Abstract

We consider stability of solitons of 3D Maxwell–Lorentz system with extended charged spinning particle. The solitons are solutions which correspond to a particle moving with a constant velocity $v\in\mathbb{R}^{3}$ with $|v|<1$ and rotating with a constant angular velocity $\omega\in R^{3}$.

Our main results are the orbital stability of moving solitons with $\omega=0$ and a linear orbital stability of rotating solitons with $v=0$.

The Hamilton–Poisson structure of the Maxwell–Lorentz system is degenerate and admits the Casimir invariants. We construct the Lyapunov function as a linear combination of the Hamiltonian with a suitable Casimir invariant. The key point is a lower bound for this function. The proof of the bound in the case $\omega\neq 0$ relies on angular momentum conservation and suitable spectral arguments including the Heinz inequality and closed graph theorem.

1 Introduction

The paper addresses the stability of solitons of 3D Maxwell–Lorentz system with an extended charged spinning particle. We choose the units such that the speed of light is $c=1$. The equations read as follows (see [37, 40]):

$$\left\{\begin{array}[]{rcl}\dot{E}(x,t)&=&{\rm curl{\hskip 1.42262pt}}B(x,t)-w(x,t)\rho(x-q(t)),\qquad{\rm div{\hskip 1.42262pt}}E(x,t)=\rho(x-q(t))\\ \dot{B}(x,t)&=&-\,{\rm curl{\hskip 1.42262pt}}E(x,t),\qquad\qquad\qquad\quad\quad\quad\quad{\rm div{\hskip 1.42262pt}}B(x,t)=0\\ m\dot{q}(t)&=&p(t),\quad\dot{p}(t)=\langle E(x,t)+w(x,t){\rm\wedge}B(x,t),\rho(x-q(t))\rangle\\ I\dot{\omega}(t)&=&\langle(x-q(t)){\rm\wedge}\big{[}E(x,t)+w(x,t){\rm\wedge}B(x,t)\big{]},\rho(x-q(t))\rangle\end{array}\right|,$$ (1.1)

where $\rho(x-q(t))$ is the charge distribution of the extended particle centered at the point $q(t)\in\mathbb{R}^{3}$, $m>0$ is the mass of the particle and $I>0$ is its moment of inertia, $\omega(t)$ is the angular velocity of the particle rotation and $w(x,t):=\dot{q}(t)+\omega(t){\rm\wedge}(x\!-\!q(t))$ is the velocity field. The brackets $\langle\cdot,\cdot\rangle$ here and below denote the integral over $x\in\mathbb{R}^{3}$ or inner product in the Hilbert spaces $L^{2}:=L^{2}(\mathbb{R}^{3})\otimes\mathbb{C}^{n}$ with suitable $n\geq 1$.

This model was introduced by Abraham in 1903–1905 (see [1, 2]) for the description of the classical extended electron coupled to its own Maxwell field. The model allows one to avoid the “ultraviolet divergence”, that is, the infiniteness of the own energy and mass of electrons which takes place in the case $\rho(x)=\delta(x)$. Using this model, Abraham was the first to discover the mass-energy equivalence (up to the doubtful factor $4/3$; see [1, 2] and also [23, Section 12.9]), which was a precursor of the Einstein theory. This system also served as the classical Landé model of spin in Old Quantum Mechanics (1900–1924): see [6] (and also [23, Chapter 14] and [24, Appendix A]). Various approximations of such systems were used to explain the famous radiation damping in classical electrodynamics: the Lorentz–Dirac equation with runaway solutions introduced by Dirac in [9] and many other approximations; see [20, Chapter 16]. The detailed account on the genesis and early investigations of the system by Dirac, Poincaré, Sommerfeld, and others can be found in [40, Chapter 3].

The first results on the corresponding long-time behavior for the system (1.1) were obtained in the 1990s. This system plays a crucial role in a rigorous analysis of radiation by moving particles; see [26, 40]. The mathematical analysis of this system is useful in connection to the related problems of nonrelativistic QED. In particular, the similarity in the renormalization of mass was pointed out by Hiroshima and Spohn [12].

We assume that the charge density $\rho(x)$ is smooth, spherically-symmetric, and not identically zero:

$$\rho\in C_{0}^{\infty}(\mathbb{R}^{3}),\qquad\rho(x)=\rho_{1}(|x|);\qquad\rho(x)=0\,\,\,\,{\rm for}\,\,|x|\geq R_{\rho};\qquad\rho(x)\not\equiv 0.$$ (1.2)

One of the basic peculiarities of the system (1.1) is its invariance under the Euclidean group generated by translations, rotations and reflections of the space $\mathbb{R}^{3}$. Moreover, the system admits solitons ${\cal S}_{v,\omega}$ discovered by Spohn [40]. The solitons are solutions moving with constant speed $v\in\mathbb{R}^{3}$ and rotating with constant angular velocity $\omega\in\mathbb{R}^{3}$. The solitons are trajectories of one-parametric subgroups of the Euclidean symmetry group. As is pointed out in [40], for $\rho(x)\not\equiv 0$, the solitons with finite energy exist only for $(v,\omega)\in\Sigma$, where

$$\Sigma=\{(v,\omega)\in\mathbb{R}^{3}\times\mathbb{R}^{3}:|v|<1\,\,{\rm and}\,\,{\rm either}\,\,\omega\|v\,\,{\rm or}\,\,\,\omega\bot v\}.$$ (1.3)

We justified the restrictions (1.3) on the parameters of solitons and calculated effective moment of inertia of solitons (6.9).

Remark 1.1.

In the case $\rho(x)\equiv 0$ we have $\ddot{q}(t)=\dot{\omega}(t)=0$ and the Maxwell equations do not depend on the motion of the particle, so the energy of the Maxwell field is conserved, and so are the particle momentum and angular momentum. The solitons read as $({E(x,t)},B(x,t),q(t),p(t),\omega(t))=(0,0,vt+q_{0},mv,\omega)$ with arbitrary $q_{0},v,\omega\in\mathbb{R}^{3}$, and they are obviously orbitally stable by the conservation laws; see Definition 6.5.

