betaFeedback

WorksMathematics

An elementary observation on the Robin problem for NLS

Abstract

We present an infinite set of explicit solutions of the integrable focusing NLS equation on the half-line that are rapidly decaying and satisfy a homogeneous Robin condition but have very different behaviors on the wall x=0x=0. We then argue that a similar phenomenon exists for a small non-homogeneous perturbation of the Robin problem. This renders a potential application of the unified transform method to non-linearizable Robin problems impossible.

NLS and the NN-soliton

We consider the focusing NLS equation, with given initial data and a simple inhomogeneous Robin condition:

iut+uxx+2∣u∣2u=0(x>0, t>0),ux(0,t)+qu(0,t)=h(t),u(x,0)=u0(x),(1)iu_t+u_{xx}+2|u|^2u=0\quad(x>0,\ t>0),\qquad u_x(0,t)+q u(0,t)=h(t),\qquad u(x,0)=u_0(x), \tag*{(1)}

with q∈Rq\in\mathbb{R}. The homogeneous case h(t)=0h(t)=0 is famously known to be integrable and have soliton solutions. We present a specific formula below.

The full-line NN-soliton

Let ζj=iηj\zeta_j=i\eta_j with η1>⋯>ηN>0\eta_1>\cdots>\eta_N>0, and let cj∈R∖{0}c_j\in\mathbb{R}\setminus\{0\}. Put

θj=−iζjx−2iζj2t,vj=(cjeθj,e−θj),Mjk=v‾j1vk1+v‾j2vk2ζk−ζj,\theta_j=-i\zeta_jx-2i\zeta_j^2t,\qquad v_j=(c_je^{\theta_j},e^{-\theta_j}),\qquad M_{jk}=\frac{\overline{v}_{j1}v_{k1}+\overline{v}_{j2}v_{k2}}{\zeta_k-\zeta_j},

and define the reflectionless NN-breather [7], [8]

U(x,t)=2i∑j,k=1Nvj1(M−1)jkv‾k2.(2)U(x,t)=2i\sum_{j,k=1}^{N}v_{j1}(M^{-1})_{jk}\overline{v}_{k2}. \tag*{(2)}

For N=1N=1 this is U(x,t)=2sgn⁡(c)⋅ηsech⁡(2ηx−log⁡∣c∣)e4iη2tU(x,t)=2\operatorname{sgn}(c)\cdot\eta\operatorname{sech}(2\eta x-\log|c|)e^{4i\eta^2t}.

The numbers ζj\zeta_j are of course the eigenvalues of the associated Zakharov-Shabat operator. By choosing them purely imaginary we ensure that the solitons have zero speed and lie ”on top of each other”. They do oscillate in time, hence the name ”breather”.

Since ζk−ζj=i(ηj+ηk)\zeta_k-\zeta_j=i(\eta_j+\eta_k), iMiM is a Gram matrix with kernel 1/(ηj+ηk)1/(\eta_j+\eta_k), which is positive definite. Hence det⁡M≠0\det M\ne0 for every xx, tt.

The half-line NN-soliton

Restricted to x>0x>0, UU solves (1) with h=0h=0 in the following cases [3]

  • q=0q=0 and cj=±1c_j=\pm1; the Robin condition degenerates into a Neumann condition.

  • q≠0q\ne0, ∣q∣<2ηj\lvert q\rvert<2\eta_j for all jj, with

cj=±(2ηj+q2ηj−q)(−1)N+1/2.c_j=\pm\left(\frac{2\eta_j+q}{2\eta_j-q}\right)^{(-1)^{N+1}/2}.

The trace U(0,⋅)U(0,\cdot) is exactly quasi-periodic, with frequencies in the Z\mathbb{Z}-module generated by ω1,…,ωN\omega_1,\ldots,\omega_N, where ωj=2ηj2\omega_j=2\eta_j^2. Indeed, at x=0x=0, θj=2iηj2t\theta_j=2i\eta_j^2t, so vj1=cjeiωjtv_{j1}=c_je^{i\omega_jt} and vj2=e−iωjtv_{j2}=e^{-i\omega_jt}. Every entry of MM and every factor in (2) is a trigonometric polynomial in the angles φj=ωjt\varphi_j=\omega_jt. The matrix iMiM is positive definite for arbitrary unimodular phases, so det⁡M≠0\det M\ne0. Hence U(0,t)U(0,t) is a rational function of the eiφje^{i\varphi_j} with non-vanishing denominator.

