Solution to Partial Differential Equations: Methods and Applications (Robert McOwen) Section 5.2
- Proof of Theorem 1. Let K(x,y,t) be the Gaussian kernel (18).
- Show that K(x,y,t) is C∞ and satisfies (∂t−Δx)K(x,y,t)=0 for x,y∈Rn and t>0.
- Prove that for every x∈Rn and t>0, ∫RnK(x,y,t)dy=1.
- For any δ>0, show that
limt→0+∫|x−y|>δK(x,y,t)dy=0
uniformly for x∈Rn. - Use (a)-(c) and g∈CB(Rn) to show u(x,t)→g(x0) as (x,t)→(x0,0+), proving the theorem.
- We firstly recall that the heat kernel K given by
K(x,y,t)=1(4πt)n/2exp(−|x−y|24t)for x,y∈Rn and t>0.
It is clear that K is C∞(Rn×Rn×R+) because all functions composing K are smooth. A direct computation gives∂K∂t(x,y,t)=−n21(4πt)n/2texp(−|x−y|24t)+1(4πt)n/2exp(−|x−y|24t)⋅|x−y|24t2,∂K∂xi(x,y,t)=1(4πt)n/2exp(−|x−y|24t)⋅−2(xi−yi)4t,∂2K∂x2i(x,y,t)=1(4πt)n/2exp(−|x−y|24t)⋅(xi−yi)24t2−12(4πt)n/2texp(−|x−y|24t).
Hence it is easy to verify that(∂t−Δx)K(x,y,t)=∂K∂t−n∑i=1∂2K∂x2i=−n21(4πt)n/2exp(−|x−y|24t)+1(4πt)n/2exp(−|x−y|24t)⋅|x−y|24t2−1(4πt)n/2exp(−|x−y|24t)⋅|x−y|24t2−n2(4πt)n/2texp(−|x−y|24t)=0.
The proof is complete. - For every x∈Rn and t>0, it is clear that
∫RnK(x,y,t)dy=1(4πt)n/2∫Rnexp(−|x−y|24t)dy=1(4πt)n/2∫Rnexp(−14tn∑i=1(xi−yi)2)dy=1(4πt)n/2n∏i=1∫Rexp(−(xi−yi)24t)dyi=1(4πt)n/2n∏i=1⋅√4πt=1.
Here we have used the fact that ∫Rexp(−k(x−y)2)dy=√πk for k>0. - Using the polar coordinate, we have
∫|x−y|>δK(x,y,t)dy=1(4πt)n/2∫Sn−1∫∞δexp(−r24t)⋅rn−1drdωSn−1=Hn−1(Sn−1)(4πt)n/2∫∞δexp(−r24t)⋅rn−1dr,
where Hn−1(Sn−1) denotes the (n−1)-dimensional Hausdorff measure of Sn−1, i.e., the surface area of Sn−1. Using the change of variable with s=r/√4t, the integral can be transformed into∫|x−y|>δK(x,y,t)dy=Hn−1(Sn−1)πn/2∫∞δ/√4tsn−1exp(−s2)ds
Since ∫∞0sn−1exp(−s2)ds is convergent and limt→0+δ√4t=∞, we obtainlimt→0+∫|x−y|>δK(x,y,t)dy=Hn−1(Sn−1)πn/2limt→0+∫∞δ/√4tsn−1exp(−s2)ds=0.
Note that this convergence rate is independent of x so the convergnece is uniform in x. - Fix x0∈Rn and ϵ>0. Since g is continuous in Rn, there exists δ>0 such that |g(y)−g(x0)|<ϵ for |y−x0|<δ and y∈Rn. For |x−x0|<δ/2, we can use (b) to find that
|u(x,t)−g(x0)|=|∫RnK(x,y,t)[g(y)−g(x0)]dy|≤∫Bδ(x0)K(x,y,t)|g(y)−g(x0)|dy+∫Rn−Bδ(x0)K(x,y,t)|g(y)−g(x0)|dy.
Clearly, we have∫Bδ(x0)K(x,y,t)|g(y)−g(x0)|dy≤ϵ∫Bδ(x0)K(x,y,t)dy≤ϵ∫RnK(x,y,t)dy=ϵ.
