∆u = f

supΩ dx |Du(x)| ≤ c(supΩ |u| + supΩ dx2 |f (x)|)
the Sobolev embedding theorem. Here k is a non-negative integer and 1 ≤ p ≤ ∞.
The first part of the Sobolev embedding theorem states that if k > l and
1 ≤ p < q ≤ ∞ are two extended real numbers such that (k − l)p < n and:
1 k −`
= −
W k,p (Rn ) ⊆ W `,q (Rn )
and the embedding is continuous.
This special case of the Sobolev embedding is a direct consequence of the
Gagliardo-Nirenberg-Sobolev inequality.
The second part of the Sobolev embedding theorem applies to embeddings in H¨older
spaces C r ,α (Rn ). If (k − r − α)/n = 1/p with α ∈ (0, 1), then one has the embedding
W k,p (Rn ) ⊂ C r ,α (Rn ).

This part of the Sobolev embedding is a direct consequence of Morrey’s inequality.
Intuitively, this inclusion expresses the fact that the existence of sufficiently many weak
derivatives implies some continuity of the classical derivatives.
In the theory of partial differential equations, elliptic operators are differential operators
that generalize the Laplace operator. They are defined by the condition that the
coefficients of the highest-order derivatives be positive, which implies the key property
that the principal symbol is invertible, or equivalently that there are no real
characteristic directions.
A linear differential operator L of order m on a domain Ω in Rd given by

Lu =

aα (x)∂ α u


(where α is a multi-index) is called elliptic if for every x in Ω and every non-zero ξ in
Rd ,

aα (x)ξ α 6= 0.


In many applications, this condition is not strong enough, and instead a uniform

If ρ is the charge density within some region Ω.ellipticity condition may be imposed for operators of degree m = 2k: (−1)k X aα (x)ξ α > C |ξ|2k . (∂ α u)|α|≤2k ) (1) is elliptic if its first-order Taylor expansion with respect to u and its derivatives about any point is a linear elliptic operator. Example 1 The negative of the Laplacian in Rd given by −∆u = − d X ∂i2 u i=1 is a uniformly elliptic operator. |α|=2k where C is a positive constant. u. Note that ellipticity only depends on the highest-order terms. The Laplace operator occurs frequently in electrostatics. the potential ∆ must . [1] A nonlinear operator L(u) = F (x.

the operator Lu = −∂i (aij (x)∂j u) + b j (x)∂j u + cu is elliptic. the p-Laplacian is a nonlinear elliptic operator defined by L(u) = − d X i=1 Elliptic regularity theorem ∂i (|∇u|p−2 ∂i u). This is the most general form of a second-order divergence form linear elliptic differential operator. These operators also occur in electrostatics in polarized media. . Example 3 For p a non-negative number. Example 2 Given a matrix-valued function A(x) which is symmetric and positive definite for every x. The Laplace operator is obtained by taking A = I .satisfy the equation −∆Φ = 4πρ. having components aij .

if f is infinitely-often differentiable. as the weak solution u might not have enough derivatives for the expression Lu to even make sense. thus. only guarantees that a weak solution u exists in the Sobolev space H k . every elliptic operator is hypoelliptic. Any differential operator exhibiting this property is called a hypoelliptic operator. u will in fact have 2k square-integrable weak derivatives. This situation is ultimately unsatisfactory. given a function f and some appropriate boundary values. it follows that f is smooth.Let L be an elliptic operator of order 2k with coefficients having 2k continuous derivatives. provided f is square-integrable. The property also means that every fundamental solution of an elliptic operator is infinitely differentiable in any neighborhood not containing 0. The existence theory for elliptic operators. The Dirichlet problem for L is to find a function u. such that Lu = f and such that u has the appropriate boundary values and normal derivatives. The elliptic regularity theorem guarantees that. As an application. suppose a function f satisfies the Cauchy-Riemann equations. General definition . then so is u. using Grding’s inequality and the LaxMilgram lemma. Since the Cauchy-Riemann equations form an elliptic operator. In particular.