There is no decaying radiation term 1^{1} because the data are reflectionless. So the quasi-periodicity is exact for all t≥0t\ge0. If the ωj\omega_j are rationally independent, the trace has NN independent frequencies. Since the ηj\eta_j are arbitrary, there are different smooth, exponentially decaying data that give different frequency sets of any size.

Remark 1.1. The above elementary example shows that (in any reasonable sense) the map taking the data for the IBVP (initial data and the constant kk) to the Dirichlet trace is very unstable in terms of long time asymptotics. Small changes in (the parameters of) the initial data can give non-decaying changes in the Dirichlet trace.

The non-homogeneous Robin problem

If hh is not zero the problem is non-linearizable in the sense of [5]. The question still arises whether it is still in a sense integrable. That might be so if the unified transform method introduced by Fokas [4] could transform the IBVP to a Riemann-Hilbert problem with data that depend continuously on the data of the IBVP and that is amenable to a steepest descent asymptotic analysis. This would be impossible however if the instability mentioned above persists, small changes in the initial data giving rise to many different non-decaying changes in the Dirichlet trace; simply because any possible scattering theory for the boundary data requires a fixed asymptotic tt-behavior. Here on the contrary we have a continuum of different asymptotically quasi-periodic behaviors.

This is indeed what the following perturbation analysis shows.

An elementary calculation gives the time derivatives of mass N=∫0∞∣u∣2 dx\mathcal{N}=\int_0^\infty|u|^2\,dx, and the energy Eq=∫0∞(∣ux∣2−∣u∣4) dx−q∣u(0,t)∣2\mathcal{E}_q=\int_0^\infty(|u_x|^2-|u|^4)\,dx-q|u(0,t)|^2 in terms of the non-homogeneous term hh.

Lemma 2.1. Let uu be a smooth solution of (1), decaying as x→∞x\to\infty. Then

ddtN=2Im⁡(u(0,t)‾h(t)),(3)\frac{d}{dt}\mathcal{N}=2\operatorname{Im}\bigl(\overline{u(0,t)}h(t)\bigr), \tag*{(3)}
ddtEq=−2Re⁡(ut(0,t)‾h(t)).(4)\frac{d}{dt}\mathcal{E}_q=-2\operatorname{Re}\bigl(\overline{u_t(0,t)}h(t)\bigr). \tag*{(4)}

Proof. Mass. From ∂t∣u∣2=−2∂xIm⁡(u‾ux)\partial_t|u|^2=-2\partial_x\operatorname{Im}(\overline{u}u_x) we get N˙=2Im⁡(u‾ux)(0)\dot{\mathcal{N}}=2\operatorname{Im}(\overline{u}u_x)(0). Substituting ux(0)=−qu(0)+hu_x(0)=-qu(0)+h, the qq-term is real and drops out. Energy. Integrating by parts,

ddt∫0∞(∣ux∣2−∣u∣4)=−2Re⁡(ut‾ux)(0)−2Re⁡∫0∞ut‾(uxx+2∣u∣2u).\frac{d}{dt}\int_{0}^{\infty}\left(\lvert u_x\rvert^2-\lvert u\rvert^4\right)=-2\operatorname{Re}(\overline{u_t}u_x)(0)-2\operatorname{Re}\int_{0}^{\infty}\overline{u_t}\left(u_{xx}+2\lvert u\rvert^2u\right).

The last integral is −2Re⁡∫ut‾(−iut)=0-2\operatorname{Re}\int\overline{u_t}(-iu_t)=0. At the boundary,

−2Re⁡(ut‾ux)(0)=2qRe⁡(ut‾u)(0)−2Re⁡(ut‾h)=qddt∣u(0,t)∣2−2Re⁡(ut‾h),-2\operatorname{Re}(\overline{u_t}u_x)(0)=2q\operatorname{Re}(\overline{u_t}u)(0)-2\operatorname{Re}(\overline{u_t}h)=q\frac{d}{dt}\lvert u(0,t)\rvert^2-2\operatorname{Re}(\overline{u_t}h),

which gives (2.2).

For h=0h=0 both quantities are of course conserved.

