Arxiv:1904.00515V1 [Math.DG] 1 Apr 2019 Wa Samnfl” H Emtia N Oooia Devel Topological and Geometrical of the Discovery the Th Manifold”

Home , Geometric measure theory

arXiv:1904.00515v1 [math.DG] 1 Apr 2019 wa samnfl” h emtia n oooia devel topological and geometrical of The discovery the th manifold”. address a they [11], is [ in deep “what Stallings Gromov provide by popularised proof), others, As many modern manifolds. among alternative, [16], an Kirby–Siebenmann for and [22] Pugh see admi it does manifold, Lipschitz or homology, topological, topology: geometric and ferential atnWinyPeettos ltForms. Flat Presentations; phrases. Cartan–Whitney and words Key 90C56. 2010 Date remn[] oado–ulvn[]adGoo [10]. Gromov and [7] Donaldson–Sullivan [9], Freedman ual oagn ude.Issections Its Lipschitz bundle”. a cotangent of probl surable smoothability smoothability the the detecting of tackle programme to the developed been has proach yWinyi i hoyo emti nerto hoy[30 theory integration geometric of theory his in Whitney by orm ihwa euaiyada setal nondegener essentially of an tion and regularity weak with coframe ota h oa ereof degree local the smoothabilit that local to the that proved [15] Heinonen–Sullivan ihteSblvrglrt condition regularity Sobolev the with prahtesotaiiypolm tcnet .H Clark H. F. connects It problem. smoothability the approach ATNWINYPEETTO,NNSOT NLSSAND ANALYSIS NON-SMOOTH PRESENTATION, CARTAN–WHITNEY pi ,2019. 2, April : ahmtc ujc Classification. Subject Mathematics onainlwrso mohblt fmnflsb Whiteh by manifolds of smoothability on works Foundational h mohblt ftplgclmnflshsln enaq a been long has manifolds topological of smoothability The nteohrhn,uigtetcnqe rmgoercmeasu geometric from techniques the using hand, other the On narcn ae 1] rn-e esetv a enadop been has perspective brand-new a [17], paper recent a In ϑ 1 1–4) swl stecnes ttmn.I xlrsthe explores It statement. analysis converse non-smooth the of as well as 219–249), asvaImrin n ieetal xtcShr Theore Tanak Sphere M. and Exotic Kondo Differentiable K. and in Immersions 1.3 via (Theorem Maps [17] in result main the of Abstract. frswt ooe euaiyo eti aiiso homol of families certain on regularity Sobolev with -forms ln emnsgvsrs oabace oeigmap covering branched a to rise gives segments along MOHBLT FMNFLS NATERMOF THEOREM A ON MANIFOLDS: OF SMOOTHABILITY exotic sn ol n eut rmgoercmaueter,w gi we theory, measure geometric from results and tools Using mohblt fMnfls o-mohAayi;Singula Analysis; Non-smooth Manifolds; of Smoothability ( i.e. F oemrhcbtntdffoopi)srcue;seMilno see structures; diffeomorphic) not but homeomorphic , la à ϑ 1 = .H lreadteeitneo pca ytm fWinyfl Whitney of systems special of existence the and Clarke H. F. utemr,Hioe–et 1]etbihdisequiva its established [13] Heinonen–Keith furthermore, ; ie aiodwt tutrso ekrglrt,eg,a e.g., regularity, weak of structures with manifold a Given rmr:4Q5 80,5R0 71,5R5 eodr:49J Secondary: 57R55; 57R12, 57R10, 58A05, 49Q15, Primary: KONDO–TANAKA ϑ 1. ϑ ∈ Introduction W r dnie with identified are IA LI SIRAN loc 1 , 2 . 1 mohstructure? smooth a t faLpcizmnfl sequivalent is manifold Lipschitz a of y g manifolds. ogy t ouedniy n h integra- the and density, volume ate aioduigtento fa“mea- a of notion the using manifold ms, oncin ewe h theory the between connections ipeytfnaetlquestion: fundamental yet simple e m ulvn[7 8 9 initiated 29] 28, [27, Sullivan em. a forms flat pet nti iecliaein culminate line this in opments ’ hoyo o-mohanalysis non-smooth of theory e’s .Ruhyspeaking, Roughly ]. 6,Siaa[4 5,Mie[20] Moise 25], [24, Shikata 26], ,Apoiaino Lipschitz of Approximation a, F olna Anal. Nonlinear ϑ nihsit h tutrsof structures the into insights ennnRcmn[4 and [14] Heinonen–Rickman . a 3]adCin 2 (also [2] Cairns and [31] ead eto ttehato dif- of heart the at uestion easml e proof new simple a ve eter,a nltcap- analytic an theory, re e yKnoTnk to Kondo–Tanaka by ted hc eeintroduced were which , ons isht Maps; Lipschitz Points; r 155 (2017), ϑ at salocal a is [19], r lence 52, ([4, 5]), originally developed for applications in optimisation and control theory, to the approximation of Lipschitz maps by diffeomorphisms. The Main Theorem 1.3 in [17] is as follows:

