Citation: Michal Fečkan, Július Pačuta, Michal Pospíśil, Pavol Vidlička. Averaging methods for piecewise-smooth ordinary differential equations[J]. AIMS Mathematics, 2019, 4(5): 1466-1487. doi: 10.3934/math.2019.5.1466
[1] | Tekin Toplu, Mahir Kadakal, İmdat İşcan . On n-Polynomial convexity and some related inequalities. AIMS Mathematics, 2020, 5(2): 1304-1318. doi: 10.3934/math.2020089 |
[2] | Hengxiao Qi, Muhammad Yussouf, Sajid Mehmood, Yu-Ming Chu, Ghulam Farid . Fractional integral versions of Hermite-Hadamard type inequality for generalized exponentially convexity. AIMS Mathematics, 2020, 5(6): 6030-6042. doi: 10.3934/math.2020386 |
[3] | Naila Mehreen, Matloob Anwar . Some inequalities via Ψ-Riemann-Liouville fractional integrals. AIMS Mathematics, 2019, 4(5): 1403-1415. doi: 10.3934/math.2019.5.1403 |
[4] | Ghulam Farid, Saira Bano Akbar, Shafiq Ur Rehman, Josip Pečarić . Boundedness of fractional integral operators containing Mittag-Leffler functions via (s,m)-convexity. AIMS Mathematics, 2020, 5(2): 966-978. doi: 10.3934/math.2020067 |
[5] | Yousaf Khurshid, Muhammad Adil Khan, Yu-Ming Chu . Conformable fractional integral inequalities for GG- and GA-convex functions. AIMS Mathematics, 2020, 5(5): 5012-5030. doi: 10.3934/math.2020322 |
[6] | Yousaf Khurshid, Muhammad Adil Khan, Yu-Ming Chu . Conformable integral version of Hermite-Hadamard-Fejér inequalities via η-convex functions. AIMS Mathematics, 2020, 5(5): 5106-5120. doi: 10.3934/math.2020328 |
[7] | Saad Ihsan Butt, Artion Kashuri, Muhammad Umar, Adnan Aslam, Wei Gao . Hermite-Jensen-Mercer type inequalities via Ψ-Riemann-Liouville k-fractional integrals. AIMS Mathematics, 2020, 5(5): 5193-5220. doi: 10.3934/math.2020334 |
[8] | Muhammad Uzair Awan, Nousheen Akhtar, Artion Kashuri, Muhammad Aslam Noor, Yu-Ming Chu . 2D approximately reciprocal ρ-convex functions and associated integral inequalities. AIMS Mathematics, 2020, 5(5): 4662-4680. doi: 10.3934/math.2020299 |
[9] | Yue Wang, Ghulam Farid, Babar Khan Bangash, Weiwei Wang . Generalized inequalities for integral operators via several kinds of convex functions. AIMS Mathematics, 2020, 5(5): 4624-4643. doi: 10.3934/math.2020297 |
[10] | Xiuzhi Yang, G. Farid, Waqas Nazeer, Muhammad Yussouf, Yu-Ming Chu, Chunfa Dong . Fractional generalized Hadamard and Fejér-Hadamard inequalities for m-convex functions. AIMS Mathematics, 2020, 5(6): 6325-6340. doi: 10.3934/math.2020407 |
For N≥2, we let Ω0 be a bounded open set of RN with boundary Γ0. Consider the heat equation with initial data the indicator function of the set Ω0:
{∂u∂t−Δu=0in RN×(0,t1]u(x,0)=1Ω0(x) on RN. | (1.1) |
for some time t1>0. In 1992, Bence-Merriman-Osher [5] provided a computational algorithm for tracking the evolution in time of the set Ω0 whose boundary Γ0 moves with normal velocity proportional to its classical mean curvature. At time t1>0, they considered
Ω1={x∈RN:u(x,t1)≥1/2}. |
Bence-Merriman-Osher [5] applied iteratively this procedure to generate a sequence of sets (Ωj)j≥0 and conjectured in [5] that their boundaries Γj evolved by mean curvature flow.
Later Evans [16] provided a rigorous proof for the Bence-Merriman-Osher algorithm by means of the level-set approach to mean curvature flow developed by Osher-Sethian [32], Evans-Spruck [17,18,19,20] and Chen-Giga-Goto [10]. For related works in this direction, we refer the reader to [4,25,28,29,31,34,36] and references therein.
Recently Caffarelli and Souganadis considered in [9] nonlocal diffusion of open sets E⊂RN given by
{∂u∂t+(−Δ)su=0in RN×(0,∞)u(x,0)=τE(x)in RN×{t=0}, | (1.2) |
where
τE(x)=1E(x)−1RN∖¯E(x). |
We consider the fractional heat kernel Ks with Fourier transform given by ˆK(ξ,t)=e−t|ξ|2s. It satisfies
{∂Ks∂t+(−Δ)sKs=0 in RN×(0,∞)Ks=δ0 on RN×{t=0}. |
It follows that the unique bounded solution to (1.2) is given by
u(x,t)=Ks(⋅,t)⋆τE(x)=∫RN Ks(x−y,t)τE(y)dy. | (1.3) |
By solving a finite number of times (1.2) for a small fixed time step σs(h), the authors in [9] find a discrete family of sets
Eh0=E,Ehnh={x∈RN:Ks(⋅,σs(h))⋆τEh(n−1)h(x)>0}, |
for a suitable scaling function σs to be defined below. It is proved in [9] that as nh→t, ∂Ehnh converges, in a suitable sense, to Γt. Here, the family of hypersurface {Γt}t>0, with Γ0=∂E, evolves under generalized mean curvature flow for s∈[12,1) and under generalized fractional mean curvature flow for s∈(0,12). We refer the reader to [9,16,27] for the notion generalized (nonlocal) mean curvature flow which considers the level sets of viscosity solutions to quasilinear parabolic integro-differential equations.
