Let $L$ be an complete lattice. A lattice homomorphism $f: L\to L$ is said to be join-incomplete if there is an infinite set $S \subseteq L$ such that $f(\bigvee_L S) \neq \bigvee_L f(S).$

If $f:L\to L$ is join-incomplete, does there exist a lattice homomorphism $g:L\to L$ such that

  • there is $S\subseteq L$ with $g(S) \subseteq S$, and
  • $g(\bigvee_L S) > \bigvee_L g(S)$


  • 2
    $\begingroup$ You use $f$ both for the given homomorphism and for the one you are looking for. Please call one of them $g$. $\endgroup$ – Goldstern Jul 22 '15 at 10:51

The answer is no.

I will refer to these lattices (the labels are not meaningful - I took the images from a paper). Figure 1 depicts a lattice $A$ with the following properties:

  1. $A$ is meet semidistributive.
  2. Every injective endomorphism of $A$ is join-complete.
  3. If $\theta$ is any nonzero congruence on $A$, then $A/\theta$ is finite. Moreover, the natural map $A\to A/\theta$ is either join-complete or else the image $A/\theta$ is not meet semidistributive.
  4. The lattice $B$ in Figure 2 is a homomorphic image of $A$ (map $\overline{\overline{a}}$ to $x$, map each $b_{2n}$ to $y$, each $c_{2n+1}$ to $z$).

So take $L$ to be the ordinal sum of $A$ and $B$ ($B$ on top of $A$). Call the least and largest elements $0$ and $1$. This lattice is complete.

Define $f\colon L\to L$ by taking the (bottom) copy of $A$ in $L$ to the (top) copy of $B$, according to the homomorphism described in 4 above, and taking the top copy of $B$ to $1$. $f$ is a join-incomplete, since $f$ does not preserve the join of the central $\omega$-chain. (In fact, $f$ does not preserve the join of any $\omega$-chain.)

Claim. $L$ has no contracting join-incomplete endomorphism.

Sketch. Assume that $g\colon L\to L$ is a join-incomplete endomorphism. $g$ must be join-incomplete restricted to $A$, since $A$ contains all nonprincipal ideals of $L$ and their joins. Hence, by 2 and 3 above, $g$ is noninjective on $A$, and $g(A)$ is not meet semidistributive. This clearly implies $g(A)\not\subseteq A$, since $A$ is meet semidistributive, but in fact you can convince yourself of the stronger fact that $g(A)\subseteq B$. This shows that $g$ is not contracting, since all nonprincipal ideals lie in $A$ while their $g$-images lie in $B$. \\


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.