Citation: Ugo Gianazza, Sandro Salsa. On the Harnack inequality for non-divergence parabolic equations[J]. Mathematics in Engineering, 2021, 3(3): 1-11. doi: 10.3934/mine.2021020
[1] | Serena Dipierro, Giovanni Giacomin, Enrico Valdinoci . The fractional Malmheden theorem. Mathematics in Engineering, 2023, 5(2): 1-28. doi: 10.3934/mine.2023024 |
[2] | Petteri Harjulehto, Peter Hästö, Jonne Juusti . Bloch estimates in non-doubling generalized Orlicz spaces. Mathematics in Engineering, 2023, 5(3): 1-21. doi: 10.3934/mine.2023052 |
[3] | Giacomo Ascione, Daniele Castorina, Giovanni Catino, Carlo Mantegazza . A matrix Harnack inequality for semilinear heat equations. Mathematics in Engineering, 2023, 5(1): 1-15. doi: 10.3934/mine.2023003 |
[4] | Prashanta Garain, Kaj Nyström . On regularity and existence of weak solutions to nonlinear Kolmogorov-Fokker-Planck type equations with rough coefficients. Mathematics in Engineering, 2023, 5(2): 1-37. doi: 10.3934/mine.2023043 |
[5] | Daniela De Silva, Ovidiu Savin . On the boundary Harnack principle in Hölder domains. Mathematics in Engineering, 2022, 4(1): 1-12. doi: 10.3934/mine.2022004 |
[6] | Arthur. J. Vromans, Fons van de Ven, Adrian Muntean . Homogenization of a pseudo-parabolic system via a spatial-temporal decoupling: Upscaling and corrector estimates for perforated domains. Mathematics in Engineering, 2019, 1(3): 548-582. doi: 10.3934/mine.2019.3.548 |
[7] | Alessia E. Kogoj, Ermanno Lanconelli, Enrico Priola . Harnack inequality and Liouville-type theorems for Ornstein-Uhlenbeck and Kolmogorov operators. Mathematics in Engineering, 2020, 2(4): 680-697. doi: 10.3934/mine.2020031 |
[8] | Idriss Mazari . Some comparison results and a partial bang-bang property for two-phases problems in balls. Mathematics in Engineering, 2023, 5(1): 1-23. doi: 10.3934/mine.2023010 |
[9] | Fernando Farroni, Gioconda Moscariello, Gabriella Zecca . Lewy-Stampacchia inequality for noncoercive parabolic obstacle problems. Mathematics in Engineering, 2023, 5(4): 1-23. doi: 10.3934/mine.2023071 |
[10] | Nicolai Krylov . On parabolic Adams's, the Chiarenza-Frasca theorems, and some other results related to parabolic Morrey spaces. Mathematics in Engineering, 2023, 5(2): 1-20. doi: 10.3934/mine.2023038 |
The literature about parabolic equations is very large, and it is extremely hard to provide a satisfactory description of all the results. Very nice books such as [3,5,8] try and collect the most significant contributions to this wide field. If one restricts the attention to the field of fully nonlinear parabolic equations, a quite extensive and recent account is given in [1].
In this paper we propose a proof of the Harnack inequality for linear parabolic equations in non-divergence form, originated long time ago from several discussions with Eugene Fabes. For simplicity we consider non-negative solutions of the equation
Lu=Dtu−Tr(A(x,t)D2u)=f | (1.1) |
in a space-time cylinder, where f is smooth, and the matrix A=(aij) is smooth and uniformly elliptic, i.e.,
λ|ξ|2≤aij(x,t)ξiξj≤|ξ|2, |
with λ∈(0,1). In the following, whenever we talk of class of L, we mean an operator as in (1.1), where the matrix A has the same lower and upper eigenvalues.
