Skip to main content

Section 14.1 Finitely Presented Modules

Previously on Modern Algebra....

“The success of your presentation will be judged not by the knowledge you send but by what the listener receives.”
―Lily Walters
When working with groups, we had the notion of the presentation of a group. We proved that every group was the quotient of a free group, where the presentation of the group was obtained by taking quotients corresponding to the relations of the presentation.
In the realm of modules, we have shown that Every \(R\)-module is the Quotient of a Free Module. We also know that every free module of rank \(n\) is isomorphic to \(R^n\text{.}\) Thus, we can obtain a presentation for a module by taking quotients of \(R^n\) corresponding to our relations, which are given in the form of a matrix (or equivalently by a homomorphism between a pair of free modules.)
Table 14.1. Notation
\(\Mat_{m,n}(R)\) The collection of all \(m\times n\) matrices with coefficients in \(R\)
\(T_A\)
\(\ker\text{,}\) \(\coker\text{,}\) \(\im\) The Kernel, Cokernel, and image of an \(R\)-module homomorphism.
\(M\oplus N\text{,}\) \(R^n\)
\(Ra\)

Subsection Module Presentations

Definition 14.2. \(R\)-Module Presentation.

Let \(R\) be a non-zero commutative ring, let \(A \in \Mat_{m,n}(R)\text{,}\) and let \(T_A: R^n \to R^m\) be the \(R\)-module homomorphism represented by \(A\) with respect to the standard bases; that is, define \(T_A(v)=A \cdot v\text{.}\) The \(R\)-module presented by \(A\) is the \(R\)-module \(\coker(T_A) = R^m/\im(T_A)\text{.}\)
Equivalently, the module presented by \(A\) is
\begin{equation*} \frac{R^m}{R v_1 + \cdots + R v_n} \end{equation*}
where \(v_1, \dots, v_n\) are the columns of \(A\text{.}\)

Example 14.3.

The \(\mathbb{Z}\)-module \(M=\mathbb{Z} / 6\) is presented by
\begin{equation*} \mathbb{Z} \stackrel{6}{\rightarrow} \mathbb{Z} \end{equation*}
Solution.
Notice \(M \cong \mathbb{Z} / \operatorname{im}\left(t_{6}\right)=\mathbb{Z} /(6)\text{.}\) Notice here we abused notation and wrote \(6\) instead of the \(1 \times 1\) matrix \([6]\text{.}\)
Conversely, we might be given a matrix and ask about what module it represents; one thing to keep in mind is that some presentations might be inefficient, either by having more generators or more relations than necessary. We want to answer to key questions: given a presentation for a module, how to find a more efficient presentation; and how to decide if two different presentations actually give us isomorphic modules. Keeping these goals in mind, let’s try a more elaborate example.

Example 14.4. \(\Z\)-Module Presentation.