In the present paper, we are interested in the normal velocity of the sets
Et:={x∈RN:κs(⋅,σs(t))⋆τE(x)>0} | (1.4) |
as they depart from a sufficiently smooth initial set E0:=E. We consider here and in the following
Ks(x,t)=t−N2sPs(t−12sx), for some radially symmetric function Ps∈C1(RN). | (1.5) |
We make the following assumptions:
C−1N,s1+|y|N+2s≤Ps(y)≤CN,s1+|y|N+2s,|∇Ps(y)|≤CN,s1+|y|N+2s+1. | (1.6) |
and
limt→0t−1κs(y,t)=CN,s|y|N+2s locally uniformly in RN∖{0}, | (1.7) |
for some constants CN,s,CN,s>0. In the Section 1.1 below, we provide examples of valuable kernels Ks satisfying the above properties.
Now, as we shall see below (Lemma 2.1), for t>0 small, ∇xu(x,σs(t))≠0 for all x∈B(y,σs(t)12s) and y∈∂E. Hence ∂Et is a C1 hypersurface, for small t>0. For t>0 and y∈∂E, we let v=v(t,y) be such that
y+vtν(y)∈∂Et∩B(y,σs(t)12s), | (1.8) |
where ν(y) is the unit exterior normal of E at y. In the spirit of the work of Evans [16] on diffusion of smooth sets, we provide in this paper an expansion of v(t,y) as t→0. It turns out that v(0,y) is proportional to the fractional mean curvature of ∂E at y for s∈(0,1/2) and v(0,y) is proportional to the classical mean curvature of ∂E at y for s∈[1/2,1).
We notice that it is not a priori clear from (1.8), that v remains finite as t→0. This is where the (unique) appropriate choice of σs(t) enters during our estimates. Here and in the following, we define
σs(t)={t2s1+2sfor s∈(0,1/2),tsfor s∈(1/2,1) | (1.9) |
and for s=1/2, σs(t) is the unique positive solution to
t=σ1/2 2(t)|log(σ1/2 (t))|. | (1.10) |
Before stating our main result, we recall that for s∈(0,12) and ∂E is of class C1,β for some β>2s, the fractional mean curvature of ∂E is defined for x∈∂E as
Hs(x):=P.V.∫RN τE(y)|x−y|N+2sdy=limε→0∫RN∖Bε(x)τE(y)|x−y|N+2sdy. |
On the other hand, if ∂E is of class C2 then the normalized mean curvature of ∂E is given, for x∈∂E, by
H(x):=N−1N+1limε→01ε|Bε(x)|∫Bε(x)τE(y)dy, |
see also (2.4) and [21]. Having fixed the above definitions, we now state our main result.
Theorem 1.1. We let s∈(0,1) and E⊂RN, N≥2. We assume, for s∈(0,1/2), that ∂E is of class C1,β for some β>2s and that ∂E is of class C3, for s∈[1/2,1). Then, as t→0, the expansion of v(t,y), defined in (1.8), is given, locally uniformly in y∈∂E, by
v(t,y)={aN,sHs(y)+ot(1)for s∈(0,1/2)bN(t)H(y)+O(1log(σ1/2 (t)))for s=1/2cN,sH(y)+O(t2s−12)for s∈(1/2,1), |
where Hs and H are respectively the fractional and the classical mean curvatures of ∂E and the positive constants aN,s, bN,1/2 and cN,s are given by
aN,s=CN,s2∫RN−1 Ps(y′,0)dy′,bN(t)=∫BN−1σ12(t)−1 |y′|2P1/2 (y′,0)dy′2|log(σ12(t))|∫RN−1 P1/2 (y′,0)dy′,cN,s=∫RN−1 |y′|2Ps(y′,0)dy′2∫RN−1 Ps(y′,0)dy′ |
and Ps(y):=κs(y,1).
Some remarks are in order. The assumption of E being of class C3 in Theorem 1.1 is motivated by the result of Evans in [16], where in the case s=1 and K1, the heat kernel, he obtained v=(N−1)H(0)+O(t12). We notice that from our argument below, we cannot improve the error term ot(1) in the case s∈(0,1/2) even if E is of class C∞. This is due to the definition of the fractional mean curvature Hs as a principal value integral. Using polar coordinates and the estimates in (1.6), it easy follows that
|SN−2|C−1N,1/22∫RN−1 P1/2 (y′,0)dy′≤limt→0bN(t)≤|SN−2|CN,1/22∫RN−1 P1/2 (y′,0)dy′. |
This then justify the choice of the scaling when s=12. In the particular case, that κ1/2 (y,t)=CNt(t2+|y|2)N+12, we have that
bN(t)=|SN−2|CN2∫RN−1 P1/2 (y′,0)dy′+O(1log(σ1/2 (t))). |
For the general case, we get the following
We next put emphasis on two valuable examples where Theorem 1.1 applies.
1) Fractional heat diffusion of smooth sets. We recall, see e.g., [6,35], that the fractional heat kernel Ks satisfies (1.5), (1.6) and (1.7) with
CN,s=s22ssin(sπ)Γ(N2+s)Γ(s)π1+N2. | (1.11) |
We recall that Ks is known explicitly only in the case s=1/2, where K1/2 (y,t)=CN,1/2t(t2+|y|2)N+12. In this case Theorem 1.1 provides an approximation of the (fractional) mean curvature motion by fractional heat diffusion of smooths sets, thereby extending, in the fractional setting, Evan's result in [16] on heat diffusion of smooth sets.
2) Diffusion of smooth sets by Harmonic extension. We consider the Poisson kernel on the half space RN+1+:=RN×(0,∞), given by
¯Ks(x,t):=t−NPs(x/t),Ps(x)=pN,s(1+|x|2)N+2s2, | (1.12) |
where pN,s:=1∫RN (1+|y|2)−N+2s2dy. Thanks to the result of Caffarelli and Silvestre in [8], the function
w(x,t)=¯Ks(⋅,t)⋆τE(x)=pN,st2s∫RN τE(y)(t2+|y−x|2)N+2s2dy |
solves
{div(t1−2s∇w)=0in RN×(0,∞)w=τEon RN×{t=0}. |
It is clear that Ks(x,t):=¯Ks(x,t12s) satisfies (1.5), (1.6) and (1.7) with CN,s=pN,s. Hence, Theorem 1.1 provides an expansion of the normal velocities of the boundary of the sets
Et:={x∈RN:¯Ks(⋅,σs(t)12s)⋆τE(x)>0}, |
where σs(t) is given by (1.9) and (1.10). Therefore this Harmonic extension yields an approximation of (fractional) mean curvature motion of smooth sets.
We conclude Section 1 by noting that the notion of nonlocal curvature appeared for the first time in [9]. Later on, the study of geometric problems involving fractional mean curvature has attracted a lot of interest, see [1,7], the survey paper [21] and the references therein. While the mean curvature flow is well studied, see e.g., [2,3,15,24,26], its fractional counterpart appeared only recently in the literature, see e.g., [11,12,13,14,27,30,33].
We finally remark that the changes of normal velocity of the nonlocal diffused sets as s varies in (0, 1/2) and [1/2, 1), appeared analogously in phases transition problems, see e.g., [22,23,33].
Unless otherwise stated, we assume for the following that E is an open set of class C1,β, with 0∈∂E and the unit normal of ∂E at 0 coincides with eN. We denote by Qr=BN−1r×(−r,r) the cylinder of RN centred at the origin with BN−1r the ball of RN−1 centred at the origin with radius r>0. Decreasing r, if necessary, we may assume that
E∩Qr={(y′,yN)∈BN−1r×R:yN>γ(y′)}, | (2.1) |
with γ∈C1,β(BN−1r) satisfying
γ(y′)=O(|y′|1+β). | (2.2) |
In the following, for f,g:R→R, we write g(t):=O(f(t)) if
|g(t)|≤C|f(t)|. |
We also write g(t)=o(f(t)) if g(t)=O(f(t)) and moreover when f(t)≠0, we have
limt→0|g(t)||f(t)|=0. |
We denote by ot(1) any function that tends to zero when t→0.
If in addition, ∂E is of class C3, then for y′∈BN−1r, we have
γ(y′)=12D2γ(0)[y′,y′]+O(|y′|3) | (2.3) |
and the normalized mean curvature of ∂E at 0 is given by
H(0)=Δγ(0)N−1=N−1N+1limε→01ε|Qε(0)|∫Qε(0)τE(y)dy. | (2.4) |
Recall that the unit exterior normal ν(y′):=ν(y′,γ(y′)) of E and the volume element dσ(y′) on ∂E∩Qr are given by
ν(y′)=(−∇γ(y′),1)√1+|∇γ(y′)|2anddσ(y′)=√1+|∇γ(y′)|2dy′. | (2.5) |
We finally note, in view of (1.5) and (1.6), that we have
0<κs(y,t)≤Ct|y|N+2s for all y∈RN∖{0},t>0, | (2.6) |
for some positive constant C=C(N,s). We start with the following result.
Lemma 2.1. Let s∈(0,1) and E be a set of class C1,β satisfying (2.1). Define
w(z,t)=∫RN κs(z−y,t)τE(y)dy. |
Then there exist t0,C>0, only depending on N,s,β and E, such that for all t∈(0,t0) and z∈Bt12s,
∂w∂zN(z,t)≥Ct−1/2s. | (2.7) |
As a consequence, for all t∈(0,t0), the set
{z∈RN:w(z,t)=0}∩Bt12sis of class C1. | (2.8) |
Proof. We fix t>0 small so that t12s<r8 and let z∈Bt12s. We write
∂w∂zN(z,t)=∫RN ∂κs∂zN(z−y,t)τE(y)dy=∫Br2(z)∂κs∂yN(z−y,t)τE(y)dy+∫RN∖Br2(z)∂κs∂yN(z−y,t)τE(y)dy. | (2.9) |
By a change of variable, (1.5) and (1.6), we have
∫RN∖Br2(z)∂κs∂yN(z−y,t)dy=t−12s∫RN∖t−12sBr2(z)∂Ps∂yN(t−12sz−y)dy=O(t−12s∫RN∖t−12sBr2(z)|t−12sz−y|−N−1−2sdy)=O(t). | (2.10) |
Integrating by parts, we have
∫Br2(z)∂κs∂yN(z−y,t)τE(y)dy=∫Br2(z)∩E∂κs∂yN(z−y,t)τE(y)dy−∫Br2(z)∩Ec∂κs∂yN(z−y,t)dy=2∫Br2(z)∩∂Eκs(z−y,t)eN⋅νE(y)dσ(y)+∫∂Br(z)κs(z−y,t)eN⋅νBr2(y)τE(y)dσ′(y). |
By a change of variable, (1.5), (1.6) and the fact that Qr/8⊂Br/4⊂Br2(z)⊂Qr, we have
∫Br2(z)∩∂Eκs(z−y,t)eN⋅νE(y)dσ(y)≥C∫BN−1r/8 κs(z′−y′,zN−γ(y′),t)dy′=Ct−12s∫BN−1r8t−12sPs(t−12sz′−y′,t−12szN−t−12sγ(t12sy′))dy′≥Ct−12s∫BN−1r8t−12s∖B2dy′1+|y′|N+2s, | (2.11) |
provided r8>t12s. Next, using (2.6) and recalling that z∈Bt12s, we then have
|∫∂Br2(z)κs(z−y,t)eN⋅νBr2(z)(y)dσ′(y)|≤∫∂Br2(z)κs(z−y,t)dσ′(y)≤tC∫∂Br2(z)1|y|N+2sdσ′(y)=O(t). |
From this and (2.11), we deduce that
∫Br(z)∂κs∂yN(z−y,t)τE(y)dy≥Ct−1/2s. |
Combining this with (2.9) and (2.10), we get
∂w∂zN(z,t)≥Ct−1/2s. |
Therefore (2.7) follows. Finally (2.8) follows from the inverse function theorem and the fact that w is of class C1 on RN×(0,∞).
In the sequel, we will need the following lemmas to estimate some error terms.
Lemma 2.2. For s∈(0,1), we let E⊂RN be a set of class C1,β, for some β>2s, as in Section 2. For r>0, we set
Jr(t):=∫Qrκs(y,t)τE(y)dyandIr(t)=∫RN∖Qr(t−1κs(y,t)−CN,s|y|N+2s)τE(y)dy. |
Then we have
|Jr(t)|≤Ctrβ−2sandlimt→0Ir(t)=0, |
where C is a positive constant depending only on N, β, s and E.
