Pierson Guthrey pguthrey@iastate.edu Part I Partial Differential Equations 1 Introductory Theory and Notation 1.1 Multiindex notation A multiindex α = (α1 , α2 , ...., αn ) ∈ Zn+ denotes Dα = ∂ |α| (∂x1 )α1 (∂x2 )α2 ...(∂xn )αn α x = n Y i xα i i=1 |x| = q v u n uX 2 2 x1 + ... + xn = t x2i i=1 α α x f = {y : y = x f (x)} For multiindices α and β, • |α| = α1 + α2 + ... + αn • α! n Q αi ! i=1 • α + β = (α1 + β1 , ..., αn + βn ) • α ≥ β if and only if αi ≥ βi for i = 1, ..., n 1.2 Senses of Solution Let Lu = P aα (x)Dα u were aα (x) ∈ C ∞ (Ω) so LT ∈ D0 (Ω) for T ∈ D0 (Ω) |α|≤M Example: If Lu = utt − uxx then the general solution is u(x, t) = F (x + t) + G(x − t) 1.2.0.1 Classical Solution Lu = f in a classical sense if u ∈ C M (Ω) and u0 (x) = f (x) ∀ x ∈ Ω Example: We require F, G ∈ C 2 (Ω) 1.2.0.2 Weak Solution Lu = f in a weak sense if u ∈ L1loc and Lu = f in D0 sense. Classical solutons are always also weak solutions Example: We require F, G ∈ L1 (Ω) 1.2.0.3 Distributional Solution Lu = f in a distributional sense if u ∈ D0 and Lu = f in D0 sense. Classical solutions and weak solutions are always also distributional solutions Example: We require F, G = δ ∈ D0 (Ω) 1 1.3 Pierson Guthrey pguthrey@iastate.edu Distributional Solutions 1.3 1.3.1 Distributional Solutions Fundamental Solutions We say E ∈ D0 (RN ) is a fundamental solution of X L= aα (x)Dα aα (x) ∈ C ∞ |α|≤M if LE = δ. • Usually fundamental solutions can be found by taking the FT of LE = δ which is ∧ X ∧ Dα E = (δ) → X (ik)α Ê(k) = |α|≤M |α|≤M 1 N (2π) 2 • Fundamental solutons are never unique since you can add solutions to the homogenous equation H and the equation L(E + H) = f will still be valid 1.3.1.1 Translational Invariance If E is a fundamental solution to and aα (x) = aα (constant coefficients) and the domain of E is all of RN , then u = E ∗ f is a solution formula for Lu = f , and E is translationally invariant which means it commutes with translation. That is, given L : C0∞ (RN ) → C ∞ (RN ), τh Lφ = Lτh φ Reason: ∀ φ ∈ C0∞ , ∀ h ∈ RN Lτh φ = T (τx (τhˇφ)) = T (τx−h φ̌) = (T ∗ φ)(x − h) = Lφ(x − h) = τh Lφ 1.3.1.1.1 Convergence If L : C0∞ (RN ) → C ∞ (RN ) such that L is translation invariant and continuous then there exists a unique T ∈ D0 (RN ) such that Lφ = T ∗ φ 1.3.2 Green’s Function Given a PDE, a fundamental solution that satisfies the boundary conditions is known as a Green’s Function for that BVP. Given X L= aα (x)Dα aα (x) ∈ C ∞ |α|≤M Then E = E(x, y) is a fundamental solution of Lx E(x, y) = δ(x − y) if ˆ ˆ Lu(x) = Lx E(x, y)f (y)dy = δ(x − y)f (y)dy = f (x) RN RN E(x, y) can be of the form E(x − y) but need not be in general. 2 1.3 Pierson Guthrey pguthrey@iastate.edu Distributional Solutions 1.3.2.1 Example ( K(x, y) = y(x − 1) 0 ≤ y ≤ x ≤ 1 x(y − 1) 0 ≤ x ≤ y ≤ 1 |x| 2 , IF Lu = u00 with u(0) = u(1) = 0 then Lx K = δ(x − y). If E(x) = then and LE(x − y) = δ(x − y) LE = δ(x) so they should differ by a solution to the homogenous equation. If H(x, y) = K(x, y) − E(x − y) then ( y(x − 1) − 21 (x − y) 0 ≤ y ≤ x ≤ 1 n H(x, y) = = y − 21 x − 12 y 0 ≤ x, y ≤ 1 1 x(y − 1) − 2 (y − x) 0 ≤ x ≤ y ≤ 1 so Lx H = 0 ∀ y. If u(x) = ´1 K(x, y)f (y)dy then u00 = f but also 0 ˆ ˆ 1 K(0, y)f (y)dy = 0 and u(1) = u(0) = 0 1 K(1, y)f (y)dy = 0 0 so we have solved the BVP using the Green’s function K(x, y) 1.3.2.2 Example: For N = 3, ∆ Further, if f ∈ C0∞ (RN ) −1 4π|x| = δ. then let u = E ∗ f and so Lu = L(E ∗ f ) = LE ∗ f = δ ∗ f = f Example: Let E(x) = −1 4π|x| so ∆E(x) = δ and so if ∆u = f then u = (E ∗ f )(x) = −1 4π ˆ R3 f (y) dy |x − y| which can be check to also be a classical solution since E ∗ f ∈ C ∞ . 