Theorem 1.1. Let M be an n-dimensional compact Riemannian manifold, and let N be an ν-dimensional Riemannian manifold, where 2 ≤ n ≤ ν. Then, a Lipschitz map F : M → N is approximable by smooth immersions if singCl F = ∅.

singCl F denotes the singular set of F in the sense of Clarke [4, 5]; see Deﬁnition 2.2. The rigorous deﬁnitions for relevant geometric-analytic notions shall be given in §2. The proof in [17] may be viewed as an intricate generalisation of the classical arguments by Grove–Shiohama [12]. In this note we present a simple, new proof of Theorem 1.2, which also establishes its converse at the same strike. Our proof is based on the geometric measure theoretic studies on the smoothability problem (see [27, 28, 29, 14, 15, 13]). In particular, we make crucial use of the results due to Heinonen–Keith [13]. We hope it may provide an avenue for further explorations on the linkages between Clarke’s non-smooth analysis [4, 5] and geometric measure theory. In summary, we shall prove:

Theorem 1.2. In the setting of Theorem 1.1, the Lipschitz map F : M → N is approximable by smooth immersions if and only if singCl F = ∅.

Alternative analytic approaches, besides the geometric measure theoretic and non-smooth analytic ones, have also been developed to study the smoothability problem; see Ball–Zarnescu [1] and the references cited therein. This paper and the subsequent developments also address the applications of manifold smoothability theorems to the modelling and analysis of liquid crystals.

2. Background

In this section we brieﬂy discuss some preliminary materials on non-smooth analysis and geometric measure theory. For comprehensive treatments, we refer to [4, 5, 17] on the former topic and to [8, 30, 6] on the latter.

Smoothability. A topological manifold M is said to possess a ℘-structure (℘ ∈ {Lipschitz, k,α ∞ C , smooth=C , analytic...}) if and only if there is an atlas {(Uα, φα) : α ∈ I} of M such that all the transition maps −1 Rn Rn φα ◦ φβ : ⊃ φβ(Uα ∩ Uβ) −→ φα(Uα ∩ Uβ) ⊂ have ℘-regularity. By deﬁnition, a topological manifold has a C0-structure. A ℘-manifold M is said to be smoothable if there is a sub-atlas with respect to which M admits a C∞-structure.

Cartan–Whitney presentation. Let O ⊂ Rν be an open subset. A k-form ω is said to be a (Whitney) ﬂat k-form if and only if ω has measurable coeﬃcients and

ω, dω ∈ L∞(O).

Here dω is understood in the weak (i.e., distributional) sense. In what follows the definition of Cartan–Whitney presentations will be given. Let us first recall a prototypical case: Consider an n-dimensional topological manifold M, and let U be an open neighbourhood of some point p ∈ M with a well-defined orientation. Then there exists a 2 local coframe {θ1,...,θn} ∈ Γ(T ∗U) such that the differential n-form θ1 ∧ . . . ∧ θn agrees with the orientation and is nondegenerate: for a constant c> 0,

1 n θ ∧ . . . ∧ θ dVgM ≥ c VolgM (U) > 0. (2.1) ZU Here and throughout, for a fibre bundle E over M, Γ(E) denotes the space of its sections. Also, dVgM is the volume measure induced by gM . The notion of Cartan–Whitney presentations serves as a generalisation of the n-form θ1 ∧ . . . ∧θn in the above. Let X be a metric space that is a “nice” n-dimensional subset of Rν, ν ≥ n. Let U be an open neighbourhood of a fixed point p in X. A (local) Cartan–Whitney presentation ν near p consists of an n-tuple ρ = (ρ1, . . . , ρn) of flat 1-forms defined in an R -neighbourhood O of p, such that O∩ X ⊂ U and