Consider the matrix
\begin{equation*} A=\left[\begin{array}{ccc} 2 & 1 & 0 \\ 3 & 9 & 5 \\ 1 & -2 & 7 \\ 0 & 1 & 2 \end{array}\right] \end{equation*}
What \(\mathbb{Z}\)-module \(M\) is presented by \(A\text{?}\)
Solution.
Formally, \(M\) is the quotient module \(M=\mathbb{Z}^{4} / \operatorname{im}\left(t_{A}\right)\text{,}\) where \(t_{A}: \mathbb{Z}^{3} \rightarrow \mathbb{Z}^{4}\) is defined by \(t_{A}(v)=A v\text{.}\) Since \(\mathbb{Z}^{4}\) is generated by its standard basis elements \(\left\{e_{1}, e_{2}, e_{3}, e_{4}\right\}\text{,}\) we deduce as in Lemma 13.10 that \(M=\mathbb{Z}^{4} / \operatorname{im}\left(t_{A}\right)\) is generated by the cosets of the \(e_{i}\text{.}\) To keep the notation short, we set \(m_{i}=e_{i}+\operatorname{im}\left(t_{A}\right)\text{.}\)
Let \(N=\operatorname{im}\left(t_{A}\right)\) and note that \(N\) is the submodule of \(\mathbb{Z}^{4}\) generated by the columns of \(A\text{:}\)
\begin{equation*} N=R\left\{\left[\begin{array}{l} 2 \\ 3 \\ 1 \\ 0 \end{array}\right],\left[\begin{array}{c} 1 \\ 9 \\ -2 \\ 1 \end{array}\right],\left[\begin{array}{l} 0 \\ 5 \\ 7 \\ 2 \end{array}\right]\right\}=R\left\{2 e_{1}+3 e_{2}+e_{3}, e_{1}+9 e_{2}-2 e_{3}+e_{4}, 5 e_{2}+7 e_{3}+2 e_{4}\right\} \end{equation*}
Since \(N\) maps to 0 under the quotient map \(q: \mathbb{Z}^{4} \rightarrow M=\mathbb{Z}^{4} / N\text{,}\) the relations of \(M\) can be written as
\begin{equation*} \begin{cases}2 m_{1}+3 m_{2}+m_{3} & =0 \\ m_{1}+9 m_{2}-2 m_{3}+m_{4} & =0 \\ 5 m_{2}+7 m_{3}+2 m_{4} & =0\end{cases} \end{equation*}
We can now see that this is a rather inefficient presentation, since we can clearly use the first equation to solve for for \(m_{3}=-2 m_{1}-3 m_{2}\text{.}\) This implies that \(M\) can be generated using only \(m_{1}, m_{2}\) and \(m_{4}\text{,}\) that is
\begin{equation*} M=R\left\{m_{1}, m_{2}, m_{3}, m_{4}\right\}=R\left\{m_{1}, m_{2}, m_{4}\right\}. \end{equation*}
This eliminates the first equation and the latter two become
\begin{equation*} \begin{cases}5 m_{1}+15 m_{2}+m_{4} & =0 \\ -14 m_{2}-16 m_{2}+2 m_{4} & =0\end{cases} \end{equation*}
Now we can also eliminate \(m_{4}\text{,}\) i.e leaving just two generators \(m_{1}, m_{2}\) that satisfy
\begin{equation*} -24 m_{1}-46 m_{2}=0. \end{equation*}
Another way to do this is to look at the matrix \(A\) and use elementary row operations to "make zeros" on the 1st and 2nd columns, as follows:
\begin{equation*} A=\left[\begin{array}{ccc} 2 & 1 & 0 \\ 3 & 9 & 5 \\ 1 & -2 & 7 \\ 0 & 1 & 2 \end{array}\right] \rightarrow\left[\begin{array}{ccc} 0 & 5 & -14 \\ 0 & 15 & -16 \\ 1 & -2 & 7 \\ 0 & 1 & 2 \end{array}\right] \rightarrow\left[\begin{array}{ccc} 0 & 0 & -24 \\ 0 & 0 & -46 \\ 1 & 0 & 13 \\ 0 & 1 & 0 \end{array}\right] \end{equation*}
Eliminating the generators \(m_{3}\) and \(m_{4}\) amounts to dropping the first two columns (which are the 3rd and 4th standard basis vectors) as well as the last two rows. As we will prove soon, this shows that the \(\mathbb{Z}\)-module presented by \(A\) is isomorphic to the \(\mathbb{Z}\)-module presented by
\begin{equation*} B=\left[\begin{array}{l} -24 \\ -46 \end{array}\right] \end{equation*}
We can go further. Set \(m_{1}^{\prime}:=m_{1}+2 m_{2}\text{.}\) Then \(m_{1}^{\prime}\) and \(m_{2}\) also form a generating set of \(M\text{.}\) The relation on \(m_{1}, m_{2}\) translates to
\begin{equation*} -24 m_{1}^{\prime}+2 m_{2}=0 \end{equation*}
given by the matrix
\begin{equation*} C=E_{1,2}(-2) B=\left[\begin{array}{c} -24 \\ 2 \end{array}\right]. \end{equation*}
Note that we have done a row operation (subtract twice row 1 from row 2 ) to get from \(B\) to \(C\text{.}\) Continuing in this fashion by subtracting 12 row 2 from row 1 we also form
\begin{equation*} D=E_{1,2}(12) C=\left[\begin{array}{l} 0 \\ 2 \end{array}\right], \end{equation*}
The last matrix \(D\) presents the module \(M^{\prime}=\mathbb{Z}^{2} / \operatorname{im}\left(t_{D}\right)\) with generators \(a, b\text{,}\) where
\begin{equation*} a=e_{1}+\operatorname{im}\left(t_{D}\right), \quad b=e_{2}+\operatorname{im}\left(t_{D}\right) \end{equation*}
and relation \(2 a=0\text{.}\) This module \(M^{\prime}\) is isomorphic to our original module \(M\text{.}\) As we will see, this proves \(M \cong \mathbb{Z} \oplus \mathbb{Z} / 2\text{.}\) An explicit isomorphism between \(M^{\prime}\) and \(\mathbb{Z} \oplus \mathbb{Z} / 2\) is given by sending \(\mathbb{Z}^{2} \rightarrow \mathbb{Z} \oplus \mathbb{Z} / 2\) by the unique \(\mathbb{Z}\)-module homomorphism defined by
\begin{equation*} e_{1} \mapsto(1,0) \text { and } e_{2} \mapsto\left(0,[1]_{2}\right) \end{equation*}
Now notice that the kernel of this homomorphism is the submodule \(\left(2 e_{2}\right) \mathbb{Z}=\operatorname{im}\left(t_{D}\right)\text{.}\) Then the first isomorphism theorem gives \(M^{\prime}=\mathbb{Z}^{2} / \operatorname{im}\left(t_{D}\right) \cong \mathbb{Z} \oplus \mathbb{Z} / 2\text{.}\)

