediate question is how it compares with finite generation. [quotetheorem:9968] The reverse implication depends strongly on the ring. Over a Noetherian ring it holds, but over a general ring the relations of a finitely generated module may require infinitely many generators. ## Permanence Under Module Operations A finiteness condition becomes useful only if it survives the constructions used every day. The first tests are quotients, finite sums, images, and extensions. These are the module-theoretic operations appearing in exact sequences. ### Quotients and Images Quotients are the most forgiving construction. If a finite list reaches every element of $M$, then its images reach every coset in $M/N$, so the quotient operation asks for no new generators. [quotetheorem:9969] This theorem is one reason finite generation is flexible. Passing to a quotient may destroy freeness or independence, but finite spanning data descends without extra hypotheses. [example: Quotienting a Free Module] Let $e_1=(1,0)$ and $e_2=(0,1)$. The $\mathbb{Z}$-module $\mathbb{Z}^2$ is generated by $e_1$ and $e_2$ because every $(a,b)\in \mathbb{Z}^2$ satisfies \begin{align*} (a,b)=a(1,0)+b(0,1)=ae_1+be_2. \end{align*} Let \begin{align*} N=\mathbb{Z}(2,0)+\mathbb{Z}(0,3)=\{(2u,3v):u,v\in \mathbb{Z}\}. \end{align*} For every coset $(a,b)+N\in \mathbb{Z}^2/N$, the displayed expression gives \begin{align*} (a,b)+N=a(e_1+N)+b(e_2+N). \end{align*} Thus $\mathbb{Z}^2/N$ is generated by the two cosets $e_1+N$ and $e_2+N$. We now identify the quotient. Define \begin{align*} \theta:\mathbb{Z}^2/N\to \mathbb{Z}/2\mathbb{Z}\oplus \mathbb{Z}/3\mathbb{Z} \end{align*} by \begin{align*} \theta((a,b)+N)=(\overline{a}\bmod 2,\overline{b}\bmod 3). \end{align*} If $(a,b)+N=(a',b')+N$, then \begin{align*} (a-a',b-b')\in N. \end{align*} So there are $u,v\in \mathbb{Z}$ with \begin{align*} (a-a',b-b')=(2u,3v). \end{align*} Hence $a-a'=2u$ and $b-b'=3v$, so $a\equiv a'\pmod 2$ and $b\equiv b'\pmod 3$. Therefore $\theta$ is well-defined. The map is a homomorphism because \begin{align*} \theta(((a,b)+N)+((c,d)+N))=\theta((a+c,b+d)+N)=(\overline{a+c}\bmod 2,\overline{b+d}\bmod 3). \end{align*} This equals \begin{align*} (\overline{a}\bmod 2,\overline{b}\bmod 3)+(\overline{c}\bmod 2,\overline{d}\bmod 3)=\theta((a,b)+N)+\theta((c,d)+N). \end{align*} It is surjective because every pair of residue classes $(\overline{r}\bmod 2,\overline{s}\bmod 3)$ is $\theta((r,s)+N)$. It is injective because if \begin{align*} \theta((a,b)+N)=(\overline{0}\bmod 2,\overline{0}\bmod 3), \end{align*} then $a=2u$ and $b=3v$ for some $u,v\in\mathbb{Z}$, so \begin{align*} (a,b)=(2u,3v)\in N. \end{align*} Thus $(a,b)+N=N$, the zero coset. Therefore \begin{align*} \mathbb{Z}^2/N\cong \mathbb{Z}/2\mathbb{Z}\oplus \mathbb{Z}/3\mathbb{Z}. \end{align*} The quotient has torsion relations \begin{align*} 2(e_1+N)=(2,0)+N=N \end{align*} and \begin{align*} 3(e_2+N)=(0,3)+N=N, \end{align*} but it remains finitely generated by the two displayed cosets. [/example] Submodules are more delicate than quotients, but images behave like quotients because $\operatorname{im} f$ is isomorphic to $M/\ker f$. This motivates the next permanence result: any homomorphism can lose generators, but it cannot create the need for infinitely many new ones in its image. [quotetheorem:9970] This is the image version of the quotient principle: a homomorphism can identify generators or send some of them to zero, but the image is still controlled by the finitely many images of the original generators. The result is one-way; knowing that the image is finitely generated says little about the whole source unless the kernel is also controlled. That limitation is why sums, extensions, and later Noetherian hypotheses become necessary. ### Sums and Extensions Finite generation should also survive combining finitely many pieces inside a common module. The sum of submodules is generated by taking the generating lists for each summand together, so the finite nature of the data is preserved. [quotetheorem:9971] Exact sequences combine submodule and quotient information, and they are the natural bookkeeping device for modules built in stages. If $0\to A\to B\to C\to 0$ is exact, generators for $A$ handle the part already inside $B$, while lifts of generators for $C$ handle representatives of the quotient. The useful question is whether finite generation can be tested from the two visible pieces of a short exact sequence. The next result records exactly which directions hold over an arbitrary ring, and it separates the reliable extension principle from the implication that will later require a Noetherian hypothesis. [quotetheorem:9974] The missing implication is the subtle one: finite generation of $B$ need not force finite generation of $A$ over an arbitrary ring. That gap is exactly what Noetherian rings repair. ## Noetherian Rings and Submodules The most important failure mode for finite generation is hidden inside submodules. If finitely generated modules are supposed to behave like finite-dimensional spaces, then submodules should also be finitely generated. Rings for which this holds are the Noetherian rings. ### Noetherian Modules To state the condition at the right level, we first name the module version. It is a finiteness condition on every submodule, not just on the whole module, so it directly targets the failure left open by exact sequences. [definition: Noetherian Module] Let $R$ be a ring, and let $M$ be a left $R$-module. The module $M$ is Noetherian if every submodule $N \subset M$ is finitely generated. [/definition] Noetherian modules make finite generation stable under passing to subobjects. To use this condition as a property of the scalar ring, we apply it to the regular module $R$ over itself; this turns questions about submodules into questions about ideals. [definition: Left Noetherian Ring] A ring $R$ is left Noetherian if $R$ is Noetherian as a left $R$-module. [/definition] The reason for naming left Noetherian rings is that they are precisely the rings over which finite generation behaves well with submodules of finite free modules. The next theorem is the promised repair of the missing implication in short exact sequences: it explains why kernels inside finitely generated modules remain finitely generated once the ring has the Noetherian property. [quotetheorem:2814] This theorem is often the reason the word "Noetherian" appears early in commutative algebra. It allows finite generation to pass through kernels, syzygies, and exact sequences. [example: Ideals in $\mathbb{Z}$] Let $I \subset \mathbb{Z}$ be an ideal. If $I=\{0\}$, then \begin{align*} I=0\mathbb{Z}. \end{align*} If $I\ne \{0\}$, choose a nonzero $a\in I$. Since $I$ is closed under multiplication by integers, $-a\in I$, so $I$ contains a positive integer. By well-ordering, let $n$ be the least positive integer in $I$. Then $n\mathbb{Z}\subset I$ because $I$ is closed under multiplication by integers. Conversely, for any $b\in I$, divide $b$ by $n$: \begin{align*} b=qn+r,\qquad q\in\mathbb{Z},\quad 0\le r<n. \end{align*} Since $b\in I$ and $qn\in I$, subtraction gives \begin{align*} r=b-qn\in I. \end{align*} The choice of $n$ as the least positive element of $I$ forces $r=0$, so $b=qn\in n\mathbb{Z}$. Hence $I\subset n\mathbb{Z}$, and therefore \begin{align*} I=n\mathbb{Z}. \end{align*} Thus every ideal of $\mathbb{Z}$ is generated by one element, so $\mathbb{Z}$ is Noetherian as a $\mathbb{Z}$-module. Now let $M$ be a finitely generated abelian group and let $N\subset M$ be a subgroup. Viewing abelian groups as $\mathbb{Z}$-modules, $M$ is a finitely generated $\mathbb{Z}$-module and $N$ is a $\mathbb{Z}$-submodule of $M$. Since $\mathbb{Z}$ is Noetherian, *[Modules over Noetherian Rings](/theorems/2814)* applies and gives that $N$ is finitely generated as a $\mathbb{Z}$-module. In ordinary group language, every subgroup of a finitely generated abelian group is again finitely generated. [/example] ### Non-Noetherian Failure Without the Noetherian hypothesis, the statement fails even for submodules of a cyclic module. This example gives the sharp warning: finite generation of the ambient module says little about arbitrary submodules over a ring with infinitely generated ideals. [example: A Cyclic Module with a Non-Finitely Generated Submodule] Let $k$ be a field and let \begin{align*} R = k[x_1, x_2, x_3, \ldots] \end{align*} be the [polynomial ring](/page/Polynomial%20Ring) in countably many variables. As an $R$-module over itself, $R$ is cyclic because every $r\in R$ satisfies \begin{align*} r=r\cdot 1. \end{align*} Consider the ideal \begin{align*} I=(x_1,x_2,x_3,\ldots)=\left\{\sum_{\ell=1}^{n} f_\ell x_{i_\ell}: n\ge 0,\ f_\ell\in R,\ i_\ell\in\mathbb{N}\right\}. \end{align*} Since ideals of $R$ are exactly the $R$-submodules of the regular $R$-module $R$, this $I$ is a submodule of the cyclic module $R$. We show that $I$ is not finitely generated. Suppose, toward a contradiction, that $g_1,\ldots,g_m$ generate $I$ as an $R$-module. Each $g_j$ is a polynomial, so it contains only finitely many variables. Taking the union of those finite sets of variables, there is an integer $N$ such that every $g_j$ lies in the subring $k[x_1,\ldots,x_N]$. Because $g_j\in I$, its constant term is $0$: explicitly, if $g_j=\sum_{\alpha} c_{\alpha}x^\alpha$ and the constant coefficient $c_0$ were nonzero, then evaluating every variable at $0$ would give $g_j(0)=c_0\ne 0$, while every element of $(x_1,x_2,\ldots)$ evaluates to $0$. Define a $k$-algebra homomorphism \begin{align*} \varphi:R\to k[x_{N+1}] \end{align*} by \begin{align*} \varphi(x_i)=0\text{ for }1\le i\le N,\qquad \varphi(x_{N+1})=x_{N+1},\qquad \varphi(x_i)=0\text{ for }i>N+1. \end{align*} Since each $g_j\in k[x_1,\ldots,x_N]$ has constant term $0$, every monomial appearing in $g_j$ contains at least one of $x_1,\ldots,x_N$, so \begin{align*} \varphi(g_j)=0. \end{align*} But $x_{N+1}\in I$, so finite generation by $g_1,\ldots,g_m$ would give elements $h_1,\ldots,h_m\in R$ with \begin{align*} x_{N+1}=h_1g_1+\cdots+h_mg_m. \end{align*} Applying $\varphi$ gives \begin{align*} x_{N+1}=\varphi(x_{N+1})=\varphi(h_1)\varphi(g_1)+\cdots+\varphi(h_m)\varphi(g_m)=0. \end{align*} This is impossible in the polynomial ring $k[x_{N+1}]$, since $x_{N+1}$ is a nonzero indeterminate. Therefore $I$ is not finitely generated. Thus the cyclic module $R$ has a submodule $I$ that is not finitely generated, showing that finite generation need not pass to submodules over a non-Noetherian ring. [/example] This example is the main reason Noetherian hypotheses are not decoration. They mark exactly when finite generation behaves well with kernels and ideals. ## Finite Generation over Principal Ideal Domains Over special rings, finite generation gives far more information than a finite spanning list. Principal ideal domains are the first major case where finitely generated modules can be classified up to isomorphism. ### Principal Ideals and Free Submodules The ring condition matters because relations among generators can be simplified using divisibility. The definition below isolates the rings where every ideal has one generator, which is the algebraic input behind Smith normal form and the classification of finitely generated abelian groups. [definition: Principal Ideal Domain] A [principal ideal domain](/page/Principal%20Ideal%20Domain) is a commutative [integral domain](/page/Integral%20Domain) $R$ in which every ideal $I \trianglelefteq R$ has the form \begin{align*} I = (a) \end{align*} for some $a \in R$. [/definition] A submodule of a finite free module can be hard to describe because its generators may interact through complicated relations. Over a principal ideal domain, divisibility gives enough control to simplify such submodules, and that control is the structural input behind the classification of quotients of finite free modules. A finitely generated module over a principal ideal domain is a quotient of $R^n$ by relations among finitely many generators. The classification theorem below records the final outcome of simplifying those relations: the module splits into free directions and torsion directions rather than an arbitrary finite presentation. ### Classification Once submodules of finite free modules are understood, the natural next question is what all finitely generated modules look like up to isomorphism. The classification theorem answers this by separating free directions from torsion directions. [quotetheorem:3233] The theorem says that a finitely generated module over a principal ideal domain splits into a free part and a torsion part. Over $\mathbb{Z}$, this becomes the classification of finitely generated abelian groups. [example: A Finitely Generated Abelian Group] Let $e_1=(1,0)$, $e_2=(0,1)$, and \begin{align*} N=\mathbb{Z}(2,0)+\mathbb{Z}(0,6)=\{(2a,6b):a,b\in\mathbb{Z}\}. \end{align*} Consider \begin{align*} M=\mathbb{Z}^2/N. \end{align*} Every element of $M$ is a coset $(u,v)+N$, and in $\mathbb{Z}^2$ we have \begin{align*} (u,v)=u(1,0)+v(0,1)=ue_1+ve_2. \end{align*} Passing to cosets gives \begin{align*} (u,v)+N=u(e_1+N)+v(e_2+N), \end{align*} so $M$ is generated as a $\mathbb{Z}$-module by $e_1+N$ and $e_2+N$. The subgroup $N$ imposes exactly the relations \begin{align*} 2(e_1+N)=(2,0)+N=N \end{align*} and \begin{align*} 6(e_2+N)=(0,6)+N=N. \end{align*} Define \begin{align*} \theta:M\to \mathbb{Z}/2\mathbb{Z}\oplus \mathbb{Z}/6\mathbb{Z} \end{align*} by \begin{align*} \theta((u,v)+N)=(\overline{u}\bmod 2,\overline{v}\bmod 6). \end{align*} If $(u,v)+N=(u',v')+N$, then $(u-u',v-v')\in N$, so there are $a,b\in\mathbb{Z}$ with \begin{align*} (u-u',v-v')=(2a,6b). \end{align*} Thus $u-u'=2a$ and $v-v'=6b$, which means $\overline{u}=\overline{u'}$ modulo $2$ and $\overline{v}=\overline{v'}$ modulo $6$. Hence $\theta$ is well-defined. For cosets $(u,v)+N$ and $(r,s)+N$, \begin{align*} \theta(((u,v)+N)+((r,s)+N))=\theta((u+r,v+s)+N). \end{align*} The right-hand side is \begin{align*} (\overline{u+r}\bmod 2,\overline{v+s}\bmod 6). \end{align*} This equals \begin{align*} (\overline{u}\bmod 2,\overline{v}\bmod 6)+(\overline{r}\bmod 2,\overline{s}\bmod 6), \end{align*} so $\theta$ is a homomorphism. It is surjective because every pair $(\overline{a}\bmod 2,\overline{b}\bmod 6)$ is the image of $(a,b)+N$. Its kernel consists of those cosets $(u,v)+N$ with $u\equiv 0\pmod 2$ and $v\equiv 0\pmod 6$, so $u=2a$ and $v=6b$ for some $a,b\in\mathbb{Z}$. Then \begin{align*} (u,v)=(2a,6b)\in N, \end{align*} so $(u,v)+N=N$. Therefore $\ker(\theta)=0$, and $\theta$ is an isomorphism: \begin{align*} M\cong \mathbb{Z}/2\mathbb{Z}\oplus \mathbb{Z}/6\mathbb{Z}. \end{align*} Since $2\mid 6$, this decomposition is already in invariant-factor form, and the example shows a finitely generated $\mathbb{Z}$-module with two visible torsion relations. [/example] Even over a principal ideal domain, finite generation is not the same as freeness. The torsion summands record the obstruction. ## Nakayama and Local Control In commutative