On the geometry at infinity of manifolds with linear volume growth and nonnegative Ricci curvature
Abstract.
We prove that an open noncollapsed manifold with nonnegative Ricci curvature and linear volume growth always splits off a line at infinity. This completes the final step to prove the existence of isoperimetric sets for given large volumes in the above setting. We also find that under our assumptions, the diameters of the level sets of any Busemann function are uniformly bounded as opposed to a classical result stating that they can have sublinear growth when the end is collapsing. Moreover, some equivalent characterizations of linear volume growth are given. Finally, we construct an example to show that for manifolds in our setting, although their limit spaces at infinity are always cylinders, the cross sections can be nonhomeomorphic.
2020 Mathematics Subject Classification:
Primary 53C21, 53C23.key words: Ricci curvature, splitting, linear volume growth, isoperimetric sets
1. Introduction
Recently there has been an interest in studying the structure at infinity on manifolds with nonnegative Ricci curvature, meaning the pointed Gromov-Hausdorff (pGH in short) limits of the form for an open manifold and a sequence of points diverging to infinity (see Definition 1.1). The motivations come from several different fields. For instance, one motivation is to understand positive scalar curvature coupled with nonnegative Ricci curvature at large scales [Zhu_Geometryofpsc, WZZZ_PSC_RLS, ZhuZhu23, zhu2023twodimension] and another is to examine the existence of isoperimetric sets in a noncompact manifold [isoregion, antonelli2023isoperimetric] with nonnegative Ricci curvature and Euclidean or linear volume growth. In both cases, a common interest is to see if any limit space at infinity (see Definition 1.1) splits off many lines, because splitting at infinity simplifies the structure of the limit spaces under our consideration, making it possible to apply some powerful techniques.
In light of this demand, we take the very first step to study in particular the line splitting at infinity for noncollapsed manifolds with linear volume growth. We introduce these notions as follows.
Definition 1.1.
Let be an open manifold. We say that is noncollapsed if
(1.1) |
We say that has linear volume growth if for some point (hence all points) ,
(1.2) |
We say that a sequence of points diverges to infinity (or it is a diverging sequence) if for any fixed point , as . We say a Ricci limit space is a limit space at infinity of if it is obtained as a pGH limit of sequence of the form for a sequence diverging to infinity. We say a metric space is a cylinder if it is a metric product in the form of with being a compact metric space.
Examples of manifolds with linear volume growth that are not necessarily a product of and a compact manifold include -manifolds with nonnegative Ricci curvature and uniformly positive scalar curvature as confirmed by [OLS23, MW_geometryofpsc].
Our first result is about splitting at infinity with the noncollapsed assumption.
Theorem 1.2.
Let be an open -manifold with . Suppose is noncollapsed (1.1), and has linear volume growth with a volume ratio upper bound as in (1.2). Then for any sequence diverging to infinity, all possible pGH limit spaces of the sequence splits as , where is a compact space with . Moreover, all possible have the same volume.
Remark 1.3.
There are extensive studies of manifolds with linear volume growth by Sormani [SormaniMiniVol, SormaniSublinear] without the noncollapsed assumption. The motivation for considering the noncollapsed assumption (1.1) is also natural. In her pioneering work, Sormani [SormaniMiniVol]*Example 26 constructed a -manifold with nonnegative Ricci curvature and linear volume growth, on which there is a Busemann function whose level sets have logarithmic diameter growth and finite codimension volume. In this example, the limit space at infinity is , it does split off an -factor but the other factor is which is noncompact as opposed to Theorem 1.2.
Intuitively, for the diameter to tend to infinity while the volume stays bounded, there should be some βdirectionβ that is shrinking, leading to a collapsing end. Therefore it is believable that can prevent the diameter of the level sets of a Busemann function from tending to infinity. Indeed, it will be shown in proposition 3.5.
Remark 1.4.
The cross sections in the limit space at infinity can indeed be nonisometric, as shown by an example of Sormani [SormaniMiniVol]*Example 27. In this example the limit space splits as but for different diverging sequence of points the -factor can carry two different metrics, and both metrics give the same volume to . We will see in Theorem 5.2 that such an example with nonisometric can have dimension and in Theorem 5.1 that the cross sections can be nonheomeomorphic.
The following example of Kasue-Washio [KasueExample]*p.913-914 helps illustrate why the above Theorem is not obvious, Specifically, it shows that with nonnegative Ricci curvature, not every diverging sequence of points produces a limit space at infinity that splits. So at least the linear volume growth condition plays a key role. The example says there is a Riemannian metric on of the form with nonnegative Ricci curvature. From its expression is translation invariant along direction, but does not split. When taking a sequence of points diverging to infinity along direction we get a limit space isometric to , the manifold we start with, which does not split. See also some discussions in [isoregion]*Section 1.4. In this example has Euclidean volume growth instead of linear volume growth. In the seemingly simple and restrictive case where has nonnegative Ricci curvature and linear volume growth, to the authorβs best knowledge the following splitting problem remains open, i.e., when the noncollapsed assumption in Theorem 1.2 is removed.
Question 1.5.
Given with and linear volume growth (1.2), is it true that for every sequence diverging to infinity, the pGH limit of splits off a line?
Unlike the rather complicated situation for Ricci curvature, in the case of manifolds with nonnegative sectional curvature, the picture for splitting is much simpler. In fact, the following two statements holds.
-
(1)
For a manifold with nonnegative sectional curvature, any limit space at infinity always splits off a line without further assumptions [antonelli2023isoperimetric]*Lemma 2.29, [zhu2023twodimension]*Corollary 4.3.
-
(2)
For a manifold with nonnegative sectional curvature, itself splits if and only if its asymptotic cone splits. See for example a proof in [isoregion]*Theorem 4.6 and the references therein.
The very reason is the monotonicity of angles, which allows us to pass large scale information to smaller scales or vice versa. It is then pointed out by the author [zhu2023twodimension]*Lemma 4.2 that the above two statements can be unified.
The above discussion gives a nice relation between asymptotic cones and limit spaces at infinity. In general we do not have such a relation for nonnegative Ricci curvature. A counterexample is again the one of Kasue-Washioβs we discussed in the introduction [KasueExample], [isoregion]*section 1.4. In this example, has a unique asymptotic cone and it splits, but itself does not. In the very special case of linear volume growth and being noncollapsed, Theorem 2.4 provides some similar relation.
Nevertheless the idea of studying the relation between asymptotic cones and limit spaces at infinity still leads us to the following interesting observations or analogs. Note that a manifold with nonegative Ricci curvature and Euclidean volume growth is noncollapsed by Bishop-Gromov inequality.
-
β’
For a manifold with Euclidean (maximal) volume growth, its asymptotic cones are metric cones and the cross sections of the asymptotic cones always have the same volume [Cheeger-Colding97I].
-
β’
For a noncollapsed manifold with linear (minimal) volume growth, its limit at infinity are cylinders and the cross sections of the cylinders always have the same volume.
-
β’
There exists a -manifold with Euclidean (maximal) volume growth and its asymptotic cones can be a cone over either or , [CN11]*Example II.
-
β’
There exists a noncollapsed -manifold with linear (minimal) volume growth and its limit spaces at infinity can be a product of with either or .
After a in depth analysis, we can establish the following list of equivalence conditions, giving a rather complete description of the geometry of linear volume growth. Also, it partially extends [antonelli2023isoperimetric]*Corollary 3.11 to nonnegative Ricci curvature and partially answers the questions in [antonelli2023isoperimetric]*Problem 1.2. However, there are remaining questions. For example we cannot show that if for some divergent sequence of points the corresponding limit space at infinity splits then for arbitrary divergent sequence of points the corresponding limit space at infinity also splits. This implication, if true, will link the isometric profile bound to the linear volume growth. See Remark 4.4 for details.
Theorem 1.6.
Let be an open -manifold with such that has noncollapsed ends (1.1). The following statements are equivalent.
-
(1)
has linear volume growth.
-
(2)
Either splits off a line or for any ray there exists such that for every .
-
(3)
Either splits off a line or there exists a ray and such that for every .
-
(4)
There exists a ray such that for any sequence , up to subsequence pGH converges to where is a compact space possibly depending on the sequence and .
-
(5)
For any sequence diverging to infinity, pGH converges to , for some that is a compact space and .
To close this section, we point out the technical difficulties we overcome in this note. A guiding principle is to use CheegerβColdingβs almost splitting theorem to replace the arguments in [antonelli2023isoperimetric]*Section 3, which exploit Alexandrov geometry. The overall idea for showing the equivalence in Theorem 1.6 is essentially the same, so we will emphasize our idea of proof of Theorem 1.2. Our proof is based on a special case of the almost splitting theorem, when the almost splitting happens along a ray. We describe it below.
Let be an open -manifold with , and be a ray. The Busemann function associated to , which will be introduced more formally in the next section, is defined as
Recall that in CheegerβGromoll splitting theorem if actually contains a line , then splits exactly as a product of and a level set of the Busemann function associated to . In the setting of the almost splitting theorem, let be a ray and . Consider the Gromov-Hausdorff limit of a convergent subsequence of . Viewing at the point , is still a ray and it extends backwards to , so we almost have a line. As an analog of CheegerβGromoll splitting theorem, it is a folklore result that the almost splitting theorem applied to gives that is Gromov-Hausdorff close to a ball of the same radius in . However, this is not exactly the statement of the almost splitting theorem. Instead of working with the actual Busemann function, the almost splitting theorem works with the harmonic replacement of it, see Theorem 2.8. The most technical part of this paper is to clarify in precise sense how this folklore result is true. This will be the theme of Section 2. The rest will be straightforward applications, as we will see in Section 3.
Acknowledgements.
