![Arxiv:1907.02372V2 [Math.AP]](http://data.docslib.org/img/2b5eca4ba990be9b0f71d1e7bfea7de5-1.webp)
arXiv:1907.02372v2 [math.AP] 27 Jun 2021 h upr fteKu n lc alnegFudto,Pro Foundation, Wallenberg Alice and Knut the of support the nry e o ntne[S0 r7] h aitoa app variational inequa The variational Fre72]. of [KS00, theory instance the for through see minimiz studied is energy, This that be function. and can given a values that above lying boundary of given whic constraint the asks with under It function [PSU12]. a monogra m al. the about the et instance Petrosyan among for and is see [Fri88], Friedman Problems, problem Boundary obstacle Free the of field setting, Euclidean the In 11 ∆ (1.1) the to solutions butional nsm oano taie group stratified a of domain some on ..akolde h upr fteUiest fPs,Pro Pisa, of University the of support the acknowledges V.M. phrases. and words Key 2020 h anqeto ecnie nti ae steotmlint optimal the is paper this in consider we question main The ahmtc ujc Classification. Subject Mathematics BTCETP RBESFRTESUB-LAPLACIAN THE FOR PROBLEMS OBSTACLE-TYPE Abstract. hr:i a ese stesblitccutrato the of counterpart subelliptic the Shahgholian. as and seen Lindgren S be Andersson, and can Garofalo it Danielli, by sharp: established previously was that eks euaiycniinon condition regularity weakest hr ∆ where ercvrthe recover we pca ntneo u eut,when results, seed relate our also first of are the and instance Sakai special setting, of a paper Euclidean 1991 the classical in the present already In Stein. and h udmna ouino ∆ of solution fundamental the PIA EUAIYO OUIN ONO-SIGN TO SOLUTIONS OF REGULARITY OPTIMAL H eoe h u-alca prtri taie ru.We group. stratified a in operator sub-Laplacian the denotes eetbihteoptimal the establish We C H 1 AETN ANN N NRA MINNE ANDREAS AND MAGNANI VALENTINO , 1 u-alca,Osal rbe,Sblitcequations, Subelliptic problem, Obstacle Sub-Laplacian, euaiyo ouin oteosal rbe nstratifie in problem obstacle the to solutions of regularity osg btcetp problem obstacle-type no-sign H The . f 1. aeytegopconvolution group the namely , ∆ Introduction 52;35R35. 35H20; H C H C G u u H 1 u , H 1 1 snneaieadstse h bv equation above the satisfies and nonnegative is e eto o oainadterminology. and notation for 2 Section see , = = , 1 euaiyi nesodi h es fFolland of sense the in understood is regularity 1 fχ neirrglrt fsltosto solutions of regularity interior fχ { u { u 6=0 6=0 } , } etKW2015.0380. KAW ject etPA21 9 ..acknowledges A.M. 49. 2018 PRA ject C la u euaiyrsl is result regularity Our alsa. oqartr oan.As domains. quadrature to d fteaoepolmare problem above the of s 1 h lsia btceproblem, obstacle classical the , 1 rpriscnb deduced be can properties h oc,atrsbrcigthe subtracting after roach, euaiyrsl u to due result regularity f h yRdius[Rod87], Rodrigues by phs iis sn h Dirichlet the using lities, ∗ s tde oisi the in topics studied ost ro euaiyo distri- of regularity erior is Γ steDrcltenergy, Dirichlet the es C H 1 , 1 taie groups. Stratified hr is Γ where , suethe assume groups, d OPTIMAL REGULARITY FOR SUBELLIPTIC NO-SIGN OBSTACLE-TYPE PROBLEMS 2 obstacle from the solution, leads to the following PDE formulation of the problem
∆u = fχ{u>0} in B1,
(1.2) u 0 in B1, ≥ u = g on ∂B1, where B1 denotes the metric unit ball with respect to the Carnot–Carathéodory distance (Definition 2.1). Our problem is a non-variational counterpart of (1.2), that is ∆u = fχ in B , (1.3) {u6=0} 1 u = g on ∂B . 1 We point out that (1.3) — which is called a no-sign obstacle-type problem — naturally appears also when considering the so-called quadrature domains [Sak91, GS05]. Two important questions on this problem concern the regularity of solutions to (1.3), and the regularity of the free boundary. In Euclidean space, the analysis of both ques- tions is essentially complete [Sak91, CKS00, PS07, ALS13]. In particular, in relation to the regularity of solutions, Andersson et al. [ALS13] show that u has the optimal C1,1 regularity if the linear problem ∆v = f has a C1,1 solution. This is the minimal regularity assumption on f in order to establish the C1,1 regularity of solutions. The main result of this paper is the sharp regularity of solutions to (1.1) also in the subelliptic setting of stratified groups.
1,1 ∞ Theorem 1.1 (CH regularity). Let u L (B1) be a distributional solution to (1.1) in R ∈ 1,1 the unit ball B1. Let f : B1 be locally summable such that f Γ CH (B1). Then there exists a universal constant→ C > 0 such that, after a modification∗ on∈ a negligible set, we have u C1,1(B ) and ∈ H 1/4 2 2 (1.4) D u ∞ C D (f Γ) ∞ + u ∞ . k h kL (B1/4) ≤ k h ∗ kL (B1) k kL (B1) 1,1 1,1 In our setting, the natural counterpart of the Euclidean C regularity is the CH regularity, where the horizontal derivatives are required to be Lipschitz continuous (Defi- nition 2.2). The function Γ denotes the fundamental solution of∆H (Definition 2.3). For further notation and terminology, we address the reader to Section 2. We wish to emphasize that u satisfies (1.1) also in the strong sense. Indeed, the distri- 1,1 butional equality ∆H (f Γ) = f joined with the assumed CH regularity of f Γ show ∞ ∗ − ∞ ∗ that f L (B1). Therefore fχ{u6=0} L (B1) and by the regularity result of Folland, ∈ 2,p∈ 1,1 [Fol75, Theorem 6.1], we get u WH,loc(B1) for every 1 p < . The CH regularity of solutions to the obstacle problem∈ in stratified groups was≤ obtained∞ by Danielli et al. [DGS03], using the variational formulation of the problem. The regularity of the free boundary was subsequently established in step two groups [DGP07]. Further results in this area have been obtained for Kolmogorov operators and parabolic non-divergence form operators of Hörmander type [DFPP08, FNPP10, FGN12, Fre13]. The no-sign obstacle- type problem in terms of the equation (1.1) does not seem to have been considered before in the subelliptic setting. OPTIMAL REGULARITY FOR SUBELLIPTIC NO-SIGN OBSTACLE-TYPE PROBLEMS 3
Our arguments are remarkably different from the ones used for the obstacle problem. For instance, in this problem without a forcing term the solution is automatically super- harmonic with respect to ∆H , while in our setting we have no such sign condition that would yield a superharmonic solution. We initiate our analysis observing that second or- der horizontal derivatives of solutions to (1.1) satisfy certain BMO estimates, that have been established by Bramanti et al. [BB05, BF13]. The subsequent step is to construct suitable approximating polynomials, starting from the second order horizontal derivatives of the solution. Indeed these polynomials yield a subquadratic growth estimate (4.10) at small scales. We point out that this estimate is valid for any bounded and W 2,p regular function, with bounded sub-Laplacian, so it might be of independent interest. As a con- sequence, we perform a suitable rescaling of the equation and then infer the crucial decay estimate of the measure of the coincidence set (Proposition 4.6), when the horizontal Hes- sian of the approximating polynomial is sufficiently large. More details on this procedure can be found at the beginning of Section 4. Although our ideas mainly follow the path set up by Andersson et al. [ALS13], and Figalli and Shahgholian [FS12], there are several difficulties related to the subelliptic set- ting. The basic one is concerned with the fact that the sub-Laplacian ∆H is degenerate elliptic. In addition, since the operator ∆H is written in terms of Hömander vector fields, we can only consider the horizontal Hessian (2.5) of the solution, that is a nonsymmetric matrix. Then the construction of the approximating polynomials starting from the av- erage of the second order noncommuting derivatives XiXju becomes more delicate and requires some preliminary algebraic work, see Section 3. Notwithstanding the technical complications, the proof has become more streamlined: we can stay clear of the projection operator used in [ALS13], and this simplifies several technical points. A suitable quan- titative decay estimate of the zero-level set (4.14) can be obtained also in our setting. Finally, we adapt Caffarelli’s polynomial iteration technique of [Caf89] to find explicit estimates of the second order horizontal derivatives, see (1.4). We finish the introduction by giving an overview of the paper. In Section 2 we introduce some basic notions on stratified groups and the related function spaces. In Section 3 we construct suitable second order homogeneous polynomials (Definition 3.2), that have an assigned horizontal Hessian (Corollary 3.3). Then some important W 2,p and BMO estimates are presented. Finally, we provide the crucial scaling estimates of Lemma 3.8. In Section 4 we prove a subquadratic growth estimate of the difference between a solution and its approximating polynomial. Then we apply the subquadratic growth estimate to 1,1 get a suitable decay of the measure of the zero-level set. Finally, we establish the CH regularity in quantitative terms, according to (1.4). Acknowledgements. A substantial part of this work was accomplished during the time A.M. spent at Scuola Normale Superiore in Pisa, with the support of the Knut and Alice Wallenberg Foundation. V.M. gratefully thanks Henrik Shahgholian, along with KTH Royal Institute of Technology, for support and hospitality during December 2018, where important parts of the present work were carried out. He also acknowledges the Institutional Research Grant from the University of Pisa. The authors wish to thank OPTIMAL REGULARITY FOR SUBELLIPTIC NO-SIGN OBSTACLE-TYPE PROBLEMS 4
Marco Bramanti for fruitful discussions concerning some a priori BMO estimates. They are also indebted to the referee for several useful remarks.
