Skip to main content

Section 14.3 Classifications

Subsection Invariant Factors

Let’s see how to classify modules over PIDs using the Smith Normal Form for their presentation matrix. First, we need a lemma on how to interpret the module presented by a matrix in Smith Normal Form.

Proof.

By Corollary 14.13, \(M\) has a presentation matrix \(A\text{.}\) By Theorem 14.14, \(A\) can be put into Smith Normal Form \(B\text{,}\) where the diagonal entries of \(B\) are \(b_{1}, \ldots, b_{\ell}\) and satisfy \(b_{1}\left|b_{2}\right| \cdots \mid b_{k}\text{.}\) Moreover, \(k\) is unique and the \(d_{i}\) are uniquely determined up to associates (ie, up to multiplication by units) by \(A\text{,}\) hence by \(B\text{.}\) By Corollary 14.13, \(M\) is isomorphic to the module presented by \(B\text{.}\) By Lemma 14.26, this is isomorphic to
\begin{equation*} M \cong R^{r} \oplus R /\left(b_{1}\right) \oplus \cdots \oplus R /\left(b_{\ell}\right) \end{equation*}
Finally, some of these \(b_{i}\) might be units; let \(d_{1}|\cdots| d_{k}\) be the nonunits among the \(b_{i}\text{,}\) and note that if \(u\) is a unit, then \(R /(u) \cong(0)\text{.}\) We conclude that
\begin{equation*} M \cong R^{r} \oplus R /\left(d_{1}\right) \oplus \cdots \oplus R /\left(d_{k}\right), \end{equation*}
as desired

Definition 14.28. Invariant Factors.

Let \(R\) be a PID, let \(r \geqslant 0, k \geqslant 0\text{,}\) and let \(d_{1}, \ldots, d_{k}\) be nonzero nonunit elements of \(R\) satisfying \(d_{1}\left|d_{2}\right| \cdots \mid d_{k}\text{.}\) Let \(M\) be any \(R\)-module such that
\begin{equation*} M \cong R^{r} \oplus R /\left(d_{1}\right) \oplus \cdots \oplus R /\left(d_{k}\right) . \end{equation*}
We say \(M\) has free rank \(r\) and invariant factors \(d_{1}, \ldots, d_{k}\text{.}\)

Remark 14.29.

The classification theorem can be interpreted as saying that \(M\) decomposes into a free submodule \(R^{r}\) and a torsion submodule \(\operatorname{Tor}(M)=R /\left(d_{1}\right) \oplus \cdots \oplus R /\left(d_{k}\right)\text{.}\)

Example 14.30. \(k[x]\)-Modules of Dimension \(4\).

Let \(k\) be a field and \(M\) a \(k[x]\)-module such that the dimension of \(M\) as a \(k\)-vector space is \(4.\) What are the possibilities for \(M\) up to isomorphism?
We have
\begin{equation*} M \cong k[x]^r \oplus k[x]/(f_1(x)) \oplus \cdots \oplus k[x]/(f_s(x)) \end{equation*}
with \(r, s, f_1, \dots, f_s\) as in the Corollary. But \(r\) must be \(0\) since \(k[x]\) is infinite dimensional as a \(k\)-vector space. Moreover, \(\dim_F k[x]/g = \deg(g)\) for any non-zero polynomial \(g\text{.}\) So we must have \(\sum_i \deg(f_i) = 4\text{.}\) There are five possibilities:
  1. \(s = 1\) and \(f_1\) is monic of degree \(4\text{.}\)
  2. \(s = 2\text{,}\) \(f_1\) is linear, \(f_2\) is cubic and \(f_1 \mid f_2\text{.}\) (So if \(f_1 = x-a\text{,}\) then \(a\) must be a root of \(f_2\)).
  3. \(s = 2\text{,}\) \(f_1 = f_2\) is quadratic.
  4. \(s = 3\text{,}\) \(f_1 = f_2\) is linear and \(f_3\) is quadratic with \(f_1 \mid f_3\text{.}\)
  5. \(s = 4\text{,}\) \(f_1 = f_2 = f_3 = f_4\) is linear.
Now suppose \(k = \Z/p\text{.}\) What is the total number of possibilities? For case \(1\text{,}\) there are \(p^4\) monic polynomial of degree \(4\text{.}\) For case \(2\text{,}\) there are \(p\) choices for \(f_1\) and \(p^2\) choices for \(f_2\) since \(f_2 = f_1 \cdot q\) for a unique quadratic \(q\text{,}\) for a total of \(p^3\) possibilities. For case \(3\) there are \(p^2\) choices. For case \(4\) there are \(p^2\) choices since there are \(p\) choices for \(f_1 = f_2\) and \(f_3 = f_1 \cdot l\) for a unique linear \(l\text{.}\) For case \(5\text{,}\) there are \(p\) choices for \(f_1\text{.}\) In total there are \(2p + 2p^2 + p^4\) such modules up to isomorphism.

Subsection Elementary Divisors

Here is a spinoff of the classification theorem.

Proof.