The author would like to thank Gioacchino Antonelli and Marco Pozzetta for sending their paper [antonelli2023isoperimetric] and Gioacchino Antonelliβs master thesis. [antonelli2023isoperimetric] inspires the author to write the present manuscript. The author would also like to thank Shouhei Honda, Jiayin Pan, and Daniele Semola for reading a preliminary version of the manuscript and providing valuable feedbacks, Zhu Ye for spotting a mistake in the previous version, Igor Belegradek and Zetian Yan for some discussions of the constructions in section 5, Sergio Zamora for helpful suggestions on Proposition 2.12. The author also appreciate the anonymous referee for patiently correcting several typos and the suggestions that greatly improve the readability.
2. Busemann function and the almost splitting Theorem
2.1. Busemann functions
Busemann functions play a fundamental role in the study of noncompact manifolds with nonnegative Ricci curvature. Some fine properties of them are studied by Sormani [SormaniMiniVol, SormaniSublinear] in the case of linear volume growth. In this section we will review some of the results from Sormaniβs work and establish some direct implications. Since we are mostly dealing with metric geometry problems, we make a convention that all geodesics mentioned in this note are minimizing geodesics.
We start by giving the definition of a Busemann function. Let be an open -manifold. We call a curve a ray on if minimizes the distance globally, i.e.
for . Given a ray , the Busemann function associated to is defined by a monotone increasing limit
It is easy to see so the limit always exists. is a -Lipschitz function by triangle inequality and when it is a subharmonic function in the sense of distribution by the Laplacian comparison theorem.
The following lemma comes handy and will be repeatedly used in the sequel.
Lemma 2.1 ([SormaniMiniVol]*Lemma 6).
Let be a ray and be its associated Busemann function. For any real number , we have for any that
In particular, if and , then .
Lemma 2.1 is simple but powerful. We will make several uses of the following consequence of it both in this and the next section.
Corollary 2.2.
Let be an -manifold with and be a ray. Fix . Let and be such that such , and let . Then
(2.1) |
Proof.
if , then since and
we see immediately . On the other hand if , then by Lemma 2.1 we have , as claimed. β
We also mention a result that directly follows from Sormaniβs work [SormaniMiniVol].
Lemma 2.3.
Let be an open -manifold with . Let be a ray in and is the associated Busemann function. Suppose has linear volume growth (1.2). Then either splits or has only one end and for each , is compact.
Proof.
If does not split, then it follows from [SormaniMiniVol]*Corollary 23 exists. Assume otherwise there is nothing to prove. Take . Notice that by Lemma 2.1, , and by [SormaniMiniVol]*Theorem 19 that is compact, we can estimate that
Therefore . Clearly is closed, so it is compact. β
Next we will see something that typically happens for nonnegative sectional curvature as discussed in the introduction. It in particular shows that the asymptotic cone of a noncompact manifold with nonnegative Ricci curvature and linear volume growth is either a ray or a line. It is frequently attributed to [SormaniSublinear] which proves the sublinear diameter growth of Busemann function level sets. However, we were not able to find a direct statement nor a proof of this fact. We include a proof here to make explicit how this fact follows from [SormaniSublinear].
Theorem 2.4.
Let be an open -manifold with . Suppose has linear volume growth (1.2). Then the asymptotic cone of is unique and isometric to or . Moreover, splits if and only if the asymptotic cone is .
Proof.
If splits then obviously the asymptotic cone is unique and isometric to with Euclidean metric. It suffices to show that if does not split then any asymptotic cone is isometric to a half line. Let , and suppose that for a sequence of radii , pGH converges to . Take a ray with . By definition converges under the rescaled metric to a ray , in the sense that for any . In particular . We claim that . If not, there is a point but not on . Let . Then there exists so that and . Let . Since does not split, we see from Lemma 2.3 that , so we can assume . Furthermore, by Corollary 2.2 we have , which in turn gives
It follows that
The last equality comes from [SormaniSublinear]*Theorem 1. Notice that for some , up to subsequence . Meanwhile , we see that , a contradiction to our assumption that is not on . β
2.2. The almost spliting theorem
We then recall some essentials of CheegerβColdingβs almost splitting theorem as it is the fundamental tool we rely on in the next section. The following discussion will be largely based on Cheegerβs book [CheegerRicBook]. We will only be concerned with the case of nonnegative Ricci curvature and the dimension .
Let be an -manifold with . For two given points we define the excess of a point with respect to , as . We now describe the setting of the almost splitting at . Let be such that
(2.2) | ||||
(2.3) |
Denote by a nonnegative error function so that when are fixed, . Notice that may change from line to line.
Let . Fix a radius . The Abresch-Gromoll inequality provides an upper bound for the excess function on .
Lemma 2.5 ([AG],[CheegerRicBook]*Theorem 9.1, [XuLocalestimates]*Lemma 2.2).
We define the harmonic replacement of on by solving the Dirichlet problem
Similar constructions can be done when the superscript is replaced by , but we will not use them.
Lemma 2.6 ([CheegerRicBook]*Lemma 9.8,[XuLocalestimates]*(2.12)).
