Research article Special Issues

On an asymptotically log-periodic solution to the graphical curve shortening flow equation

  • With the help of heat equation, we first construct an example of a graphical solution to the curve shortening flow. This solution y(x,t) has the interesting property that it converges to a log-periodic function of the form

    Asin(logt)+Bcos(logt)

    as t, where A, B are constants. Moreover, for any two numbers α<β, we are also able to construct a solution satisfying the oscillation limits

    lim infty(x,t)=α,   lim supty(x,t)=β,   xK

    on any compact subset KR.

    Citation: Dong-Ho Tsai, Xiao-Liu Wang. On an asymptotically log-periodic solution to the graphical curve shortening flow equation[J]. Mathematics in Engineering, 2022, 4(3): 1-14. doi: 10.3934/mine.2022019

    Related Papers:

    [1] Takeyuki Nagasawa, Kohei Nakamura . Asymptotic analysis for non-local curvature flows for plane curves with a general rotation number. Mathematics in Engineering, 2021, 3(6): 1-26. doi: 10.3934/mine.2021047
    [2] M. M. Bhatti, Efstathios E. Michaelides . Oldroyd 6-constant Electro-magneto-hydrodynamic fluid flow through parallel micro-plates with heat transfer using Darcy-Brinkman-Forchheimer model: A parametric investigation. Mathematics in Engineering, 2023, 5(3): 1-19. doi: 10.3934/mine.2023051
    [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] Franco Flandoli, Eliseo Luongo . Heat diffusion in a channel under white noise modeling of turbulence. Mathematics in Engineering, 2022, 4(4): 1-21. doi: 10.3934/mine.2022034
    [5] Yuji Hamada, Takeshi Sato, Takashi Taniguchi . Multiscale simulation of a polymer melt flow between two coaxial cylinders under nonisothermal conditions. Mathematics in Engineering, 2021, 3(6): 1-22. doi: 10.3934/mine.2021042
    [6] Glen Wheeler . Convergence for global curve diffusion flows. Mathematics in Engineering, 2022, 4(1): 1-13. doi: 10.3934/mine.2022001
    [7] Massimiliano Giona, Luigi Pucci . Hyperbolic heat/mass transport and stochastic modelling - Three simple problems. Mathematics in Engineering, 2019, 1(2): 224-251. doi: 10.3934/mine.2019.2.224
    [8] Gennaro Ciampa, Gianluca Crippa, Stefano Spirito . Propagation of logarithmic regularity and inviscid limit for the 2D Euler equations. Mathematics in Engineering, 2024, 6(4): 494-509. doi: 10.3934/mine.2024020
    [9] Luca Franzoi, Riccardo Montalto . Time almost-periodic solutions of the incompressible Euler equations. Mathematics in Engineering, 2024, 6(3): 394-406. doi: 10.3934/mine.2024016
    [10] Bruno Bianchini, Giulio Colombo, Marco Magliaro, Luciano Mari, Patrizia Pucci, Marco Rigoli . Recent rigidity results for graphs with prescribed mean curvature. Mathematics in Engineering, 2021, 3(5): 1-48. doi: 10.3934/mine.2021039
  • With the help of heat equation, we first construct an example of a graphical solution to the curve shortening flow. This solution y(x,t) has the interesting property that it converges to a log-periodic function of the form

    Asin(logt)+Bcos(logt)

    as t, where A, B are constants. Moreover, for any two numbers α<β, we are also able to construct a solution satisfying the oscillation limits

    lim infty(x,t)=α,   lim supty(x,t)=β,   xK

    on any compact subset KR.



    The goal of this paper is to make a comparison between the geometric heat equation (graphical curve shortening flow equation) and the Euclidean heat equation (the usual heat equation on R). Based on a nice result due to Nara-Taniguchi [9] (see Theorem 1.1), which says that under some suitable assumption on the initial condition u0(x), xR, the solutions of both equations give rise to the same asymptotic behavior, we can construct a solution y(x,t) of the graphical curve shortening flow equation so that it is asymptotically log-periodic (i.e., functions which are periodic in logt) as t. Moreover, for any two numbers α<β, we can construct another solution y(x,t) prescribing the oscillation behavior lim infty(x,t)=α, lim supty(x,t)=β on any compact subset xKR (see Theorem 2.1, Theorem 2.7, Corollary 2.9).