Proof. Since τE=1E−1RN∖¯E, we get
Jr(t)=∫Qr∩Eκs(y,t)dy−∫Qr∩Ecκs(y,t)dy=∫BN−1r ∫rγ(y′)κs((y′,yN),t)dyNdy′−∫BN−1r ∫γ(y′)−rκs((y′,yN),t)dyNdy′=∫BN−1r (∫rγ(y′)κs((y′,yN),t)dyN−∫γ(y′)−rκs((y′,yN),t)dyN)dy′=∫BN−1r (∫rγ(y′)κs((y′,yN),t)dyN−∫−γ(y′)−rκs((y′,yN),t)dyN−∫γ(y′)−γ(y′)κs((y′,yN),t)dyN)dy′=∫BN−1r (∫rγ(y′)κs((y′,yN),t)dyN+∫γ(y′)rκs((y′,−yN),t)dyN−∫γ(y′)−γ(y′)κs((y′,yN),t)dyN)dy′. |
Since the map y⟼Ks(y,t) is radial, we have κs(y′,yN,t)=κs(y′,−yN,t) so that
∫rγ(y′)κs((y′,yN),t)dyN+∫γ(y′)rκs((y′,−yN),t)dyN=0. |
Therefore
Jr(t)=−∫BN−1r ∫γ(y′)−γ(y′)κs((y′,yN),t)dyNdy′=−2∫BN−1r ∫γ(y′)0κs((y′,yN),t)dyNdy′. |
Then, by (2.6),
|Jr(t)|≤2CN,st|∫BN−1r ∫γ(y′)01|(y′,yN)|N+2sdyNdy′|≤Ctrβ−2s. |
where C is a positive constant depending on N, β and s and which may change from a line to another. Next, using (2.6), (1.7) and the dominate convergence theorem, we obtain
limt→0Ir(t)=0. |
This then ends the proof.
Lemma 2.3. Let s∈(0,1) and let x=vteN∈∂Et∩Bσs(t)12s, with Et given by (1.4). Then
vt=O((σs(t))1+2s2s)as t→0. |
Proof. Since x=vteN∈∂Et, we have that u(x,σs(t))=0. By the fundamental theorem of calculus, we have
u(x,σs(t))=u(0,σs(t))+vt∫10∂u∂xN(θvteN,σs(t))dθ=0 |
so that
vt∫10∂u∂xN(θvteN,σs(t))dθ=−u(0,σs(t)). | (2.12) |
We write
u(0,σs(t))=∫RN κs(y,σs(t))τE(y)dy=∫Qrκs(y,σs(t))τE(y)dy+∫Qcrκs(y,σs(t))τE(y)dy. |
Then by Lemma 2.2 and (2.6), we have
|∫Qrκs(y,σs(t))τE(y)dy|≤Cσs(t) and ∫Qcrκs(y,σs(t))τE(y)dy=O(σs(t)) | (2.13) |
for some constant C depending on r. Furthermore by (2.7), we have
∂u∂xN(θvteN,σs(t))≥C(σs(t))−1/2s. | (2.14) |
Therefore, by (2.12), (2.13) and (2.14), we obtain
|vt|=u(0,σs(t))|∫10∂u∂xN(θvten,σs(t))dθ|≤Cσs(t)(σs(t))−12s=C(σs(t))1+2s2s. |
This then ends the proof.
Lemma 2.4. Under the assumptions of Lemma 2.3, we have
∫10∫BN−1r∫γ(y′)−vt0yN∂Ks∂yN(y′,θyN,σs(t))dydθ=O(σs(t))as t→0. |
Proof. Let θ∈[0,1]. By (1.5), (1.6) and a change of variable, we have
∫BN−1r∫γ(y′)−vt0yN∂Ks∂yN(y′,θyN,σs(t))dy=(σs(t))−N−12s∫BN−1r∫γ(y′)−vt0yN∂Ps∂yN(y′(σs(t))−12s,θyN(σs(t))−12)dy≤C∫BN−1r(σs(t))−1/2s∫(σs(t))−1/2s(γ((σs(t))1/2sy′)−vt)0yN1+|y′|N+2s+1dy≤C(σs(t))−1/s∫BN−1r(σs(t))−1/2s(γ((σs(t))1/2sy′)−vt)21+|y′|N+2s+1dy=O(σs(t))+O(v2t2(σs(t))−1/s)+O(vt). |
Applying Lemma 2.3, we get
∫γ(y′)−vt0yN∂Ks∂yN(y′,θyN,σs(t))dydθ=O(σs(t)), |
as t→0. This then ends the proof.
In this section, we start by the following preliminary result.
Lemma 3.1. Let s∈(0,1/2). We assume that E is of class C1,β for some β>2s satisfying (2.1). Then, for all θ∈[0,1], we have
∫RN ∂κs∂yN(y′,yN−vtθ,σs(t))τE(y)dy=2(σs(t))−1/2s∫RN−1 Ps(y′,0)dy′+O((σs(t))2s−12s). | (3.1) |
Proof. We have
∫RN ∂κs∂yN(y′,yN−vtθ,σs(t))τE(y)dy=∫Br∂κs∂yN(y′,yN−vtθ,σs(t))τE(y)dy+∫Bcr∂κs∂yN(y′,yN−vtθ,σs(t))τE(y)dy, |
where Br is the ball of RN centered at the origin and of radius r>0. By integration by parts, we have
∫Br∂κs∂yN(y′,yN−vtθ,σs(t))τE(y)dy=2∫∂E∩Brκs(y′,yN−vtθ,σs(t))νN(y)dσ(y)+∫∂Brκs(y′,yN−vtθ,σs(t))eN⋅νBr(y)τE(y)dσ′(y). |
Therefore
∫RN ∂κs∂yN(y′,yN−vtθ,σs(t))τE(y)dy=2∫∂E∩Brκs(y′,yN−vtθ,σs(t))νN(y′)dσ(y′) +∫Bcr∂κs∂yN(y′,yN−vtθ,σs(t))τE(y)dy+∫∂Brκs(y′,yN−vtθ,σs(t))yNrτE(y)dσ′(y). | (3.2) |
Then by a change of variable and (2.5), we have
∫∂E∩Brκs(y′,yN−vtθ,σs(t))νN(y)dσ(y)=∫BN−1r κs(y′,γ(y′)−vtθ,σs(t))dy′. |
By the Fundamental Theorem of calculus, we can write
κs(y′,γ(y′)−vtθ,σs(t))=κs(y′,0,σs(t))+(γ(y′)−vtθ)∫10∂κs∂yN(y′,θ′(γ(y′)−vtθ)),σs(t))dθ′. | (3.3) |
In the following, we let
β(y′):=γ(y′)−vtθ. | (3.4) |
Then we have
∫∂E∩Brκs(y′,yN−vtθ,σs(t))νN(y)dσ(y)=∫BN−1r κs(y′,β(y′),σs(t))dy′=∫BN−1r κs(y′,0,σs(t))dy′+∫10∫BN−1r β(y′)∂κs∂yN(y′,θ′β(y′),σs(t))dy′dθ′. | (3.5) |
Therefore By a change of variable, (1.5) and (1.6), we have
∫BN−1r κs(y′,0,σs(t))dy′=∫BN−1r (σs(t))−N2sPs(y′(σs(t))−1/2s,0)dy′=(σs(t))−1/2s∫RN−1 Ps(y′,0)dy′+(σs(t))−1/2s∫RN−1∖BN−1r(σs(t))−1/2s Ps(y′,0)dy′=(σs(t))−1/2s∫RN−1 Ps(y′,0)dy′+O((σs(t))−1/2s(σs(t))1+2s2s). | (3.6) |
By a change of variable, (1.6) and (3.4), we have
∫BN−1r β(y′)∂κs∂yN(y′,θ′β(y′),σs(t))dy′=(σs(t))−1/s∫BN−1r(σs(t))−1/2s β(y′(σs(t))1/2s)∂Ps∂yN(y′,θ′(σs(t))−1/2sβ(y′(σs(t))1/2s))dy′. |
We use (2.2), (3.4) and Lemma 2.3 to get
(σs(t))−1/sβ(y′(σs(t))1/2s)=O(|y′|1+β(σs(t))β−12s)−vtθ(σs(t))−1/s=O(|y′|1+β(σs(t))2s−12s)+O((σs(t))2s−12s)in BN−1r(σs(t))−1/2s. |
Then by (1.6), we have
∫BN−1r β(y′)∂κs∂yN(y′,θ′β(y′),σs(t))dy′=O((σs(t))2s−12s∫BN−1r(σs(t))−1/2s |y′|1+β∂Ps∂yN(y′,θ′(σs(t))−1/2sβ(y′(σs(t))1/2s))dy′)+O((σs(t))2s−12s∫BN−1r(σs(t))−1/2s ∂Ps∂yN(y′,θ′(σs(t))−1/2sβ(y′(σs(t))1/2s))dy′)=O((σs(t))2s−12s∫RN−1 1+|y′|1+β1+|y′|N+2s+1dy′)=O((σs(t))2s−12s). |
Hence
∫10∫BN−1r β(y′)∂κs∂yN(y′,θ′β(y′),σs(t))dy′dθ′=O((σs(t))2s−12s)as t→0. | (3.7) |
It follows from (3.5), (3.6) and (3.7) that
∫∂E∩Brκs(y′,yN−vtθ,σs(t))νN(y)dσ(y)=(σs(t))−1/2s∫RN−1 Ps(y′,0)dy′+O((σs(t))2s−12s)as t→0. | (3.8) |
By a change of variable and the fact that |τE(y)|≤1, we have
∫Bcr∂κs∂yN(y′,yN−vtθ,σs(t))τE(y)dy=O((σs(t))−1/2s∫Bcr(σs(t))−1/2s∂Ps∂yN(y′,yN−vt(σs(t))−1/2sθ)dy) =O((σs(t))−1/2s∫Bcr(σs(t))−1/2s11+|y|N+2s+1dy)=O(σs(t)). | (3.9) |
We use (1.7) to get, as t→0,
|∫∂Brκs(y′,yN−vtθ,σs(t))yNrτE(y)dσ′(y)|≤∫∂Brκs(y′,yN−vtθ,σs(t))dσ′(y) ≤σs(t)∫∂BrCN,s|(y′,yN−vtθ)|N+2sdσ′(y)=O(σs(t)). |
Therefore, the expansion (3.1) follows immediately from (3.2), (3.8), (3.9) and the above estimate. This ends the proof.
The following result completes the proof of Theorem 1.1 in the case s∈(0,1/2).
Proposition 3.2. Under the assumptions of Lemma 3.1, we have
v=aN,sHs(0)+ot(1)as t→0, | (3.10) |
where Hs(0) is the fractional mean curvature of ∂E at the point 0 and the positive constant aN,s is given by
aN,s=CN,s2∫RN−1 Ps(y′,0)dy′. |
Proof. We put with x=vteN and we recall that
u(x,σs(t))=∫RN κs(y−x,σs(t))τE(y)dy=0. |
By the fundamental theorem of calculus, we have
κs(y−x,σs(t))=κs(y,σs(t))−vt∫10∂κs∂yN(y′,yN−vtθ,σs(t))dθ. |
Then
u(x,σs(t))=˜Jr(t)+σs(t)˜Ir(t)+σs(t)CN,s∫RN∖QrτE(y)|y|N+2sdy−vt∫RN ∫10∂κs∂yN(y′,yN−vtθ,σs(t))τE(y)dθdy, | (3.11) |
where ˜Jr(t)=Jr(σs(t)), ˜Ir(t)=Ir(σs(t)), while Ir(t) and Jr(t) are given by Lemma 2.2. Moreover, by Lemma 3.1, we have
∫RN ∂κs∂yN(y′,yN−vtθ,σs(t))τE(y)dy=2(σs(t))−1/2s∫RN−1 Ps(y′,0)dy′+O((σs(t))2s−12s). |
Therefore
∫10∫RN ∂κs∂yN(y′,yN−vtθ,σs(t))τE(y)dydθ=2(σs(t))−1/2s∫RN−1 Ps(y′,0)dy′+O((σs(t))2s−12s) as t→0. | (3.12) |
Putting (3.12) in (3.11), we obtain that
u(x,σs(t))=˜Jr(t)+σs(t)˜Ir(t)+σs(t)CN,s∫RN∖QrτE(y)|y|N+2sdy−vt(2(σs(t))−1/2s∫RN−1 Ps(y′,0)dy′+O((σs(t))2s−12s))=σs(t)[(σs(t))−1˜Jr(t)+˜Ir(t)+CN,s∫RN∖QrτE(y)|y|N+2sdy −2vt(σs(t))−2s−12s∫RN−1 Ps(y′,0)dy′+O(σs(t))]. |
Recalling that σs(t)=t2s1+2s and using the fact that u(x,σs(t))=0, we have
0=t−2s1+2s˜Jr(t)+˜Ir(t)+CN,s∫RN∖QrτE(y)|y|N+2sdy−2v∫RN−1 Ps(y′,0)dy′+O(t2s1+2s) as t→0. |
As a consequence,
|CN,sHs(0)−2v∫RN−1 Ps(y′,0)dy′|≤|CN,sHs(0)−CN,s∫RN∖QrτE(y)|y|N+2sdy|+|t−2s1+2s˜Jr(t)|+˜Ir(t)+O(t2s1+2s). |
Therefore by Lemma 2.2, taking the limsups as t→0 and as r→0 respectively, we obtain
lim supt→0|CN,sHs(0)−2v∫RN−1 Ps(y′,0)dy′|=0. |
Hence
v=CN,sHs(0)2∫RN−1 Ps(y′,0)dy′+ot(1) as t→0. |
This then ends the proof.
We have the following result.
Proposition 4.1. We consider E a set of class C3 satisfying the condition in Section 2. For s∈(1/2,1), we have
v=cN,sH(0)+O(t2s−12),as t→0. | (4.1) |
where H(0) is the normalized mean curvature of ∂E at 0 and the positive constant cN,s is given by
cN,s=∫RN−1 |y′|2Ps(y′,0)dy′2∫RN−1 Ps(y′,0)dy′. |
Proof. We let x=vteN∈∂Et and we expand
u(x,σs(t))=∫RN κs(y−x,σs(t))τE(y)dy=∫Qrκs(y−x,σs(t))τE(y)dy+∫Qcrκs(y−x,σs(t))τE(y)dy, | (4.2) |
By (2.6) and Lemma 2.3, we have
∫Qcrκs(y−x,σs(t))τE(y)dy=O(σs(t)) as t→0. |
Therefore
u(x,σs(t))=∫Qrκs(y−x,σs(t))τE(y)dy+O(σs(t)). | (4.3) |
By a change of variable, the fact that τE=1E(x)−1RN∖¯E(x) and x=vteN, we have
∫Qrκs(y−x,σs(t))τE(y)dy=∫E∩Qrκs(y−x,σs(t))dy−∫Ec∩Qrκs(y−x,σs(t))dy=∫BN−1r ∫γ(y′)−rκs(y−x,σs(t))dy−∫BN−1r ∫rγ(y′)κs(y−x,σs(t))dy=∫BN−1r ∫γ(y′)−vt−r−vtκs(y,σs(t))dy−∫BN−1r ∫r−vtγ(y′)−vtκs(y,σs(t))dy=2∫BN−1r ∫γ(y′)−vt0κs(y,σs(t))dy+∫BN−1r ∫0−r−vtκs(y,σs(t))dy−∫BN−1r ∫r−vt0κs(y,σs(t))dy=2∫BN−1r ∫γ(y′)−vt0κs(y,σs(t))dy+∫BN−1r ∫−r+vt−r−vtκs(y,σs(t))dy. |
The last line is due to the fact that the map yN→κs(y,σs(t)) is even so that
∫r−vt0κs(y,σs(t))dyN=−∫−r+vt0κs(y,σs(t))dyN. |
Therefore we have
∫Qrκs(y−x,σs(t))τE(y)dy=2∫BN−1r ∫γ(y′)−vt0κs(y,σs(t))dy+∫BN−1r ∫−r+vt−r−vtκs(y,σs(t))dy. | (4.4) |
By (2.6) and the fact that vt=ot(1), we have
∫BN−1r ∫−r+vt−r−vtκs(y,σs(t))dy=O(σs(t)). | (4.5) |
By a change of variable, the Fundamental Theorem of Calculus, (1.5) and (2.3), we have
∫BN−1r ∫γ(y′)−vt0κs(y,σs(t))dy=∫BN−1r ∫γ(y′)−vt0κs(y′,0,σs(t))dy+∫BN−1r ∫γ(y′)−vt0∫10yN∂κs∂yN(y′,θyN,σs(t))dydθ=∫BN−1r κs(y′,0,σs(t))(12γyiyj(0)yiyj+O(|y′|3)−vt)dy′+∫BN−1r ∫γ(y′)−vt0∫10yN∂κs∂yN(y′,θyN,σs(t))dydθ=Δγ(0)2(N−1)∫BN−1r |y′|2κs(y′,0,σs(t))dy′−vt∫BN−1r κs(y′,0,σs(t))dy′+∫BN−1r ∫γ(y′)−vt0∫10yN∂κs∂yN(y′,θyN,σs(t))dydθ+O(∫BN−1r |y′|3κs(y′,0,σs(t))dy′). |
Therefore, recalling (2.4),
∫BN−1r ∫γ(y′)−vt0κs(y,σs(t))dy=H(0)2∫BN−1r |y′|2κs(y′,0,σs(t))dy′−vt∫BN−1r κs(y′,0,σs(t))dy′+∫BN−1r ∫γ(y′)−vt0∫10yN∂κs∂yN(y′,θyN,σs(t))dydθ+O(∫BN−1r |y′|3κs(y′,0,σs(t))dy′). | (4.6) |
By a change of variable and (1.5), we have
∫BN−1r |y′|2κs(y′,0,σs(t))dy′=(σs(t))12s∫BN−1r(σs(t))−12s |y′|2Ps(y′,0)dy′ | (4.7) |
and
∫BN−1r κs(y′,0,σs(t))dy′=(σs(t))−12s∫BN−1r(σs(t))−12s Ps(y′,0)dy′. | (4.8) |
Moreover by (1.6), we get
(σs(t))12s∫RN−1∖BN−1r(σs(t))−12s |y′|2Ps(y′,0)dy′+(σs(t))−12s∫RN−1∖BN−1r(σs(t))−12s Ps(y′,0)dy′=O(σs(t))as t→0 | (4.9) |
and
∫BN−1r |y′|3κs(y′,0,σs(t))dy′=O(σs(t)). | (4.10) |
By Lemma 2.4, we get
∫BN−1r∫γ(y′)−vt0∫10yN∂κs∂yN(y′,θyN,σs(t))dydθ=O(σs(t)). | (4.11) |
Combining (4.4), (4.6), (4.7), (4.8), (4.9) and (4.11), we obtain
∫Qrκs(y−x,σs(t))τE(y)dy=(σs(t))12sH(0)2∫RN−1 |y′|2Ps(y′,0)dy′−vt(σs(t))−12s∫RN−1 Ps(y′,0)dy′+O(σs(t)). | (4.12) |
By (4.3) and (4.12), we obtain
u(x,σs(t))=(σs(t))1/2sH(0)∫RN−1 |y′|2Ps(y′,0)dy′−2vt(σs(t))−1/2s∫RN−1 Ps(y′,0)dy′+O(σs(t))=(σs(t))−1/2s[(σs(t))1/sH(0)∫RN−1 |y′|2Ps(y′,0)dy′−2vt∫RN−1 Ps(y′,0)dy′+O((σs(t))1+2s2s)]. |
Since x=vtν∈∂Et, we have u(x,σs(t))=0. Now, from the definition of σs(t)=ts, we deduce that
H(0)∫RN−1 |y′|2Ps(y′,0)dy′−2v∫RN−1 Ps(y′,0)dy′+O(t2s−12)=0. |
Thus
v=cN,sH(0)+O(t2s−12), |
where
cN,s=∫RN−1 |y′|2Ps(y′,0)dy′2∫RN−1 Ps(y′,0)dy′. |
This then ends the proof.
As usual, we consider the function
u(x,t)=κ1/2 (⋅,t)⋆τE(x) |
and recall that
Et:={x∈RN:u(x,σ1/2 (t))≥0}. |
To alleviate the notations, for the following of this section, we write σ1/2 (t):=σ(t).
Proposition 5.1. For s=1/2, we have
v=bN(t)H(0)+O(1log(σ1/2 (t)))as t→0, | (5.1) |
where H(0) is the mean curvature of ∂E at 0.
Proof. Recall that x=vtν→0 as t→0, thanks to Lemma 2.3. We write
u(x,σ(t))=∫RN κ1/2 (y−x,σ(t))τE(y)dy=∫Qrκ1/2 (y−x,σ(t))τE(y)dy+∫Qcrκ1/2 (y−x,σ(t))τE(y)dy, | (5.2) |
where Qr=BN−1r×(−r,r). By (2.6), we have
∫Qcrκ1/2 (y−x,σ(t))τE(y)dy=O(σ(t)) as t→0. |
Then, we have
u(x,σ(t))=∫Qrκ1/2 (y−x,σ(t))τE(y)dy+O(σ(t)). | (5.3) |
By a change of variable and (2.3), we have
∫Qrκ1/2 (y−x,σ(t))τE(y)dy=∫E∩Qrκ1/2 (y−x,σ(t))dy−∫Ec∩Qrκ1/2 (y−x,σ(t))dy=∫BN−1r ∫γ(y′)−rκ1/2 (y−x,σ(t))dy−∫BN−1r ∫rγ(y′)κ1/2 (y−x,σ(t))dy=∫BN−1r ∫γ(y′)−vt−r−vtκ1/2 (y,σ(t))dy−∫BN−1r ∫r−vtγ(y′)−vtκ1/2 (y,σ(t))dy=2∫BN−1r ∫γ(y′)−vt0κ1/2 (y,σ(t))dy+∫BN−1r ∫0−r−vtκ1/2 (y,σ(t))dy−∫BN−1r ∫r−vt0κ1/2 (y,σ(t))dy=2∫BN−1r ∫γ(y′)−vt0κ1/2 (y,σ(t))dy+∫BN−1r ∫−r+vt−r−vtκ1/2 (y,σ(t))dy. |
The last line is due to the fact that the map yN→κ1/2 (y,σ(t)) is even so that
∫r−vt0κ1/2 (y,σ(t))dyN=−∫−r+vt0κ1/2 (y,σ(t))dyN. |
Therefore we have
∫Qrκ1/2 (y−x,σ(t))τE(y)dy=2∫BN−1r ∫γ(y′)−vt0κ1/2 (y,σ(t))dy+∫BN−1r ∫−r+vt−r−vtκ1/2 (y,σ(t))dy. | (5.4) |
Using (2.6), we find that
∫BN−1r ∫−r+vt−r−vtκ1/2 (y,σ(t))dy=O(σ(t)). | (5.5) |
By a change of variable, the fundamental theorem of calculus, (1.5) and (2.3), we have
∫BN−1r ∫γ(y′)−vt0κ1/2 (y,σ(t))dy=∫BN−1r ∫γ(y′)−vt0κ1/2 (y′,0,σ(t))dy+∫BN−1r ∫γ(y′)−vt0∫10yN∂κ1/2 ∂yN(y′,θyN,σ(t))dydθ=∫BN−1r κ1/2 (y′,0,σ(t))(12γyiyj(0)yiyj+O(|y′|3)−vt)dy′+∫BN−1r ∫γ(y′)−vt0∫10yN∂κ1/2 ∂yN(y′,θyN,σ(t))dydθ=Δγ(0)2(N−1)∫BN−1r |y′|2κ1/2 (y′,0,σ(t))dy′−vt∫BN−1r κ1/2 (y′,0,σ(t))dy′+∫BN−1r ∫γ(y′)−vt0∫10yN∂κ1/2 ∂yN(y′,θyN,σ(t))dydθ+O(∫BN−1r |y′|3κ1/2 (y′,0,σ(t))dy′). |
Therefore
∫BN−1r ∫γ(y′)−vt0κ1/2 (y,σ(t))dy=H(0)2∫BN−1r |y′|2κ1/2 (y′,0,σ(t))dy′ −vt∫BN−1r κ1/2 (y′,0,σ(t))dy′+∫BN−1r ∫γ(y′)−vt0∫10yN∂κ1/2 ∂yN(y′,θyN,σ(t))dydθ+O(σ(t)). | (5.6) |
By (1.5), (1.6) and a change of variable, we have
∫BN−1r∫γ(y′)−vt0yN∂κ1/2 ∂yN(y′,θyN,σ(t))dy=O(∫BN−1rσ(t)−1 ∫σ(t)−1(γ(y′σ(t))−vt)0yN(1+|(y′,θyN)|2)N+22dy)=O(∫BN−1rσ(t)−1 σ(t)−2(γ(y′σ(t))−vt)2(1+|y′|2)N+22dy′)=O(σ−2(t)∫BN−1rσ(t)−1 (σ(t)2|y′|2−vt)2(1+|y′|2)N+22dy′)=O(σ(t))+O(vt)+O(v2t2(σ(t))−2). |
Now Lemma 2.3 yields vt=O(σ(t)2) and thus
∫BN−1r∫γ(y′)−vt0∫10yN∂κ1/2 ∂yN(y′,θyN,σ(t))dydθ=O(σ(t)). | (5.7) |
We get from (5.3), (5.4), (5.5), (5.6) and (5.7) that
u(x,σ(t))=σ(t)H(0)∫BN−1rσ(t)−1 |y′|2P1/2 (y′,0)dy′−2vt∫BN−1rσ(t)−1 P1/2 (y′,0)dy′+O(σ(t)) as t→0. |