Proof.

Note: This proof was not covered in class. Assume \(B\) is obtained from \(A\) by a single one of the steps listed above. We need to prove that there is an isomorphism \(R^m/\im(T_A) \cong R^{m'}/\im(T_B)\) of \(R\)-modules.
  1. In this case, \(B = E A\) for some elementary matrix \(E\text{.}\) More generally, let \(E\) be any invertible matrix such that \(B = EA\text{.}\) Then \(T_E: R^m \to R^m\) is an isomorphism and it maps \(\im(T_A)\) bijectively onto \(\im(T_B)\text{.}\) It follows that the kernel of the composition \(R^m \xra{T_E} R^m \onto R^m/\im(T_B)\) is \(\im(T_A)\) and hence by the first isomorphism theorem it induces an isomorphism
    \begin{equation*} R^m/\im(T_A) \xra{\cong} R^m/\im(T_B). \end{equation*}
  2. In this case, \(B=AE\) for some elementary matrix \(E\text{.}\) More generally, assume \(E\) is any invertible matrix such that \(B = AE\text{.}\) Since \(T_E\) is an isomorphism, we have
    \begin{equation*} \im(T_B) = \im(T_{AE})=\im(T_A\circ T_E)=\im(T_A) \end{equation*}
    and so \(R^m/\im(T_A) = R^{m}/\im(T_B)\text{.}\) (For this one we get equality, not merely an isomorphism.)
  3. For notational simplicity, let us assume \(i = j =1\text{;}\) that is, the first column of \(A\) is \(e_1\) and \(B\) is obtained by deleting the first row and column of \(A\text{,}\) giving a \((m-1) \times (n-1)\) matrix. So
    \begin{equation*} A = \begin{bmatrix} 1 & r \\ 0 & B \\\end{bmatrix} \end{equation*}
    where \(r\) denotes some row vector and \(0\) denotes a column of all \(0\)’s. Let \(p: R^{m} \onto R^{m-1}\) and \(q: R^{n} \onto R^{n-1}\) be projection onto the last \(m-1\) and \(n-1\) components, respectively. Because of the nature of \(A\) and \(B\text{,}\) the diagram (page 40 in notes) commutes. Moreover, the kernel of \(q\) is \(Re_1\) and the kernel of \(p\) is \(Re_1\text{,}\) and since the first column of \(A\) is \(e_1\text{,}\) \(T_A\) maps the kernel of \(q\) bijectively onto the kernel of \(p\text{.}\) A “diagram chase’’ shows that \(\coker(T_A) \cong \coker(T_B)\text{.}\) In detail: Since the diagram commutes, \(p(\im(T_A)) \subseteq \im(T_B)\) and hence \(p\) induces an \(R\)-module homomorphism
    \begin{equation*} \ov{p}: R^m/\im(T_A) \to R^{m-1}/\im(T_B). \end{equation*}
    (by the \(0\)-th isomorphism theorem). Since \(p\) is onto, so is \(\ov{p}\text{.}\) Suppose \(v + \im(T_A) \in \ker(\ov{p})\text{.}\) So, \(p(v) \in \im(T_B)\text{.}\) Say \(p(v) = B w\text{.}\) Since \(q\) is onto, \(w = q(u)\) for some \(u\text{.}\) Then
    \begin{equation*} p(v - Au) = p(v) - pA(u) = p(v) - Bq(u) = p(v) - Bw = p(v) - p(v) = 0, \end{equation*}
    and thus \(v - Au \in \ker(p)\text{.}\) As noted above, \(T_A\) maps \(\ker(q)\) onto \(\ker(p)\) and hence
    \begin{equation*} v - Au = Ay \end{equation*}
    for some vector \(y\text{.}\) This proves \(v = A(u+y) \in \im(T_A)\) and hence that \(v +\im(T_A) = 0\) in \(R^m/\im(T_A)\text{.}\) This proves \(\ov{p}\) is one-to-one.
  4. The columns of \(B\) generate the same submodule of \(R^m\) as do the columns of \(A\text{,}\) and thus \(\im(T_A) = \im(T_B)\) and \(R^m/\im(T_A) =R^m/\im(T_B)\text{.}\)
  5. Since the isomorphism relation is reflexive, the statements of parts 3. & 5. show that parts 4.& 6. are true as well.

Remark 14.6.

The converse is true for some rings \(R\text{,}\) including Euclidean domains.

Remark 14.7.

In fact, if \(A\) and \(B\) are equivalent matrices, then \(\coker(T_A) \cong \coker(T_B)\text{,}\) as I shall prove below. This implies both (1) and (2) from the Theorem.

Proof.