    The curve shortening flow (CSF) is a geometric heat equation which can be used to deform curves in the plane. This flow arises naturally in phase transitions and plays an important role in the thermomechanics of evolving phase boundaries in the plane; see [8]. Mathematically, CSF is also an interesting topic on its own with fundamental importance. Since 1986, it has been studied thoroughly by Gage-Hamilton [7], Grayson [6], Angenent [1] and many others. See the bibliography in the book by Chou-Zhu [3] for literature.

    A family of smooth embedded curves γ(φ,t):I×[0,T)R2 (IR is some interval) is said to evolve under CSF if it satisfies the equation

    γt(φ,t)=κ(φ,t)N(φ,t),   (φ,t)I×[0,T),   (CSF), (1.1)

    where κ(φ,t) is the curvature of γ(φ,t) and N(φ,t) is its unit normal vector. We use the convention that for a parametrized curve γR2 its Frenet frame {T,N}, T=dγ/ds, has positive orientation and its curvature κ is defined according to the identity κ=dT/ds,N, where s is the arc length parameter of γ. By the identity dT/ds=κN, one can write (1.1) as

    γt=κN=Ts=2γs2. (1.2)

    Since Eq (1.2) resembles the familiar classical heat equation ut(x,t)=uxx(x,t) and the second derivative operator 2/s2 involves the geometry of the evolving curve γ(,t), we call it a "geometric heat equation". As a contrast, the classical heat equation ut=uxx is called an "Euclidean heat equation".

    Denote the tangent angle of the curve γ(φ,t) as θ(φ,t). Then we have

    T(φ,t)=(cosθ(φ,t),sinθ(φ,t)),   N(φ,t)=(sinθ(φ,t),cosθ(φ,t)) (1.3)

    and Eq (1.1) has the form (in the following, "angle" means "tangent angle")

    γt=κN=θsN=(angle)sN,   κ=θs,   (CSF). (1.4)

    It is also known that if γ(φ,t)R2 is a family of graphical curves* evolving under CSF and is represented by the graphs of some function y(x,t) over some interval J of the x-axis, then on its domain (in general the interval J on which y(x,t) is defined may also depend on time) the function y(x,t) will satisfy the equation

    yt(x,t)=yxx(x,t)1+y2x(x,t)=x(tan1(yx(x,t)))=(angle)x,   xJ,   t>0,   (CSF), (1.5)

    *Our convention on the orientation of a graphical curve γ is that its tangent vector is pointing in the positive direction of x-axis.

    where y2x(x,t) means (yx(x,t))2. Note that the right hand side of (1.5) is the derivative of the tangent angle θ(x,t) of the graph (we have tanθ(x,t)=yx(x,t)). In terms of the curvature κ=yxx(1+y2x)3/2 for a graph, (1.5) can also be written as

    yt(x,t)=1+y2x(x,t)κ(x,t),   (CSF), (1.6)

    where κ(x,t) is the curvature of the graph at the point (x,y(x,t)). By (1.5), if y(x,t) is a solution of (1.5), so are y(x,t), y(x,t) and λy(λ1x,λ2t) (for any constant λ>0).

    On the other hand, for the Euclidean heat equation ut=uxx, xJ, we can interpret it geometrically as

    ut(x,t)=uxx(x,t)=x(ux(x,t))=tanθx=(slope)x,   xJ,   t>0 (1.7)

    and similar to (1.6), we have

    ut(x,t)=uxx(x,t)=(1+u2x(x,t))3κ=1+u2x(x,t)(sec2θ)κ. (1.8)

    Similar to the equivalent relation between the flow Eq (1.1) and the graph Eq (1.6), one can verify that if γ(φ,t)R2 is a family of graphical curves (represented by the graph of some function u(x,t) over some interval J of the x-axis) evolving under the anisotropic curve shortening flow (ACSF) of the form

    γt(φ,t)=(sec2θ(φ,t))κ(φ,t)N(φ,t),   (φ,t)I×[0,T),   (ACSF), (1.9)

    where θ(φ,t) is the tangent angle of γ(φ,t), then the graph function u(x,t),  xJ, will satisfy the equation (1.7). Moreover, the converse is also true.