1.3.2.3 Example: In Ω = R2 , Lu = utt − uxx u(x, 0) = h(x) ut (x, 0) = g(x) Let E(x, t) = 21 H (() t − |x|), where H is the Heaviside function. So ( 0 t < |x| E(x, t) = 1 (Indicator function for the forward light cone) t > |x| 2 If f ∈ C0∞ , u = E ∗ f is a soluion of Lu = f . ˆ ˆ ˆ 1 ∞ ∞ u(x, t) = E ∗ f = E(x − y, t − s)dyds = H(t − s − |x − y|)f (y, s)dyds 2 −∞ −∞ R2 So u(x, t) = 1 2 ˆ t ˆ x+s−t f (y, s)dyds −∞ (Indicator function for backward light cone) x+t−s 3 1.3 Pierson Guthrey pguthrey@iastate.edu Distributional Solutions If we assume f (x, t) = 0 for t < 0, this becomes u(x, t) = 1 2 ˆ tˆ x+s−t (Integral of cone with vertex at x, y ) f (y, s)dyds x+t−s 0 Note u(x, 0) = ut (x, 0) = 0 Also (E ∗ g) = ´∞ (x) and ∂ ∂t (E ∗ h)(x, t) (x) −∞ E(x − y, t)g(y)dy = 1 2 ´ x+t x−t g(y)dy (Part of D’Alembert’s Solution Formula) = 12 (h(x + t) + h(x − t)) so solution of PDE is u(x, t) = (E ∗ f )(x, y) + (E ∗ g) + (x) 1.3.3 ∂ (E ∗ h)(x, t) ∂t (x) Symbol of Operator L Define the Symbol of L to be N P (k) = (2π) 2 X (ik)α |α|≤M with P (k)Ê(k) = 1. • The coefficients of the operator L yield a unique symbol and vice versa P N (ik)α is the principal symbol of L and excludes lower order terms • Pm (k) = (2π) 2 |α|=M L is elliptic if Pm (k) = 0 if and only if k = 0 for k ∈ RN : that is Pm (k) has no real roots except 0 • A fundamental solution should satisfy (assuming P (k) 6= 0) 1 1 =⇒ E(x) = Ê(k) = N P (k) (2π) 2 But this is not clear if 1 p ˆ RN eikx dk P (k) ∈ /S • Malgrange Ehrenpreis Theorem If L 6= 0 everywere then a fundamental solution to Lu = f exists. • Theorem If L is elliptic, f ∈ C ∞ (RN ), u ∈ D0 (RN ) and Lu = f , then u ∈ C ∞ (RN ) 1.3.3.1 N Example: L = ∆ then P (k) = −(2π) 2 N P j=1 kj2 so P (k) = 0 when k1 = k2 = 0 so is therefore elliptic 1.3.3.2 Example: Lu = utt − uxx then P (k) = 2π(kx2 − kt2 ) so P (k) = 0 when kt = ±kx and is therefore not elliptic. 1.3.4 Duhamel’s Principle Duhamel’s principle is a general method for obtaining solutions to inhomogeneous linear evolution equations (Ex: see heat equation) by convolution of a fundamental solution with the inhomogenous term. 4 1.4 Pierson Guthrey pguthrey@iastate.edu Characteristic Curves 1.3.4.1 Regularity of Solutions For ∆u = 0 all solutions are classical, weak, and distributional solutions. Thus, the regularity of the solution depends on the PDE. 1.4 Characteristic Curves of a PDE are the lines on which the solution is constant. 1.5 Types of PDEs 1.5.1 Linear PDEs A PDE of order m is linear if it can be expressed as X Lu(x) = aα (x)Dα u(x) = f (x) |α|≤m n An order k linear PDE over R will have 1.5.2 n+k−1 k−1 distinct terms Semilinear PDEs A PDE of order m is semilinear if it can be expressed as X aα (x)Dα u(x) + a0 Dk−1 u, ..., Du, u, x = f (x) |α|=m 1.5.3 Quasilinear PDEs A PDE of order m is quasilinear if it can be expressed as X aα (Dα u(x), ..., Du, u, x) + a0 Dk−1 u, ..., Du, u, x = f (x) |α|=m 1.5.4 Nonlinear PDEs A PDE of order m is nonlinear if it depends nonlinearly upon the highest order derivatives. PDEs are classified by order and type (elliptic, hyperbolic, parabolic) 1.6 First Order Equations (m = 1) 1.6.1 1.6.1.1 Linear Equations Constant Coefficients ~ · ~θ = ∂u = aux + buy = 0 ∇u ∂~θ A directional derivative vanishes for all points (x, y). A characteristic curve though point (x0 , 0) obeys ay = b(x −x0 ), so x0 = bx−ay b If u(x, 0) = f (x), then u(x, y) = u(x0 , 0) = f (x0 ) = f 5 bx−ay b 1.6 Pierson Guthrey pguthrey@iastate.edu First Order Equations (m = 1) 1.6.1.2 Nonconstant Coefficients a(x, y)ux + b(x, y)uy = 0 A directional derivative at each point vanishes. A characteristic curve parameterized by (x(t), y(t)) satisfying the ODE system dx = a(x, y) dt dy = b(x, y) dt d dx dy u(x(t), y(t)) = ux + uy = aux + buy = 0 dt dt dt u is a solution of the PDE =⇒ the curves are characteristics =⇒ u is constant along these curves Let f (x) = u(x, 0), if (x, y) and (x0 , 0) lie on the same characteristic then u(x, y) = u(x0 , 0) = f (x0 ) 1.6.1.3 Cauchy Problem A solution u of the PDE is specified by f (s) on a curve γ • γ can be defined parametrically by x = φ(s) and y = ψ(s) then u(φ(s), ψ(s)) = f (s) • γ can be nowhere characteristic nor can it be tangent to the characteristic direction (x0 , y 0 ) = (a(x, y), b(x, y)) because if f = 0 anywhere, then f touches a characteritic, which means it is a characteristic • Solution: Existence isn’t guaranteed. If γ is nowhere tangent to (a, b) and (a, b) 6= (0, 0) then a unique solution exists locally (near γ) Solve the ODE for (x(t, s), y(t, s)), a fixed s which is a characteristic through (φ(s), ψ(s)) dx = a(x, y) dt dy = b(x, y) dt x(0, s) = φ(s) y(0, s) = ψ(s) Perform a transformation (x(t, s), y(t, s)) → (t(x, y), s(x, y)) Which means u(0, s) = f (s) becomes u(x, y) = f (s(x, y)) Solution is valid at a point (x, y) if it can be conncted to γ by a characteristic curve 1.6.2 Semilinear Equations a(x, y)ux + b(x, y)uy = c(x, y, u) So we have an ODE system dx = a(x, y) dt dy = b(x, y) dt d (u(x(t), y(t))) = a(x, y)ux + b(x, y)uy = c(x, y, u) = c(x(t), y(t), u(x(t), y(t))) dt Parameterize γ (see Cauchy Problem), solve system to get t(x, y) and s(x, y) and solve u0 = c(x, y, u) u(s, 0) = f (s) using previously obtained x(t, s) and s(t, s) to set u(t, s) and substitute to get u = u(x, y) 6 1.7 Pierson Guthrey pguthrey@iastate.edu Second Order Equations (m = 2) 1.7 Second Order Equations (m = 2) Typically made easier using a linear transformation of coordinates Let η = φ(x, y), ξ = ψ(x, y) be an invertible transformation (x, y) → (η, ξ). We need to find ux and uxx in terms of the new system. Using chain rule, ux = uη φx + uξ ψx uxx = uηη φ2x + uη + uξη ψx φx + uξ + uηξ φx ψx + uη + uξξ ψx2 + uξ uxx = uηη φ2x + 2uξη ψx φx + uξξ ψx2 + 2uη + 2uξ Now we must find uy and uyy in the new system. uy = uη φy + uξ ψy uyy = uηη φ2y + uη + uξη ψy φy + uξ + uηξ φy ψy + uη + uξξ ψy2 + uξ uyy = uηη φ2y + 2uξη ψy φy + uξξ ψy2 + 2uη + 2uξ Now we must compute the mixed derivative uxy based on ux uxy = uηη φy φx + uη + uξη φy ψx + uη + uηξ ψy φx + uξ + uξξ ψy ψx + uξ uxy = uηη φy φx + uξη (φy ψx + ψy φx ) + uξξ ψy ψx + 2uη + 2uξ Adding everything together, a second order equation in the original coordinates auxx + 2buxy + cuyy + dux + f uy + hu = g(x, y) becomes in the new coordinate system: Auηη + Buηξ + Cuξξ + Duη + F uξ + Hu = g(η, ξ) where A(η, ξ) =aφ2x + bφx φy + cφ2y B(η, ξ) =2aφx ψx + b(φy ψx + φx ψy ) + 2cψy φy C(η, ξ) =aψx2 + bψx ψy + cψy2 D(η, ξ) =dφx + f φy + 2a + 2b + 2c F (η, ξ) =dψx + f ψy + 2a + 2b + 2c H(η, ξ) =h If the transformation η = φ(x, y), ξ = ψ(x, y) is linear, it can be expressed in the form η α β x = ξ γ δ y with αδ − βγ 6= 0 to ensure invertibility. This means φx = α, φy = β, ψx = γ, ψy = δ, so A(η, ξ) =aα2 + bαβ + cβ 2 B(η, ξ) =2aαγ + b(βγ + αδ) + 2cβδ C(η, ξ) =aγ 2 + bγδ + cδ 2 D(η, ξ) =dα + f β + 2a + 2b + 2c F (η, ξ) =dγ + f δ + 2a + 2b + 2c H(η, ξ) =h 7 1.7 Pierson Guthrey pguthrey@iastate.edu Second Order Equations (m = 2) 1.7.1 Types of Equations 1.7.1.1 Parabolic Equations b2 −ac = 0 With equations of this type, one can choose b(η, ξ) = c(η, ξ) = 0 to obtain uηη = 0 or equivalently b(η, ξ) = a(η, ξ) = 0 to get uξξ = 0 1.7.1.2 Hyperbolic Equations b2 −ac > 0 0 to obtain With equations of this type, one can choose a(η, ξ) = c(η, ξ) = uξη + D|1| u = 0 or equivalently b(η, ξ) = 0 to get uξξ − uηη + D|1| u = 0 Where D|1| u is the lower order terms If D|1| u = 0, the solutions to these equations are of the form u(η, ξ) = Φ(η) + Ψ(ξ) for some functions Φ(η), Ψ(ξ) 1.7.1.3 Elliptic Equations b2 − ac < 0 to obtain With equations of this type, one can choose a(η, ξ) = c(η, ξ) = 0 uξη + D|1| u = 0 or equivalently b(η, ξ) = 0 to get uξξ − uηη + D|1| u = 0 Where D|1| u is the lower order terms 1.7.2 Separation of Variables Seeking separable solutions u(x1 , ..., xn ) = X1 (x1 ) × ... × Xn (xn ) works for many kinds of PDEs in multiple dimensions. (0 ≤ θ < 2π, r ≥ 0) 1.7.3 Boundary Conditions For a well-posed mth order PDE, you will have up to m side conditions, usually in the form of F (u) = 0 on ∂Ω Boundary conditions are said to be homogenous if they are closed under linear combinations, such as u = 0 on ∂Ω The most common boundary conditions are: 1.7.3.1 Dirichlet Conditions (First Type) u = g on ∂Ω 8 1.7 Pierson Guthrey pguthrey@iastate.edu Second Order Equations (m = 2) 1.7.3.2 Neumann (Second Type) ∂u = ∇u · ~n = g on ∂Ω ∂n where ~n is the unit outward normal 1.7.3.3 Robin (Third Type), aka Mixed ∂u + σu = ∇u · ~n + σu = g on ∂Ω ∂n 1.7.4 Common Well-Posed Problems The following equations are well-posed and have classic solutions 1.7.4.1 Poisson’s Equation ∆u = f (Poisson’s Equation) ∆u = 0 (Laplace’s Equation) Solutions are harmonic functions, typically with Dirichlet, Neumann, or Robin conditions (rectangular coordinates) uxx + uyy + uzz = f 1 1 urr + ur + 2 uθθ = f (polar coordinates) r r Requires product solution families to be 2π periodic in θ: ( c1 rn cos (nθ) + c2 rn sin (nθ) + c3 r−n cos (nθ) + c4 r−n sin (nθ) n = 1, 2, 3... un (θ, r) = c1 + c2 log(r) n=0 Coefficients are found using boundary conditions, initial conditions, and Fourier theory ∂u 1 ∂2u ∂2u 1 ∂ r + 2 2 + 2 =0 (cylindrical coordinates) r ∂r ∂r r ∂φ ∂z 1 ∂ r2 ∂r r2 ∂u ∂r + 1 ∂2 2 r sin(θ) ∂θ2 sin(θ) ∂u ∂θ + r2 1 ∂2u =0 2 sin (θ) ∂φ2 (spherical coordinates) 1.7.4.1.1 Using Distributions −1 u(x) = 4π|x| , note u ∈ L1loc (R3 ) since ˆ ˆ R ˆ 2π ˆ π |u(x)| dx = R 0 0 −π 1 2 C r sin(θ)dφdrdθ = R2 4πr 2 ˆ Claim: ∆u = δ. Reason: u(∆φ) = ˆ u∆φ = lim →0 u∆φdx <|x|<R Since φ has compact support and so for some R φ(|x| > R) = 0. By Green’s Identity this becomes ! ˆ ˆ ˆ ∂u ∂φ lim u∆φdx = lim φ(x)∆udx + φ − u ds →0 <|x|<R →0 ∂n <|x|<R |x|= ∂n 9 1.7 Pierson Guthrey pguthrey@iastate.edu Second Order Equations (m = 2) Note that ∆u ∝ ∆ 1r = 0, and ˆ ˆ ∂φ ∂φ ∂φ 1 1 u ds ≤ ds = max 4π2 ≤ C → 0 |x|= ∂n 4π |x|= ∂n 4π ∂n ∂u since the 4π2 is the surface area of a sphere. We also see that since ∂n = − ∂u ∂r , ˆ ˆ ˆ ˆ ∂u 1 ∂u 1 φ ds = − ds = φ ds = − φ φds 2 4π2 |x|= |x|= ∂n |x|= ∂r |x|= 4π which is the´ average of φ over the sphere |x| = which must tend to the value at the center. That is, as 1 φds → φ(0), so → 0, 4π 2 |x|= ∆u(φ) = u(∆φ) = φ(0) = δ(φ) In higher dimensions, ( u(x) = −1 2π AN log(x) N = 2 1 N ≥3 |x|N −2 Where AN is the surface area of the N -Sphere. 1.7.4.2 Fundamental Solution E(x) = |x| 2 N =1 1 2π log |x| CN |x|N −2 N =2 N ≥3 N If ∆ is replaced by the positive operator −∆, Where CN = N (N??? −1)an and an is the volume of B1 (0) in R the the FS is E− (x) = −E(x) 1.7.4.3 Wave Equation utt − c2 uxx = 0 Typically with Dirichlet, Neumann, or Robin conditions in space and some u(~x, 0) =f (~x) ut (~x, 0) =g(~x) x∈Ω η = x + ct and ξ = x − ct, then u(x, t) = Φ(x + ct) + Ψ(x − ct) If initial conditions u(x, 0) = f (x) and ut (x, 0) = g(x) are given, ´x then f (x) = Φ(x) + Ψ(x) and g(x) = Φ0 (x) + Ψ0 (x) and so 0 g(y)dy = Φ(x) + Ψ(x) + C ´ x Combining, F (x) = 12 f (x) + 0 g(y)dy + C giving us D’Alembert’s Solution Formula ˆ x+ct 1 1 u(x, t) = (f (x + ct) + f (x − ct)) + g(y)dy 2 2c x−ct 10 1.7 Pierson Guthrey pguthrey@iastate.edu Second Order Equations (m = 2) 1.7.4.4 Fundamental Solution 1 2 H(t − |x|) N = 1 H(t−|x|) 1 N =2 E(x, t) = 2 √t2 −|x|2 δ(t−|x|) N =3 4π|x| Example: For N = 3 ˆ (E ∗ f ) = R4 δ(s − |y|) f (x − y, t − s)dsdy = 4π |y| Assume f (x, t) < 0 for t < 0, so ˆ (E ∗ f ) = R3 ,|y|<t 1 f (x − y, t − |y|)dy = 4π |y| ˆ R3 ,s=|y| ˆ B(x,t) 1 f (x − y, t − |y|)dy 4π |y| 1 f (y, t − |x − y|)dy 4π |x − y| So the solution of utt − ∆u = f , u(x, 0) = h(x), ut (x, 0) = g(x) can be shown to be ∂ (E ∗ h) + (E ∗ f ) ∂t (x) u(x, t) = (E ∗ g) + (x) 1.7.4.4.1 Uniqueness Showing the uniqueness of solutions to this problem is equivalent to showing that the homogenous problem utt − c2 uxx = 0 ut (~x, 0) = u(~x, 0) = 0 with homogenous boundary conditions has only a trivial solution. 