Here we have used (b) again. On the other hand, if |y−x0|≥δ, we observe that|y−x0|≤|y−x|+|x−x0|≤|y−x|+δ2≤|y−x|+12|y−x0|,
which implies |y−x|≥12|y−x0|. Hence, by (c), we have∫Rn−Bδ(x0)K(x,y,t)|g(y)−g(x0)|dy≤2supy∈Rn|g(y)|∫Rn−Bδ(x0)K(x,y,t)dy→0as t→0+,
which means∫Rn−Bδ(x0)K(x,y,t)|g(y)−g(x0)|dy≤ϵfor 0<t≤t0,
where t0 is sufficiently small. Therefore, |u(x,t)−g(x0)|<2ϵ for |x−x0|<δ/2 and 0<t<t0 and the proof is complete.
Remark. The original statement is that "u(x,t)→g(x) as t→0+", which cannot be regarded as the continuous extension of u to t≥0. A suitable continuous extension of u has been shown. - Let g(x) be bounded and continuous for x∈Rn and define u by (19).
- Show |u(x,t)|≤sup{|g(y)|:y∈Rn}.
- If, in addition, ∫Rn|g(y)|dy<∞, show that limt→∞u(x,t)=0 uniformly in x∈Rn.
- By part (b) of Exercise 1, we have
|u(x,t)|=|∫RnK(x,y,t)g(y)dy|≤∫RnK(x,y,t)|g(y)|dy≤supy∈Rn|g(y)|∫RnK(x,y,t)dy=supy∈Rn|g(y)|.
Here we have used the fact that K≥0 in Rn×(0,∞). - A direct estimate implies
|u(x,t)|=|∫RnK(x,y,t)g(y)dy|≤1(4πt)n/2∫Rnexp(−|x−y|24t)|g(y)|dy≤1(4πt)n/2∫Rn|g(y)|dy.
Here we have used the fact that K≥0 in Rn×Rn×(0,∞) and exp(−|x−y|24t)≤1 for x,y∈Rn and t∈(0,∞). Since limt→∞1(4πt)n/2∫Rn|g(y)|dy=0 and is independent of x∈Rn, we obtain limt→∞u(x,t)=0 uniformly in x∈Rn. The proof is complete. - Let g(x)∈Ck(Rn) with Dαg uniformly bounded on Rn for each |α|≤k. Show that u defined by (19) satisfies u∈Ck,[k/2](Rn×[0,∞)).
- For Dαxu, we find
Dαxu(x,t)=Dαx1(4πt)n/2∫Rnexp(−|x−y|24t)g(y)dy=1(4πt)n/2Dαx∫Rnexp(−|y|24t)g(x−y)dy=1(4πt)n/2∫Rnexp(−|y|24t)(Dαg)(x−y)dy=1(4πt)n/2∫Rnexp(−|x−y|24t)(Dαg)(y)dy.
Since Dαg is bounded and continuous in Rn, we can apply Theorem 1 for Dαxu(x,t), which can be continuously extended to Rn×{0}. - For ∂[(m+1)/2]tu, we can use ∂tK(x,y,t)=ΔxK(x,y,t) to find
∂[(m+1)/2]tu(x,t)=∂[(m+1)/2]t∫RnK(x,y,t)g(y)dy=∫Rn(Δ[(m+1)/2]xK)(x,y,t)g(y)dy=∫Rn(Δ[(m+1)/2]yK)(x,y,t)g(y)dy=∫RnK(x,y,t)(Δ[(m+1)/2]g)(y)dy.
Since Δ[(m+1)/2]g is bounded and continuous in Rn, we can apply Theorem 1 for ∂[(m+1)/2]tu(x,t), which can be continuously extended to Rn×{0}. - Formally check that
u(x,t)=∞∑k=01(2k)!x2kdkdtke−1/t2
satisfies ut=uxx for t>0, x∈R with u(x,0)=0 for x∈R. (Proving the convergence of the infinite series is not easy; see [John, 1] or [Widder].) - Prove the following weak maximum principle for the heat equation in U=Rn×(0,T): Let u be bounded and continuous on ˉU=Rn×[0,T] with ut,uxixj∈C(U) and ut−Δu≤0 in U. Then
M≡sup(x,t)∈Uu(x,t)=supx∈Rnu(x,0)≡m.