Assume \(n \leq m\) and define \(g: R^m \to R/R \cdot d_1 \oplus R/R \cdot d_2 \oplus \cdots \oplus R/ R \cdot d_n \oplus R^{m-n}\) to be the map sending \((r_1, \dots, r_m)^\tau\) to \((\ov{r}_1, \dots, \ov{r}_n, r_{n+1}, \dots, r_m)\) where \(\ov{r}_i = r_i + R \cdot d_i\) for \(1 \leq i \leq n\text{.}\) (I.e., \(g\) is the unique \(R\)-map sending the \(i\)-th standard basis vector \(e_i\) to \((0, \dots, 0, \ov{1}, 0, \dots, 0)\) with \(\ov{1}\) in the \(i\)-th position, for \(1 \leq i \leq n\text{,}\) and to \(e_i\) itself for \(i > n\text{.}\)) Then \(g\) is clearly onto and the kernel \(\ker(g)\) of \(g\) is the set of those tuples \((r_1, \dots, r_m)^\tau\) such that \(r_i = x_i d_i\) for some \(x_i\) for all \(1 \leq i \leq n\) and \(r_j = 0\) for \(j > n\text{.}\) Given such a tuple,
\begin{equation*} (r_1, \dots, r_m)^\tau = x_1 (d_1, 0, \dots, 0)^\tau + x_2 (0,d_2, 0, \dots, 0)^\tau + \cdots + x_n (0, \dots, 0, d_n, 0, \dots, 0)^\tau. \end{equation*}
This proves \(\ker(g)\) is contained
\begin{equation*} R \cdots (d_1, 0, \dots, 0)^\tau + R \cdot (0,d_2, 0, \dots, 0)^\tau + \cdots + R \cdot (0, \dots, 0, d_n, 0 \dots, 0)^\tau = \im(T_A). \end{equation*}
Arguing backwards we see that the opposite containment also holds, so that in fact \(\ker(g) = \im(T_A)\text{.}\)
By the [cross-reference to target(s) "thm-fit-module" missing or not unique],
\begin{equation*} \coker(T_A) = R/\im(T_A) = R/\ker(g) \cong R/R \cdot d_1 \oplus \cdots \oplus R/R \cdot d_n \oplus R^{m-n}. \end{equation*}
If \(n \leq m\) then, by deleting columns of all \(0\)’s, we may reduce to the case when \(n = m\text{,}\) which is included in the first case.

Subsection When Modules are Finitely Presented: Noetherian Rings

“A chain is no stronger than its weakest link.”
―William James

Remark 14.9.

We now address the question of which modules have finite presentations. Any such module must be finitely generated (since the cosets of \(e_1, \dots, e_m\) generate \(R^m/\im(T_A)\) for any \(m \times n\) matrix \(A\)). If \(M\) is finitely generated, say by \(m\) elements, then we can find a surjective \(R\)-module homomorphism
\begin{equation*} \pi: R^m \onto M. \end{equation*}
Provided the kernel of \(\pi\) is also finitely generated, say by \(n\) elements, then we may find a surjection
\begin{equation*} g: R^n \onto \ker(\pi). \end{equation*}
The composition \(R^n \xra{g} \ker(\pi) \subseteq R^m\) is a map between free \(R\)-modules and is thus equal to \(T_A\) for some \(n \times m\) matrix \(A\text{.}\) Then \(\im(T_A) = \ker(\pi)\) and hence by the [cross-reference to target(s) "thm-fit-module" missing or not unique]
\begin{equation*} R^m/\im(A) = R^m/\ker(\pi) \cong M, \end{equation*}
so that \(M\) is finitely presented.
So the real question is: For a given ring \(R\text{,}\) is it the case that for all \(m\text{,}\) every submodule of \(R^m\) is finitely generated? The answer is "no" in general, but it does hold for many rings of interest:

Proof.

Assume every ideal is finitely generated and that such a chain is given. Let \(J = \cup_n I_n\text{.}\) For \(a \in J, r \in R\) we have \(a \in I_n\) for some \(n\) and hence \(ra \in I_n \subseteq J\text{.}\)
If \(a,b \in J\text{,}\) then \(a \in I_n\) and \(b \in I_m\) for some \(n,m\) and hence there is a \(N\) such that \(a,b \in I_N\text{.}\) It follows that \(a \pm b \in I_N \subseteq J\text{.}\) Finally \(0 \in J\text{.}\) Thus \(J\) is an ideal. Thus by assumption \(J\) is finitely generated, say \(J = Rx_1 + \cdots + Rx_m\) for some \(x_1, \dots, x_m \in J\text{.}\)
Each \(x_i\) belongs to one of the \(I_n\)’s and hence, since there are only a finite number of such elements and ideas are nested, there is an \(N\) such that \(x_1, \dots, x_m \in I_N\text{.}\) It follows that \(I_N = J\) and hence \(I_N = I_{N+1} = I_{N+2} = \cdots\text{.}\)
Assume \(R\) has the acc for ideals and let \(J\) be any ideal. Pick any element \(a_1\in I\) and set \(I_1 = Ra_1\text{.}\) If \(J = I_1\) we are done. If not, pick \(a_2 \in I \setminus I_1\) and set \(I_2 = Ra_1 + Ra_2\text{.}\) If \(J = I_2\) we are done and if not pick \(a_3 \in I\setminus I_2\) and let \(I_3 = Ra_1 + Ra_2 + Ra_3\text{.}\) In this way we form a strictly ascending chain \(I_1 \subset I_2 \subset I_3 \subset\text{,}\) and this process cannot be continued forever since \(R\) has the acc. When it terminates, we have \(J = Ra_1 + Ra_2 + \cdots + Ra_N\) for some \(a_1, \dots, a_N\) and thus \(J\) is finitely generated.