The harmonic function is called -splitting function in the literature. It satisfies
-
(1)
[XuLocalestimates]*(1.8). for some constant ;
-
(2)
[CheegerRicBook]*Lemma 9.10. ;
-
(3)
[CheegerRicBook]*Lemma 9.13. .
Under these conditions, it is noticed in several works that one can improve the gradient bound to
(2.4) |
See for example [XuLocalestimates]*Lemma 2.6 or [HP23]*Lemma 4.3 (substitute therein).
The most important ingredient of the proof of the almost splitting theorem is the following almost Pythagorean theorem.
Proposition 2.7 ([CheegerRicBook]*Lemma 9.16).
Let , with for some , and a choice of closest point to , then
(2.5) |
It is worth mentioning that in its proof, we have the following relation between distances and levels of , as shown in the inequality between and of [CheegerRicBook] as a consequence of Lemma 2.5 and Lemma 2.6, for in Proposition 2.7, it holds
(2.6) |
We now state the almost splitting theorem.
Theorem 2.8 ([CheegerRicBook]*Theorem 9.25).
Let be a -manifold with . Let , , so that , are satisfied. Set
where is a closest point to . Equip with the ambient metric and with the standard Euclidean metric. Take the product metric on . Then is a -Gromov-Hausdorff approximation between and where is a choice of closest point to and .
Remark 2.9.
We will also use the sequential version of the almost splitting theorem. Let be sequences of points so that the excess of with respect to tends to , and . Then pGH converges to . In particular, let there exists some with so that if is the harmonic replacement on of and is a choice of closest point of to , then pGH converges to . Here the distance on the product space restricts to .
Remark 2.10.
If with is also noncollapsed in the sense of (1.1), then the we get from Remark 2.9 is a noncollapsed ( in short) space. To show this we need to incorporate the splitting theorem of spaces [Gigli_splitting] and the volume convergence theorem of De PhilippisβGigli [DPG17]. We will use this observation without further mentioning the argument.
To close this section, we derive two consequences of the almost splitting theorem along a given ray. The first is a quantitative way of saying that the -factor split off in the limit space at infinity comes from the ray along which the limit is take. The second is to bound the diameters of Busemann function level sets given the information at infinity.
Proposition 2.11.
Let be an open -manifold with and be a ray. If for a divergent sequence , pGH converges to then for any , converges to along with this pGH convergence.
Proof.
We can assume otherwise we take . Let , , , and for some . Then we define , and to be the harmonic replacement of in . Denote by a choice of closest point to and a choice of closest point to . From the construction in Theorem 2.8, it suffices to prove that and along .
it is clear that the excess of with respect to , is . In what follows we will use the error function which is now independent of the excess and we abbreviate as . We first estimate using (2.6) and Lemma 2.6 that
Then,since , by the gradient estimate (2.4) we get
(2.7) |
and from Lemma 2.6 we infer that
(2.8) |
Combining together (2.7) and (2.8), we have . It in turn gives
(2.9) |
by (2.6). When is large we will have so .
We also observe that for large . Indeed, incorporating (2.9) we have
when is large. Now by the almost Pythagorean theorem, Proposition 2.7, we have
This completes the proof that since by Theorem 2.8, Remark 2.9 along .
β
Proposition 2.12.
Let be an open -manifold with and be a ray. If for any divergent sequence , up to subsequence pGH converges to for some compact , then .
Proof.
Let . We observe that . Indeed, if not we have a contradicting sequence so that and each is a pGH limit of for as . Then we can find a diagonal sequence denoted by so that up to subsequence converges to , but this can not be compact since it is an with infinite diameter, a contradiction. Set .
Now we bound the diameters of level sets of . We first set up the almost splitting along the ray . For some fixed , let to be chosen, let , , , and . Take and let be the harmonic replacement of in . Take . Since here we have , , we can use the error function abbreviated as . For each , there exists depending on so that
This is possible thanks to the local uniform convergence of
as for each fixed , . From Remark 2.9 we see that
By Lemma 2.6, we have that
Altogether we have . Let be a choice of closest point on to , . we can estimate that
(2.10) | ||||
Then since are arbitrary, we get from (2.10) that
Taking , hence , we obtain that
(2.11) |
There exists such that for , . We claim that holds for all . If not, suppose there exists and . We take so that . Let be a geodesic with and . Then for all ,
(2.12) |
Here we have used that is monotone increasing. To ease the notation we set for . In particular . By the assumption, it holds that
(2.13) |
On the other hand we can estimate for . Let be a choice of closest point of on then . By definition and by Corollary 2.2, , so . Observe also that is a continuous function and , by intermediate value theorem there is a so that . Then by Corollary 2.2 we have
It immediately follows from (2.12) and (2.13) that . However, this is a contradiction to we have derived. Thus we have shown that holds for all . Combine this with (2.11) we have completed the proof.
β
3. Structure of limit spaces at infinity
In this section the final goal is to prove Theorem 1.6. We need to further derive some properties of Busemann functions under the volume noncollapsed assumption.
The first lemma is standard, see [antonelli2023isoperimetric]*Corollary 3.11 or [SchoenYauBook]. We provide a proof here for completeness. In what follows, for any and , we denote the annulus by .
