ProofComplete

Proof of the Structure Theorem via Smith Normal Form

We prove that every matrix over a PID can be reduced to Smith normal form by row and column operations, from which the structure theorem follows.


Smith Normal Form

Theorem9.5Smith normal form

Let RR be a PID and AA an m×nm \times n matrix over RR. There exist invertible matrices PGLm(R)P \in GL_m(R) and QGLn(R)Q \in GL_n(R) such that PAQPAQ is in Smith normal form:

PAQ=diag(d1,d2,,dk,0,,0),PAQ = \mathrm{diag}(d_1, d_2, \ldots, d_k, 0, \ldots, 0),

where k=rank(A)k = \mathrm{rank}(A), each di0d_i \neq 0, and d1d2dkd_1 \mid d_2 \mid \cdots \mid d_k.


Proof

Proof

The proof uses three types of elementary operations (each achievable by multiplication by an invertible matrix):

  1. Swap two rows (or columns).
  2. Multiply a row (or column) by a unit.
  3. Add a multiple of one row (or column) to another.

Step 1: If A=0A = 0, we are done. Otherwise, by swapping rows/columns, we may assume a110a_{11} \neq 0.

Step 2: Make a11a_{11} divide all entries in the first row and column. If a11a1ja_{11} \nmid a_{1j}, perform the Euclidean algorithm: write a1j=a11q+ra_{1j} = a_{11} q + r with 0r0 \neq r and φ(r)<φ(a11)\varphi(r) < \varphi(a_{11}) (where φ\varphi is the Euclidean function; in a general PID, use that ideals are principal to argue). Replace column jj by column jqj - q \cdot column 11, getting rr in position (1,j)(1,j). Swap columns to put rr in position (1,1)(1,1). The "size" of a11a_{11} strictly decreases, so this terminates. Similarly for column entries.

Step 3: After Step 2, a11a_{11} divides all entries in row 1 and column 1. Subtract multiples to zero them out: column jj \leftarrow column j(a1j/a11)j - (a_{1j}/a_{11}) \cdot column 1 for j2j \geq 2, and similarly for rows.

Step 4: If a11a_{11} does not divide some aija_{ij} with i,j2i, j \geq 2: add row ii to row 1, creating a nonzero entry a1j=aija_{1j} = a_{ij} in row 1 not divisible by a11a_{11}. Return to Step 2. Since the ideal (a11)(a_{11}) strictly increases each time this happens, and RR is Noetherian, this terminates.

Step 5: Now a11a_{11} divides all entries of AA. The matrix has the form (a1100A)\begin{pmatrix} a_{11} & 0 \\ 0 & A' \end{pmatrix}. Apply the procedure recursively to AA'.

Step 6: The resulting diagonal entries d1,,dkd_1, \ldots, d_k satisfy didi+1d_i \mid d_{i+1} by construction (Step 4 ensures each diagonal entry divides all later ones). \blacksquare


Deriving the Structure Theorem

RemarkFrom Smith normal form to the structure theorem

Given a finitely generated RR-module MM with nn generators, MRn/im(A)M \cong R^n / \mathrm{im}(A) for some matrix AA (the relation matrix). Reducing AA to Smith normal form diag(d1,,dk,0,,0)\mathrm{diag}(d_1, \ldots, d_k, 0, \ldots, 0) gives:

MR/(d1)R/(dk)Rnk.M \cong R/(d_1) \oplus \cdots \oplus R/(d_k) \oplus R^{n-k}.

The divisibility chain d1dkd_1 \mid \cdots \mid d_k gives the invariant factors.

ExampleComputing Smith normal form

A=(244661210416)A = \begin{pmatrix} 2 & 4 & 4 \\ -6 & 6 & 12 \\ 10 & -4 & -16 \end{pmatrix} over Z\mathbb{Z}. After row/column operations: Smith form diag(2,6,12)\mathrm{diag}(2, 6, 12). So Z3/im(A)Z/2Z/6Z/12\mathbb{Z}^3 / \mathrm{im}(A) \cong \mathbb{Z}/2 \oplus \mathbb{Z}/6 \oplus \mathbb{Z}/12.