    Note that the flow (1.9) is defined only for graphical curves with angle θ(π/2,π/2). On the other hand, the classical curve shortening flow (1.1) can be defined for rather general curves, say simple closed curves in the plane. However, here we are confining the curve shortening flow to graphs and Eq (1.5) is not defined at points where the gradient blows up or equivalently the angle is vertical. Similar to (1.4), one can rewrite (1.9) as

    γt=(tanθ)sN=(slope)sN,   (ACSF). (1.10)

    A major important connection between the two Eqs (1.5) and (1.7) (or between the two flows (1.1) and (1.9)) is the following equivalence result due to Nara-Taniguchi [9]:

    Theorem 1.1. ([9]) Let γ(0,1) be a constant. Assume that y0(x), xR, is a bounded function lying in the space C2+γ(R). Then there exists a classical solution y(x,t) to Eq (1.5) (with y(x,0)=y0(x), xR) on the domain R×[0,). Moreover the solution y(x,t) is smooth on R×(0,), continuous on R×[0,), and satisfies

    supxR|y(x,t)RΓ(xξ,t)y0(ξ)dξ|Ct,   t>0 (1.11)

    for some constant C>0 depending only on y0. Here Γ(ξ,t) is the heat kernel given by Γ(ξ,t)=(1/4πt)exp(ξ2/(4t)).

    In the following we give an interesting example of bounded y0(x), lying in the space C2+γ(R), which gives explicit y(x,t). Therefore, we can check property (1.11) directly.

    Example 1.2. Let

    y0(x)=sinh1(sinx)=log(sinx+sin2x+1),   xR, (1.12)

    which is an odd 2π-periodic smooth function on R. Since there exists a constant M>0 such that

    |y0(x), y0(x), y0(x), y0(x)|M,    xR, (1.13)

    we have y0(x)C2+γ(R) for any γ(0,1). Under the evolution of Eq (1.5), one can verify that the solution y(x,t) is given explicitly by

    y(x,t)=sinh1(etsinx)=log[(etsinx)+(etsinx)2+1],   xR,   t[0,), (1.14)

    which decays to 0 uniformly on xR with exponential rate et as t. On the other hand, since y0(x) is an odd 2π-periodic smooth function on R, Fourier series theory also implies that the convolution solution RΓ(xξ,t)y0(ξ)dξ converges uniformly to the constant 12πππy0(x)dx=0 uniformly on xR with exponential rate et as t. Hence (1.11) is verified.

    Remark 1.3. The solution (1.14) is known as the hairclip solution of the curve shortening flow. It is a graphical entire solution defined on xR. It has been derived in Broadbridge-Vassiliou [2] and Doyle-Vassiliou [5] using symmetry and separation method. Also see Tsai-Wang [10] for a simple ODE method to derive it.

    With the help of Theorem 1.1, one can find some interesting solution y(x,t) of the graphical curve shortening flow. The upshot is that, as t, the solution y(x,t) approaches a non-constant bounded function Y(t) which satisfies Y(λt)=Y(t) for some constant λ>1 for all t(0,). The function Y(t) will be periodic in logt. 

    The theorem below is motivated by the initial condition (1.12) in the hairclip solution in Example 1.2. Roughly speaking, we switch the role of sine function and logarithmic function, and in order for logx to be defined on (,), we change it as log(x2+1). More precisely, we have:

    Theorem 2.1. Let y0(x) be given by

    y0(x)=sin(log(x2+1)),   x(,) (2.1)

    and consider Eq (1.5) with the above initial condition. Then the solution y(x,t) of this Cauchy problem is defined on R×[0,), which is smooth on R×(0,) and continuous up to t=0. Moreover, y(x,t) satisfies the following two properties:

    (1). (Asymptotic behavior as t.) There exist two positive constants A, B such that

    limtsupxK|y(x,t)(Asin(logt)+Bcos(logt))|=0 (2.2)

    for any compact set KR.

    (2). (Asymptotic behavior as |x|.) For fixed t>0 we have

    lim sup|x||y(x,t)sin(log(x2+1))|Ct (2.3)

    for some constant C>0 depending only on y0.

    Remark 2.2. The limit function Y(t):=Asin(logt)+Bcos(logt), t(0,), is not periodic in time t(0,). Instead, it satisfies Y(e2πt)=Y(t) for all t(0,). One can see that a function Y(t) defined on (0,) satisfying the identity Y(e2πt)=Y(t) for all t(0,) if and only if the function F(s)=Y(es) is defined on (,) and satisfies the identity F(s+2π)=F(s) for all s(,). Since we have Y(t)=F(logt), t(0,), we can say that Y(t) is 2π-periodic in logt, or for simplicity,  just log-periodic.