Lemma 3.1.
Let be an open -manifold with . Then for every and , it holds
(3.1) |
Furthermore, if has linear volume growth (1.2), and is noncollapsed 1.1 then for any sequence diverging to infinity, any pGH limit space of the sequence has linear volume growth with
(3.2) |
In particular if splits isometrically as , then and is compact.
It is not always the case that the same volume growth order is carried to a limit space at infinity. For example the paraboloid has volume growth of order , while its limit space at infinity is regardless of divergent sequence which has volume growth order .
Proof.
Let be the following bounded Lipschitz function with compact support.
(3.3) |
The distributional Laplacian comparison implies that
(3.4) |
So we can rearrange terms in the above inequality to derive that
(3.5) |
Now given diverging to infinity. We may assume that converges to up to taking a subsequence. Fix a radius then for large enough , . Applying (3.5) to , yields that
(3.6) |
Let , we get from the volume convergence [Colding97, Cheeger-Colding97I, DPG17] and volume growth assumption (1.2) that
(3.7) |
This is the sought estimates. If isometrically, then we have that . Since is with finite -volume, it is compact. β
We will turn to the study of Busemann functions with the noncollapsed condition (1.1). Some pathological behaviors are excluded as a result of this extra assumption. Note that the proposition we will prove below is not possible without (1.1) as mentioned in Remark 1.3.
Proposition 3.2.
Proof.
Assume the statement is not true, then there exists a sequence of points such that as . Clearly diverges to infinity. There exists such that . Let be a geodesic joining with . There are now two cases.
-
(1)
If up to subsequence . The limit space splits off a line coming from by the almost splitting theorem, Theorem 2.8 and Proposition 2.11. Write , where is an space. is compact thanks to Lemma 3.1. We set . The segments locally uniformly converge to a ray emanating from , and for any , .
We claim that for any , . Suppose not, there is a and a such that . Set , then . The last equality comes from the product metric structure. By proposition 2.11, . Meanwhile, by the choice of ,
when is large, a contradiction. This completes the proof of the claim.
However, our claim implies
which simplifies to for any . This implies that , which is impossible since is compact.
-
(2)
If up to subsequence , then up to subsequence converges to a ray emanating from . After a reparameterization, we can take . The convergence of yields that
(3.8) We show in this case must split. If does not split, then is compact due to Lemma 2.3. However, for any , we see from (3.8) that
Let , it follows that which contradicts the compactness of .
β
It is expected that the constant is independent of the ray .
Motivated by Proposition 3.2, we also consider that for a fixed ray there exists such that for any , as an assumption. This assumption is more general, for example it is allowed that the end of collapses to . It serves as a weaker substitution of linear volume growth (1.2) and noncollapsed condition (1.1). We will refer to it as the bounded distance to a fixed ray condition.
Lemma 3.3.
Let be an open -manifold with , and be a ray in . If there exists such that for any , , then does not split and the Busemann function associated to has finite minimum and its level sets have uniform diameter bound .
Proof.
If splits, write . Also we can set . Take , let (resp. ) be such that (resp. ) and (resp ) be such that (resp. ). Using our assumption and applying Corollary 2.2, we can estimate that
(3.9) | ||||
Now fix and let in (3.9), we see that . Likewise, fix and let in (3.9) we see that . Then since is continuous. We find by intermediate value theorem that there exists such that
In particular . However, we will prove that the diameters of the level sets of are uniformly bounded. This will be a contradiction.
To this end, we first show that has a finite minimum. Given let be such that , then by Corollary 2.2, , which in turn gives , as desired. Let , and for . Let (resp. ) be such that (resp. ), Corollary 2.2 implies . It follows from our assumption that
This estimate is independent of , we take take supremum over to get and then take supremum over . This completes the proof. β
Lemma 3.4.
Let be an open -manifold with , and be a ray in . If there exists such that for any , , then for any such that , is compact.
Proof.
It suffices to show for every . Here the distance on restricts to , which is isometric to . Fix then there is a diverging sequence such that . Let be such that . We have
Note that so are uniformly bounded, in particular, as . We also infer that there exists a subsequence which we do not relabel so that exists and . By Proposition 2.11, . Also notice that up to subsequence , thus we have . Letting in , we see that . This reads , yielding the desired bound . β
The next proposition provides a new perspective to understand Sormaniβs sublinear diameter growth of the level sets of a Busemann function [SormaniSublinear]. It confirms the intuition in Remark 1.3.
Proposition 3.5.
Proof.
If splits as , then is a compact -manifold of nonnegative Ricci curvature. Notice that the translations in the -factor are isometries, so we can bring any point into via an isometry, so the limit space at infinity is unique and it is . The uniform bound on now follows from Proposition 2.12 and Lemma 2.1.
If does not split, then by proposition 3.2 there is such that for any , . The desired result follows from the proof of Lemma 3.3.
β
Remark 3.6.