′ ⋆ (ρ1 ∧ . . . ∧ ρn) ≥ c > 0 a.e. on O∩ X (2.2) for some constant c′ > 0. What does it mean by “nice” for X? On one hand, in the above definition ρ is defined on O ⊂ Rν, so we have to ensure that its local restrictions to X make sense. On the other hand, U ⊂ X needs to have a good sense of orientation, so that the Hodge-star in (2.2) is well-defined. Indeed, by the work [23] of Semmes, the following conditions ensure that X is nice enough to make sense of the above definition of Cartan–Whitney presentation:

X is a locally Ahlfors n-regular, locally linearly contractible homology n-manifold, n ≥ 2. (2.3)

Here, X is locally Ahlfors n-regular if and only if X has Hausdorﬀ dimension n, and for every compact K ⋐ X there exist numbers rK > 0, CK ≥ 1 such that

−1 n n n CK r ≤H (B(x, r)) ≤ CK r for each metric ball B(x, r) ⊂ X with x ∈ K and r < rK. X is locally linearly contractible ⋐ ′ ′ if and only if for every compact K X there exist numbers rK > 0, CK ≥ 1 such that every ′ ′ metric ball B(x, r) ⊂ X with x ∈ K and r < rK contracts to a point inside B(x,CKr). Finally, X is homology n-manifold if and only if it is separable, metrisable, locally compact, locally contractible, and that for each x ∈ X the following identity on homology groups holds:

n n H•(X, X ∼ {x}; Z) =∼ H•(R , R \ {0}; Z). (2.4)

′ ′ The quadruple (CK , rK; CK , rK ) is called the local data of X on K. One may refer to §§1– 3 in Heinonen–Keith [13] for detailed discussions. The punchline is: a local Cartan–Whitney presentation ρ can be deﬁned on X ⊂ Rν with weak regularity as in (2.3).

Sobolev space. Next let us deﬁne the Sobolev space W 1,2 on X satisfying (2.3): it is the norm completion of Lipschitz functions φ : X → R with respect to

kφk1,2 := kφkL2(X) + kapDφkL2(X), where the approximate diﬀerential apDφ is a.e. deﬁned on X as in 3.2.19 of Federer [8].

Measureble cotangent structure; the theorem of Heinonen–Keith on smoothability. Let X be as in (2.3). A result due to Cheeger [3] implies that X is n-rectifiable. For any flat 1-form ω defined on a open subset O ⊂ Rν such that O∩ X =: U is non-empty, one can define 3 ∗ ∗ the restriction ω U as a map from U to Tx U. The space Tx U is viewed as a measurable section of T ∗U ⊂ T ∗Rν, i.e. the measurable cotangent bundle; see §3.4 in [13] and p.303, Theorem 9A in [30]. Furthermore, in the pioneering works [27, 28, 29] Sullivan introduced the notion of a measurable cotangent structure. It consists of a pair (E, ι), where E is an oriented rank-n Lipschitz vector bundle over X, and ι is a module map over Lip(X, R) from Lipschitz sections of E to flat

1-forms on X, such that the following holds: If σ1,...,σn : X → E are Lipschitz sections such that σ1 ∧ . . . ∧ σn determines the chosen orientation on E, then for every α (index of an oriented, trivialised atlas {Uα}) the ﬂat n-form

1 n ι(σ1)|α ∧ . . . ∧ ι(σn)|α = ταdx ∧ . . . ∧ dx satisﬁes

ess infK τα > 0 ∗ in each compact subset K ⋐ Uα. In practice, one considers E = T Uα as in the above paragraph. The main result in [13] can be summarised as follows: Theorem 2.1 (Theorems 1.1 and 1.2, [13]). (1) If X ⊂ Rν as in (2.3) admits a Cartan– 1,2 Rn Whitney presentation in Wloc (X), then it is locally bi-Lipschitz parametrised by . (2) An oriented Lipschitz manifold is smoothable if and only if it admits a measurable cotan- 1,2 gent structure with local frames in Wloc (X).

Non-smooth analysis. Let (X, dX ) and (Y, dY ) be metric spaces. For a function φ : X → Y , one can deﬁne its Lipschitz norm as usual with respect to the metrics dX , dY . As an example, let (M, gM ) and (N, gN ) be n- and ν-dimensional Riemannian manifolds, and denote by distM and distN the Riemannian distance functions on M, N induced by the Riemannian metrics, respectively. The Lipschitz norm of φ : M → N is

distN Äφ(x), φ(y)ä kφkLip(M,N) := sup . x6=y, x,y∈M distM (x,y)

By Rademacher’s theorem, at dVgM -a.e. point x ∈ M a Lipschitz function φ : M → N has well-deﬁned diﬀerential dxφ : TxM → Tφ(x)N.

