## Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

Subscribe to the OCW Newsletter

Enter Email

Help | Contact Us

Courses

About

Donate

Featured Sites

Search

Advanced Search

Home » Courses » Mathematics » Introduction to Partial Differential Equations » Lecture Notes

Lecture Notes

COURSE HOME SYLLABUS CALENDAR LECTURE NOTES ASSIGNMENTS

The lecture notes for this course were prepared by Dale Winter, a student in the class, in collaboration with Prof. Colding.

LEC # 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 TOPICS Harmonic Functions and the Harnack Inequality (PDF) The Gradient Estimate (PDF) The Hopf Maximum Principle (PDF) The Poincare Inequalities (PDF) The Cacciopolli Inequality (PDF) More General Operators (PDF) Consequences of Cacciopolli (PDF) Maximum Principles and Gradient Estimates (PDF) Hopf and Harnack for L-harmonic Functions (PDF) An Improved Gradient Estimate for Harmonic Functions (PDF) More on Harmonic Functions on a Ball (PDF) Solving the Laplace Equation in R 2: The Dirichlet Problem (PDF) The Heat Equation (PDF) A Gradient Estimate for the Heat Equation on a Ball (PDF) Campanato's Lemma and Morrey's Lemma (PDF) Five Inequalities for Harmonic Functions (PDF) Regularity of L-harmonic Functions Part I (PDF) Regularity of L-harmonic Functions Part II (PDF) Regularity of L-harmonic Functions Part III (PDF) Smoothness of L-harmonic Functions (PDF) The Mean Value Inequality Revisited Part I (PDF) The Mean Value Inequality Revisited Part II (PDF) Moser's Approach (PDF)

COURSES

» Find by Topic » Find by Course Number » Find by Department » Audio/Video Courses » OCW Scholar Courses » Most Visited Courses

ABOUT

» About OpenCourseWare » Site Stats » OCW Stories » OCW Consortium » Media Coverage » Newsletter

DONATE

» Make a Donation » Why Donate? » Become a Course Champion » Our Supporters » Other Ways to Contribute

FEATURED SITES

» Highlights for High School » edX

**ABOUT MIT OPENCOURSEWARE
**

MIT OpenCourseWare makes the materials used in the teaching of almost all of MIT's subjects available on the Web, free of charge. With more than 2,000 courses available, OCW is delivering on the promise of open

TOOLS

» Help & FAQs » Contact Us

file:///Volumes/Data/Users/lhpham/Documents/18-152-fall-2005/contents/lecture-notes/index.htm[17/01/13 11:11:28 AM]

**Lecture Notes | Introduction to Partial Differential Equations | Mathematics | MIT OpenCourseWare
**

» New Courses » Translated Courses » Supplemental Resources » Press Releases » OCW's Next Decade » Shop OCW » Become a Corporate Sponsor » Advanced Search » Site Map » Privacy & Terms of Use

sharing of knowledge.

OUR CORPORATE SUPPORTERS

© 2001–2012 Massachusetts Institute of Technology Your use of the MIT OpenCourseWare site and materials is subject to our Creative Commons License and other terms of use.

file:///Volumes/Data/Users/lhpham/Documents/18-152-fall-2005/contents/lecture-notes/index.htm[17/01/13 11:11:28 AM]

Lecture One: Harmonic Functions and the Harnack Inequality

1

The Laplacian

Let Ω be an open subset of Rn , and let u : Ω → R be a smooth function. We deﬁne the Laplacian by �u = The equation �u = 0 (2) is called the Laplace equation, and functions which satisfy it are said to be harmonic. Harmonic functions turn out to be very important, and much of this course will be devoted to their study. Also of interest are functions with nonnegative or nonpositive laplacian. These are termed sub and superharmonic respectively.

n � ∂2u i=1

∂x2 i

.

(1)

2

The Maximum principle

The maximum principle is simply the statement that the gradient of a function at a max ¯r (x) with imum is zero. Formally, if u is a twice diﬀerentiable function on a closed ball B a maximum at x, then �u = 0 at x and �u ≤ 0 at x. The one dimensional case should be familiar, and proofs of other cases are analogous. (3)

3

Dirichlet Energy

� Recall that the Dirichlet energy of a function v : Ω → R is given by Ω |�v |2 . We will show that harmonic functions correspond to critical points of Dirichlet energy. For ﬁxed v and ∞ (Ω) (i.e. φ an inﬁnitely diﬀerentiable real valued function on Ω) we deﬁne for any φ ∈ C0 � Eφ (t) = |�(v + tϕ)|2 . (4)

Ω

1

� � �(v − u) · �(v + u) = − (v − u)�(v + u). Ω Ω (7) for all functions v satisfying u = v on ∂ Ω. by Stokes’ theorem. In other words harmonic functions have the smallest Dirichlet energy for their boundary values. Proposition 3. it is clear that ∂ Ω φ�v · dS = 0.Now compute � Eφ (t) = Ω |�v |2 + t2 � Ω |�φ|2 + 2t � �v · �φ. These results give our correspondence. (9) Apply these to get 2 . Ω (6) d Thus. Applying Stokes’ theorem to this gives � � �v · �φ = − φ�v. dt Ω � Since φ = 0 on ∂ Ω. The converse is also true but we will not prove it here. then dt |t=0 Eφ (t) = 0 for all φ. Therefore. Ω � d (5) |t=0 Eφ (t) = 2 �v · �φ. Ω Ω so which we apply to (5) to get d |t=0 Eφ (t) = −2 dt � φ�v. � Proof is by calculation. If u is a harmonic function on Ω then � � 2 |�u| ≤ |�v |2 .1 Let Ω ⊂ Rn be open. The following proposition makes it more explicit. if �v = 0. Clearly ∂ Ω (v − u)�(v + u) · dS = 0. (8) Ω Ω Similarily � �(v − u) · �(v − u) = − Ω Ω � (v − u)�(v − u).

Proof First note that d 1 dr vol ∂Br (x) � u= ∂Br (x) � u Br (x) (11) 1 vol ∂Br (x) � ∂Br (x) ∂u . ∂n (12) This is simply saying that the derivative of the average of u is the average of the outward normal derivative. 4 The Mean Value Property The following property of harmonic properties turns out to be very useful Theorem 4.1 Let x ∈ Rn and take Br0 (x) a ball around x.� Ω |�v |2 − |�u|2 = � Ω �(v − u) · �(v + u) � = − (v − u)�(v + u) �Ω = − (v − u)�(v − u) � Ω = |�(v − u)|2 Ω ≥ 0 as required. If u is a harmonic function on Br0 (x) then � 1 u(x) = u. (10) vol ∂Br (x) ∂Br (x) and 1 u(x) = vol Br (x) for all 0 < r ≤ r0 . Using this we can calculate � � d 1 1 ∂u u = dr vol ∂Br (x) ∂Br (x) vol ∂Br (x) ∂Br (x) ∂ n � 1 = �u · dS vol ∂Br (x) ∂Br (x) � 1 = �u vol ∂Br (x) Br (x) = 0 3 .

Corollary 4. and we’re done. show that subharmonic functions take their maximum on the boundary. and set c = u(y ). Proof If u has no interior maximum then it’s maximum must be on the boundary. For the second statement calculate � � �� r � u ds u = Br (x) 0 ∂Bs (x) r � = 0 u(x) vol ∂Bs (x)ds = u(x) vol Br (x). and show that super harmonic functions take they’re minimum on the boundary. By a very similar argument we can extend this result to shapes other than spheres. or 1 u(x) ≥ vol ∂Br (x) � 1 u and u(x) ≥ vol Br (x) ∂Br (x) � u Br (x) (13) (14) for superharmonic u.2 If x ∈ Rn and u is a harmonic function on Br (x) then u takes both it’s maximum and it’s minimum value on the boundary ∂Br (x). and we conclude that u = c on each sphere. By a similar argument we can also show � � 1 1 u(x) ≤ u and u(x) ≤ u vol ∂Br (x) ∂Br (x) vol Br (x) Br (x) for subharmonic u. By the maximum principle c is the average of u over any sphere surrounding y . One consequence of the mean value property is the following. Since y is a maximum we also have u ≤ c on each of these spheres. 4 . This gives u(x) = lim s→0 � u ∂Br (x) 1 vol ∂Bs (x) � u= ∂Bs (x) 1 vol ∂Br (x) � u ∂Br (x) as required.by Stokes’ theorem. Thus 1 vol ∂Br (x) is constant on 0 < r ≤ r0 . The argument for the minimum is similar. Now take a sphere that intersects with the boundary. and u takes it’s maximum on this intersection. Else take y an interior maximum.

by modifying the above argument. This is a very powerful result about harmonic functions. y to get u(x) ≤ 33n u(y ) as required. Note that u is positive and harmonic on Br (x) and that Br/3 (w) ⊂ Br (x). y ∈ Br (0). There is a constant C depending only on the dimension n such that sup u ≤ C inf u.1 Let B2r (0) be an open ball in Rn . Theorem 5. we can obtain a similar result for any radius s < 2r. For example. with several consequences. By the mean value property we have � 1 u vol Br/3 (w) Br/3 (w) � 3n u vol Br (x) Br/3 (w) � 3n u vol Br (x) Br (x) u(w) = = ≤ ≤ 3n u(x). In fact. We must show that u(x) ≤ Cu(y ). Br (0) Br (0) (15) for all functions u that are nonnegative and harmonic on B2r (0). if we have u ≥ 0 and �u = 0 on B2r (0) and inf Br (0) u = 0. Let d ≤ 2r be the distance between x and y . (16) 5 . z and z.5 Harnack Inequality Another useful property of harmonic functions is the Harnack inequality. Proof Pick x. By a similar calculation we compare w. then u is identically 0 on Br (0). and pick w and z one and two thirds of the way from x to y respectively.

�u > +|Hess u|2 . Proposition 1. ∂xi ∂xj i. Let u be a real valued function on some open subset of Rn . �u > +|Hess u|2 .j In particular. 1 .j � ∂2u ∂2u ∂ 3 u ∂u + ∂xi ∂j ∂xi ∂j ∂x2 i ∂xj ∂xj i. Rn 2 and u : Ω → R.j Using this we can prove the following.1 (The Bochner formula). Proof Proof is by calculation 1 �|�u|2 = 2 � � ∂u ∂u 1 � ∂2 ∂xj ∂xj 2 ∂x2 i i. y > indicates the usual dot product of x and y .Lecture Two: The Gradient Estimate 1 The Bochner Formula For an m × m real matrix we can deﬁne a norm by taking � |A|2 = a2 ij .j (2) = � i. then 1 �|�u|2 =< ��u.j � ∂ ∂u = (�u) + |Hess u|2 ∂xj ∂xj j = < ��u. i. then we can take the norm of the (1) � � ∂ 2 u �2 |Hess u| = . 2 where < x. if Ω is some subset of Hessian.j � ∂ � ∂ 2 u ∂u � = ∂xi ∂xi ∂xj ∂xj i.

and introduce a test function φ with φ = 0 on the boundary of B2 (0) and φ > 0 on the interior. by the maximum principle.j �(φ2 |�u|2 ) = |�u|2 �(φ2 ) + 2 � a2 ij + 4aij bij .j φ ∂φ ∂u . Theorem 2. Take u harmonic on B2 (0). Now proceed as follows. ∂xi ∂xj xi ∂xj ∂2u ∂φ ∂u ∂xi ∂xj . Note that �|x|2 = 2n. so �(p+�|x|2 ) > 0 for all � > 0.j φ ∂φ ∂ 2 u ∂u ∂xi ∂xj xi ∂xj � i. (4) 2 2 Note that a2 ij + 4aij bij + 4bij = (aij + 2bij ) ≥ 0. p + �|x|2 has no interior maximum. Step 2. Therefore. Prove the result for r = 1. and apply Bochner to simplify. Calculate �(φ2 |�u|2 ) = φ2 �|�u|2 + |�u|2 �(φ2 ) + 2�(φ2 ) · �(|�u|2 ) = 2φ2 |Hess u|2 + 2φ2 < ��u.2 The Gradient Estimate We now prove a gradient estimate for harmonic functions. so it’s maximum occurs on the boundary. Apply this to (4) to give �(φ2 |�u|2 ) ≥ |�u|2 �(φ2 ) − 8 � ij b2 ij (5) 2 . Let p : B2r (x0 ) → R be subharmonic. Show that a subharmonic function on a ball takes it’s maximum on the boundary. �u > +|�u|2 �(φ2 ) + 8 = 2φ2 |Hess u|2 + |�u|2 �(φ2 ) + 8 ∂ u Deﬁne aij = φ ∂x and bij = i ∂xj 2 � i. We can rewrite to get �� i.1 There are dimensional constants c(n) such that sup |�u| ≤ Br (x0 ) c(n) sup |u|. We will work with �(φ2 |�u|2 ). Step 1. Letting � → 0 we see that p takes its maximum on the boundary as well. r B2r (x0 ) (3) for all harmonic functions u on B2r (x0 ) ⊂ Rn Proof Note that it suﬃces to check the case x0 = 0.

by (11). (9) δB2 (0) B1 (0) Let h(n) = infB1 (0) φ > 0 and rearrange to give k (n) supδB2 (0) u2 ≥ supB1 (0) |�u|2 . This completes the proof. Step 3. Therefore. If u is harmonic on B2r (0) deﬁne u ˜(x) = u(x/r). (8) By step 1 supB2 (0) (φ2 |�u|2 + k (n)u2 ) occurs on the boundary. supB1 (x0 ) |�u so supBr (0) r|�u| ≤ c(n)supB2r (0) |u|. h(n)2 Finally. in our original notation. take square roots to get (11) supB1 (0) |�u| ≤ c(n)supB2 (0) |u| � �1/2 k(n) . � � ∂φ �2 � ∂u �2 + |�u|2 �(φ2 ) �(φ |�u| ) ≥ −8 ∂xi ∂xj 2 2 ij (6) (7) ≥ −8|�φ|2 |�u|2 + �(φ2 )]|�u|2 . Let k (n) = | inf B2 (0) (−8|�φ|2 + �(φ2 ))|. Observe that �(u2 ) = 2u�u+2|�u|2 = 2|�u|2 since �u = 0.or. and note that u ˜ is harmonic on B2 (0). It is important to note that c(n) really is a dimensional as required. ˜ | ≤ c(n)supB2 (0) |u ˜|. so �(φ2 |�u|2 + k (n)u2 ) ≥ 0 on B2 (0). Extend this to general r. so we get sup k (n)u2 ≥ sup φ2 |�u|2 . (13) (12) (10) 3 . with c(n) = h 2 (n) constant: although it depends on a choice of φ it doesn’t depend on u. Furthermore φ vanishes on the boundary.

On the annulus B1 \ B1/2 1 2 2 . Now consider �f (w) = f �� (w)|�w|2 + f � (w)�w = 4α2 e−α|x| |x|2 − 2nαe−α|x| . Deﬁne w : Rn −→ R by w(x) = |x|2 and f : R −→ R by f (t) = e−αt − e−α so that v = f (w). Let 2 v (x) = e−α|x| − e−α . so v = 0 on ∂B1 (0) and v > 0 on the interior. Theorem 1. First consider the case r = 1. (1) Proof We prove this from the maximum principle. Now apply this to u.Lecture Three: The Hopf Maximum Principle 1 Lecture Three: The Hopf Maximum Principle In this lecture we will state and prove the Hopf Maximum Principle.1 If u is an harmonic function on the closure of Br (0) ⊂ Rn . Picking α = 4n and restricting v to 1 ≥ |x| ≥ 1/2 we obtain �v ≥ 2αe−α (α/2 − n) ≥ 8n2 e−4n . and x0 on the boundary of Br (0) is a strict maximum of u (ie u(x0 ) > u(y ) for all y �= x0 ) then ∂u k (x0 ) ≥ (u(x0 ) − u(0)) ∂n r for some strictly positive dimensional constant k .

and the maximum principle applies. Deﬁne w(x) by w(x) = u(x0 ) − u(x). (2) so u + �v is subharmonic on this anulus. (3) We know that u + �v has a maximum on the outer boundary and it has to be at x0 (since v = 0 on the outer boundary). Therefore u(x0 ) − u(0) ≤ (u(x0 ) − maxB1/2 (0) u(x)). so if we choose � very small we can arrange that u + �v also takes it’s maximum on the outer boundary. It follows that ∂ (u + �v ) ≥0 ∂n and therefore that ∂u ∂v (x0 ) ≥ −� . 2C (n)(e−n − e−4n ) ∂n 2 (6) (5) . For this we need u(x0 ) + �v (x0 ) ≥ max∂B1/2 (u(x) + �v (x)) so that u(x0 ) ≥ max∂B1/2 u(x) + �(e−n − e−4n ). therefore the Harnack inequality holds. 2(e−n − e−4n ) ∂n (4) Finally we apply the Harnack inequality to get this in terms of u(0). C (n) Substituting this into (4) we obtain ∂u 8ne−4n (x0 ) ≥ (u(x0 ) − u(0)). and we get w(0) ≤ maxB1/2 (0) w(x) ≤ C (n)minB1/2 (0) w(x) for an appropriate dimensional constant C (n). Recall that u has a strict maximum on the outer boundary. Note that w is harmonic and nonnegative. We can choose �= u(x0 ) − max∂B1/2 u(x) 2(e−n − e−4n ) .�(u + �v ) = ��v > 0. Therefore the maximum of u + �v on the annulus B1 \ B1/2 occurs on the boundary. ∂n ∂n Calculating ∂v ∂n and substituting in for � we obtain ∂u 8ne−4n (x0 ) ≥ (u(x0 ) − max∂B1/2 u(x)).

so ∂u 1 8ne−4n (x0 ) ≥ (u(x0 ) − u(0)) r 2C (n)(e−n − e−4n ) ∂n as required.This completes the proof for r = 1. We will get the general case by scaling. Also if x0 ∈ ∂Br (0) is a strict maximum of u then x ˜0 = x0 /r is a strict maximum of u ˜ on the boundary. (9) 3 . If u is harmonic on Br (0) and we deﬁne u ˜(y ) = u(ry ) then u ˜ is harmonic on B1 (0). Therefore ∂u ˜ (˜ x0 ) ≥ ∂n ≥ By the chain rule ∂u ˜ x0 ) ∂n (˜ 8ne−4n (˜ u(˜ x0 ) − u ˜(0)) 2C (n)(e−n − e−4n ) 8ne−4n (u(x0 ) − u(0)). 2C (n)(e−n − e−4n ) (7) (8) ∂u = r ∂n (x0 ).

1 If u : Br → R is a C 1 function with u = 0 on ∂Br then � � 2 2 u ≤ C (n)r |�u|2 . By the Fundamental Theorem of Calculus � s f (s) = f � (x). 1 The DirichletPoincare Inequality Theorem 1. �� h2 �1/2 �� g2 �1/2 � (4) . Here the statement becomes � r � r 2 2 f ≤ kr (f � )2 −r −r (2) where f is a C 1 function satisfying f (−r) = f (r) = 0.Lecture Four: The Poincare Inequalities In this lecture we introduce two inequalities relating the integral of a function to the integral of it’s gradient. g = |f � | to get �� |f (s)| ≤ −r s hg ≤ (f ) � 2 �1/2 (r + s) 1/2 �� ≤ r (f ) −r � 2 �1/2 (2r)1/2 . Apply this with −r Recall the CauchySchwarz inequality h = 1. Br Br (1) We will prove this for the case n = 1. (5) Squaring both sides gives |f (s)| ≤ 2r −r 2 � r (f � (s))2 . (6) 1 . (3) −r Therefore |f (s)| ≤ � �� s |f � (x)|. They are the DirichletPoincare and the NeumannPoincare in equalities.

�r 1 Take a diﬀerentiable function f with A = 2 r −r f . r] with f (c) = A. r] to give � r � |f (s)|2 ≤ 4r2 −r r |f � (s)|2 (7) −r as required. 2 The NuemanPoincare Inequality � u then Theorem 2. � r � 2 2 (f (s) − A) ≤ (2r) −r r |f � (s)|2 . c From this we get |f (s) − A| ≤ �s c |f � (t)| ≤ �r −r |f � (t)|. If we deﬁne g (x) = Br (u − x)2 we can rearrange to get g (x) = Br u2 − 2x Br u − x2 . (9) Squaring and integrating then gives our result. Apply CauchySchwarz again to give r �� |f (s) − A| ≤ 2r (f ) −r � 2 �1/2 . Br Br (8) Again we give the proof in the case n = 1. and see that it occurs at x = Br . and we deﬁne A = 1 volBr Br � � (u − A)2 ≤ C (n)r2 |�u|2 . u 2 . so the volBr mean really is the best constant approximation to a function.1 If u is C 1 on Br .and ﬁnally we integrate over [−r. We then R ﬁnd the minimum by taking the derrivative. There is another interesting fact related to the NeumannPoincare � � � inequality. (10) −r It is not diﬃcult to extend these proofs to higher dimensional cubes. Note by the intermediate value theorem that there is a point c in [−r. We have � s f (s) = f (c) + f � (s).

and φ : B2r → R is nonnegative with φ = 0 on ∂B2r . then � � 2 2 |u|2 |�φ|2 . B2r 1 . (3) � � � Clearly ∂B2r φ2 u�u · dS = 0.1 Let u : B2r → R satisfy u�u ≥ 0. The statement is this.2 If u : B2r → R satisﬁes u�u ≥ 0. (4) B2r B2r B2r and so � B2r φ |�u| 2 2 � ≤ −2 � ≤ 2 φu�φ�u B2r (5) (6) φ|u||�φ||�u|.Lecture Five: The Cacciopolli Inequality 1 The Cacciopolli Inequality The Cacciopolli (or Reverse Poincare) Inequality bounds similar terms to the Poincare inequalities studied last time. r Br B2r \Br First prove a Lemma. Then � � 4 |�u|2 ≤ 2 u2 . From this � � � 2 φu�φ · �u − �(φ u)�u = −2 0≤− φ2 |�u|2 . Theorem 1. Lemma 1. but the other way around. so apply Stokes’ theorem to get B2r φ2 u�u + B2 r �(φ2 u) · �u = 0. (2) φ |�u| ≤ 4 B2r B2r (1) Proof Consider 0≤ � B2r φ2 u�u.

(8) To complete the proof of theorem 1. and consider � B2r |�(φu)| 2 � = B2r |φ�u + u�φ|2 φ2 |�u|2 + u2 |�φ|2 + 2uφ�φ · �u. Substitute this into the lemma to obtain the result. so |�φ| = 0 on Br and |�φ| = 1/r on B2r \ Br .�� 2 �1/2 �� 2 �1/2 � for any functions f and g (this is one form g Recall the inequality f g ≤ f of the CauchySchwarz inequality).1 Applications of the Cacciopolli Inequality Bounding the growth of a harmonic function One nice consequence of the Cacciopolli Inequality is the following inequality bounding the rate at which a harmonic function can decay. � = B2r Apply CauchySchwarz and lemma 1. and apply it above to get � B2r φ |�u| ≤ 2 B2r 2 2 �� φ |�u| 2 2 �1/2 �� B2r |u| |�φ| 2 2 �1/2 . namely � � 4 |�u|2 ≤ 2 u2 .1 There are strictly positive dimensional constants k (n) such that � � u2 ≥ (1 + k (n)) u2 B2r Br (10) for all harmonic functions u : B2r → R. (9) r Br B2r \Br 2 2. if r < x ≤ 2r. Proof Let φ be a test function as before.2 to get 2 . Proposition 2. (7) Dividing and squaring then gives � � 2 2 φ |�u| ≤ 4 B2r B2r |u|2 |�φ|2 .1 pick � 1 φ(x) = 2r −|x| r if |x| ≤ r.

r B2r \Br C (n)r2 B2r (11) (12) Since (φu)2 is a positive function we can reduce the area of the integration. Br B2r \Br (14) and (1 + k (n)) � Br u2 ≤ � B2r u2 . (13) φ u ≤ k (n) Br B2r \Br for k (n) = 1 10C (n) . so � � 2 k (n) u ≤ u2 . therefore � � 2 2 u2 . Proposition 2.� B2r |�(φu)| 2 � ≤ B2r φ |�u| + B2r 2 2 � � u |�φ| + 2 B2r 2 2 �� φ |�u| 2 2 �1/2 �� B2r u |�φ| 2 2 �1/2 � ≤ 2 B2r � φ2 |�u|2 + 2 u |�φ| . 3 . . B2r 2 2 u2 |�φ|2 . 2. B2r Br (16) for all harmonic functions u : B2r → R. Finally note that φ = 1 on Br . B2r ≤ 10 Now make the same choice of φ as before to give � � 10 2 |�(φu)| ≤ 2 u2 r B2r B2r \Br and apply DirichletPoincare to the left hand side to get � � 10 1 2 2 φ u ≤ 2 u2 .2 Bounding the growth of the energy of a harmonic function We will now prove a similar inequality for the Dirichlet energy of a harmonic function. (15) This completes the proof.2 There are dimensional constants c(n) such that � � 2 |�u| ≥ (1 + θ(n)) |�u|2 .

(19) r |�u| ≤ 4 Br B2r \Br Together (15) and (16) give 1 4d(n) as required. (17) To do this we use two inequalities. (18) B2r \Br B2r \Br Secondly we use Cacciopolli.Proof It suﬃces to show that � c(n) Br |�u| ≤ B2r \Br 2 � |�u|2 . � Br |�u| ≤ B2r \Br 2 � |�u|2 (20) 4 . noting that if �u = 0 then �(u + A) = 0. namely if A = volB2r \Br B2r \Br � � (u − A)2 ≤ d(n)r2 |�u|2 . Firstly we will state and use � without proof the Neumann 1 u then Poincare inequality for an annulus. to give � � 2 2 (u − A)2 . and �(u + A) = �u.

so this really is a weaker condition. We say that u is weakly harmonic on Ω if � �φ · �u = 0 (1) Ω for all diﬀerentiable functions φ with φ = 0 on ∂ Ω. We will consider operators of the form � � ∂ � ∂u Lu = aij . 2 Generalizations of the Laplacian Over the last few lectures we have proved some results about the laplacian and harmonic functions. Ω (4) and it easy to run this argument backwards for C 2 functions. The two deﬁnitions of harmonic are linked by Stokes’ theorem.j 1 .Lecture Six: More General Operators 1 The Weak deﬁnition of a harmonic function It is sometimes useful to weaken the notion of a harmonic function somewhat. If we take u to be harmonic then � (2) φ�u = 0 Ω and � φ�u · dS = 0. One way of doing this is with the notion of a weakly harmonic function. Now we will try to generalise some of these. (5) ∂xi ∂xj i. Let u be a diﬀerentiable function on some set Ω. ∂Ω (3) Applying Stokes’ theorem to this we recover the weak deﬁnition � �φ · �u = 0. One nice thing about this deﬁnition is that we don’t need u to be as smooth as we did for the old deﬁnition.

where � �φ · (A�u) = 0.2 The Cacciopolli Inequality for uniformly elliptic operators The ﬁrst result that we will generalise is the Cacciopolli inequality. These operators are said to be uniformly elllptic. A�u > 2 . 2. This is almost exactly the same inequality as for harmonic functions. There is also the concept of a weakly L harmonic function. An operator written like this is said to be in divergence form since Lu = div(A�u). Such functions are called Lharmonic. Theorem 2.1 If L is uniformly elliptic with λ|v|2 ≤< Av. (6) Ω by a similar argument to section 1 any Lharmonic function is weakly Lharmonic. We will often be interested in functions satisfying Lu = 0. We can extend many of the results we proved for harmonic functions to functions solving any uniformly elliptic operator. A�u > − B2 r B2 r � ≤ − B2 r � ≤ −2 φ2 < �u.1 Uniformly Elliptic Operators Operators where the matrix A satisﬁes λ|v|2 ≤< Av. Calculate � 0 ≤ �B2 r ≤ B2 r φ2 uLu φ2 u(� · A�u) < �(φ2 u). v >≤ Λ|v |2 for some real 0 < λ ≤ Λ and for all vectors v are of particular interest. 2.where A is a symmetric n by n matrix with entries aij (not necessarily constant). Note that if A is the identity matrix then L is simply the laplacian. λ r B2r \Br Br Proof Again we start oﬀ by introducing a test function φ with φ ≥ 0 and φ = 0 on the boundary of B2r . v >≤ Λ|v |2 and u satisﬁes uLu ≥ 0 on B2r then � � 4Λ2 2 |�u| ≤ 2 2 u2 . If there is a lower bound but not an upper bound the operator is simply said to be elliptic. A�u > � φu < �φ.

�u > ≤ 2Λ �1/2 �� u < �φ. �φ > 2 B2 r �1/2 φ < �u. φA�u > | B2 r At each point A is a real symmetric matrix with positive eigenvalues. let � 1 if |x| ≤ r. uA�φ >< φ�u. φA�u >)1/2 �1/2 �� < u�φ. A�u > < u�φ. so it deﬁnes a good norm. Therefore � � 2 λ φ < �u. φA�u > B2 r φu < �φ. φ(x) = . 2r −|x| if r < x ≤ 2r r Then � Br |�u| 2 � ≤ ≤ φ2 |�u|2 u2 |�φ|2 � B2r \Br B2 r � 4Λ2 λ2 B2 r 4Λ2 ≤ 4 2 2 λ r u2 as required. Now now work on the right hand side. �u > . �u > ≤ −2 B2 r B2 r φu < �φ. (11) Now pick φ and proceed exactly as we did for the previous Cacciopolli Inequality i. < φ�u. 3 .e. � λ B2 r φ < �u. Therefore we can apply CauchySchwarz twice to the last line to get � λ B2 r φ < �u. uA�φ > B2 r (8) �� ≤ 2 B2 r �1/2 . A�u > (7) by uniform elipticity. �u > 2 � ≤ |2 B2 r (< u�φ. φA�u > (9) Applying uniform ellipticity and rearranging gives � λ B2 r 2 �� B2 r φ < �u. (10) 2 Divide and square to get λ2 � B2 r φ2 |�u|2 ≤ 4Λ2 � B2 r u2 |�φ|2 .by Stokes’ theorem. �u > 2 � ≤ −2 �B2 r ≤ −2 � ≤ |2 < u�φ.

not just when t/m is a power of 2. and let L be a uniformly elliptic second order operator with Lf = � · A�f. and λ|v|2 ≤ v · Av ≤ Λ|v|2 . |�u|2 (2) This is very similar to a result for harmonic functions from Lecture 5. but we will not give the proof here. We start by stating two results without proof. Deﬁne α = then � �α � � t 2 u2 . u ≥ s Bs Bt > 0. 1 . There are positive constants c. and d depending only on the λ dimension and the ratio Λ such that � � u2 ≥ (1 + c) u2 (1) B2r Br 2 and � B2r � Br |�u| ≥ (1 + d) for all u with Lu = 0 on B2r . and let t = 2m s. (3) B 2m s m log(1+c) Bs log(1+c) log 2 We can rewrite (1 + c)m = em log(1+c) = 2 log 2 . Proposition 1. Instead we will work on some of the consequences.1 Let B2r be a ball in Rn . (4) It runs out that a slightly weaker version of (4) holds for all t. it is clear from (1) that if u is L harmonic (ie Lu = 0) on B2m s then � � 2 m u ≥ (1 + c) u2 .Lecture Seven: Consequences of Cacciopolli 1 Consequences of Cacciopolli In this lecture we continue to generalise some of our results about the Laplacian to uni formly elliptic second order operators.

Since u2 is positive we have � � u. 2 . If � u2 < ∞ Rn then u = 0 on Rn . u ≥ 2s Bs Bt Proof Take 2m s ≤ t ≤ 2m+1 s. s u ≥ as required. (9) Corollary 1. 2 � t 2s �α � Bs u2 (8) We can do exactly the same calculation for the Dirichlet energy to give Proposition 1. If u is L harmonic |�u | ≥ There is a nice corollary to this 2 � t 2s �β � Bs |�u|2 .3 Let t > s > 0 be real numbers and let β = on Bt then � Bt log(1+d) log 2 .Proposition 1. Bt (7) =2 (m+1)α −α s ≥2 −α � �α t .4 Take u : Rn → R an L harmonic function. If u is L harmonic on Bt then � �α � � t 2 u2 . u≤ B 2m s Bt (5) (6) We can estimate this bound by (4).2 Let t > s > 0 be real numbers. so 2mα Also 2 Plugging this into (7) gives � Bt mα � u≤ Bs � u.

if r < x ≤ 2r . so the DirichletPoincare applies. Then � u2 = � > 0. λ B2r ≤ � (11) (12) Notice that φu is zero on the boundary of B2r . Now we will prove one of the two inequalities at the start of this lecture. Also note that |�(φu)|2 = |u�φ + φ�u|2 ≤ 2|u�φ2 | + 2|φ�u|2 by CauchySchwarz.Proof Suppose u is not identically zero. Recall from the proof of the Cacciopolli inequality last time that if u is L harmonic on B2r then � �2 � � Λ 2 2 φ2 |�u|2 u |�φ| ≥ 4 λ B2r B2r for all φ ≥ 0 with φ = 0 on the boundary. (14) r2 B2r λ B2r Pick � φ(x) = as usual to give c(n) r2 � 2 u2 ≤ 2 r Br �� � � � Λ 2 4 +1 u2 . B1 (x0 ) so � Bt (x0 ) u = which goes to inﬁnity as t gets large. Putting these together gives � B2r 2 (10) |�(φu)| 2|u�φ2 | + 2|φ�u|2 B2r � � � �� Λ 2 ≤ 2 4 +1 u2 |�φ|2 . λ B2r \Br 3 (15) 1 2r −|x| r if |x| ≤ r. say u(x0 ) �= 0. and we have � � c(n) 2 |�(φu)|2 . (13) (φu) ≤ r2 B2r B2r so �� � � � � c(n) Λ 2 2 (φu) ≤ 2 4 +1 u2 |�φ|2 . 2 � t 2s �α �.

Therefore ⎛ ⎞ � Br n) ⎝1 + �� c(� �⎠ Λ 2 2 2λ + 1 as expected. u ≤ B2r 2 � u2 (16) 4 .

Lecture Eight: Maximum principles and gradient estimates 1 The Maximum Principle for more general operators Let u : Br → Rn be a C 2 function. 2 The gradient estimate for Lharmonic functions Recall the gradient estimate sup |�u| ≤ Br c(n) sup |u| r B2r for harmonic functions u. Since x is a maximum for all i. When A is the identity matrix this is exactly the maximum principle from lecture 1. and Lu ≤ 0 as expected.j Aij ∂2u ∂xi ∂xj (1) for some real n × n symmetric matrix A with continuously diﬀerentiable entries. 1 . We will prove a similar estimate for uniformly elliptic operators. and let L be a uniformly elliptic diﬀerential operator taking Lu = � i. If not then we pick coordnates at x so that A(x) is diagonal. If x is an internal maximum of u then �u = 0 and Lu ≤ 0 at x. Since all the eigenvalues of A(x) are positive we have Lu = � i bi ∂2u |(x) ∂x2 i ∂2u |(x) ∂x2 i (2) is nonpositive for some collection of positive constants bi .

1 Let L be a uniformly elliptic operator as above. � ijk Aij ∂u +2 ∂i ∂xj ∂xk ∂xk ∂3u � k ⎛ ⎞ 2 2 � ∂ u ∂ u ⎠ ⎝ Aij . Substituting � ∂3u ∂u Aij + 2λ ∂xi ∂xj ∂xk ∂xk ik L|�u| ≥ 2 2 � ijk � ∂2u ∂xi ∂xk �2 . One key part of the proof in the harmonic case was the Bochner formula. (7) Now work on the ﬁrst term. and then extend to general r. n. Proof This proof follows essentially the same steps as the proof for the earlier gradient estimate from lecture 2.Proposition 2. As before we will ﬁrst prove the case r = 1. with λ|v|2 ≤ v · (Av) ≤ Λ|v|2 for some real 0 < λ ≤ Λ. � � � ∂ � � ∂Aij ∂ 2 u ∂u ∂3u ∂u ∂2u ∂u Aij = Aij − ∂xk ∂xi ∂xj ∂xk ∂xi ∂xj ∂xk ∂xk ∂xk ∂xi ∂xj ∂xk ijk ijk = � ∂Aij � ∂ ∂u ∂u − (Lu) ∂xk ∂xk ∂xk ∂xi ∂xj ∂xk k ijk ijk ∂2u = − � ∂Aij ijk ∂u ∂xk ∂xi ∂xj ∂xk 2 ∂2u . ∂xj ∂xk ∂xi ∂xk ij (6) The last term is �vk · A�vk ≥ λ|�vk |2 by uniform ellipticity. When constants are intrioduced it is important to noctive that they depend only on A. Calculate ∂2 |�u|2 = ∂xi ∂xj ∂ � ∂ 2 u ∂u 2 ∂xi ∂xj ∂xk ∂xk k (4) (5) = 2 Therefore � k ∂3u ∂u ∂2u ∂2u + . Step 1. ∂xi ∂xj ∂xk ∂xk ∂xj ∂xk ∂xi ∂xk L|�u|2 = 2 Let vk = in gives ∂u ∂xk . We will prove a similar result for L harmonic functions. but each step is now more complicated. λ and Λ. There are constants C which depend only on the operator and the dimension of the space such that C sup |�u| ≤ sup |u| (3) r B2r Br for all Lharmonic functions u on B2r .

∂xi ∂xk 2 2 ik (13) Step 2. (9) c2 ij − 2cij dij + = λ � √ cij √ − λdij λ �2 ≥ 0.since Lu = 0. (10) 1 2 λd2 ij − 2cij dij ≥ − cij . L|�u|2 � �2 � � ∂ 2 u �2 1 � � ∂Aij ∂u +λ ≥ − λ ∂xk ∂xk ∂xi ∂xk ij k ik � � ∂ 2 u �2 1� (�Aij · �u)2 + λ ≥ − λ ∂xi ∂xk ij ik � � ∂ 2 u �2 1� 2 2 ≥ − |�Aij | |�u| + λ λ ∂xi ∂xk ij ik ⎛ ⎞ � � ∂ 2 u �2 1 � ≥ − ⎝ . |�Aij |2 ⎠ |�u|2 + λ λ ∂xi ∂xk ij ik Pick c1 = supB2r 1 λ �� � 2 > 0. λ L|�u|2 ≥ − � 1� 2 cij + λ d2 ij . We will also need to estimate 3 . λ ij i. in the old notation. ∂2u We can rewrite (8) as � ij L|�u|2 ≥ −2 Note that λd2 ij and so cij dij + 2λ � ij d2 ij .j (11) Apply this to (9) to get (12) or. Together with (7) this gives L|�u| ≥ −2 Let cij = � ∂Aij ∂u k ∂xk ∂xk 2 � ∂Aij ijk � ∂ 2 u ∂u + 2λ ∂xk ∂xi ∂xj ∂xk ik � ∂2u ∂xi ∂xk �2 . (8) and dij = ∂xi ∂xj . We have |� A | ij ij � � ∂ 2 u �2 L|�u| ≥ −c1 |�u| + λ .