- Use the maximum principle of (a) to show that the solution (19) is the unique solution that is bounded in Rn×[0,T].
- Prove the following weak maximum principle for the heat equation in U=Rn×(0,T): Let u be bounded and continuous on ˉU=Rn×[0,T] with ut,uxixj∈C(U) and ut−Δu≤0 in U. Then
- Clearly, M≥m follows from the fact that (Rn×{0})⊊ˉU. It suffices to show that M≤m. For this purpose, we fix x0∈Rn and consider the function
v(x,t)=u(x,t)−ϵ(3nt+|x−x0|2)for (x,t)∈ˉU and ϵ>0.
It is easy to verify thatvt(x,t)−Δv(x,t)=ut(x,t)−Δu(x,t)−ϵn<0in U.
Since u is bounded, lim|x|→∞v(x,t)=−∞ for any t∈[0,T], which implies there exists R>0 (independent of t) such that v(x,t)≤−1+infx∈Rnu(x,0) for |x−x0|≥R and t∈R. On the other hand, v(x0,t)=u(x0,0)≥infx∈Rnu(x,0)>v(x,t) for |x−x0|≥R and t∈[0,T]. Hence for any τ∈(0,T), we can apply the weak maximum principle for heat equation on the bounded domain:supRn×(0,τ)v(x,t)=supBR(x0)×(0,τ)v(x,t)=maxˉBR(x0)×[0,τ]v(x,t)=max(ˉBR(x0)×{0})∪(∂BR(x0)×[0,τ])v(x,t).
Let (x∗,t∗)∈(ˉBR(x0)×{0})∪(∂BR(x0)×[0,τ]) be the maximum point of v. Suppose by contradiction that (x∗,t∗)∈∂BR(x0)×(0,τ]. Then vt(x∗,t∗)≥0 and Δv(x∗,t∗)≤0, which implies vt(x∗,t∗)−Δv(x∗,t∗)≥0. This leads a contradiction. Due to the arbitrariness of τ∈(0,T), we haveu(x0,t)−3ϵnT≤u(x0,t)−3ϵnt=v(x0,t)≤sup(x,t)∈Rn×[0,T]v(x,t)=maxx∈ˉBR(x0)v(x,0)≤supx∈Rnv(x,0)=supx∈Rn[u(x,0)−ϵ|x−x0|2]≤supx∈Rnu(x,0)=m.
Letting ϵ→0+, we obtain u(x0,t)≤m for any (x0,t)∈Rn×[0,T]. Therefore, we obtain M≤m and complete the proof. - Suppose ˜u and u are bounded solution of the heat equation ut(x,t)−Δu(x,t)=0 in U with initial condition u(x,0)=g(x) for x∈Rn, where g∈CB(Rn). Let w(x,t)=˜u(x,t)−u(x,t) for (x,t)∈Rn×R+. Then w is also bounded and satisfies
{wt(x,t)−Δw(x,t)=0for (x,t)∈U,w(x,0)=0for x∈Rn.
By the result of part (a) for w and −w, we have0=−supx∈Rn(−w)(x,0)=−sup(x,t)∈U(−w)(x,t)=inf(x,t)∈Uw(x,t)≤sup(x,t)∈Uw(x,t)=supx∈Rnw(x,0)=0,
which shows w≡0, and hence ˜u≡u in ˉU. Therefore, we complete the proof of the uniqueness of bounded solution of heat equation with initial condition. - For |x|≠0, show that ˜K (given in (22)) is smooth (C∞) in t∈R. Conclude that ˜K is smooth for (x,t)≠(0,0).
- Heat conduction in a semi-infinite rod with initial temperature g(x) leads to the equations
{ut=uxxfor x>0 and t>0,u(x,0)=g(x)for x>0.