Values on forced solitons

When a non-homogenous hh term appears, we can still consider the evolution of the eigenvalues of the Zakharov-Shabat operator. They of course can move in time.

Theorem 2.2. For the Robin-compatible 1-soliton of the first section with A=2ηA=2\eta,

N=A−q,Eq=−13(A3−q3).(5)\mathcal{N}=A-q,\qquad\mathcal{E}_q=-\frac{1}{3}\left(A^3-q^3\right). \tag*{(5)}

Proof. Write u=Asech⁡(Ax+c)eiA2tu=A\operatorname{sech}(Ax+c)e^{iA^2t} with s=tanh⁡c=q/As=\tanh c=q/A. Then

N=A(1−s),∫0∞∣ux∣2=A33(1−s3),∫0∞∣u∣4=A3(23−s+s33),∣u(0)∣2=A2(1−s2).\mathcal{N}=A(1-s),\qquad\int_{0}^{\infty}\lvert u_x\rvert^2=\frac{A^3}{3}(1-s^3),\qquad\int_{0}^{\infty}\lvert u\rvert^4=A^3\left(\frac{2}{3}-s+\frac{s^3}{3}\right),\qquad\lvert u(0)\rvert^2=A^2(1-s^2).

Combining these gives N=A−q\mathcal{N}=A-q and Eq=−13(A3−q3)\mathcal{E}_q=-\frac{1}{3}(A^3-q^3).

The above formulae easily generalize to the case of several solitons.

First-order frequency shifts

Let h=εh1h=\varepsilon h_1 with h1h_1 integrable and rapidly decaying, and let the unforced solution be a ”wall” 1-soliton UU. Assume for the moment that the forced solution will be close to a 1-soliton with amplitude A+δAA+\delta A, plus radiation. 5

Integrate (2.1) and (2.2) with the unforced trace inserted, and use (2.3). To first order,

δA=2Im⁡∫0∞U(0,t)‾h(t) dt,A2δA=2Re⁡∫0∞Ut(0,t)‾h(t) dt,(6)\delta A=2\operatorname{Im}\int_{0}^{\infty}\overline{U(0,t)}h(t)\,dt,\qquad A^2\delta A=2\operatorname{Re}\int_{0}^{\infty}\overline{U_t(0,t)}h(t)\,dt, \tag*{(6)}

and δω=2AδA\delta\omega=2A\delta A.

Since U(0,t)=A2−q2 eiA2tU(0,t)=\sqrt{A^2-q^2}\,e^{iA^2t} and Ut=iA2UU_t=iA^2U, the two relations in (2.4) coincide. They give

δA=2A2−q2Im⁡∫0∞h(t)e−iA2t dt.\delta A=2\sqrt{A^2-q^2}\operatorname{Im}\int_{0}^{\infty}h(t)e^{-iA^2t}\,dt.

This calculation can be generalized to the case of NN solitons, where we will end up for a system for the different δωj\delta\omega_j and δAj\delta A_j. In the general case the energy calculation becomes relevant. All δωj\delta\omega_j and δAj\delta A_j are real.

One reason to assume that a small perturbation does not destroy solitons is the following. All solitons corresponding to the eigenvalues iηji\eta_j correspond to zeros of some scattering data that depend continuously on the perturbation. These could move somewhat under a small enough perturbation, but cannot disappear. For this to happen they would have to hit the real line; but all zeros are at least at distance qq away from the real line.

Now, because of symmetries (ζ→−ζ‾)(\zeta\to-\overline{\zeta}), the eigenvalues have to also remain on the imaginary line, unless there is a collision of two imaginary eigenvalues. This does not happen for a small enough perturbation. Breathers remain breathers even if frequencies can change.

Note also that even if when there is collision of two imaginary eigenvalues that gives rise to eigenvalues with non-zero real parts, the leading order asymptotics at x=0x = 0 still come from the remaining eigenvalues with zero real parts!

If hh has compact support and the perturbation size is small, clearly one can ensure that eigenvalues remain imaginary. For decaying hh not of compact support, some eigenvalues can eventually leave the imaginary line. But in general not all of them will.

To summarize, we end up with a situation similar to the unforced case: continua of solutions with different non-decaying asymptotically quasi-periodic asymptotics at the wall x=0x = 0. This renders a unified transform impossible!