Our main results are i) the orbital stability of solitons ${\cal S}_{v,0}$ with $|v|<1$ and ii) the linear orbital stability of solitons ${\cal S}_{0,\omega}$ with $\omega\in\mathbb{R}^{3}$ in the class of solutions with zero total momentum. In particular, the total momentum is zero for solutions with “symmetric” initial data:

$$q(0)=p(0)=0,\qquad E(-x,0)\equiv-E(x,0),\qquad B(-x,0)\equiv B(x,0).$$ (1.4)

Our approach essentially relies on the Hamilton–Poisson structure of the system (1.1) which was constructed in [27]. The structure is degenerate, so the system admits the Casimir invariants, and the methods of [11] are not applicable to the system.

The proof of the orbital stability of solitons ${\cal S}_{v,0}$ relies on the reduction to the comoving frame and lower bound for the Hamiltonian as the Lyapunov function. The Hamilton functional of the reduced system depends on the parameter $P$ which is the conserved total momentum. Thus, we restrict the Hamiltonian to the states with the fixed value of $P$.

To prove the stability of solitons ${\cal S}_{0,\omega}$, we employ conservation of angular momentum $J$. The angular momentum is a functional on the phase space of the reduced system only for the total momentum $P=0$. So, we consider the stability in the class of solutions with $P=0$. In particular, all solutions with initial states (1.4) have zero total momentum.

The linear orbital stability of the solitons refers to the stability of the linearized reduced dynamics in the tangent space to the manifold of states with constant angular momentum. The manifold is not well-defined, but the tangent space is; see Section 9. So, the nonlinear stability problem on the manifold is not correct, as opposed to the linearized problem on the tangent space.

To prove the linear orbital stability, we construct the Lyapunov function $L$ for the linearized equations as the sum of the quadratic part of the Hamiltonian with the Casimir functional. Such a strategy is an analog of the energy–Casimir method [13, 36]; see Remark 10.4. We prove the conservation of $L$ in the linearized dynamics and establish a suitable lower bound for $L$. A lower bound provides a priori estimate and existence of global solutions, and at the same time the stability of the linearized dynamics. The bound is proved by suitable spectral arguments including the Heinz inequality from the theory of interpolation [21] and the closed graph theorem. Further comments on our methods and open questions can be found in Appendix A.

Let us comment on related work. The mathematical theory of Maxwell–Lorentz system (1.1) started after Nodvick’s paper [37], where the Hamilton least action principle and conservation laws have been established in the Euler angles representation. The coordinate-free proof of the conservation laws was given by Kiessling [22]. The coordinate-free proof of the Hamilton principle was given in [18] using the technique of [4, 38] which relies on the Lie algebras and the Poincaré equation. This technique was developed in [19], where the general theory of invariants was constructed for the Poincaré equations on manifolds and applied to the system (1.1).

In 1990’–2000’, the stability of solitons was studied for the Maxwell–Lorentz system (1.1) without spinning (the first three lines with $w(t)\equiv\dot{q}(t)$). In [5], Bambusi and Galgani proved the existence and orbital stability of solitons for such a system. The proof relies on the transition to a comoving frame and a lower bound for the reduced Hamiltonian. The global convergence to solitons for the same system with relativistic kinetic energy of the particle was proved in [15]. In [32], the global attraction to stationary states was established in presence of an external confining potential. All these results were obtained under the Wiener condition on the Fourier transform of the charge density:

$$\hat{\rho}(k):=(2\pi)^{-3/2}\int e^{ikx}\rho(x)dx\neq 0,\qquad k\in\mathbb{R}^{3},$$ (1.5)

which allows us to apply the Wiener Tauberian theorem. This method was introduced first in [31, 33] for the system of the particle coupled to the scalar wave field instead of the Maxwell field.

The condition means a strong coupling of the charged particle to the eigenfunctions $e^{ikx}$ of the continuous spectrum of the free field. The results [15, 32] provide the first rigorous proof of the radiation damping in classical electrodynamics. For the survey, see [40] and [25, 26].

In [17, 29], the asymptotic stability of the solitary manifold was proved for the 3D and 2D Maxwell–Lorentz systems without spinning: for large times, any solution with initial data close to the solitary manifold is close to the sum of a soliton and a dispersive wave. The proof relies on the Buslaev–Perelman–Sulem method of symplectic projection onto solitary manifold [7, 8] and uses the Wiener condition (1.5). The adiabatic effective dynamics of solitons for the 3D system was proved in [35].

For the Maxwell–Lorentz system with spinning particle at rest ($q(t)\equiv 0$), the global convergence to rotating solitons was established in [16] and [34]. In [16], the convergence is proved in the case of sufficiently small charge density $\rho$. In [34], Kunze extended the result to the same system without the smallness condition. This condition is replaced by the assumption that the mass of the particle does not belong to a discrete set of resonances.

The stability of solitons for the Maxwell–Lorentz system (1.1) with moving and spinning particle was not considered previously.