In fact, in the previous Proposition, if splits as for compact, then any ray lies entirely in for some . This is because the projection of any ray onto is a minimizing geodesic parametrized on , hence is either a ray or a point, and it cannot be a ray since is compact. In this case all the Busemann function level sets are isometric to hence trivial have uniformly bounded diameter.
Now we can provide a proof of the existence of splitting for any sequence of points diverging to infinity in the setting of Theorem 1.2. We will postpone the proof of equal volume of the cross sections to Proposition 3.8.
First part of proof of Theorem 1.2.
Suppose that does not split, otherwise let for some compact . As already observed, by translations along the -factor, we see that the unique limit space at infinity of is regardless of divergent sequences, and
Take a ray , and the associated Busemann function . Let be an arbitrary sequence, Proposition 3.2 asserts that there exists such that . Let . It is readily checked that
given the uniform boundedness of the level sets of due to Proposition 3.5. Meanwhile, For large so that , choose large so that and that . The Abresch-Gromoll excess estimate [AG]*Proposition 2.3 asserts that the excess of with respect to is controlled as follows
In particular, as , then by Remark 2.9 up to subsequence the limit space of splits off a line. We write . The compactness of and upper bound follows from Lemma 3.1. β
It then comes to the core of this note. Lemma 3.7 below allows us to study the isoperimetric problem with the exact same strategy as in [antonelli2023isoperimetric]. It is the most difficult step to generalize the results in [antonelli2023isoperimetric] from to .
Lemma 3.7.
Let be an open -manifold with . Let be a ray in and be the associated Busemann function. Suppose there exists such that for every , and does not split, then for any sufficiently small , there exists such that for satisfying it holds , where is such that .
Proof.
We set . We first observe from Corollary 2.2 that
(3.10) |
Then we prove by contradiction. Assume there is a so that there is a divergent sequence with . Here, we have simplified notations in the way that and . We also have from (3.10). So up to taking subsequence, we can assume that and with , granted Proposition 2.11.
Let , and for some to be chosen, we define
and be the harmonic replacement of in . We aim to prove that for some . By the almost splitting theorem, Theorem 2.8, we only need to show . Here makes sense because . We see that by definition, for any there exists so that . It follows from Lemma 2.6 that , so follows.
Now by the definition of we have . Let , we have
This is a contradiction to , and we conclude the proof.
β
We are at a good stage to generalize the result of [antonelli2023isoperimetric]*Theorem 3.10 to nonnegative Ricci curvature.
Proposition 3.8.
Let be an open -manifold with such that is noncollapsed (1.1) and does not split. Let be a ray in and be the associated Busemann function. Suppose there exists such that for every . Take a sequence such that converges to for some compact space and . Denote by the projection onto the -factor, the following holds.
-
(1)
locally uniformly converges to along the pGH convergence . In particular for any with the (characteristic functions of) corresponding (sub)level sets converge.
(3.11) (3.12) -
(2)
For any , has finite perimeter, and for any with , the following monotonicity holds
(3.13) (3.14) Moreover the following rigidity holds. If for some , then there exists such that is isometric to (with the metric given in its structure).
-
(3)
Let be an arbitrary diverging sequence in , and the pGH limit space of be , for some compact space and , then .
The proof of item (1) and item (2) can be done verbatim as in [antonelli2023isoperimetric]*Theorem 3.10 given Lemma 3.7, so we skip the proof.
For the rigidity, it is the same as [antonelli2023isoperimetric]*Theorem 3.10. At an intuitive level, If for some , then becomes a harmonic function in , then a rigidity theorem of Kasue [KasueRigidity] can be invoked to deduce the rigidity we seek for.
proof of item (3).
We are finally in position to achieve the goal of this section.
Proof of Theorem 1.6.
We show and .
(3) (1) We infer from Lemma 3.3 that the dichotomy in (3) are mutually exclusive. Assume does not split and
(3.15) |
Take . By definition , so . We invoke also item (2) of Proposition 3.8 and the coarea formula to estimate that
(3.16) |
The limit space at infinity taken along splits as by Theorem 2.8. We deduce from Lemma 3.4 that is compact hence has finite volume because is . In particular, is finite.
is Theorem 1.2.
is trivial.
If splits off this is clear. Assume does not split. We claim that the provided by (4) is the ray we search for and argue by contradiction. It follows from the proof of Proposition 3.2 that there exists a ray with
(3.17) |
The first item there is excluded because of our assumption.
We claim that is bounded from above. If not, there exists such that . Then it is easily seen by (3.17) that . On the other hand, up to subsequence converges to and is compact. Proposition 2.12 shows that . However, this would force , a contradiction. The claim is proved. Let us assume for some . There is a such that . The proof of Lemma 2.3 can be applied to show that is compact. Now has at least two noncompact connected components contained in and because both of them contain a ray, which means has at least ends. So splits, a contradiction. β
4. Application to isoperimetric problem
The results in previous section clear an obstacle of studying the existence of isoperimetric sets of a given large volume. Therefore we are able to generalize [antonelli2023isoperimetric]*Theorem 5.13 to nonnegative Ricci curvature. Except for the fact that the limit space at infinity is unique. Note that for isoperimetric problem we only consider noncollapsed spaces because of [isoregion]*Proposition 2.18. It is worth pointing out that most of the technical work has been done in [antonelli2023isoperimetric].