Assume that g is continuous and bounded for x≥0.- If g(0)=0 and the rod has its end maintained at zero temperature, then we must include the boundary condition u(0,t)=0 for t>0. Find a formula for the solution u(x,t).
- If the rod has its end insulated so that there is no heat flow at x=0, then we must include the boundary condition ux(0,t)=0 for t>0. Find a formula for the solution u(x,t). Do you need to require g′(0)=0?
- The problem can be formulated as
{ut=uxxfor x>0 and t>0,u(x,0)=g(x)for x>0,u(0,t)=0for t>0.
Here g is continuous and bouneded for x≥0 with g(0)=0. To solve this problem, we use the reflection method to extend g by˜g(x)={g(x)if x≥0,−g(−x)if x<0.
Note that ˜g is continuous and bounded in Rn since g(0)=0. Now we pay attention on the following pure initial value problem{˜ut=˜uxxfor x∈R and t>0,˜u(x,0)=˜g(x)for x∈R.
By Theorem 1 at page 149 in the textbook, we get˜u(x,t)=1√4πt∫Rexp(−(x−y)24t)˜g(y)dy=1√4πt[∫∞0exp(−(x−y)24t)˜g(y)dy+∫0−∞exp(−(x−y)24t)˜g(y)dy]=1√4πt[∫∞0exp(−(x−y)24t)g(y)dy−∫0−∞exp(−(x−y)24t)g(−y)dy]=1√4πt[∫∞0exp(−(x−y)24t)g(y)dy−∫∞0exp(−(x+y)24t)g(y)dy]=1√4πt∫∞0[exp(−(x−y)24t)−exp(−(x+y)24t)]g(y)dy=1√πt∫∞0exp(−x2+y24t)sinh(xy2t)g(y)dy.
It is clear that ˜u(0,t)=0 because sinh(0)=0. Therefore, the solution u can be obtained from the restriction of ˜u over [0,∞)×[0,∞), i.e., u=˜u|[0,∞)×[0,∞). - The problem can be formulated as
{ut=uxxfor x>0 and t>0,u(x,0)=g(x)for x>0,ux(0,t)=0for t>0.
To solve this problem, we use the reflection method to extend g by˜g(x)={g(x)if x≥0,g(−x)if x<0.
Clearly, ˜g is continuous and bounded in R. Now we pay attention on the following pure initial value problem{˜ut=˜uxxfor x∈R and t>0,˜u(x,0)=˜g(x)for x∈R.
By Theorem 1 at page 149 in the textbook, we get˜u(x,t)=1√4πt∫Rexp(−(x−y)24t)˜g(y)dy=1√4πt[∫∞0exp(−(x−y)24t)˜g(y)dy+∫0−∞exp(−(x−y)24t)˜g(y)dy]=1√4πt[∫∞0exp(−(x−y)24t)g(y)dy+∫0−∞exp(−(x−y)24t)g(−y)dy]=1√4πt[∫∞0exp(−(x−y)24t)g(t)dy+∫∞0exp(−(x+y)24t)g(y)dy]=1√4πt∫∞0[exp(−(x−y)24t)+exp(−(x+y)24t)]g(y)dy=1√πt∫∞0exp(−x2+y24t)cosh(xy2t)g(y)dy.
To verify that the boundary condition holds, we notice that˜ux(x,t)=1√πt∫∞0[exp(−x2+y24t)⋅−x2tcosh(xy2t)+exp(−x2+y24t)⋅sinh(xy2t)⋅y2t]g(y)dy,
which implies˜ux(0,t)=1√πt∫∞0[exp(−y24t)⋅0+exp(−y24t)⋅0⋅y2t]g(y)dy=0.
Therefore, the solution u can be obtained from the restriction of ˜u over [0,∞)×[0,∞), i.e., u=˜u|[0,∞)×[0,∞). Clearly, we do not need to require g′(0)=0. - If u∈C2,1(Rn×[0,∞)) and f∈C(Rn+1), show that (29) is equivalent to (27).
- Use (30) to prove Theorem 3.
- Verify the following elementary fact: If 0<α<1 and β≥0, then there exists a constant M=M(α,β)>0 so that zβe−z≤Me−αz for all z≥0.
