I am looking at two seemingly same (but not quite) Riesz representation theorems:

(Wikipedia) Let $X$ be a locally compact Hausdorff space. Let $C_c(X)$ be the space of compactly supported continuous functions. Then a positive continuous linear functional $\Lambda : C_c(X) \to \mathbb R$ can be represented as an integral that is, $\Lambda$ corresponds to a (unique) measure $\mu$ (regular and Borel) such that $\Lambda (f) = \int_X f d \mu$ for all $f \in C_c(X)$.

(My lecture notes) Let $X$ be a locally compact, $\sigma$-compact metric space. Then a positive linear functional $\Lambda : C_c(X) \to \mathbb R$ can be represented by a (unique) measure $\mu$ (locally finite and positive) such that $\Lambda (f) = \int_X f d \mu$ for all $f \in C_c(X)$.

I can see that metric implies Hausdorff. But I find it impossible to remember all the conditions, namely, those about $X$ (just locally compact or $\sigma$-compact as well?) and those about the measure (surely every measure is positive by definition but requiring it to be Borel and regular seems to be stronger than locally finite).

My question(s): Which of these two versions is more general? Or are they the same? And how can I tell them apart, that is, remember which version comes with which assumptions on the space and the measure?

Thanks for your help.

  • 2
    $\begingroup$ The assumptions of your lecture notes are equivalent to local compactness and second countability of $X$ (Hausdorff is self-understood anyway), so would simplify the hypotheses to remember a bit. Certainly this assumption is far more restrictive than Wikipedia's and it makes the proof quite a bit less technical (hence simpler). The unpleasant stuff and a good discussion of the proof of the Riesz-Markov theorem you might run into without some simplifying assumptions on $X$ beyond local compactness is given in Arveson's notes. $\endgroup$
    – t.b.
    Aug 16, 2012 at 16:37
  • $\begingroup$ @t.b. If you copy paste this comment into an answer I'd upvote and accept it. $\endgroup$ Aug 16, 2012 at 16:45

2 Answers 2


I think the first version is more general. It clearly has weaker assumptions, as you noted. As for the conclusions, they both yield measures that correspond to the functional in exactly the same way, and are unique, which are described as following:

  1. Regular and Borel.
  2. Locally finite and positive.

In the first case, I think we can safely assume that "positive" is implicit (because it almost always is, if not stated otherwise, and because otherwise the resulting functional would not be positive!). Similarly, if the measure was not locally finite, then I believe that the resulting „functional” would not be bounded (or even finite), so we can also assume that implicitly, so the first version does not provide anything more with the assumptions of the second one.

As for the second case, I believe that there is also implicit assumptions of regularity and Borelness, because otherwise the measure would be unlikely to be unique, because by the first version we already have a Borel measure, but a Borel measure can always (except for trivial cases, e.g. when every set is Borel) be extended to a larger $\sigma$-field, defying uniqueness (remember that the integral of a continuous functions depends only on the Borel part of the measure), and any finite Borel measure on a metric space is already regular, and this should extend to $\sigma$-finite in the obvious way. (Note that locally finite measure on a $\sigma$-compact space is $\sigma$-finite.)

Summing it all up, unless I made some large mistakes:

  1. The first theorem is more general, because it applies to more cases.
  2. With the assumptions of the second theorem, they are equally strong.

To my taste, adjectives in a situation like this should be re-read in terms of what they accomplish (or don't) in terms of the goal at hand.

That is, locally compact, Hausdorff assures a good supply of continuous, compactly-supported functions, by Urysohn's lemma. (Otherwise, there might be too few continuous functions to approximate characteristic functions of compact sets.)

The sigma-compactness is (in my world) completely reasonable, and avoids various "pathologies" both in terms of outer regularity and measures on product spaces. It is not obvious in advance that it is needed, but... it is. (The interaction of these adjectives in metric spaces is a source of many homework exercises.)

And the outer and inner regularity, and "Borel"-ness, are certainly desirable, and unsurprising in a context that the purported measure is strongly related to continuous functions (assuming that the space is nice enough so there aren't pathologies!)

  • 1
    $\begingroup$ I think it is true that $\sigma$-compactness avoids many pathologies. However I don't know if I can safely assume $\sigma$-compactness everywhere in integration theory on locally compact Hausdorff spaces. In other words, is there no meirt in developing the theory without $\sigma$-compactness condition? $\endgroup$ Aug 16, 2012 at 16:54
  • 1
    $\begingroup$ In my own experience, I've never found reason to worry about loss of sigma-compactness: real Lie groups and p-adic groups occuring in (my) practice are sigm-compact, as are adele groups. Arithmetic quotients are. I've never seen a tangible reason compelling concern about not-sigma-compact spaces on which to do genuine measure theory. A different complication, apparently of some use, as in Feynman and other delicate "integrals", is loss of local compactness entirely... and a restricted notion of "measure", but I don't know anything about that. $\endgroup$ Aug 16, 2012 at 17:14

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .