= Solution
Use the natural <partial order>
$$
x\le y\quad\Longleftrightarrow\quad x+y=y.
$$
The paper's shorthand $x\le x+y$ specifies the join order: more formally, $x\le z$ means $z=x+y$ for some $y$, which is equivalent to the displayed condition by <idempotence>. Associativity, commutativity and idempotence make addition the <join> operation. Distributivity makes multiplication monotone, while absorption gives $xy\le x$ and $xy\le y$.
Let $g_1,\ldots,g_r$ generate the <incline>. Distributivity writes every element as a nonempty finite join of monomials
$$
m(v)=g_1^{v_1}\cdots g_r^{v_r},\qquad v\in\mathbb N^r\setminus\{0\}.
$$
There is no assumed multiplicative identity: an exponent zero means omit that factor, and at least one factor remains. If $u\le v$ coordinatewise, absorption gives $m(v)\le m(u)$. If the two vectors are equal the conclusion is immediate; otherwise multiply $m(u)$ by the additional factors and apply absorption. For a polynomial with finite exponent support $A$, put $U_A=\{v:(\exists u\in A)\ u\le v\}$, its upward closure. If $U_A\supseteq U_B$, every monomial in $B$ is below some monomial in $A$, and consequently the join represented by $B$ is below the join represented by $A$.
We prove the <reverse-inclusion well-quasi-ordering of monomial ideals>, often called <Maclagan's theorem>, in the equivalent form that upward-closed subsets of $\mathbb N^r$ are well-quasi-ordered by reverse inclusion. Include the empty subset. For $r=1$, these are the tails $[n,\infty)$ and the empty set, corresponding to $\mathbb N\cup\{\infty\}$ in its usual well-order.
Induct on $r$. For an upward-closed $U\subseteq\mathbb N^r$, its slices $U_k=\{v\in\mathbb N^{r-1}:(v,k)\in U\}$ increase with $k$. Their union is upward closed and has finitely many minimal elements by the <Dickson lemma>; all those elements have appeared by some finite slice. Thus the slices eventually stabilize. Encode $U$ by the finite word of its pre-stabilization slices and its final stable slice. By the induction hypothesis, the slice order given by reverse inclusion is a <well-quasi-ordering>. The <Higman lemma> and <finite product closure of well-quasi-orderings> make these word-and-tail encodings well-quasi-ordered.
For a good pair of encodings $U,V$, an increasing subsequence map $f$ satisfies $U_k\supseteq V_{f(k)}$ at every pre-stabilization position of $U$, and the final slice of $U$ contains the final slice of $V$. Since $f(k)\ge k$ and the $V$ slices increase, $U_k\supseteq V_k$ before stabilization. Afterwards the final-slice comparison also gives $U_k\supseteq V_k$. Hence $U\supseteq V$, proving the induction. The use of <Higman lemma> is essential: finite-subset reverse domination does not preserve an arbitrary well-quasi-order, as the <Rado order> shows.
Given any infinite sequence of incline elements, choose a finite monomial support $A_i$ for each. The result just proved gives $i<j$ with $U_{A_i}\supseteq U_{A_j}$, hence $x_i\ge x_j$. \b[Every finitely generated incline is therefore well-quasi-ordered by its reversed natural order.]
Back to article page