For φ : (X, dX ) → (Y, dY ) as above, its generalised diﬀerential is the set-valued function:

∂φ(x) := conv lim dxi φ : dxi φ exists and distM (xi,x) → 0 . i→∞ n o Here conv denotes the convex hull. Note that for any x ∈ X, each element of m ∈ ∂φ(x) can be identiﬁed with a matrix; thus we may introduce the following Deﬁnition 2.2. The singular set of φ à la Clarke ([4, 5]) is m singCl φ := x ∈ X : there exists ∈ ∂φ(x) that is not of the maximal rank . n o A function φ : X → Y is said to be approximable by smooth immersions if for any ǫ > 0 there exists a smooth immersion ιǫ : X → Y such that distN (φ(x), ιǫ(x)) ≤ ǫ for any x ∈ X, and that kιǫkLip(X,Y ) ≤ (1 + ǫ)kφkLip(X,Y ).

3. Proof

This section is devoted to the proof of Theorem 1.2. We shall establish the more general 4 Theorem 3.1. Let X ⊂ Rν be a locally Ahlfors n-regular, locally linearly contractible homological 1,2 n-manifold; ν ≥ n ≥ 2. Assume the existence of a Cartan–Whitney presentation in Wloc (X). ′ Let F : X → Rν be a Lipschitz map, ν′ ≥ n. Then the image F (X) is a smoothable topological n-manifold if and only if singCl F = ∅.

Proof for Theorem 3.1 ⇒ Theorem 1.2. Let M be an n-dimensional compact Riemannian manifold. It admits a Ck-isometric embedding into Rν, for k ≥ 3 and ν large enough, by Nash’s theorem [21]. Cover M by finitely many coordinate charts and fix one such chart U. Denote by {∂/∂x1,..., ∂/∂xn}⊂ Γ(TU) a local orthonormal frame on U, and let {dx1,...,dxn}⊂ Γ(T ∗U) be 1 n i the corresponding coframe. Clearly {dx ,...,dx } are flat 1-forms with sup1≤i≤n kdx kL∞(U) ≤ C < ∞ and d(dxi) = 0. Moreover, since dx1 ∧ . . . ∧ dxn is a volume n-form, hence Eq. (2.2) is 1 n ∞ 1,2 verified. Thus, {dx ,...,dx } constitute a Cartan–Whitney presentation in C (U) ⊂ Wloc (M). Furthermore, since F : M → N is Lipschitz and F (U) is precompact in the manifold N, by shrinking U if necessary, we can take F (U) lying in one single geodesic normal ball B on N. As the exponential map on B is a C∞-diffeomorphism, by composing with it we may assume ′ that F maps into the Euclidean space Rν . Finally, by the local nature of the statement of Theorem 1.2, it remains to show that under the assumption that F (U) satisfies (2.3), F |U is approximable by smooth immersions if and only if F (U) is smoothable. For the forward implication, we may utilise verbatim the arguments on p.32 in [13]; in particular, the proof of Eqs. (8.9) and (8.10) and an application of the results in ∞ [18, 27]. For the converse, one may pass to a sub-atlas of the C -structure and take ιǫ ≡ F . Thus Theorem 1.2 follows.

We are now ready to show Theorem 3.1. Heuristically, the key idea is that singCl F = ∅ prevents F from pinching necks.

Proof of Theorem 3.1. Fix an open neighbourhood U ⊂ X, on which there is a given Cartan– 1,2 ν′ Whitney presentation ρ = (ρ1, . . . , ρn) ∈ W (U). Let F : U → R be a Lipschitz map. Again, by the local nature of the statement, we may assume that F (U) is orientable and show that

F (U) is smoothable if and only if singCl (F |U)= ∅. In the sequel we view F as deﬁned on U. To this end, consider the pushforward n-tuple of ﬂat 1-forms:

F#ρ = ÄF#ρ1,...,F#ρnä.

In view of Theorem 2.1 (2), it then suffices to prove the equivalence between singCl F = ∅ and the following two conditions altogether: F#ρ defines a measurable cotangent structure à la Sullivan [27, 28, 29] on F (U), and that 1,2 Rν′ F#ρ ∈ Wloc ( ). (3.1) The Sobolev regularity condition (3.1) is automatic, as W 1,2-tensors are preserved under pushforward via Lipschitz functions. In the sequel, we show that singCl F = ∅ if and only if F#ρ yields a measurable cotangent structure, momentarily assuming that F (U) satisfies (2.3).

Let us first suppose singCl F = ∅ and deduce that F#ρ induces a measurable cotangent structure. By definition, for each x ∈ U, every element of the generalised differential ∂F (x) is of maximal rank. Denote by E the set of points on U where dF do not exist; Hn(E) = 0 by

Rademacher’s theorem. In addition, clearly a necessary condition for singCl F = ∅ is that the 5 ′ diﬀerential dF : TU → T Rν (deﬁned in the distributional sense; see §2) is invertible at Hn-a.e. point on U. Our goal is to show that

ess infF (U) ⋆ F#Äρ1 ∧ . . . ∧ ρnä ≥ c1 > 0 (3.2) under the assumption: ess infU ⋆ Äρ1 ∧ . . . ∧ ρnä ≥ c0 > 0. (3.3) Suppose (3.2) were false. Then, for any ǫ > 0 there would be a set Σ ⊂ U with Hn(Σ) > 0 and ⋆F#(ρ1 ∧ . . . ∧ ρn) < ǫ on Σ. Without loss of generality we may take Σ to be the metric ball B(p, 2r) ⊂ U with an Hn-null set Γ removed, such that Γ ⊃ B(p, 2r) ∩ E. After passing to subsequences if needed, one can ﬁnd a convergent sequence of points {qi} ⊂ Σ ∼ Γ such that qj → q ∈ B(p, r) and that det ÄdF (qj)ä < ǫ/c0, (3.4) where c0 is as in (3.3). Indeed, observe the identity

1 n ⋆F#(dx ∧ . . . ∧ dx ) = det dF ¶ ⋆ Äρ1 ∧ . . . ∧ ρnä© wherever dF is invertible; the determinant is well-defined as F is Lipschitz. Thus, in view of Eq. (3.3), we have 1 n ⋆F#(dx ∧ . . . ∧ dx ) ≥ c0 det dF outside an Hn-null set on U. Thus Eq.(3.4) follows. However, in the limits of ǫ ց 0 and j = j(ǫ) ր ∞, Eqs. (3.3), (3.4) and the rank-nullity theorem imply (via a diagonalisation argument) that dF cannot be of maximal rank at the limiting point q. Thus q ∈ singCl F , which yields a contradiction. Conversely, assume that F is not approximable by smooth immersions; we shall find a ′ ′ E point q in singCl F . By definition, it suffices to find a sequence {qj} ⊂ U ∼ such that ′ ′ ′ qj → q and that limjր∞ dF (qj) has rank less than n. Indeed, again due to Theorem 2.1, the non-approxmability of F leads to

ess infF (U) ⋆ F#Äρ1 ∧ . . . ∧ ρnä = 0. (3.5) In view of the Lipschitzness of F and the precompactness of U, the essential inﬁmum in Eq. (3.5) ′ ′ is attained at a point. That is, for some q ∈ U there holds ⋆F#(ρ1 ∧ . . . ∧ ρn)ÄF (q )ä = 0. ′ E Moreover, for any ǫ > 0, there exists a sequence {qj} ⊂ U ∼ depending possibly on ǫ, such ′ ′ ′ ′ that qj → q (hence F (qj) → F (q )) and that ′ ⋆F#(ρ1 ∧ . . . ∧ ρn)ÄF (qj)ä < ǫ.

Since (ρ1, . . . , ρn) is a Cartan–Whitney presentation on U, by Eq. (3.3) we may assume that all ′ the qj chosen above satisfy ′ ⋆Äρ1 ∧ . . . ∧ ρnä(qj) ≥ c0 > 0. By an analogous estimate as for Eq. (3.8), we can again bound

′ ǫ det dF (qj) ≤ /c0. ′ In particular, the determinant of dF are well-deﬁned at each point qj. By sending ǫ ց 0, j = j(ǫ) ր ∞ and using a standard diagonalisation argument, we ﬁnd that any pointwise

6 m ′ subsequential limit of dF (qj) veriﬁes rank m ≤ n − 1.