2 g = (g . Lw = Lu − Lϕ = gˆ + Di fˆi .uniform (strongly) elliptic. j = 1. v ) ≤ c|u|1. v ) = Ω [(aij (x)Dj + b i (x)u)Di v − (c i (x)Di u + d(x)u)v ] dx for all v ∈ C01 (Ω).2 |v |1. f n ) Suppose that u ∈ W 1. x ∈ Ω. aij (x)ξi ξj ≥ λ|ξ|2 .2 R Lu = g + Di f i . n. n. function f belong L2 (Ω). w ∈ W01. n uniformly lipschitz continuous in Ω. F (v ) = Ω (f i Di v − gv )dx. b i .2 (Ω) g . 2.2 (Ω) weak solution Lu = f . ie. 2. u = ϕ on ∂Ω. u . d ∈ L∞ (Ω). · · · . w = u − ϕ. · · · .2 Iu v = Ω uvdx Suppose that ϕ ∈ W 1. has unique solution. ξ ∈ Rn . (2) R L(u.strong and weak derivatives ?? ]∠^ Lu = Di (aij (x)Dj + b i (x)u) + c i (x)Di u + d(x)u We suppose that operator L uniform (strongly) elliptic. Then generalized Dirichlet problem Lu = g + Di f i in Ω. coefficients aij . Then for every Ω0 ⊂⊂ Ω. coefficients c i . F ∈ H ∗ R H = W01. i = 1. 2. L(u. f 1 . · · · . i. · · · . f i ∈ L2 (Ω). i = 1.

xi xj n−1 ∆u + b |x| 2 Dij u = 0. Then for every Ω0 ⊂⊂ Ω. if ∂Ω enough smooth ϕ ∈ W 2.2 (Ω0 ) and the estimate holds |u|W k=2.1 (Ω).2 (Ω) ) .2 (Ω) ∩ Wloc (Ω) such that Lu = f in Ω and u − ϕ ∈ W01.2 u ∈ W 1.2 (Ω). b i belong C k. c ∈ L∞ (Ω). b i .2 (Ω0 ) + |f |L2 (Ω) ) .e. d belong C k−1. aij ∈ C 0.2 (Ω0 ) and the estimate holds |u|W 2. Lu = f a. coefficients aij .2 (Ω). k ≥ 1.1 (Ω). However for discontinuous coefficient the result doe not hold as the following example shows.1 (Ω). there is unique 2. By induction. b = −1 + 1−λ .2 (Ω).2 (Ω) the result holds under condition that aij are continuous on Ω.2 (Ω0 ) ≤ c(|u|W 1. coefficients c i . u belong W k+2. Let Lu = (aij (x)Dij u + b i Di u + c(x)u be uniform (strongly) elliptic. in Ω.belong W 2. where Lu = (aij (x)Dij u + (Dj aji + b i (x) + c i (x))Di u + (Di b i + d(x))u. Then for L2 (Ω) and ϕ ∈ W 1.2 (Ω0 ) ≤ c(|u|W 1. c ≤ 0.2 (Ω0 ) + |f |W k. function f belong W k.

Eberhard Hopf proved in 1927 that if a function satisfies a second order partial differential inequality of a certain kind in a domain of Rn and attains a maximum in the domain then the function is constant. the comparison technique he introduced for this purpose. xn ) be a C 2 function which satisfies the differential inequality Lu = X ij aij (x) X ∂u ∂2u + bi ≥0 ∂xi ∂xj ∂xi i in an open domain Ω. It is usually thought that the Hopf maximum principle applies only to linear differential operators L. where the symmetric matrix aij = aij (x) is locally uniformly positive definite in Ω and the coefficients aij . In particular. Let u = u(x). If u takes a maximum value M in Ω then u = M. which coincide on S. The simple idea behind Hopf’s proof.for n > 2(2 − λ) has two solutions u1 (x) = 1 and u2 (x) = |x|λ in W 2. Generalizing the maximum principle for harmonic functions which was already known to Gauss in 1839. bi = bi (x) are locally bounded. this is the point of view taken by Courant and Hilbert’s . x = (x1 . · · · . has led to an enormous range of important applications and generalizations.2 (B).

