o. Basic definitions, examples and problems
2.5. D. On the homotopy role of the Lq-norm of the differential
1)/
on H for q<
N. Let us look at the mapsf
withIIDf
IHllLq
~ c for some q < N. Now, we do not have the full finiteness-contractibility result but one obtains a weaker homotopy conclusion by restrictingf
to the k-skeleton Vk C V brought to a sufficiently general position with respect tof.
Then the boundIIDf I HllLq
~ c implies a similar bound on Vk , namelywhere c' = const ·c, where the "tangent bundle" T(Vk) is understood as the set of the vectors tangent to the (smooth!) simplices in Vk and where the Lq-norm of the differential V
f
on the intersection T(Vk) n H is obtained by integration over V k. Now, if the intersection T(Vk) n H induces a C-C structure of formal Hausdorff dimension N' ~ q on each k-face of Vk , then the bound (*)' has non-trivial homotopy effect onf' = f I
Vk and hence onf·
Namely, ifIIVf I HliLq
is sufficiently small, then the restrictionf
I Vk is contractible whenever Vk can be stably brought into a position where dimHau (Vk, T(Vk) n H)) ~ q. Here the stability means the existence of a measure J-L on the space of embeddings (Le. positions) Vk in V having dimHau ~ q and such that the push-forward of the measure J-Lxdvk to V is absolutely continuous with respect to Lebesgue (or equivalently C-C Hausdorff) measure in V.If V is Riemannian (i.e. dim V
=
dimHau V) then the above stability is trivially satisfied with N' = k and so every smooth mapf :
V ----+ W withIIVfllLq
small is null-homotopic on some (and hence every) k-skeleton Vk C V with k ~ q. In fact the converse is also known to be true. Every smooth mapf :
V ----+ W sending a small neighbourhood of Vk to a point can be composed with a suitable diffeotopy 'Pt : V ----+ V, 0 ~ t<
00,such that
IIV f
0 'PtIlL
q t---+(X) ----+ 0 for every q<
k+
1. (If V= sn
andVO C V is the south pole, one uses the north pole south pole push for 'Pt. In general one uses such pushes in the cells in V - Vk . First every n-cell B is radially pushed from the center bo E B toward the boundary, so that in the limit for t ----+ 00 all of B - {bo} goes to aBo Then one
2.5 C-C SPACES SEEN FROM WITHIN 181
composes the above push toward Vn - 1 with a similar push of a small neighbourhood of vn-1 toward Vn - 2 and so on, see p.388 in [E-L] and references therein. Probably, a similar construction can be carried over for certain C-C manifolds, e.g. the contact ones.
Now, let us look at the maps f with IIDf
I
HllLq<
C with a possibly large c and observe that there are at most finitely many homotopy classes of restrictions off
to Vk in the following two cases, (i) q>
N' and (ii) q = N' and 7r1 (W) acts trivially on 7rk(W), where N' is, as earlier, the minimal integer so that Vk can be made stably of formal Hausdorff dimension N' for the C-C metric associated to T(Vk)n
H. For example, N' = k in the Riemannian case. If V is a contact C-C manifold, then N'=
k for 2k<
dim V and N'=
k+
1 for 2k ~ dim V as follows from the discussion in 3.4.B, and see 4.? for the general C-C case.The space Fq,c' Let us look at the homotopy property of the space Fq,c of the maps f with
liD
f I HllLq ~ c. We have just seen that the zero-dimensional homotopy (i.e. connected components) of Fq,c are strongly affected by q and c but, probably, there is no additional link between the geometry of mapsf
(encoded into q and c) and higher homotopies of spaces of these f for q<
N. Namely, if fa : V -* W is a family of smooth map parametrized by a compact polyhedron A :3 a such that each fa can be individually contracted to Fq,c then, conjecturally, the whole family can be continuously moved to Fq,c (or, possibly, to Fq,cl for c' = c'(c, A)) in the case where q<
N. This appears easy in the Riemannian case.For example, the above diffeotopy 'Pt of V works for families of maps fa : V --t W which send a fixed (i.e. independent of a) skeleton Vk to a point. In general, if IIDfallLq is small for all a, we can only have fa almost constant on Yak depending on a. In fact we can make Yak constant in a on each simplex of a suitable subdivision of A and then 'Pa,t can be probably build using some induction on skeletons (or partition of unity) in A.