2 The Maxwell potentials and the well-posedness

As is well-known [20], the second line of the system (1.1) implies the following expressions of the Maxwell fields in terms of the potentials $A(x,t)=(A_{1}(x,t),A_{2}(x,t),A_{3}(x,t))$ and $\Phi(x,t)$:

$$B(x,t)={\rm curl{\hskip 1.42262pt}}A(x,t),\qquad E(x,t)=-\dot{A}(x,t)-\nabla\Phi(x,t).$$ (2.1)

We choose the Coulomb gauge

$${\rm div{\hskip 1.42262pt}}A(x,t)=0.$$ (2.2)

Then the first line of the system (1.1) becomes

$$-\ddot{A}(x,t)-\nabla\dot{\Phi}(x,t)=-\Delta A(x,t)-w(x,t)\rho(x-q(t)),\qquad\Delta\Phi(x,t)=\rho(x-q(t)).$$ (2.3)

Hence, $\Phi(x,t)$ is the Coulomb potential $\Phi(x,t)=-\displaystyle\frac{1}{4\pi}\int\frac{\rho(y-q(t))}{|x-y|}dy$. The gradient $\nabla\dot{\Phi}(x,t)$ in the first equation in (2.3) can be eliminated by the application of the orthogonal projection ${\cal P}$ in $L^{2}$ onto the divergence-free vector fields. Substituting the expressions (2.1) into the system (1.1) and using the notation $j(x,t):={\cal P}([\dot{q}(t)+\omega(t){\rm\wedge}(x-q(t))]\rho(x-q(t)))$ for the current density, we obtain the system

$$\left\{\begin{array}[]{rcl}\ddot{A}(x,t)&\!\!=&\!\!\Delta A(x,t)+j(x,t)\\ m\ddot{q}(t)&\!\!=&\!\!\langle-\dot{A}(x,t)\rho(x\!-\!q(t))+j(x,t){\rm\wedge}{\rm curl{\hskip 1.42262pt}}A(x,t)\rangle\\ I\dot{\omega}(t)&\!\!=&\!\!\langle(x\!-\!q(t)){\rm\wedge}[-\dot{A}(x,t)\rho(x\!-\!q(t))+j(x,t){\rm\wedge}{\rm curl{\hskip 1.42262pt}}A(x,t)]\rangle\end{array}\right|.$$ (2.4)

In the last two equations, the terms with $\Phi(x,t)$ cancel since

$$\langle\nabla\Phi(x,t),\rho(x\!-\!q)\rangle=\langle-ik\frac{\hat{\rho}(k)}{k^{2}},\hat{\rho}(k)\rangle=0,\quad\langle(x\!-\!q)\nabla\Phi(x,t),\rho(x\!-\!q)\rangle=-\langle\nabla{\rm\wedge}k\frac{\hat{\rho}(k)}{k^{2}},\hat{\rho}(k)\rangle=0.$$

Indeed, the first equality holds since the function $\hat{\rho}(k)$ is even. The second equality follows from the rotation invariance (1.2).

Denote the spaces of real functions $C^{k}=C^{k}(\mathbb{R}^{3})\otimes\mathbb{R}^{3}$ with $k=0,1,\ldots$, the Sobolev spaces $H^{s}=H^{s}(\mathbb{R}^{3})\otimes\mathbb{R}^{3}$ with $s\in\mathbb{R}$, and $\dot{H}^{1}=\dot{H}^{1}(\mathbb{R}^{3})\otimes\mathbb{R}^{3}$ being the completion of $C_{0}^{\infty}:=C_{0}^{\infty}(\mathbb{R}^{3})\otimes\mathbb{R}^{3}$ with the norm $\|A\|_{\dot{H}^{1}}=\|\nabla A\|_{L^{2}}$.

Definition 2.1.

Denote i) ${\cal F}^{0}\!:=\!\{A\!\in\!L^{2}\!:{\rm div{\hskip 1.42262pt}}A(x)\!\equiv\!0\}$ the Hilbert space endowed with the norm of $L^{2}$, and ii) $\dot{\cal F}^{1}\!:=\!\{A\!\in\!\dot{H}^{1}\!:{\rm div{\hskip 1.42262pt}}A(x)\!\equiv\!0\}$ the Hilbert space which is the completion of $\{A\in C_{0}^{\infty}:{\rm div{\hskip 1.42262pt}}A(x)\equiv 0\}$ with respect to the norm $\|\nabla A\|_{L^{2}}$.

Remark 2.2.

The space $C_{0}^{\infty}\cap{\cal F}^{0}$ is dense in ${\cal F}^{0}$ (see Appendix C).

Denote the Hilbert spaces

$${\mathbb{Y}}:=\dot{\cal F}^{1}\oplus{\cal F}^{0}\oplus\mathbb{R}^{3}\oplus\mathbb{R}^{3}\oplus\mathbb{R}^{3},\qquad{\mathbb{V}}=L^{2}\oplus H^{-1}\oplus\mathbb{R}^{3}\oplus\mathbb{R}^{3}\oplus\mathbb{R}^{3}.$$ (2.5)

For $n=1,2,3$ let us denote

$$\langle\dot{A}(x),\nabla_{*}A(x)\rangle_{n}\!=\!\langle\dot{A}(x),\nabla_{n}A(x)\rangle,\quad\langle((x\!-\!q){\rm\wedge}\nabla)_{*}A(x),\dot{A}(x)\rangle_{n}\!=\!\langle((x\!-\!q){\rm\wedge}\nabla)_{n}A(x),\dot{A}(x)\rangle.$$ (2.6)

The notations are suitable versions of the basic operation [3, (10)] from the theory of coadjoint representation for the groups of translations and rotations, respectively. Denote by $p$ and $\pi$ the momentum and angular momentum of the particle in the Maxwell field,

$$p=m\dot{q}+\langle A(q+y),\rho(y)\rangle,\qquad\pi=I\omega+\langle y{\rm\wedge}A(q+y),\rho(y)\rangle.$$ (2.7)

Proposition 2.3.