Let us first briefly recall the problem without the attempt to exhaust literature in this field of study. Let be an open -manifold with . The isoperimetric problem asks for a given volume , the existence of a set of finite perimeter that realizes
We call the isoperimetric profile of . A set of finite perimeter such that is called an isoperimetric set. One of the difficulties in establishing the existence results for isoperimetric sets is that a minimizing sequence may escape to infinity. It is now understood that an asymptotic mass decomposition technique can be applied to spaces to study how a minimizing sequence can escape to infinity, see [antonelli2022isocomparison]*Theorem 4.1, [antonelli2023isoperimetric]*Theorem 2.22 and references therein. For , this technique can be pushed further [antonelli2022isocomparison], deriving
Proposition 4.1 ([antonelli2023isoperimetric]*Proposition 2.25).
Let be an open -manifold with and noncollapsed ends (1.1). Given , let be a minimizing sequence for with bounded perimeters. Then exactly one of the following happens
-
(1)
either there exists an isoperimetric set for ,
-
(2)
or there exist a limit space at infinity and an isoperimetric set so that .
Therefore, the focus is to understand the isoperimetric sets in the limit space at infinity. It seems that this is even harder than studying it on manifolds. However, splitting drastically simplifies the structure of a limit space at infinity. This is most powerful when has linear volume growth. As shown in Theorem 1.2, in this case any limit space at infinity is a cylinder , and the isoperimetric sets are well understood in such cylinders by [antonelli2023isoperimetric]*Appendix A.
We set for any limit space at infinity . It is justified by Theorem 1.2 that all βs have the same volume. The following is a generalization of [antonelli2023isoperimetric]*Theorem 5.13 to Ricci curvature lower bound.
Theorem 4.2.
Let be an open -manifold with such that is noncollapsed (1.1), has linear volume growth (1.2), and does not split. Then the following holds.
-
(1)
There exists , such that an isoperimetric set exists for any . (recall that is from (1.1).)
-
(2)
For every , we have . If for some , the equality holds, then it holds for all and there exists an isoperimetric set of volume such that is a space and that is isometric to .
-
(3)
Let be a sequence of isoperimetric sets with , . Let be the pGH limit space of . Then converges to in and converges to along the pGH convergence. Moreover there exists a sequence of isoperimetric sets so that and
Again, the proof will be almost the same as the original one [antonelli2023isoperimetric]*Theorem 5.13. Instead of reproducing the proof, we only point out how results in this note are applied. We refer the readers to the aforementioned paper for a detailed proof.
For item (1). The only subtlety compared to the original proof is the loss of uniqueness of the limit spaces at infinity, but this is resolved by Theorem 1.2 which says even different limit spaces at infinity have the same volume. One shows that for a minimizing sequence it cannot be the case (2) in Proposition 4.1. If this was the case, then one can use sublevel sets as a competitor for .
For item (2). Nothing need to be changed. This is the Kasue-type rigidity in Proposition 3.8 item (2).
For item (3). With Theorem 1.2, Lemma 3.7 and Proposition 3.8, (1). The proof can also be done verbatim.
The following consequence is pointed out by the referee.
Corollary 4.3.
Proof.
Take a ray emanating from . We let be such that guaranteed by Proposition 3.2. Then the level sets of have a uniform diameter upper bound and thanks to Lemma 3.3. Recall from the proof of Theorem 1.6, specifically (3). We obtained there that for sufficiently large , and
To show the reverse inequality, we see from the uniform diameter upper bound that . By co-area formula we have
it follows that
This completes the proof since is contained in item (3) of Theorem 4.2. β
Remark 4.4.
We are unfortunately unable to add the statement
There exists such that for any , . |
to the list of equivalence conditions in Theorem 1.6. The argument in [antonelli2023isoperimetric]*Corollary 3.11 (4) (1) or (3) can also be applied in our setting to show that this statement implies that there exists a diverging sequence (in fact, many such sequences exist) for which the limit space at infinity splits off a line. However, to go from one sequence to all sequences seems to be out of reach. Because different sequence may have different βescape speedβ, i.e., the growth of for fixed and . For Alexandrov spaces, the escape speed does not matter and it is shown in [antonelli2023isoperimetric]*Theorem 5.13(1) that the limit space at infinity is unique. For Ricci limit spaces, we do not know if two divergent sequences of different escape speed are related to each other or not.
Finally, we mention that large volume is a necessary condition for the existence of isoperimetric sets. A counterexample [antonelli2023nonexistence] has been found when the volume is small.
5. Examples
In this section we will borrow the examples from [CN11]. Recall that topologically is the connected sum of and another with reversed orientation. The examples we will construct are all noncollapsed, since all the limit spaces at infinity have the same dimension as the manifold we start with.
