Let $A$ be a non-trivial integral domain. Define the relation $\sim$ on the set of pairs $A \times A\setminus\{0_A\}$ as follows:

$$(a_1,b_1) \sim (a_2,b_2) \overset{\text{def}}{\Longleftrightarrow} a_1b_2=a_2b_1.$$

It turns out that $\sim$ is an equivalence relation on $A \times A\setminus\{0_A\}$. Addition and multiplication procedure is defined as follows.

$$(a_1,b_1)+(a_2,b_2) \overset{\text{def}}{=} (a_1b_2+a_2b_1,b_1b_2)\\(a_1,b_1)\cdot(a_2,b_2)\overset{\text{def}}{=}(a_1a_2,b_1b_2).$$

If one wishes to define such operations similarly on the set of equivalence classes by $\sim$, that is on the set $(A \times A\setminus\{0_A\})/\!\sim$, one must prove the operations agree with the relation $\sim$. In other words, it must be shown these procedures give a well-defined function, not depending on the choice of representative from an equivalence class.

Here is how I would prove the result in the case of addition.

Let $(a,b)\sim(a_1,b_1)$ and $(c,d) \sim (c_1,d_1)$ be any pairs in $A \times A\setminus\{0_A\}$. We need to show that $(a,b)+(c,d)$ is $\sim$-equivalent to $(a_1,b_1)+(c_1,d_1)$, that is $(ad+bc)b_1d_1 = (a_1d_1+b_1c_1)bd.$

Hence, look at the expression $E:=(ad+bc) b_1d_1$. Using distributivity in $A$, we have $E=(ad)b_1d_1+(bc)b_1d_1$. Using commutativity (and associativity) of multiplication, $E=(ab_1)dd_1+(cd_1)bb_1$. But because $(a,b)\sim(a_1,b_1)$ and $(c,d) \sim (c_1,d_1)$, we may replace $ab_1=a_1b$, and $cd_1=c_1d$. Therefore, $E=(a_1b)dd_1+(c_1d)bb_1$. Again via distributivity (and commutativity, associativity), finally $E=(a_1d_1+b_1c_1)bd$. QED

Here is how E. B. Vinberg does it in A Course of Algebra, page 130.

Define now addition and multiplication of pairs by the following rules: $$(a_1,b_1)+(a_2,b_2) = (a_1b_2+a_2b_1,b_1b_2)\\(a_1,b_1)(a_2,b_2)=(a_1a_2,b_1b_2).$$ We will prove that the equivalence relation defined above agrees with these operations. By the preceding discussion, it suffices to show that when we multiply both entries in one of the pairs $(a_1,b_1)$ or $(a_2,b_2)$ by the same element $c$, their sum and product get replaced by equivalent pairs. But it is clear that when we do this, both entries in the sum and the product are multiplied by $c$.

(Emphasis added by me).

Q: Why does it suffice to show only what Vinberg says?

To emphasise, "the preceding discussion" is quoted in either my previous question in yellow quote boxes, or here in this post. The order of the book is preserved. I thought it would be a poor idea to again quote the full passage here due to length. Of course, I am willing to do so if necessary; in such a case, please leave an appropriate comment.


Vinberg implicitly defines a relationship which we'll call $\sim_1:$

$(a_1,b_1)\sim_1 (a_2,b_2)$ if $\exists c\in A\setminus \{0\}$ such that $a_1c=a_2,b_1c=b_2.$

This is not an equivalence relation. ($\sim_1$ is actually a pre-order.)

Vinberg shows in the prior discussion that $\sim_1$ has the property:

Lemma 1: If $(a_1,b_1)\sim_1(a_2,b_2)$ then $(a_1,b_1)\sim (a_2,b_2)$

and also the property:

Lemma 2: $(a_1,b_1)\sim (a_2,b_2)$ if and only if there exists $(a_3,b_3)$ such that $(a_1,b_1)\sim_1 (a_3,b_3)$ and $(a_2,b_2)\sim_1 (a_3,b_3).$

Those two properties are the key.

Now Vinberg is saying we only need to show:

Lemma 3: For $p\sim_1 p_1$ and any $q$ that: $$\begin{align}p+q&\sim p_1+q\text{ and }\\ q+p&\sim q+p_1\end{align}\tag{1}$$

and similarly for multiplication.

From Lemma 3 we prove the general case:

Theorem: If $p\sim p_1$ and $q\sim q_1$ then $p+q\sim p_1+q_1.$

Proof: By Lemma 2, there must have $p_2,q_2$ such that $p\sim_1 p_2, p_1\sim_1 p_2, q\sim_1 q_2, q_1\sim_1 q_2.$

Then we have: $$p+q\sim p_2+q\sim p_2+q_2$$ by (1), and so $p+q\sim p_2+q_2.$

Likewise, we have $p_1+q_1\sim p_2+q_2.$

So we've shown: $p+q\sim p_1+q_1.$

The same works for multiplication.