Exercise. Determine the homotopy structure of the space of maps
f :
81x[O,1] -* 82 with
IIDfllLq ~f
UIIDfllq)t~
c for given c>
0 and1 ~ q
<
2.182 MIKHAEL GROMOV § 2 corresponding result in the C-C category remains conjectural. A closely related (apparently more global) question is that of the homotopy content of FfI and the homotopy structure of the inclusions Coo n FqH C FfI and Coo nF,;;c C F,;;c C FfI where F,;;c = {f E FqH
IIIDI I HllLq ::;;
c}, where the space FfI is given L;J -topology arizing from the normIIV f I HllLq
viaan embedding of W into some Euclidean space. One asks in this regard, for example, which homotopy invariants of smooth maps extends to
frH
and are FfI -continuous (or continuous in some weaker topology on Fq ).
2.5.E'. Examples of
F,JI
-non-density of smooth maps inF,JI.
(7) corollary in 2.5.A applied to concentric spheres in the ball B=
Bn.Now let V be a smooth (topological) ball around the origin in a nilpo-tent Lie group with a one-parametric self-similarity such that each orbit of this self-similarity transversally meets sn-1
=
aV at a single point.Then the radial projection fo : V ~ sn-1 along the orbits is in FfI for all q
<
N = dimHau V and neither this fo nor anyI
= cp 0 fo for a non-null-homotopic cp can be approximated by smooth maps in Ff!-l (under the standing assumption of (codim I)-stability of V).7 Compare [Sh-Uhl], [Beth] and [HajASM].
2.5 C-C SPACES SEEN FROM WITHIN 183
2.5.F. Space
FJ!.
Let us look at the homotopy structure of a mapf
E FJ!. We start with the (well known) Riemannian case (where H =T(V)) and observe that if f E F,JI then the restriction f I V' has IIVf I TV'IILq(VI)
<
00 for a generic hypersurface V' in V and so for q>
dim V-1 this restriction in continuous. However, the homotopy class off
I V' may jump under small perturbations of V' in V for q<
dim V. Let us show this does not happen for q=
n=
dim V. We observe that for each point v E V and every small positive c there is a sphere 88 around v of radius 8 in the interval c ::s; 8 ::s; 2c, such that the Ln-norm off
on this sphere 88 with the normalized metric (= dist /8) is bounded by the Ln-norm off
on the ball B2e (by integrating IIDflln over the annulas between 8e and 82e ).It follows, that Diamf(88) is bounded in terms of
fB2E
IIVf(v)lIn dv, as we, in fact, have seen earlier and therefore for every c>
0 one can cover V by balls of radii between c and 2c, such that every ball among these at most v = v( n) neighbours. We assume without loss of generality that the union ~ of the boundary spheres of these balls is connected and we partition V into the connected components of the complement V -~, say V = UUi , where the boundary of each Ui has Diamf(8Ui ) ::s; vDiamf(88 )which uniformly (in i) goes to zero as c -+ O. Then one can regularize
f
by using some standard continuous extension of
f
I 8Ui for each i to all of Ui within the (small) ball of radius p=
Diamf(8Ui ) in W. Furthermore, given a hypersurface V' C V, one can do the same to the (finer) partition into the connected components U: of V - (~ U V') as all Diamf(UD are necessarily small for small (now, depending also on V') c. Thus the (continuous) restrictionf
I V' : V' -+ W admits a continuous extension to all of V which, in particular, imply the contractibility off
I V' for small spheres in V (which are contractible in V). Then, obviously, the homotopy type of f I V'is invariant under the deformations of V' in V and also under homotopies off
in the space F!! for H = T(V).The same reasoning applies to (codim I)-stable C-C manifolds V. For example, if V is a contact C-C manifold, then every map
f
E FJ!, N = n+
1, has a well defined homotopy class of the restriction off
to the (n - I)-skeleton of V, provided n=