i) For any initial state $Y(0)=(A(x,0),\dot{A}(x,0),q(0),p(0),\omega(0))\in{\mathbb{Y}}$, the system (2.4) admits a unique solution

$$Y(t)=(A(x,t),\dot{A}(x,t),q(t),p(t),\omega(t))\in C(\mathbb{R},{\mathbb{Y}})\cap C^{1}(\mathbb{R},{\mathbb{V}}).$$ (2.8)

ii) The map $W(t):Y(0)\mapsto Y(t)$ is continuous in ${\mathbb{Y}}$ for every $t\in\mathbb{R}$, and the energy and momentum are conserved:

$${\cal E}(t) = \frac{1}{2}\int[\dot{A}^{2}(x,t)+|{\rm curl{\hskip 1.42262pt}}A(x,t)|^{2}]dx+\frac{1}{2}m\dot{q}^{2}(t)+\frac{1}{2}I\omega^{2}(t)=\mathop{\rm const}\nolimits,\qquad t\in\mathbb{R},$$ (2.9)

$$P(t) = -\langle\dot{A}(x,t),\nabla_{*}A(x,t)\rangle+p(t)+\langle\rho(x-q(t)),A(x,t)\rangle=\mathop{\rm const}\nolimits,\qquad\,\,\,t\in\mathbb{R}.$$ (2.10)

iii) Let for $|\alpha|\leq 3$ and $|\beta|\leq 2$,

$$A(x,0)\!\in\!C^{3}\!\cap\!\dot{\cal F}^{1},\quad\dot{A}(x,0)\!\in\!C^{2}\!\cap\!{\cal F}^{0};\quad|\partial_{x}^{\alpha}A(x,0)|+|\partial_{x}^{\beta}\dot{A}(x,0)|\leq C(1+|x|)^{-3},\quad x\in\mathbb{R}^{3}.$$ (2.11)

Then $\partial_{x}^{\alpha}\partial_{t}^{l}A(x,t)\in C^{2-(|\alpha|+l)}(\mathbb{R}^{3}\times[0,T])\!\otimes\!\mathbb{R}^{3}$ for any $T>0$, $l=0,1,2$, and $|\alpha|+l\leq 2$, and

$$\left\{\begin{array}[]{rcll}|\partial_{t}^{l}A(x,t)|&\leq&C(1+|x|)^{-2},&l\leq 2\\ \\ |\partial_{x}^{\alpha}\partial_{t}^{l}A(x,t)|&\leq&C(1+|x|)^{-3},&\alpha\neq 0,\,\,|\alpha|+l\leq 2\end{array}\right|,\qquad x\in\mathbb{R}^{3},\,\,\,|t|<T.$$ (2.12)

For such solutions,

$$J(t)=\langle A(x,t){\rm\wedge}\dot{A}(x,t)\rangle-\langle((x-q(t)){\rm\wedge}\nabla)_{*}A(x,t),\dot{A}(x,t)\rangle+q(t){\rm\wedge}P+\pi(t)=\mathop{\rm const}\nolimits,\,\,t\in\mathbb{R}.$$ (2.13)

The items i) and ii) were proved in [27], and (2.13) is proved in [22] by a partial integration; see also [28]. The bounds (2.12) justify the calculations. Let us comment on the proof of (2.12). In the first equation in (2.4), the current reads

$$j(x,t):=j_{0}(x,t)+\omega(t){\rm\wedge}(x-q(t))\rho(x-q(t)),\qquad j_{0}(x,t)={\cal P}([\dot{q}(t)\rho(\cdot-q(t))].$$

In the Fourier transform, $\hat{j}_{0}(k,t)=e^{iq(t)}[\dot{q}(t)-\frac{k(\dot{q}(t)\cdot k)}{k^{2}}]\hat{\rho}(k)$. Hence, $|\partial_{k}^{\beta}[k^{\alpha}\hat{j}_{0}(k,t)]|\sim|k|^{|\alpha|-|\beta|}$ as $|k|\to 0$. Therefore, $\partial_{k}^{\beta}[k^{\alpha}\hat{j}_{0}(k,t)]\in L^{1}(\mathbb{R}^{3})$ for $|\beta|\leq 2+|\alpha|$, so

$$|\partial_{x}^{\alpha}j_{0}(x,t)|\leq C_{\alpha}(1+|x|)^{-2-|\alpha|},\quad x\in\mathbb{R}^{3},\qquad\forall\alpha.$$

Now the decay (2.12) follows from the Kirchhoff formula for solutions to the wave equation.

3 The Hamilton–Poisson structure

In this section we recall the Hamilton–Poisson structure of the system (2.4) which has been constructed in [27]. The construction relies on the Hamilton least action principle for the system (2.4) established in [18, 37]. Using this principle and the Lie–Poincaré technique [3, 14], we have shown in [27] that the system (2.4) admits the following representation:

$$\left\{\begin{array}[]{rcl}\dot{A}&\!\!=&\!\!\Pi,\qquad\dot{\Pi}=\Delta A+j\\ m\dot{q}&\!\!=&\!\!p-\langle A(q+y),\rho(y)\rangle,\quad\,\,\dot{p}=\langle[\dot{q}+\omega{\rm\wedge}y]\rho(y),\nabla_{*}A(q+y)\rangle\\ \dot{\pi}&\!\!=&\!\!\omega{\rm\wedge}\pi\end{array}\right|.$$ (3.1)

Let us rewrite the energy (2.9) as the Hamiltonian functional:

$$\!\!H(A,\Pi,q,p,\pi)=\frac{1}{2}\langle\Pi,\Pi\rangle+\frac{1}{2}\langle{\rm curl{\hskip 1.42262pt}}A,{\rm curl{\hskip 1.42262pt}}A\rangle+\frac{1}{2}m\dot{q}^{2}+\frac{1}{2}I\omega^{2}$$ (3.2)

$$\!\!=\!\! \!\!\frac{1}{2}\!\int[|\Pi(x)|^{2}\!+\!|{\rm curl{\hskip 1.42262pt}}A(x)|^{2}]dx\!+\!\frac{1}{2m}[p\!-\!\langle A(x),\rho(x\!-\!q)\rangle]^{2}\!+\!\frac{1}{2I}[\pi\!-\!\langle(x\!-\!q){\rm\wedge}A(x),\rho(x\!-\!q)\rangle]^{2}.\qquad$$