    Proof. (1). We first note that y0(x)C2+γ(R) for any γ(0,1) since there exists a constant M>0 such that (1.13) holds for y0(x). Hence Theorem 1.1 is applicable and to look at the asymptotic behavior of y(x,t), it suffices to look at that of y0(x) under the heat equation ut=uxx. We have for t>0 that

    u(0,t)=14πtex24ty0(x)dx=14πez24y0(tz)dz=14πez24sin[logt+log(z2+1t)]dz=A(t)sin(logt)+B(t)cos(logt), (2.4)

    where

    {A(t)=14πez24cos[log(z2+1t)]dz,B(t)=14πez24sin[log(z2+1t)]dz, (2.5)

    and we note that

    {limtA(t)=14πez24cos(log(z2))dz=A0.26682limtB(t)=14πez24sin(log(z2))dz=B0.12278. (2.6)

    Thus we conclude

    limt|u(0,t)(Asin(logt)+Bcos(logt))|=0, (2.7)

    where the constants A, B are from (2.6).

    Since |y0(x)|1 for all xR, we have the following gradient estimate for the convolution solution of the heat equation with initial data y0(x): 

    |ux(x,t)|1πt,    (x,t)R×(0,), (2.8)

    which, for each fixed x, implies

    limt|u(x,t)(Asin(logt)+Bcos(logt))|limt{|u(x,t)u(0,t)|+|u(0,t)(Asin(logt)+Bcos(logt))|}=0. (2.9)

    Thus for each fixed x, u(x,t) also approaches to the same function as t. Moreover, the above convergence is uniform in xK for any compact set KR. This fact, together with Theorem 1.1, gives the proof of (2.2).

    (2). By (2.10) below and letting x=ty, we have

    u(x,t)=u(ty,t)=A(y,t)sin(logt)+B(y,t)cos(logt), (2.10)

    where now

    {A(y,t)=14πez24cos(log((z+y)2+1t))dzB(y,t)=14πez24sin(log((z+y)2+1t))dz,   y=xt.

    For fixed z(,) and fixed t>0, we have

    {lim|y||ez24cos(log((z+y)2+1t))ez24cos(log(y2+1t))|=0lim|y||ez24sin(log((z+y)2+1t))ez24sin(log(y2+1t))|=0

    and the Lebesgue Dominated Convergence Theorem implies, for fixed t>0, the limits

    {lim|y||A(y,t)14πez24cos(log(y2+1t))dz|=0lim|y||B(y,t)14πez24sin(log(y2+1t))dz|=0.

    Since (4π)1ez2/4dz=1, we conclude

    {lim|y||A(y,t)cos(log(y2+1t))|=0lim|y||B(y,t)sin(log(y2+1t))|=0

    and so

    lim|y||u(ty,t)[cos(log(y2+1t))sin(logt)+sin(log(y2+1t))cos(logt)]|=lim|y||u(ty,t)sin(log(ty2+1))|=lim|x||u(x,t)sin(log(x2+1))|=0, (2.11)

    which, together with (1.11), implies

    lim sup|x||y(x,t)sin(log(x2+1))|lim sup|x|{|y(x,t)u(x,t)|+|u(x,t)sin(log(x2+1))|}Ct.

    Hence (2.3) follows.

    Remark 2.3. Note that the convergence in (2.2) cannot be uniform in the whole space x(,). If we choose x=t, t>0, in the identity

    u(x,t)=14πez24y0(x+tz)dz=14πez24sin[logt+log(x2t+2xzt+z2+1t)]dz, (2.12)

    then the function u(t,t) will converge to ˜Asin(logt)+˜Bcos(logt) as t, where ˜A, ˜B are constants different from the A, B in (2.6), given by

    {˜A=14πez24cos(log(z+1)2)dz0.2003˜B=14πez24sin(log(z+1)2)dz0.24081.

    The following says that it is impossible to find an initial data y0(x)C2+γ(R) for some γ(0,1) so that we have convergence to a periodic function.

    Lemma 2.4. There does not exist y0(x)C2+γ(R) for some γ(0,1) such that under the graphical curve shortening flow Eq (1.5) we have

    limt|y(0,t)P(t)|=0, (2.13)

    where P(t) is a non-constant periodic function.

    Proof. Assume (2.13) holds for some P(t). Then by Theorem 1.1 we will have

    limt|14πez24y0(tz)dzP(t)|=0, (2.14)

    which will give a contradiction due to

    ddt(14πez24y0(tz)dz)=14πez24y0(tz)z2tdz=O(1t)0   as   t. (2.15)

    Remark 2.5. In view of Lemma 2.4, it seems reasonable to see that for y0(x)C2+γ(R), instead of converging to Asint+Bcost, we get convergence to Asin(logt)+Bcos(logt), which is the result of Theorem 2.1. The proof of Lemma 2.4 also says that for heat equation with initial data in C2+γ(R), the solution u(x,t), for fixed x, cannot converge to a periodic function as t. On the other hand, it is possible if we allow the initial data to be unbounded. Let

    u0(x)=λ+Aex2cos(x2)+Bex2sin(x2),   xR,

    where λ, A, B are arbitrary constants. Then the function

    u(x,t)=λ+Aex2cos(t+x2)+Bex2sin(t+x2),   (x,t)R2,

    is a solution of the heat equation with u(x,0)=u0(x), xR. For each fixed x, u(x,t) is a non-constant 2π-periodic function in time. However, as seen in Section 3, for unbounded initial data, solutions to equation (1.5) and solutions to the heat equation (1.7) may not have the same asymptotic behavior as t.

    Remark 2.6. This is for comparison. Let M>0 be a constant and y0(x)C2+γ(R) be such that for |x|M it is equal to sin(|x|α) for some constant α(0,1] (if α>1, the function will not lie in the space C2+γ(R)). Then as |x|, sin(|x|α) oscillates faster than sin(log(x2+1)). Evolving y0(x) under the heat equation, we have

    u(0,t)=14πt|x|<Mex24ty0(x)dx+14π|z|M/tez24sin((t)α|z|α)dz=14πt|x|<Mex24ty0(x)dx+1απ(M/t)αθ1α1eθ2/α4sin((t)αθ)dθ,

    where, similar to the Riemann-Lebesgue lemma, we can prove that limtu(0,t)=0. Hence by Theorem 1.1, under Eq (1.5), we have limty(x,t)=0 uniformly on compact set KR if we choose the above initial data. Thus the moral is: a bounded slow-oscillation function, under either the heat Eq (1.7) or (1.5), has more chance to preserve its profile as t. In the extreme case that when y0(x)c is a constant (which has no oscillation at all), we have y(x,t)c and the profile is unchanged at all. See Remark 2.8 also.

    Finally, we note that if we take the initial data of the heat equation as

    psin(λlog(x2+1))+qcos(λlog(x2+1)),   x(,), (2.16)

    where p, q, λ>0 are constants, then in the limit we have

    limt|u(x,t)[(pAλqBλ)sin(λlogt)+(pBλ+qAλ)cos(λlogt)]|=0,   x(,), (2.17)

    where

    Aλ=14πez24cos(λlog(z2))dz,   Bλ=14πez24sin(λlog(z2))dz (2.18)

    with

    limλ0+Aλ=1,     limλ0+Bλ=0,     limλAλ=limλBλ=0. (2.19)

    Thus in the limit λ, the solution u(x,t) will converge to zero since the two functions in (2.16) are no longer of slow-oscillation. This matches with the moral stated in Remark 2.6. By Theorem 1.1, the solution y(x,t) of Eq (1.5) with initial data (2.16) also satisfies (2.17), uniformly on any compact set KR of x.

    Another result similar to Theorem 2.1 is the following:

    Theorem 2.7. Let r0(x) be given by

    r0(x)=sin(log(log(x2+2))),   x(,) (2.20)

    and consider Eq (1.5) with the above initial condition. Then the solution y(x,t) of this Cauchy problem is defined on R×[0,), which is smooth on R×(0,) and continuous up to t=0. Moreover, y(x,t) satisfies the following two properties:

    (1). (Asymptotic behavior as t.) On any compact set KR we have

    limtsupxK|y(x,t)sin(log(logt))|=0. (2.21)

    (2). (Asymptotic behavior as |x|.) For fixed t>0 we have

    lim sup|x||y(x,t)sin(log(log(x2+2)))|Ct, (2.22)

    for some constant C>0 depending only on r0.

    Proof. (1). We have r0(x)C2+γ(R) for any γ(0,1); hence Theorem 1.1 is applicable. The solution u(x,t) of the heat equation with initial data r0(x) satisfies

    u(0,t)=14πez24sin{log[logt+log(z2+2t)]}dz,   t(0,).

    For convenience, let β be the quantity β=log[1+1logtlog(z2+2t)]. Then the above becomes

    u(0,t)=14πez24sin[log(logt)+β]dz=A(t)sin(log(logt))+B(t)cos(log(logt)),

    where

    {A(t)=14πez24cosβdz,   limtA(t)=1B(t)=14πez24sinβdz,   limtB(t)=0. (2.23)

    Thus we have limt|u(0,t)sin(log(logt))|=0 and (2.2) follows from the gradient estimate (2.8).

