Let $\mathcal L$ be a bounded modular lattice. Suppose also that $\mathcal L$ is complete and upper-continuous (i.e. for any directed subset $\{x_i:i\in I\}$ of $\mathcal L$ and any $x\in \mathcal L$, $x\wedge\bigvee_{i\in I}x_i=\bigvee_{i\in I}(x\wedge x_i)$).

Is there any known sufficient condition for $\mathcal L$ to admit an order-preserving function $f\colon\mathcal L\to [0,1]$, such that:

  1. $f(x)=0$ if and only if $x=0$;
  2. if $a\leq b$ in $\mathcal L$, $f(a)=f(b)$ implies $a=b$;
  3. $f$ commutes with (possibly infinite) joins of directed families.

Is there any known structure theorem for lattices admitting such a map?

Of course, the obvious case is when $\mathcal L$ is an atom, in which case $\mathcal L\cong\{0<1\}$. I am much more interested in situations where the image of $f$ is a dense subset of $[0,1]$.

Motivating Example. Let $D$ be an Archimedean valuation domain, and denote by $v:D^*\to \mathbb R_{\geq0}$ the valuation on $D$. It is possible to define a function $L_v:Mod(D)\to \mathbb R_{\geq0}\cup\{\infty\}$ via the following steps:

  1. given an ideal $I$ of $R$, let $v(I):=\inf\{v(d):d\in I\}$;
  2. given a finitely generated right $D$-module $F$, let $\{x_1,\dots,x_n\}$ be a generating set with the minimum possible cardinality. Define inductively
    • $F_0=0$;
    • $F_1=x_1D$;
    • if $F_k$ is defined for some $k<n$, then let $F_{k+1}=F_k+x_{k+1}D$;

in particular $F_n=F$. Let also $B_j=Ann(x_j+F_{j-1})=\{d\in D: x_jd\in F_{j-1}\}$. We let $$L_v(F)=v(B_1\cdots B_n)\,;$$ 3. for a given right $D$-module $M$, we let $L_v(M)=\sup\{L_v(F):F\leq M\text{ finitely generated}\}$.

One can prove that $L_v$ is additive on short exact sequences (i.e. $L_v(B)=L_v(A)+L_v(C)$ if $0\to A\to B\to C\to 0$ is exact) and it is upper-continuous (i.e. $L_v(M)=sup_iL_v(M_i)$ when $M=\bigcup_iM_i$ with $\{M_i\}$ a directed set of submodules of $M$).

Using these properties, one can see that $$Ker(L_v):=\{M\in Mod(D):L_v(M)=0\}$$ is a hereditary torsion class, and so we can define a new category $$\mathfrak A:=Mod(D)/Ker(L_v)$$ as the Gabriel quotient of $Mod(D)$ over $Ker(L_v)$. Let $Q_v\colon Mod(D)\to \frak A$ be the quotient functor. The function $L_v$ induces a new function $$\bar{L_v}\colon \mathfrak A\to \mathbb R_{\geq0}\cup \{\infty\}$$ just letting $\bar{L_v}(X)$ be $L_v(M)$ for any $M\in Mod(D)$ such that $Q_v(M)=X$. This new function has the same additivity and upper-continuity properties of $L_v$ and, furthermore, it has the property that $\bar L_v(X)=0$ iff $X=0$.

All the objects $X\in \mathfrak A$ such that $\bar {L_v}(X)<\infty$ do have a lattice of subobjects with the properties described in the question (the wanted function to $[0,1]$ is just a rescaling of the values of $\bar L_v$ on subobjects).

  • $\begingroup$ I don't think I understand 3. Is join of f(a) and f(b) Max? In which case I do not see 2 holding for interesting examples. Gerhard "Not Sure Where Order Is" Paseman, 2017.07.02. $\endgroup$ – Gerhard Paseman Jul 2 '17 at 18:41
  • 1
    $\begingroup$ Well yes, if $a\leq b$, then $a\vee b=b$. Since $[0,1]$ is totally ordered any finite join in it is a maximum. I'll add an interesting (at least to me) example coming from commutative algebra in a moment. $\endgroup$ – Simone Virili Jul 2 '17 at 18:44
  • 3
    $\begingroup$ Note that one obvious necessary condition is that there be no embedding of $(\omega_1, <)$ into $\mathcal{L}$, since there is no embedding of $(\omega_1, <)$ into $[0, 1]$. And in fact this follows from (2) alone. $\endgroup$ – Noah Schweber Jul 2 '17 at 18:51
  • $\begingroup$ Yes, you are right of course, another necessary condition is that there is no pair $0\neq x\leq y$ in $\mathcal L$ such that the lattice $[x,y]$ (with the induced order) has infinite Goldie dimension (=uniform dimension). I can find some obstructions of this kind, but I cannot see non-obvious sufficient conditions for the moment $\endgroup$ – Simone Virili Jul 2 '17 at 19:13
  • 1
    $\begingroup$ Sorry, I forgot to restrict 3 to "directed sets"... I have edited the question $\endgroup$ – Simone Virili Jul 2 '17 at 19:23

Mathematical economists have been interested in special cases of the following general question, which they call "existence of utility representations."

$\quad$ Given $(X, \precsim)$ an ordered set, when does there exist a strictly increasing function $f: X \to \mathbb R$?

A classical result is that for $\precsim$ a complete preorder (total, transitive; not necessarily antisymmetric), it is necessary and sufficient that $X$ has a countable order-dense subset, and furthermore $f$ can be taken continuous with respect to the interval topology on $X$ and standard topology on $\mathbb R$. The first statement of this was Debreu (1954) (N.B. the proof in that paper has a nontrivial error.)

Economists tend to be interested in how structural axioms on the preorder translate to canonical functional forms for $f$ (and how to interpret these axioms), but it looks like others have continued this line of research in other directions.

Here's a recent paper that looks like it gives a general answer to your question (e.g. Theorem 3.1).

$\quad$ Continuous Utility Representation Theorems in Arbitrary Concrete Categories

$\quad$ Bosi, G. & Herden, G. Appl Categor Struct (2008) 16: 629. doi:10.1007/s10485-007-9097-0

I don't know if the characterization will be useful for your purposes, and don't know the non-economics literature well, but it seems to exist.


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.