The Hamiltonian is well-defined and Fréchet-differentiable on the Hilbert phase space ${\mathbb{Y}}$ defined in (2.5). The system (3.1) admits the Hamilton–Poisson representation

$$\dot{Y}={\cal J}(Y)DH(Y),\qquad Y=(A,\Pi,q,p,\pi),\qquad{\cal J}(Y)=\left(\begin{array}[]{ccccc}0&1&0&0&0\\ -1&0&0&0&0\\ 0&0&0&1&0\\ 0&0&-1&0&0\\ 0&0&0&0&-\pi{\rm\wedge}\end{array}\right).$$ (3.3)

Note that the operator ${\cal J}(Y)$ is not invertible for all $Y$. In detail, the system (3.3) reads as

$$\dot{A}=D_{\Pi}H,\quad\dot{\Pi}=-D_{A}H;\qquad\dot{q}=D_{p}H,\quad\dot{p}=-D_{q}H;\qquad\dot{\pi}=-\pi{\rm\wedge}D_{\pi}H.$$

The system admits the Casimir invariants

$$C(Y)=\pi^{2},\qquad Y=(A,\Pi,q,p,\pi).$$ (3.4)

Indeed, $DC=(0,0,0,0,2\pi)$, and so

$$\partial_{t}C(Y(t))=\big{(}DC(Y(t)),\dot{Y}(t)\big{)}\!\! \!\!=\!\! \big{(}DC(Y(t)),{\cal J}(Y(t))DH(Y(t))\big{)}$$ (3.5)

$$\!\!=\!\! \!\!-\big{(}{\cal J}^{*}(Y(t))DC(Y(t)),DH(Y(t))\big{)}=0$$

since the structural operator ${\cal J}(Y)$ is skew-symmetric by (3.3).

4 The canonical transformation to the comoving frame

Denote the fields in the comoving frame

$${\bf A}(y,t):=A(q(t)+y,t),\qquad{\bm{\Pi}}(y,t):=\Pi(q(t)+y,t).$$ (4.1)

We are going to change the variables in the system (3.3) via the map

$$T:(A,\Pi,q,p,\pi)\mapsto({\bf A},{\bm{\Pi}},q,P,\pi).$$ (4.2)

Here $P$ is the conserved total momentum (2.10): using (2.6), $P$ can be written as

$$P=p-\langle{\bm{\Pi}},\nabla_{*}{\bf A}\rangle.$$ (4.3)

Hence, we can define

$${\bf H}({\bf A},{\bm{\Pi}},q,P,\pi):=H(A,\Pi,q,p,\pi),\qquad p=P+\langle{\bm{\Pi}},\nabla_{*}{\bf A}\rangle.$$ (4.4)

Lemma 4.1.

The system (3.3) in the new variables is equivalent to a similar system with the Hamiltonian ${\bf H}$:

$$\dot{\bf Y}(t)={\cal J}({\bf Y}(t))D{\bf H}({\bf Y}(t)),\qquad{\bf Y}(t)=({\bf A}(t),{\bm{\Pi}}(t),q(t),P(t),\pi(t)).$$ (4.5)

Proof.

First, the map (4.2) is canonical, i.e., it leaves the canonical form unchanged:

$$\langle{\bm{\Pi}},\dot{\bf A}\rangle+P\cdot\dot{q}+\pi\cdot\omega=\langle\Pi,\dot{A}\rangle+p\cdot\dot{q}+\pi\cdot\omega.$$ (4.6)

Indeed, $\dot{\bf A}(y,t)=\dot{A}(q(t)+y,t)+\dot{q}(t)\cdot\nabla A(q(t)+y,t)$ by (4.1). Hence, the identity (4.6) reduces to

$$\langle\Pi(q+y),\dot{q}\cdot\nabla A(q+y)\rangle+P\cdot\dot{q}=p\cdot\dot{q},$$

which holds by (4.3). Second, the identities (4.6) and (4.4) imply that the Lagrangians corresponding to the Hamiltonians ${\bf H}$ and $H$ are related via the map $T$: ${\bf l}({\bf A},q,\dot{\bf A},\dot{q},\omega)\equiv l(A,q,\dot{A},\dot{q},\omega).$ Therefore, the Lagrangian actions also coincide:

$$\int_{a}^{b}{\bf l}({\bf A},q,\dot{\bf A},\dot{q},\omega)dt=\int_{a}^{b}l(A,q,\dot{A},\dot{q},\omega)dt.$$

Hence, the solutions of the systems (3.3) and (4.5) also are related by this transformation since they are critical trajectories of the action. ∎

Remark 4.2.

Similar canonical transformations have been applied in [5, 15] to the systems of type (1.1) without spinning and in [31] to the system of the particle coupled to the scalar field.

5 The reduced system

The Hamiltonian (4.4) does not depend on the variable $q$. Hence, the equations for $q$ and $P$ in the system (4.5) reduce to

$$\dot{q}=D_{P}{\bf H},\qquad\dot{P}=0,$$ (5.1)

which correspond to the conservation of the conjugate momenta $P$. As a result, the system (4.5) reduces to the following family of the Hamiltonian systems with the parameter $P\in\mathbb{R}^{3}$:

$$\dot{\bf Z}(t)={\mathbb{J}}({\bf Z}(t))D{\bf H}_{P}({\bf Z}(t)),\qquad{\bf Z}(t)=({\bf A}(t),{\bm{\Pi}}(t),\pi(t)),$$ (5.2)

where

$${\mathbb{J}}({\bf Z})=\left(\begin{array}[]{ccccc}0&1&0\\ -1&0&0\\ 0&0&-\pi{\rm\wedge}\end{array}\right),\quad{\bf Z}=({\bf A},{\bm{\Pi}},\pi),$$ (5.3)

and ${\bf H}_{P}$ denotes the reduced Hamiltonian ${\bf H}_{P}({\bf A},{\bm{\Pi}},\pi):={\bf H}({\bf A},{\bm{\Pi}},q,P,\pi)$. According to (4.4) and (3.2),

$${\bf H}_{P}({\bf A},{\bm{\Pi}},\pi) = \frac{1}{2}\!\int\![|{\bm{\Pi}}(y)|^{2}+|{\rm curl{\hskip 1.42262pt}}{\bf A}(y)|^{2}]dy$$ (5.4)

$$+\frac{1}{2m}[P+\langle{\bm{\Pi}},\nabla_{*}{\bf A}\rangle\!-\!\langle{\bf A},\rho\rangle]^{2}\!+\frac{1}{2I}[\pi-\langle y{\rm\wedge}{\bf A}(y),\rho(y)\rangle]^{2}.$$