    (2). Similar to (2.10), if we let x=ty, we have 

    u(x,t)=u(ty,t)=A(y,t)sin(log(logt))+B(y,t)cos(log(logt)),

    where now

    A(y,t)=14πez24cosσdz,   B(y,t)=14πez24sinσdz,

    with

    σ=log[1+1logtlog((y+z)2+2t)],   y=xt. 

    For fixed z(,) and fixed t>0, by the limit

    lim|y|[log((y+z)2+2t)log(y2+2t)]=0,

    we have

    lim|y||cosσcosρ|=0,   lim|y||sinσsinρ|=0,

    where

    ρ=log[1+1logtlog(y2+2t)],   y=xt. 

    The Lebesgue Dominated Convergence Theorem implies, for fixed t>0, the limits

    lim|y||A(y,t)cosρ|=0,   lim|y||B(y,t)sinρ|=0

    and so

    lim|y||u(ty,t)sin[log(logt)+ρ]|=lim|y||u(ty,t)sin[log(log(ty2+2))]|=0. (2.24)

    Hence we have

    lim|x||u(x,t)sin(log(log(x2+2)))|=0,

    which, together with (1.11), implies the following estimate for fixed t>0: 

    lim sup|x||y(x,t)sin(log(log(x2+2)))|lim sup|x|{|y(x,t)u(x,t)|+|u(x,t)sin(log(log(x2+2)))|}Ct.

    Hence (2.3) follows. The proof is done.

    Remark 2.8. The function sin(log(log(x2+2))) oscillates even more slowly than sin(log(x2+1)) as |x|. Under the heat equation, its profile is totally unchanged as t. On the other hand, as t the profile of sin(log(x2+1)) is slightly changed into a linear combination of sin(logt) and cos(logt). Similarly, if r0(x) is given by r0(x)=cos(log(log(x2+2))), we have

    limtsupxK|y(x,t)cos(log(logt))|=0 (2.25)

    for any compact set KR. By analogy, if we take the initial condition in the space C2+γ(R) as

    sin{log[log(log(x2+m))]}   (k copies of log, k3),

    where m>0 is some constant so that the function is defined on x(,), then the limit of u(0,t), as t, is sin{log[log(logt)]}. The same result holds if we replace the sine function by cosine function.

    As a consequence of Theorem 1.1 and Theorem 2.7, we can prescribe the oscillation limits of y(x,t) as t.

    Corollary 2.9. For any two numbers α<β, if we choose the initial data as

    r0(x)=βα2sin[log(log(x2+2))]+β+α2C2+γ(R),   x(,)  (2.26)

    and consider Eq (1.5) with the above initial condition, then on any compact set KR we have

    limtsupxK|y(x,t)(βα2sin(log(logt))+β+α2)|=0. (2.27)

    In particular, for any fixed x0(,), we have

    α=lim infxr0(x)=lim infty(x0,t)<lim supty(x0,t)=lim supxr0(x)=β. (2.28)

    Thus, in the limit, the oscillation of y(x0,t) can attain any two arbitrary numbers α<β. 

    Proof. This is because the solution of the heat equation with initial data (2.26) satisfies (2.27). The result follows due to Theorem 1.1.

    Theorem 1.1 fails if the initial condition y0(x) is not in the space C2+γ(R). This is not discussed in the paper [9]. In the following we give one example of unbounded y0(x), x(,) to demonstrate this. This example is related toa travelling wave solution for the heat equation (1.7) along the y-direction.

    We choose the initial data to be unbounded, with y0(x)=x2/2, x(,). It is not in the space C2+γ(R) for any γ(0,1). Under the heat equation the solution is given by u(x,t)=t+x2/2, which is a travelling wave solution translating in the positive y-direction. On the other hand, the curve y0(x)=x2/2 divides the plane into two regions, each with infinite area. By the main theorem in p. 472 of [4], if we evolve y0(x)=x2/2 under equation (1.5), the solution y(x,t) is defined for all time t>0, i.e., defined on R×(0,), which is smooth in R×(0,) and continuous on R×[0,). 

    It seems quite obvious that the initial curve y0(x)=x2/2, under the evolution of (1.7) and (1.5) respectively, will reveal different behavior as t. Nonetheless, a simple analysis on the angle function θ(x,t) is useful in general and can provide us a rigorous proof of the above claim. Moreover, as soon as the evolution of θ(x,t) is known, the evolution of the curvature κ(x,t) will come out immediately. See equations (3.7) and (3.8) below.

