6

Let $(X_\alpha)_{\alpha <\kappa}$ be an inverse system of abelian groups.


If $\kappa = \omega$ (or by extension if $\kappa$ is of countable cofinality), then the Mittag-Leffler condition is a criterion guaranteeing that $\varprojlim^1_{\alpha <\kappa} X_\alpha = 0$. Namely, one says that $(X_n)_{n < \omega}$ is Mittag-Leffler if for every $n < \omega$, there exists $n' \geq n$ such that for every $n'' \geq n'$, one has $\operatorname{Im}(X_{n''} \to X_n) = \operatorname{Im}(X_{n'} \to X_n)$. In other words, $(X_n)_{n<\omega}$ is Mittag-Leffler if "$\operatorname{Im}(X_m \to X_n)$ stabilizes for $m \geq n$ sufficiently large". Two special cases are worth mentioning:

  1. If every map $X_m \to X_n$ is an epimorphism, then $(X_n)_{n<\omega}$ is Mittag-Leffler.

  2. If every map $X_m \to X_n$ with $m > n$ is zero, then $(X_n)_{n<\omega}$ is Mittag-Leffler.

And to reiterate: the point of the Mittag-Leffler condition is the following theorem of Grothendieck:

Theorem: If $(X_n)_{n<\omega}$ is Mittag-Leffler, then $\varprojlim^1_{n<\omega} X_n = 0$.


Now consider the case where $\kappa$ is an uncountable regular cardinal. In this case, the Mittag-Leffler condition as formulated above is not sufficient to guarantee that $\varprojlim^1_{\alpha<\kappa} X_\alpha = 0$. Moreover, in this case it's possible to have $\varprojlim^n_{\alpha<\kappa} (X_\alpha) \neq 0$ for $n \geq 2$. Nevertheless, an analog of (1) above is available. That is Neeman shows (Triangulated Categories, Appendix A) the following:

  1. (bis) If $X_\alpha \to \varprojlim_{\beta<\alpha} X_\beta$ is an epimorphism for each $\alpha <\kappa$, then $\varprojlim^n_{\alpha<\kappa} = 0$ for all $n \geq 1$.

(When $\alpha = \beta+1$ is a successor, this reduces to $X_{\beta+1} \to X_\beta$ being an epimorphism as in (1); when $\alpha$ is a limit, this is an additional condition.) In fact, Neeman simply calls the condition of (1 bis) "Mittag-Leffler". I'd resist this choice of terminology, because unlike in the countable case, (1 bis) does not contain as a special case the following analog of (2):

  1. (bis) if $X_\alpha \to X_\beta$ is zero for all $\kappa > \alpha > \beta$, then $\varprojlim^1_{\alpha <\kappa} X_\alpha = 0$.

Question: Let $\kappa$ be an uncountable regular cardinal, and let $(X_\alpha)_{\alpha < \kappa}$ be an inverse system of abelian groups (or even vector spaces). Is there a natural condition, analogous to the Mittag-Leffler condition when $\kappa = \omega$, which one can put on $(X_\alpha)_{\alpha<\kappa}$ ensuring that $\varprojlim^1_{\alpha < \kappa} X_\alpha = 0$?

In particular, is there such a condition which contains both (1 bis) and (2 bis) as special cases?

Tim Campion
  • 60,951
  • 3
    Being pro-isomorphic to an inverse system of abelian groups whose bonding morphisms are surjections. This condition is called Mittag-Leffler, AFAIK, and ensures that all derived limits vanish. In the countable case it is equivalent to what yo said. – Fernando Muro Nov 25 '20 at 22:36
  • 2
    @FernandoMuro Thanks! After consulting SGA 4.8.12 and working things through, it seems to me that being pro-isomorphic to a strict pro-object is equivalent to the Mittag-Leffler condition as I stated it, but for any filtered indexing category. On the other hand, is it really the case that the derived limits of a strict pro-object in $Ab$ always vanish? I don't think this is true in $Set$ unless you assume countable cofinality, as witnessed by the existence of Aronszajn trees. – Tim Campion Nov 26 '20 at 01:46
  • 2
    My bad, you need the strong Mittag-Leffler condition from Edwards, D. A., and H. M. Hastings. Czech and Steenrod Homotopy Theories with Applications to Geometric Topology. Lecture Notes in Math. 542. Springer Verlag, 1976. See definition 4.8.3. – Fernando Muro Nov 26 '20 at 15:03
  • @TimCampion Not to hijack, but what do Aronszajn trees have to do with derived limits? – Kevin Carlson Nov 28 '20 at 23:46
  • 1
    @KevinArlin Maybe the connection is not as tight as I have it in my head, but let $\kappa$ be the height of a tree $T$. There is a functor $F: \kappa^{op} \to Set$ which sends $\alpha$ to the set of branches of height $\alpha$. Then $\varprojlim F$ is the set of cofinal branches of $T$. If $T$ is $\kappa$-Aronszajn, then $\varprojlim F = \emptyset$, but $F(\alpha) \neq \emptyset$ for each $\alpha$. I had convinced myself that the transition maps for $F$ were moreover surjective, but now I realize that is not the case. – Tim Campion Nov 29 '20 at 14:35
  • 1
    However, at least in the standard constructions, each $x \in F(\alpha)$ admits lifts to $y \in F(\beta)$ for arbitrarily large $\beta$, at least if $\alpha$ is a successor ordinal. Anyway, if no $\kappa$-Aronszajn trees exist (i.e. $\kappa$ has the tree property), it becomes much easier to perform the inductive step of various constructions at $\kappa$. But since there are always smaller cardinals (such as $\omega_1$) without the tree property, I've never been able to leverage this properly. – Tim Campion Nov 29 '20 at 14:36
  • 1
    For instance I once spent way too much time trying to use a weakly compact cardinal (=inaccessible cardinal with the tree property) to address this question. – Tim Campion Nov 29 '20 at 14:36
  • 1
    I would like to clarify that the previous reference to SGA is SGA4, Exposé I. Préfaisceaux, Section 8.12. –  Feb 10 '21 at 18:50

0 Answers0