′ ′ ′ So, as {qj} converges to q , we immediately obtain that q ∈ singCl F . We are now left to prove that F (U) satisﬁes the structural assumptions in (2.3). For this purpose, we shall make crucial use of Eq. (3.2) established above. Indeed, by the Lipschitzness of F let us rewrite Eqs. (3.3) and (3.2) as

c0 ≤ ess infU ⋆ Äρ1 ∧ . . . ∧ ρnä ≤ ess supU ⋆ Äρ1 ∧ . . . ∧ ρnä ≤ C0, (3.6)

c1 ≤ ess infU ⋆ F#Äρ1 ∧ . . . ∧ ρnä ≤ ess supU ⋆ F#Äρ1 ∧ . . . ∧ ρnä ≤ C1. (3.7) where C0 depends on the ﬂat norm of ρ, and C1 additionally on the Lipschitz norm of F . Indeed, Eqs. (3.6)(3.7) imply that c C 0 < λ := 1 ≤ det(dF ) ≤ 1 =: Λ < ∞ Hn − a.e. on U. (3.8) C0 c0

As a result, given any metric ball B(x, r) ⊂ F (U), we can ﬁnd radii 0 < r− < r+ such that the “ following inclusions hold outside at most a Hn-null set:

F ÄB(x, r−)ä ⊂ B(x, r) ⊂ F ÄB(x, r+)ä. (3.9) “ Here and hereafter, we always use B(•, •) to denote the metric balls in F (U); the notation B(•, •) “ is reserved for the metric balls in U.

To proceed, notice that one may take r± to be equal to r modulo a multiplicative factor depending only on Λ and λ. Thus, by the area formula ([8], 3.2.20), there exists a constant

0 < c3 < ∞ depending only on Λ, λ, the Lipschitz norm of F and the local data of X such that n n n c3r ≤ H (B(x, r)) ≤ c3r . This gives the local n-Ahlfors regularity of F (U). The local linear contractibility of F (U) follows similarly from (3.9). Finally, Let B ⊂ U be the set on which Eq. (3.8) fails. Then Eq. (2.4) clearly holds on

U ∼ B, since F is a homeomorphism thereon and hence leaves H•(X, X ∼ {∗}; Z) invariant. For ∗∈ B, by excision we have

H• F (U) ∪ {∗} ∼ B, F (U) ∼ B; Z =∼ H• F (U), F (U) ∼ {∗}; Z . Utilising the facts that F is Lipschitz (hence continuous) on U, that F is a bi-Lipschitz homeomorphism onto F (U) ∼ B by Eq. (3.8), and that U is a Lipschitz n-manifold modulo reparametri- sations (thanks to Theorem 2.1 (1)), we deduce that F (U) ∪ {∗} ∼ B deformation retracts onto

F (U) ∼ B. This allows us to compute the relative homology from the reduced homology H•: ‹ B B B Z ∼ F (U) ∪ {∗} ∼ Z ∼ Rn Rn Z H• F (U) ∪ {∗} ∼ , F (U) ∼ ; = H• ; = H•Ä ; ∼ {0}; ä. ‹ F (U) ∼ B ! Hence F (U) is a homology n-manifold. The proof is now complete.

Acknowledgement. This work has been done during Siran Li’s stay as a CRM–ISM postdoc- toral fellow at Centre de Recherches Mathématiques, Université de Montréal and Institut des Sciences Mathématiques. The author would like to thank these institutions for their hospitality. 7 References