algebra, finite generation becomes especially powerful after localising at a prime or working over a local ring. The guiding principle is that a finitely generated module over a local ring is controlled by its reduction modulo the maximal ideal. To make local control meaningful, we first need the setting in which a ring has a single distinguished maximal ideal. This packages the idea that all noninvertible behavior is concentrated at one algebraic point. [definition: Local Ring] A commutative ring $R$ is local if it has a unique maximal ideal $\mathfrak{m}$. [/definition] The next object is needed because local finite-generation questions are usually not answered inside $R$ itself, but after collapsing the maximal ideal to zero. That quotient is a field, so it turns a local module question into a linear algebra question. [definition: Residue Field] Let $(R, \mathfrak{m})$ be a commutative local ring. The residue field of $R$ is the quotient field \begin{align*} R/\mathfrak{m}. \end{align*} [/definition] The passage from $M$ to $M/\mathfrak{m}M$ forgets higher-order information. The theorem needed here says that finite generation prevents a nonzero module from disappearing into its maximal-ideal multiple, which is what allows residue-field calculations to control the original module. [quotetheorem:2168] [Nakayama's lemma](/theorems/2935) is a finiteness theorem. Without finite generation, the conclusion can fail: a module may equal its own maximal-ideal multiple without being zero. [example: Why Finite Generation Matters in Nakayama] Here $\mathbb{Z}_{(p)}$ denotes the localisation of $\mathbb{Z}$ at the prime ideal $(p)$; equivalently, it is the subring of $\mathbb{Q}$ consisting of rational numbers whose denominator is not divisible by $p$. Let $R=\mathbb{Z}_{(p)}$ and let $M=\mathbb{Q}$, with $R$ acting on $\mathbb{Q}$ by ordinary multiplication. Multiplication by $p$ is surjective on $M$: if $y\in \mathbb{Q}$, then $y=p(y/p)$ and $y/p\in \mathbb{Q}$. Hence every element of $M$ lies in $pM$, while $pM\subset M$ by definition of scalar multiplication, so \begin{align*} M=pM. \end{align*} We show that $M$ is not finitely generated over $R$. Suppose $q_1,\ldots,q_k\in \mathbb{Q}$ generated $M$ as an $R$-module. For each nonzero $q_j$, write $q_j=a_j/b_j$ with $a_j,b_j\in\mathbb{Z}$ and $b_j\ne 0$. Factor the power of $p$ out of $b_j$, so $b_j=p^{n_j}c_j$ with $n_j\ge 0$ and $p\nmid c_j$. Then \begin{align*} q_j=\frac{a_j}{p^{n_j}c_j}=\left(\frac{a_j}{c_j}\right)p^{-n_j}. \end{align*} Since $p\nmid c_j$, the fraction $a_j/c_j$ lies in $\mathbb{Z}_{(p)}$. For any zero generator $q_j=0$, set $n_j=0$ and $u_j=0$. Thus each generator has the form \begin{align*} q_j=u_jp^{-n_j} \end{align*} with $u_j\in \mathbb{Z}_{(p)}$ and $n_j\ge 0$. Choose $N$ with $N\ge n_j$ for every $j$. Then \begin{align*} q_j=u_jp^{-n_j}=u_jp^{N-n_j}p^{-N}. \end{align*} Because $N-n_j\ge 0$, the coefficient $u_jp^{N-n_j}$ lies in $\mathbb{Z}_{(p)}$, so $q_j\in p^{-N}\mathbb{Z}_{(p)}$. Therefore every $R$-linear combination of the generators has the form \begin{align*} r_1q_1+\cdots+r_kq_k=\left(r_1u_1p^{N-n_1}+\cdots+r_ku_kp^{N-n_k}\right)p^{-N} \end{align*} with the parenthesized coefficient in $\mathbb{Z}_{(p)}$. Hence the $R$-span of $q_1,\ldots,q_k$ is contained in $p^{-N}\mathbb{Z}_{(p)}$. But $p^{-(N+1)}\notin p^{-N}\mathbb{Z}_{(p)}$. Indeed, if $p^{-(N+1)}=p^{-N}r$ for some $r\in\mathbb{Z}_{(p)}$, then multiplying by $p^N$ in $\mathbb{Q}$ gives $1/p=r$. This is impossible because elements of $\mathbb{Z}_{(p)}$ are fractions with denominator not divisible by $p$, while $1/p$ has denominator divisible by $p$. Thus $p^{-(N+1)}$ is not in the span of the alleged finite generating set, contradicting that the $q_j$ generate $\mathbb{Q}$. So this