I am reading "Introduction to the Theory of Computation 3rd Edition" by Michael Sipser.
How can we modify the proof if the alphabet in $M_1$ and the alphabet in $M_2$ are not the same?
We cannot calculate $\delta((r_1, r_2), a) := (\delta_1(r_1, a), \delta_2(r_2, a))$ if $a \in \Sigma_1 - \Sigma_2$ for example.
The proof given really gives the proof for the following theorem:
The following proposition generalizes the above theorem for languages $A_1$ and $A_2$ over different alphabets:
Proof:
Lemma: For alphabets $\Sigma_1$ and $\Sigma_2$, if $A$ is a regular language over $\Sigma_1$, then $A$ is a regular language over $\Sigma_1\cup\Sigma_2$.
Proof: Let $M_1$ recognize $A$ where $M_1=(Q,\Sigma_1,\delta_1,q_0,F)$. We will construct another DFA $M_2$ that will recognize $A$ over the alphabet $\Sigma_1\cup\Sigma_2$. Let $M_2=(Q\cup\{q'\}, \Sigma_1\cup\Sigma_2,\delta_2,q_0,F)$ where
$M_2$ recognizes $A$ so $A$ is a regular language over $\Sigma_1\cup\Sigma_2$.
By our lemma, languages $A_1$ and $A_2$ are regular languages over $\Sigma_1\cup\Sigma_2$. By Theorem $1$, $A_1\cup A_2$ is a regular language over alphabet $\Sigma_1 \cup\Sigma_2$. This completes the proof of the proposition.