First, using Theorem 14.27 write \(M\) in invariant factor form \(M \cong R^{r} \oplus R /\left(d_{1}\right) \oplus \cdots \oplus R /\left(d_{k}\right)\text{.}\) Then write each invariant factor as a product of prime powers
\begin{equation*} d_{i}:=\prod_{j=n_{i}}^{n_{i+1}} p_{j}^{e_{j}}, \end{equation*}
and recall that by the CRT we have
\begin{equation*} R /\left(d_{i}\right) \cong R /\left(p_{n_{i}}^{e_{n_{i}}}\right) \oplus \cdots \oplus R /\left(p_{n_{i+1}}^{e_{n_{i+1}}}\right) . \end{equation*}
Substituting into the invariant factor form gives the desired result. Uniqueness follows from the uniqueness of the invariant factor form and of the prime factorizations of each \(d_{i}\text{.}\)

Proof.

Note that \(I \cdot J \subseteq I \cap J\) holds in general for any pair of ideals. If \(I + J = R\) then \(a + b = 1\) for some \(a \in I\) and \(b \in J\text{.}\) Given \(x \in I \cap J\) we have \(xa + xb = x\) with \(xa \in I \cdot J\) and \(xb \in I \cdot J\text{,}\) which proves \(x \in I \cdot J\text{.}\)
For the second assertion define \(g: R \to R/I \oplus R/J\) to be the \(R\)-module homomorphism \(g(r) = (r +I, r + J)\text{.}\) Note that the kernel of \(g\) is \(I \cap J\) which equals \(I\cdot J\) by the first assertion. I claim \(g\) is onto: Pick \((x + I, y +J) \in R/I \oplus R/J\text{.}\) With \(a,b\) chosen as above, we have
\begin{equation*} g(xb + ya) = (xb + ya + I, xb + ya + J) = (xb + I, ya + J) = (x + I, y +J). \end{equation*}
The last equation holds since \(xa + xb = x\) and thus \(xb + I = xb + xa + I = x + I\) and similarly \(ya + J = y + J\text{.}\)
For the final assertion, recall that when \(R\) is a UFD, two principle ideals \(Rx\) and \(Ry\) satisfy \(Rx + Ry = R\) if and only if \(x\) and \(y\) are relatively prime. Also, for \(I = Rx\) and \(J = Ry\text{,}\) we have \(I\cdot J = R(xy)\text{.}\)

Definition 14.33. Elementary Divisor (Module).

With the notation in Theorem 14.31 above, the elements \(p_1^{e_1},\ldots, p_s^{e_s}\) of \(R\) are the elementary divisors of \(M\text{.}\)

Example 14.34. \(\Z/90\).

Since \(\Z/90 \cong \Z/9 \oplus \Z/2 \oplus \Z/5\text{,}\) so the elementary divisors of \(\Z/90\) are \(9\text{,}\) \(2\) and \(3\text{.}\) The only invariant factor of \(\Z/90\) is \(90\text{.}\)

Example 14.35. Direct Sums and \(\Z\).

Consider the group \(G = \Z/90 \oplus \Z/9 \oplus \Z/6 \oplus \Z/3 \oplus \Z/4\text{.}\) This is neither in IFF nor in EDF. Applying the Sunzi Remainder Theorem, we have
\begin{equation*} G \cong \Z/9 \oplus \Z/2 \oplus \Z/5 \oplus \Z/9 \oplus \Z/2 \oplus \Z/3 \oplus \Z/3 \oplus \Z/4, \end{equation*}
and this gives the EDF. The elementary divisors of \(G\) are \(9,2,5,9,2,3,3,4\) (ordering does not matter).
To find the IFF we start by finding the largest prime power order for each prime in the list of orders of the summands. These are \(9, 5, 4\text{.}\) The CRT gives
\begin{equation*} G \cong \Z/(9 \cdot 4 \cdot 5) \oplus \Z/2 \oplus \Z/9 \oplus \Z/2 \oplus \Z/3 \oplus \Z/3. \end{equation*}
Then we find the highest prime power orders for each prime among the left-over summands: \(9, 2.\) By the CRT we have
\begin{equation*} G \cong \Z/(9 \cdot 4 \cdot 5) \oplus \Z/(2 \cdot 9) \oplus \Z/2 \oplus \Z/3 \oplus \Z/3. \end{equation*}
The highest orders of the prime power order not yet used are \(2\) and \(3\text{,}\) and we have
\begin{equation*} G \cong \Z/(9 \cdot 4 \cdot 5) \oplus \Z/(2 \cdot 9) \oplus \Z/(2 \cdot 3) \oplus \Z/3. \end{equation*}
which can be rearranged to give
\begin{equation*} G \cong \Z/3 \oplus \Z/6 \oplus \Z/18 \oplus \Z/180. \end{equation*}
This is the IFF, and the invariant factors of \(G\) are \(3,6,18,180\text{.}\)

Example 14.36. Finding IFs and EDs.

Find the invariant factor form and the elementary divisor form of the \(k[x]\)-module
\begin{equation*} M = k[x]/(x^2-1) \oplus k[x]/(x-1) \oplus k[x]/(x^2 + 1) \oplus k[x]/(x^2 + 2x + 1) \end{equation*}
first when
  1. \(k = \Z/2\text{,}\)
  2. \(k = \R\) and
  3. \(k = \C\text{.}\)