Precisely, let KR(x0) be the R−cube in Rn, centered at x0, and QR(x0,t0)=KR(x0)×(t0−R2,t0). We have
Theorem 1.1. Let u be a non-negative solution of Lu=f in Q2R(x0,t0). Then there exists a constant c=c(λ,n) such that
supQR(x0,t0−2R2)u≤c{infQR(x0,t0)u+Rn/(n+1)‖f‖Ln+1(Q2R(x0,t0))}. | (1.2) |
The proof we present follows the usual strategy: Local L∞ estimates for subsolutions, weak Harnack inequalities for supersolutions and has some resemblance with the proof in [8]. The main differences are the use of the Green function and a more, we believe, systematic and elementary use of the growth lemmas in Section 3. There is no problem in extending the proof to operators with bounded drift and zero order terms.
The Green function for the operator L can be introduced due the following theorem of Krylov ([2]).
Theorem 1.2. Let u be the solution of the following problem in Qr,T=Kr(0)×(0,T):
{Lu=finQr,Tu=0on∂pQr,T. |
Then
‖u‖L∞(Qr,T)≤c(λ,n)rn/(n+1)‖f‖Ln+1(Qr,T). | (1.3) |
From (1.3) one gets the representation formula
u(x,t)=∫Qr,tGr,T(x,t;ξ,s)f(ξ,s)dξds |
(x,t)∈Qr,t,t≤T, where Gr,T is the Green function for the cylinder Qr,T, and the estimate
sup(x,t)∈Kr(0)×(0,T)(∫Qr,t[Gr,T(x,t;ξ,s)](n+1)/ndξds)n/(n+1)≤c(λ,n)rn/(n+1). | (1.4) |
As a consequence of the Krylov estimate (1.3), an easy check shows that it is enough to prove (1.2) with f=0. In turn, this inequality follows by the combination of the local L∞ estimate (2.1) and the weak Harnak inequality (3.1).
Theorem 2.1. Let u be a non-negative subsolution in Q2r(x0,t0). Then, for all p>0,
‖u‖L∞(Qr/2(x0,t0))≤c(p,λ,n)(1|Qr(x0,t0)|∫Q2r(x0,t0)updxdt)1/p. | (2.1) |
Proof. The function
v(x,t)=u(x0+rx,t0−4r2+r2t) |
satisfies an equation Lrv=0 with Lr in the same class of L, in the cylinder Q2(0,4). Let Qs=Qs(0,4). We want to show that
‖u‖L∞(Q1/2)≤c(p,λ,n)(∫Q1updxdt)1/p. | (2.2) |
It is well-known that it is enough to prove that
‖u‖L∞(Qr)≤c(p,λ,n)(ρ−r)2‖u‖L2(n+1)(Qρ). | (2.3) |
for 1/2≤r≤ρ≤1. For completeness, we show that (2.3) implies (2.2).
From the Young inequality
ab≤(ηa)θθ+(b/η)γγ,a,b≥0,η>0,1θ+1γ=1 |
with η=(εθ)1/θ, and
θ=2(n+1)2(n+1)−p,γ=2(n+1)p,p<2(n+1), |
since γθγ/θ>1, we get
ab≤εaθ+ε−γ/θbγ. |
Choosing
a=(supQρu)(2(n+1)−p)/2(n+1),b=c(p,λ,n)(ρ−r)2(∫Qρupdxdt)1/2(n+1) |
and using (2.3) we get
supQru≤εsupQρu+c(ε,p,λ,n)(ρ−r)4(n+1)/p(∫Qρupdxdt)1/p. |
This inequality is of the form
f(r)≤εf(ρ)+H(ρ−r)−α, | (2.4) |
where
f(ρ)=supQρu,α=4(n+1)/p, and H=c(ε,p,λ,n)(∫Qρupdxdt)1/p. |
Let r0=r, rj+1=rj+(1−τ)(ρ−r)τj, j≥0, with ε<τα<1. Then
f(r)≤εf(r1)+H(1−τ)−α(ρ−r)−α≤ε2f(r2)+H(1−τ)−α(ρ−r)−α(ετ−α+1)....≤εkf(rk)+H(1−τ)−α(ρ−r)−αk−1∑l=0(ετ−α)l. |
Letting k→∞, we get (2.2).
