# Does commutativity imply Associativity?

Does commutativity imply associativity? I'm asking this because I was trying to think of structures that are commutative but non-associative but couldn't come up with any. Are there any such examples?

NOTE: I wasn't sure how to tag this so feel free to retag it.

• Commutative operations that are associative are the exception. But an important exception! Let $x\ast y=|x-y|$. – André Nicolas Jun 20 '12 at 20:41
• Not even in the presence of an identity element and an opposite, see this. In fact, William's answer is already in that post ;) – Bruno Stonek Jun 20 '12 at 21:58
• The interchange law $(x * y) \cdot (z * w) = (x \cdot z) * (y \cdot w)$, in the presence of a two-sided common unit element, implies commutativity and associativity of $*$ and $\cdot$. (In fact, they have to be the same operation!) – Zhen Lin Jun 20 '12 at 22:14
• See my 3 February 2009 sci.math post A natural example of a commutative, non-associative operator (see Google archive version or Math Forum archive version) for some examples and references. – Dave L. Renfro Jun 22 '12 at 21:31

## 12 Answers

Consider the operation $(x,y) \mapsto xy+1$ on the integers.

• @marlu Could you explain this some more please? It doesn't say why, or (for people like me who don't know much about maths), what that arrow even means. – user673679 Jan 12 '15 at 13:13
• Well, associativity and commutativity are properties of maps $X\times X \to X$ for a set $X$. In other words, such a map takes two elements as an "input" and returns a single element. In my example, the set under consideration is the set of integers and the map sends each pair of integers $(x,y)$ to $xy+1$. Commutativity means $xy + 1 = yx + 1$ for all $x$ and $y$, which is satisfied. Associativity would mean $x(yz+1) + 1 = (xy+1)z+1$ for all $x$, $y$ and $z$, but it's easy to find examples where this equation does not hold, so the operation is not associative. – marlu Jan 14 '15 at 4:14
• Short, sweet. Nice. – AJFarmar Mar 20 '15 at 20:55
• But wait, this is unfair — you're considering commutativity under addition, but associativity under both addition and multiplication, they're different operations! ☹ – Hi-Angel Jan 14 '16 at 18:28
• @Hi-Angel: Commutativity and associativity are properties of a single binary operation; in this case, the operation defined by marlu above. Denoting this operation by, say, the symbol $\odot$, it is clearly commutative since $x \odot y = xy+1 = yx+1 = y \odot x$, but not associative since $(x \odot y) \odot z = (xy+1)z+1 = xyz+z+1 \ne xyz+x+1 = x(yz+1)+1 = x \odot (y \odot z)$ whenever $x \ne z$. – Ilmari Karonen Mar 4 '18 at 18:11

A basic example is the "midpoint" binary operation: $a*b = \frac{a+b}{2}$

In general, if $P(u,v)$ is any polynomial in two variables with rational coefficients, then $x*y = P(x+y,xy)$ is rarely associative - I'd be curious under what conditions on $P$ this operation would be associative.

My example is $P(u,v)=\frac{u}{2}$ and Marlu's example is $P(u,v)=1+v$.

• In general, a symmetric function $P(x,y)$ is rarely associative. You could take $P(x,y)=\arctan xy+e^{(x+y)^9}$, too. – Mariano Suárez-Álvarez Sep 14 '12 at 16:38

Arguably the most important example of a commutative but non-associative structure is that of finite-precision floating point numbers under addition. (a + -a) + b is always equal to b but a + (-a + b) can differ from b since the sum -a + b can involve a loss of precision (this is especially true if a and b are nearly but not quite equal, -a + b could work out to 0 even though the corresponding real sum is nonzero). The lack of associativity of floating point arithmetic is a constant complicating factor in numerical analysis.

• This is a nice example, because it reminds us how important these concepts are in practical life, even though we often do not think about them..(+1) – Björn Friedrich Jun 14 '16 at 16:46

The easiest Jordan algebra is symmetric square matrices with the operation $$A \ast B = (AB + BA)/2,$$ similar to a Lie algebra but with a plus sign.

http://en.wikipedia.org/wiki/Jordan_algebra

Consider the commutative operation $\texttt{vs}$ on the set $\{\textbf{rock}, \textbf{paper}, \textbf{scissors}\}$ abbreviated $\{r,p,s\}$ defined by $$\begin{array}{c|ccc} \texttt{vs} & r&p&s\\ \hline r & r & p & r \\ p & p & p & s \\ s & r & s & s \end{array}$$ It is not associative since, for example, $$\textbf{paper} \texttt{ vs } (\textbf{scissors} \texttt{ vs } \textbf{rock}) = \textbf{paper}$$ but $$(\textbf{paper} \texttt{ vs } \textbf{scissors}) \texttt{ vs } \textbf{rock} = \textbf{rock}.$$

Let $A = \{e,x,y\}$. Define $\cdot$ on $A$ to be $a\cdot e=a$ for all $a$, $e\cdot a= a$ for all a, and $a\cdot b=e$ for all $a$ and $b$ such that $a\neq e$ and $b\neq e$, (i.e. $a,b \in \{x,y\}$).

This operation is commutative, $e$ is the identity, (everything even has an inverse), but is not associative since $(x \cdot y) \cdot y = e \cdot y = y$ and $x \cdot (y \cdot y) = x \cdot e = x$.

