3
$\begingroup$

$\let\op=\operatorname$In $\op{Set}$, we have an $(\op{Epi},\op{Mono})$ orthogonal factorization system. Strikingly, if we reverse the roles, we get the no-less-important $(\op{Mono},\op{Epi})$ weak factorization system.

In the $\infty$-category $\op{Spaces}$ of spaces, the most direct analog of the $(\op{Epi},\op{Mono})$ orthogonal factorization system on $\op{Set}$ is the $(\text{Effective Epi}, \op{Mono})$ orthogonal factorization system, but this is just the $(-1)$th in a whole tower: for each $n \in \mathbb Z_{\geq -2}$, we have an $(\text{$n$-connected}, \text{$n$-truncated})$ factorization system [1].

It seems that, just as in the analogous case in $\op{Set}$, one can take the left half of each these orthogonal factorization systems, and view it as the right half of a weak factorization system $(\mathcal L_n, \text{$n$-connected})$ [2]. To see this, one shows that the $n$-connected maps are precisely the maps which are weakly right orthogonal to the maps $\{S^k \to 1 \mid -1 \leq k \leq n\}$, and applies the small object argument to obtain factorizations.

In $\op{Set}$, we have the cute fact that the resulting weak factorization system $(\op{Mono},\op{Epi})$ is just the original orthogonal factorization system $(\op{Epi},\op{Mono})$ with the left and right classes swapped. This is not the case in $\op{Spaces}$, even when $n=-1$: a map $A \xrightarrow i B$ of spaces is weakly left orthogonal to the effective epimorphisms if and only if it is a coproduct inclusion $A \to A \amalg S$ where $S$ is discrete; this is more restrictive than being a monomorphism [3]. I don't know how to characterize the left class $\mathcal L_n$ for $n\geq 0$ as cleanly. In fact, unlike the case in $\op{Set}$, I don't think we have either containment $\mathcal L_n \subseteq \text{$n$-truncated}$ or $\text{$n$-truncated} \subseteq \mathcal L_n$ in general. This leads to my

Questions: Let $n \in \mathbb Z_{\geq -2}$.

  1. Is there a good characterization of the class of maps $\mathcal L_n$, i.e., the maps of spaces which are weakly left orthogonal to the $n$-connected maps?

  2. What would be a good name for the maps of $\mathcal L_n$?

Note that by the small object argument, the maps of $\mathcal L_n$ are precisely the retracts of transfinite composites of cobase-changes of coproducts of the maps $\{S^k \to 1 \mid -1 \leq k \leq n\}$. So in some sense this is a quite explicit class of maps. By "characterization" I suppose I mean something which can be "checked directly" without having to find all the data of a construction of this form.


[1] Here a map is said to be $n$-truncated or $n$-connected if its fibers are all so. This convention is off by one from the most classical convention.

[2] Some care should be taken with the definition of a weak factorization system $\infty$-categorically: say that a morphism $A \xrightarrow i B$ is weakly orthogonal to a morphism $X \xrightarrow p Y$ if the map $\op{Hom}(B,X) \to \op{Hom}(B,Y) \times_{\op{Hom}(A,Y)} \op{Hom}(A,X)$ is an effective epimorphism. Spelled out, this says that if we have a commutative square—i.e., morphisms $A \xrightarrow u X$, $B \xrightarrow v Y$ along with a homotopy $\gamma: pu \sim vi$, then there exists a lift, i.e., $B \xrightarrow w X$ and homotopies $\alpha: wi \sim u$, $\beta: pw \sim v$ and (here's the only subtle part) a homotopy of homotopies from the composite $\beta \ast \alpha$ to $\gamma$. Then a weak factorization system is, as usual, a pair of classes of morphisms $(\mathcal L, \mathcal R)$ which are complements to each other with respect to weak orthogonality, such that every morphism admits a factorization as a morphism in $\mathcal L$ followed by a morphism in $\mathcal R$.

[3] Recall that a monomorphism of spaces is a coproduct inclusion $A \to A \amalg S$ where $S$ may be an arbitrary space.

$\endgroup$
8
  • 1
    $\begingroup$ Good grief. $\mathcal L_n$ is just the class of maps which are retracts of relative cell complexes of dimension $\leq n+1$. I'd probably call these "retracts of $n+1$-skeletal maps" or something. $\endgroup$ – Tim Campion Feb 15 '20 at 18:25
  • 2
    $\begingroup$ Is this your answer to your own question? I have trouble parsing the meaning of "good grief" here …. $\endgroup$ – LSpice Feb 15 '20 at 22:17
  • 1
    $\begingroup$ @LSpice. Yeah, I think I answered my own question... Maybe I should delete it, but I spent so much time typing it out... And now you've invested time in prettying it up, too! (Thanks, btw) Actually, I suppose it's still an interesting question how to characterize these maps, but $\mathcal L_n$ is so familiar that if a good characterization were known it would have been known in the '50's, and I don't think such a thing is known... Although maybe in the simply-connected case you can say something about dimension vs. homology dimension or something... $\endgroup$ – Tim Campion Feb 15 '20 at 22:18
  • 1
    $\begingroup$ To clarify, "Good grief" is my sheepish exclamation when I realized that I found a ridiculous roundabout way to ask for a characterization of the retracts of relative $n+1$-dimensional complexes without recognizing that that's what I was asking! $\endgroup$ – Tim Campion Feb 15 '20 at 22:28
  • 3
    $\begingroup$ Yes, there is a characterization in terms of (co)homology in most cases. Of course for a map in this class the relative homology and cohomology vanishes in degrees above $n+1$. Conversely, if relative $H^j$ vanishes for all $j\ge n+1$, for all coefficient systems on the codomain, then the map is in that class. This holds for all $n>2$, at least. $\endgroup$ – Tom Goodwillie Feb 15 '20 at 23:46
1
$\begingroup$

This question was answered in the comments $\mathcal L_n$ comprises those maps which are retracts of relative $\leq n+1$-dimensional relative cell complexes.

Tom Goodwillie explains in the comments a cohomological characterization of $\mathcal L_n$ for sufficiently large $n$.

$\endgroup$

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.