dim V ~ 5. (It is unclear what happens for 3-dimensional contact manifolds.)184 MIKHAEL GROMOV § 2
2.5.F'. Regularization of
F{:f
-maps. The above process of filling small Ui-holes by continuous maps (extending f I aUi ) allows us to ap-proximate everyf
E FfJ by a continuous map, sayf' :
V ----7 W, such that any two such approximation toI
are mutually homotopic. More precisely, we have the following proposition (which is well known in the Riemannian case and is due, I believe, to K. Uhlenbeck).Let V be a compact (codim 1 )-stable C-C manifold (i.e. with sufficiently many (N - 1)- dimensional hypersurfaces for N
=
dimHau V, e.g. V is Riemannian or contact of dimension;) 5) and f : V ----7 W be a map withIIV
fI
HIILN :( c<
00. Then for every E:>
0 there is a decomposition of V, say V=
VE U V1 - E with the following three properties.(1) VE is an open subset in V of meSN VE :( E: and V1 - E = V - V1 - E;
furthermore, each connected component U of VE has Diam(U) :( D where D ----7 0 for E: ----7 O.
(2) The restriction I
I
V1 -E is a continuous, moreover, C/3-Holder map forf3
= liN (where the implied Holder constant may depend on E).Furthermore the image f(aU), of the boundary of every component U ofVE, has Diamf(aU) :( D' where D' ----7 0 for E: ----7 O.
(3) IfW has locally bounded geometry (e.g. compact) and E is sufficiently small, then f I V1- E admits a continuous extension say flO : V ----7 W which is contained in FfJ and, moreover, has
IIV
IEI
HIILN :(IIV
II
HIILN" (In fact this IE may be chosen taut on VE relative to the boundary aVe C VI-E.)
The proof follows by our earlier argument and is left to the reader.
Notice that the maps
f
converge tof
in FfJ for E: ----7 0 and so they are all mutually homotopic for small E: by the weak homotopy stability observed in 2.5.A. Also notice that the regularization f f--+ flO applies to families of maps and shows that the space FfJ is homotopy equivalent to the space of continuous maps V ----7 W (where V and Ware compact and V is (codim 1 )-stable, i.e. has many "nice" hypersurfaces as we always assume). In fact the inclusion of the space C1 of smooth maps V ----7 W into FfJ is a homotopy equivalence. Furthermore the space FfJ (as well as C1 C FfJ with the induced topology) is locally contractible. (Of course, this all is well known for Riemannian manifolds V.)2.5 C-C SPACES SEEN FROM WITHIN 185
Example. Let V be homeomorphic to the sphere
sn
and W be an n-dimensional Riemannian manifold with locally bounded geometry. Sup-pose there exists a L~-mapf : sn
---+ W which has degree 1 in the(a) The degree 1 condition can be more succinctly expressed if V is Rie-mannian by
Iv
f*(w) = 1 for the normalized oriented volume form w on W but I do not know how to do this in the general C-C case.(b) Here and in future N refers to the Hausdorff dimension of V if it is equiregular and to the formal dimension max ~i (ni - ni-l) otherwise.
But in fact many of our results hold true for N = dimHau under milder (genericity) assumptions than equiregularity.
2.5.G. Restriction of
LIJ
-maps to k-dimensional submanifolds in codim-stable manifolds for q<
Nand k<
q. If V is Riemannian then the restrictionf
I V is obviously continuous for generic V' C V of dimension k<
q (wheref
E F[(V)). Furthermore if k ~ q - 1, then the homotopy type of this restriction is well defined. This is derived from the case codim V'=
1 as follows. First, let co dim V'=
2, take two V' which are transversal intersections of hypersurfaces, e.g. to those with trivial normal bundles and then we localize in a standard fashion to make the intersection trick work for all V'. Moreover, one easily shows with a properly localized intersection argument thatf
defines the homotopy type of the restriction off
to the k-skeleton of V for k = ent(q - 1) (i.e.k is biggest integer ~ q - 1).