To show (2.3), take a cutoff φ∈C∞0(Qρ), φ=1 on Qr, 0≤φ≤1. We have
|DαDltφ|≤c(α,l)(ρ−r)|α|+l. |
If G2(x,t,y,s) is the Green function for the cylinder Q2, we can write
u(x,t)φ(x,t)=∫t0∫K2G2(x,t;y,s)L(uφ)(y,s)dyds. |
Since Lu≤0, we have
L(uφ)=φLu−2A∇u⋅∇φ+uLφ≤−2A∇u⋅∇φ+uLφ. |
Then, using (1.4) and choosing ψ∈C∞0(Qρ∖¯Qr), ψ=1 on the support of φ,
u(x,t)φ(x,t)≤c(ρ−r)(∫t0∫K2G2(x,t;y,s)|∇u(y,s)|2ψ2(y,s)dyds)1/2+c(ρ−r)2(∫Qρu2(n+1)dyds)1/2(n+1). | (2.5) |
Let Eu=∑ni,j=1aij(x,t)D2iju. We have
∫t0∫K2G2(x,t;y,s)|∇u|2ψ2dyds≤λ−1∫t0∫K2G2(x,t;y,s)ψ2A∇u⋅∇u dyds=λ−1∫t0∫K2G2(x,t;y,s)[12E(u2)−uEu]ψ2dyds=12λ−1∫t0∫K2G2(x,t;y,s)[(E(u2ψ2)−Ds(u2ψ2))]dyds+λ−1∫t0∫K2G2(x,t;y,s)[12Ds(u2ψ2)−uEu ψ2−12u2E(ψ2)]dyds−4λ−1∫t0∫K2G2(x,t;y,s)(A∇u⋅∇ψ)uψdyds=I+II+III. |
Let (x,t)∈Qr. Then I vanishes, since equals −u2(x,t)ψ2(x,t)=0.
For III we find
III≤cελ−1∫t0∫K2G2(x,t;y,s)|∇u|2ψ2dyds+cελ−1∫t0∫K2G2(x,t;y,s)|∇ψ|2u2dyds. |
Using (1.4)
III≤cελ−1∫t0∫K2G2(x,t;y,s)|∇u|2ψ2dyds+cελ−11(ρ−r)2(∫Qρu2(n+1)dyds)1/(n+1). |
For II, since Lu≤0, observe that
12Ds(u2ψ2)−uEu ψ2=uψ2(Dsu−Eu)+ψu2Dsψ≤ψu2Dsψ |
so that
II≤cλ−11(ρ−r)2(∫Qρu2(n+1)dyds)1/(n+1). |
Chosing ε sufficiently small, we conclude.
In this section we prove the weak Harnack inequality for non-negative supersolutions.
Theorem 3.1. Let u be a non-negative supersolution in Q2(0,4). Then there exists p0=p0(λ,n), p0>0, such that
(∫Q1(0,1)up0dyds)1/p0≤c(λ,n)infQ1/2(0,2)u. | (3.1) |
Proof. We may assume that infQ1/2(0,2)u=1. Let
Γz={(x,t)∈Q1(0,1): u(x,t)>z}. |
For any p0>0 we have
∫Q1(0,1)up0dyds=p0∫∞0zp0−1|Γz|dz≤c(p0)+∫∞1zp0−1|Γz|dz. |
As we will show in Section 5, we have
Γz⊂Γz1∪Γz2∪Γz3, | (3.2) |
where
Γz1={(x,t)∈Q1(0,1): |Γz|≤cz},Γz2={(x,t)∈Q1(0,1): |Γz|≤cz1/M},Γz3={(x,t)∈Q1(0,1): |Γz|≤1ρ|Γγz|}, |
with c,M,ρ,γ depending only on λ and n, and ρ>1,0<γ<1. Then
∫Q1(0,1)up0dyds≤c(p0)+p0∫∞1zp0−1|Γz∩Γz1|dz+p0∫∞1zp0−1|Γz∩Γz2|dz+p0∫∞1zp0−1|Γz∩Γz3|dz≤c(p0)+c1(p0)+c2(p0,M)+1ργp0∫Q1(0,1)up0dyds. |
Choosing p0≪1 such that ργp0>1, we get
∫Q1(0,1)up0dyds≤c(λ,n). |
The next sections will be devoted to the proof of (3.2).