[1] J. M. Ball, A. Zarnescu, Partial regularity and smooth topology-preserving approximations of rough domains, Calc. Var. 56 (2017), Art. 13, 32 pp. [2] S. S. Cairns, Homeomorphisms between topological manifolds analytic manifolds, Ann. of Math. 41 (1940), 796–808. [3] J. Cheeger, Differentiability of Lipschitz functions on metric measure spaces, Geom. Funct. Anal. 9 (1999), 428–517. [4] F. H. Clarke, Generelized gradients and applications, Trans. Amer. Math. Soc. 205 (1975), 247–262. [5] F. H. Clarke, Optimization and Nonsmooth Analysis, in: Canadian Mathematical Society Series of Mono- graphs and Advanced Texts, John Wiley and Sons Inc., New York, 1983. [6] G. David and S. Semmes, Fractured Fractals and Broken Dreams: Self-similar Geometry Through Metric and Measure, Oxford Lecture Series in Mathematics and its Applications, vol. 7, Clarendon, Oxford, 1997. [7] S. K. Donaldson and D. P. Sullivan, Quasiconformal 4-manifolds, Acta Math. 163 (1989), 181–252. [8] H. Federer, Geometric Measure Theory, Die Grundlehren der mathematischen Wissenschaften, vol. 153, Springer, New York, 1969. [9] M. H. Freedman, The topology of four-dimensional manifolds, J. Differ. Geom. 17 (1982), 357–453. [10] M. Gromov, Metric Structures for Riemannian and Non-Riemannian Spaces, Birkhäuser, Boston, MA, 1999. [11] M. Gromov, Manifolds: Where do we come from? What are we? Where are we going? In: The Poincaré Conjecture, pp.81–144, Clay Math. Proc., 19, Amer. Math. Soc., Providence, RI, 2014. [12] K. Grove and K. Shiohama, A generalized sphere theorem, Ann. of Math. 106 (1977), 201–211. [13] J. Heinonen and S. Keith, Flat forms, bi-Lipschitz parametrizations, and smoothability of manifolds, Publ. Math. de l’IHÉS 133 (2011), 1–37. [14] J. Heinonen and S. Rickman, Geometric branched covers between generalized manifolds, Duke Math. J. 113 (2002), 465–529. [15] J. Heinonen, D. Sullivan, On the locally branched Euclidean metric gauge, Duke Math. J. 114 (2002), 15–41. [16] R. C. Kirby and L. C. Siebenmann, Foundational Essays on Topological Manifolds, Smoothings, and Trian- gulations, Ann. of Math. Stud., vol. 80, Princeton University Press, Princeton, 1977. With notes by John Milnor and Michael Atiyah. [17] K. Kondo and M. Tanaka, Approximation of Lipschitz maps via immersions and differentiable exotic sphere theorems, Nonlinear Anal. 155 (2017), 219–249. [18] A. A. G. Michael, On the smoothing problem, Tsukuba J. Math. 25 (2001), 13–45. [19] J. Milnor, On manifolds homeomorphic to the 7-sphere, Ann. of Math. 64 (1956), 399–405. [20] E. E. Moise, Geometric topology in dimensions 2 and 3. Graduate Texts in Mathematics, vol. 47. Springer- Verlag, New York-Heidelberg, 1977. x+262 pp. [21] J. F. Nash, Jr., The imbedding problem for Riemannian manifolds, Ann. of Math. 63 (1956), 20–63. [22] C. Pugh, Smoothing a topological manifold, Topol. Appl. 124 (2002), 487–503. [23] S. Semmes, Finding curves on general spaces through quantitative topology, with applications to Sobolev and Poincaré inequalities, Selecta Math. 2 (1996), 155–295. [24] Y. Shikata, On a distance function on the set of differentiable structures, Osaka J. Math. 3 (1966), 65–79. [25] Y. Shikata, On the smoothing problem and the size of a topological manifold, Osaka J. Math. 3 (1966), 293–301. [26] J. R. Stallings, The piecewise-linear structure of Euclidean space, Proc. Cambridge Philos. Soc. 58 (1962), 481–488. [27] D. Sullivan, Exterior d, the local degree, and the smoothability, in: Prospects in Topology (Princeton, NJ, 1944), Ann. of Math. Stud., vol. 138, pp.328–338, Princeton Univ. Press, Princeton, 1995. [28] D. Sullivan, The Ahlfors–Bers Measurable Riemann Mapping Theorem for Higher Di- mensions, Lecture at the Ahlfors celebration, Stanford University, September 1997, http://www.msri.org/publications/ln/hosted/ahlfors/1997/sullivan/1/index.html. [29] D. Sullivan, On the foundation of geometry, analysis, and the differentiable structure for manifolds, in Topics in Low-Dimensional Topology (University Park, PA, 1996), pp. 89–92, World Sci. Publishing, River Edge, 1999. [30] H. Whitney, Geometric Integration Theory, Princeton University Press, Princeton, 1957.

8 [31] J. H. C. Whitehead, Manifolds with transverse ﬁelds in Euclidean space, Ann. of Math. 73 (1961), 154–211.

Siran Li: Department of Mathematics, Rice University, MS 136 P.O. Box 1892, Houston, Texas, 77251-1892, USA • Department of Mathematics, McGill University, Burnside Hall, 805 Sherbrooke Street West, Montreal, Quebec, H3A 0B9, Canada. E-mail address: [email protected]