module satisfies $M=pM$ but is not finitely generated. The example shows exactly why the finite generation hypothesis in Nakayama's lemma is essential. [/example] The practical version of Nakayama turns generator counting into a vector-space problem over the residue field. The possible obstruction is that elements which span only after reducing modulo $\mathfrak{m}$ might miss elements of $M$ before reduction. Finite generation is exactly the hypothesis that prevents this obstruction: if the residual classes span $M/\mathfrak{m}M$, their chosen lifts already generate $M$. [quotetheorem:9976] This theorem is one reason finite generation is central in algebraic geometry and commutative algebra. Local questions about modules can often be reduced to linear algebra over residue fields. ## Tensor Products, Localisation, and Base Change Finitely generated modules are stable under changing scalars in the usual ways. This matters because algebra rarely studies a module over only one ring: one localises, extends scalars, reduces modulo ideals, and compares fibers. ### Extension of Scalars The most basic scalar change is extension of scalars along a ring homomorphism. If $M$ is generated by finitely many elements over $R$, then after scalars are extended to $S$, those same elements should generate through tensors with $1 \in S$. [quotetheorem:9978] The generators after base change are the tensors $1 \otimes m_i$. Relations may change, and new relations may appear, but finite spanning survives. Reduction modulo an ideal is the most concrete version of this construction: the new scalars are $R/I$, and tensoring simply records what remains after every element of $I$ acts as zero. [example: Reducing a Module Modulo an Ideal] Let $R$ be a commutative ring, let $I \trianglelefteq R$ be an ideal, and let $M$ be generated by $m_1,\ldots,m_k$ as an $R$-module. Define \begin{align*} \Phi:(R/I)\otimes_R M\to M/IM \end{align*} by \begin{align*} \Phi((r+I)\otimes m)=rm+IM. \end{align*} This is well-defined because if $r-r'\in I$, then \begin{align*} rm-r'm=(r-r')m\in IM, \end{align*} and the formula is $R$-balanced since \begin{align*} \Phi(((r+I)a)\otimes m)=ram+IM=\Phi((r+I)\otimes am). \end{align*} Define \begin{align*} \Psi:M/IM\to (R/I)\otimes_R M \end{align*} by \begin{align*} \Psi(m+IM)=(1+I)\otimes m. \end{align*} If $m-m'\in IM$, write $m-m'=a_1n_1+\cdots+a_\ell n_\ell$ with $a_t\in I$ and $n_t\in M$. Then \begin{align*} (1+I)\otimes(m-m')=(a_1+I)\otimes n_1+\cdots+(a_\ell+I)\otimes n_\ell=0, \end{align*} so $\Psi$ is well-defined. Now \begin{align*} \Phi(\Psi(m+IM))=\Phi((1+I)\otimes m)=m+IM \end{align*} and \begin{align*} \Psi(\Phi((r+I)\otimes m))=\Psi(rm+IM)=(1+I)\otimes rm=(r+I)\otimes m. \end{align*} Thus \begin{align*} M/IM\cong (R/I)\otimes_R M. \end{align*} The quotient is generated over $R/I$ by the residue classes $m_1+IM,\ldots,m_k+IM$: if $m=r_1m_1+\cdots+r_km_k$, then \begin{align*} m+IM=(r_1+I)(m_1+IM)+\cdots+(r_k+I)(m_k+IM). \end{align*} Equivalently, $(R/I)\otimes_R M$ is generated by $(1+I)\otimes m_1,\ldots,(1+I)\otimes m_k$. Reduction modulo $I$ therefore keeps the same finite generating list, but changes the scalars from $R$ to $R/I$. [/example] ### Localisation Localisation is the scalar change that focuses attention near a prime ideal or multiplicative set. To state the finite-generation result for localisation, we first need the module being produced by allowing denominators from a chosen multiplicative set. [definition: Localisation of a Module] Let $R$ be a commutative ring, let $S \subset R$ be a multiplicative subset, and let $M$ be an $R$-module. The localisation of $M$ at $S$ is the $S^{-1}R$-module $S^{-1}M$ whose elements are fractions \begin{align*} \frac{m}{s}, \qquad m \in M,\ s \in