- Use (b) to verify the following estimates:
∂t˜K(x,t)≤M1t˜K(x,2t),∂xi˜K(x,t)≤M2√t˜K(x,2t),∂xixj˜K(x,t)≤M3t˜K(x,2t).
- Use (30) and (c) to confirm the Remark following Theorem 3.
- In order to prove Theorem 3, we need to check that function u defined by
u(x,t)=∫t0∫Rn˜K(x−y,t−s)f(y,s)dyds
satisfies nonhomogeneous heat equation with homogeneous initial condition:{ut=Δu+f(x,t)for x∈Rn and t>0,u(x,0)=0for x∈Rn.
Here f∈C2,1(Rn×[0,T]) is bounded for every T>0 and ˜K is defined in (22) (also see Exercise 6). In addition, we setM=max{sup(x,t)∈Rn|ft(x,t)|,sup(x,t)∈Rn|∇f(x,t)|,sup(x,t)∈Rn|D2xf(x,t)|}.
Firstly we change variable to writeu(x,t)=∫t0∫Rn˜K(y,s)f(x−y,t−s)dyds.
Since f∈C2,1(Rn×[0,T]) is bounded for every T>0 and ˜K=˜K(y,s) is smooth near s=t>0, we getut(x,t)=∫t0∫Rn˜K(y,s)ft(x−y,t−s)dyds+∫Rn˜K(y,t)f(x−y,0)dy,uxixj(x,t)=∫t0∫Rn˜K(y,s)fxixj(x−y,t−s)dyds.
Clearly, ut and uxixj are continuous so u∈C2,1(Rn×(0,∞)). Then we findut(x,t)−Δu(x,t)=∫t0∫Rn˜K(y,s)[∂t−Δx]f(x−y,t−s)dyds+∫Rn˜K(y,t)f(x−y,0)dy=(∫ϵ0+∫tϵ)∫Rn˜K(y,s)[−∂s−Δy]f(x−y,t−s)dyds+∫Rn˜K(y,t)f(x−y,0)dy
For 0≤t≤T, we note that|∫ϵ0∫Rn˜K(y,s)[−∂s−Δy]f(x−y,t−s)dyds|≤2M∫ϵ0∫Rn˜K(y,s)dyds=2Mϵ.
Using the integrating by parts, we find∫tϵ∫Rn˜K(y,s)[−∂s−Δy]f(x−y,t−s)dyds=∫tϵ∫Rn[∂s−Δy]˜K(y,s)f(x−y,t−s)dyds+∫Rn˜K(y,ϵ)f(x−y,t−ϵ)dy−∫Rn˜K(y,t)f(x−y,0)dy,
which implies∫tϵ∫Rn˜K(y,s)[−∂s−Δy]f(x−y,t−s)dyds+∫Rn˜K(y,t)f(x−y,0)dy=∫Rn˜K(y,ϵ)f(x−y,t−ϵ)dy.
Here we have used the fact that ˜Ks(y,s)=Δy˜K(y,s) for y∈Rn and ϵ≤s≤t. Thus, by Proposition we obtainut(x,t)−Δu(x,t)=limϵ→0+∫Rn˜K(y,ϵ)f(x−y,t−ϵ)dy=f(x,t).
Here we have used the fact that (21), i.e., limt→0+˜K(x,t)=δ0(x)=δ(x) as distribution on Rn. Finally, we note that |u(x,t)|≤tM→0 as t→0+, which gives u(x,0)=0. The proof is complete. - Consider the function g:[0,∞)→R defined by
g(z)=zβe(α−1)zfor z≥0.
It is easy to seeg′(z)=βzβ−1e(α−1)z+zβe(α−1)z(α−1)=g(z)z(β+(α−1)z).
By solving g′(z)=0, we get z=β/(1−α). Since g(z)≥0, we have g′(z)>0 on (0,(1−α)−1β) and g′(z)<0 on ((1−α)−1β,∞), which implies g attina its maximum value at z=(1−α)−1β. Therefore, we findg(z)≤g(β1−α):=M(α,β)=M,
which gives zβe−z≤Me−αz for all z≥0. - Some direct computations give
∂t˜K(x,t)=∂∂t[(4πt)−n/2exp(−|x|24t)]=(−n2)(4πt)−(n+2)/2exp(−|x|24t)+(4πt)−n/2exp(−|x|24t)⋅|x|24t2=(−n8πt+|x|24t2)(4πt)−n/2exp(−|x|24t)≤2n/2t(4π(2t))−n/2⋅|x|24texp(−|x|24t)≤2n/2t(4π(2t))−n/2⋅M(12,1)exp(−|x|28t)=2n/2M(1/2,1)t˜K(x,2t):=M1t˜K(x,2t),
where M1=2n/2M(1/2,1). On the other hand, we have∂xi˜K(x,t)=∂∂xi(4πt)−n/2exp(−|x|24t)=(4πt)−n/2exp(−|x|24t)⋅−xi2t≤2n/2√t(4π(2t))−n/2|x|2√texp(−|x|24t)2n/2√t(8πt)−n/2⋅M(12,12)exp(−|x|28t)=2n/2M(1/2,1/2)√t˜K(x,2t):=M2√t˜K(x,2t),
where M2=2n/2M(1/2,1/2). Moreover, we have∂xixj˜K(x,t)=∂∂xj[(4πt)−n/2exp(−|x|24t)⋅−xi2t]=(4πt)−n/2exp(−|x|24t)[xixj4t2−δij2t]≤2n/2t⋅(4π(2t))−n/2⋅|x|24texp(−|x|24t)≤2n/2t(8πt)−n/2⋅M(12,1)exp(−|x|28t)=2n/2M(1/2,1)t⋅˜K(x,2t):=M3t˜K(x,2t),
where M3=2n/2M(1/2,1). The proof is complete. - I am not sure why we need to use (c). In my opinion, we need to prove the equality ut=Δu+f(x,t) so we cannot use too weak inequality unless we can drop those terms. I think there is a formal derivation as follows.
ut(x,t)=∫Rn(Δx˜K)(x−y,0)f(y,t)dyds+∫t0∫Rn˜Kt(x−y,t−s)f(y,s)dyds=f(x,t)+∫t0∫Rn(Δx˜K)(x−y,t−s)f(y,s)dyds=Δu(x,t)+f(x,t)in Rn×(0,T).
Since f is continuous and bounded in Rn×[0,T], we have |u(x,t)|≤tsup(y,s)∈Rn×[0,T]|f(y,s)|, which implies limt→0+u(x,t)=0 for any x∈Rn. Therefore, we complete the proof. - If f(x,t) is continuous and bounded on Rn+1 and vanishes for t sufficiently negative [i.e., f(x,t)=0 for t≤τ<0], show that the convolution u=˜K∗f defines a continuous weak solution of the nonhomogeneous heat equation ut=Δu+f(x,t) in Rn+1.
- Find a formula for the solution of the initial value problem
{ut=Δu−ufor x∈Rn and t>0,u(x,0)=g(x)for x∈Rn,
where g is continuous and bouneded. Is the solution bounded? Is it the only bounded solution?
Solution
Solution
Solution
We prove the statement by mathematical induction. Clearly, the statement holds true for k=0 due to Theorem 1 at page 149 in the textbook. Suppose the statement holds true for k=m, i.e., u∈Cm,[m/2](Rn×[0,∞)) when g∈Cm(Rn). Now we suppose that g∈Cm+1(Rn). Then by the inductive assumption, we haveu∈Cm,[m/2](Rn×[0,∞))∩Cm+1,[(m+1)/2](Rn×(0,∞))⊊Cm,[m/2](Rn×[0,∞))∩C∞,∞(Rn×(0,∞)).
To show that u∈Cm+1,[(m+1)/2](Rn×[0,∞)), we need to prove Dαxu (|α|=m+1) and ∂[(m+1)/2]tu can be continously extended to Rn×{0}. Since u∈C∞(Rn×(0,∞)), it is clear that Dαxu and ∂[(m+1/2)]t satisfies the heat equation:(Dαxu)t=Δ(Dαxu)in Rn×(0,∞),(∂[(m+1)/2]tu)t=Δ(∂[(m+1)/2]tu)in Rn×(0,∞).