2. Basic facts and notation A stratified group is a simply connected, real nilpotent Lie group G, whose Lie algebra has a special stratification. We denote by V the subspaces of , having the properties: G i G = V V V and [V ,V ]= V G 1 ⊕ 2 ⊕···⊕ ι 1 j j+1 for j =1,...,ι and Vι+1 =0. Let us denote by n the topological dimension of G and by m the dimension of V1. We choose a graded basis X1,X2,...,Xn of , that is characterized by the property that G
Xmj−1+1,...,Xmj is a basis of Vj for all j = 1,...,ι, where we have set m0 = 0, m1 = m and mj = j i=1 dim Vi. We notice that mι = n and with these definitions, if mk−1 < j mk, then k N is uniquely determined and we define the positive integer ≤ P∈ (2.1) dj := k. Through the exponential mapping of G, one can construct a diffeomorphism from Rn n to G. Hence we have defined a graded basis e1, e2,...,en of R and graded coordinates x1, x2,...,xn that define the point x =(x1, x2,...,xn) of G. This allows us to identify G with Rn, as it will be understood in the sequel. In addition, one may also verify that the Lebesgue measure of Rn through the graded co- ordinates yields the Haar measure of the group G. The notation A denotes the Lebesgue measure of a measurable set A Rn. | | The diffeomorphism associated⊂ to graded coordinates has also the property that the group operation on G, when read in Rn, is given by a special polynomial group operation (2.2) xy = x + y + BCH(x, y), where the precise form of the vector polynomial BCH : Rn Rn Rn is given by the important Baker–Campbell–Hausdorff formula, in short BCH× formula,→ see for instance [Var84]. The degree of xj is the integer dj defined in (2.1) and we define intrinsic dilations as follows n 2 2 ι ι dj δrx =(rx1,...,rxm,r xm+1,...,r xm2 ,...,r xmι−1+1,...,r xn)= r xjej jX=1 for any r> 0. The notion of degree fits the algebraic properties of dilations, since
(2.3) δr(xy)=(δrx)(δry) for all x, y Rn. By the form of dilations, for every measurable set A Rn we have ∈ ⊂ δ (A) = rQ A | r | | | n G for all r> 0, where Q = j=1 dj can be proved to be the Hausdorff dimension of . P OPTIMAL REGULARITY FOR SUBELLIPTIC NO-SIGN OBSTACLE-TYPE PROBLEMS 5
The metric structure of Rn is given by a control distance. We say that γ : [0, T ] Rn, an absolutely continuous curve, is admissible if for a.e. t [0, 1] there holds → ∈ m γ˙ (t)= bi(t)Xi(γ(t)) Xi=1 m 2 and i=1 bi(t) 1. We denote by (x, y) the family of all admissible curves whose image ≤ H Rn containsP x, y. By Chow’s theorem, (x, y) is nonempty for every x, y , hence the following “control distance” H ∈ d(x, y) = inf T > 0 γ : [0, T ] Rn, γ (x, y) → ∈ H is well defined. It is also possible to check that d is actually a distance, corresponding to the well known Carnot–Carathéodory distance. Since left translations preserve the “horizontal velocity”, d is also left invariant, namely d(x, y) = d(zx,zy) for all x, y, z Rn. Furthermore, dilations are Lie group homo- morphisms, hence the Carnot–Carathéodory∈ distance is homogeneous in the sense that d(δ x, δ y)= rd(x, y) for every x, y Rn and r > 0. r r ∈ n Definition 2.1 (Metric balls). For x R and r> 0, we denote by Br(x) the open ball of center x and radius r> 0 with respect∈ to d. Precisely, this is the set y Rn : d(x, y) From the properties of d and δr, it is easy to observe that Br(x)= xδr(B1). Dilations also allow us to introduce a natural notion of homogeneity, so we may say that a polynomial p : Rn R is k-homogeneous if → p(δ x)= rkp(x) forall x Rn and r > 0. r ∈ The number k N is the degree of p. Moreover, any vector field Xj of the fixed graded basis can be identified∈ with a first order differential operator of the form n (2.4) Xj = ∂xj + aji∂xi i=m +1 Xdj for every j = 1,...,n. The functions a : Rn R are homogeneous polynomials of ji → degree di dj 1 and in particular Xj(0) = ej for all j =1,...,n. In the sequel Ω will be understood− ≥ as an open bounded subset of G, that can be also identified with an open subset of Rn, if not otherwise stated. Given a function u : Ω R and considering the vector fields Xj as differential operators, we may introduce the horizontal→ gradient and the horizontal Hessian X1X1u X1X2u X1Xmu X X u X X u ··· X X u 2 2 1 2 2 ··· 2 m (2.5) hu =(X1u,...,Xmu) and Dhu = . . .. . , ∇ . . . . X X u X X u m 1 m m ··· ··· OPTIMAL REGULARITY FOR SUBELLIPTIC NO-SIGN OBSTACLE-TYPE PROBLEMS 6 respectively, whenever they are pointwise defined. More generally, we can define higher order differential operators considering for I =(i ,...,i ) Nn the function 1 n ∈ XI u := Xin Xi1 u. 1 ··· n Definition 2.2 (Folland–Stein spaces). Let Ω Rn be an open set. We denote by C1 (Ω) ⊂ H the space of all functions u : Ω R such that the horizontal derivatives Xju exist on → 1,α Ω for all j = 1,...,m and are continuous. If 0 < α 1, then CH (Ω) is the space of 1 ≤ functions u in CH (Ω) such that there exists C > 0 with the property that X f(x) X f(y) Cd(x, y)α | j − j | ≤ for every x, y Ω and j =1,...,m. ∈ 2 Notice that Dhu is not symmetric, since the vector fields Xj do not commute in general. We say that Xju are the horizontal derivatives and XiXju are the second order horizontal derivatives. The symmetrized horizontal Hessian is defined as 1 D2,su = D2u + D2uT . h 2 h h The sub-Laplacian is defined as m 2 ∆H u = Xj u. jX=1 Functions satisfying ∆H u = 0 are called as usual harmonic functions. Definition 2.3 (Fundamental solution). We say that Γ C∞(G 0 ) is a fundamental solution for ∆ if it is locally summable, it vanishes at infinity∈ and\{ satisfie} s ∆ Γ= δ , H H − 0 where δ0 denotes the Dirac distribution centered at the origin. 1/(2−Q) The fundamental solution Γ defines a gauge dG = Γ , that is 1-homogeneous with respect to dilations and continuous on Rn. We can readily check that there exists a constant c0 > 1 such that (2.6) c−1d (x) d(x, 0) c d (x) 0 G ≤ ≤ 0 G for all x Rn. Defining d (x, y) := d (x−1y) we also introduce the gauge ball ∈ G G (2.7) BG(x)= y Rn : d (x, y) Proof. We consider the function φ defined in [BLU07, (5.50e)], where we choose ϕ appear- ∞ −1 ing in the definition of φ, such that ϕ Cc (]3 , 1[), ϕ 0 and R ϕ(t)dt = 1. It follows that φ is smooth and bounded on Rn, along∈ with all of its≥ derivatives,´ and it is compactly G −Q supported in B1 , see the definition (2.7). We also consider φr(z) := r φ(δ1/rz), that is G ˆ −1 Rn compactly supported on Br . We finally set φr(z) := φr(z ) for all z and r > 0. Thus, using [BLU07, (5.50a),(5.50d)], for every x B , we get ∈ ∈ λ −1 −1 ϑ(x)= φh(x y)ϑ(y)dy = φˆh(y x)ϑ(y)dy. ˆ G ˆ Bh (x) Ω We can differentiate the last integral an arbitrary number of times, due to the smoothness of φˆh, getting the smoothness of ϑ and the following estimate I I ˆ −1 I ˆ ∞ X ϑ(x) = X φh(y x)ϑ(y)dy X φh L (Rn) ϑ L1(Ω) | | ˆ ≤k k k k Ω for all x Ω′. This concludes the proof. ∈ In our setting, we need the notion of Sobolev function adapted to the horizontal vector k,p fields X1,...,Xm, see [Fol75]. The horizontal Sobolev space WH (Ω) consists of those p functions u L (Ω) for which, for all js 1,...,m and s 1,...,k , there exists a function v ∈ Lp(Ω) such that ∈ { } ∈ { } j1,...,jk ∈ k u(y)(Xj1 Xjk φ)(y) dy =( 1) vj1,...,jk (y)φ(y) dy ˆΩ ··· − ˆΩ ∞ for any function φ Cc (Ω). Also in the more general setting of Hörmander vector fields some Sobolev embedding∈ theorems hold, see [GN96, Theorem 1.11 and (3.19)], or [Lu96, Theorem 1.1]. The next theorem specializes these embedding results for stratified groups. Theorem 2.5. Let p > Q, where Q is the Hausdorff dimension of G and let Ω′ ⋐ Ω be any open and relatively compact subset. Then there exists C > 0, depending on Ω′, such that for every u W 1,p(Ω), up to a modification of u on a negligible set, we have ∈ H Q 1− p u(x) u(y) C u W 1,p(Ω) d(x, y) , | − | ≤ k k H for every x, y Ω′. ∈ The following (1,1)-Poincaré inequality holds, (2.10) u(y) uBr(x) dy cr hu(y) dy ˆBr(x) | − | ≤ ˆBr(x) |∇ | 1 for every u C (Br(x)). This inequality follows from [Jer86], see also [LM00]. For any measurable∈ function u that is summable on a measurable set A Ω, we use the notation ⊂ 1 u := u(y) dy = u(y) dy. A A ˆ A | | A OPTIMAL REGULARITY FOR SUBELLIPTIC NO-SIGN OBSTACLE-TYPE PROBLEMS 8 Definition 2.6. For u L1(Ω), we define the BMO seminorms ∈ [u]BMO(Ω) := sup u(y) uBr(x0) dy, x0∈Ω,r>0 Br(x0)∩Ω | − | [u] := sup u(y) u dy BMOloc(Ω) Br(x0) Br(x0)⊂Ω Br(x0) | − | and for 1 p< the corresponding BMOp norms ≤ ∞ u p := [u] + u p , k kBMO (Ω) BMO(Ω) k kL (Ω) u BMOp (Ω) := [u]BMO (Ω) + u Lp(Ω). k k loc loc k k p p p p The spaces BMO (Ω) and BMOloc(Ω) consist of all L functions on Ω with finite BMO p and BMOloc norm, respectively. See [BF13] for more information on BMO functions in the subelliptic setting. 3. Preparatory results We first study the relationship between the coefficients of a 2-homogeneous polynomial 2,p and its second order horizontal derivatives. Then, by some WH and BMO estimates, we show how to control the horizontal Hessian of a Sobolev function by the horizontal Hessian of a suitable 2-homogeneous harmonic polynomial (Corollary 3.7). Finally, in Lemma 3.8 we establish a quantitative control on the growth of these polynomials at small scales. We need first to find 2-homogeneous polynomials with assigned second order horizontal derivatives. To do this, we first observe that (2.2), combined with (2.3) and (2.1), setting n BCH(x, y)= qj(x, y)ej, j=Xm+1 imply that any qj is a homogeneous polynomial of degree dj. Due to the BCH formula, one can also prove that ql is a 2-homogeneous polynomial with respect to the variables x1,...,xm,y1,...,ym for all l = m +1,...,m2 and q (x, y)= q (y, x). l − l From the definition of left invariant vector field, we get ∂ql ajl(x)= (x, 0), ∂yj for all j =1,...,m and l = m +1,...,m2. As a consequence, we get ∂a ∂2q ∂2q ∂a (3.1) jl = l = l = il ∂xi ∂xi∂yj −∂xj ∂yi − ∂xj for all i, j = 1,...,m and l = m +1,...,m2. Notice that the partial derivatives in the previous equalities are all constant functions. Equalities (3.1) will be important in the proof of Proposition 3.1. OPTIMAL REGULARITY FOR SUBELLIPTIC NO-SIGN OBSTACLE-TYPE PROBLEMS 9 n Every polynomial on R , thought of as equipped with dilations δr, is the sum of homoge- neous polynomials and the maximum among these degrees is the degree of the polynomial. Polynomials of degree one are just affine functions ℓ of the form ℓ(x)= α + β, π(x) h i with β =(β , β ,...,β ) Rn, α R and we have used the projection 1 2 m ∈ ∈ (3.2) π : Rn Rm, π(x)=(x ,...,x ). → 1 m A homogeneous polynomial of degree two must have the form 1 m m2 (3.3) p(x)= c x x + c x , 2 ij i j l l i,jX=1 l=Xm+1 where cij and cl are real numbers, with cij = cji for all i, j =1,...,m. Proposition 3.1. Let p : Rn R be a 2-homogeneous polynomial of the form (3.3) and → let us consider the basis Xm+1,...,Xm2 of V2. Then we have 1 m2 c = (X X p + X X p) and X X p = c + γl c , ij 2 i j j i i j ij ij l l=Xm+1 l where γij are proportional to the structure constants of the Lie algebra, namely m2 l (3.4) [Xi,Xj]= 2γijXl l=Xm+1 and i, j =1,...,m. Proof. We first define the symmetrized second order derivative X X + X X (X X )s := i j j i , i j 2 so that we can write 1 (3.5) X X =(X X )s + [X ,X ], i j i j 2 i j for every i, j = 1,...,m. Since XiXj and Xl are homogeneous differential operators of order 2 and p has degree 2, the horizontal derivatives XiXjp and Xlp are constants. By (3.3)− and (2.4), we get m n m2 m m2 X p = c x + a ∂ c x = c x + a c , j ji i ji xi l l ji i ji i Xi=1 i=Xm+1 l=Xm+1 Xi=1 i=Xm+1 for j =1,...,m. As a consequence, taking into account the form of the vector fields (2.4) and of the polynomial (3.3), for any i, j =1,...,m and l = m +1,...,m2, we get m2 (3.6) XiXjp = cij + ∂xi ajscs. s=Xm+1 OPTIMAL REGULARITY FOR SUBELLIPTIC NO-SIGN OBSTACLE-TYPE PROBLEMS 10 To establish the previous equality, we have also observed that the polynomials aji are homogeneous of degree d d = 1, therefore they are only depending on their first i − j m variables. In particular, all the partial derivatives ∂xl aji are vanishing whenever the integers l and i take values from m +1 to m2 and j = 1,...,m. Combining (3.6) and (3.1), we also obtain the first of the following equalities XiXjp + XjXip =2cij and Xlp = cl, with 1 i, j m and m +1 l m2. The latter directly follows from the form of (3.3). In conclusion,≤ ≤ by virtue of (3.4),≤ ≤ (3.5) and (3.6), we have obtained that m2 m2 s l l XiXjp =(XiXj) p + γijXlp = cij + γijcl, l=Xm+1 l=Xm+1 hence concluding the proof. Definition 3.2. For B (x ) ⋐ Ω and u W 2,1 (Ω), we define the matrix r 0 ∈ H,loc 1 P x0 :=(D2u) (∆ u) I Rn×n, r h Br(x0) − m H Br(x0) m ∈ 2 where Im stands for the identity matrix and the (i, j) entry of (Dhu)Br(x0) is the average (XiXju)Br(x0). Associated to the ball Br(x0), we also define the coefficients r,x0 XiXju + XjXiu 1 r,x0 cij := δij (∆H u)Br(x0) and cl =(Xlu)Br(x0). 2 Br(x0) − m These numbers define the 2-homogeneous polynomial 1 m m2 px0 (x)= cr,x0 x x + cr,x0 x , r 2 ij i j l l i,jX=1 l=Xm+1 x0 that we will show to be related to Pr . x0 Corollary 3.3. In the assumptions of Definition 3.2, the 2-homogeneous polynomial pr is harmonic and 2 x0 x0 Dhpr = Pr . Proof. By Proposition 3.1, we have m2 x0 r,x0 l r,x0 XiXjpr = cij + γijcl , l=Xm+1 l r,x0 where γii = 0 and by definition of cii we get m m x0 r,x0 1 ∆H p = c = (XiXiu) (∆H u)B (x ) =0. r ii Br(x0) − m r 0 Xi=1 Xi=1 OPTIMAL REGULARITY FOR SUBELLIPTIC NO-SIGN OBSTACLE-TYPE PROBLEMS 11 Finally, we observe that X X u + X X u 1 m2 X X px0 = i j j i δ (∆ u) + γl X u i j r ij H Br(x0) ij l 2 Br(x0) − m l=m+1 X Br(x0) XiXju + XjXiu 1 [Xi,Xj]u = δij (∆H u)Br(x0) + 2 Br(x0) − m 2 ! Br(x0) 1 =(XiXju) δij (∆H u)B (x ), Br(x0) − m r 0 m2 l having taken into account that [Xi,Xj]= l=m+1 2γijXl from Proposition 3.1. 2,p P The following WH estimates go back to the work of Folland [Fol75], see also the work by Bramanti and Brandolini [BB00] for more general hypoelliptic operators. Theorem 3.4 ([BB00]). Let 1 (3.8) X X u p C(σ, p) ∆ u p + u p . k i j kBMO (Bσ) ≤ k H kBMOloc(B1) k kBMOloc(B1) Remark 3.6. Note that the nonvariational form of the operator in [BF13, Theorem 2.10] needs a priori that the solution u and its horizontal derivatives are BMO. For our purposes, it is very important that the BMO regularity of u is established with no a priori assumptions. This can be obtained for the sub-Laplacian operator, since its distributional form allows us to apply a mollification argument. In the sequel, we will also use the Frobenius norm M for a matrix M of coefficients | | mij, setting M = m 2. | | | ij| sXij With this definition we easily notice that ∆ u D2u . | H |≤| h | Corollary 3.7. Let 1 X X u BMOp(B ) for i, j =1,...,m and whenever x B , 0 2 x (3.9) D u(y) P 0 dy C(σ, p) ∆ u ∞ + u p , | h − r | ≤ k H kL (B1) k kBMOloc(B1) Br(x0)∩Bσ x0 where the matrix Pr is introduced in Definition 3.2. p Proof. Theorem 3.5 immediately implies that XiXju BMO (Bσ) and (3.8) holds. Thus, we obtain the following estimates ∈ 2 x0 Dhu(y) Pr dy Br(x0)∩Bσ | − | 1 D2u(y) (D2u) dy + (∆ u) I h h Br(x0) H Br(x0) m ≤ Br(x0)∩Bσ | − | m| | 2 1 [D u] + ∆ u ∞ ≤ h BMO(Bσ) √mk H kL (B1) C(σ, p) ∆H u BMOp (B ) + u BMOp (B ) + ∆H u L∞(B ) ≤ k k loc 1 k k loc 1 k k 1 Ce(σ, p) ∆H u L∞(B ) + u BMOp (B ) . ≤ k k 1 k k loc 1 This completes the proof. 2 2 x0 Heuristically, if Dhu is not bounded around x0, since the difference of Dhu and Pr x0 + is controlled, the BMO estimate tells us that also Pr becomes unbounded as r 0 . x0 → Hence we will turn our attention to Pr . In the following lemma, we will derive a general “scaling estimate” for the difference P x0 P x0 . In particular, when r = 2r we get a | r1 − r2 | 2 1 uniform bound on the growth of P x0 on dyadic scales. | r | Lemma 3.8 (Scaling estimates). Let 1 Q x x r2 P 0 P 0 C(λ ,p) ∆ u ∞ + u p . | r1 − r2 | ≤ r 1 k H kL (B1) k kBMOloc(B1) 1 OPTIMAL REGULARITY FOR SUBELLIPTIC NO-SIGN OBSTACLE-TYPE PROBLEMS 13 Proof. x0 x0 Due to the BMO estimate (3.9) with σ = λ1, we can estimate Pr1 Pr2 as follows | − | P x0 P x0 dx | r1 − r2 | Br1 (x0)∩Bλ1 D2u(y) P x0 dy + D2u(y) P x0 dy ≤ | h − r1 | | h − r2 | Br1 (x0)∩Bλ1 Br1 (x0)∩Bλ1 2 x0 Br2 (x0) Bλ1 2 x0 Dhu(y) Pr1 dy + | ∩ | Dhu(y) Pr2 dy ≤ B (x )∩B | − | Br1 (x0) Bλ1 B (x )∩B | − | r1 0 λ1 | ∩ | r2 0 λ1 Q r2 C(λ1,p) 1+ ∆H u L∞(B ) + u BMOp (B ) . ≤ r k k 1 k k loc 1 1 The last inequality follows by taking into account our conditions on the radii r1 and r2. Indeed, we have B (x ) B B (x ) r Q | r2 0 ∩ λ1 | | r2 0 | = 2 . B (x ) B ≤ B (x ) r | r1 0 ∩ λ1 | | r1 0 | 1 Finally, with a slight abuse of notation, we denote the constant 2C(λ1,p) again by C(λ1,p) in the inequality of the lemma, concluding the proof. 1,1 4. Proof of CH regularity This section represents the core of the paper. We establish the sub-quadratic growth of the difference u(y) u(x ) u(x ), π(x−1y) px0 (x−1y) − 0 − h∇h 0 0 i − r 0 x0 on the ball Br(x0), where pr is the harmonic polynomial introduced in Definition 3.2. We 2 x0 show that when the norm of Dhpr is sufficiently large, then the measure of the coincidence set u = 0 decays in a quantitative way. This is one of the central facts, that leads us { } 1,1 to the dichotomy argument of [ALS13] to reach the CH regularity. There are indeed 2 x0 + two cases: (i) when Dhpr is uniformly bounded as r 0 , we immediately infer the | | → 2 x0 regularity from the subquadratic growth, (ii) if otherwise Dhpr grows without bound as r 0+, then the coincidence set is “small” and we show| that a| suitable adaptation of → 1,1 Caffarelli’s polynomial iteration technique can lead us to the CH regularity. In the sequel, whenever we consider a function u with essentially bounded sub-Laplacian 1 ∆H u, then it is understood that u is chosen to be of class CH . The following remark rigorously justifies this convention. Remark 4.1. Let Ω Rn be an open set and let u : Ω R be a locally summable ′ ⊂ ∞ → 2,p function such that ∆H u L (Ω). From [Fol75, Theorem 6.1], we have u WH,loc(Ω) for every p′ > 1. In view∈ of Theorem 2.5, by standard arguments, we can modif∈ y u on a 1,α ′ ′ ⋐ negligible set such that u CH (Ω ) for any relatively compact open set Ω Ω, where ′ ∈ Q we have fixed some p > Q and α =1 ′ . In particular, we have shown that, after the − p modification, u C1 (Ω). ∈ H OPTIMAL REGULARITY FOR SUBELLIPTIC NO-SIGN OBSTACLE-TYPE PROBLEMS 14 p ∞ Lemma 4.2 (Sub-quadratic growth). Assume u BMOloc(B ) such that ∆ u L (B ). ∈ 1 H ∈ 1 Let λ, σ (0, 1) and fix p > 1. Then there exist r0 > 0 and a universal constant C(λ,σ,p)∈> 0, such that for any x B and 0 x0 −1 ∞ p 2 sup u(y) pr (x0 y) C(λ,σ,p) ∆H u L (B1) + u BMOloc(B1) r . y∈B (x )| − | ≤ k k k k σr 0 Proof. We fix x B and λ′ = (1+λ)/2, so that for 0 2 2 1 Dhur,x0 L (B1) = B1 Dhur,x0 (x) dx k k | | B1 | | 2 x0 (4.2) = B1 Dhu(y) Pr dy | | Br(x0)| − | ∞ p C(λ,p) ∆H u L (B1) + u BMO (B ) . ≤ k k k k loc 1 Now we wish to apply the Poincaré inequality (2.10) to ur,x0 ℓr,x0 , where ℓr,x0 is an affine function to be properly defined. If we let − ℓ (x) :=(u ) + ( u ) , π(x) , r,x0 r,x0 B1 h ∇h r,x0 B1 i where π(x)=(x1,...,xm), it follows that u ℓ 1 c u ( u ) dx, r,x0 r,x0 L (B1) h r,x0 h r,x0 B1 k − k ≤ B1 ∇ − ∇ since the average over B1 of the linear part of ℓr,x0 is zero. Again, from the Poincaré inequality, using (4.2), we get 2 1 1 ur,x0 ℓr,x0 L (B1) C Dhur,x0 L (B1) (4.3) k − k ≤ k k C(λ,p) ∆H u L∞(B ) + u BMOp (B ) . ≤ k k 1 k k loc 1 x0 For the sequel, we setu ˆr,x0 := ur,x0 ℓr,x0 . Since both pr and ℓr,x0 are harmonic, we observe that − ∆H uˆr,x0 (x)=(∆H u)(x0δrx)= f(x0δrx) OPTIMAL REGULARITY FOR SUBELLIPTIC NO-SIGN OBSTACLE-TYPE PROBLEMS 15 for a.e. x B , where we have set f := ∆ u L∞(B ). We set g (x) = f(x δ x)χ ∈ 1 H ∈ 1 r,x0 0 r B1 and we consider the decompositionu ˆr,x0 =v ˆr,x0 +w ˆr,x0 , where vˆ = g Γ andw ˆ =u ˆ + g Γ r,x0 − r,x0 ∗ r,x0 r,x0 r,x0 ∗ and Γ is the fundamental solution for ∆H , introduced in Definition 2.3. The explicit form ofv ˆr,x0 allows us to get the estimate −1 −1 Q vˆr,x0 (x) = Γ(z x)gr,x0 (z)dz = Γ(z x)gr,x0 (z)dz C gr,x0 L 0 (B1) | | ˆRn ˆ ≤ k k B1 for every x B , where Q = Q + 1 and C > 0 can be seen as a universal constant. The ∈ 1 0 previous estimate follows from the Hölder inequality, setting q = Q0/Q and taking into account the (2 Q)-homogeneity of Γ. Indeed, it holds − 1/q −1 q (4.4) Γ(z x)gr,x0 (z)dz Γ gr,x0 LQ0 (B ) ˆ ≤ ˆ | | ! k k 1 B1 B2 for every x B . As a consequence, we have proved that ∈ 1 (4.5) vˆ ∞ C ∆ uˆ Q . k r,x0 kL (B1) ≤ k H r,x0 kL 0 (B1) Sincew ˆr,x0 is harmonic, from [BLU07, (5.52)] we have the mean value type formula −1 wˆr,x0 (x)= Ψ(x z)ˆwr,x0 (z) dz G B − (x) (1 σ)/c0 for any x Bσ, whenever 0 <σ< 1 and with c0 > 1 defined in (2.6). We point out that the function∈ Ψ is 0-homogeneous with respect to dilations and smooth on Rn 0 , see [BLU07, Definition 5.5.1] for more information. Notice that with our assumptions\{ we} G have the inclusion B (x) B1. For every x Bσ, it holds (1−σ)/c0 ⊂ ∈ −1 wˆr,x0 (x) = Ψ(x z)ˆwr,x0 (z) dz G | | B − (x) (1 σ)/c0 ∞ Ψ L (B1) wˆr,x0 (z) dz G ≤k k B − (x) | | (1 σ)/c0 1 wˆr,x0 L (B1) Ψ ∞ k k C(σ) wˆ 1 . ≤k kL (B1) BG (x) ≤ k r,x0 kL (B1) | (1−σ)/c0 | The constant C(σ) only depends on σ and it blows up as σ 1−. By the triangle inequality and (4.5) we obtain that → wˆ ∞ C(σ) wˆ 1 k r,x0 kL (Bσ) ≤ k r,x0 kL (B1) 1 1 (4.6) C(σ) uˆr,x0 L (B1) + vˆr,x0 L (B1) ≤ k k k k C (σ) uˆ 1 + ∆ uˆ Q . ≤ 1 k r,x0 kL (B1) k H r,x0 kL 0 (B1) OPTIMAL REGULARITY FOR SUBELLIPTIC NO-SIGN OBSTACLE-TYPE PROBLEMS 16 We conclude from both (4.5) and (4.6) that ∞ ∞ ∞ uˆr,x0 L (Bσ) vˆr,x0 L (Bσ ) + wˆr,x0 L (Bσ) (4.7) k k ≤k k k k C (σ) uˆ 1 + ∆ uˆ Q . ≤ 2 k r,x0 kL (B1) k H r,x0 kL 0 (B1) Differentiatingv ˆ , seen as an integral, it turns out thatv ˆ C1 (B ). Again arguing r,x0 r,x0 ∈ H 1 as in the proof of (4.4), from the Hölder inequality and the (1 Q)-homogeneity of XjΓ, we get the estimate − −1 (4.8) X vˆ (x) = X Γ(y x) g (y)dy C g Q j r,x0 ˆ j r,x0 r,x0 L 0 (B1) | | B1 ≤ k k for every j =1,...,m, x B1 and a fixed geometric constant C > 0. By Proposition 2.4, ∈ we get a constant C3(σ) > 0, such that (4.9) wˆ ∞ C (σ) wˆ 1 . k∇h r,x0 kL (Bσ) ≤ 3 k r,x0 kL (B1) Combining (4.7), (4.8) and (4.9), along with the third inequality of (4.6), we