    In the following we evolve y0(x)=x2/2 under Eq (1.5). The angle function θ(x,t)=tan1(yx(x,t))(π/2,π/2) of the evolving curve y(x,t) satisfies the equation

    tθ(x,t)=t(tan1(yx(x,t)))=tyx(x,t)1+y2x(x,t)=(cos2θ(x,t))θxx(x,t), (3.1)

    where we have used the identity yt(x,t)=θx(x,t) from (1.5) in (3.1). As a consequence, letting v(x,t)=θx(x,t), we have

    tv(x,t)=(tθ)x(x,t)=(cos2θ(x,t))vxx(x,t)(sin2θ(x,t))v(x,t)vx(x,t) (3.2)

    and for w(x,t)=θxx(x,t)=vx(x,t) we have the equation

    tw(x,t)={(cos2θ(x,t))wxx(x,t)2(sin2θ(x,t))v(x,t)wx(x,t)(sin2θ(x,t))w2(x,t)2(cos2θ(x,t))v2(x,t)w(x,t). (3.3)

    For the initial data y(x,0)=x2/2, we have  and θ(x,0)=tan1yx(x,0)=tan1x and so v(x,0)=θx(x,0)=1/(1+x2)>0 on R. We also have

    w(x,0)=θxx(x,0)=2x(1+x2)2={<0,   x(0,),=0,   x=0,>0,   x(,0). (3.4)

    In particular, we see that v(x,0) is an even positive function in x(,) and w(x,0) is an odd function in x(,) with a simple zero at x=0. By the symmetry of y0(x)=x2/2 and its curvature κ0(x) (both are even functions) and the geometry of equation (1.6), θ(x,t) and w(x,t) must remain odd functions in x(,) as long as the solution exists. Also v(x,t) must remain an even function in x(,).

    Note that the initial data w(x,0), over x(,), has only one simple zero at x=0. Since we know that the number of zeros (counted with multiplicity) for solutions to Eq (3.3) cannot increase with time (see Angenent [1], p. 607), we must have

    w(x,t)=θxx(x,t)={<0,   x(0,),   t(0,), =0,   x=0,   t(0,),>0,   x(,0),   t(0,), (3.5)

    which implies

    tθ(x,t)=(cos2θ(x,t))w(x,t)={<0,   x(0,),   t(0,),=0,   x=0,   t(0,),>0,   x(,0),   t(0,). (3.6)

    The above says that, if we evolve y0(x)=x2/2 under (1.5), the angle function θ(x,t) (with θ(x,0)=tan1x) is decreasing in time for x(0,) and increasing in time for x(,0). Thus the asymptotic behavior for y0(x)=x2/2, evolving by (1.7) and (1.5) respectively, will be different as time goes on.

    To end this section, we point out that, in terms of the angle function θ(x,t) and its evolution equation (3.1), one can compute the evolution equation of the curvature κ as follows:

    Lemma 3.1. (The evolution equation of the curvature for CSF and ACSF.) Assume y(x,t)C(R×(0,Tmax))C0(R×[0,Tmax)) is a solution to Eq (1.5) on the domain R×[0,Tmax). Then the curvature κ(x,t) of the graph y(,t) satisfies the equation

    tκ=(cos2θ)κxx+κ3,   κ=κ(x,t),   (x,t)R×(0,Tmax). (3.7)

    Similarly, if u(x,t)C(R×(0,Tmax))C0(R×[0,Tmax)) satisfies the heat Eq (1.7) on R×[0,Tmax), then its curvature κ(x,t) satisfies

    tκ=κxx+6(tanθ)θxκx+(6cos4θ3cos2θ)κ3,   κ=κ(x,t),   (x,t)R×(0,Tmax). (3.8)

    Remark 3.2. Since Eq (1.5) is a geometric equation, we see that Eq (3.7) looks better than Eq (3.8).

    Proof. For (3.7), by (1.6) we can express the curvature as

    κ(x,t)=θx(x,t)1+y2x(x,t)=θx1+tan2θ=(cosθ)θx=(sinθ)x, (3.9)

    and get

    {κx=(cosθ)θxx(sinθ)θ2xκxx=(cosθ)θxxx3(sinθ)θxθxx(cosθ)θ3x. (3.10)

    By the second identity in (3.10) and (3.1), we conclude  

    tκ=(sinθ)θtθx+(cosθ)(θt)x=(sinθ)(cos2θ)θxxθx+(cosθ)[(cos2θ)θxxx(sin2θ)θxθxx]=(sinθ)(cos2θ)θxθxx+(cos2θ)[κxx+3(sinθ)θxθxx+(cosθ)θ3x](cosθ)(sin2θ)θxθxx=(cos2θ)κxx+(cos3θ)θ3x=(cos2θ)κxx+κ3, (3.11)

    which gives (3.7).

    For (3.8), we first have ut=uxx=(tanθ)x and κ=(cosθ)θx. Similar to the derivation of (3.11), we need to compute tθ and tθx first. We have

    tθ=t(tan1ux)=(tu)x1+u2x=(cos2θ)(tanθ)xx=(cos2θ)((sec2θ)θx)x=θxx+2(tanθ)θ2x (3.12)

    and

    tθx=(tθ)x=θxxx+2(sec2θ)θ3x+4(tanθ)θxθxx (3.13)

    and similar to (3.11), together with (3.10), we get

    tκ=(sinθ)[θxx+2(tanθ)(θx)2]θx+(cosθ)[θxxx+2(sec2θ)θ3x+4(tanθ)θxθxx]=(sinθ)θx[θxx+2(tanθ)(θx)2]+(cosθ)(κxxcosθ+3sinθcosθθxθxx+θ3x+2(sec2θ)θ3x+4(tanθ)θxθxx)=κxx+6(sinθ)θxθxx+(2sin2θcosθ+cosθ+2secθ)θ3x=κxx+6(sinθ)θx(κxcosθ+sinθcosθθ2x)+(2sin2θcosθ+cosθ+2secθ)θ3x=κxx+6(tanθ)θxκx+(6cos4θ3cos2θ)κ3,

    which gives (3.8).

    We thank the referee for his careful reading of our paper and useful comments and suggestions. In particular, for his correction of some mistake in the original version of the article. We also thank Professor James McCoy of the University of the Newcastle, Australia, for his kind invitation of submitting the article. Research of the first author is supported by MoST of Taiwan under grant number 108-2115-M-007-013-MY2. Research of the second author is supported by NSF of China 11871148.

    The authors declare no conflict of interest.



    [1] S. Angenent, On the formation of singularities in the curve shortening flow, J. Diff. Geom., 33 (1991), 601–633.
    [2] P. Broadbridge, P. J. Vassiliouz, The role of symmetry and separation in surface evolution and curve shortening, SIGMA, 7 (2011), 1–19.
    [3] K. S. Chou, X. P. Zhu, The curve shortening problem, Chapman: Hall/CRC, 2001.
    [4] K. S. Chou, X. P. Zhu, Shortening complete plane curves, J. Diff. Geom., 50 (1998), 471–504.
    [5] P. W. Doyle, P. J. Vassiliou, Separation of variables for the 1-dimensional non-linear diffusion equation, Int. J. Non-Linear Mech., 33 (1998), 315–326. doi: 10.1016/S0020-7462(97)00013-9
    [6] M. A. Grayson, The heat equation shrinks embedded plane curves to round points, J. Diff. Geom., 26 (1987), 285–314.
    [7] M. E. Gage, R. Hamilton, The heat equation shrinking convex plane curves, J. Diff. Geom., 23 (1986), 69–96.
    [8] M. E. Gurtin, Thermomechanics of evolving phase boundaries in the plane, New York: The Clarendon Press, 1993.
    [9] M. Nara, M. Taniguchi, The condition on the stability of stationary lines in a curvature flow in the whole plane, J. Diff. Eq., 237 (2007), 61–76. doi: 10.1016/j.jde.2007.02.012
    [10] D. H. Tsai, X. L. Wang, On some simple methods to derive the hairclip and paperclip solutions of the curve shortening flow, Acta Math. Sci., 39 (2019), 1674–1694. doi: 10.1007/s10473-019-0616-5
  • Reader Comments
  • © 2022 the Author(s), licensee AIMS Press. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0)
通讯作者: 陈斌, bchen63@163.com
  • 1. 

    沈阳化工大学材料科学与工程学院 沈阳 110142

  1. 本站搜索
  2. 百度学术搜索
  3. 万方数据库搜索
  4. CNKI搜索

Metrics

Article views(2346) PDF downloads(115) Cited by(0)

Other Articles By Authors

/

DownLoad:  Full-Size Img  PowerPoint
Return
Return

Catalog