Trouble understanding direct sum in smith normal form proof

72 Views Asked by At

I'm struggling to understand this step in this proof in Rotman's advanced modern algebra. The related theorems are:

Let R be a euclidean ring, let F be a finitely generated free R-module, and let S be a submodule of F. Then there exists a basis $z_1, . . . , z_n$ of F and nonzero $σ_1, . . . , σ_q$ in R, where 0 ≤ q ≤ n, such that $σ_1 | · · · | σ_q$ and $σ_1z_1, . . . , σ_qz_q$ is a basis of S.

Let R be a euclidean ring, let Γ be the n × t presentation matrix associated to an R-map $λ: R^t → R^n$ relative to some choice of bases, and let M = cokerλ. If Γ is R-equivalent to a Smith normal form diag$(σ_1, . . . , σ_q)$ ⊕ 0, then those $σ_1, . . . , σ_q$ that are not units are the invariant factors of M.

The proof of the second theorem using the first is:

If diag$(σ_1, . . . , σ_q)$ ⊕ 0 is a Smith normal form of Γ, then there are bases $y_1, . . . , y_t$ of $R^t$ and $z_1, . . . , z_n$ of $R^n$ with $λ(y_1) = σ_1z_1, . . . , λ(y_q) = σ_qz_q$ and $λ(y_j)$ = 0 for all $y_j$ with j > q, if any. Now R/(0) $\cong$ R and R/(u) = {0} if u is a unit. If $σ_s$ is the first $σ_i$ that is not a unit, then M $\cong$ $R^{n−q}$$R/(σ_s)$ ⊕· · ·⊕ $R/(σ_q)$ a direct sum of cyclic modules for which $σ_s | · · · | σ_q$.

From the first line of the proof I gather that $σ_1z_1, . . . , σ_qz_q$ is a basis of the free submodule im(λ) so then M $\cong R^t/< σ_1z_1, . . . , σ_qz_q >$ ? What I don’t understand is how the bottom line (if this is correct) decomposes into $(σ_s)$ ⊕· · ·⊕ $(σ_q)$ i.e. where does z go and why is it not $(σ_sz_1)$ ⊕· · ·⊕ $(σ_qz_q)$ (or are these equivalent).

Any help would be appreciated.

1

There are 1 best solutions below

0
On

Where does "$z$" go? $R/(\sigma z)$ doesn't mean anything as $z\notin R$. You can think of these quotients as coming from $Rz/(\sigma z)$, where $Rz$ is the span of $z$ in $R^n$, and these are isomorphic to $R/(\sigma)$. This is basically what is happening.

Formally, consider the map $R^n\to R^{n-q}\oplus\bigoplus_{j=s}^qR/(\sigma_j)$ as defined by $\sum_ir_iz_i\mapsto(\sum_{i>q}r_iz_i,\overline{r_s},\overline{r_{s+1}},\cdots,\overline{r_q})$. This is obviously a map of $R$-modules and is well defined since the $(z_\bullet)$ are a basis. It clearly also is surjective. Therefore, the only thing we need to do is prove the kernel of this map is $\mathrm{im}(\lambda)$.

By linear independence, $\sum_{i>q}r_iz_i$ is zero iff. $r_i=0$ for all $i>q$. The other terms are zero iff. $r_j=x_j\cdot\sigma_j$ for some $x_j\in R$, for all $s\le j\le q$. $r_i$ for $i<s$ can be absolutely anything. Therefore, the kernel is the span of $\{z_1,z_2,\cdots,z_{s-1},\sigma_sz_s,\cdots,\sigma_qz_q\}$. As $\sigma_i$ is a unit for $i<s$, this is equal to the span of $\{\sigma_1z_1,\cdots,\sigma_{s-1}z_{s-1},\sigma_sz_s,\cdots,\sigma_qz_q\}$. Oh, but that's exactly the image of $\lambda$!

Therefore $M\cong R^{n-q}\oplus\bigoplus_{j=s}^qR/(\sigma_j)$ and all terms of this direct sum, except possibly $R^{n-q}$, are nonzero modules.