In view of the proof of (3.2), we need two fundamental lemmas. The first one states that if u is a non-negative supersolution in Q2r=Q2r(x0,t0) and it is greater than z on a sizable portion of Qr, then infu>cz on a full smaller cylinder. Therefore, a measure-theoretical information is converted in a pointwise information.
Lemma 4.1. Let u≥0, Lu≥0 in Q2r. There exist ξ∈(0,1) and c>0, both depending only on λ and n, such that, if
|{(x,t)∈Qr:u(x,t)>z}|≥ξ|Qr|, |
then
infQr/2u>cz. |
Proof. We may assume r=1,z=1,x0=0,t0=1. As in the previous Section, we let
Γ1={(x,t)∈Q1: u(x,t)>1}. |
Consider the function
wΓ1(x,t)=∫Γ1G1(x,t;y,s)dyds. |
If we let Q1=Q1(0,1), we have
LwΓ1(x,t)=χΓ1(x,t) and wΓ1(x,t)=0 on ∂pQ1. |
From Theorem 1.2
supQ1wΓ1(x,t)≤c0. |
Since u≥0 on ∂pQ1 and u>1 on Γ1, the maximum principle gives
u(x,t)≥c−10wΓ1(x,t) in Q1. |
Thus, it is enough to show that wΓ1(x,t)≥c>0 in Q1/2. We have:
wΓ1(x,t)=wQ1(x,t)−wQ1∖Γ1(x,t)≥wQ1(x,t)−|Q1∖Γ1|1/(n+1)(∫Q1G1(x,t;y,s)(n+1)/ndyds)n/(n+1)≥wQ1(x,t)−c(1−ξ)1/(n+1). |
Take a cutoff ψ∈C∞0(Q1), ψ=1 on Q1/2. Then v=wQ1−ψ/‖Lψ‖∞ satisfies
Lv=1−Lψ‖Lψ‖∞≥0 in Q1 and v=0 on ∂pQ1. |
Thus, wQ1(x,t)≥ψ/‖Lψ‖∞ in Q1 and, since ψ=1 on Q1/2, ‖Lψ‖∞=c1(λ,n)>0, we conclude that
wΓ1(x,t)≥wQ1(x,t)−c(1−ξ)1/(n+1)≥c−11−c(1−ξ)1/(n+1)≥c2(λ,n)>0, |
provided ξ=ξ(λ,n) is suitably chosen.
The second lemma is a variant of the so-called growth lemma: if u is strictly positive in a small ball at level t=0, this positivity expands parabolically upwards.
The growth lemma was originally introduced by Landis ([6]), first in the context of elliptic equations, and later extended to parabolic ones (for an exposition of these ideas, we refer the interested reader to [7]). The deep significance of the growth lemma was later shown by Krylov and Safonov, in their celebrated proof of the local Hölder continuity of solution to equations like (1.1) ([4]). A similar and slightly easier version was used by Safonov for the corresponding proof in the elliptic case ([9]). Since then, the growth lemma has become a sort of standard tool in the regularity theory of elliptic and parabolic equations in non-divergence form.
Lemma 4.2. Let u≥0,Lu≥0 in B2r(0)×(0,4r2). Assume that u(x,0)≥1 for |x|≤εr,0<ε≤1. Then, there exist M=M(λ,n) and c=c(λ,n) such that, for each α∈(0,4−ε2),
u(x,αr2)≥cεMfor|x|≤√α+ε24r. |
Proof. We may assume r=1. Let
ψ(x,t)={(1−4|x|2t+ε2)4ε2q(t+ε2)q if 4|x|2t+ε2≤1,0 otherwise. |
Claim: if q=q(λ,n) is large, then Lψ≤0 in Rn+1+. Indeed,
Lψ=ε2q(t+ε2)q+1(1−4|x|2t+ε2)2{16|x|2t+ε2(1−4|x|2t+ε2)−q(1−4|x|2t+ε2)2−768Ax⋅xt+ε2+32(1−4|x|2t+ε2)(trA)}. |
If 1−δ≤4|x|2t+ε2≤1, with a suitable δ=δ(λ,n), then Lψ≤0. If 0<4|x|2t+ε2≤1−δ, then Lψ≤0 for q large.