1.7.4.4.2 Four Point Property for solutions to the wave equation uxx − utt = 0 on Ω ⊂ R2 containing the tilted rectangle with vertices (x, t), (x + h − k, t + h + k), (x + h, t + h), u(x − k, t + k), u(x, t) + u(x + h − k, t + h + k) = u(x + h, t + h) + u(x − k, t + k) 1.7.4.4.3 Using Distributions Let F ∈ L1loc (R), then u(x, t) = F (x + t) is one solution to the wave equation in D0 (R2 ) since ˆ ˆ T (φ) = (utt − uxx )(φ) = u(φtt − φxx ) = F (x + t)(φtt (x, t) − φxx (x, t))dxdt R2 By change of coordinates, ξ = x + t, η = x − t, φtt − φxx = −4φξη , dxdt = ˆ ˆ ∞ =2 ∞ F (η) −∞ φξη dξdη = 0 −∞ since φ has compact support 11 ∂(x,t) ∂(ξ,η) = − 21 dξdη 1.7 Pierson Guthrey pguthrey@iastate.edu Second Order Equations (m = 2) 1.7.4.5 Heat Equation ut − ∆u = 0 (~x, t) ∈ Ω × [0, ∞) Typically with Dirichlet, Neumann, or Robin conditions in space and some u(~x, 0) = f (~x), x ∈ Ω Look for separable solutions of the form u(t, x) = Φ(t)Ψ(x) where Φ(t) = c1 ekt + c2 e−kt Ψ(x) = c1 sin(kx) + c2 cos(kx) Solutions will usually involve k = kn so the set of product solutions is n o∞ 2 2 2 2 ∞ {Φn (t)Ψn (x)}0 = c1 ekn t sin(kn x) + c2 ekn t cos(kn x) + c3 e−kn t sin(kn x) + c4 e−kn t sin(kn x) 0 Where the c1 , ..., c4 is chosen to satify the boundary and initial conditions using Fourier theory. 1.7.4.5.1 Uniqueness Showing the uniqueness of solutions to this problem is equivalent to showing that the homogenous problem ut − ∆u = 0 ut (~x, 0) = u(~x, 0) = 0 with homogenous boundary conditions has only a trivial solution. 1.7.4.6 Fundamental Solution E(x, t) = 1.7.4.7 H(t) (2πt) N 2 e −|x|2 4t Reason: 1.7.4.7.1 ∧ Using FT For Lu = ut − uxx with u(x, 0) = f (x) taking the partial FT in x yields ˆ 1 ∧ u(x, t)e−ikx dx and (ut ) = (û)t û(k, t) = N N 2 (2π) R 2 2 So (−∆u) = |k| û so ût + |k| û = 0, which is an ODE for û for fixed k. Solving, ∧ ! ˆ |x−y|2 1 1 −|k|2 t −1 ˆ û(k, t) = f e =⇒ u(x, t) = F (f ∗ g) = e− 4t f (y)dy N N (2π) 2 (4πt) 2 RN 2 where ĝ(k) = e−|k| t so g(x) = |x|2 4t N (2t) 2 e− . This is valid for f ∈ Lp (RN ) for some 1 ≤ p ≤ ∞, t > 0, and lim u(x, t) = f (x) a.e. t→0+ 12 1.7 Pierson Guthrey pguthrey@iastate.edu Second Order Equations (m = 2) 1.7.4.7.2 Using FT and Duhamel’s Principle By Duhamel’s Principle, we can do the same for Lu = ´t h. Let v(x, t, s) satisfy vt − ∆v = 0 x ∈ RN , t > 0, v(x, 0, s) = h(x, s) and let u(x, t) = 0 v(x, t − s, s)ds. ˆ ˆ t t vt (x, t − s, s)ds = h(x, t) + ut = v(x, 0, t) + 0 ∆v(x, t − s, s)ds = h(x, t) + ∆u 0 So for ut − ∆u = h(x, t) for t > 0 with u(x, 0) = 0, we have v(x, t, s) = where E = ˆ 1 (2πt) N 2 −|x|2 4t N (2πt) 2 H(t)e e −|x−y|2 4t ˆ tˆ 1 h(y, s)dy =⇒ u(x, t) = RN 0 RN (2π(t − s)) N 2 e −|x−y|2 4(t−s) h(y, s)dyds = E ∗ h and h(x, t) = 0 for t < 0. So our solution is u(x, t) = (E ∗ h)(x, t) + (E ∗ f )(x, t) (x) N F √xt if So E is the fundamental solution to the Heat Equation. Note: for t > 0, E(x, t) = √1t |x|2 ´ − F (x) = e 4N . Here F ≥ 0, F ∈ L1 , and RN F (x)dx = 1. By previous discussion of approximate identities let k = (2π) 1 √ t 2 k N F (kx) → δ as k → ∞ =⇒ E(x, t) → δ ∈ D0 as t → 0+ so (E ∗ f )(x, t) = E(·, t) ∗ f → f as t → 0+ (similar to approximate identities) (x) 1.7.4.8 Maximum Principle The solution u is bounded by the extremes of the initial condition and the Dirichlet boundary conditions. 