The simplest examples of commutative but nonassociative operations are the NOR and NAND operations (joint denial and alternative denial) in propositional logic. Quoting from my answer to the question A conjecture in equational logic:

Namely, the $$2$$-element structure $$\{a,b\}$$, where $$aa=b$$ and $$ab=ba=bb=a$$, is commutative but not associative; in fact, for any $$x\in\{a,b\}$$, we have $$(ax)b=b(xa)=a,$$ $$a(xb)=(bx)a=b.$$ This is the unique (up to isomorphism) binary operation on a $$2$$-element set which is commutative but not associative; it can be interpreted as either of the truth-functions NOR or NAND.

The NAND is commutative but not associative.

\begin{eqnarray*} \begin{array}{|c|c|c|c|c|c|c|c|c|c|c|c|} \hline A & B & & A \text{ nand } A & (A \text{ nand } A) \text{ nand } B & & A \text{ nand } B & A \text{ nand } (A \text{ nand } B) \\ \hline 0&0& &1&1& &1&1 \\ \hline 0&1& &1&0& &1&1 \\ \hline 1&0& &0&1& &1&0 \\ \hline 1&1& &0&1& &0&1 \\ \hline \end{array} \end{eqnarray*}

• Please excuse ... but I went to the trouble of constructing the truth table ... feel free to remove it. – Donald Splutterwit Sep 26 '17 at 20:17
• @DonaldSplutterwit I would use three variables instead were I you. – Kenny Lau Sep 26 '17 at 20:18
• I already mentioned the examples NAND and NOR in my answer in 2015, so what was your point? – bof Aug 19 at 2:14

The arithmetical, geometrical and harmonic mean operations on $$\mathbb{R}$$, $$\mathbb{R^+}$$ and $$\mathbb{R_*^+}$$ resp.:

$$a*b=\dfrac{a+b}{2}$$

$$a*b=\sqrt{ab}$$

$$a*b=\dfrac{ab}{a+b}$$

Remark: the non-associativity of arithmetical mean has a kind of physical interpretation by placing more weight either on the last or on the first term:

$$(a*b)*c=\dfrac{\tfrac{a+b}{2}+c}{2}=\dfrac{a+b+2c}{4}$$

whereas

$$a*(b*c)=\dfrac{2a+b+c}{4}$$

(none of them being equal to $$\dfrac{a+b+c}{3}$$ !...)

For $x,y \in \mathbb{Z}$, define $x*y = xy + 1$.

Then * is clearly commutative.

As for associativity, $$a*(b*c) = a*(bc + 1) = a(bc + 1) + 1 = abc + a + 1$$ $$(a*b)*c = (ab + 1)*c = (ab + 1)c + 1 = abc + c + 1$$ so associativity fails for any triple $(a,b,c)$ with $a \ne c$.

Here's another example . . .

For $x,y \in \mathbb{Z}$, define $x y = x^2y^2$.

Once again, commutativity is obvious.

For associativity, $$a*(b*c) = a*(b^2c^2) = a^2(b^2c^2)^2 = a^2b^4c^4$$ $$(a*b)*c = (a^2b^2)*c = (a^2b^2)^2c^2 = a^4b^4c^2$$ so associativity fails if $a,b,c \ne 0$, and $|a| \ne |c|$.

One last example . . .

For $x,y \in \mathbb{Z}$, define $x*y = -x-y$.

Commutativity is clear.

For associativity, $$a*(b*c) = a*(-b-c) = -a-(-b-c) = -a + b + c$$ $$(a*b)*c = (-a-b)*c = -(-a-b)-c = a + b -c$$ so associativity fails if $a \ne c$.

Using mace4 on the assumption x*y=y*x. and goal (x*y)*z=x*(y*z). gives:

% Interpretation of size 2

*(0,0) = 1.
*(0,1) = 0.
*(1,0) = 0.
*(1,1) = 0.

c1 = 0.

c2 = 0.

c3 = 1.


i.e. 0*0=0, 0*1=0, 1*0=0, 1*1=0, with (x,y,z)=(0,0,1) being the counter-example.

(0*0)*1 = 1*1 = 0
0*(0*1) = 0*0 = 1


P.S. Mace4 is a software used for constructing finite models and finding counter-examples.

• Sorry, but your answer uses a "mace4" software that is "not very well known" (euphemism)... – Jean Marie Sep 26 '17 at 20:03
• @JeanMarie so what? – Kenny Lau Sep 26 '17 at 20:04
• So, I give you a [+1] to your answer, which is good... but... – Jean Marie Sep 26 '17 at 20:05
• (Thanks for using French :)) You should say more about "mace4" system you use : we have no idea on the category of software it belongs to for example ... – Jean Marie Sep 26 '17 at 20:08
• @JeanMarie I've added a postscriptum. – Kenny Lau Sep 26 '17 at 20:10

Given ideals $I$ and $J$ in a Lie algebra $\mathfrak{g}$ over a commutative ring $K$, the ideal product $[I,J]$ defined as the image of the composite $$I \otimes_K J \hookrightarrow \mathfrak{g} \otimes_K \mathfrak{g} \xrightarrow{[-, -]} \mathfrak{g}$$ is commutative, but not associative in general. It is easy to see why, because the ideal product is given by $$[I,J] \equiv \{ \Sigma_{i=0}^n [x_i, y_i] \mid \text{x_i \in I, y_i \in J, n \in \mathbb{N}} \},$$ and $[x,y]=-[y,x]$, but the bracket is not associative in general.