In detail, the system (5.2) reads as

$$\left\{\begin{array}[]{rcl}\dot{\bf A}(y,t)&\!\!=&\!\!D_{\bm{\Pi}}{\bf H}_{P}({\bf Z}(t))={\bm{\Pi}}(y,t)+(v(t)\cdot\nabla){\bf A}(y,t)\\ \dot{\bm{\Pi}}(y,t)&\!\!=&\!\!-D_{\bf A}{\bf H}_{P}({\bf Z}(t))=\Delta{\bf A}(y,t)+(v(t)\cdot\nabla){\bm{\Pi}}(y,t)+{\bf j}(y,t)\\ \dot{\pi}(t)&\!\!=&\!\!-\pi(t){\rm\wedge}D_{\pi}{\bf H}_{P}({\bf Z}(t))=-\pi(t){\rm\wedge}\omega(t)\end{array}\right|,$$ (5.5)

where ${\bf j}(y,t)=j(q(t)+y,t)$ while $v(t)$ and $\omega(t)$ are given by

$$v(t)=\frac{1}{m}[P-\langle{\bm{\Pi}}(t),\nabla_{*}{\bf A}(t)\rangle-\langle{\bf A}(y,t),\rho(y)\rangle],\qquad\omega(t)=\frac{1}{I}[\pi(t)-\langle y{\rm\wedge}{\bf A}(y,t),\rho(y)\rangle].$$ (5.6)

The Hamiltonian (5.4) is well-defined and Fréchet-differentiable on the reduced phase space

$${\mathbb{Z}}=\dot{\cal F}^{1}\oplus{\cal F}^{0}\oplus\mathbb{R}^{3},$$ (5.7)

and it is conserved along trajectories of the system (5.5).

6 The solitons

Here we calculate the solitons of the system (3.1), which are solutions of the form

$$S_{v,\omega}(t)=({\bf A}_{v,\omega}(x-vt),{\bm{\Pi}}_{v,\omega}(x-vt),vt,p_{v},\pi_{v,\omega}).$$ (6.1)

The second line of (3.1) and the definition (2.7) imply that

$$p_{v}=mv+\langle{\bf A}_{v,\omega}(y),\rho(y)\rangle,\qquad\pi_{v,\omega}=I\omega+\langle y{\rm\wedge}A_{v,\omega}(q+y),\rho(y)\rangle.$$ (6.2)

The corresponding conserved (total) momentum is given by (2.10):

$$P_{v,\omega}=-\langle{\bm{\Pi}}_{v,\omega},\nabla_{*}{\bf A}_{v,\omega}(t)\rangle+mv+\langle\rho(y),{\bf A}_{v,\omega}(y)\rangle.$$ (6.3)

In the comoving frame, the solitons ${\bf S}_{v,\omega}=({\bf A}_{v,\omega},{\bm{\Pi}}_{v,\omega},\pi_{v,\omega})\in{\mathbb{Z}}$ become stationary solutions of the reduced system (5.5):

$$\left\{\begin{array}[]{rcl}0&\!\!=&\!\!D_{\bm{\Pi}}{\bf H}_{P_{v,\omega}}({\bf S}_{v,\omega})={\bm{\Pi}}_{v,\omega}(y)+(v\cdot\nabla){\bf A}_{v,\omega}(y)\\ 0&\!\!=&\!\!-D_{\bf A}{\bf H}_{P_{v,\omega}}({\bf S}_{v,\omega})=\Delta{\bf A}_{v,\omega}(y)+(v\cdot\nabla){\bm{\Pi}}_{v,\omega}(y)+{\bf j}(y,t)\\ 0&\!\!=&\!\!-\pi_{v,\omega}{\rm\wedge}D_{\pi}{\bf H}_{P_{v,\omega}}({\bf S}_{v,\omega})=-\pi_{v,\omega}{\rm\wedge}\omega\end{array}\right|.$$ (6.4)

The first two equations imply that

$$\Delta{\bf A}_{v,\omega}(y)-(v\cdot\nabla)^{2}{\bf A}_{v,\omega}(y)=-{\bf j}(y,t)=-{\cal P}[v\rho(\cdot)]+\omega{\rm\wedge}y\rho(y)].$$ (6.5)

Note that ${\cal P}[y{\rm\wedge}\omega\rho(y)]=y{\rm\wedge}\omega\rho(y)$ due to the spherical symmetry (1.2), while on the Fourier transform side one has $\widehat{{\cal P}[v\rho]}(k)=(v-\frac{k(v\cdot k)}{k^{2}})\hat{\rho}(k)$. Hence, solving (6.5) on the Fourier transform side, we obtain:

$$\hat{\bf A}_{v,\omega}(k)=\frac{(v-\frac{k(v\cdot k)}{k^{2}}-i\omega{\rm\wedge}\nabla)\hat{\rho}(k)}{k^{2}-(v\cdot k)^{2}}.$$ (6.6)

Now the first equation of (6.4) gives ${\bm{\Pi}}_{v,\omega}(y)=-v\cdot\nabla{\bf A}_{v,\omega}(y).$ The conditions (1.2) imply that

$${\bf A}_{v,\omega}\in\dot{\cal F}^{1},\quad{\bm{\Pi}}_{v,\omega}\in{\cal F}^{0},\qquad{\rm for}\,\,\,|v|<1,\,\,\omega\in\mathbb{R}^{3}.$$ (6.7)

Remark 6.1.