Proof.

I will just prove the following special case (since it is all we need): For each \(n \geq 1\text{,}\) every submodule of \(R^n\) is finitely generated. The base case \(n=1\) holds by definition (and Lemma ), since a submodule of \(R^1\) is the same thing as an ideal.
Assume \(n > 1\) and the result holds for \(R^{n-1}\text{.}\) Let \(M\) be any submodule of \(R^n\text{.}\) Define
\begin{equation*} \pi: R^n \onto R^1 \end{equation*}
to be the projection onto the last component of \(R^n\text{.}\) The kernel of \(\pi_n\) may be identified with \(R^{n-1}\) and so \(N := \ker(\pi) \cap M\) is a submodule of \(R^{n-1}\text{,}\) and it is therefore finitely generated by assumption. The image \(\pi(M)\) of \(M\) under \(\pi\) is a submodule of \(R^1\text{,}\) that is, an ideal of \(R\text{,}\) and so it too is finitely generated by assumption (and Lemma ).
Furthermore, by the first isomorphism theorem \(M/\ker(\pi)\cong \pi(M)\) is also finitely generated. By a homework problem, we deduce that \(M\) is a finitely generated module.
% I’ll just sketch the general case (which I don’t think we’ll actually need): let \(T\) be any finitely generated \(R\)-module and \(N \subseteq T\) any submodule. % Since \(T\) is finitely generated, there exists a surjective \(R\)-module homomorphism \(q: R^n \onto T\) for some \(n\text{.}\) Then \(q^{-1}(N)\) is a submodule of \(R^n\) and % hence it is finitely generated by the case we already proved. Moreover, \(q\) induces a surjective \(R\)-module homomorphism \(q: q^{-1}(N) \onto N\text{,}\) % and hence \(N\) is isomorphic to a quotient of a finitely generated \(R\)-module and thus it is also finitely generated.
The converse is also true.

Remark 14.12.

If \(R\) is not noetherian, there there exists an ideal \(I\) that is not finitely generated (by Lemma 14.10). This gives an example of a non-finitely-generated submodule, namely \(I\text{,}\) of a finitely generated module, namely \(R\text{.}\)

Proof.

We basically already proved this, but let me recap it:
If \(M\) is finitely generated, then for some \(m\) we can find a surjective \(R\)-module homomorphism
\begin{equation*} \pi: R^m \onto M. \end{equation*}
Since we assume \(R\) is noetherian, the kernel of \(\pi\) is also finitely generated by Theorem 14.11, and so we may find a surjection of \(R\)-modules
\begin{equation*} g: R^n \onto \ker(\pi) \end{equation*}
for some \(n\text{.}\) The composition \(R^n \xra{g} \ker(\pi) \subseteq R^m\) is equal to \(T_A\) for some \(n \times m\) matrix \(A\text{.}\) Since \(\im(T_A) = \ker(\pi)\text{,}\) the First Isomorphism Theorem for Modules gives an isomorphism
\begin{equation*} R^m/\im(T_A) = R^m/\ker(\pi) \cong M. \end{equation*}