in some cases. In the later sections of his original paper. Luca Martinazzi. 0 ≤ α ≤ 1 if for every point x0 ∈ ∂Ω there exists a ball B = B(x0 ) and mapping ψ from B onto D such that ([?].α (B).α (Ω). leads to uniqueness statements in the Dirichlet problem for the mean curvature operator and the MongeAmpre equation.α .α . Theorem 6. for k=1 it is proved in Gilbarg-H¨ ormander in Arch. then P[f ] ∈ C k. Harmonic Maps. . If f ∈ C k. k ≥ 2. An Introduction to the Regularity Theory for Elliptic Systems. GERMANY (Kalaj) . 2012 Scuola Normale Superiore Pisa u ∈ C k (Ω) if all derivatives of order ≤ k have continuous extension to Ω. (ii) ψ(B ∩ ∂Ω ⊂ ∂R+ (iii) ψ ∈ C k. p. (i) ψ(B ∩ Ω) ⊂ R+ n ). A bounded domain Ω in Rn and its boundary belong to class C k.19 [?]:Let Ω be of class C k. Let C k (Ω) denote family of functions (mappings) which belong C k (Ω) and all derivatives of order ≤ k have continuous extension to Ω.Mariano Giaquinta.Methods of Mathematical Physics. however. Anal. ψ −1 ∈ C k. 95): n. Ration. Hopf considered a more general situation which permits certain nonlinear operators L and.α (Ω). Mech.α (D).

Let Ω and Ω0 be domain in R n and Ω b Ω0 . H¨ormander. The LaplaceBeltrami operator. Lars Intermediate Schauder estimates. 35J25 (see also references in [?]) Theorem (Lemma 6. u = ϕ on T . ϕ ∈ C 2. David. An explicit formula in local coordinates is possible.α (Ω ∪ T ). Then u ∈ C 2. f belong C ∞ (Ω) and let u ∈ W 1. Rational Mech. like the Laplacian. c i .Gilbarg.2 (Ω) be weak solution of strongly elliptic equation Lu = f . 74 (1980). one can prove: Theorem Let aij . d. f and coefficients of L belong C α (Ω). 4.18 [?]) Let boundary of Ω contain part T of class C 2.α (Ω) u ∈ C 0 (Ω) ∩ C 2 (Ω). b i . Then for every f ∈ C k (Ω) (resp f ∈ H k (Ω)) there is an extension F ∈ C k (Ω0 ) (resp F ∈ H k (Ω0 ) ) finite in Ω0 such that |F |H k (Ω0 ) ≤ C |f |H k (Ω) . is the divergence of the gradient: ∆f = div grad f . Then u also belongs to C ∞ (Ω). Using the Sobolev embedding theorem. Lu = f on Ω. no. Anal. Arch. . 297–318.α .

given in an oriented coordinate system xi by voln := p |g | dx 1 ∧ . Xn−1 ). d(iX ω) = divω X ω . Here |g | := |det(gij )| is the absolute value of the determinant of the metric tensor gij . · · · . ∧ dx n where the dx i are the 1-forms forming the dual basis to the basis vectors ∂i := ∂ ∂x i and ∧ is the wedge product. Xn−1 ) = ω(X . · · · .Suppose first that M is an oriented Riemannian manifold. one can define n − 1-form iX ω by iX ω(X1 . . . The orientation allows one to specify a definite volume form on M. For given n-form ω. X1 . The divergence divX of a vector field X on the manifold is then defined as the scalar function with the property (divX ) voln := LX voln where LX is the Lie derivative along the vector field X .

one obtains p  1 |g |X i divX = p ∂i |g | where the Einstein notation is implied. . as hgradf (x). Here. ·i on the manifold. vx i = df (x)(vx ) for all vectors vx anchored at point x in the tangent space Tx M of the manifold at point x. one has (gradf )i = ∂ i f = g ij ∂j f where g ij are the components of the inverse of the metric tensor. The gradient of a scalar function is the vector field grad f that may be defined through the inner product h·. In local coordinates.If ωg is the volume form of (M. d is the exterior derivative of the function . In local coordinates. so that g ij gjk = δki with δki the Kronecker delta. so that the repeated index i is summed over. it is a 1-form taking argument vx . g ). then the number divg X := divωg X is known as the divergence of X .

in local coordinates  p 1 ∆f = div grad f = p ∂i |g |g ij ∂j f . when M is the circle. The Laplacian commutes with isometries. where ∆f = g ij ∂ij2 f . The Euler-Lagrange differential equation for the energy functional is a non-linear elliptic partial differential equation. between two compact Riemannian manifolds. the formula for the LaplaceBeltrami operator ∆ applied to a scalar function is. For example. Hence. |g |   p p p |g |g ij ∂j f = ∂i |g |g ij ∂j f + |g |g ij ∂ij2 f . The map from the circle to the equator of the standard 2-sphere is a harmonic map. and so are the maps that take the circle and map it . is a harmonic map if it is a critical point for the energy functional intM |du|2 dµM .Combining the definitions of the gradient and divergence. Since ∂i ∆f = ∆f + lower order terms. The norm of the differential |du| is given by the metric on M and N and dµM is the measure on M. u is a closed geodesic iff u is harmonic. A map u : M → N. then the Euler-Lagrange equation is the same as the geodesic equation. Typically. the class of allowable maps lie in a fixed homotopy class of maps.

and if it does it may not be unique.around the equator n times. for any integer n. A higher-dimensional example is a meromorphic function on a compact Riemann surface. the harmonic maps have been classified. h) and φ as above. g ). Note that these all lie in the same homotopy class. Thus by Hodge’s theorem for surfaces. a harmonic representative exists for each homotopy class. Given Riemannian manifolds (M. or of a harmonic function when the range is a Euclidean space. For surfaces. with respect to the induced metric on the bundle T ∗ M ⊗ φ−1 TM. which is a harmonic map to the Riemann sphere. A harmonic map may not always exist in a homotopy class. there are no non-trivial harmonic maps from the sphere to the torus. When N is negatively curved. the energy density of e(φ) at a point x in M is defined as 1 e(φ) = kdφk2 2 where the kdφk2 is the squared norm of the differential of φ. (N. and is also unique. A harmonic map between Riemannian manifolds can be viewed as a generalization of a geodesic when the domain dimension is one. The total energy of E (φ) is given by . and are precisely the holomorphic maps and the anti-holomorphic maps.

This generalizes the classical Dirichlet energy. more typically. 2 ∂x ∂x j If M is compact. This definition is extended to the case where M is not compact by requiring the restriction of φ to every compact domain to be harmonic. in local coordinates the right hand side of this equality reads e(φ) = 1 ∂φα ∂φβ e(φ) = g ij hαβ i . then φ is called a harmonic map if it is a critical point of the energy functional E . or. requiring that φ be a critical point of the energy functional in the Sobolev . 2 Using the Einstein summation convention. The energy density can be written more explicitly as 1 traceg φ∗ h.integrating the density over M Z 1 E (φ) = e(φ) dvg = 2 M Z kdφk2 dvg M where dvg denotes the measure on M induced by its metric.

(In this case.space H 1. N). In terms of the physical analogy. that φ is a curve (or a closed curve) on N. it corresponds to the direction in which the ”rubber” manifold M will tend to move in N in seeking the energy-minimizing configuration.2 (M. The quantity τ (φ) is a section of the bundle φ−1 (TN) known as the tension field of φ.e. Examples Identity and constant maps are harmonic. the map φ is harmonic if it satisfies the Euler-Lagrange equations associated to the functional E . Then φ is a harmonic map if and only if it is a geodesic. Equivalently. Assume that the source manifold M is the real line R (or the circle S1 ). (3) where ∆g is the Laplace-Beltrami operator on B n and Γ0ν αβ the Christoffel symbols on 0 M. i. the rubber-and-marble . We define the tension field τ (u) of u by coordinates α β τ (u)ν = ∆g u ν + g ij Γ0ν αβ ◦ u ui uj . These equations read def τ (φ) = traceg ∇dφ = 0 where ∇ is the connection on the vector bundle T ∗ M ⊗ φ−1 (TN) induced by the Levi-Civita connections on M and N.

One of the original ideas in grid generation methods for .e. one ”releases” it. ∇dφ∗ h itself vanishes. This ”physical” observation leads to the following mathematical problem: given a homotopy class of maps from M to N. a minimal submanifold M of N is a harmonic immersion of M in N.) In theoretical physics. If. Then φ is a harmonic map if and only if it is a harmonic function in the usual sense (i. not just its trace). harmonic maps correspond to instantons. Every totally geodesic map is harmonic (in this case. a quantum field theory whose action is given by the Dirichlet energy is known as a sigma model. then it gives a harmonic coordinate system. Every holomorphic map between Khler manifolds is harmonic. after applying the rubber M onto the marble N via some map φ.analogy described above reduces to the usual elastic band analogy for geodesics.) Assume that the target manifold N is Euclidean space Rn (with its standard metric). Once existence is known. does it contain a representative that is a harmonic map? Existence results on harmonic maps between manifolds has consequences for their curvature. how can a harmonic map be constructed explicitly? (One fruitful method uses twistor theory. More generally. a solution of the Laplace equation). If φ is a diffeomorphism onto an open set in Rn. it will try to ”snap” into a position of least tension. In such a theory. This follows from the Dirichlet principle. Every minimal surface in Euclidean space is a harmonic immersion.

In 1975 Yau [65] proved that. Since the fundamental works of Cheng and Yau [20]. but also in the non-linear setting of p-harmonic functions. The interplay between the existence of non-constant harmonic functions on a complete Riemannian manifold M and the global geometrical structure of M has inspired researchers in geometric analysis for more than thirty years. in a sense. [38]. A crude idea is that harmonic functions are. Greene and Wu [30] conjectured that a . Greene and Wu [29]. every positive harmonic function is constant. see e.computational fluid dynamics and computational physics was to use either conformal or harmonic mapping to generate regular grids. and [24]. [67] in the mid-70s. on a complete Riemannian manifold M of non-negative Ricci curvature. [37]. regarding the negative curvature case. In other words. and Yau [65]. whereas they should exist in abundance on simply connected manifolds with negative sectional curvature K ≤< −a2 < 0. M has the so-called strong Liouville property. On the other hand. Some of these results have their counterparts in the case of p-harmonic functions as well. There is a vast literature on such Liouville-type theorems for harmonic functions and therefore we just refer to the survey articles [31] by Grigoryan and [53] by Li. rare on manifolds satisfying various non-negative curvature assumptions.g. there has been a lot of research related to various Liouville-type problems not only for harmonic functions.

α) = {x ∈ M \ o : ^(v . connected and simply connected Riemannian n-manifold. the exponential map expo : To M → M is a diffeomorphism for every point o ∈ M. let Co (v . In particular. called the cone topology. By the Cartan-Hadamard theorem. There is ¯ = M ∪ S(∞) defined as follows. The class of γ1 is denoted by γ1 (∞). o) is the distance function to a fixed point o ∈ M. γ˙ x (0)) < α} . It is well-known that M can be compactified by adding a sphere at infinity. on M For any point o ∈ M and any unit vector v ∈ To M. of non-positive sectional curvature. where r = d(. n ≥ 2. M is diffeomorphic to R n . The sphere at infinity is defined as the set of all equivalence classes of geodesic unit speed rays in M. γ2 (t)) < 1. two such rays γ1 and γ2 are equivalent if supt≥0 d(γ1 (t). Recall that a Cartan-Hadamard manifold is a complete. so that the resulting space M = M ∪ S(∞) will be homeomorphic to a closed Euclidean ball.Cartan-Hadamard manifold M admits a nonconstant bounded harmonic function if the sectional curvatures of M have an upper bound K (x) ≤ −Cr −2 (x) outside a compact set for some constant C > 0. denoted by S(∞). a natural topology.

r > 0. meaning that it satisfies Laplace’s equation ∆x i = 0. s)..be the cone about v of angle α > 0. q ∈ M. the cone topology is independent of the choice of o ∈ M and. xn ) on a Riemannian manifold each of whose coordinate functions x i is harmonic. Here ∆ is the LaplaceBeltrami operator. a branch of mathematics. and truncated cones Co(v . That is to ask whether every continuous function on S(∞) has a (unique) harmonic extension to M. . α > 0. form a basis for the cone topology. equipped with this topology. with v ∈ To M. in general. harmonic coordinates are a coordinate system (x1 . Furthermore. Of course. Equivalently. M is homeomorphic to the closed unit ball B¯ n ⊂ Rn and S(∞) to the sphere S n−1 = ∂B n . A way to approach the conjecture of Greene and Wu is to study whether the so-called Dirichlet problem at infinity (or the asymptotic Dirichlet problem) is solvable on a Cartan-Hadamard manifold. r ).. regarding a coordinate system . γ˙ x (0)) is the angle between the vectors v and γ˙ x (0) in To M. where γ x is the unit speed geodesic from o = γ x (0) to x and ^(v . s > 0. α) B(o. see [26]. the answer. Then geodesic balls B(q. is no since the simplest Cartan-Hadamard manifold R n admits no positive harmonic functions other than In Riemannian geometry..

roughly meaning that it minimizes the elastic energy of ”stretching” M into Rn . They were then introduced into Riemannian geometry by Sabitov & efel (1976) and later were studied by DeTurck & Kazdan (1981). the latter being a special case of the former. Harmonic coordinates are characterized in terms of the Christoffel symbols by means of the relation . The elastic energy is expressed via the Dirichlet energy functional Z E [ϕ] = |dϕ|2 dV . M In two dimensions. the coordinate system is harmonic if and only if ϕ is a harmonic map of Riemannian manifolds. harmonic coordinates have been well understood for more than a century. The essential motivation for introducing harmonic coordinate systems is that the metric tensor is especially smooth when written in these coordinate systems. and are closely related to isothermal coordinates. Harmonic coordinates in higher dimensions were developed initially in the context of general relativity by Einstein (1916) (see harmonic coordinate condition).as a local diffeomorphism ϕ : M → Rn .

then they are in that same Hlder space when expressed in harmonic coordinates. a result which follows easily from standard results on the existence and regularity of solutions of elliptic partial differential equations. In particular.α when expressed in some coordinate system. the equation ∆u j = 0 has a solution in a ball around any given point p. Harmonic coordinates always exist (locally). such that u j (p) and ∂u j /∂x i (p) are all prescribed. for any coordinate system at all. The basic regularity theorem concerning the metric in harmonic coordinates is that if the components of the metric are in the Hlder space C k. ∆x k = −g ij Γkij . .g ij Γkij = 0 and indeed.

This is known as the harmonic coordinate condition in physics. harmonic coordinates are solutions of the wave equation instead of the Laplace.In general relativity. .