PAA by Douglas Heggie
2011
Notes of Paul Boeck
Last changes: May 16, 2011
Contents
1 The Generalized Riemann Integral 2
2 Fourier transforms 3
2.1 Definition and basic properties . . . 3
2.2 Gaussians . . . 5
2.3 Inversion Theorem . . . 6
2.4 Plancherel’s Theorem . . . 7
2.5 An application: Heisenberg’s Uncertainty Principle . . . 8
3 The Heat Equation and Laplaces Equation 9 3.1 Existence and uniqueness for the heat equation . . . 9
3.2 Laplace’s equation . . . 11
3.3 Laplace equation in the upper half–plane . . . 11
3.4 The maximum principle . . . 12
3.5 Uniqueness for Laplace’s equation . . . 12
3.6 Mean Value Theorem . . . 13
3.7 Green’s Functions, Laplace’s eqn . . . 14
4 The Wave equation 15 4.1 1 Space Dimension . . . 15
4.2 1 Dimension in the Half Line . . . 15
4.3 The Wave Equation in 3 Dimensions . . . 16
5 Distributions 17 5.1 Definitions and basic properties . . . 17
5.2 Addition and multiplications of distributions . . . 18
5.3 Differentiation of distributions . . . 19
5.4 Convergence of Distributions . . . 19
5.5 Higher–dimensional distributions . . . 20
5.6 Fourier transform of distributions . . . 21
6 Nonlinearity 22
Index 23
1 THE GENERALIZED RIEMANN INTEGRAL
1 The Generalized Riemann Integral
Definition 1.1
Let f :[a,∞)→Rbe a function that is Riemann–integrable on each[a,b], b>a. We define Z ∞
a f(x)dx= lim
b→∞ Z b
a f(x)dx provided that the limit exists.
Example 1.2 • f(x) = x12,x ∈[1,∞). ThenRb
1 f(x)dx=hx−2+1−2+1ib
1= (b−1−1) =1−b−1−−−→b→∞ 1
• 1x,x∈[1,∞),Rb 1 1
xdx= [lnx]b1=lnb−−−→b→∞ ∞ =⇒ the integral does not exist.
Remark Integrals with∞limits sometimes are called "improper" integrals.
Definition 1.3
Let f : (−∞,a] → Rbe a function which is Riemann–integrable on each [b,a] with b < a. We define Ra
−∞f(x)dx = limb→−∞Ra
b f(x)dx provided the limit exists.
Definition 1.4
Let f :R→Rbe a given function. We defineR∞
−∞f(x)dx=R0
−∞f(x)dx+R∞
0 f(x)dx provided both limits exist.
Proposition 1.5 The integralR∞
a f(x)dx exists iffRb0
b f(x)dx→0as b,b0→∞.
Proof: Problem 2.
Proposition 1.6 IfR∞
a |f(x)|dx exists, so doesR∞
a f(x)dx.
Proof: Assume R∞
a |f(x)|dx exists. Then Prop. 1.5. implies Rb0
b |f(x)|dx −−−−→b,b0→∞ 0. But
Rb0
b f(x)dx ≤ Rb0
b |f(x)|dx. Hence Prop 1.5 impliesR∞
a f(x)dxexists.
Proposition 1.7 (Comparison)
Let f,g be Riemann–integrable on every[a,b]and suppose that|f(x)| ≤g(x)on[a,∞)andR∞
a g(x)dx exists. Then so does R∞
a f(x)dx.
Proof: Rb
a |f(x)|dx ≤ Rb
a g(x)dx −−−→b→∞ R∞
a g(x)dx from below. Obviously Rb
a |f(x)|dx is a non–decreasing function ofband bounded above byR∞
a g(x)dx. HenceR∞
a |f(x)|dxexists, henceR∞
a f(x)dxexists by Prop.
1.6.
Example 1.8 • TestR∞
0 cosx 1+x2dx.
cosx 1+x2
≤ x12 andRb 0 dx
1+x2 = [arctanx]b0 =arctanb −−−→b→∞ π2. Hence convergence by Prop. 1.7.
• TestR∞
0 e−x2dx. Nowx2≥xifx ≥1 and so ifb>1:Rb
0 e−x2dx=R1
0 e−x2dx+Rb
1 e−x2dxwhere|e−x2| ≤e−x (ifx>1).
Rb
1 e−xdx= [−e−x]1b=e−b+e−1−−−→b→∞ e−1. Hence the whole integral exists.
• TestR∞
1 xke−xdx. e−x = e1x ≤ xn1/n! for anyn≥0 and allx ≥0. Choosen> k+1. Thusxke−x ≤n!xk−n and Rb
1 xk−ndx= k−n+11 hxk−n+1ib
1= k−n+11 ek−n+1−1 b→∞
−−−→ −k−n+11 . Hence convergence.
Theorem 1.9 (Leibniz’ rule for improper integrals) Suppose f(x,τ)and fx(x,τ)are continuous functions fora ≤ x ≤ bandc ≤ τ < ∞. Define F[x,t) = Rt
c f(x,τ)dτand supposeF(x,t) → ϕ(x) ≡ R∞
c f(x,τ)dτandFx(x,t) → ψ(x)uniformly forx∈[a,b].