S, \end{align*} where $\frac{m}{s} = \frac{m'}{s'}$ if there exists $t \in S$ such that \begin{align*} t(s'm - sm') = 0 \end{align*} in $M$. [/definition] If $M$ has finitely many generators before localisation, the same generators, now viewed as fractions with denominator $1$, are the only candidates needed after localisation. The theorem records that no hidden infinite generating set appears when denominators are added. [quotetheorem:9980] Preservation under localisation is only one direction. In geometry and commutative algebra, the harder question is whether finite generation checked on a finite family of local pieces can be assembled into finite generation before localisation. The next theorem gives the standard finite-cover form of that local-to-global principle. [quotetheorem:9981] This theorem captures the geometric intuition that finite generation is a quasi-compact condition. The finite cover hypothesis is essential: it says there are only finitely many local tests to reconcile, so local finite-type behavior can be promoted to a global finite generating set. On affine schemes, this is the algebraic reason that checking finite type behavior on finitely many basic open sets is enough once the local data comes from the same global module. ## Beyond and Connected Topics Finitely generated modules are the entry point to the broader finiteness hierarchy in algebra. The next step is finite presentation, where relations are required to be finitely generated as well as generators. This distinction becomes important over non-Noetherian rings and in homological algebra, where kernels of maps between finite free modules are the first syzygies. Noetherian methods turn finite generation into a stable theory. In [Cambridge III Commutative Algebra](/page/Cambridge%20III%20Commutative%20Algebra), finitely generated [modules over Noetherian rings](/theorems/2903) appear alongside localisation, primary decomposition, Nakayama's lemma, and dimension theory. The guiding theme is that finite module data behaves like finite algebraic data on affine schemes. The abelian group case belongs to the algebra learned in [Cambridge IB Groups, Rings and Modules](/page/Cambridge%20IB%20Groups%2C%20Rings%20and%20Modules). There the structure theorem over $\mathbb{Z}$ is the prototype: finitely generated modules over a principal ideal domain decompose into free and torsion parts. Homological algebra studies finite generation through resolutions. In [Homological Algebra I: Complexes and Resolutions](/page/Homological%20Algebra%20I%3A%20Complexes%20and%20Resolutions), a finite free cover $R^n \to M$ becomes the first map in a projective or free resolution, and the relations become the first syzygy module. Finite generation alone controls the start of the resolution, while stronger finiteness conditions control later terms. Finite-dimensional vector spaces are the cleanest special case. In [Cambridge IB Linear Algebra](/page/Cambridge%20IB%20Linear%20Algebra), finite spanning sets can often be reduced to bases. Module theory keeps the spanning idea but abandons the guarantee of bases, which is why generators and relations replace coordinates and dimension. ## References Androma, [Cambridge IB Groups, Rings and Modules](/page/Cambridge%20IB%20Groups%2C%20Rings%20and%20Modules). Androma, [Cambridge III Commutative Algebra](/page/Cambridge%20III%20Commutative%20Algebra). Androma, [Homological Algebra I: Complexes and Resolutions](/page/Homological%20Algebra%20I%3A%20Complexes%20and%20Resolutions). Androma, [Cambridge IB Linear Algebra](/page/Cambridge%20IB%20Linear%20Algebra). Atiyah and Macdonald, *Introduction to Commutative Algebra* (1969). Eisenbud, *Commutative Algebra with a View Toward Algebraic Geometry* (1995). Lang, *Algebra* (2002).