Thanks to (1.6), we have
∫RN−1∖BN−1rσ(t)−1 P1/2 (y′,0)dy′=O(σ(t)2)and∫BN−1σ(t)−1∖BN−1rσ(t)−1 |y′|2P1/2 (y′,0)dy′=O(1)as t→0. |
This implies that
u(x,σ(t))=σ(t)H(0)∫BN−1σ(t)−1 |y′|2P1/2 (y′,0)dy′−2vt(σ(t))−1∫RN−1 P1/2 (y′,0)dy′+O(σ(t)). | (5.8) |
Using polar coordinates and (1.6), we then have
∫BN−1σ(t)−1 |y′|2P1/2 (y′,0)dy′≍CN,1/2ωN−2∫1/σ(t)0mN(1+m2)N+12dm, | (5.9) |
where ωN−2:=|SN−2|. By the change of variable ρ=1m, we have
∫1/σ(t)0mN(1+m2)N+12dm=∫+∞σ(t)1ρ(1+ρ2)N+12dρ =∫1σ(t)1ρ(1+ρ2)N+12dρ+∫+∞11ρ(1+ρ2)N+12dρ =∫1σ(t)1ρ(1+ρ2)N+12dρ+O(1)=−log(σ(t))+O(1). |
Letting bN(t):=∫BN−1σ(t)−1 |y′|2P1/2 (y′,0)dy′−2log(σ(t))∫RN−1 P1/2 (y′,0)dy′, by (5.8), (5.9) and the above estimate, we obtain, as t→0,
u(x,σ(t))=σ(t)H(0)∫BN−1σ(t)−1 |y′|2P1/2 (y′,0)dy′−2vt(σ(t))−1∫RN−1 P1/2 (y′,0)dy′+O(σ(t))=(σ(t))−1[σ2(t)H(0)∫BN−1σ(t)−1 |y′|2P1/2 (y′,0)dy′−2vt∫RN−1 P1/2 (y′,0)dy′+O(σ2(t))]. |
Since x=vtν∈∂Et, we have u(x,σ(t))=0. Recalling that t=σ2(t)|log(σ(t))|, we finally get
0=σ(t)log(σ(t))[H(0)∫BN−1σ(t)−1 |y′|2P1/2 (y′,0)dy′|log(σ(t))|−2v∫RN−1 P1/2 (y′,0)dy′+O(1log(σ(t)))]. |
Hence
v=bN(t)H(0)+O(1log(σ(t))) as t→0. |
The proof is then ended.
This work is supported by the Alexander von Humboldt foundation and the German Academic Exchange Service (DAAD). Part of this work was done while the authors were visiting the International Center for Theoretical Physics (ICTP) in December 2019 within the Simons associateship program.
The authors declare no conflict of interest.
[1] | F. M. Arscott, A. F. Filippov, Differential Equations with Discontinuous Righthand Sides: Control Systems, Netherlands: Springer, 2013. |
[2] | M. Bernardo, C. Budd, A. R. Champneys, et al. Piecewise-Smooth Dynamical Systems: Theory and Applications, London: Springer Science & Business Media, 2008. |
[3] | L. C. Evans, Partial Differential Equations, American Mathematical Society, 2010. |
[4] |
M. Fečkan, A Galerkin-averaging method for weakly nonlinear equations, Nonlinear Anal., 41 (2000), 345-369. doi: 10.1016/S0362-546X(98)00281-8
![]() |
[5] |
S. Ma, Y. Kang, Periodic averaging method for impulsive stochastic differential equations with Lévy noise, Appl. Math. Lett., 93 (2019), 91-97. doi: 10.1016/j.aml.2019.01.040
![]() |
[6] |
J. G. Mesquita, A. Slavík, Periodic averaging theorems for various types of equations, J. Math. Anal. Appl., 387 (2012), 862-877. doi: 10.1016/j.jmaa.2011.09.038
![]() |
[7] | N. A. Perestyuk, V. A. Plotnikov, A. M. Samoilenko, et al. Differential Equations with Impulse Effects, Multivalued Right-Hand Sides with Discontinuities, Berlin: Walter de Gruyter GmbH & Co. KG, 2011. |
[8] | J. A. Sanders, F. Verhulst, J. Murdock, Averaging Methods in Nonlinear Dynamical Systems, 2 Eds, New York: Springer, 2007. |
1. | Dario Mazzoleni, Benedetta Pellacci, Calculus of variations and nonlinear analysis: advances and applications, 2023, 5, 2640-3501, 1, 10.3934/mine.2023059 |