Theorem 5.1.
There exists an open -manifold with nonnegative Ricci curvature and linear volume growth such that both and are possible limit spaces at infinity of . In particular the cross sections are not homeomorphic.
Proof.
It is constructed in [CN11]*Example II a family of metrics on , so that
-
(1)
.
-
(2)
, i.e., the volume are the same for .
-
(3)
.
Moreover converges in GH topology to with singular points. Based on this model, we construct a cylinder whose cross sections oscillate between and for . To this end, we construct a Riemannian metric on of the form
Step 1. We start with the case when is large.
Define . clearly, every real number in is a possible limit of as . The Ricci tensors are computed in [CN11]*Lemma 2.1. We use subscript to represent vector fields along direction, to represent vector fields along . In what follows we already incorporated the fact that .
(5.1) | ||||
(5.2) | ||||
(5.3) |
Let for to be determined. We can easily verify the linear volume growth. Now let us verify the nonnegative Ricci curvature. By our definition of we have and
provided that and .
Next note that
as , and is bounded. |
We have the following control
(5.4) |
Here the last two terms are zero when for , and when it also decays faster than by our choice of . In conclusion we have for large . Finally, we are left with the verification of . First we see from (5.2) that
(5.5) |
Then we adapt the strategy of [CN11]*Lemma 2.1 to show the positivity of . Fix a point and find an orthonormal basis at this point. Write every possible mixed direction as for . It holds
Choose to be an increasing sequence so that and choose to be an increasing sequence so that , and fix an arbitrary point . It is immediate that the pGH limit along is ), and the pGH limit along is .
Step 2. This step is to transit the Riemannian metric on from some to , which is known to be closable. Assume that the previous construction is done in for some large and satisfying , i.e., will be chosen among , . We extend our construction to with the following properties.
-
β’
can be smoothly extended on so that , and for . In particular at and has no contribution to the second fundamental form at .
-
β’
and remains positive for any .
To achieve the two items above, let be a smooth function on such that when , , when , and on and its derivative of all orders vanish at . Set
On , we define . Then it is readily checked that is smooth at , , and . Now we check the Ricci tensors. First by (5.1) we have
This is still true provided that is large enough since our modification only contributes some terms with exponential decay. For the same reasoning works, we just need to choose large so that . It again remains to see . Recall (5.2), we can again use the same trick as the one right below (5.5).
We now compute the second fundamental form at . Since , we have that
Step 3. For the last step, we would like to glue the manifold with boundary we constructed to the so-called constructed in [CN11]*Example II. This would give us a manifold without boundary. For the gluing, note that it is shown that the second fundamental form of is strictly positive, say . By choosing big enough, we can have .
So far we have already chosen by combining all the requirements in the last two steps. It remains to determine the constant in . We want to be isometric to . This can be achieved by choosing some since and is fixed and has small diameter. This means the previous bound for still holds so the bounds for and also hold. In conclusion after scaling the nonnegativity of the Ricci tensors of equipped with is preserved. Now by Perelman [PerelmanLargeBetti], we can glue and along the boundary, so we are done with our construction. β
We proceed to the next example. As suggested by Daniele Semola, we can also construct a manifold with nonisometric limit spaces at infinity in dimension , reducing the dimension requirement in Sormaniβs example [SormaniMiniVol]*Example 27, which is .
Theorem 5.2.
For a given volume , there exists an open -manifold with nonnegative Ricci curvature and linear volume growth such that the cross sections of limit spaces at infinity can be with infinitely many different metrics of the form for and . Moreover with any of these metrics, the volume of is .
Proof.
It follows from [CN11]*Example I and our previous construction. For a parameter , we denote the spherical suspension over a manifold by with Riemannian metric for . Consider with parameters , where is the radius of . We normalize the given to be , as the construction is the same for all other volumes. The set
is a smooth curve in that contains . For the construction we can fix a parametrization so that and corresponds to . There are several ways to smooth while keeping positive Ricci curvature, for example removing a neighborhood of each singularity and gluing in a spherical cap, which is possible by Perelman [PerelmanLargeBetti]. For convenience we use Ricci flow starting from a -dimensional Alexandrov space [RichardSmoothing] instead, since it naturally gives rise to a one-parameter family of metrics with the same lower curvature bound and a uniform control of distances hence volumes [RichardSmoothing]*Theorem 1.1. Let be the possibly singular metric corresponding to , and be the Ricci flow starting from . Along with our choice of parametrization of , we can also assume the derivatives of satisfies (3) of Theorem 5.1. We consider the metric on of the form
Again take , and . It is easy to see that as the possible limits of are exactly points in . The construction is now the same as the previous one. To see the bounds of derivatives as in (3) of Theorem 5.1, we recall classical curvature estimates along Ricci flow which in turn implies that
Here we have used the evolution equation of Ricci curvature under Ricci flow and denotes the derivative w.r.t. the second subscript of .
As in Step 2 of the previous example, after completing the contruction for large , we let the metric to transit from to and then we can glue in a disk to close it.
β