Remark. In fact, when g∈Ck(Rn), Dαx∂stu are continuous for multi-index α and nonnegative integer s satisfying |α|+2s≤k.
Solution
Here we suppose that the differentiation and summation can be intercahgned freely. A direct computation showsut(x,t)=∂∂t∞∑k=01(2k)!x2kdk+1dtk+1e−1/t2=∞∑k=11(2k−2)!x2k−2dkdtke−1/t2=∞∑k=11(2k)!⋅(2k)(2k−1)x2k−2dkdtke−1/t2=∞∑k=11(2k)!(d2dx2x2k)dkdtke−1/t2=∞∑k=01(2k)!(d2dx2x2k)dkdtke−1/t2=∂2∂x2∞∑k=01(2k)!x2kdkdtke−1/t2=uxx(x,t).
Next we prove that limt→0+u(x,t)=0. For this purpose, it suffices to show limt→0+dkdtke−1/t2=0 for all k∈N∪{0}, which implieslimt→0+u(x,t)=limt→0+∞∑k=01(2k)!x2kdkdtke−1/t2=∞∑k=01(2k)!x2klimt→0+dkdtke−1/t2=0=u(x,0)for x∈R.
Let f(t)=e−1/t2 for t≠0 and f(0)=0. Then for n∈N, we use the mathematical induction to prove that there exist a(n)1,…,a(n)2n−1 such thatf(n)(t)={2n−1∑k=1a(n)ktn+k+1e−1/t2if t≠0,0if t=0.
When n=1, we have f′(t)=2t3e−1/t2 for t≠0. For t=0, we observe thatf′(0)=limh→0f(h)−f(0)h=limh→0e−1/h2h=limh→01/he1/h2=limh→0−1h2−2h3e1/h2=limh→0h2e1/h2=0.
Thus, we getf′(t)={2t3e−1/t2if t≠0,0if t=0.
Here a(1)1=2. Now we suppose that the statement holds true for n=m, i.e., there exists a(m)1,…,a(m)2m−1 such thatf(m)(t)={2m−1∑k=1a(m)kxm+k+1e−1/x2if t≠0,0if t=0.
For x≠0, we havef(m+1)(t)=2m−1∑k=1−a(m)k(m+k+1)tm+k+2e−1/t2+2m−1∑k=1−2a(m)ktm+k+4e−1/t2=2m−1∑k=1−a(m)k(m+k+1)tm+k+2e−1/t2+2m+1∑k=3−2a(m)k−2tm+k+2e−1/t2=2m+1∑k=1a(m+1)ktm+k+2e−1/t2,
where a(m+1)1=−(m+2)a(m)1, a(m+1)2=−(m+3)a(m)2, a(m+1)2m=−2a(m)2m−2, a(m+1)2m+1=−2a(m)2m−1, and a(m+1)k=−(m+k+1)a(m)k−2a(m)k−2 for 3≤k≤2m−1. For x=0, we havef(m+1)(0)=limh→0f(m)(h)−f(m)(0)h=limh→02m−1∑k=1a(m)khm+k+2e−1/h2=2m−1∑k=1limh→0a(m)khm+k+2e−1/h2=0.
Here we have used the same technique to calculuate the limit. Therefore, by mathematical induction, we complete the proof of statement, which tells us f(n)(0)=0, i.e., limt→0dkdtke−1/t2=0 for all k∈N.Solution
Solution
Recall that˜K(x,t)={1(4πt)n/2e−|x|2/4tfor t>0,0for t≤0.
To show ˜K is smooth in t for |x|≠0, it suffices to prove ∂mt˜K(x,0)=0 for |x|≠0 and m∈N. The major observation is based on the calculation in the solution of Exercise 4, which shows that∂m˜K∂tm(x,t)={m∑i=0(didti(4πt)−n/2)(∂m−i∂tm−ie−|x|2/4t)for t>0,0for t<0.
Clearly limt→0∂m˜K∂tm(x,t)=0 and hence ˜K is smooth in t. On the other hand, for t≠0, ˜K is obviously smooth for all x∈Rn. Therefore, ˜K is smooth for (x,t)≠(0,0).The original statement does not follow English grammar so I correct it directly.