∂xk ∂xi ∂xj ∂xk (15) (16) (17) Note that Lφ2 is bounded on B2 . and pick φ a test function with φ = 0 on ∂B2 . We need a bound for L(φ2 |�u|2 ).L(u2 ) = ∂u2 ∂xi ∂xj i. so we can pick a positive constant c3 such that c1 − kφ2 ≥ −c3 . (19) L(φ |�u| ) ≥ −c3 |�u| + λφ ∂xj ∂xk ∂xk ∂xi ∂xj ∂xk ∂ u ∂u ∂φ Now set γjk = φ ∂x and δjk = Aij ∂x .j � � ∂ ∂u = Aij 2u ∂xi ∂xj ∂2u ∂u ∂u = 2uAij + 2Aij ∂xi ∂xj ∂xj ∂xi ∂u ∂u = 2Aij ∂xj ∂xi � Aij since Lu = 0. Calculate L(φ2 |�u|2 ) = Aij ∂ 2 (φ2 |�u|2 ) ∂xi ∂xj ∂ 2 |�u|2 ∂ 2 φ2 ∂φ ∂ |�u|2 = φ2 Aij + |�u|2 Aij + 2Aij φ ∂xi ∂xj ∂xi ∂xj ∂xi ∂xj ∂u ∂φ ∂ 2 u = φ2 L(|�u|2 ) + L(φ2 )|�u|2 + 4Aij φ . 2 (20) Use an absorbing inequality to show that 4 . say Lφ2 ≥ c2 . and rewrite as j ∂xk k ∂xi 2 L(φ2 |�u|2 ) ≥ −c3 |�u|2 + λγik + 4γjk δjk . Then � 2 �2 ∂u ∂φ ∂ 2 u ∂ u 2 2 2 2 + 4Aij φ . ∂xk ∂xi ∂xj ∂xk L(φ |�u| ) ≥ (c2 − c1 φ )|�u| + λφ 2 2 2 2 2 (18) The function φ2 is also bounded on B2 . Assume r = 1. Now apply uniform ellipticity to get L(u2 ) ≥ 2λ|�u|2 . Applying this bound together with inequality (13) we get � ∂2u ∂xj ∂xk �2 + 4Aij φ ∂u ∂φ ∂ 2 u . (14) Step 3. and φ > 0 on the interior.

k i ≥ −c3 |�u| − c4 n k |�φ|2 |�u|2 . Substitute in for δjk and we have �2 ∂u ∂φ Aij L(φ2 |�u|2 ) ≥ −c3 |�u|2 − c4 ∂xk ∂xi i j. By steps 2 and 3 c5 2 u ) ≥ 0. Therefore it takes it’s maximum on the boundary. Thus � � � ∂u ∂φ �2 L(φ |�u| ) ≥ −c3 |�u| − c4 nk ∂xk ∂xi 2 2 2 2 j. We have shown that L(φ2 |�u|2 + sup(φ2 |�u|2 + B2 2 2 2 (22) c5 2 c5 2 u ) = sup(φ2 |�u|2 + u ).2 L(φ2 |�u|2 ) ≥ −c3 |�u|2 − c4 δjk (21) for some positive constant c4 . and since |�φ|2 is bounded on B2 L(φ2 |�u|2 ) ≥ −c5 |�u|2 for some constant c5 which depends only on the dimension and L. 2λ 2 λ ∂B2 (24) Recall that φ is zero on the boundary.k i since for any collection of real numbers the average of the squares is greater than the square of the averages. Apply the maximum principle. The functions Aij are bounded on B2 . Thus sup |�u|2 ≤ sup(φ2 |�u|2 + B1 c5 2 u ) 2 λ B1 c5 2 ≤ sup(φ2 |�u|2 + u ) 2 λ B2 c5 2 ≤ sup(φ2 |�u|2 + u ) 2λ ∂B2 5 . (23) 2λ c5 2 so. by the maximum principle. φ2 |�u|2 + 2 λ u has no interior maxima. so there is a constant k with |Aij | ≤ k . and choose it to be identically one on B1 .k � � �� ∂u ∂φ 2 2 ≥ −c3 |�u| − c4 n Aij ∂xk ∂xi � � � j. Step 4.

and deﬁne u� (x) = u(rx). Scale for general r. 6 . 2λ B2 sup |�u| ≤ B1 c5 sup |u| 2λ B2 (25) as expected.≤ ≤ Finally take square roots to obtain c5 sup u2 2λ ∂B2 c5 sup u2 . step 5. Assume u is Lharmonic on B2r . Plugging u� into (25) we get r sup |�u| ≤ Br c5 sup |u| 2λ B2r (26) as required. Then u� is L harmonic on B2 .

Theorem 1. There are constants C which depend only on L and the dimension such that ∂u C (1) |(x) ≥ (u(x) − u(0)).Lecture Nine: Hopf and Harnack Revisited 1 The Hopf maximum principle for uniformly elliptic oper ators The next result that we will generalise is the Hopf Maximum principle. Proof This proof is similar to the earlier version. r ∂n We will actually prove that ∂u C (2) |(x) ≥ (u(x) − sup u).1 (The Hopf Maximum principle for uniformly elliptic operators) Let u be an L harmonic function on Br (0) with a strict maximum at x ∈ ∂Br (0) . Calculate Lv = Aij ∂2v ∂xi ∂xj � ∂ � 2 = Aij −2αxi e−α|x| ∂xj � � 2 = −2Aii α + 4Aij α2 xi xj e−α|x| � � 2 ≥ −2Aii α + 4α2 λ|x|2 e−α|x| 1 2 (3) .1 will then follow easily once we have a harnack inequaltity for elliptic operators. We will prove the case r = 1 and claim that the general result follows by scaling exactly as it did for the previous Hopf maximum principle. Let α be a constant. r ∂n Br Theorem 1. As before we will consider uniformly elliptic operators L taking Lf = Aij ∂2f ∂xi ∂xj with λ|v|2 < v · Av ≤ Λ|v|2 for some real 0 < λ ≤ Λ. though a bit more complicated. and deﬁne v (x) = e−α|x| − e−α .

so the maximum of u + �v is at x. substitute in. Then sup Br |�u| c ≤ r u (9) for some dimensional constant c (ie c depends on the dimension of the space. (4) and so picking α large we can get Lv ≥ 0 on the annulus. Clearly this is subharmonic on B1 \ B1/2 . 2(eα/4 − e−α ) Br (7) (8) The result then follows from the Harnack inequality. so it takes it’s maximum on either the inner or the outer boundary. We’ll pick � so that it occurs at x. 2(eα/4 − e−α ) (5) Also note that v is zero on the outer boundary. Proposition 2. It is based on a gradient estimate that is slightly stronger than the one we proved. It follows that ∂ (u + �v ) � � � ≥ 0. We need u(x) + �v (x) ≥ sup(u + �v ) Br Evaluating v gives u(x) ≥ sup(u + �(e−α/4 − e−α )).1 Let u be a positive harmonic function on B2r . ∂n (x) Calculate ∂v � ∂ n � (x) (6) � = −2αe−α . 2 .by uniform elipticity. and rearrange to get ∂v � ∂u � � � ≥ −� � � ∂ n (x) ∂ n (x) 2αe−α ≥ (u(x) − sup u). 2 Another proof of the Harnack inequality We will now give an alternative proof of the Harnack inequality. Consider u + �v . Br therefore choose �= u(x) − supBr u . but not on u). Restricting to B1 \ B1/2 we have � � 2 Lv ≥ α2 λ − 2αAii e−α|x| .

and let γ1 be the straight path from x to 0. and γ2 the straight path from 0 to y . Pick x and y in Br . Note that |�u| u = |�(log u)|. Taking exponents e−2c ≤ and so c r 0 u(y ) ≤ e2c .We will prove this next time. and calculate �� � � � � � � (log u ) · dx � (log u ) · dx + | log u(y ) − log u(x)| = � � � γ2 γ1 � 1 � 1 ≤ |x| |�(log u(sx)|ds + |y | |�(log u(sy ))|ds 0 (10) (11) (12) (13) ≤ (|x| + |y |) ≤ 2c. 3 . u(x) (14) sup u ≤ e2c inf u Br Br (15) as required. We can derive the Harnack inequality from this as follows.

As usual we take a nonnegative test function φ : B2 → R with φ = 0 on |�u|2 u ∂B2 . Consider �(wφ4 ) + 2�v · �(wφ4 ) = �(wφ4 ) + 2φ4 �v · �w + 2w�v · �φ4 . Theorem 1. ∂xi ∂xj Now try to ﬁnd our quartic bound. We start by bounding �(wφ4 ) + 2�v · �(wφ4 ) by a quartic polynomial in w1/2 φ. Deﬁne v = log u and w = |�v |2 . Proof We will prove the result for r = 1 and claim that the general case follows immedi ately by scalng.1 There are dimensional constants c such that sup Br |�u| c ≤ r u (1) for all positive harmonic functions u : B2r → R. Calculate �w = �|�v |2 � 2 �2 ∂ v = 2 + 2 < ��v. (4) (2) (3) 1 . Therefore �(wφ4 ) = φ4 �w + 2�φ4 · �w + w�φ4 � 2 �2 ∂ v 4 = 2φ − 2φ4 < �w. �v > +2�φ4 · �w + w�φ4 .An Improved Gradient Estimate for Harmonic Functions 1 The new gradient estimate Last lecture we used an improved form of the gradient estimate for harmonic functions. �v > ∂xi ∂xj � 2 �2 ∂ v − 2 < �w. �v > = 2 ∂xi ∂xj by the Bochner formula. Note that �v = � u and �v = − u2 = −w . We will now prove it.

**Substitute for �(wφ4 ) from (3) to get �(wφ ) + 2�v · �(wφ ) = 2φ
**

4 4 4

�

∂2v ∂xi ∂xj

�2

+ 2�φ4 · �w + w�φ4 + 2w�v · �φ4 .

(5)

We need to write the second term out in terms of partial derivatives. Calculate ∂φ4 ∂ |�v |2 ∂xi ∂xi ∂φ4 ∂v ∂ 2 v = 4 , ∂xi ∂xj ∂xj ∂xi

2�φ4 · �w = 2

**so � ∂2v ∂xi ∂xj �2 +4 ∂φ4 ∂v ∂ 2 v + w�φ4 + 2w�v · �φ4 . (6) ∂xi ∂xj ∂xj ∂xi
**

2

�(wφ ) + 2�v · �(wφ ) = 2φ

4

4

4

∂ v and bij = Use an absorbing inequality to simplify. Let aij = φ2 ∂x i ∂xj 2 2 aij + 4aij bij ≥ −4bij . Together with (6) we have

∂φ4 ∂v ∂xi ∂xj .

Note that

�(wφ ) + 2�v · �(wφ ) ≥ φ ≥ φ

4

4

4

� � �

∂2v ∂xi ∂xj ∂2v ∂xi ∂xj ∂2v ∂xi ∂xj

�2 �2 �2

4 4 − 4b2 ij + w �φ + 2w �v · �φ

4

− 4|�φ4 |2 |�v |2 + w�φ4 + 2w�v · �φ4 − 16φ6 |�φ|2 |�v |2 + (4φ3 �φ + 12φ2 |�φ|2 )w − 8φ3 w|�v ||�φ|

≥ φ4

since φ and w are both nonnegative. Observe that φ, |�φ|, and �φ are bounded, so there are constants c1 , c2 , c3 such that � ∂2v ∂xi ∂xj �2

�(wφ ) + 2�v · �(wφ ) ≥ φ

4

4

4

− c1 φ2 |�v |2 + c2 φ2 w − c3 φ3 w|�v |.

(7)

Recall that for any collection of real numbers the average of the squares is greater than the square of the average. Thus for any matrix A � 2 � 2 �� � Aij Aii Aii 2 . ≥ ≥ n n n

2

Apply this above to give � ∂2v ∂xi ∂xj �2 1 ≥ n �� ∂2v ∂x2 i �2 = (�v )2 , n

and substitute this into (7) ; �(wφ4 ) + 2�v · �(wφ4 ) ≥ φ4 (�v )2 − c1 φ2 |�v |2 + c2 φ2 w − c3 φ3 w|�v |. n (8)

Observe that �v = −|�v |2 = −w, so we have �(wφ4 ) + 2�v · �(wφ4 ) ≥ φ4 w2 − c1 φ2 w + c2 φ2 w − c3 φ3 w3/2 . n (9)

This is the bound we were looking for. Now apply this. Since φ is zero on the boundary of B2 , φ4 w takes it’s maximum in the interior. Let x be the maximum. At x, �(φ4 w) = 0 and 0 ≥ �(φ4 w), so 0 ≥ 1 (φw1/2 )4 + (c2 − c1 )(φw1/2 )2 − c3 (φw1/2 )3 . n (10)

This is a quartic polynomial in φw1/2 with positive leading coeﬃcient. Such polynomials are positive for large argument, so there is a constant k with |φ(x)(w(x))1/2 | ≤ k . Note that k depends only on the coeﬃcients of the polynomial. The coeﬃcients themselves depend only on dimension, so k also depends only on dimension. Choose 0 ≤ φ ≤ 1. Then sup φ4 w = (φ(x))4 w(x) ≤ (φ(x)(w(x))1/2 )2 ≤ k 2 .

B2

(11)

Finally we choose φ to be identically one on B1 , so |�u|2 u2

sup

B1

= sup φ4 w

B1

(12) (13) (14)

**≤ sup φ4 w ≤ k and take square roots to give our result.
**

B2 2

The gradient estimate we proved earlier follows easily from this; this is a stronger result. As we saw last time the Harnack inequality is also a reasonably straightforward consequence. The only major annoyance is that we needed u > 0.

3

Solving the Laplace equation in R2: The Dirichlet problem

In this lecture we will study something called the Dirichlet problem for discs in R2 . Given a continuous function f on the boundary of a disc Br we will try to construct a harmonic function u on the entire disc so that f and u agree on ∂Br . This turns out to be a lot easier in polar coordinates.

1

The laplacian in polar coordinates.

The ﬁrst problem is to calculate the laplacian in polar coordinates. If u : R2 → R, then ∂ 2 u 1 ∂u 1 ∂2u + + . (1) ∂r2 r ∂r r2 ∂θ2 This is a straightforward consequence of the chain rule, but the calculation is rather tedious. Now look for harmonic functions of the form u(r, θ) = g (r)h(θ). Plug this into (1) and �u = h Multiplying by

r2 u

g ∂2h ∂ 2 g h ∂g + 2 2 = 0. + 2 ∂r r ∂r r ∂θ

(2)

we get 1 ∂2h r2 ∂ 2 g r ∂g + + = 0. g ∂r2 g ∂r h ∂θ2 (3)

The ﬁrst two terms depend only on radius, and the last term depends only on θ, so both must be constant. We have h�� = c, (4) h and r2 g �� rg � + = −c. (5) g g We can solve these separately. From (4) we get √ ⎧ (√c)θ + be−( c)θ if c > 0 ⎨ ae h= a + bθ if c = 0, or √ √ ⎩

a cos( −cθ) + b sin( −cθ) if c < 0, 1

It turns out that all harmonic equations on the disc are of this form. From above we need c = −k 2 . Plugging this into (5) we get rm (m2 − k 2 ) = 0. Together with our solution for h we have u constant or u(r. Our solutions are h constant or h = a cos kθ + b sin kθ for some integer k . Then v is harmonic on BR . In order for g to be well deﬁned at the origin we need a1 = 0. so g = a0 + a1 log r . Once again we need a1 = 0 so that g is well deﬁned at the origin. (8) We have found lots of harmonic functions on R2 . so f= �∞ k −k k=0 r (R ck ∞ � k=0 (ck cos kθ + dk sin kθ) (9) Deﬁne v = cos kθ + R−k dk sin kθ).but h has to be periodic with period 2π . It also restricts the possibilities if c < 0. Then we get rg �� + g � = 0 1 This is a ﬁrst order equation in g’ which we can solve to get g � = a r . We can expand f as a fourier series. (6) so we need m = ±k . 2 . when we need c = −k 2 for some integer k . First we’ll deal with the case c = 0. and v = f on the boundary as required. and g (r) = a0 rk + a1 r−k . Since the sum of any collection of harmonic functions is still harmonic we also get solutions of the form v= ∞ � k=0 rk (ak cos kθ + bk sin kθ). This follows immediately from the fact that two harmonic functions which agree on the boundary agree over the entire set. θ) = rk (a cos kθ + b sin kθ) (7) for some integer k . Take a continuous function f on the boundary of a disc BR . and we can now solve the Dirichlet problem fairly easily. This immediately excludes the possibility that c > 0. Solving (5) is a little trickier. We’ll try for solutions of the form g = rm . Now deal with c �= 0. so if c = 0 then the only solutions are constants. and if c = 0 we need h to be constant.

= F t. �u = 1 A gradient estimate for a torus |�u|2 1 ∂u n − ≤ . since this turns out to be easier than doing the proof for a ball. Also deﬁne F = t(|�f |2 − ft ). Observe that estimate (� − ∂t �F = t(�|�f |2 − �ft ) � 2 �2 ∂ f = 2t + 2t < ��f.The Heat equation. One example is the heat equation. Note that we actually want to bound ∂ )F . The heat equation is ∂u . and calculate Proof For this proof we will use the notation gt = ∂ )f ∂t ∂g ∂t .[t1 > t2] We’ve spent a lot of time concentrating on the laplace equation. �f > −t�ft ∂xi ∂xj 1 We need to (3) (4) . (� − 1 ∂u �u |�u|2 − − 2 u u ∂t u |�u|2 = − 2 u = −|�f |2 . u ∂t 2t u2 Theorem 1. which we will study in this lecture. Consider a function u : Rn × R → R of both time and space. (1) ∂t In this lecture we will prove a gradient estimate and a Harnack inequality for functions satisfying the heat equation on a torus T n = S 1 × S 1 × · · · × S 1 . but there are other important PDE’s.1 If u is positive and satisﬁes the heat equation on the cylinder T n × R then (2) Deﬁne f = log u.

and let η (t) = (x2 . Therefore �F ≥ 2t(�f )2 + 2t < ��f. t1 ) − (x2 . for any matrix Aij . so Ft = |�f |2 − ft + t(2�f · �ft ) − t(�f + |�f |2 )t = |�f | − ft − t�ft . ds (13) 2 . 2 (7) (8) (9) (10) (11) (12) 2 A Harnack inequality for a torus Now we’ll try to get a Harnack inequality out of this. u 2t u2 which is what we wanted. �F ≤ 0 and Ft = 0. n (5) We also have �f = −|�f |2 + ft = − F t . ∂t nt t At a maximum of F we have �F = 0. so �F ≥ Now work on Ft . nt (6) Ft = |�f |2 − ft + t(2�f · �ft ) − tftt .P � ( i Aii )2 by Bochner. t1 ) and (x2 . t1 ) − f (x2 . and it essentially because the average of the square is greater than the square of the average). t2 ) with t2 ≥ t1 . �f > −t�ft . �f > −t�ft .j A2 ≥ (We saw this in lecture ij n 10. t2 ) + t((x1 . Therefore (� − 2 nF (F 2 − ) nt 2 Therefore F ≤ n 2 . Recall that. i. t2 ) = 0 1 df (η ) ds. Substituting in for F gives 0≥ n |�u|2 ut − ≤ . Clearly 2F 2 − 2 < �F. �f > − . Putting together (6) and (9) we get ∂ 2F 2 F )F ≥ − 2 < �F. Pick (x1 . Note that �f + |�f |2 = ft . t2 )) be the straight line path from one to the other. Then � f (x1 .

By inequality (12) ft (t1 − t2 ) ≤ n (t2 − t1 ) − |�f |2 (t2 − t1 ). |x2 − x1 | n t2 + log . 2t Together with (13) we get � f (x1 . 2t (14) The integrand is a quadratic in |�f |2 with negative leading coeﬃcient.Calculate df (η ) ds = �f · (x1 − x2 ) + ft (t1 − t2 ). t1 ) − f (x2 . u(x1 . 4(t2 − t1 ) (16) and for the second � � � 1 n 1 n t1 1 n 1 n t2 (t2 − t1 ) ds = (t2 − t1 ) ds = − dv = log . x1 ) ≤ Taking exponents we get a harnack inequality.t1 ) − log u(x2 . t2 ) � t2 t1 �n/2 exp � |x2 − x1 | 4(t2 − t1 ) � . t2 ) ≤ 0 1 |�f ||x2 − x1 | − |�f |2 (t2 − t1 ) + n(t2 − t1 ) ds.t1 ) ≤ u(x2 . + s ( t − t ) t1 2 2 2 t t v 2 2 1 2 0 0 t2 Putting these together we have log u(x1 . (15) |x2 − x1 | − 2(t2 − t1 ) 2(t2 − t1 ) 2t We split this up. t1 2 4(t2 − t1 (17) (18) (19) 3 . t2 ) ≤ 0 1 � � |x2 − x1 | |x2 − x1 | 2 n (t2 − t1 ) + (t2 − t1 ) ds. for the ﬁrst part � 0 1 |x2 − x1 | |x2 − x1 | − 2(t2 − t1 ) � |x2 − x1 | 2(t2 − t1 ) �2 (t2 − t1 )ds = |x2 − x1 | . so � f (x1 . so it has a maximum |x2 − x1 | at |�f | = 2( t2 −t1 ) . t1 ) − f (x2 .

Take u : B2 × [0. Let φ : B2 → R with φ = 0 on ∂B2 and φ = 1 on B1 .Heat equation gradinet estimate on a ball 1 generalising the proof from last time Last time we proved a gradient estimate for functions that satisfy the heat equation on a cylinder. +F (�− ∂ (�− )φk F ≥ φk ∂t F − 2 < �F.∞) → R with �u = ∂u ∂t . We have ∂ k ∂ ∂ )φ F = φk (� − )F + F (� − )φk + 2�F · �φk . or goes any further in the proof. ∂t ∂t ∂t Last time we showed that � 2 �2 F ∂ ∂ f − 2 < �F. (� − )F ≥ 2t ∂t ∂xi ∂xj t (� − and we can use the trick of saying � �2 � ∂2f 1 � ∂2f (�f )2 F2 ≥ = = n ∂xi ∂xj n nt2 ∂x2 i so � F2 +t nt � ∂2f ∂xi ∂xj �2 � (1) (2) ∂ k )φ +2�F ·�φk . and F = |�f |2 − ∂f ∂t . To do this we will need to introduce a cutoﬀ function. �f > − . To do this we needed to know that the function didn’t have a maximum on a boundary. �f > − t 1 . and try to estimate (� − ∂t )φ F . As before we deﬁne ∂ k f = log u. can you point out mhow I’m meant to ﬁnish it? thanks. We’ll now generalise this proof to work on a ball Br . ∂t (3) k and then I think we’re meant to use the maximum principle on φ f to ﬁnish it oﬀ but neither my notes nor Yaims actually states the result we’re aiming for. Again we will prove the case r = 1.

Lecture 14: A gradient estimate for the heat equation on a ball. At this maximum d(φr F ) ≥ 0. 1 Adapting the proof of the gradient estimate to Rn Last time we proved a gradient estimate for solutions of the heat equation on a torus. and note that the result we u|2 t want follows immediately from this . T ]. d dt (2) 0 ≥ − (φr F ) φr − φr dF dt (3) (4) ≥ φr F + 2∇F · ∇φr + F We need to estimate some of these. Calculate (φr F ) ≤ 0. ∞) → R is positive and satisﬁes the heat equation then t |∇u|2 ut − u2 u ≤ n . If F is non positive the result is trivially true. Let φ be a u2 cutoﬀ function on the ball B1 (0) with 0 < φ < 1 on the interior and φ = 0 on the boundary. Deﬁne f = log u and F = t |∇ −u u . We work on an interval [0. and φr ∇F = −F ∇φr dt We’ll use these to get a bound on F . We start with 1 . 2 (1) We will sketch the proof. In this lecture we will adapt the argument to prove the same theorem on Rn . We can stretch this to get a cutoﬀ function φr on Br (0) by taking φr (x) = φ(x/r) . The calculations are very similar to last time. Theorem 1.1 If u : Rn × [0. so we can assume that φr F has an interior maximum without loss of generality.

Calculate f= + 2t∇( f ) · ∇f − t ft u u 2 ∂ ∂u/∂xi ∂xi u − |∇u|2 u2 = −F t to get (8) F = 2t Recall the inequality ( Ai i)2 ≤ n to the hessian of f to give F = ≥ ∂2f ∂xi ∂xj − 2∇F · ∇f − t ft . so F 2F 2F 2 2F + ∇φr · ∇f − − |∇φr | + F nt φr t φr 2F 2 1 |∇φr |2 φr ≥ F φr + ∇φr · ∇f − − 2 + 2 nt φr t φr φr 2 0 ≥ φr φr .F = t |∇u|2 ut − u2 u ∂2f ∂xi ∂xj = 2 (5) (6) (7) = t (|∇f |2 − ft ) = 2t by Bochner. We have Ft = |∇f |2 − ft + t(2∇f · ∇ft ) − tftt . (15) (16) . so Ft = |∇f |2 − ft + t(2∇f · ∇ft ) − t( f + |∇f |2 )t F − t ft . 10 and 13 together we get 0 ≥ φr 2F 2 F − 2∇ F · ∇ f − nt t + 2∇F · ∇φr + F φr . = t Putting 4. and apply it 2t ( f )2 − 2∇F · ∇f − t ft n 2F 2 − 2∇F · ∇f − ft . (14) (11) (12) (13) Recall that φr ∇F = −F ∇φr . (A2 ii ) for all matrices A from last time. and f + |∇f |2 = ft . nt (9) (10) We also need an estimate on Ft .

3 . and get 0≥F 2F 1 − 2 |∇f |2 − nt t .∂f 1 ∂φr r ∂f − Now use an absorbing inequality ∂φ ∂xi ∂xi ≥ − ∂xi ∂xi that 1 ∇φr · ∇f ≥ − |∇φr |2 − |∇f |2 2 2 for all > 0. Consequently 0 ≥ F φr 2F 2 − nt φr 1 |∇φr |2 + |∇f |2 − 1 |∇φr |2 φr −2 + 2 t φr φr . (18) Let r → ∞ so that |∇φr | and φr tend to zero and φr → 1. (19) Finally we let → 0 and recover 0≥ F t 2F −1 . to show (17) for all > 0. n (20) From this we getF ≤ n/2 as required.

0) ≤ −2T Ω 2 � |�u(·. Campanato’s lemma. If u solves the heat equation on Ω ⊂ Rn with u = 0 on the boundary then � � � � d 2 u = 2 uut = 2 u�u = −2 |�u|2 . we will now bound the decay of the Dirichlet energy and the L2 norm (ie u2 ) of solutions. (5) 2 Campanato’s Lemma Before discussing Campanato’s lemma we need the concept of Holder continuity. 0)| ≤ Ω 2 � u (·. t)|2 .The decay of solutions of the heat equation. so � � � d 2 2 −2 |�u(·. (2) dt Ω Ω Ω Ω and d dt � Ω |�u|2 = 2 � �u · �ut = −2 Ω � ut �u = −2 Ω � Ω |�u|2 . (3) Also note that the dirichlet energy is decreasing. Deﬁnition Let f : Rn → R be a continuous function. T )|2 . f is α continuous at x0 if there are constants k and 0 < α ≤ 1 such that f (x0 + x) − f (x0 ) ≤ k |x|α 1 (6) . 0)| ≤ u ≤ −2 |�u(·. dt Ω Ω Ω and � −2T Ω (4) |�u(·. T ) − Ω 2 � u (·. and Morrey’s Lemma 1 The decay of solutions of the heat equation A few lectures ago we introduced the heat equation �u = ut (1) for functions of both�space and time.

2r | 2 �2 � 1 = f f− vol B2r (x) B2r (x) Br (x) ��2 � � � � 1 1 = f− f vol Br (x) Br (x) vol B2r (x) B2r (x) � �2 � 1 ≤ |f − Ax.r − Ax.for small x0 . Let x ∈ Rn .2r | . so (f − Ax.1 (Campanato’s Lemma) If there is a constant c such that � (f − Ax. vol B2r (x) � 2n+2α c2 and we can pick a new dimensional constant C = so that vol B1 (x) |Ax.r )2 ≤ cr2α+n Br (x) (7) (8) for all x and for small r then f ∈ C α .2r to � B2r (x) 22n |Ax. Ax. Apply this with g = 1 and h = f − Ax.r − Ax.r = vol Br (x) Br (x) Campanato’s Lemma is then Lemma 2. and deﬁne � 1 u.2r | vol Br (x) Br (x) � �2 � 2n ≤ |f − Ax. Proof First we will bound |Ax.r − Ax. If f is everywhere Holder continuous for some constant α then we say that f ∈ C α. 8.2r | ≤ Crα .r − Ax.2r | Calculate |Ax.2r )2 .2r |2 ≤ |Ax. (13) 22n c(2r)2α+n .r − Ax.2r | ≤ vol B2r (x) Now use the condition. 2 (14) (15) . vol B2r (x) B2r (x) � 1 vol Br (x) � �2 ≤ �� g2 2 (9) (10) (11) (12) By CauchySchwarz get �� gh � �� � h2 . Now take f continuous on Rn .

1 − 2− α (20) 1 Csα 2−α .s | ≤ vol Br (x) 2 (21) (22) (23) � Br (x) (f − Ay.s )2 (24) Pick r = |x − y | and s = 2|x − y | so that Br (x) ⊂ Bs (y ).2−i r . so we can use the usual formula to get |Ax.s � � vol Br (x) Br (x) � � �2 � � � 1 � � = � (f − Ay.s )2 vol Br (x) Bs (y) � vol Bs (y ) 1 ≤ (f − Ay. 1−2−α 1 − 2−kα α −α Cr 2 . 1 − 2− α (19) Now pick another point y ∈ Rn .2−k r − Ax.s )2 vol Br (x) vol Bs (y ) Bs (y) 3 (25) (26) .r − Ay.r − Ay.r − Ay.2−k r − Ax.s |. Clearly we also have |f (y ) − Ay. Note that Ax.r | ≤ 1 Crα 2−α .r − Ay.r | ≤ Now let k → ∞ to get |f (x) − Ax. Therefore � 1 |Ax.s |2 ≤ (f − Ay.2−k r − Ax.We apply this to our problem.s |2 = � f − Ay.2−i−1 r − Ax.s | . (18) This is simply the sum of a geometric series.s | ≤ Now we’ll estimate |Ax. vol Br (x) Br (x) and apply CauchySchwaz as before to get 1 |Ax. Calculate � �2 � � � 1 � � |Ax.s )� � vol Br (x) Br (x) � �2 � � 1 ≤ |f − Ay.r = so i= k −1 � i=0 k −1 i= � i=0 �i=k−1 i=0 Ax.2−i r | C (2−i−1 r)α k −1 i= � i=0 (16) (17) ≤ Cr 2 α −α 2−iα .2−i−1 r − Ax. |Ax.r | ≤ ≤ |Ax.

r − Ay.s | + |Ay. Proof This is a straightforward consequence of Campanato and Poincare. (29) (28) vol Bs (y ) 2α+n cs . At last we can calculate |f (x) − f (y )| ≤ |f (x) − Ax.s |2 ≤ 2n c(2|x − y |)2α+n . If there is a constant c1 with � |�f |2 ≤ c1 rn−2−2α (33) Br (x) for all x and small r then f ∈ C α. The result then follows by Campanato.r ) ≤ c2 r |�f |2 ≤ c1 c2 rn+2α (34) Br (x) Br (x) by hypothesis. so we can take 2|x − y | ≤ 1 to get |Ax.1 (Morrey’s lemma) Let f : Rn → R be continuous.s |2 ≤ Plugging in our values for r and s we have |Ax. vol Br (x) (27) (30) (31) (32) 3 Morrey’s Lemma Morrey’s lemma is very similar to Campanato’s lemma.s |2 ≤ 2n c(2|x − y |)2α .and.r − Ay. and we’re only interested in small |x − y |.r | + |Ax. |Ax. By Poincare there is a constant c2 such that � � 2 2 (f − Ax.r − Ay.s − f (y )| √ 1 1 ≤ Crα 2−α + 2n/2 c(2|x − y |)α + Csα 2−α − α 1−2 1 − 2−α � � 1 C 2−α n/2+α √ + 2 C |x − y |α . but uses a condition on Dirichlet energy rather than L2 norm. 4 . Lemma 3.r − Ay. c + ≤ 1 − 2− α 1 − 2− α This completes the proof. by our hypothesis.

and x ∈ Rn .t . (5) Proof Pick y ∈ Br (x).r | ≤ c4 2 � r �n+2 � s Bs (x) |�f − (�f )x. There are constants ci such that � � r �n � f 2 ≤ c1 f 2. (4) for all functions f that are harmonic on Bs (x) with Ax.1 Let r and s be real numbers with 0 < r ≤ s.2 If f is harmonic on B2r (x) then 2n sup f ≤ vol B2r (x) Br (x) 2 � Br (x) f 2. (1) s Br (x) Bs (x) � � r �n+2 � (f − Ax. By the mean value property (from lecture 1) � 1 f (y ) = f.r )2 ≤ c2 (f − Ax.Five inequalities for Harmonic functions In this lecture we will prove ﬁve inequalities for harmonic functions. vol Br (y ) Br (y) so 1 (6) .t the averages of f and �f over Bt (x) respectively. (�f )x. (2) s Br (x) Bs (x) � � r �n � |�f |2 .s |2 . the mean value inequality. Proposition 1.s )2 . (3) |�f |2 ≤ c3 s Bs (x) Br (x) and � Br (x) |�f − (�f )x. Before proving these we will prove another inequality. 1 Bounding integrals of Harmonic functions Proposition 1.

If r ≤ s ≤ 2r then � Br (x) f ≤ Bs (x) 2 � f2 � � 2r n ≤ f2 s Bs (x) � �n � n r ≤ 2 f 2. Note that Br (y ) ⊂ B2r (x). so we can expand the area of integration to get f 2 (y ) ≤ ≤ Therefore sup f 2 ≤ Br (x) � 1 f2 vol Br (y ) B2r (x) � 2n f 2. s Bs (x) � (14) (15) (16) If instead 2r ≤ s then sup f 2 ≤ Br (x) sup f 2 Bs/2 (x) 2n (17) � Bs (x) ≤ vol Bs (x) 2 f2 (18) . vol B2r (x) B2r (x) (11) (12) 2n vol B2r (x) � B2r (x) f2 (13) as required. Now we’ll use this to get our ﬁrst inequality.� f (y ) = � = � ≤ ≤ 2 1 vol Br (y ) 1 vol Br (y ) � f Br (y ) �2 (7) �2 f Br (y ) �2 �� (8) (9) (10) � �� � �2 �� 1 2 2 f 1 vol Br (y ) Br (y ) Br (y ) � 1 f2 vol Br (y ) Br (y) by Cauchy Schwarz.

Now we’ll ∂f ∂xi prove 2. ∂f = Note that � ∂x i � r �n s f 2 ≤ 2n � r �n � s Bs (x) f2 (21) Br (x) ∂ ∂xi �f = 0. (22) Now apply this. Clearly |f (z ) − f (y )| ≤ |z − y | supBr (x) |�f | ≤ 2r supBr (x) |�f |. First consider the case 4r ≤ s.s )2 . For r ≤ s ≤ 4r we simply note that 3 . (26) 2r sup |�f | ≤ 4r2 sup |�f |2 ≤ 4r2 sup |�f |2 Bs/4 (x) 1 ≤ 4r vol Bs/2 (x) 2 � Bs/2 (x) � � 2 Apply Cacciopolli to get B (x) |�f |2 ≤ s1 2 B (x) (f − Ax. Since inequality to get 2 is harmonic we can apply the mean value � B2r (x) 1 sup |�f | ≤ vol B2r (x) Br (x) |�f |2 . Therefore 3 follows immediately from 1. By the intermediate value theorem there is y ∈ Br (x) with f (y ) = Ax.r )2 ≤ 1 vol Br (x) Br (x) � Br (x) � Br (x) �2 (23) (24) (25) |�f |2 .r . so s s/2 � � 1 4r2 2 (f − Ax. s vol Bs/2 (x) Bs (x) vol Br (x) Br (x) and � Br (x) (27) (f − Ax.by the mean value inequality Therefore 1 vol Br (x) � Br (x) f2 ≤ ≤ 1 vol Br (x) 2n vol Bs (x) . Pick z ∈ Br (x).s ) .r )2 ≤ 2n+2 � r �n+2 � s Bs (x) (f − Ax. so � Br (x) � 2n vol Bs (x) � Bs (x) � f2 (19) (20) � Bs (x) f 2.r ) ≤ 2 (f − Ax.s )2 (28) as required. Therefore 1 vol Br (x) � Br (x) (f − Ax. and the ration of the volumes is � for large s as well.

The ﬁnal inequality. 2. In this case the constants ci depend on the operator. We can also prove 1. 3.� Br (x) (f − Ax. 4 .r ) ≤ 4 2 n+2 � r �n+2 � s Bs (x) (f − Ax. 4. operator taking Lu = Aij ∂x i ∂xj Proofs are omitted. follows from 2 in exactly the same way that 3 follows from 1. (29) This completes the proof of 2.r )2 . and 4 for L harmonic operators when L is a uniformly elliptic ∂2u .r ) ≤ 4 2 n+2 � r �n+2 � s Br (x) (f − Ax.

Lv ∂xj ∂xj ∂xi ∂xj (4) � is also a uniformly elliptic operator with the same constants λ. and deﬁne � 1 �x.r = u. but we need to do some simpliﬁcation ﬁrst. Λ as L. First calculate 1 . 1 Operators in divergence form Theorem 1. To apply Morrey we need to estimate |�u|2 . (5) vol Br (x) Br (x) The inequality (a�+ b)2 ≤ 2a2 + 2b2 will be very useful in this proof. Let v Note that L � an L harmonic function with v − u on the boundary of a ball Bs (x0 ). (2) vol Br (x) Br (x) and (�f )x.r = A v. We’ll consider operators in divergence form. Then u is Holder continuous.r = 1 vol Br (x) � (�f ). As before we deﬁne � 1 Ax.1 Let L be a uniformly elliptic operator taking Lu = ∂ ∂u Aij ∂xj ∂xj (1) with 0 < λ|v |2 ≤ v · (Av ) ≤ Λ|v |2 as usual. Eventually we will prove this via Morrey’s lemma. Br (x) (3) � taking Given a point x0 ∈ Rn we can deﬁne a new operator L 2 � = ∂ Aij (x0 ) ∂v = Aij (x0 ) ∂ v .Lecture 17: Holder continuity of LHarmonic operators part I Over the next few lectures we’ll prove that L harmonic functions are Holder continuous. and let u be an L harmonic function.

Lemma 1.r |2 .r ) (A 0 0 2 �� �2 1 = (v − u) vol Br (x0 ) Br (x0 ) � ≤ (v − u)2 Br (x0 ) (8) (9) by CauchyScwarz.r ) ≤ 6 (u − v ) + k (12) (v − A 0 s Br (x0 ) Br (x0 ) Bs (x0 ) Similarly we can show � � 2 |�u − (�u)x0 .2 Let ||Aij − Aij (x0 )|| = supBs (x0 ) |Aij − Aij (x0 )|.r − Ax .r )2 .r(7) (A 0 0 ) .r )2 .r ) ≤ 6 (u − v ) + 4 (v − A 0 Br (x0 ) Br (x0 ) Br (x0 ) (10) Pick s ≥ r and expand the area of integration to get � � � (u − Ax0 . (u − Ax0 .r )2 ≤ 6 (u − v )2 + 4 Br (x0 ) Bs (x0 ) Bs (x0 ) �x . � ≤ 2 Br (x0 ) (u − v ) + 4 2 Work on the last term.r )2 + 4 (v − A 0 � Br (x0 ) (6) 2 �x .r ) 2 � ≤ 2 Br (x0 ) (u − v ) + 2 Br (x0 ) 2 � � Br (x0 ) (v − Ax0 .r | ≤ 6 Br (x0 ) Br (x0 ) |�(u − v )| + k 2 � � r �n+2 � s Bs (x0 ) � |�v − ( �v )x0 . Then � Bs (x0 ) |�(u − v )| ≤ and � Bs (x0 ) 2 � ||Aij − Aij (x0 )|| λ ||Aij − Aij (x0 )|| λ 2 �2 � Bs (x0 ) |�v |2 (14) |�(u − v )| ≤ 2 � �2 � Bs (x0 ) |�u|2 (15) . (13) Now we need to estimate � |�(u − v )|2 . (u − Ax0 .r )2 .� Br (x0 ) (u − Ax0 . We’ll prove a lemma that will be useful next time. (v − A 0 (11) � is in the correct form for the inequalities we proved last time to apply. so we Note that L can use them to get � � � r �n+2 � 2 2 �x .r − Ax . We have � Br (x0 ) �x . Therefore we can absorb this into the ﬁrst term of (8) to get � � � 2 2 �x .r )2 �x .

Bs (x0 ) (18) Plugging this into 17 gives � λ Bs (x0 ) |�(u − v )| ≤ Bs (x0 ) 2 � Aij ∂ (v − u) ∂v .We will prove only the ﬁrst of these statements. |�(u − v )|2 Bs (x0 ) Finally divide and square to get � ||Aij − Aij (x0 )|| |�(u − v )| ≤ λ Bs (x0 ) 2 � �2 � Bs (x0 ) |�v |2 (23) as required. Proof Calculate � � 2 λ |�(u − v )| ≤ ∂ (v − u) ∂ (u − v ) (16) ∂xi ∂xj Bs (x0 ) Bs (x0 ) � � ∂ (v − u) ∂v ∂ (v − u) ∂u − Aij . 3 . Clearly ∂Bs (x0 ) (u − v )A�u · dS = 0. By Stokes’ theorem we get Aij � ∂ (v − u) ∂u Aij =− ∂xi ∂xj Bs (x0 ) � ∂ ∂u (u − v ) Aij = ∂xi ∂xj Bs (x0 ) � (u − v )Lu = 0. and (19) By a similar calculation to 18 we get � λ Bs (x0 ) � Bs (x0 ) Aij (x0 ) ∂ (v −u) ∂v ∂xi ∂xj |�(u − v )|2 ≤ ∂ (v − u) ∂v ∂xi ∂xj Bs (x0 ) � � � � ∂ (v − u) ∂v � � � ≤ ||Aij − Aij (x0 )|| � ∂xi ∂xj � Bs (x0 ) �1/2 �� �� � (Aij − Aij (x0 )) ≤ ||Aij − Aij (x0 )|| Bs (x0 ) (20) (21) �1/2 |�v |2 )1 (22) /2. The proof of the other is very similar. (17) ≤ Aij ∂x ∂x ∂xi ∂xj i j Bs (x0 ) Bs (x0 ) � Work on the ﬁrst term. ∂xi ∂xj = 0.

we need to eliminate v to use Morrey. � � � r �n � � 2 |�u| ≤ c1 ||Aij − Aij (x0 )|| + c2 |�u|2 (6) s Br (x0 ) Bs (x0 ) 1 . |�u|2 ≤ 6 λ s λ Bs (x0 ) Br (x0 ) (5) By choosing s small we can get ||Aij − Aij (x0 )|| small. λ s Br (x0 ) Bs (x0 ) (2) However. and so.Lecture 18: Regularity of L harmonic functions Part II 1 Finishing the proof � r �n � s Bs (x0 ) Last time we got as far as showing � � 2 |�u| ≤ 6 Br (x0 ) Br (x0 ) |�(u − v )| + k 2 � |�v |2 . Calculate � Bs (x0 ) |�v | 2 � ≤ 2 � ≤ Bs (x0 ) � |�u| + 2 2 Bs (x0 ) � ||Aij − Aij (x0 )|| 2 + 2 λ |�(v − u)|2 � �� 2 Bs (x0 ) (3) |�u|2 (4) by the other part of our lemma. so we need to replace the integral. (1) and we can apply the lemma to get � � � �2 � � r �n � || A − A ( x ) || ij ij 0 |�u|2 ≤ 6 + k� |�v |2 . Substituting this back in we get � �� � � �2 � �2 � � � r �n || A − A ( x ) || || A − A ( x ) || ij ij 0 ij ij 0 + k� 2 + 2 |�u|2 .

c be positive constants. If instead τ k+1 s ≤ r ≤ τ k s then 1 � r �γ φ(s) (6) τ s which is what we needed. Lemma 1. Proof Choose 0 < τ < 1 such that � ≤ τ α . (5) (4) (3) When r = τ k s this is precisely what we wanted with c� = 1.Lecture 18: Regularity of L harmonic functions Part III 1 Finishing the proof In order to ﬁnish the proof from last time we need a lemma. φ(r) ≤ φ(τ k s) ≤ τ kγ φ(s) ≤ 1 . Note that by taking τ very small we can get γ as close to α as we like. Pick γ so that 2cτ α−γ ≤ 1 and we have φ(τ k s) ≤ τ kγ φ(s). Therefore φ(τ k s) ≤ (2cτ α )k φ(s). For all 0 < γ < α there is � > 0 such that �� r �α � φ(r) ≤ c + � φ(s) (1) s for 0 < r < s implies � r �γ φ(s) (2) φ(r) ≤ c� s for some constant c� .1 Let φ be a positive and increasing function on the positive reals. Then φ(τ s) ≤ c(τ α + �)φ(s) ≤ 2cτ α φ(s). and let α.

(8) φ(r) ≤ k � s In our old notation this is � � �n−2+2β � 2 � r |�u| ≤ k |�u|2 .� Now we apply this to what we were doing last time. (9) s Br (x0 ) Bs (x0 ) and Holder continuity follows by Morrey’s lemma. We showed that �� r �n � + k ||Aij − Aij (x0 )|| φ(s) (7) φ(r) ≤ s By picking s we can get ||Aij − Aij (x0 )|| as small as we like so we will be able to apply our lemma. Let φ(r) = Br (x0 ) |�u|2 . 2 . Pick 0 < β < 1 and set γ = n − 2 + 2β . By our lemma there is a constant k � with � r �n−2+2β φ(s).

Theorem 1. The proof is a little involved. ∂xi ∂xj (1) If u : Rn → R is an L harmonic function then u is holder continuous. and deﬁne the operator L 2 � = ∂ Aij (x0 ) ∂f = Aij (x0 ) ∂ f . and let v be an L harmonic function with v = u on ∂Bs (x0 ). � by Proof Pick x0 ∈ Rn . � � r �n � |�v |2 ≤ k |�v |2 (3) s Br (x0 ) Bs (x0 ) for all r < s. so we will ﬁrst give a sketch of the proof. Note that the inequalities we proved in lecture 16 apply to v so. speciﬁcally they must be Holder continuous. Lf ∂xi ∂xj ∂xi ∂xj (2) Pick s > 0.Lecture 20: Holder continuity of Harmonic functions. 1 . 1 Holder continuity of Harmonic functions In this lecture we will show that harmonic functions need to have a degree of regularity. The aim is to use Morrey’s lemma. We use this and the inequality (a + b)2 ≤ 2a2 + 2b2 to estimate � Br (x0 ) |�u|2 ≤ 2 � |�v |2 + 2 |�(u − v )|2 Br (x0 ) Br (x0 ) � � r �n � 2 ≤ 2k |�v | + 2 |�(u − v )|2 s Bs (x0 ) B (x ) � r 0 � r �n � ≤ 2k |�v |2 + 2 |�(u − v )|2 .1 Let L be a uniformly elliptic operator in divergence form taking Lu = ∂ ∂u Aij . s Bs (x0 ) Bs (x0 ) � (4) (5) (6) Now use a lemma which we will prove later. and then go back to ﬁll in the details. in particular.

2 Let ||A − A(x0 )|| = supBs (x0 ). Therefore. and let α. (9) Now we need to estimate this last integral in terms of u.Lemma 1.i. For 0 < γ < α there is δ > 0 such that �� r �α � φ(r) ≤ c1 + δ φ(s) (14) s for 0 < r < s implies � r �γ φ(r) ≤ c2 φ(s). We have � � � |�v |2 ≤ 2 |�u|2 + 2 |�(u − v )|2 Bs (x0 ) B (x ) Bs (x0 ) � s 0� � �� ||A − A(x0 )|| 2 n |�u|2 ≤ 2+2 λ Bs (x0 ) (10) (11) by lemma 1. for some constant k � and for any δ > 0 we can pick a small s so that � � � r �n �� |�u|2 ≤ k � +δ |�u|2 .3 Let φ be a positive and increasing function on the positive reals. (13) s Br (x0 ) Bs (x0 ) We need one more lemma. Then � Bs (x0 ) |�(u − v )| ≤ and � Bs (x0 ) 2 � n||A − A(x0 )|| λ n||A − A(x0 )|| λ �2 � Bs (x0 ) |�v |2 (7) |�(u − v )| ≤ By the ﬁrst of these we get � Br (x0 ) 2 � �2 � Bs (x0 ) |�u|2 (8) � |�u|2 ≤ 2k � r �n s � +2 n||A − A(x0 )|| λ �2 � � Bs (x0 ) |�v |2 .2. Plugging this back into ?? gives � Br (x0 ) � |�u| ≤ 2 2k � r �n s � +2 ||A − A(x0 )|| λ �2 � � 2+2 � n||A − A(x0 )|| λ �2 � � Bs (x0 ) |�u|2 .j |Aij − Aij (x0 )|. (15) s where c2 is some constant that depends on c1 . 2 . c be positive constants. α and γ . Lemma 1. (12) By choosing s small we can get n||A − A(x0 )|| as small as we like.

We will prove this later. ∂xi ∂xj (23) 3 . Lemma 1.2. (19) Proof We will prove the ﬁrst equation. (21) ≤ Aij ∂x ∂x ∂xi ∂xj i j Bs (x0 ) Bs (x0 ) � Work on the ﬁrst term. then � Br (x0 ) |�u|2 ≤ c � r �n−2+2β s C. Calculate � � ∂ (u − v ) ∂ (u − v ) 2 Aij |�(u − v )| ≤ λ (20) ∂xi ∂xj Bs (x0 ) Bs (x0 ) � � ∂ (u − v ) ∂u ∂ (v − u) ∂v − Aij . By Stokes’ theorem we get � ∂ (u − v ) ∂u =− Aij ∂xi ∂xj Bs (x0 ) � ∂ ∂u (u − v ) Aij = ∂xi ∂xj Bs (x0 ) � (u − v )Lu = 0.3. Clearly ∂Bs (x0 ) (u − v )A�u · dS = 0. We wish to show that � � � � n||A − A(x0 )|| 2 2 |�(u − v )| ≤ |�v |2 λ Bs (x0 ) Bs (x0 ) and � Bs (x0 ) (18) |�(u − v )| ≤ 2 � n||A − A(x0 )|| λ �2 � Bs (x0 ) |�u|2 .In other words for any 0 < γ < α we can prove ?? by proving ?? for a suﬃciently small � δ . Bs (x0 ) (22) Plugging this into ?? gives � λ Bs (x0 ) |�(u − v )| ≤ Bs (x0 ) 2 � Aij ∂ (v − u) ∂v .2 and 1. Now prove lemma’s 1. Pick 0 < β < 1 and apply this to ?? with φ(r) = Br (x0 ) |�u|2 and γ = n − 2 + 2β to get � � r �n−2+2β � 2 |�u|2 . The proof of the second is analogous. (16) |�u| ≤ c s Bs (x0 ) Br (x0 ) � �n−2+2β � 2 Let C = 1 s Bs (x0 ) |�u| . (17) so u ∈ C β by Morrey’s lemma.

We will show that if φ is a positive and increasing function on R+ and �� r �α � φ(r) ≤ c1 + δ φ(s) (33) r� for r < r� and 0 < δ < 1 then � r �γ φ(s) (34) φ(r) ≤ c2 (γ ) s � � 2c1 . and i � � ∂ (v − u) ∂v 2 λ |�(u − v )| ≤ (Aij − Aij (x0 )) ∂xi ∂xj Bs (x0 ) Bs (x0 ) � � � � � ∂ (v − u) ∂v � � � ≤ ||A − A(x0 )|| � ∂xi ∂xj � . . . Note that � |w| = n(v1 + . �� Bs (x0 ) |�(u − v )|2 �1/2 �� Bs (x0 ) Finally divide and square to get � Bs (x0 ) |�(u − v )|2 ≤ � n||A − A(x0 )|| λ �2 � Bs (x0 ) |�v |2 (32) as required. + v2 v1 n (28) ≤ n3/2 n ≤ n|v | (29) since the square of the mean is less than or equal to the mean of the square. where γ = α 1 + log log δ 4 . .3. and c2 is a constant depending on γ . . + vn )2 (26) � 2 (v1 + . From this we � get i. . + vn for all i.� v −u) ∂v By a similar calculation to ?? we get Bs (x0 ) Aij (x0 ) ∂ (∂x ∂xj = 0. namely that if u. . + vn ) ≤ n3/2 (27) n2 � 2 + . v are n vectors then ij ui vj ≤ n|u||v |. . Lemma 1. Bs (x0 ) i. Let w be the vector with wi = v1 + v2 + . .j ui vj = u · w ≤ |u||w| ≤ n|u||v | as expected. Applying this to �(u − v ) and �v gives � λ Bs (x0 ) |�(u − v )|2 ≤ n||A − A(x0 )|| ≤ n||A − A(x0 )|| � |�(u − v )||�v | Bs (x0 ) (30) �1/2 |�v |2 (31) .j (24) (25) � Now we need a minilemma.

Pick γ = α 1 + � log 2c1 log δ (35) (36) � so that 2c1 τ α−γ = 1 and we have φ(τ k s) ≤ τ kγ φ(s). If instead τ k+1 s ≤ r ≤ τ k s then 1 � r �γ φ(s) (38) τ s which is what we needed. φ(r) ≤ φ(τ k s) ≤ τ kγ φ(s) ≤ 5 . Then φ(τ s) ≤ c(τ α + δ )φ(s) ≤ 2cτ α φ(s). (37) When r = τ k s this is precisely what we wanted with c2 = 1. Finally note that by using a small δ we can get γ as close as we like to α (though the constant will become nastier). Therefore φ(τ k s) ≤ (2c1 τ α )k φ(s).Proof Choose τ = δ 1/α so that δ = τ α .

Let φ be a test function on BR (x0 ) which is zero on the boundary. Let u be a function satisfying u ≥ 0. and Moser. The argument is an iterative one due to De Georgi. � 0≥ B R ( x0 ) Aij ∂φ2 u ∂u . 0≥ Aij φ2 ∂x ∂x ∂x i j i ∂xj BR (x0 ) BR (x0 ) and −2 BR (x0 ) (5) � Aij φu ∂φ ∂u ≥ ∂xi ∂xj 1 � BR (x0 ) Aij φ2 ∂u ∂u .Lecture 21: The mean value inequality for uniformly elliptic operators part I 1 The mean value inequality: Iterative argument In this lecture we will prove a mean value inequality for uniformy elliptic operators in divergence form . ∂xi ∂xj (4) We can simplify this a bit to get � � ∂φ ∂u ∂u ∂u +2 Aij φu . As usual we take L an operator with Lu = ∂ ∂u Aij ∂xi ∂xj (1) and λ|v |2 ≤ Aij vi vj ≤ Λ|v |2 for all vectors v . and R a ﬁxed positive number. since the ﬁrst term is nonnegative. Lu ≥ 0. ∂xi ∂xj (6) . Nash. Clearly � φ2 uA�u · dS = 0 (2) BR (x0 ) so. by Stokes’ theorem. Take x0 a point. � BR (x0 ) φ uLu + BR (x0 ) 2 � Aij ∂φ2 u ∂u =0 ∂xi ∂xj (3) and.

For Simplicity we λ2 will assume that n ≥ 3. (10) so rearrange to get � 4Λ2 φ |�u| ≤ 2 λ BR (x0 ) 2 2 � BR (x0 ) u2 |�φ|2 . λ φ2 |�u|2 ≤ −2 Aij φu ∂xi ∂xj BR (x0 ) BR (x0 ) (7) Now work on the other term. We’ll apply it slightly diﬀerently this time. 2 . At each point the matrix A deﬁnes a good metric. i ∂xj so � � λ φ2 |�u|2 ≤ 2 φu (�φ · A�φ)1/2 (�u · A�φu)1/2 . as we proved it on the way to the Cacciopolli inequality in lecture 6. (9) Uniform ellipticity then gives � λ BR (x0 ) φ2 |�u|2 ≤ 2Λ �� BR (x0 ) u2 |�φ|2 �1/2 �� BR (x0 ) �1/2 φ2 |�u|2 . and we get −φuAij ∂x ≤ φu (�φ · A�φ)1/2 (�u · A�φu)1/2 . (13) BR (x0 ) |�(φu)| ≤ k BR (x0 ) 2 � u2 |�φ|2 (14) for a constant k = 2 + 8Λ .Apply uniform ellipticity to the right hand side to get � � ∂φ ∂u . (11) This should be familiar. so ∂φ ∂u CauchySchwarz applies. Consider � BR (x0 ) |�(φu)| 2 � = BR (x0 ) � ≤ 2 Combining this with 11 we get � BR (x0 ) 2 |φ�u + u�φ|2 � 2 2 φ |�u| + 2 BR (x0 ) (12) u2 |�φ|2 . (8) BR (x0 ) BR (x0 ) Use CauchySchwarz again in the form � λ BR (x0 ) 2 2 � fg ≤ �� �1/2 �� 2 �1/2 f 2) g ) to get �1/2 �� BR (x0 ) �� BR (x0 ) �1/2 φ � u · A� u 2 φ |�u| ≤ 2 u � φ · A� φ 2 . but a similar result holds in the other cases. Now we need to use the Sobolev inequality.

(16) (φu) n−2 ≤ c BR (x0 ) BR (x0 ) BR (x0 ) for some constant c ˜. (17) Note that if u is L harmonic then u+ is L harmonic almost everywhere. (R − r)2 AR. and let |Ar.k \Br (x0 ) BR (x0 ) Now we’ll introduce another important inequality: the Holder Inequality. and ⎩ R −r 0 outside BR (x0 ) so that |�φ| = R1 −r on BR (x0 ). and let w be a function with compact support on Ω. q real numbers satisfying � fg ≤ �� f p ! p + 1 q = 1. g be functions. Also pick r < R. Note that if u is Lharmonic then u − k is also L harmonic.k | be the volume of this set. and p. 3 . Then �� |w| Ω 2n n−2 � n−2 n � ≤c Ω |�w|2 .e. Theorem 1.1 Let Ω ⊂ Rn with n ≥ 3. Then (22) �1/p �� g q �1/q . f+ = sup(f. 0). and claim without proof that everything we’ve done today goes through for the positive part of a harmonic function as well as for completely harmonic functions. Putting all this together we get �� |(u − k )+ | Ar. (15) We won’t prove this here. Apply it with w = φu (this has compact support because φ does) to get � � � 2n 2 |�(φu)| ≤ c ˜ u2 |�φ|2 .Theorem 1. and 0 elsewhere.k = Br (x0 ) ∩ {u > k }. i.2 Let f. Deﬁne Ar.k 2n n−2 � n−2 n �� ≤ BR (x0 ) |φ(u − k )+ | � 2n n−2 � n−2 n (19) (20) (21) ≤ c ˜ ≤ |�φ|2 ((u − k )+ )2 � c ˜ ((u − k )+ )2 . and set ⎧ on Br (x0 ) ⎨ 1 R−|x| φ= (18) on BR (x0 ) \ Br (x0 ). For any function f deﬁne f+ to be the positive part.

then u(x) > k . 2 (23) Set f = (u − k )+ and Ω = Ar.k | n (R − r)2 2 � AR.k ≤ ((u − h)+ )2 ((u − h)+ )2 AR. 4 . and u(x) − h > k − h.h 2 �2/n � �2/n � ((u − k )+ )2 (29) ((u − h)+ )2(30) AR. which is the case p = q = n n 2.k | (27) ((u − h)+ ) ≥ Ar. Note that if h < k then Ar.h .This is simply a generalisation of the CauchySchwarz inequality. Plugging this back into 26 we get �� Ar.k and we get � Ar. (28) for all h < k .k ((u − k )+ )2 ≤ ≤ ≤ �� ((u − k )+ ) Ar.k 2 2n n−2 � n−2 n |Ar.k ⊂ Ar.h ((u − h)+ )2 .k ((u − k )+ )2 .k | ≤ 1 (k − h)2 � Ar.h �(1+2/n) (31) ≤ Next lecture we will actually do the induction argument. q = 2 and any function f on any set Ω to get � Ω f ≤ Ω 2 �� f 2n n−2 � n−2 n |Ω| n .h �� AR.k .k ((u − h)+ )2 ≤ 1 (k − h)2 � Ar.k | n (R − r)2 c ˜|Ar.h � Ar.k ((u − k )+ ) 2 ≤ c ˜ 2 (R − r) (k − h)4/n c ˜ (R − r)2 (k c ˜ (R − r)2 (k − h)4/n − h)4/n ((u − h)+ ) �� AR. Take x ∈ Ar. Apply this with p = n− 2 .k Ar. Therefore � � 2 (k − h)2 = (k − h)2 |Ar.k | n ((u − k )+ )2 2 (24) (25) (26) c ˜|Ar.k \Br (x0 ) � AR.k and |Ar.

(2) Deﬁne φ(m) = and � = 2/n.k �� AR. Deﬁne rm = 1 + 2−m and km = (2 − 2−m )k for some constant k . so √ c ˜ φ(m + 1) ≤ (φ(m))1+� .h �(1+2/n) ((u − h)+ )2 (1) for all h < k and r < R.Lecture 22: The mean value inequality for uniformly elliptic operators part II 1 The mean value inequality: Iterative argument continued In this lecture we will complete the proof of the mean value inequality. (rm − rm+1 )(km+1 − km )2/n √ Substituting for rm .km+1 2 �(1+2/n) ((u − km )+ ) 2 . 2 /n − m − 1 (2 k) �� Arm .km ((u − km )+ )2 �1/2 (3) (4) Now use induction to show that φ(n) → 0 as n → ∞. rm+1 . km+1 gives �� Arm . Suppose there is some constant (0) γ > 1 with φ(m) ≤ φ γ m . Last time we had � c ˜ ((u − k )+ )2 ≤ 2 (R − r) (k − h)4/n Ar. Applying the above inequality to rm . km . Then � � 2m+1 c φ(0) 1+� φ(m + 1) ≤ (2−m−1 k )2/n γ m �� � � 2m+1 c φ(0) � φ(0) ≤ γm (2−m−1 k )2/n γ m 1 (5) (6) .km � c ˜ ((u−km+1 )+ ) ≤ 4/n 2 (k − r ) ( r m m+1 m+1 − km ) Arm+1 . km and renaming c = c ˜ gives φ(m + 1) ≤ 2m+1 c (φ(m))1+� .

sup u ≤ k B1 (x0 ) � �� A2. (13) 2 . Therefore n→∞ lim φ(n) ≤ lim φ(0) = 0. B1 (x0 ) 1 (11) and. writing out φ(0) and �.0 �1/2 (u+ ) 2 (12) ≤ k This is the mean value inequality. � �� u B2 (x0 ) �1/2 2 .so if � 2m+1 c (2−m−1 k )2/n � φ(0) γm �� � ≤ 1 γ 1+� � (7) and k > (21+� cγ (φ(0))� ) � = 1 (0) then we get φ(n) ≤ φγ n for all n.2k (10) and conclude that u ≤ 2k on B1 . (9) Note that lim rm = 1. It suﬃces to pick γ > 2 2k � φ(0)for appropriate k � . and lim km = 2k so we get � ((u − 2k )+ )2 = 0 A1. Putting in our value for k we obtain sup u ≤ (21+� c) � φ(0).km ((u − km )+ )2 → 0 as n → ∞. n→∞ γ n (8) so � Arm .

∂xi ∂xj (5) Let k = l1 + l2 and apply Cauchy Schwarz to the right hand side to get k+1 k+1 4k (k + 1)2 � ∂u 2 ∂u 2 η 2 Aij ≤2 ∂xi ∂xj Bs (x0 ) �� ∂u ∂u Aij η 2 u2l1 ∂xi ∂xj Bs (x0 ) �1/2 �� ∂η ∂η Aij u2l2 ∂xi ∂xj Bs (x0 ) (6) �1/2 . take an open ball Bs (x0 ). Therefore � � ∂u ∂η ∂u ∂u +2 ηuk Aij .k > 1 a positive constant. As before we take L a uniformly elliptic second order operator in divergence form. We have � Aij Bs (x0 ) ∂u ∂η 2 uk =0 ∂xi ∂xj (1) (this is from the weak deﬁnition of L harmonic). Let x0 ∈ Rn . 1 . If we pick β = � k+1 2 �2 u get k+1 k+1 k −1 ∂u ∂u ∂u 2 ∂u 2 Aij = Aij ∂xi ∂xj ∂xi ∂xj (4) and applying this to 2 gives 4k (k + 1)2 � ∂u 2 ∂u 2 η Aij = −2 ∂xi ∂xj Bs (x0 ) 2 k+1 k+1 � Bs (x0 ) ηuk Aij ∂u ∂η . and η be a cutoﬀ function (ie η : Bs (x0 ) → R with η = 0 on the boundary). and u an L harmonic function on Rn .Lecture 23: Moser’s approach to the mean value inequality 1 The mean value inequality: Moser’s Approach In this lecture we will give an alternative proof of the mean value inequality. 0=k η 2 uk−1 Aij ∂xi ∂xj ∂xi ∂xj Bs (x0 ) Bs (x0 ) Note that Aij ∂uβ ∂uβ ∂u ∂u = β 2 u2β −2 Aij ∂xi ∂xj ∂xi ∂xj k+1 2 we (2) (3) for all constants β .

(13) (ηu ) l2 2n n−2 � n−2 n � ≤ c Bs (x0 ) |�(ηul2 )|2 �� Bs (x0 ) (14) u2l2 |�η |2 (15) 8Λ ≤ c 2+ λ � 2 . Choosing l1 = = k+1 2 we have �� �1/2 �� �1/2 ∂η ∂η Aij u2l2 . (10) Since k ≥ 1. Thus � Bs (x0 ) 4Λ η |�u | ≤ λ Bs (x0 ) l2 )|2 . (11) We need to estimate � Bs (x0 ) |�(ηu Note that (12) |�(ηul2 )|2 ≤ 2u2l2 |�η |2 + 2η 2 |�ul2 |2 . ∂xi ∂xj (9) Use uniform ellipticity to simplify this to � 2l2 − 1 l2 �2 � 4Λ η |�u | ≤ λ Bs (x0 ) 2 l2 2 � Bs (x0 ) u2l2 |�η |2 . l2 k+1 k+1 �� ∂ul1 +1 ∂ul1 +1 Aij η 2 ∂xi ∂xj Bs (x0 ) �1/2 �� ∂η ∂η Aij u2l2 ∂xi ∂xj Bs (x0 ) (7) �1/2 . u2l2 |�η |2 .Apply 3 with β = l1 + 1 to get 4k (k + 1)2 � ∂u 2 ∂u 2 2 η 2 Aij ≤ l1 + 1 ∂xi ∂xj Bs (x0 ) k −1 2 . Apply 11 to get � Bs (x0 ) |�(ηu )| ≤ From this we can use Sobolev to estimate �� Bs (x0 ) l2 � 8Λ 2+ λ �� Bs (x0 ) u2l2 |�η |2 . we have l2 ≥ 1 and so � 2 � 2l2 −1 l2 l2 2 �2 ≥ 1. ∂xi ∂xj Bs (x0 ) (8) 2l2 − 1 l2 2 � 2 ∂ul2 ∂ul2 ≤ η 2 Aij l2 ∂xi ∂xj Bs (x0 ) ∂ul2 ∂ul2 Aij η 2 ∂xi ∂xj Bs (x0 ) so dividing and squaring gives � 2l2 − 1 l2 �2 � ∂ul2 ∂ul2 η Aij ≤4 ∂xi ∂xj Bs (x0 ) 2 � Bs (x0 ) Aij u2l2 ∂η ∂η .

so. i.for some dimensional constant c. Let rm = 1 + 2−m and pm = 2χm . Pick r < s and let η be the usual linear cutoﬀ function. Then � 2m+2 �1/pm ||u||Lpm+1 (Brm+1 (x0 )) ≤ c ˜2 ||u||Lpm (Brm (x0 )) . and outside Bs (x0 ) . (22) This is useful notion of length for integrable functions. 3 . by induction. ||u||Lpm+1 (Brm+1 (x0 )) � � 1 m 2i+2 2χi ||u||L2 (B2 (x0 )) ≤ Πi=0 (2 c ˜) ≤ 2 Pm i+1 i=0 2χm (24) (25) (26) c Pm c i=0 2χm ||u||L2 (B2 (x0 )) . (21) We can deﬁne the Lq norm of a function f by �� ||f ||Lq (BR (x0 )) = f BR (x0 ) �1/q q . � ≤ c ˜ Bs (x0 )\Br (x0 ) u2l2 |�η |2 ≤ ≤ � c ˜ u2l2 (s − r)2 Bs (x0 )\Br (x0 ) � c ˜ 2l2 u (s − r)2 Bs (x0 ) n n−2 for an appropriate constant c ˜ that depends on L and n. ⎧ ⎨ 1 ⎩ This gives �� Br (x0 ) s−|x| s− r φ= 0 on Br (x0 ) on Bs (x0 ) \ Br (x0 ). (16) (ul2 ) 2n n−2 � n−2 n �� ≤ Bs (x0 ) (ηul2 ) 2n n−2 � n−2 n (17) (18) (19) (20) > 1 and p = 2l2 .e. Deﬁne χ = We’ve shown that �� Br (x0 ) �1/pχ upχ ≤ � c ˜ (s − r)2 �1/p �� Bs (x0 ) �1/p up . and in this notation we have � �1/p c ˜ ||u||Lp (Bs (x0 )) (23) ||u||Lpχ (Br (x0 )) ≤ (s − r)2 for all r < s.

4 . Therefore if the L2 norm of u is ﬁnite on B2 (x0 ) then the L∞ norm onB1 (x0 ) is ﬁnite. is bounded by a dimensional constant. and indeed.Both of these sums converge as m → ∞ by the ration test.

Sign up to vote on this title

UsefulNot usefulClose Dialog## Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

Loading