Applying the maximum principle, we find
u(x,t)≥ψ(x,t) in Q1. |
In particular, for 0≤t=α≤4−ε2 and |x|≤√α+ε2/4, letting M=2q yields
u(x,α)≥(34)414M/2εM=c(λ,n)εM, |
since M=M(λ,n), due to its definition in terms of q.
Remark 4.3. The extension to a fully nonlinear parabolic equation such as
Dtu−F(D2u)=0, |
where F falls within the same class of ellipticity as the matrix governing the problem in (1.1), should not require too much of an effort. Indeed, as shown in [1,Section 4], the key-point is the proof of the weak Harnack inequality. Since the barrier employed in the proof of Lemma 4.2 is a radial function, the same argument works also in the fully nonlinear case considered above. The difficult step is represented by Lemma 4.1: This should require the construction of a second, proper barrier, as in [1], and then the analogue of the Krylov-Safonov estimates. We refrain to speculate further on this issue here, since it goes beyond the limits of the present manuscript. We plan to address this topic in a separate paper.
In this section we prove (3.2), concluding the proof of the weak Harnack inequality. As it will be clear in the following, Lemma 5.1 is a straightforward consequence of Lemmas 4.1–4.2: whenever one has at disposal these two results, the structure of the equation plays no further role in the proof.
We let Γz and ξ as in Lemma 4.1, and introduce the notation
Q+s(x0,t0)=Bs(x0)×(t0+bs2,t0+s2η], |
where b and η are to be chosen depending only on λ,n. We have
Lemma 5.1. If |Γz|>ξ then
|Γz|≤czinfQ1/2(0,2)u | (5.1) |
with c=c(λ,n).
On the other hand, if |Γz|≤ξ, there exist c1>0, γ∈(0,1), ρ>1, M>0, all depending only on λ and n, such that, either
|Γz|M≤c1zinfQ1/2(0,2)u, | (5.2) |
or
|Γz|≤1ρ|Γγz|. | (5.3) |
Proof. Let |Γz|>ξ. Then, Lemma 4.1 gives
infQ1/2(0,1)u≥cz |
and Lemma 4.2 with ε=1 gives
infQ1/2(0,2)u≥c1z≥c1z|Γz| |
which is (5.1).
Let now |Γz|≤ξ. We apply the Calderon-Zygmund decomposition to f=χΓz to find a sequence of non overlapping cylinders Qrj contained in Q1=Q1(0,1), satisfying the following conditions:
i)|Γz∖∪Qrj|=0;
ii)|Γz∩Qrj|>ξ|Qrj|;
iii) each Qrj is contained in a predecessor ˜Qj⊂Q1 such that the ˜Qj are non overlapping and
|Γz∩˜Qj|≤ξ|˜Qj|. |
Let D=⋃j˜Qj and D+=⋃j˜Q+j. From ii) and Lemma 4.1 we infer
infQrj/2u≥c(λ,n)z, |
and from Lemma 4.2 we get
inf˜Q+ju>γz, |
where γ∈(0,1) depends on b and η in the definition of Q+j. In turn, b is chosen accordingly to Lemma 4.2 and η will be chosen later. Thus
u>γz in D+. | (5.4) |
Now we use the following lemma, whose proof we postpone to the end.
Lemma 5.2. With the same notation as before, we have
|D+|≥1−bηη+1|D|. | (5.5) |
Let δ be a small positive number and assume first that
|D+∖Q1|≤δ|Γz|. |
Then, from Lemma 5.2,
|D+∩Q1|=|D+|−|D+∖Q1|≥|D+|−δ|Γz|≥1−bηη+1|D|−δ|Γz|. | (5.6) |
Since |Γz∖D|=0,
|D|=∑j|˜Qj|≥1ξ∑j|˜Qj∩Γz|=1ξ|Γz|. |
It follows form (5.4) that
|Γγz|≥(1−bη(η+1)ξ−δ)|Γz|. |
Choosing η and δ such that
ρ=1−bη(η+1)ξ−δ>1, |
(5.3) follows.
Now assume that
|D+∖Q1|>δ|Γz|. | (5.7) |
We distinguish two cases.
Case a) There exists j such that r2j≥δ2b|Γz|. Since
infQrj/2u≥cz, |
from Lemma 4.2 with ε=rj/2,r=2, we get
infQ1/2(0,2)u≥c(rj2)Mz≥c|Γz|M/2z, |
which gives (5.2).
Case b) For all j, r2j≤δ2b|Γz|. We show that there exist rj and c=c(δ,η) such that
r2j≥c|Γz|. | (5.8) |
It is enough to show that the height of one of the ˜Q+j is greater than δ|Γz|/2. To prove it we show that the bottom of at least one of the ˜Q+j is at time level t<1+δ|Γz|/2, and its top at a time level t>1+δ|Γz|.
Indeed, since r2j≤δ2b|Γz|, the bottom of ˜Q+j is at a time level t<1+br2j<1+δ|Γz|/2. If the top of every ˜Q+j were at a time level ≤1+δ|Γz|, we would have
D+⊂Q1∪[B1(0)×(0,1+δ|Γz|)], |
so that
|D+∖Q1|≤δ|Γz| |
contradicting (5.7).
Acting as in case a) we obtain (5.2).
We are left with the proof of Lemma 5.2. Let
˜Qj=K˜rj(xj)×(t0−˜r2j,t0] |
and set
ˆQj=˜Qj∪[K˜rj(xj)×(t0,t0+b˜r2j]],ˆD=∪ˆQj. |
Note that
˜Q+j=K˜rj(xj)×(t0+b˜r2j,t0+1η˜r2j]=K˜rj(xj)טI+j |
and |˜I+j|=1−ηbη˜r2j. Moreover,
ˆQj∪˜Q+j=K˜rj(xj)×(t0−˜r2j,t0+1η˜r2j]=K˜rj(xj)׈I+j |
and |ˆI+j|=1+ηη˜r2j=1+η1−ηb|˜I+j|.
Let x∈K˜rj(xj). Then
|{t: (x,t)∈D+∪ˆD}|≤1+η1−ηb|˜I+j|=1+η1−ηb|{t: (x,t)∈D+}|, |
so that, integrating with respect to x, we get
|D+|≥1−bηη+1|D+∪ˆD|, |
which implies (5.5).
The authors declare no conflict of interest.
[1] | Imbert C, Silvestre L (2013) An introduction to fully nonlinear parabolic equations, In: An Introduction to the Kähler-Ricci Flow, Cham: Springer, 7-88. |
[2] | Krylov NV (1983) Boundedly inhomogeneous elliptic and parabolic equations in a domain. Izv Akad Nauk SSSR Ser Mat 47: 75-108. |
[3] | Krylov NV (1987) Nonlinear Elliptic and Parabolic Equations of the Second Order, Dordrecht: D. Reidel Publishing Co. |
[4] | Krylov NV, Safonov MV (1980) A property of the solutions of parabolic equations with measurable coefficients. Izv Akad Nauk SSSR Ser Mat 44: 161-175. |
[5] | Ladyzenskaja OA, Solonnikov VA, Ural'tzeva NN (1967) Linear and Quasilinear Equations of Parabolic Type, Providence: American Mathematical Society. |
[6] | Landis EM (1968), Harnack's inequality for second order elliptic equations of Cordes type. Dokl Akad Nauk SSSR 179: 1272-1275. |
[7] | Landis EM (1998) Second Order Equations of Elliptic and Parabolic Type, Providence: American Mathematical Society. |
[8] | Lieberman GM (1996) Second Order Parabolic Differential Equations, World Scientific. |
[9] | Safonov MV (1980) Harnack's inequality for elliptic equations and Hölder property of their solutions. Zap Nauchn Sem Leningrad Otdel Mat Inst Steklov 96: 272-287. |
1. | Diego Maldonado, Linear parabolic operators of Monge-Ampère type I: Nondivergence-form degenerate/singular PDEs, 2023, 345, 00220396, 161, 10.1016/j.jde.2022.11.032 |