Proof:

(i) Let $ \tilde G$ be the coproduct (direct sum) of the abelian groups $ \{G_{\alpha}\}$ and we regard (for simplifying notations) the groups $ G_{\alpha}$ as being subgroups of $ \tilde G$ and $ i_{\alpha} : G_{\alpha}\longrightarrow G$ the inclusion maps. Declare $ x_{\alpha}\in G_{\alpha}$ and $ x_{\beta}\in G_{\beta}$ as being equivalent if there exists $ \gamma \in \Lambda$ such that $ \gamma \geq \alpha$, $ \gamma \geq \beta$ and

$\displaystyle f_{\alpha\gamma}(x_{\alpha}) = f_{\beta\gamma}(x_{\beta}). \eqno(40.4)
$

Lemma (40.1) states that this is a well defined equivalence relation. We denote by $ \sim$ the equivalence relation just defined and define $ N$ to be the subgroup generated by

$\displaystyle \{
x_{\alpha} - x_{\beta}\;/\; x_{\alpha} \sim x_{\beta}
\}.
$

Finally let $ G = {\tilde G}/N$ and $ \eta:{\tilde G}\longrightarrow G$ be the quotient map. We claim that $ G$ is the inductive limit with respect to the maps $ f_{\alpha}$ given by the composition

$\displaystyle \begin{CD}
G_{\alpha} @> i_{\alpha}>> {\tilde G} @> \eta >> {\tilde G}/N.
\end{CD} \eqno(40.5)
$

We now check the conditions (1) and (2) in definition (40.2). For $ \alpha \leq \beta$ we derive from

$\displaystyle (f_{\beta\beta}\circ f_{\alpha\beta})(x_{\alpha}) = f_{\alpha\beta}(x_{\alpha}).
$

the useful piece of information

$\displaystyle f_{\alpha\beta}(x_{\alpha}) \sim x_{\alpha},\quad x_{\alpha}\in G_{\alpha}. \eqno(40.6)
$

Hence $ f_{\alpha\beta}(x_{\alpha}) - x_{\alpha} \in N$ whereby we conclude

$\displaystyle \eta(f_{\alpha\beta}(x_{\alpha})) = \eta(x_{\alpha}),
$

which in turn implies $ f_{\beta}\circ f_{\alpha\beta} = f_{\alpha}$. Turning to the universal property (2) assume given an abelian group $ H$ and a family of group homomorphisms $ g_{\alpha} : G_{\alpha}\longrightarrow H$ such that

$\displaystyle g_{\beta}\circ f_{\alpha\beta} = g_{\alpha},\quad \alpha\leq \beta.\eqno(40.7)
$

We first use the defining property of the coproduct to get a group homomorphism $ \phi: {\tilde G}\longrightarrow H$ such that the following diagram commutes:

$\displaystyle \xymatrix{
G_{\alpha} \ar[rr]^{i_{\alpha}}\ar[rd]_{g_{\alpha}} & & {\tilde G} \ar[ld]^{\phi}\\
& H
}
$

That is $ \phi\circ i_{\alpha} = g_{\alpha}$. From (40.4) and (40.7) we get $ g_{\alpha}(x_{\alpha}) = g_{\beta}(x_{\beta}),$ or in view of the fact that we have identified $ G_{\alpha}$ as a subgroup of $ {\tilde G}$, $ \phi(x_{\alpha}) = \phi(x_{\beta}).$ Hence there is a group homomorphism $ \psi:{\tilde G}/N\longrightarrow H$ such that

$\displaystyle \psi\circ\eta = \phi. \eqno(40.8)
$

Upon applying this to an arbitrary $ x_{\alpha}\in G_{\alpha}$ we get using (40.5) that $ \psi\circ f_{\alpha} = g_{\alpha}$ for every $ \alpha \in \Lambda$. The homomorphism satisfying (40.8) is unique since the elements $ \{f_{\alpha}(x_{\alpha})/\alpha\in \Lambda$    and $ x_{\alpha} \in G_{\alpha}\}$ generate the group $ {\tilde G}/N$.

We now prove (ii) which we shall use in the next lecture. Since $ x \in N$, there exists a finite set of indices $ \alpha_1, \beta_1, \alpha_2, \beta_2,\dots,\alpha_k, \beta_k$ such that

$\displaystyle x = \sum_{j = 1}^k (x_{\alpha_j} - x_{\beta_j}) \eqno(40.9)
$

where $ x_{\alpha_j} \sim x_{\beta_j}$ for each $ j$. Thus for each $ j$ there is a $ \gamma_j$ exceeding both $ \alpha_j$ and $ \beta_j$ such that $ f_{\alpha_j\gamma_j}(x_{\alpha_j}) = f_{\beta_j\gamma_j}(x_{\beta_j}).$ Since (40.9) spells out a relation in the direct sum of the groups $ G_{\beta}$, it decomposes into a bunch of equations namely
$\displaystyle x$ $\displaystyle =$ $\displaystyle \sum_{\alpha_i = \alpha} x_{\alpha_i} - \sum_{\beta_i = \alpha} x_{\beta_i}$  
0 $\displaystyle =$ $\displaystyle \sum_{\alpha_i = \lambda} x_{\alpha_i} - \sum_{\beta_i = \lambda} x_{\beta_i},\quad \lambda \neq \alpha$  

The index $ \lambda$ runs through a finite subset of $ \alpha_1, \beta_1,\dots,\alpha_k, \beta_k$. Taking $ \delta$ to be sufficiently large and applying $ f_{\alpha\delta}$ to the first and $ f_{\lambda\delta}$ to the second of the above displayed equations and adding we get

$\displaystyle f_{\alpha\delta}(x) = \sum_{j = 1}^k (f_{\alpha_i\delta}(x_{\alpha_j}) - f_{\beta_i\delta}(x_{\beta_j}))\eqno(40.10)
$

Using lemma (40.1) we see that if $ \delta$ is sufficiently large each of the summands on the right hand side of (40.10) is in $ N$ and so $ f_{\alpha\delta}(x) = 0$ as asserted.
nisha 2012-03-20