1.7.4.9 Shrodinger Equation ut − i∆u = 0 (~x, t) ∈ Ω × [0, ∞) Typically with Dirichlet, Neumann, or Robin conditions in space and some u(~x, 0) = f (~x), x ∈ Ω 1.7.4.10 Fundamental Solution E(x, t) = 1.7.4.11 H(t) (2πt) N 2 −|x|2 4it Helmholtz Equation ∆u − k 2 u = 0 1.7.4.12 π ei(N −2) 4 e (~x, t) ∈ Ω × [0, ∞) Fundamental Solution ( E(x, t) = 1 2π K0 (k |x|) −e−k|x| 4π|x| N =2 N =3 Where Ki is the ith order Modified Bessel Function. Note as k → 0, E(x) tends toward the E(x) corresponding to L = ∆ 13 Pierson Guthrey pguthrey@iastate.edu 1.7.4.13 Euler-Tricomi Equation uxx − xuyy = 0 Hyperbolic in the half plane x > 0, parabolic at x = 0, elliptic in x < 0 2 Characteristics are along y ± 23 x 3 = C Particular solutions are u =c1 xy + c2 x + c3 y + c4 u =c1 3y 2 + x3 + c2 y 3 + x3 y + c3 6xy 2 + x4 1.7.4.14 Biharmonic Oscillator ∆2 u = 0 (~x, t) ∈ Ω × [0, ∞) 1.7.4.15 Fundamental Solution ?? 1.7.4.16 Klein Gordon Equation Lu = utt − uxx + u 1.7.4.17 Fundamental Solution E(x, t) = 2 (~x, t) ∈ Ω × [0, ∞) p 1 H(t − |x|)J0 ( t2 − x2 ) 2 Distribution Theory in DE If f 0 (x) = 0 for a ≤ x ≤ b then f (x) = c classically If T 0 = 0 ∈ D0 (a, b) then T = c ´b ´b Reason: Choose φ0 ∈ C0∞ with a φ0 (x)dx = 1. If φ ∈ C0∞ (a, b) let ψ(x) = φ(x) − a φ(y)dyφ0 (x). Note ´b ´x ψ(x)dx = 0. Let ζ = a ψ(s)ds so ζ 0 = ψ(x) since ζ ∈ C ∞ (a, b) and further ζ ∈ C0∞ since a ´ ζ(a) = ζ(b) = 0 b 0 0 0 and ζ = 0 for x < a or x > b. Then 0 = T (ζ) = −T (ζ ) = −T (ψ) = −T (φ) + a φydy T (φ0 ) so ´b ´b T (φ) = a T (φ0 )φ(y)dy = a cφ(y)dy = c in D0 sense. 2.1 ODEs in D0 sense Let g(x) = ´x a T0 = f f (s)ds (antiderivative of f ). Claim g 0 = f in D0 (a, b). Reason: ˆ Tg0 (φ) f ∈ L1loc (R) 0 ˆ b b ˆ 0 = −Tg (φ ) = − ˆ x 0 g(x)φ (x)dx = − f (s)dsφ (x)dx = − a a a b a So the general solution of T 0 = f in D0 (a, b) is ˆ T = x f (s)ds + C a 14 f (s)φ(s)ds = Tf (φ) b φ0 (x)dxds f (s) a Using FTC and compact support of φ, ˆ b ˆ 0 Tg (φ) = − f (s)(φ(b) − φ(s))ds = a ˆ b s Pierson Guthrey pguthrey@iastate.edu 3 Advanced Theory 4 New Notes Notation Let A ∈ Cn×m , B ∈ Cn×m , then the Hadamar product A:B= m X n X aij bij i=1 i=1 2 kAk = |A : A| Sn are the symmetric n × n matrices. Sn ⊂ Cn×n . B(x, r) is the unit ball in Rn centered at x B(x, 1) is the unit ball in Rn centered at x. B(0, 1) is the unit ball in Rn centered at the origin. S n−1 is the unit sphere in Rn . S n−1 = ∂B(0, 1) ei is the ith coordinate vector in Rn . Rn+ = {y ∈ Rn : yn > 0} is the upper half space of Rn . Let U, V, W be open subsets of Rn . V ⊂⊂ U means V is compactly contained in U . That is V ⊂ U , V is compact. ∂U is the boundary of U . U = ∂U ∪ U UT = U × (0, T ] n ΓT = U T − UT (parabolic boundary) α(n) = Γ πn 2+1 is the volume of the B(0, 1) in RN . nα(n) is the (2 ) surface area of ∂B(0, 1) in RN \0 = \ ∪ {0} Du is the Gradient of u D2 uP is the Hessian of u n The Laplacian of u is ∆u = i=1 uxi xi = tr (D2 u) Airy eqn -3rd order linear ut + uxxx = g(x) KdV eqn -3rd order semilinear ut + uux + uxxx = g(u) Monge-Ampere eqn - fully nonlinear, second order det(D2 u) = f (u) Transport equation solutions ut + b · Du = f, (~x, t) ∈ Rn × (0, ∞) Laplace Poisson eqn ∆u = f Heat Equation ut − ∆u = f Wave Eqn utt − ∆u = f Wave Eqn ˆ ˆ Φ(y)∆x f (x − y)dy + ∆u = Φ(y)∆x f (x − y)dy RN \B(0,) B(0,) The first part vanishes by FTC, and the second part becomes ˆ ˆ ˆ ∂ ∂ = ∆y Φ(y)∆x f (x − y)dy − Φ(y)f (x − y)dS(y) + Φ(y) f (x − y)dS(y) ∂n ∂n RN \B(0,) ∂B(0,) ∂B(0,) 15 Pierson Guthrey pguthrey@iastate.edu ∂ ∂n f (x − y) = 0 away from the origin. We also note that d −1 1 −Dn Φ(y) = − − log(r) |r= = dr 2π 2π ˆ So ∆u(x) = − ∂B(0,) f (x − y) dS(y) = −f (x) 2π ˆ ∂B(x,) 1 dS(y) = −f (x) 2π So ∆u(x) → f (x) uniformly as → 0. If u ∈ C 2 (U ) is Harmonic, then it satisfies the Mean Value Property ´ udS ∂B(x,r) u(x) = udS = ´ dS ∂B(x,r) ∂B(x,r) For all B(x, r) ⊂ U Proof. Set ´ ϕ(ν) = v(ν)dS(y) = ∂B(0,1) v(x + rz)dS(z)rn−1 = k∂B(0, 1)k ∂B(x,r) v(x + rz)dS(z) ∂B(0,1) Since k∂B(0, 1)k = nα(n)rn−1 The average value of a Harmonic function over a sphere is the value at the center of the sphere. ϕ0 (r) = ∂B(0,1) Since y−x r = z n Du(y) · Du(x + rz)zdS(z) = ∂B(x,r) y−x dS(z) r (?) it is the outward unit normal ´ = ∂B(x,r) ∂u dS = ∂n ∂B(x,r) ∆uds |∂B(x, r)| =0 This implies ϕ(r) is a constant. lim ϕ() = lim ?/ =⇒ φ(r) = →0 ffl Corollary: Same holds on B(x, r). u(x) = B(x,r) u(y)dy. Proof. Use the shell method ! ˆ ˆ ˆ →0 r u(y)dy = B(x,r) udS 0 ds = ... = u(x)α(n)rn = u(x) |∂B(x, r)| ∂B(x,s) ffl Theorem. Converse. If u ∈ C 2 (U ) satisfies u(x) = ∂B(x,r) uds ∀ B(x, r) ⊂ U then u is harmonic. Proof. If not, there exists r > 0 : ∆u > 0 on B(x0 , r). As before, ´ ∆udy B(x0 ,r) 0 ϕ (r) = 0 = > 0 ⇒⇐ |∂B(x0 , r)| Strong Maximum Principle Suppose u ∈ C 2 (U ) ∩ C(U ) is harmonic on U , and assume U is bounded. Let M = maxU u. 1. max∂U = M 2. If U is connected and ∃ x0 ∈ U : u(x0 ) = M , then u = M . Corollary, Strong Minimum principle also holds. Suppose u ∈ C 2 (U ) ∩ C(U ) is harmonic on U , and assume U is bounded. Let m = minU u. 16 Pierson Guthrey pguthrey@iastate.edu 1. min∂U = m 2. If U is connected and ∃ x0 ∈ U : u(x0 ) = m, then u = m. ffl Proof of Strong Maximum Principle If ∃ x0 ∈ U : u(x0 ) = M then if B(x0 , r) ⊂ U , M = B(x0 ,r) u(y)dy ≤ M with equality if and only if u = M on B(x0 , r). This implies u = M on B(x0 , r). This implies if S = {x ∈ U : u(x) = M } =⇒ S is relatively open in U . u is continuous u−1 (M ) is closed in U =⇒ S is relatively open, closed. If U is connected, nonempty then U = S. So (2) holds. The continuity implies that (1) holds. Corollary. Consider ( ∆u = 0, x ∈ U v= g, x ∈ ∂U Assume U is connected, u ∈ C 2 (U ) ∩ C(U ). • If g ≥ 0, then u ≥ 0. • If also ∃ s0 ∈ ∂U : g(s0 ) > 0, then u > 0 in U . Theorem 5. This problem has at most one solution. If u1 , u2 are solutions, then y = u1 − u2 solves this problem with g = 0 (the homogenous case). Maximum principle implies y ≥ 0, minimum principle implies y ≤ 0, so y = 0 =⇒ u1 = u2 . Mollifiers The standard mollifier ( −1 2 ce(|x| −1) |x| < 1 η( x) = 0 |x| ≥ 1 η( x) ∈ C0∞ (RN ) ´ If f : U → R ∈ L1loc , define f = η ∗ f = U η (x − y)f (y)dy in U . η = η (x) is a delta sequence ´ • RN η dx = 1 1 n η x • supp η = B(0, ) ⊂ B(0, 1) • η ≥ 0. Theorem p 714 • f ∈ C ∞ (U ), U = {x ∈ U : dist(x, ∂U ) > } • f → f a.e. as → 0 • If f ∈ C(U ) then f → f uniformly on compact subsets of U p • f ∈ LP loc (U ) =⇒ f → f in Lloc (U ) Theorem If U ∈ C(U ) satisfies the mean value property on each B(x, r) ⊂ U then u ∈ C ∞ by showing ˆ ˆ 1 |x − y| U (x) =? (x), x ∈ U (x) = η (x − y)u(y)dy = n η u(y)dy B(x,) U 1 = n = 1 n ˆ η r ˆ 0 ˆ η 0 r ! U ds dr ∂B(x,r) U (x) |∂B(x, r)| dr 17 Pierson Guthrey pguthrey@iastate.edu ˆ = U (x) η dy = U (x) B(0,) If u is harmonic on U then |Dα u(x0 )| ≤ n+1 ck kukL1 (B(,x0 ,r)) r n+k for all B(x0 , r) ⊂ U . |α| = k. where C0 = 1 α(n) k nk) and Ck = (2 α(n) Proof k = 0 |U (x0 )| = ´ |u| dx C0 B(x0 ,r) u(x)dx ≤ = n kU kL1 (B(x0 ,r) n α(n)r r B(x0 ,r) Proof k = 1 ´ ´ u · ni dx ds r B(x , ) B(x0 ,( r ) 2n ( ) 0 |Uxi (x0 )| = uxi (x)dx ≤ ≤ kuk n ∞ r = kU kL1 (B(x0 ,r)) = kU kL1 (B(x0 ,r)) r r B(x0 , r ) B(x0 , ) α(n) ... Liouville’s Theorem. Suppose u : Rn → R is harmonic and bounded. Then u ≡ const. . Proof. Let B = B(x0 , r) √ C1 C nC1 |D(x0 )| ≤ n+1 kU kL1 (B) ≤ n+1 kU kL∞ (B) |B| ≤ kU kL∞ (R) ∀ r r r r =⇒ |D(x0 )| → 0r → ∞ =⇒ |Du| = 0 =⇒ u ≡ const. Theorem 9. Let f ∈ C02 (Rn ), n ≥ 3. Then any bdd solution of −∆u = f in Rn satisfies ˆ u= Φ(x − y)f (y)dy + C Rn ´ Proof. For n ≥ 3 Φ is bdd, Φ → 0 as |x| → ∞ =⇒ Rn Φ(x − y)f (y)dy is bounded. Let v be another bdd solution to ∆v = f , then y = u − v satisfies ∆y = 0, y bdd. so y ≡ const. . Theorem 10. u harmonic in U implies u analytic P on U . u is analytic on U if for all x0 ∈ U , u(x) is given by Dα α its Taylor series in a neighborhood of x0 . Ie u(x) = u (x0 )α!