Co dim-stability. Call V stable in codimension n - k if for each tangent k-plane T' C T(V) there exists a germ of smooth sub manifold V' tangent to T' such that the formal Hausdorff dimension of V' at most equals that of V minus n - k (compare 2.4.B). This means, intuitively, that generic k-dimensional submanifolds V' have codimHau V' = codimtop V'.
186 MIKHAEL GROMOV §2
Example. Contact manifolds are (n - k )-stable for k ~ 3.
If V is (n - k )-stable then the restriction of
f
E F: to generic V' of dimension k<
q is continuous and if k ::;; q - 1 the homotopy class of this restriction is well defined by the above Riemannian argument. This applies in particular to k-dimensional sub manifolds in contact manifolds for k ~ 3 (where the situation for k ::;; 2 remains unclear).Restriction to horizontal submanifolds V' in a contact V. Such V' are plentiful for k
<
n/2, n = dim V, and the restrictionf I
V'iscontinuous for generic horizontal V' provided q
>
k (compare 3.1). Now we claim that if k ::;; q - 1 and k+
1<
dim V /2, then the homotopy class of this (continuous) restrictionf I
V'is stable under (horizontalf) deformations of V'. This implies (via the horizontal triangulation of V, see 3.4.B) that eachf
EF:
has well defined homotopy class of the restriction off
to the k-skeleton of V for every k<
(dim V /2) - 1 andq~k+1.
Proof. To grasp the idea we start with the Riemannian case and indicate another way of reducing the case codim V'
>
1 to that of co dim V'=
1.For example, let V' be a small k-dimensional sphere in V and let us show that, generically, the restriction
f
I V', which is continuous for q>
k, is contractible for q ~ k+
1. We make this sphere V' the boundary of a small (k+
I)-ball V{c
V and observe that the restrictionf
I V{ is genericly contained in the space F;revD on V{. Then we restrict further, to a generic sphere in V{ concentric to aV{, and obtain contractible (as well as continuous)f
on a generic small k-sphere in V. This argument easily generalizes to all (generic) V' C V and then it extends into the contact framework with the provisions of 3.1, 3.4.A.2.5.G/. On singularities of
f
E F:. One can imagine, following Karen Uhlenbeck, every mapf
EF:
as being regular (e.g. continuous) away from a certain (pole-like) singularity ~ f C V which, in the case where V is Riemannian, has "dim" ~f ::;; ent(n - q). This is justified by the following three facts(1) generic V' of dimension k
<
q misses ~f asf I
V'is continuous;(2) if k ::;; q - 1, then generic I-parameter families of V's miss ~f and so the homotopy class of
f I
V'is well defined;2.5 C-C SPACES SEEN FROM WITHIN 187
(3) the blow-up construction in 2.5.E' applied to the balls Bk+1 xb E Bk+l xBn-k-l
=
V, b E Bn-k+1, gives usJ
E Fq , with q=
k+
1- cfor all c
>
0, and with (n - k - I)-dimensional singularity.Now, in the C-C case, one should think of "£f as some virtual subset (in V or in some auxiliary jet space over V) so that generic submanifolds V' and generic horizontal submanifolds of the same topological dimen-sion have different chances to meet "£f. (Notice that both, "generic" and
"generic horizontal" submanifolds V' of Hausdorff dimension.e, miss "£j,
J
E F:, if .e<
q, and the restrictionJ
I Viis homotopically sound if.e~q-l.)