Thenϕis differentiable andψ(x) =ϕ0(x) =R∞
c fx(x,τ)dτ.
Remark Basically dxd R∞
c f(x,τ)dτ=R∞
c ∂
∂xf(x,τ)dτprovided that integral on the right has uniform convergence.
2 FOURIER TRANSFORMS
2 Fourier transforms
2.1 Definition and basic properties
Definition 2.1.1
Let f :R→Cbe such thatR∞
−∞|f(x)|dx exists. We define theFourier transformof f by fˆ:R→C, where fˆ(ξ) =
Z ∞
−∞f(x)e−iξxdx (ξ∈R) Remark (i) The integral exists, becauseR∞
−∞
f(x)eiξx
dxexists, and using Prop 1.7.
(ii) ˆf is a bounded function, because for allξ∈R
|fˆ(ξ)|=
Z ∞
−∞f(x)e−iξxdx
≤ Z ∞
−∞|f(x)e−iξx|dx= Z ∞
−∞|f(x)|dx Example 2.1.2 Take f(x) =
(1 if|x| ≤1
0 otherwiseIts F.T. is fˆ(ξ) =
Z ∞
−∞f(x)e−ξxdx= Z 1
−1eiξxdx=
−1 iξe−iξx
1
−1
=−1 iξ
e−iξ−eiξ
=−1
iξ(−2isinξ) = 2
ξsinξ if ξ6=0 Forξ=0, ˆf(0) =R1
−1 dx=2=limξ→02 sinξ
ξ . Thus ˆf(ξ) = (2 sinξ
ξ ifξ6=0 2 ifξ=0
Remark Sometimes we just write ˆf = 2 sinξ ξ on the understanding that ˆf(0) =limξ→0fˆ(ξ). Definition 2.1.3
TheSchwartz space(of rapidly decreasing functions) denotedS(R)is defined to be the set of all functions f :R→Cwhich are infinitely differentiable and have the following property: For all integers k,l ∈ (0, 1, 2, . . .)the function|x|k|f(l)(x)|is bounded.
Example 2.1.4 The function f(x) =e−x2 is inS(R). For f is obviously infinitely differentiable.
Also fixk,l∈ {0, 1, 2, . . .}, we show that|x|k|f(l)(x)|is bounded, as follows.
First, do
l=0 :|x|k|f(x)|=|x|ke−x2 = |x|k
ex2 = |x|k
1+x2+12x4+· · ·+(2k)!x2k +· · · ≤ |x|k 1+(2k)!x2k which is bounded.
Now considerl > 0. First, observe that f(x) = e−x2 =⇒ f0(x) = −2xe−x2 =⇒ f00(x) = (−2+4x2)e−x2. Clearly (or by induction) f(l)(x) = hl(x)e−x2, wherehl(x)is a polynomial. Hence|x|k|f(l)(x)| = |xkhl(x)|e−x2 =
|ql,k(x)|e−x2, whereql,kis another polynomial. Writeqk,l(x) = ∑dn=0cnxk. Thenqk,l(x)e−x2 = ∑dn=0cnxne−x2, each term is bounded
Proposition 2.1.5 If f ∈ S(R)thenR∞
−∞|f(x)|dx exists.
Proof: Since|x|k|f(l)(x)|is bounded for eachk,l ∈(0, 1, 2, . . .),|f(x)|andx2|f(x)|are bounded. (l =0,k=0 and 2). Hence(1+x2)|f(x)|is bounded, i.e.∃Ms.t.(1+x2)|f(x)| ≤M. Hence|f(x)| ≤ 1+xM2.R∞
−∞ dx
1+x2 exists,
hence result, by Prop 1.7.
Corollary 2.1.6
The Fourier transformf exists for any fˆ ∈ S(R).
Proof: See Def 2.1.1 and remarks and Prop 2.1.5.
2.1 Definition and basic properties 2 FOURIER TRANSFORMS
Example 2.1.7 Let f : R→ Cbe infinitely differentiable and suppose that f vanishes outside some interval[a,b]. Then f ∈ S(R).
For, given anyk,l the function|xk||f(l)(x)|vanishes outside[a,b]and is continuous and hence bounded on[a,b]. Hence it is bounded onR.
Example 2.1.8 The functione−|x|is not inS(R)as it is not differentiable atx=0.
Proposition 2.1.9
S(R)is closed under (i) differentiation and under (ii) multiplication by polynomials.
Proof: (i) Let f ∈ S(R). Then f is infinitely differentiable. Therefore f0is infinitely differentiable. Also for any integersk,l≥0 we havexk(f0)(l)(x) =xkf(l+1)(x), which is bounded.
(ii) See Problem 9.
Proposition 2.1.10
Let f ∈ S(R)and let h∈R. Define g(x) =e−ixhf(x). Thengˆ(ξ) = fˆ(ξ+h). Proof: gˆ(ξ) =R∞
−∞g(x)e−iξxdx=R∞
−∞f(x)e−i(ξ+h)dx= fˆ(ξ+h)
Proposition 2.1.11
Let f ∈ S(R), h∈Rand define g(x) = f(x+h). Thengˆ(ξ) = fˆ(ξ)eiξh Proof: gˆ(ξ) =R∞
−∞g(xe−iξxdx=R∞
−∞f(x+h)e−iξxdx. Change variable: Letx0 =x+h.
Then ˆg(ξ) =R∞
−∞f(x0)e−iξ(x0−h)dx0=R∞
−∞e−iξx0eiξhdx0 =eihξfˆ(ξ).
Remark Props 2.1.10-11 are calledshift–theorems.
Proposition 2.1.12
Let f ∈ S(R). The(df0)(ξ) =iξfˆ(ξ).
Remark analogue ofcn(f0) =incn(f)complex Fourier series.
Proof: (df0)(ξ) =R∞
−∞f0(x)e−iξxdx=f(x)e−iξx∞
−∞−R∞
−∞f(x)(−iξ)e−iξxdx. Now f ∈ S(R) =⇒ |x f(x)| ≤ M =⇒ |f(x)| ≤ |x|M for allx6=0.Hence limx→±∞|f(x)e−iξx|=0. Finally(df0)(ξ) =iξfˆ.
Proposition 2.1.13
Let f ∈ S(R). Then(fˆ)0(ξ) =gˆ(ξ)where g(x) =−ix f(x). Proof: (fˆ)0(ξ) = dξd R∞
−∞ f(x)e−iξxdx=R∞
−∞ ∂
∂ξ f(x)e−iξx
dx=R∞
−∞(−ix f(x))e−iξxdxby Prop 1.9.
Remark \x f(x)(ξ) =ifˆ0(ξ). Corollary 2.1.14
If f ∈ S(R)thendf(k)= (iξ)kf .ˆ
Proof: Induction onk. Prop 2.1.12 givesk=1. Also, assuming 2.1.14,\f(k+1)=(\f(k))0 =iξdf(k)=iξ(iξ)kfˆ.
Corollary 2.1.15
If f ∈ S(R)thenf is infinitely differentiable.ˆ fˆ(k)(ξ) =gˆk(ξ)where gk(x) = (−ix)kf(x). Proof: Induction based on Prop 2.1.13.
Remark x\kf(x)(ξ) =ikfˆ(k)(ξ).
Example Let f(x) = e−x2. f0(x) = −2xe−x2 = −2x f(x) (∗).The FT of f0 isiξfˆ(Prop 2.1.12) and the FT ofx f(x) isifˆ0. Hence(∗) =⇒ iξfˆ= −2ifˆ0. We are using the fact that the Fourier transform, being an integral, is a linear operation. Hence ddξfˆ=−ξ2fˆ. Separable differential eqn:
Z dfˆ fˆ −
Z ξ
2dξ =⇒ ln ˆf =−ξ
2
4 +const. =⇒ fˆ(ξ) =Ce−ξ2/4 Cconst.
2.2 Gaussians 2 FOURIER TRANSFORMS
Z∞
−∞
e−x2−iξxdx x2+iξx=x+iξ22+ξ42
= Z∞
−∞
e−(x+iξ2)2−ξ2/4dx=e−ξ2/4 Z∞
−∞
e−(x+iξ2)2dx x0=x+iξ 2
=e−ξ2/4
∞+iξ/2
Z
−∞+iξ/2
e−x02dx0
Now ˆf(0) =R∞
−∞f(x)dxby Def 2.1.1.C=R∞
−∞e−x2dx=√
π. Hence ˆf(ξ) =√ πeξ2/4. Proposition 2.1.16
If f ∈ S(R)thenfˆ∈ S(R). Proof: We need to show
(a) ˆf is infinitely differentiable and
(b) |ξ|k|fˆ(l)(ξ)|is bounded for all integersk,l≥0
To proof (b) we have (iξ)kfˆ(l)(ξ) = (iξ)kgˆl(ξ)where gl(x) = (−ix)lf(x). by Cor 2.1.15. Now gl ∈ S(R) because it is a product off ∈ S(R)by a polynomial (Prop 2.1.9).
fˆ(k)=
Z ∞
−∞f(x)e−iξxdx
≤ Z ∞
−∞|f(x)||e−iξx|dx
Also(iξ)kgˆl(ξ) = dg(k)l (ξ)by Cor 2.1.14. But since gl ∈ S(R)we haveg(k)l ∈ S(R)(Prop 2.1.9 again). Hence
dg(k)l
≤R∞
−∞|g(k)l |dx(by Def 2.1, Remark ii), i.e.
ξkfˆ(l)(ξ)=(iξ)kfˆ(l)
=(iξ)kgˆl(ξ)=
gd(k)l (ξ)
is bounded
Remark The mapF :S(R)→ S(R), defined byF(f) = fˆis linear (as integration). We shall see it is 1–1 and onto and (almost) an isometry.
2.2 Gaussians
i.e. functions of the forme−ax2,a>0.
Definition 2.2.1
Forδ>0define Kδ(x) = √1
2πδe−x2/2δ. Lemma 2.2.2
Kˆδ(ξ) =e−δξ2/2. Proof:
Z ∞
−∞
√1
2πδe−x2/2δ−iξxdx x=√ 2δx0
= Z ∞
−∞
√1
πe−x02−2ξx0
√2δdx=e−ξ22δ/4=e−ξ2δ/2
Remark K1 δ
(x)Def 2.2.1= q
δ
2πe−δx2/2 Lemma 2.2.2= q
δ
2πKˆδ(x). Also ˆK1 δ
(ξ) =√
2πδKδ(ξ).
Example What is the function whose FT isξe−δξ2/2? We use 2.1.12: bf0 = iξfˆ. Let f(x) = Kδ(x) = √1
2πδe−x2/2δ. Hence bf0 =iξe−δξ2/2. Required function is 1iK0δ(x) =−i√2πδ1 (−2x2δ)e−x2/2δ= ix
δKδ(x).
2.3 Inversion Theorem 2 FOURIER TRANSFORMS
Proposition 2.2.3
The family Kδ,δ>0has the properties (i) ∀δ>0 :R∞
−∞Kδ(x)dx=1 (ii) ∀η>0 :R
|x|>ηKδ(x)dx→0asδ→0.
Definition 2.2.4
For f,g∈ S(R)define theirconvolutionby
f∗g:R→C where (f∗g)(x) = Z∞
−∞
f(x−y)g(y)dy
Theorem 2.2.5 (Convolution Theorem) Let f,g∈ S(R)then(\f∗g)(ξ) = fˆ(ξ)gˆ(ξ). Proposition 2.2.6
If f ∈ S(R)then f∗Kδ → f uniformly onRasδ→0.
2.3 Inversion Theorem
Proposition 2.3.1 (Multiplication Formula) Let f,g∈ S(R). Then
Z∞
−∞
f(x)gˆ(x)dx= Z∞
−∞
fˆ(x)g(x)dx
Theorem 2.3.2 (Inversion Theorem) Let f ∈ S(R). Then∀x∈R f(x) = 1
2π Z ∞
−∞fˆ(ξ)eixξdξ
Remark This is the analogue of the following result for a well–behaved 2π–periodic function f. If cn = 1
2π Z π
−π f(x)e−inxdx then f(x) =
∑
∞−∞cneinx Proof: First prove it forx=0, i.e. f(0) = 2π1 R∞
−∞fˆ(ξ)dξ(∗). Letg(x) =Gδ(x) = 2π1 e−δx2/2where Z ∞
−∞ f(x−y)Kδ(y)
| {z }
ˆ g(y)
dy→ f(x) Kδ= √1
2πδe−x2/2δ) F√1
2πδe−x2/2δ is e−δx2/2 F
r δ
2πeδx2/2 is e−x2/2δ (δ→ 1
δ) Gˆδ(ξ) =Kδ(ξ)(remark to Lemma 2.2.2). ThusR∞
−∞f(x)Kδ(x)dx= R∞
−∞fˆ(x)e−δx2/22π dx. Now letδ→0. RHS→
1 2π
R∞
−∞fˆ(x)dxand LHS=R∞
−∞ f(x)Kδ(x)dx.
Z ∞
−∞f(x−y)Kδ(y)dy→ f(x)
= Z ∞
−∞f(−y)Kδ(−y)y (x=−y)
= Z ∞
−∞f(0−y)Kδ(y)dy sinceKδ(y) =Kδ(−y)
= (f∗Kδ)(0)→ f(0) as δ→0 by Prop 2.2.6 Hence f(0) = 2π1 R∞
−∞ fˆ(ξ)dξ.
Fix x0 ∈ R. Define g(x) = f(x+x0). Apply (∗) to g : g(0) = 2π1 R∞
−∞gˆ(ξ)dξ. But g(0) = f(x0) and ˆ
g(ξ) =eiξx0fˆ(ξ)(Prop 2.1.11). Hence f(x0) = 2π1 R∞
−∞fˆ(ξ)eiξx0dξ.
2.4 Plancherel’s Theorem 2 FOURIER TRANSFORMS
Corollary 2.3.3
(i) If f ∈ S(R)and fˆ(0) =0(i.e. fˆ(ξ) =0for allξ∈R) then f =0.
(ii) If f1,f2∈ S(R)and fˆ1= fˆ2then f1= f2 Proof: (i) f(x) = 2π1 R∞
−∞0eiξxdξ=0
(ii) apply (i) to f1−f2.
Definition 2.3.4
For f ∈ S(R)we have defined its FT by(Ff)(ξ) = fˆ(ξ) =R∞
−∞f(x)e−ixξdx.
We now also define theinverse Fourier transformof f by(F−1f)(ξ) = 2π1 R∞
−∞f(x)eixξdx= fˇ(ξ). Remark (i) We can show that if f ∈ S(R), then ˇf ∈ S(R).
(ii) The inversion theorem (2.3.2) states thatF−1(Ff) = f. Similarly we can show thatF(F−1f) = f. Corollary 2.3.5
The mapF :S(R)→ S(R), f 7→ Ff = f is 1–1 and onto (i.e. bijection).ˆ Proof:onto: If f ∈ S(R)letg=F−1f. ThenFg= f.
1–1: Ff1=Ff2 =⇒ f1= f2(Cor 2.3.3(ii))
Remark The inversion theorem is true for many f ∈ S(/ R). Example 2.3.6 Let f(x) =
(e−x ,x>0
0 ,x≤0. Then ˆf(ξ) =1+iξ1 (Tutorial Q4b). Hence f(x) = 2π1 R∞
−∞ eiξx
1+iξdξ. f(−x) =
1 2π
R∞
−∞e−iξxdξ
1+iξ and withx↔ξwe have f(−ξ) = 2π1 R∞
−∞e−iξxdx 1+ix
2.4 Plancherel’s Theorem
Recall:For 2π–periodic functions f :R→Cwe haveParceval’s identity 1
2π Z π
−π
|f(x)|2dx=
∑
∞−∞
|cn|2 Theorem 2.4.1 Iff ∈ S(R)then
Z ∞
−∞|f(x)|2dx= 1 2π
Z ∞
−∞|fˆ(ξ)|2dξ Remark (i) In terms of theL2–normkfk2=R∞
−∞|f(x)|dx1/2
,Plancherel’s Theoremsays thatkfk2= √1
2πkfˆk2. (ii) Except for the factor √1
2π, which can be removed by slight redefinition ofF, F is an isometry (preserving length).
Proof: (idea)
Z∞
−∞
|f(x)|2dx= Z∞
−∞
f(x)f(x)dx= 1 2π
Z∞
−∞
fˆ(ξ)eixξdξf(x)dx
But fˆ(ξ) = Z∞
−∞
f(x)e−ixξdx
fˆ(ξ) = Z∞
−∞
f(x)e−ixξdx= Z∞
−∞
f(x)e−ixξdx= Z∞
−∞
f(x)eixξdx
Hence Z∞
−∞
|f(x)|2dx= 1 2π
Z∞
−∞
fˆ(ξ)fˆ(ξ)dξ= 1 2π
Z∞
−∞
|fˆ(ξ)|2dξ
2.5 An application: Heisenberg’s Uncertainty Principle 2 FOURIER TRANSFORMS
2.5 An application: Heisenberg’s Uncertainty Principle
Theorem 2.5.1 Let f ∈ S(R)be such that R∞
−∞|f(x)|2dx=1. Then R∞
−∞x2|f(x)|2dx
! ∞ R
−∞ξ2|f(ξ)|2dξ
!
≥ π2
Proof:
1= Z∞
−∞
1· |f(x)|2dx= Z∞
−∞
dx
dx|f(x)|2dx
=− Z∞
−∞
x
|f(x)|20 dx integration by parts
=− Z∞
−∞
x
f(x)f(x)0 dx=− Z∞
−∞
x
f0(x)f(x) +f(x)f0(x)dx
Hence 1=
Z∞
−∞
x
f0(x)f(x) +f(x)f0(x)dx
≤ Z∞
−∞
|x||f0(x)||f(x)|+|f(x)||f0(x)| dx=2 Z∞
−∞
|x||f(x)||f0(x)|dx
Cauchy Schwarz
≤2
Z∞
−∞
|x|2|f(x)|2dx
1/2
Z∞
−∞
|f0(x)|2dx
1/2
(∗)
=2
Z∞
−∞
|x|2|f(x)|2dx
1/2
1 2π
Z∞
−∞
|bf0(ξ)|2dξ
1/2
by Theorem 2.4.1
=2
Z∞
−∞
|x|2|f(x)|2dx
1/2
1 2π
Z∞
−∞
|iξfˆ(ξ)|2dξ
1/2
by Theorem 2.1.12
Squaring: π 2 ≤
Z∞
−∞
|x|2|f(x)|2dx
Z∞
−∞
|ξ|2|fˆ(ξ)|2dξ
Remark In quantum mechanics, the position of a particle moving on thex–axis is known only probabilistically, in terms of awave functionψ:R→C,|ψ(x)|2dxis the probability of finding the particle in[x,x+ dx]. Thus|ψ(x)|2is the probability density function and soR∞
−∞|ψ(x)|2dx=1. Its width is characterized by∆xwhere (∆x)2=
Z∞
−∞
x2|ψ(x)|2dx
Similarly its velocity (strictly, momentum) has a probability distribution whose width is∆p, where (∆p)2=
Z∞
−∞
| −i¯h∂ψ
∂x|2dx where ¯h= 2πh wherehisPlanck’s constant. Hence(∗)
=⇒ 1<4(∆x)2 ∆p
¯ h
2
, i.e. ∆x,∆x≥ ¯h 2
Hence, if the uncertainty in position is small, the uncertainty in velocity is large and conversely.
3 THE HEAT EQUATION AND LAPLACES EQUATION
Example 2.5.2 Let f(x) =pKδ(x) =q√1
2πδe−x2/2δ= (2πδ)11/4e−x2/4δ(Def. 2.2.1). Then Z∞
−∞
|f(x)|2dx= Z∞
−∞
Kδ(x)x=1 Prop 2.2.3
Also Z∞
−∞
x2|f(x)|2dx= Z∞
−∞
x2
√2πδe−x2/2δdx= √1 2πδ
Z∞
−∞
d
dx(e−x2/2δ)(−δ)x dx
= r
δ 2π
he−x2/2δxi∞
−∞− Z∞
−∞
e−x2/2δdx
=???
Next, f(x) = 1 (2πδ)1/4
√
4πδK2δ(x) = (8πδ)1/4K2δ(x) and so fˆ(ξ) = (8πδ)1/4e−δξ2 Lemma 2.2.2 By a similar calculationR∞
−∞|ξ|2|fˆ(ξ)|2dξ= 2δπ. Hence Heisenbergs inequality becomes equality.
3 The Heat Equation and Laplaces Equation
ut=uxx uxx+uyy=0
Interpretation LetT(x,t)be the temperature at timet, at position x along a rod which conducts heat. Then for common substancesk∂∂x2T2 = ∂T∂t. By scalingt= tk∗,∂∂x2T2 = ∂t∂T∗ (heat equation).
NextT(x,y,t)be the temperature at timetat position(x,y)on a flat object which conducts heat. Then
∂T
∂t =k ∂2T
∂x2 +∂
2T
∂y2
If the temperature settles into steady, ∂T∂t =0 and so∂∂x2T2 +∂∂y2T2 =0 (Laplaces equation).
For the heat equation, the usual initial condition isu(x, 0) = f(x)∈ S(R).
3.1 Existence and uniqueness of solutions of the heat equation
Proposition 3.1.1 (Existence forut =uxx)
Let f ∈ S(R). Define u(x,t) = (f ∗Ht)(x)where Ht(x) = √1
4πte−x2/4tfor t>0theHeat Kernel, i.e.
u(x,t) = √1 4πt
Z∞
−∞
f(y)e−(x−y)2/4tdy Then
(i) ut=uxxif t>0 (ii) limt→0+u(x,t) = f(x)
Proof: (i) By Leibniz’s rule (Prop 1.9) ut(x,t) = =
Z∞
−∞
f(y)∂
∂t √1
4πte−(x−y)2/4t
dy
& similarly uxx(t) = Z∞
−∞
f(y) ∂
2
∂x2 1
√4πte−(x−y)2/4t
dy
3.1 Existence and uniqueness for the heat equation 3 THE HEAT EQUATION AND LAPLACES EQUATION
Now, sincex−yis independent oft
∂
∂t 1
√te−(x−y)2/4t
= 1/2
t1/2e−(x−y)2/4t+(x−y)2
4t5/2 e−(x−y)2/4t
∂
∂x 1
√te−(x−y)2/4t
= (x−y)
2t3/2 e−(x−y)2/4t =⇒ ∂
2
∂x2 1
√te−(x−y)2/4t
= ∂
∂t 1
√te−(x−y)2/4t
Hence result.
(ii) SinceHt(x) =K2t(x), (ii) is an application of Theorem 2.2.6.
Remark The expressionu(x,t) = √1 4πt
Z∞
−∞
f(y)e−(x−y)2/4tdyis an (integral) superposition of solutions of the form
√1
4πte−(x−y)2/4t.
Remark 3.1.2 For fixedt > 0,uis a convolution of 2 functions inS(R)and so is in S(R)(see problem 17). Also u(x, 0)∈ S(R).
Proposition 3.1.3 (Uniqueness for the heat equation) Let u(x,t), v(x,t)have the following properties:
(i) both are defined for x∈R, t≥0and are continuous
(ii) both solve the heat equation in the region x∈R, t>0with u(x, 0) =v(x, 0). (iii) both have the property in Remark 3.1.2 (inS(R))
Then u=v.
Proof: Letw=u−v. Thenwhas properties (i)–(iii) withw(x, 0) =0. DefineE[w](t) =R∞
−∞|w(x,t)|2dx. Then E[w]0(t) =
Z ∞
−∞
∂
∂t|w(x,t)|2dx Z ∞
−∞(wtw¯+ww¯t)x= Z ∞
−∞(wxxw¯+ww¯xx)dx
= Z ∞
−∞((ww¯)xx−2wxw¯x)dx=
(ww¯)x ∞
−∞
| {z }
=0 asw∈ S(R)
−2 Z ∞
−∞|wx|2dx≤0
ThereforeE[w]is a decreasing function oft. Hence∀t≥0 0≤E[w](t)≤E[w](0) =
Z ∞
−∞|w(x, 0)|2dx=0 HenceE[w](t) =R∞
−∞|w(x,t)|2dx=and sow(x,t) =0. Henceu(x,t) =v(x,t).
Remark 3.1.5 Though we assumed f ∈ S(R)the formulau(x,t) = √1 4πt
Z ∞
−∞ f(y)e−(x−y)2/4tdyis also correct for many f ∈ S/ (R).
Example 3.1.6 Take f =e3x∈ S/ (R). Thenu(x,t) = √1 4πt
Z ∞
−∞e3ye−(x−y)2/4tdywhere the exponent is 3y1
4t(x2−2xy+y2) =−1
4t(y2−2y(x+6t) +x2) =−1
4t((y−(x+6t))2−x2−12xt−36t2+x2)
=−1
4t(y−x−6t)2+3x+9t u(x,t) = √1
4πt Z ∞
−∞e−4t1(y−x−6t)2+3x+9tdy= √1
4πte3x+9t Z ∞
−∞e−4t1(y−x−6t)2dy
= √1
4πte3x+9t Z ∞
−∞e−y02/4tdy0 (y0 =y−x−6t)
=e3x+9t by Prop 2.2.3, since the integrand isK2t(y0)
3.2 Laplace’s equation 3 THE HEAT EQUATION AND LAPLACES EQUATION
3.2 Laplace’s equation u
xx+ u
yy= 0
Definition 3.2.1
A setΩ ⊆R2is said to be open iff for every(x,y)∈ Ω, there is an r>0, s.t. the disc D((x,y),r), i.e. the disc of radius r, centered at(x,y), is contained inΩ.
Definition 3.2.2
LetΩ⊆R2be open. A function u:Ω→R(orC) is said to beharmoniciff it is twice differentiable and s.t. uxx+uyy=0.
Example 3.2.3 If f(z) is holomorphic with real and imaginary parts u,v, then u,v satisfy the Cauchy–Riemann equationux=vy,uy=−vxand souxx+uyy=vyx−vxy=0. Similarlyvxx+vyy=0, i.e.uandvare harmonic.
f(z) =
∑
∞ 0anzn=
∑
∞ 0anrneinθ =
∑
∞ 0anrn(cosnθ+isinnθ) zn =rncosnθ+irnsinnθ
3.3 Laplace equation in the upper half–plane
Ω={(x,y):y,x∈R,y>0}=R2+and we have∇2u=0,u= f(x).
Consider the following boundary value problem: given a functionf, find a functionu(x,y)s.t.uxx+uyy=0 inR2+ andu(x, 0) = f(x).
Definition 3.3.1
ThePoisson kernelis defined to be Py(x) = π1x2+yy 2, x∈R, y>0
Remark 3.3.2 For fixedy >0,Py(x)is infinitely differentiable, but it doesn’t decay rapidly enough as|x| →∞and is not inS(R).
Proposition 3.3.5
The Poisson kernel has the properties (i) ∀y>0R∞
−∞Py(x)dx=1 (ii) F(Py)(ξ) =Pˆy(ξ) =e−y|ξ| (iii) for fixedδ>0,Ry
x>δPy(x)dx→0as y→0+. Proof: (i) 1
π Z∞
−∞
y
y2+x2dx= 1 π
tan−1 x
y ∞
−∞
= 1 π
π
2 −−π 2
=1
(ii) tutorial Ex 2.1 (iii) By symmetry,
Z
|x|>δ
Py(x)dx= 2 π
Z∞
δ
y
x2+y2dx= 2 π
tanx
y ∞
δ
= 2 π
π
2 −tan−1δ y
→ 2 π
π 2 −π
2
=0
asy→0+.
Theorem 3.3.6 Let f ∈ S(R). Defineu(x,y) = (f∗Py)(x),x ∈R,y>0. Then
(i) u(x,y)is twice differentiable in the upper half planey>0 anduxx+uyy=0 (i.e.uis harmonic) (ii) lim
y→0+u(x,y) = f(x)for allx∈R
(iii) if we extendu(x,y)tox∈R,y≥0 by defining it to be f(x)wheny=0, the extendeduis continuous.
Proof: Similar to the heat equation (Prop. 3.1.1).
Again, this formula to applies to many f ∈ S/ (R).
3.4 The maximum principle 3 THE HEAT EQUATION AND LAPLACES EQUATION
Example 3.3.7 Letu(x, 0) =H(x) =
(1 ifx>0
0 ifx<0 (Heaviside step function). Then
u(x,y) = (H∗Py)(x) = Z∞
−∞
H(z)Py(x−y)dz= y π
Z∞
−∞
H(z) dz
y2+ (x−2)2 Def 3.3.1
= y π
Z∞ 0
dz
y2+ (2−x)2 = 1 π
tan−12−x y
∞
0
= 1 π
π
2 −tan−1
−x y
= 1 π
π
2 +tan−1x y
z=x+iy
= 1 2+ 1
π π
2 −argz
=1− 1
πargz=Im
i− 1 πlnz
Ex 3.2.3
3.4 The maximum principle
In this section,Ω⊆R2is open and bounded. We define theclosureofΩby ¯Ω= Ω∪∂Ωwhere∂Ωis the set of all boundary points ofΩ. The set ¯Ωis closed and bounded. (Closed means its complement is open.)
Such sets in the plane are calledcompact. It can be shown that a continuous function on a compact set always has a maximum and a minimum.
Lemma 3.4.1
Let u: ¯Ω→Rbe continuous onΩ¯ and such that uxx+uyy≥0inΩ. Then the max. of u onΩ¯ is attained on∂Ω.
Remark such a function is calledsubharmonic.
Proof: Suppose that the max is attained at an interior point(x0,y0)∈Ω. Thenuxx(x0,y0)≤0 anduyy(x0,y0)≤ 0. Henceuxx(x0,y0) +uyy(x0,y0)≤0. This is a contradiction.
Theorem 3.4.2 (Max Principle) Letu: ¯Ω→Rbe continuous and harmonic inΩ. Then the max ofuon ¯Ωis attained on the boundary∂Ω.
Remark It may also be attained at an interior point, e.g.u=constant.
Proof: Letε > 0. Definev(x,y) = u(x,y) +ε(x2+y2). Thenvx = ux+2εx,vxx = uxx+2ε, similarlyvyy = uyy+2ε. Hencevxx+vyy=4ε>0 By Lemma 3.4.1 the max ofvis attended on∂Ω. Hence for any(x0,y0)∈Ω, v(x0,y0)≤max∂Ωv(x,y)i.e.
u(x0,y0) +ε(x20+y20)≤max
(x,y)∈∂Ω
n
u(x,y) +ε(x2+y2)o≤max
∂Ω u(x,y) +εM
where M > 0 is a bound for x2+y2on∂Ω. Henceu(x0,y0) ≤ max∂Ωu(x,y) +ε(M−(x20+y20)). Now let
ε→0+:u(x0,y0≤max∂Ωu(x,y).
Corollary 3.4.3
Under the same assumptions the minimum of u is attained on∂Ω.
Proof: Apply Theorem 3.4.2 to−u.
3.5 Uniqueness for Laplace’s equation in the upper half plane
Given f(x)∈ S(R), letu(x,y) = (f∗Py)(x). Then by Theorem 3.3.6u(x,y)satisfies (i) uxx+uyy=0 fory>0
(ii) u(x,y)→ f(x)asy→0+
Remark (iii) It is also true thatu(x,y)→0 asx2+y2→∞,y>0 (Problem 33) Theorem 3.5.1 u(x,y)satisfying (i)-(iii) is unique.
3.6 Mean Value Theorem 3 THE HEAT EQUATION AND LAPLACES EQUATION
Proof: Given f, suppose∃two suchu, i.e.u1(x,y)andu2(x,y). Letw=u1−u2. Thenwxx+wyy=0 inR2+ = {(x,y)∈ R2,y >0}. w(x, 0) = u1(x, 0)−u2(x, 0) = f(x)− f(x) =0. wis continuous in{(x,y)∈R2,y ≥0} and limx2+y2→∞w(x,y) =0−0=0.
We want to showw=0. Suppose on the contrary that∃(x0,y0)∈R2+s.t.w(x0,y0)6=0, sayw(x0,y0)>0. (For w(x0,y0)<0, the proof is similar). PickRlarge enough so that
(i) x20+y20<R2
(ii) ∀(x,y)on the semicirclex2+y2=R2,y≥0,w(x,y)≤ 12w(x0,y0). Apply maximum principle towon the setΩR={(x,y):x2+y2<R,y>0}
max
x2+y2=R2,y>0
≤ 12w(x0,y0) max
AB w(x, 0) =0 whereABis the straight part of the semicircle max∂ΩR
≤ w(x0,y0) =⇒ w(x0,y0)≤ 12w(x0,y0)<w(x0,y0)
This clearly is a contradiction.
3.6 Mean Value Theorem
Theorem 3.6.1 SupposeΩ ⊆R2is open and bounded andu:Ω →Ris harmonic. Let(x0,y0)∈Ωand letR>0 be such thatD((x0,y0),R)⊆Ω. Then for anyr∈[0,R]we have
u(x0,y0) = 1 2π
Z 2π
0 u(x0+rcosθ,y0+rsinθ)dθ Proof: Define
f(r) = 1 2π
Z 2π
0 u(x0+rcosθ,y0+rsinθ)
| {z }
U(r,θ)
dθ
r,θare polar coords with origin(x0,y0). Thus sinceuis harmonic, ∂∂r2U2 +1r∂U∂r +r12∂2U
∂θ2 =0 d f
dr = 1 2π
d dr
Z 2π
0 U(r,θ)dθ= 1 2π
Z 2π 0
∂U
∂r(r,θ)dθ (1) d2f
dr2 = 1 2π
Z 2π 0
∂2U
∂r2(r,θ)dθ Hence rd f
dr +r2d2f dr2 = 1
2π Z 2π
0
r∂U
∂r +r2∂2U
∂r2
dθ
=− 1 2π
Z 2π 0
∂2U
∂θ2 dθ=− 1 2π
∂U
∂θ 2π
0
=0 (2)
SinceU(r,θ) = U(r,θ+2π) =⇒ ∂U∂θ(r,θ) = ∂U∂θ(r,θ+2π). Hence d fdr +rddr2f = 0 =⇒ drd(rd fdr) = 0 =⇒ rd fdr =const. (1) =⇒ d fdr is bounded inD((x0,y0),R). Taking lim asr→0 we find the constraint is 0 and so
d f
dr =0 =⇒ f =constc.
Evaluating atr=0, f(r) = 2π1 R2π
0 u(x0,y0)dθ=u(x0,y0).
Example 3.6.2 Suppose u(x,y) is cts in x2+y2 ≤ 1, harmonic in x2+y2 > 1 andu(x,y) = 2x2+3y2 on the boundaryx2+y2=1. Its value at the origin isu(0, 0) =. . .= 52.
Remark u(cosθ, sinθ) = 2 cos2θ+3 sin2θ = 2+sin2θ. maxx2+y2=1u(x,y) = 3 and minu(x,y) = 2. Hence 2≤u(x,y)≤3 inx2+y2≤1 by Maximum principle.