Solution
Solution
Suppose that u∈C2,1(Rn×[0,∞)) and f∈C(Rn+1) satisfies (29), i.e.,∫∞0∫Rn(uvt+uΔv+fv)dxdt=0for v∈C∞c(Rn+1).
Using the intergration by parts, we find0=∫∞0∫Rn(uvt+uΔv+fv)dxdt=∫Rn[u(x,t)v(x,t)|t=∞t=0−∫∞0utvdt]dx+∫∞0∫Rn(Δu)vdxdt+∫∞0∫Rnfvdxdt=−∫Rnu(x,0)v(x,0)dx+∫∞0∫Rn(−ut+Δu+f)vdxdt.
Due to arbitrariness of v∈C∞c(Rn+1), we get −ut+Δu+f≡0 in Rn×(0,∞) and u(x,0)=0 on Rn, which gives (27).Conversely, we suppose that u∈C2,1(Rn×[0,∞)) and f∈C(Rn+1) satisfies (27), i.e.,
{ut=Δu+f(x,t)for (x,t)∈Rn×(0,∞),u(x,0)=0for x∈Rn.
Multiplying it by v∈C∞c(Rn+1) and then integrating it over Rn×[0,∞), we get∫∞0∫Rn(−ut+Δu+f(x,t))vdxdt=0.
Using the integration by parts, we have∫∞0∫Rn(uvt+uΔv+fv)dxdt=0,
which means (29). Here we have used the condition that v has compact support in Rn+1.Therefore these calculations end the proof.Solution
Solution
Since f is defined in Rn+1 and vanishes for t sufficiently negative, the convolution u can be denoted asu(x,t)=∫t0∫Rn˜K(x−y,t−s)f(y,s)dyds=∫∞−∞∫Rn˜K(x−y,t−s)f(y,s)dyds=∫Rn+1˜K(x−y,t−s)f(y,s)dyds.
Here we have used the fact that ˜K(x−y,t−s)=0 for s≥t. Using the integration by parts, we can verify that∫Rn+1∫Rn+1[˜K(x−y,t−s)f(y,s)(ϕt(x,t)+Δϕ(x,t))]dydsdxdt+∫Rn+1f(x,t)ϕ(x,t)dxdt=0
for any ϕ∈C∞c(Rn+1). Hecne we need to show that u is continuous in Rn+1. Clearly, u(x,t)=0 for t<0. Let us consider the function v(x,t,s)=∫RnK(x,y,t−s)f(y,s)dy, which is continuous in (x,s)∈Rn×[0,t] and by Theorem 1 with the intial data g(y)=f(y,s). Therefore, the solution u=∫t0v(x,t,s)ds is also continuous in (x,t)∈Rn+1. The proof is complete.Solution
Let v(x,t)=etu(x,t) for x∈Rn and t>0. Then v satisfies{vt=Δvfor x∈Rn and t>0,v(x,0)=g(x)for x∈Rn.
Since g is continuous and bouneded in Rn, we can apply Theorem 1 to getv(x,t)=∫RnK(x,y,t)g(y)dy=1(4πt)n/2∫Rnexp(−|x−y|24t)g(y)dy,
which givesu(x,t)=e−t(4πt)n/2∫Rnexp(−|x−y|24t)g(y)dy.
Since v is bounded and e−t≤1, this solution u is also bounded in Rn×[0,∞). To show the uniqueness of bounded solution, we suppose that ˜u is also a bounded solution of{˜ut=Δ˜u−˜ufor x∈Rn and t>0,˜u(x,0)=g(x)for x∈Rn.
Let w=e−t(˜u−u) in Rn×[0,∞). Then w is also bounded in Rn×[0,∞) and satisfies{wt=Δwfor x∈Rn and t>0,w(x,0)=0for x∈Rn.
By the result of part (b) of Exercise 5, we have w≡0 in Rn×[0,T] for any T>0, which means w≡0 and hence ˜u≡u in Rn×[0,∞).
沒有留言:
張貼留言