Numerical tests

We have asked Claude (Anthropic) to provide numerical runs where indeed small non-homogeneities just shift amplitudes and frequencies of breather solutions. We mention three examples below.

They used a Crank–Nicolson plus Strang splitting scheme with ghost-node Robin closure u−1=u1+2dx(qu0−h)u_{-1} = u_1 + 2dx\left(qu_0-h\right) and an absorbing layer. Frequencies were measured from the trace by a zero-padded FFT with peak interpolation. They compared forced and unforced runs on the same grid, thus removing the O(dx2)O(dx^2) shift of the computed frequencies.

One wall soliton, q=0.5q = 0.5, A=1A = 1, h=εe−t/2h = \varepsilon e^{-t/2} or resonant h=iεe−t/2eith = i\varepsilon e^{-t/2}e^{it}. The amplitude A′A' was measured from the frequency, the trace amplitude and the mass; the three agree to 10−410^{-4}. A scan over q∈{−0.5,0,0.75,0.95}q \in\{-0.5, 0, 0.75, 0.95\} with ε=0.01\varepsilon= 0.01 gave measured-to-predicted ratios between 0.97 and 1.01.

Two wall solitons, q=0.3q = 0.3, η=(1.6,1.0)\eta= (1.6, 1.0), so ω=(2.56,1)\omega= (2.56, 1). They used dx=0.02dx = 0.02, dt=0.002dt = 0.002, t≤450t \le450, and measured frequencies on [150,450][150, 450]. Halving ε\varepsilon roughly halves the deviation of the ratios from 1, as expected for an O(ε2)O(\varepsilon^2) remainder. A single forcing moves the two frequencies independently, and here in opposite directions. The late trace is again quasi-periodic, with both frequencies shifted.

Two wall solitons, q=−0.3q = -0.3, η=(0.8,0.5)\eta= (0.8, 0.5), h=ε(1−e−t)(1+t)−ph = \varepsilon(1-e^{-t})(1+t)^{-p}, t≤400t \le400, ε=0.05\varepsilon= 0.05. Different values of pp. Same phenomena: no new zeros, no non-imaginary zeros. For larger ε=0.3\varepsilon= 0.3 and p=0.5p = 0.5 more solitons can appear, but there are always at least two (sometimes more) imaginary eigenvalues.

Concluding remark

The above elementary discussion is contrasted with the analysis of the Dirichlet problem. In that case (see [1] and [2]) one can show that some kind of integrability does persist when perturbing the linearisable Dirichlet data (zero data).

Acknowledgements

We are grateful for several discussions with Claude (Anthropic).

References

  1. [1]D.C. ANTONOPOULOU, S. KAMVISSIS, On the Dirichlet to Neumann problem for the 1-dimensional Cubic NLS Equation on the Half-Line, Nonlinearity, 28 (2015), pp. 3073–3099.
  2. [2]D.C. ANTONOPOULOU, S. KAMVISSIS, Addendum to: On the Dirichlet to Neumann problem for the 1-dimensional Cubic NLS Equation on the Half-Line, Nonlinearity, 29 (2016), pp. 3206–3214.DOI
  3. [3]G. BIONDINI AND A. BUI, On the Nonlinear Schrödinger Equation on the Half Line with Homogeneous Robin Boundary Conditions, Stud. Appl. Math. 129 (2012)DOI
  4. [4]A.S. Fokas, A Unified Transform Method for Solving Linear and Certain Nonlinear PDEs, Proc. R. Soc. Lond. A 453 (1997), pp. 1411–1443.DOI
  5. [5]A. S. Fokas, A. R. Its and L.-Y. Sung, The nonlinear Schrödinger equation on the half-line, Nonlinearity, 18 (2005), pp. 1771–1822.
  6. [6]Jae Min Lee, Jonatan Lenells, The nonlinear Schrödinger equation on the half-line with homogeneous Robin boundary conditions, Proc. LMS, v.126, 2022DOI
  7. [7]Jianke Yang, General N-solitons and their dynamics in several nonlocal nonlinear Schrodinger equations, https://arxiv.org/pdf/1712.01181
  8. [8]Jianke Yang, Nonlinear Waves in Integrable and Non integrable Systems, SIAM, Philadelphia, 2010.DOI

Paper details

vlnc:W_njfk3pau
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC0-1.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents