# Example of a naturally ordered semiring which is not an m-semiring

A semiring is an algebraic structure equipped with a commutative and associative addition operator $+$ with a neutral element $0$, and an associative multiplication operator $\times$ with a neutral element $1$, such that $0$ is absorptive for $\times$ and that $\times$ distributes over $+$. We can define a preorder on a semiring $K$ by $x \leq y$ iff $\exists z$ such that $x + z = y$. We call the semiring $K$ naturally ordered if the relation $\leq$ is a partial order, i.e., if it is antisymmetric.

We further say that the semiring $K$ is an m-semiring if the following holds: for any $x, y \in K$, the set $\{z \in K \mid x \leq y + z\}$ has a smallest element for $\leq$.

What is an example of a naturally ordered semiring which is not an m-semiring?

Such an example is claimed by Example 4 of this paper, page 6, but (as confirmed with one author of this paper) there is a problem with the example. (In "Clearly, this is not bounded below", the authors mean the usual order on $\mathbb{R}$, but the natural order $\leq$ is the reverse of the usual order.)

(Edit: note that this question can be posed just for monoids already, as it is independent from the $\times$ operator; but requiring that a suitable $\times$ operator can be defined makes it slightly harder to build a counterexample.)

Here is an example:

Let $K$ be $\mathbb{N} \cup \{42'\}$ where $42'$ is an element distinct from each element of $\mathbb{N}$. For any $x$ in $K$ we will define $h(x)$ by $x$ if $x \in \mathbb{N}$ and $42$ if $x = 42'$; we denote $+$ (resp $\times$) the usual addition (resp., multiplication) in $\mathbb{N}$. Now, we define the commutative binary relation $\oplus$ on $K$ as follows:

• $0 \oplus a = a = a \oplus 0$ for every $a \in K$
• $a \oplus b = h(a) + h(b)$ for every $a,b \neq 0$

We can easily see that $(K, \oplus, 0)$ is a commutative monoid.

In what follows, we will ensure that $h$ is a homomorphism of semirings from $K$ to $\mathbb{N}$.

The preorder on this monoid defined by $x\leq y$ iff $\exists z ~ x \oplus z=y$ is then:

• for $a,b \in \mathbb{N}$ then $a\leq b$ iff $a$ is less than or equal to $b$ in the usual sense
• $0 \leq 42'$
• $42'$ is incomparable with $1$, with $2$, ..., with $42$
• $42'\leq 42'$
• $42' \leq 43$, $42'\leq 44$, ...

This preorder is antisymmetric, so that $(K, \oplus, 0)$ is naturally ordered.

We define the commutative binary relation $\otimes$ on $K$ as follows:

• $1 \otimes a = a \otimes 1 = a$ for every $a \in K$
• $a \otimes b = h(a) \times h(b)$ for every $a,b \neq 1$

Then $(K, \oplus, \otimes, 0, 1)$ is naturally ordered (as we just proved), and we must now prove that it is actually a semiring.

Indeed associativity of $\otimes$, $1$ being neutral for $\otimes$, $0$ an annihilator for $\otimes$ are all easy to check, so let us look at distributivity: is $a \otimes ( b \oplus c )$ equal to $a \otimes b \oplus a \otimes c$?

First, note that since $(\mathbb{N}, +, \times, 0, 1)$ is a semiring, $h(a) \times ( h(b) + h(c) ) = h(a) \times h(b) + h(a) \times h(c)$. And $h$ has been defined so that $h(x\oplus y)=h(x)+h(y)$ and $h(x\otimes y)=h(x)\times h(y)$, so we have: $h(a \otimes ( b \oplus c )) = h(a \otimes b \oplus a \otimes c)$. This means that if distributivity were not to hold, it would be because we obtained $42$ on one side of the equation and $42'$ on the other. But, with a single operation, $42'$ can only be obtained as a product on the left-hand side with $42' \otimes 1$ or $1 \otimes 42'$. We thus distinguish the cases:

• if $a = 1$, then distributivity holds trivially
• if $b \oplus c=1$, then $\{b,c\}=\{0,1\}$ and distributivity holds

Similarly, $42'$ can only be obtained as a sum on the right-hand side with $42' \oplus 0$ or $0 \oplus 42'$. You can do a case-by-case analysis and check that distributivity holds.

However, $(K, \oplus, \otimes, 0, 1)$ does not satisfy the additional m-semiring property that for every $x$ and $y$, the set $S(x,y) = \{z \in K \text{~s.t~} x \leq y \oplus z\}$ has a smallest element. Indeed we have $42 \in S(142,100)$ and $42' \in S(142,100)$ but $42$ and $42'$ are incomparable. And no element smaller than both (it would have to be $0$ by the definition of $\leq$) is in the set.

Remark: it should even possible to make the example finite by cutting at 142 and defining, e.g., $a {\oplus'} b = \max(a \oplus b, 142)$.