For $|v|\geq 1$, formula (6.6) shows that ${\bf A}_{v,\omega}\not\in\dot{\cal F}^{1}$, hence the corresponding solitons ${\bf S}_{v,\omega}\in{\mathbb{Z}}$ do not exist.

It remains to check the last equation in (3.1). It is equivalent to the relation

$$\pi_{v,\omega}\|\omega.$$ (6.8)

Lemma 6.2.

Let all conditions (1.2) hold. Then the relation (6.8) holds if and only if $(v,\omega)\in\Sigma$. In this case,

$$\pi_{v,\omega}=I_{\rm eff}\omega,\qquad I_{\rm eff}=I+\delta I,\qquad\delta I=\left\{\begin{array}[]{ll}\delta I^{\|}=\displaystyle\int\frac{k_{3}^{2}+k_{2}^{2}}{[k^{2}-v^{2}k_{1}^{2}]k^{2}}|\nabla\rho(k)|^{2}dk,&\omega\|v\\ \\ \delta I^{\bot}=\displaystyle\int\frac{k_{3}^{2}+k_{2}^{2}}{[k^{2}-v^{2}k_{2}^{2}]k^{2}}|\nabla\rho(k)|^{2}dk,&\omega\bot v\end{array}\right|.$$ (6.9)

Above, $I_{\rm eff}=I_{\rm eff}(v)$ is the effective moment of inertia of the soliton. We will prove this lemma in Appendix B. By (1.2), we have

$$I_{\rm eff}>I.$$ (6.10)

Corollary 6.3.

The soliton ${\bf S}_{v,\omega}\in{\mathbb{Z}}$ exists and is unique if $(v,\omega)\in\Sigma$. The solitons do not exist for $(v,\omega)\not\in\Sigma$.

The following lemma is a particular case of Lemma 5.2 from [27].

Lemma 6.4.

The total momentum (6.3) admits the representation

$$P_{v,\omega}=m_{\rm eff}(|v|,|\omega|)v,\qquad(v,\omega)\in\Sigma,$$ (6.11)

where the effective mass $m_{\rm eff}$ is a srictly increasing continuous function of $|v|\in[0,1)$ and $|\omega|\in[0,\infty)$.

Now we formulate the definition of the orbital stability of solitons (6.1) in terms of reduced trajectories corresponding to solutions $Y(t)=(A(x,t),\Pi(x,t),q(t),p(t),\pi(t))\in C(\mathbb{R},{\mathbb{Y}})$ of the system (3.1). Introduce the distance

$$d(Y(t),S_{v,\omega}(t)):=\|{\bf Z}(t)-{\bf S}_{v,\omega}\|_{{\mathbb{Z}}}+|\dot{q}(t)-v(Y(t))|+|\omega-\omega(Y(t))|.$$ (6.12)

Here ${\bf Z}(t)=({\bf A}(y,t),{\bm{\Pi}}(y,t),\pi(t))\in C(\mathbb{R},{\mathbb{Z}}),$ where ${\bf A}$, ${\bm{\Pi}}$ are defined according to (4.1), while $v(Y(t))$, $\omega(Y(t))$ are expressed by (5.6). Suppose that the initial data $Y(0)$ is close to $S_{v,\omega}(0)$ in the following sense:

$$d(Y(0),S_{v,\omega}(0))<\varepsilon.$$ (6.13)

Definition 6.5.

The soliton (6.1) is orbitally stable if for any $r>0$ there is $\varepsilon>0$ such that, for any solution $Y(t)$ of the system (3.1) the inequality (6.13) implies that

$$d(Y(t),S_{v,\omega}(t))<r,\qquad t\in\mathbb{R}.$$ (6.14)

Obviously, for $\rho(x)\equiv 0$ all solitons are orbitally stable; see Remark 1.1.

7 The Lyapunov function

To prove the stability of a soliton ${\bf S}_{v,\omega}$, we must construct a Lyapunov function $\Lambda({\bf Z})$ which is an invariant and admits a lower bound:

$$\delta\Lambda:=\Lambda({\bf S}_{v,\omega}+\delta{\bf Z})-\Lambda({\bf S}_{v,\omega})\geq{\varkappa}\|\delta{\bf Z}\|_{{\mathbb{Z}}}^{2},\qquad\delta{\bf Z}\in{\mathbb{Z}},\qquad\|\delta{\bf Z}\|_{\mathbb{Z}}\ll 1$$ (7.1)

with some ${\varkappa}={\varkappa}(v)>0$. We will construct the Lyapunov function as a perturbation of the Hamiltonian ${\bf H}_{v,\omega}:={\bf H}_{P_{v,\omega}}$ by a suitable Casimir invariant.

