# On the definition of locally compact for non-Hausdorff spaces

It seems that there are different conventions in the literature as to what is a locally compact space (when the space is not supposed Hausdorff).

The two main non equivalent definitions I've seen are :

• (LC1) every point has a compact neighborhood

• (LC2) every neighborhood of any point contains a compact neighborhood of the point.

I am wondering if there is a reason as to why one might prefer one definition to another. More precisely, in practice, what definition is really useful (yields interesting results), or are they are both important in their own right ? In that case why not give a different name to those definitions ?

Naively, when one looks at the definition of locally connected space, one does not use the version (LCn1)"every point has a connected neighborhood", but always (LCn2) "every neighborhood of any point contains a connected neighborhood of the point". Is there a deep reason as to why LCn1 is never considered, but LC1 is ?

I am aware that for Hausdorff space, LC1 and LC2 are equivalent, and since LC1 is easier to check, one might prefer that as a definition, but this argument is unconvincing if LC2 is actually more useful for non-Hausdorff space.

• LCn1 seems clearly "wrong" since it is satisfied by every connected space, even those such as comb space or the topologist's sine curve which violate the whole idea of local connectedness. – Nate Eldredge Feb 25 '17 at 7:26
• You are probably aware that in most french books (following Bourbaki, and including Wikipedia), a locally compact space is Hausdorff by definition -- with good reasons in my view. – abx Feb 25 '17 at 10:14
• @abx : what are those good reasons ? – Phil-W Feb 25 '17 at 10:41
• The reason why LC2 and LCn2 are the "right" definitions is because they actually capture what "locally X" means. We don't care about "a" neighbourhood. Things hold locally if they hold on arbitrary small neighbourhoods. That LC1 is equivalent to LC2 in Hausdorff spaces is simply an "accident" that is rooted in the non-obvious interplay between compactness, separation axioms ($T_2+compact\implies T_3$) and a similarly accidental side effect of $T_3$ being equivalent to "locally closed". – Johannes Hahn Feb 25 '17 at 11:03
• @Phil-W: compare with the 3 possible definitions given by the english Wikipedia page, which concludes by "In almost all applications, locally compact spaces are indeed also Hausdorff". – abx Feb 25 '17 at 11:11

To me, the second definition of local compactness is much to be preferred for the simple reason that such locally compact spaces $X$ are exponentiable in $Top$, meaning that $X \times -: Top \to Top$ has a right adjoint $(-)^X: Top \to Top$ (even without the Hausdorff condition), and all this implies (such as $X \times -$ preserving coequalizers). In fact the necessary and sufficient condition for exponentiability, called core compactness, is only a mild generalization of local compactness (and equivalent to it under the Hausdorff assumption).
• Similarly for local connectedness of $X$: one may argue that rather than being "a thing in itself" it is "needed" for the constant sheaf functor $\mathrm{Sets}\to\mathrm{Sheaves}(X)$ to have a left adjoint. And for that, LCn1 is useless, one really needs LCn2. – მამუკა ჯიბლაძე Feb 25 '17 at 9:46