establish the first of the following inequalities: 1 1 Q uˆr,x0 C (B ) C4(σ) uˆr,x0 L (B1) + ∆H uˆr,x0 L 0 (B ) k k H σ ≤ k k k k 1 ∞ p C5(σ,p,λ) ∆H u L (B1) + u BMO (B ) . ≤ k k k k loc 1 The second inequality is a consequence of (4.3). Since ur,x0 (0) = Xjur,x0 (0) = 0, by our x0 x0 assumptions on u, and taking into account that pr (0) = Xjpr (0) = 0, we immediately 1 infer from the CH estimate above that m ℓ (0) + X ℓ (0) | r,x0 | | i r,x0 | Xi=1 m = ℓ (0) u (0) + X ℓ (0) X u (0) | r,x0 − r,x0 | | i r,x0 − i r,x0 | Xi=1 1 ∞ p uˆr,x0 C (B ) C5(σ,p,λ) ∆H u L (B ) + u BMO (B ) . ≤k k σ ≤ k k 1 k k loc 1 It follows that ∞ ∞ p ℓr,x0 L (Bσ) C6(σ,p,λ) ∆H u L (B ) + u BMO (B ) . k k ≤ k k 1 k k loc 1 As a consequence, it follows that x0 1 x0 −1 u(x0δrx) pr (δrx) 2 sup u(y) pr (x0 y) = sup −2 r | − | x∈B r y∈Bσr(x0) σ = sup ur,x0 (x) x∈Bσ| | sup uˆr,x0 (x) + sup ℓr,x0 (x) ≤ x∈Bσ| | x∈Bσ| | C(σ,p,λ) ∆H u L∞(B ) + u BMOp (B ) . ≤ k k 1 k k loc 1 This finishes the proof. OPTIMAL REGULARITY FOR SUBELLIPTIC NO-SIGN OBSTACLE-TYPE PROBLEMS 17 Corollary 4.3. Assume u L∞(B ) such that ∆ u L∞(B ) and fix 0 <λ,σ < 1. If ∈ 1 H ∈ 1 we consider π as in (3.2), then there exists r0 > 0 such that the affine function ℓx0 (z) := u(x )+ u(x ), π(z) , x B 0 h∇h 0 i 0 ∈ λ satisfies the following properties. There exists a universal constant C(λ, σ) > 0 such that for every r (0,r ] the following estimate holds ∈ 0 x0 −1 x0 −1 ∞ ∞ 2 (4.10) sup u(y) ℓ (x0 y) pr (x0 y) C(λ, σ) ∆H u L (B1) + u L (B1) r , y∈B (x )| − − | ≤ k k k k σr 0 x0 where pr is as in Definition 3.2. Proof. Our assumptions allow us to apply Lemma 4.2 to y u(y) ℓx0 (x−1y) with p = 2. → − 0 Then there exist r0,C(λ, σ) > 0 such that x0 −1 x0 −1 x0 2 ∞ 2 sup u(y) ℓ (x0 y) pr (x0 y) C(λ, σ) ∆H u L (B1) + u ℓ BMO (B1) r y∈B (x )| − − | ≤ k k k − k loc σr 0 for every r (0,r ]. In addition, we have ∈ 0 x0 x0 ′ 2 ∞ ∞ u ℓ BMO (B ) C u ℓ L (B1) C ( u L (B1) + hu(x0) ). k − k loc 1 ≤ k − k ≤ k k |∇ | We set f = ∆ u L∞(B ) and write v = f Γ, getting H ∈ 1 ∗ u(x ) (u + v)(x ) + (f Γ)(x ) . |∇h 0 |≤|∇h 0 | |∇h ∗ 0 | Arguing as in [GT01, Lemma 4.1], we establish v(x ) = (f Γ)(x ) ∆ u ∞ Γ 1 , |∇h 0 | |∇h ∗ 0 |≤k H kL (B1)k∇h kL (B2) therefore we have u(x ) (u + v)(x ) + ∆ u ∞ Γ 1 . |∇h 0 |≤|∇h 0 | k H kL (B1)k∇h kL (B2) Since u + v is harmonic in B1, by (2.9), it follows that (u + v)(x ) C ( u ∞ + v ∞ ) |∇h 0 | ≤ 0 k kL (B1) k kL (B1) C ( u ∞ + ∆ u ∞ Γ 1 ). ≤ 0 k kL (B1) k H kL (B1)k kL (B2) This immediately leads us to our claim. Remark 4.4. Notice that under the same assumptions of Corollary 4.3, we can assume that for every λ, σ (0, 1) and any x0 Bλ, there existr ˜0 > 0 and C > 0, only depending on λ and σ, such that∈ for all r (0, r˜∈] the following estimate holds ∈ 0 x0 −1 x0 −1 ∞ ∞ 2 sup u(y) ℓ (x0 y) pr (x0 y) C(λ, σ) ∆H u L (B1) + u L (B1) r , y∈BG (x )| − − | ≤ k k k k σ0r 0 for σ = σ/c (0, 1/c ) and additionally we have the inclusion BG (x ) B . This is a 0 0 ∈ 0 r0 0 ⊂ 1 consequence of the definition of c0 in (2.8). If we set r0 :=r ˜0/c0, replacing r by c0r, we may rephrase the previous estimate as follows x0 −1 x0 −1 2 ∞ ∞ 2 (4.11) sup u(y) ℓ (x0 y) pc0r(x0 y) C(λ, σ)c0 ∆H u L (B1) + u L (B1) r G | − − | ≤ k k k k y∈Bσr(x0) for every 0 We introduce now the important definition of coincidence set: Λ := x B : u(x)=0 . { ∈ 1 } We will perform a blow-up of Λ around a fixed point x0 B1/2, considering the rescaled and translated coincidence sets ∈ Λ (x ) := x BG : u(x δ x)=0 , r 0 { ∈ 1 0 r } G for 0 < r r0 and some r0 > 0 such that Br (x0) B1. Notice that in the previous definition the≤ gauge distance is used for technical reasons,⊂ related to the existence of solutions to the Dirichlet problem with respect to the sub-Laplacian. The next result is a technical lemma, that will be used both to get the decay estimates in Proposition 4.6 and to establish the regularity in Theorem 4.8. 1,1 Lemma 4.5. Let f be such that f Γ CH (B1) and let u solve (1.1) in B1. Then for every , there exists ∗ such∈ that for every we have G 0 <λ,σ < 1 r0 > 0 x0 Bλ Br0 (x0) B1 and the following holds. Let us consider the translated and rescaled∈ function ⊂ u(x δ x) ℓx0 (δ x) px0 (δ x) (4.12) u (x) := 0 r − r − c0r r , r,x0 r2 where x0 is introduced in Definition 3.2 and x0 . For each pc0r ℓ (z)= u(x0)+ hu(x0), π(z) r (0,r ] we also define v as the solution to h∇ i ∈ 0 r,x0 ∆ v = f , in BG, (4.13) H r,x0 r,x0 σ v = u , on ∂BG, r,x0 r,x0 σ where G . Then there exists a universal constant , de- fr,x0 (x) = f(x0δrx)χBσ C(λ, σ) > 0 pending on λ and σ, such that 2 2 ∞ G ∞ ∞ Dhvr,x0 L (B ) C(λ, σ)( Dh(f Γ) L (B1) + u L (B1)). k k σ2 ≤ k ∗ k k k Proof. Due to Remark 4.4, there exists r0 > 0 such that for every x0 Bλ we have G ∈ Br0 (x0) B1 and (4.11) holds for every r (0,r0]. We write the solution to the Dirichlet problem⊂ (4.13) in the form ∈ vr,x0 = ηr,x0 + ζr,x0 , where ζr,x0 solves G ∆H ζr,x0 =0, in Bσ , ζ = u η , on ∂BG r,x0 r,x0 − r,x0 σ and we have defined η = f Γ. r,x0 − r,x0 ∗ G Indeed, the open set Bσ is regular with respect to ∆H , see [BLU07, Proposition 7.2.8]. From the identity D2v = D2(f Γ) + D2ζ , h r,x0 − h r,x0 ∗ h r,x0 OPTIMAL REGULARITY FOR SUBELLIPTIC NO-SIGN OBSTACLE-TYPE PROBLEMS 19 2 2 G taking into account the equality Dh(f Γ)(x0δrx)= Dh(fr,x0 Γ)(x) for a.e. x Bσ2 and the estimate (2.9) we obtain that ∗ ∗ ∈ 2 2 2 ∞ ∞ ∞ Dhvr,x0 L (BG ) Dh(fr,x0 Γ) L (BG ) + Dhζr,x0 L (BG ) k k σ2 ≤k ∗ k σ2 k k σ2 2 ∞ G ∞ G Dh(f Γ) L (B (x0)) + C(σ) ζr,x0 L (B ). ≤k ∗ k σ2r k k σ Now we combine the maximum principle and the Dirichlet problem (4.13) to get 2 2 ∞ G ∞ ∞ G Dhvr,x0 L (B ) Dh(f Γ) L (B1) + C(σ) ur,x0 + fr,x0 Γ L (∂Bσ ). k k σ2 ≤k ∗ k k ∗ k Due to the version of the sub-quadratic growth in (4.11), taking into account the definition (4.12) and the immediate estimate f Γ ∞ G C f ∞ , k r,x0 ∗ kL (Bσ ) ≤ k kL (B1) where C > 0 only depends on Γ, it follows that 2 2 ∞ G ∞ ∞ ∞ Dhvr,x0 L (B ) Dh(f Γ) L (B1) + C(λ, σ)( u L (B1) + f L (B1)). k k σ2 ≤k ∗ k k k k k In conclusion, we have established the following estimate 2 2 ∞ G ∞ ∞ Dhvr,x0 L (B ) C(λ, σ)( Dh(f Γ) L (B1) + u L (B1)), k k σ2 ≤ k ∗ k k k concluding the proof. Proposition 4.6 (Decay of the coincidence set). Let f be such that f Γ C1,1(B ) ∗ ∈ H 1 and let u solve (1.1). Then for every β > 0, there exist Cβ > 0 and r0 > 0 so that if 0 w = f χ G G r,x0 − r,x0 Λr(x0) ∗ Bσ OPTIMAL REGULARITY FOR SUBELLIPTIC NO-SIGN OBSTACLE-TYPE PROBLEMS 20 G G where GBσ is the Green function of Bσ , according to [BLU07, Definition 9.2.1]. From the G definition of Green function, we have GBσ 0. In addition, taking into account [BLU07, Proposition 9.2.12(iv)], we also notice that≥ the maximum principle gives −1 G G (x, y) Γ(x y) Bσ ≤ for every x, y BG with x = y. Then a standard convolution estimate yields ∈ σ 6 1/Q (4.15) w ∞ G C f ∞ G χ Q C f ∞ Λ (x ) k r,x0 kL (Bσ ) ≤ k kL (Brσ(x0))k Λr(x0)kL (B1) ≤ k kL (B1)| r 0 | 2,p for some geometric constant C > 0. The WH estimates (3.7) give a universal constant C1, depending on σ, such that 2 2Q 2Q 2Q G 2Q G Dhwr,x0 (x) dx C1 fr,x0 χΛr(x0) L (Bσ ) + wr,x0 L (Bσ ) ˆBG | | ≤ k k k k σ2 2Q 2 C2 f ∞ ( Λr(x0) + Λr(x0) ) ≤ k kL (B1) | | | | 2Q C3 f ∞ Λr(x0) . ≤ k kL (B1)| | The second inequality is again a consequence of a convolution estimate, joined with (4.15). G Since Λr(x0) B1 , the third inequality is also established. Furthermore, taking the second| order|≤| horizontal| derivatives in the definition (4.12), we get the equality P x0 =(D2u)(x δ x)+ D2w (x) D2v (x). c0r h 0 r h r,x0 − h r,x0 − G and also Λrσ2 (x0) = δσ 2 (Λr(x0) Bσ2 ). In addition, arguing as in [GT01, Lemma 7.7], 2 ∩ we can establish that (Dhu)(x0δrx) = 0 a.e. on the coincidence set Λr(x0). Taking into account all previous facts, we get 2Q x0 2Q σ Λ 2 (x ) P | rσ 0 || c0r| G x0 2Q = Λ (x ) B 2 P | r 0 ∩ σ || c0r| x0 2Q = Pc0r dx G ˆΛr(x0)∩B | | σ2 2 2 2 2Q = (Dhu)(x0δrx)+ Dhwr,x0 (x) Dhvr,x0 (x) dx ˆ G Λr(x0)∩B | − | σ2 2 2 2Q = Dhwr,x0 (x) Dhvr,x0 (x) dx ˆ G Λr(x0)∩B | − | σ2 Q 2 2Q 2 2Q 4 Dhwr,x0 (x) + Dhvr,x0 (x) dx G ≤ ˆΛr(x0)∩B | | | | σ2 2Q 2 2Q C2(σ) f ∞ Λr(x0) + Λrσ2 (x0) D (f Γ) L∞(B ) + u L∞(B ) . ≤ k kL (B1)| | | | k h ∗ k 1 k k 1 OPTIMAL REGULARITY FOR SUBELLIPTIC NO-SIGN OBSTACLE-TYPE PROBLEMS 21 Consequently, 2Q 2Q x 2Q 2 σ P 0 C (σ) D (f Γ) ∞ + u ∞ | c0r| − 2 k h ∗ kL (B1) k kL (B1) 2Q Λrσ2 (x0) Λr(x0) . C2(σ) f ∞ | |≤| | k kL (B1) βQ We see that the coefficient in front of Λ 2 (x ) is bigger than 2 if | rσ 0 | 2Q 2Q x0 2Q βQ 2Q 2 (4.16) σ P C2(σ)2 f ∞ + C2(σ) D (f Γ) L∞(B ) + u L∞(B ) . | c0r| ≥ k kL (B1) k h ∗ k 1 k k 1 2 ∞ ∞ By the simple inequality Dh(f Γ) L (B1) f L (B1), a few more computations lead us to the following sufficientk condition∗ k ≥ k k x −2Q βQ 2Q−1 1/2Q 2 P 0 C (σ)σ (2 +2 ) D (f Γ) ∞ + u ∞ , | c0r| ≥ 2 k h ∗ kL (B1) k kL (B1) h i to get (4.16) to hold. Finally, we choose σ = 1/√2, then the proof follows by choosing 2Q Q βQ 2Q−1 the constant Cβ in our statement equal to C2(1/√2)2 (2 +2 ) and replacing c0r by r. q 1,1 To carry out the proof of the CH regularity, we need a Calderón type second order differentiability, according to the next definition. 1 1 Definition 4.7. We say that u Lloc(Ω) is twice L differentiable at x0 if there exists a polynomial t of degree less than∈ or equal to two, such that 1 u(z) t(z) dz 0 as r 0+. r2 | − | → → Br(x0) The polynomial t has the following form m 1 m m2 t(x)= c + v (x x )+ c (x x )(x x )+ c (x x ), 0 l l − 0l 2 ij i − 0i j − 0j l l − 0l Xl=1 i,jX=1 l=Xm+1 x =(x ,...,x ), x =(x ,...,x ) and c ,c ,c R. 1 n 0 01 0n 0 ij l ∈ It is possible to show that any u W 2,1 (Ω) is twice L1 differentiable a.e. in Ω. ∈ H,loc Furthermore, if the function is twice L1 differentiable at a Lebesgue point x Ω of all 0 ∈ functions XiXju, Xju and u, then the corresponding polynomial is unique and it has the following form m 1 m u(x )+ X u(x )(x x )+ ((X X + X X )u)(x )(x x )(x x ) 0 j 0 j − 0j 2 i j j i 0 i − 0i j − 0j jX=1 i,jX=1 m2 + X u(x )(x x ), l 0 l − 0l l=Xm+1 see [Mag05] for more information. We are now in the position to prove the optimal interior regularity of solutions to the no-sign obstacle-type problem (1.1). OPTIMAL REGULARITY FOR SUBELLIPTIC NO-SIGN OBSTACLE-TYPE PROBLEMS 22 1,1 ∞ Theorem 4.8 (C regularity). Let u L (B1) be a distributional solution to (1.1) in R ∈ 1,1 the unit ball B1. Let f : B1 be locally summable such that f Γ CH (B1). Then there exists a universal constant→ C > 0 such that, after a modification∗ on∈ a negligible set, we have u C1,1(B ) and ∈ H 1/4 2 2 D u ∞ C D (f Γ) ∞ + u ∞ . k h kL (B1/4) ≤ k h ∗ kL (B1) k kL (B1) Proof. We consider Cβ as in Proposition 4.6 and fix β = 4. We consider a priori the following constant 2 K = C D (f Γ) ∞ + u ∞ . 4 k h ∗ kL (B1) k kL (B1) Combining the Hölder inequality and Theorem 3.4, taking into account that the constant Cβ in Proposition 4.6 is bounded from below by a universal positive constant independent ′ of β, we can find a universal constant C1 > 0 such that 2 ′ (4.17) D u 1 C K. k h kL (B1/2) ≤ 1 Let r0 > 0 be the minimum among the r0’s of Remark 4.4, Lemma 4.5 with λ =1/4 and Proposition 4.6 with β = 4. We fix an integer i0 such that (4.18) i 3+log c , 0 ≥ 2 0 −i0 such that 2 r0, where c0 is the geometric constant appearing in (2.6). Then (4.17) provides us with≤ a universal constant C 1 such that 1 ≥ y (4.19) P −i C K | 2 0 | ≤ 1 for all y B . Notice that C actually depends on i . However, this integer is fixed ∈ 1/4 1 0 throughout the proof. We have chosen i0 to satisfy also (4.18) in view of the subsequent application of Lemma 3.8 with λ = 3/4. We can fix x B such that u is twice L1 1 0 ∈ 1/4 differentiable at x0. Using [Mag05, Theorem 3.8] for p = 1 and k = 2, the set of these differentiability points has full measure in B . We can further write u = v w, such that 1 − ∆H v = f and ∆H w = fχΛ 1,1 1 on B1, where v = f Γ. By assumption v CH (B1), hence it is also a.e. twice L − ∗ ∈ 1 differentiable, therefore we can further assume v is twice L differentiable at x0, having the set of these points full measure in B1. Now, only two cases may occur. OPTIMAL REGULARITY FOR SUBELLIPTIC NO-SIGN OBSTACLE-TYPE PROBLEMS 23 x0 Case 1: lim inf|P2−k | ≤ C1K. At our point x0, we have k→∞ 2 2 Dhu(x0) = lim Dhu(y) dy k→∞ | | B −k (x0) 2 (∆ u) − x0 H B2 k (x0) = lim P2−k + Im k→∞ m x0 1 lim inf P −k + (∆ u) − 2 H B2 k (x0) ≤ k→∞ | | √m| | 1 x0 = ∆H u(x0) + lim inf P −k √m| | k→∞ | 2 | 1 f ∞ + C K ≤ √mk kL (B1) 1 2 D (f Γ) ∞ + C K. ≤k h ∗ kL (B1) 1 Therefore 2 2 ∞ ∞ Dhu(x0) (C1C4 + 1)( Dh(f Γ) L (B1) + u L (B1)). | x0 | ≤ k ∗ k k k Case 2: lim inf|P2−k | > C1K. Then the following integer is well defined k→∞ x0 k := min k N : k i , P −j > C K, for all j k . 0 { ∈ ≥ 0 | 2 | 1 ≥ } The positive integer k0 possibly depends on x0. We notice that from the definition of k0, x0 we have P − C K. The strict inequality k > i follows by (4.19). In view of our | 2 k0+1 | ≤ 1 0 0 choice of i0, that satisfies (4.18) and then i0 > 3, we can apply Lemma 3.8 with λ1 =3/4. Indeed, we have B1/4 = Bλ1/3 and −k0+1 2 −2 2 < min λ1, 1 λ1 =2 , 3 − −k0 −k0+1 so Lemma 3.8 with r1 =2 and r2 =2 yields x0 x0 2 − − ∞ ∞ P k0 P k0+1 + C Dh(f Γ) L (B1) + u L (B1) , (4.20) | 2 |≤| 2 | k ∗ k k k 2 (C C + C) D (f Γ) ∞ + u ∞ . ≤ 1 4 k h ∗ kL (B1) k kL (B1) We consider the “rescaled function” defined in Lemma 4.5: x − x0 − 0 − u(x0δ k0 x) ℓ (δ k0 x) p −k (δ k0 x) 2 2 c02 0 2 (4.21) u0(x) := − − . 4−k0 −k −k This function coincides with u2 0 ,x0 of the same lemma. Now we set f0(x) := f(x0δ2 0 x), G that is also defined on B1 . We can find a harmonic function h0 such that 2k0 − v0 =2 v(x0δ2 k0 x)+ h0 and v0 satisfies the Dirichlet problem ∆ v = f in BG, (4.22) H 0 0 σ v = u , on ∂BG, 0 0 σ OPTIMAL REGULARITY FOR SUBELLIPTIC NO-SIGN OBSTACLE-TYPE PROBLEMS 24 1 with 0 <σ< 1. Notice that v0 is also twice L differentiable at 0, being a consequence of 1 1 the twice L differentiability of v at x0. For the same reason, the twice L differentiability 1 of u at x0 gives the twice L differentiability of u0 at 0. From Lemma 4.5 with λ =1/4, there exists Cσ > 0 such that 2 2 ∞ G ∞ ∞ (4.23) Dhv0 L (B ) Cσ Dh(f Γ) L (B1) + u L (B1) . k k σ2 ≤ k ∗ k k k For the sequel, it is now important to remark that the difference (4.24) w := v u 0 0 − 0 is twice L1 differentiable at the origin. Then we know the existence of a polynomial m 1 m m2 R(x)= w (0) + X w (0) x + (X X + X X )w (0) x x + X w (0) x 0 j 0 j 2 i j j i 0 i j l 0 l jX=1 i,jX=1 l=Xm+1 such that we get 1 (4.25) w (z) R(z) dz 0 r2 | 0 − | → Bκr as r 0+ and for an arbitrary κ> 0. The definition of w immediately gives → 0 G ∆H w0 = f0χΛ − (x ) in B , 2 k0 0 σ w =0 on ∂BG. 0 σ Claim: for a fixed 0 <α< 1, there exist l 1 and C > 0, depending on α and on 0 ≥ universal constants, such that for τ =2−l0 and for every k N 0 , there exist harmonic ∈ \{ } polynomials qk with the property that 2 (2+α)(k−1) ∞ G ∞ ∞ (4.26) w0 qk L (B ) C Dh(f Γ) L (B1) + u L (B1) τ k − k τk ≤ k ∗ k k k where the constants are independent of x0. To prove (4.26) by induction, we need first to establish the case k = 1. Here we choose the null harmonic polynomial q1 =0. We consider the decomposition (4.24) and observe ∞ that standard L estimates for v0 are available, since it solves (4.22). Indeed, we may further decompose v into the sum of z = f Γ and of a harmonic function h such that 0 0 − 0 ∗ 0 G h0 ∂Bσ = u0 z0. Then we apply the sub-quadratic growth estimate (4.11) of Remark 4.4 where| we fix−λ =1/4. This leads us to the following estimate 2 v ∞ C D (f Γ) ∞ + u ∞ . k 0kL (Bσ ) ≤ 1σ k h ∗ kL (B1) k kL (B1) Then using again estimate (4.11), we obtain w0 L∞(BG) v0 L∞(BG ) + u0 L∞(BG ) k k σ ≤k k r1 k k r1 2 C D (f Γ) ∞ + u ∞ . ≤ 2σ k h ∗ kL (B1) k kL (B1) OPTIMAL REGULARITY FOR SUBELLIPTIC NO-SIGN OBSTACLE-TYPE PROBLEMS 25 −l0 Taking σ = τ = 2 , the estimate (4.26) is established for k = 1. We may take l0 N possibly larger, such that ∈ 1 (4.27) τ =2−l0 . ≤ 2 BG 1/Q | 1 | In view of Proposition 2.4, there exists a universal constant c> 0 such that 3 (4.28) DhH L∞(BG ) c H L∞(BG) k k 1/2 ≤ k k 1 G for any harmonic function H on B1 . Now we assume the statement (4.26) is true for any fixed k 1 and define ≥ w (δ k x) q (δ k x) w (x) := 0 τ − k τ k τ (2+α)(k−1) G on B1 . We choose the harmonic function hk such that G ∆H hk = 0 in B1 , h = w on ∂BG. k k 1 From the definition of wk, we get 2−α(k−1) − ∆H wk = τ f(x0δ2 k0 τ k )χΛ − (x0) · 2 k0 τk G on B1 , and the induction assumption yields 2 ∞ G ∞ ∞ (4.29) wk L (B ) C D (f Γ) L (B1) + u L (B1) . k k 1 ≤ k h ∗ k k k G −i0 Clearly wk hk vanishes on ∂B1 . Taking into account our choice of i0 such that 2 r0, the decay estimate− of the coincidence set (4.14) applies in particular for β = 4 and≤ for every r (0, 2−k0 ], that is ∈ Λ (x ) (4.30) Λ (x ) | r 0 |. | r/2 0 | ≤ 24Q Arguing as before for v0, we can decompose wk hk into the sum of a harmonic function and a convolution with the fundamental solution,− therefore standard convolution estimates yield the first of the following inequalities 2−α(k−1) ∞ − − wk hk L (BG) C τ f(x0δ2 k0 l0k )χΛ − − (x0) LQ(BG) k − k 1 ≤ k 2 k0 l0k k 1 −α(k−1) 1/Q Cτ f ∞ Λ −k −l k (x ) ≤ k kL (B1)| 2 0 0 0 | αl0(k−1) −4l0k 1/Q C2 f ∞ 2 Λ −k (x ) ≤ k kL (B1) | 2 0 0 | G 1/Q k(4−α)+α C B f ∞ τ ≤ | 1 | k kL (B1) C 2+α f ∞ τ , ≤ 2 k kL (B1) where the third inequality is a consequence of (4.30) and the last inequality follows from (4.27). Combining the estimate (4.28) for harmonic functions, the maximum principle OPTIMAL REGULARITY FOR SUBELLIPTIC NO-SIGN OBSTACLE-TYPE PROBLEMS 26 and our induction assumption as stated in (4.29), we get 3 Dhhk L∞(BG ) c hk L∞(BG) c wk L∞(BG) k k 1/2 ≤ k k 1 ≤ k k 1 2 cC D (f Γ) ∞ + u ∞ . ≤ k h ∗ kL (B1) k kL (B1) Define qk(x) as the second order Taylor polynomial of hk at the origin. In particular, qk is harmonic. The previous estimates joined with the application of the stratified Taylor inequality stated in [FS82, Corollary 1.44 with k = 2 and x = 0] give ′ 2 3 h q ∞ G C cC D (f Γ) ∞ + u ∞ τ k k − kkL (Bτ ) ≤ 2 k h ∗ kL (B1) k kL (B1) C 2 2+α D (f Γ) ∞ + u ∞ τ , ≤ 2 k h ∗ kL (B1) k kL (B1) where we have chosen l0 possibly larger, such that the following conditions 1/(1−α) 3 3 −l0 −l0 1 b τ = b 2 < 1/2 and τ =2 ′ ≤ 2cC2 ! ′ also hold. The constants b and C2 are from [FS82, Corollary 1.44 with k = 2 and x = 0] and this corollary is applied with the gauge distance dG. We stress that l0 does not 2 depend on either k0 or x0. This is very important for the final estimate of Dhu(x0). As a consequence, we obtain w q ∞ G w h ∞ G + h q ∞ G k k − kkL (Bτ ) ≤k k − kkL (Bτ ) k k − kkL (Bτ ) C 2+α C 2 2+α f ∞ τ + D (f Γ) ∞ + u ∞ τ ≤ 2 k kL (B1) 2 k h ∗ kL (B1) k kL (B1) 2 2+α C D (f Γ) ∞ + u ∞ τ . ≤ k h ∗ kL (B1) k kL (B1) Taking into account the definition of wk, we have proved that w0(δτ k ) qk(δτ k ) 2 ∞ ∞ 2+α (2+· α−)(k−1) · qk C Dh(f Γ) L (B1) + u L (B1) τ , τ − ∞ G ≤ k ∗ k k k L (Bτ ) from which we infer that (2+α)(k−1) 2 (2+α)k − ∞ G ∞ ∞ w0 qk τ qk(δτ k ) L (B ) C Dh(f Γ) L (B1) + u L (B1) τ . k − − · k τk+1 ≤ k ∗ k k k If we define the new polynomial (2+α)(k−1) qk+1(x) := qk(x)+ τ qk(δτ −k x), then the induction step is proved and this concludes the proof of our claim. By the same previous argument, we have another universal constant c′ > 0 such that 2 max hk ∞ G , hhk ∞ G , D hk ∞ G L (B1/2) L (B1/2) h L (B1/2) k k k∇ k k k ′ (4.31) c wk L∞(BG) ≤ k k 1 ′ 2 c C D (f Γ) ∞ + u ∞ . ≤ k h ∗ kL (B1) k kL (B1) OPTIMAL REGULARITY FOR SUBELLIPTIC NO-SIGN OBSTACLE-TYPE PROBLEMS 27 We introduce the following notation: m 1 m m2 q (x)= ak + bkx + ck x x + ckx , k i i 2 ij i j l l Xi=1 i,jX=1 l=Xm+1 m m m2 k 1 q (x)= ak + b x + ck x x + ckx . k i i 2 ij i j l l Xi=1 i,jX=1 l=Xm+1 From the definition of q¯k and taking into account the estimates (4.31), we get k k k k ′ 2 ∞ ∞ (4.32) max a , bi , cij, cl c C Dh(f Γ) L (B1) + u L (B1) . i,j,l { } ≤ k ∗ k k k Consequently, differentiating the equality (2+α)(k−1) q (x) q (x)= τ q (δ −k x), k+1 − k k τ with the differential operators Xi, XiXj for i, j =1,...,m and Xl for l = m +1,...,m2, and evaluating all the equalities at the origin, we get k+1 k ′ 2 (2+α)(k−1) a a c C D (f Γ) ∞ + u ∞ τ , | − | ≤ k h ∗ kL (B1) k kL (B1) k+1 k ′ −1 2 ∞ ∞ (1+α)(k−1) max bi bi c Cτ Dh(f Γ) L (B1) + u L (B1) τ , 1≤i≤m| − | ≤ k ∗ k k k k+1 k ′ −2 2 ∞ ∞ α(k−1) max cij cij c Cτ Dh(f Γ) L (B1) + u L (B1) τ , 1≤i,j≤m| − | ≤ k ∗ k k k k+1 k ′ −2 2 ∞ ∞ α(k−1) max cl cl c Cτ Dh(f Γ) L (B1) + u L (B1) τ . m+1≤l≤m2| − | ≤ k ∗ k k k Representing any of these coefficients as γk, we notice that they are Cauchy sequences converging to some γ. In addition, for any of them we have ′ −2 k c Cτ 2 (α+l)(k−1) (4.33) γ γ D (f Γ) ∞ + u ∞ τ . | − | ≤ 1 τ α k h ∗ kL (B1) k kL (B1) − k k k k k In these estimates, we have set l = 0 when γ = cl ,cij, l = 1 for γ = bi and l = 2 for k k γ = ai . As a consequence, the polynomials qk uniformly converge on compact sets to a polynomial q˜, that has the form m 1 m m2 q˜(x)=˜a + ˜b x + c˜ x x + c˜ x . i i 2 ij i j l l Xi=1 i,jX=1 l=Xm+1 Any coefficient γ of q˜, can be written for instance as γ2 + γ γ2. By (4.33), we can find a universal estimate for γ γ2, that depends on α. We also− observe that the coefficients − − of q2 are given by the formula q2 = q1 δτ 1 . The estimate (4.32) for k =2 and the fact −l0 ◦ that τ =2 can be universally fixed, independently of x0, finally lead us to the estimate 2 2 (4.34) D q˜ C D (f Γ) ∞ + u ∞ | h | ≤ τ,α k h ∗ kL (B1) k kL (B1) OPTIMAL REGULARITY FOR SUBELLIPTIC NO-SIGN OBSTACLE-TYPE PROBLEMS 28 for a suitable geometric constant Cτ,α > 0 depending on the constants of (4.33) and (4.32). From (4.33), defining ′ −2 2 ∞ ∞ c Cτ Dh(f Γ) L (B1) + u L (B1) C = k ∗ k k k , τ,α,f,u 1 τ α − we can establish the following quantitative estimate on small balls: m (α+2)(k−1) (α+1)(k−1) qk q˜ L∞(BG ) Cτ,α,f,u τ + τ xi L∞(BG ) k − k τk+1 ≤ k k τk+1 Xi=1 m m2 α(k−1) α(k−1) +τ xixj L∞(BG ) + τ xl L∞(BG ) k k τk+1 k k τk+1 i,jX=1 l=Xm+1 C C τ (2+α)(k−1), ≤ τ,α,f,u where we have used the propere intrinsic homogeneity of all the monomials xi, xixj and xl, with i, j =1,...,m and l = m +1,...,m2. We would like to prove that 1 lim 2 w0(z) q˜(z) dz =0. r→0 r G Bτr | − | Let us consider r = τ k and then 1 1 1 2 w0(z) q˜(z) dz 2 w0(z) qk(z) dz + 2 q˜(z) qk(z) dz r G r G r G Bτr | − | ≤ Bτr | − | Bτr | − | −2k −2k =τ w0(z) qk(z) dz + τ q˜(z) qk(z) dz BG | − | BG | − | τk+1 τk+1 2 αk −2k ∞ ∞ ∞ G C Dh(f Γ) L (B1) + u L (B1) τ + τ q˜ qk L (B ) ≤ k ∗ k k k k − k τk goes to zero as k . If we choose κ > 0 sufficiently small, then we have proved that →∞ 0 1 1 w (z) R(z) dz w (z) R(z) dz 0. 2k 0 2k 0 τ B | − | ≤ τ G | − | → κ τ2k B 0 τk By the uniqueness of the second order polynomial R satisfying (4.25) with κ = κ0, we get q˜ = R. Taking into account (4.23) with σ = τ and (4.34) with α =1/2, we obtain a new constant C > 0, such that 2 2 2 2 D u (0) D v (0) + D w (0) C( D (f Γ) ∞ + u ∞ ). | h 0 |≤| h 0 | | h 0 | ≤ k h ∗ kL (B1) k kL (B1) As a consequence, since k0 > i0 and i0 satisfies (4.18) we can apply Lemma 3.8 with −k0 −k0 r1 = 2 and r2 = c02 , so that taking into account the estimate (4.20) and the definition (4.21) of u0, we finally obtain a possibly larger constant, that we still denote by C > 0, such that 2 2 x0 2 ∞ ∞ D u(x0) D u0(0) + P −k C( D (f Γ) L (B ) + u L (B )), | h |≤| h | | c02 0 | ≤ k h ∗ k 1 k k 1 concluding the proof. OPTIMAL REGULARITY FOR SUBELLIPTIC NO-SIGN OBSTACLE-TYPE PROBLEMS 29 References [ALS13] John Andersson, Erik Lindgren, and Henrik Shahgholian, Optimal regularity for the no-sign obstacle problem, Comm. Pure Appl. Math. 66 (2013), no. 2, 245–262. MR 2999297 [BLU07] A. Bonfiglioli, E. Lanconelli, and F. Uguzzoni, Stratified Lie groups and potential the- ory for their sub-Laplacians, Springer Monographs in Mathematics, Springer, Berlin, 2007. MR 2363343 [BB00] Marco Bramanti and Luca Brandolini, Lp estimates for nonvariational hypoelliptic operators with VMO coefficients, Trans. Amer. Math. Soc. 352 (2000), no. 2, 781–822. MR 1608289 [BB05] , Estimates of BMO type for singular integrals on spaces of homogeneous type and applications to hypoelliptic PDEs, Rev. Mat. Iberoamericana 21 (2005), no. 2, 511–556. MR 2174915 [BF13] Marco Bramanti and Maria Stella Fanciullo, BMO estimates for nonvariational operators with discontinuous coefficients structured on Hörmander’s vector fields on Carnot groups, Adv. Differential Equations 18 (2013), no. 9-10, 955–1004. MR 3100057 [Caf89] Luis A. Caffarelli, Interior a priori estimates for solutions of fully nonlinear equations, Ann. of Math. (2) 130 (1989), no. 1, 189–213. MR 1005611 (90i:35046) [CKS00] Luis A. Caffarelli, Lavi Karp, and Henrik Shahgholian, Regularity of a free boundary with ap- plication to the Pompeiu problem, Ann. of Math. (2) 151 (2000), no. 1, 269–292. MR 1745013 [DGP07] Donatella Danielli, Nicola Garofalo, and Arshak Petrosyan, The sub-elliptic obstacle problem: C1,α regularity of the free boundary in Carnot groups of step two, Adv. Math. 211 (2007), no. 2, 485–516. MR 2323535 [DGS03] Donatella Danielli, Nicola Garofalo, and Sandro Salsa, Variational inequalities with lack of ellipticity. I. Optimal interior regularity and non-degeneracy of the free boundary, Indiana Univ. Math. J. 52 (2003), no. 2, 361–398. MR 1976081 [DFPP08] Marco Di Francesco, Andrea Pascucci, and Sergio Polidoro, The obstacle problem for a class of hypoelliptic ultraparabolic equations, Proc. R. Soc. Lond. Ser. A Math. Phys. Eng. Sci. 464 (2008), no. 2089, 155–176. MR 2434049 [FS12] Alessio Figalli and Henrik Shahgholian, A General Class of Free Boundary Problems for Fully Nonlinear Elliptic Equations. [Fol75] G. B. Folland, Subelliptic estimates and function spaces on nilpotent Lie groups, Ark. Mat. 13 (1975), no. 2, 161–207. MR 0494315 [FS82] G. B. Folland and Elias M. Stein, Hardy spaces on homogeneous groups, Mathematical Notes, vol. 28, Princeton University Press, Princeton, N.J.; University of Tokyo Press, Tokyo, 1982. MR 657581 [Fre72] Jens Frehse, On the regularity of the solution of a second order variational inequality, Boll. Un. Mat. Ital. (4) 6 (1972), 312–315. MR 0318650 (47 #7197) [Fre13] Marie Frentz, Regularity in the obstacle problem for parabolic non-divergence operators of Hörmander type, J. Differential Equations 255 (2013), no. 10, 3638–3677. MR 3093377 [FGN12] Marie Frentz, Elin Götmark, and Kaj Nyström, The obstacle problem for parabolic non- divergence form operators of Hörmander type, J. Differential Equations 252 (2012), no. 9, 5002–5041. MR 2891355 [FNPP10] Marie Frentz, Kaj Nyström, Andrea Pascucci, and Sergio Polidoro, Optimal regularity in the obstacle problem for Kolmogorov operators related to American Asian options, Math. Ann. 347 (2010), no. 4, 805–838. MR 2658144 [Fri88] Avner Friedman, Variational principles and free-boundary problems, second ed., Robert E. Krieger Publishing Co., Inc., Malabar, FL, 1988. MR 1009785 [GN96] Nicola Garofalo and Duy-Minh Nhieu, Isoperimetric and Sobolev inequalities for Carnot- Carathéodory spaces and the existence of minimal surfaces, Comm. Pure Appl. Math. 49 (1996), no. 10, 1081–1144. MR 1404326 OPTIMAL REGULARITY FOR SUBELLIPTIC NO-SIGN OBSTACLE-TYPE PROBLEMS 30 [GT01] David Gilbarg and Neil S. Trudinger, Elliptic partial differential equations of second order, Springer, 2001. [GS05] Björn Gustafsson and Harold S. Shapiro, What is a quadrature domain?, Quadrature domains and their applications, Oper. Theory Adv. Appl., vol. 156, Birkhäuser, Basel, 2005, pp. 1–25. MR 2129734 [Jer86] David Jerison, The Poincaré inequality for vector fields satisfying Hörmander’s condition, Duke Math. J. 53 (1986), no. 2, 503–523. MR 850547 [KS00] David Kinderlehrer and Guido Stampacchia, An introduction to variational inequalities and their applications, Classics in Applied Mathematics, vol. 31, Society for Industrial and Applied Mathematics (SIAM), Philadelphia, PA, 2000, Reprint of the 1980 original. MR 1786735 [LM00] Ermanno Lanconelli and Daniele Morbidelli, On the Poincaré inequality for vector fields, Ark. Mat. 38 (2000), no. 2, 327–342. MR 1785405 [Lu96] Guozhen Lu, Embedding theorems into Lipschitz and BMO spaces and applications to quasi- linear subelliptic differential equations, Publ. Mat. 40 (1996), no. 2, 301–329. MR 1425620 [Mag05] Valentino Magnani, Differentiability from the representation formula and the Sobolev-Poincaré inequality, Studia Math. 168 (2005), no. 3, 251–272. MR 2146126 [PS07] Arshak Petrosyan and Henrik Shahgholian, Geometric and energetic criteria for the free boundary regularity in an obstacle-type problem, Amer. J. Math. 129 (2007), no. 6, 1659– 1688. MR 2369892 [PSU12] Arshak Petrosyan, Henrik Shahgholian, and Nina Uraltseva, Regularity of free boundaries in obstacle-type problems, Graduate Studies in Mathematics, vol. 136, American Mathematical Society, Providence, RI, 2012. MR 2962060 [Rod87] José-Francisco Rodrigues, Obstacle problems in mathematical physics, North-Holland Math- ematics Studies, vol. 134, North-Holland Publishing Co., Amsterdam, 1987, Notas de Matemática [Mathematical Notes], 114. MR 880369 [Sak91] Makoto Sakai, Regularity of a boundary having a Schwarz function, Acta Math. 166 (1991), no. 3-4, 263–297. MR 1097025 [Var84] V. S. Varadarajan, Lie groups, Lie algebras, and their representations, Graduate Texts in Mathematics, vol. 102, Springer-Verlag, New York, 1984, Reprint of the 1974 edition. MR 746308 Dipartimento di Matematica, Università di Pisa, Largo Bruno Pontecorvo 5, 56127, Pisa, Italy Email address: [email protected] KTH Royal Institute of Technology, 100 44 Stockholm, Sweden Email address: [email protected]