It is easier to show the stronger statement:

For $p\sim_1 p_1$ and any $q$, $$\begin{align}p+q&\sim_1 p_1+q\text{ and }\\ q+p&\sim_1 q+p_1,\end{align}\tag{1'}$$

and then deduce Lemma 3 from (1') using Lemma 1.

  • $\begingroup$ Hmm. Interesting answer. Of course, the caveat with my type of question "What person X had in mind", is always that some inherent subjectivity is woven in from the beginning. Yet there are a few missing links that don't yet fall in place for me if this answer is indeed what Vinberg had in mind. The first is a smaller quibble, Vinberg first mentions and supports the commutativity of $+$ a bit further along in the text. Secondly, and more importantly, where is the multiplication with the element $c\in A\setminus\{0_A\}$ used in your argument? $\endgroup$ – Linear Christmas Jul 2 '19 at 18:49
  • $\begingroup$ To be clear, I am not saying your answer is wrong at all, it proves the result itself nicely. But could you further clarify why you feel this is what Vinberg had in mind? I need more convincing on the latter part. $\endgroup$ – Linear Christmas Jul 2 '19 at 18:49
  • $\begingroup$ Okay, I've read it more completely, and corrected my answer. @LinearChristmas $\endgroup$ – Thomas Andrews Jul 2 '19 at 19:36
  • $\begingroup$ I haven't read any further at the moment but lemma 1 seems unclear (and may not hold depending on what is meant). It is true that $\sim$ is the smallest equivalence relation $\approx$ such that $\sim_1$ is true. Is this what you had in mind? $\endgroup$ – Linear Christmas Jul 2 '19 at 19:58
  • $\begingroup$ Why is Lemma 1 false? It just says that $(a_1,b_1)\sim(ca_1,cb_2)$ for non-zero $c.$ It does not imply the other way around - that all $p\sim q$ have $p\sim_1 q.$ Lemma 2 gives the more complicated definition if $\sim$ in terms of $\sim_1.$. @LinearChristmas $\endgroup$ – Thomas Andrews Jul 2 '19 at 20:01

Recall that the scaling relation $\,\sim:\,$ is defined as $\, (a,b) \sim: (c,d)\iff (c,d) = (ea,eb)\,$ for some $\,e\neq 0,\,$ i.e. $\,\large \frac{a}b \sim: \frac{e\,a}{e\,b}.\,$ They have equal cross-multiples $\,eab\,$ so $\,f\sim:g\,\Rightarrow\, f\sim g.$

The Lemma in the prior question shows that every cross-multiplication equivalence $\,f_1\sim f_2\,$ can be decomposed into a pair of scaling relations, i.e. $\,f_1\sim f_2\iff f_1\sim:f:\sim f_2\ $ for some $\,f,\,$ i.e. $\,f_1,\,f_2\,$ are cross-multiplication equivalent $\iff$ they have a common scaling $\,f.\,$

Thus it suffices to prove that addition and multiplication are compatible with the scaling relation, which follows from scaling symmetry of the addition & multiplication formulas due to their linear form, i.e. $\, s(f_1)\sim: \color{#c00}e\,s(f_1) = s(\color{#c00}ef_1) = s(f)\,$ below, where we prove compatibility for the first argument of addition using the sum function$\ s(x) := x + g_1,\, $ for $\,g_1 = (c,d).$

$\ \ \ \ \ \ \ \begin{align}f_1 + g_1\ \ \ \ \ &\sim: \ \ \ \ \ f + g_1 \\[.2em] f_1 \ \ \ \sim:\ \ \ \ f \ \ \ \ \, \smash[t]{\color{#0a0}{\overset{\rm C}\Longrightarrow}}\, \ \ \ \ \ \ \ \ s(f_1)\ \ \ \ \ \ \ & \sim:\ \ \ \ \ \ \ s(f)\\[.2em] \ {\rm i.e.}\ \ \ \ (a,b)\sim:(ea,eb)\,\Rightarrow\, (a,b)+(c,d)&\sim: (\color{#c00}ea,\color{#c00}eb)+(c,d)\ \ = \ s(\color{#c00}ef_1) \\[.2em] {\rm by}\ \ \ \ (ad\!+\!cb,\,bd) &\sim: (\color{#c00}ead\!+\!\color{#c00}ecb,\,\color{#c00}ebd)\ \ = \ \color{#c00}e\,s(f_1) \end{align}\ \ \ \ \ \qquad$

${\rm Then}\ \ f_1\sim f_2\,\Rightarrow\, s(f_1)\sim s(f_2)\,$ follows by applying $\,\smash[t]{\color{#0a0}{\overset{\rm C}\Rightarrow}}\,$ to a $\,\sim:\,$ decomposition of $\, f_1 \sim f_2\,$

$\ \ \ \ \ \ \ \ \ \, f_1\sim f_2\,\Rightarrow\begin{align}f_1\sim: f\\[.2em] f_2\sim: f\end{align}$ $\:\color{#0a0}{\overset{\rm C}\Rightarrow}\,\begin{align}s(f_1)\sim: s(f)\\[.2em] s(f_2)\sim: s(f)\end{align}$ $\,\Rightarrow\begin{align}s(f_1)\sim s(f)\\[.2em] s(f_2)\sim s(f)\end{align}$ $\,\color{#08ff}\Rightarrow\! \begin{align} s(f_1)\,&\sim\, s(f_2),\,\ {\rm i.e.}\\[.2em] f_1+g_1&\sim \color{#08f}{f_2+g_1}\end{align}$

Similarly (or using symmetry and commutativity) we get $\ g_1\sim g_2\,\Rightarrow\, \color{#08f}{f_2+g_1}\sim f_2+ g_2\,$ thus

$\rm\color{#08f}{transitivity}$ of $\,\sim\,$ yields $\,\ \ f_1\sim f_2,\ g_1\sim g_2\,\Rightarrow\, f_1+g_1\sim f_2+g_2\qquad $

which means $\,\sim\,$ is compatible with addition. Multiplication compatibility follows similarly.

Remark $ $ These tedious proofs are usually "left to the reader" in most expositions. One can avoid this by instead using a more algebraic construction of fraction rings via quotients of polynomial rings, where we adjoin an inverse $\,x_a\,$ for each $\,a\neq 0\,$ via extension rings $\, A_j[x_a]/(ax_a-1).\,$

In this approach the proofs follow easily from universal properties of polynomial and quotient rings. The two approaches are related by the fact that the fraction pairs correspond to normal forms in these quotients rings, where every element is equivalent to a monomial $\,a\, x_{a_1}\cdots x_{a_k}\,$ (essentially by choosing a $ $ common "denominator"), $ $ denoted by the $ $ "fraction" $\,a/(a_1\cdots a_k)\,$ or, set-theoretically, by the pair $\,(a,\,a_1\cdots a_k),\,$ analogous to Hamilton's pair-representation of complex numbers $\,(a,b),\,$ corresponding to normal forms (least degree reps) $\,a+bx\,$ in $\,\Bbb R[x]/(x^2\!+1)\cong C.\,$ For more on this viewpoint see here (there we consider a more general construction which inverts elements in some specified subset $\,S\subseteq A)$

  • $\begingroup$ I have gone over your answer as well. It seems you took advantage of the theorem Thomas Andrew's calls "stronger statement" at the end of their answer (your implication C). That's great because the answers then complement each other. Your answer seems correct to me (besides the part starting with "Remark" as it is a bit over my head, and thus cannot directly verify). Now on to more specific questions. You say "addition formula is multilinear". What is the formal logical statement you have in mind for this "multilinearity"? (The general definition but exactly in the form it it used here). $\endgroup$ – Linear Christmas Jul 5 '19 at 15:50
  • $\begingroup$ @Lin Yes, that's essentially what I do. Re: multilinear, here this means that the sum formula's numerator $\,p(a,b) = ad+cb\,$ satisfies $\,p(ea,eb) = \color{#c00}e\,p(a,b)\,$ and its denom $\,q(a,b) = bd\,$ satisifies $\,q(ea,eb) = \color{#c00}e\,q(a,b),\,$ which makes it clear why scaling $\,(a,b)\,$ by $\,e\,$ has the effect of scaling the sum by $\,\color{#c00}e\,$ (cf. answer's red $\,\color{#c00}e's$). Presumably such multilinearity is what Vinberg implicitly refers to when he wrote "But it is clear that when we do this, both entries in the sum and the product are multiplied by $e$". $\endgroup$ – Bill Dubuque Jul 5 '19 at 16:49
  • $\begingroup$ I tried to present it in a form that make the argument clear, e.g. above you should visualize how the ">"-shaped subgraph of $\,\sim:\,$ (pointing $\,f_1, f_2\,$ at their common scaling $\,f)$ is preserved by applying $\,s\,$ to it, and that lifts to the same ">" shaped subgraph of $\,\sim\,$ (erase the ":"s using $\,f\sim:g\,\Rightarrow\,f\sim g),\,$ which yields the the sought result by transitivity (around the ">" in $\,s(f_1)\sim s(f)\sim s(f_2)$ with $\,s(f_1),s(f_2)$ pointed at their common scaling $\,s(f))$. This allows us to "see" the proof in a single glance (can't animate here). $\endgroup$ – Bill Dubuque Jul 5 '19 at 16:49
  • $\begingroup$ @LinearChristmas If there are any points that remain unclear then let me know, esp. if you can't see the proof in a single glance (as intended). $\endgroup$ – Bill Dubuque Jul 5 '19 at 16:56
  • $\begingroup$ So it is not "multilinearity" in the most common sense of the word wrt multilinear functions. In the usual sense, it would also require $p(ea,eb) = e^2p(a,b)$ instead of scaling simply by $e$, correct? As for seeing the proof at a single glance, I am not sure what you mean by that. I think I do understand each step in your argument, if that's what you mean. And by ">"-shaped subgraph did you have in mind "<"-shaped graph? If not, then perhaps something is rendered differently, or I am misunderstanding what you said. $\endgroup$ – Linear Christmas Jul 5 '19 at 17:14

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.