(x − x0 ) . Idea of pf, use Cauchy estimates α to show that (the Taylor remainder term) RN (x0 , x) → 0 as N → ∞. Harmek’s theorem. For all connected open sets V ⊂⊂ U , for all nonnegative Harmonic functions u defined on U , there exists C > 0 for which (C = CV ) sup u ≤ C inf u V V In particular, 1 u(y) ≤ u(x) ≤ Cu(y) ∀ x, y ∈ V C Proof. Let r = dist(V,∂U ) . 4 Let x, y ∈ V , |x − y| ≤ r. ˆ 1 1 u(x) = ≥ udz = n n α(n)(2r) 2 B(x,2r) B(y,r) udz = B(y,r) 1 u(y) 2n so 21n u(y) ≤ u(x) ≤ 2n u(y), where the upper inequality is obtained by symmetry of the first inequality. Then cover V with a finite set of balls of radius 2r (since V is compact and connected), then apply this theorem to N each ball {Bi }i=1 with Bi ∩ Bi−1 6= ∅, i = 2, 3, ...N . 18 Pierson Guthrey pguthrey@iastate.edu Green’s function for −∆u = f 5 −∆u = f ∀ x ∈ U, u = g ∀ x ∈ ∂U for some open bounded U where ∂U ∈ C 1 . Let V = U − B(x, ), x ∈ U . Apply Green’s formula to Φ(y − x), u(y). ˆ ˆ ∂ ∂ u(y)∆Φ(y − x) − Φ(y − x)∆(y) = u(y) Φ(y − x) − Φ(y − x) u(y)dS(y) ∂n ∂n V ∂V Which as → 0, we obtain ˆ ˆ ∂ ∂ Φ(y − x)f (x)dy u(x) = Φ(y − x) u(y) − g(y) Φ(y − x)dS(y) + ∂n ∂n U ∂U Corrector function. Suppose we can find φ∗ (y) that solves ∆y u = 0Φ(y − x) ∈ U, φ∗ (y) = Φ(y − x) ∈ ∂U Apply Green’s function to u1 φ∗ : ˆ ˆ − φ∗ (y)∆u(y)dy = U ∂U u(y) ∂ ∗ ∂ φ (y) − Φ(y − x) u(y)dS(y) ∂n ∂n Combining this with the above equation, ˆ ˆ ∂ ∗ ∂ u(x) = u(y) φ (y) − Φ(y − x) dS(y) + (ψ ∗ (y) − Φ(y − x)) ∆u(y)dS(y) ∂n ∂n ∂U U So we have the Green’s function G(x, y) = Φ(y − x) − φ∗ (y) x, y ∈ U x 6= y. So ˆ ˆ ∂ u(x) = − u(y) G(x, y)dS(y) − G(x, y)∆u(y)dy, x ∈ U ∂n ∂U U Theorem 12. If u ∈ C 2 (U ) solves the Poisson problem, then ˆ ˆ ∂ u(x) = − g(y) G(x, y)dS(y) − G(x, y)f (y)dy, x ∈ U ∂n ∂U U That is, IF a solution exists and if φ∗ can be found, the above is the solution. 5.1 Properties of Green’s functions • G(x, y) = G(y, x) Green’s functions on a half-space. Let Rn+ = {(x1 , ..., xn ) : xn > 0}. If x ∈ Rn+ , define The reflection of x as x̃ = (x1 , ..., −xn ). Define φ∗ (y) = Φ(y − x̃), x, y ∈ Rn+ . Then ∆φ∗ = 0, x ∈ R∗n , φ∗ (y)|∂Rn+ = Φ(y − x) = Φ(y − x̃) Consequently, G(x, y) = Φ(y − x) − Φ(y − x̃) is the Green’s function. So a solution to the Poisson Problem is 1 yn − xn yn + xn ∂ G = −Gyn (x, y) = − (yn Φ(y − x) − yn Φ(y − x̃)) = − n ∂n nα(n) |y − x|n |y − x̃| 19 5.1 Pierson Guthrey pguthrey@iastate.edu Properties of Green’s functions so −2xn ∂ G|∂Rn+ = ∂n nα(n) |x − y n | Define the Poisson Kernel K(x, y) = 2xn nα(n) |x − y n | ˆ then ˆ u(x) = K(x, y)g(y)dy + yn =0 Assume g ∈ C(Rn−1 ) ∩ L∞ (Rn−1 ) then u = ´ ?? G(x, y)f (y)dy U K(x, y)g(y)dy satisfies n−1 ∞ • u = C ∞ (Rn−1 + ) ∩ L (R+ ) • ∆u = 0 in Rn+ • limx→x0 u(x) = g(x0 ) ∀ x0 ∈ ∂Rn+ 5.1.1 Green’s function on the unit ball If x ∈ Rn \ {0} define x̃ = x . |x|2 x̃ is the dual point to x with respect to ∂B(0, 1). Note x, x̃ are parallel vectors, 1 = |x| . ∗ and |x̃| We say R : x → x̃ is the inversion map. Find φ (y) such that ∆φ∗ (y) = 0 ∈ B 0 (0, 1), φ∗ (y) = Φ(y − x) ∈ ∂B(0, 1) Note that Φ(y − x̃) is harmonic in y in B 0 (0, 1) for all x ∈ B(0, 1) so Φ(|x| (y − x̃)) is harmonic in y in B 0 (0, 1). Define the Green’s function for the unit Ball G(x, y) = Φ(y − x) − Φ(|x| (y − x̃)) = Φ(y − x) − φ∗ (y) where 2 −1 1 − |x| ∂ G(x, y) = ∂n nα(n) |x − y|n Define Poisson Kernel for B(0, 1) by 2 K(x, y) = Then u = ´ ∂B(0,1) 1 1 − |x| nα(n) |x − y|n K(x, y)u(y)dy and so ∆u = 0 ∈ B(0, 1), and u = g ∈ ∂B(0, 1). 20