We conclude by observing that the restriction problem is still not solved in full generality for C-C manifolds. Namely, when does every map
J
EFqH restrict to a continuous map on a generic submanifold V' E V of dimension k? When is the homotopy class of the restriction of
J
to the k-skeleton of V well defined? When does there exist at least one submanifold V' C V of dimension k for whichJ
I Vi is continuous?2.5.H. On local estimates for taut L~-maps. Consider a taut map J on an co-ball B(co) C V where
IB
IIDJ(v) I HIIN dv ~ Co for a small Co>
0 and let us evaluate the diameters of the J-images of the concen-tric c-ball B(c) for c ~ 1/2 co. SinceJ
is taut, this diameter does not exceed the infimum of those for the concentric spheres S(p) for p E [co,c]and Diam J(S(p)) is bounded by constp-l Is(p) IIDJ(v) I HIIN dv, pro-vided the induced C-C geometry on these spheres has formal Hausdorff dimension N - 1. (In fact, the C-C spheres are usually non-smooth and so the formal dimension makes no sense. The true condition we need is, of course, (codim I)-stability which allows us to approximate the spheres by piecewise smooth hypersurfaces of formal dimension N - 1.) Then we integrate over p E [co, c] and conclude to the uniform continuity of Jon B(cO /2) with the logarithmic modulus of continuity,
dist(f(v),J(v' )) ~ consto co1(-10gdist(v,v'))-k for all v, Vi in B(cO /2).
188 MIKHAEL GROMOV §2
2.5.H'. HOlder estimates. Let us additionally assume that our
f
min-imizes the energyf
I--t IIDf I HllfN and prove that thenf
is Holder. It is well known (and obvious by the previous discussion) that the Holder bound for taut maps issues from the following monotonicity inequalityr
IIVf(v) I HIIN dv~
Ar
IIVf(v) I HIIN dv (*)} B(2e) } B(e)
for 10 :::;; 100/2 and some A> 1 independent of c. Now, since
f
is minimizing, no extension of f from the sphere 8(210) = oB(2c) to a map f' on B(2c) may have IB(2e) IIVf'(v) I HIIN dv<
IB(2e) IIVf(v) I HIIN dv, and so the following lemma yields (*).Modification lemma. Every taut map fo : B(2c) -+ W which lands in a (small) ball within the range of the convexity radius of W can be modified to a map f' : B(2c) -+ W agreeing with fo on 8(2c), landing in the same (small) ball in Wand having
r
IID!,(v) I HIIN dv:::;; Cr
IIDfo(v) I HIIN dv, (*+)} B(2e) } A(e)
where A(c) denotes the annulus B(2c) - B(c) in V.
Proof. Let Bo be the minimal (convex!) ball in W containing the image fo (B (~ c)) and Wo be the center of Woo Observe that the radius of Bo
1
is bounded by Ro
=
Rad Bo :::;; Co (IA(e) IIVfo(v) I HIIN dV) N by our earlier argument (using (codim I)-stability and the tautness of fo). Then we use the notation W I--t 8w, 8 E [0,1], for the geodesic scaling of the ball Bo toward the center (i.e. 8w stands for the geodesic convex combination 8w+
(1- 8)wo) and define f' on B(2c) by "compressing" fo by means of the cut-off distance function. Namely, take{ I for v E B (2c) - B (~ c) , d(v)
=
(c/2)-1 dist(v,B(c)) for v E B(~ c)
and f' (v) = d( v) fo (v). The horizontal differential of f' clearly satisfies IIV!'(v) I HII :::;; C 1 Ro
+
IIVfo(v) I HIIall v E A(c) and Vf'(v) = 0 on B(c). Thus (*+) follows from the above bound on Ro and the volume bound meSN B(2c) ;S ION.
2.5 C-C SPACES SEEN FROM WITHIN 189
Remarks
(a) The minimizing property of
f
can be replaced by "quasi-minimizing".This means that, for every relative compact domain U C V and every map
l' :
V ---+ W obtained by a homotopy of V fixed outside U, the total energy ofl'
on U can not significantly smaller than that off,
i.e.