In this section we calculate $\delta{\bf H}_{v,\omega}={\bf H}_{v,\omega}({\bf S}_{v,\omega}+\delta{\bf Z})-{\bf H}_{v,\omega}({\bf S}_{v,\omega})$ for $\delta{\bf Z}=(\alpha(y),\beta(y),\gamma)\in{\mathbb{Z}}$, where

$${\rm div{\hskip 1.42262pt}}\,\alpha(y)={\rm div{\hskip 1.42262pt}}\,\beta(y)=0$$ (7.2)

according to (2.2). We have

$$\delta{\bf H}_{v,\omega} = \frac{1}{2}\int[|{\bm{\Pi}}_{v,\omega}+\beta|^{2}+|{\rm curl{\hskip 1.42262pt}}({\bf A}_{v,\omega}+\alpha)|^{2}]\,dy-\frac{1}{2}\int[|{\bm{\Pi}}_{v,\omega}|^{2}+|{\rm curl{\hskip 1.42262pt}}{\bf A}_{v,\omega}|^{2}]\,dy$$

$$+\frac{1}{2m}[P_{v,\omega}+\langle{\bm{\Pi}}_{v,\omega}+\beta,\nabla_{*}({\bf A}_{v,\omega}+\alpha)\rangle-\langle{\bf A}_{v,\omega}(y)+\alpha,\rho(y)\rangle]^{2}$$

$$-\frac{1}{2m}[P_{v,\omega}+\langle{\bm{\Pi}}_{v,\omega},\nabla_{*}{\bf A}_{v,\omega}\rangle-\langle{\bf A}_{v,\omega}(y),\rho(y)\rangle]^{2}$$

$$+\frac{1}{2I}[\pi_{v,\omega}+\gamma-\langle y{\rm\wedge}({\bf A}_{v,\omega}(y)+\alpha),\rho(y)\rangle]^{2}-\frac{1}{2I}[\pi_{v,\omega}-\langle y{\rm\wedge}{\bf A}_{v,\omega}(y),\rho(y)\rangle]^{2}.$$

After rearrangements, we obtain

$$\delta{\bf H}_{v,\omega}=\frac{1}{2}\int(|\beta|^{2}+|{\rm curl{\hskip 1.42262pt}}\alpha|^{2})\,dy+\int({\bm{\Pi}}_{v,\omega}\cdot\beta+{\rm curl{\hskip 1.42262pt}}{\bf A}_{v,\omega}\cdot{\rm curl{\hskip 1.42262pt}}\alpha)\,dy$$

$$+\frac{1}{2m}[(p_{v,\omega}+\delta p)^{2}-p_{v,\omega}^{2}]+\frac{1}{2I}[(M_{v,\omega}+\delta M)^{2}-M_{v,\omega}^{2}],$$ (7.3)

where

$$\left\{\begin{array}[]{rcl}p_{v,\omega}&:=&P_{v,\omega}+\langle{\bm{\Pi}}_{v,\omega},\nabla_{*}{\bf A}_{v,\omega}\rangle-\langle{\bf A}_{v,\omega}(y),\rho(y)\rangle=mv\\ \delta p&:=&\langle\beta,\nabla_{*}{\bf A}_{v,\omega}\rangle+\langle{\bm{\Pi}}_{v,\omega},\nabla_{*}\alpha\rangle+\langle\beta,\nabla_{*}\alpha\rangle-\langle\alpha(y),\rho(y)\rangle\\ M_{v,\omega}&:=&\pi_{v,\omega}-\langle y{\rm\wedge}{\bf A}_{v,\omega}(y),\rho(y)\rangle=I\omega\\ \delta M&:=&\gamma-\langle y{\rm\wedge}\alpha,\rho(y)\rangle\end{array}\right|$$ (7.4)

due to (6.3) and (6.2). Using the first two equations of (6.4) and taking into account (7.2), we obtain:

$$\int({\bm{\Pi}}_{v,\omega}\cdot\beta+{\rm curl{\hskip 1.42262pt}}{\bf A}_{v,\omega}\cdot{\rm curl{\hskip 1.42262pt}}\alpha)\,dy=\langle{\bm{\Pi}}_{v,\omega},\beta\rangle-\langle\Delta{\bf A}_{v,\omega},\alpha\rangle$$

$$=-\langle(v\cdot\nabla){\bf A}_{v,\omega},\beta\rangle-\langle{\bm{\Pi}}_{v,\omega},(v\cdot\nabla)\alpha\rangle+\langle v\rho(y)+y{\rm\wedge}\omega\rho(y),\alpha\rangle$$

$$=-v\cdot\delta p-\omega\cdot\delta M+\omega\cdot\gamma+\langle\beta,(v\cdot\nabla)\alpha\rangle.$$

Substituting into (7), after rearrangements, we get $\delta{\bf H}_{v,\omega}={\bf J}_{1}+{\bf J}_{2},$ where

$${\bf J}_{1} = \frac{1}{2}\int(|\beta|^{2}+|\nabla\alpha|^{2})dy+\langle\beta,v\cdot\nabla\alpha\rangle,$$ (7.5)

$${\bf J}_{2} = \frac{m}{2}[(v+\delta p/m)^{2}-v^{2}-2v\cdot\delta p/m]+\frac{I}{2}[(\omega\!+\!\delta M/I)^{2}-\omega^{2}-2\omega\delta M/I]+\omega\cdot\gamma$$ (7.6)

$$= \frac{(\delta p)^{2}}{2m}+\frac{(\delta M)^{2}}{2I}=\frac{(\delta p)^{2}}{2m}+\frac{1}{2I}[\gamma-\langle y{\rm\wedge}\alpha,\rho(y)\rangle]^{2}+\omega\cdot\gamma.$$

As a result,

$$\delta{\bf H}_{v,\omega}=\frac{1}{2}\!\int\!(|\beta|^{2}\!\!+\!|\nabla\alpha|^{2})dy\!+\!\langle\beta,v\!\cdot\!\nabla\alpha\rangle\!+\!\frac{(\delta p)^{2}}{2m}\!+\!\frac{1}{2I}[\gamma\!-\!\langle y{\rm\wedge}\alpha,\rho(y)\rangle]^{2}+\omega\cdot\gamma.$$ (7.7)

The first term ${\bf J}_{1}$ admits a lower bound

$${\bf J}_{1}\geq\frac{1-|v|}{2}\int(|\beta|^{2}+|\nabla\alpha|^{2})\,dy.$$ (7.8)

8 Orbital stability of solitons with $\omega=0$

In the case $\omega=0$, the solitons ${\bf S}_{v,0}$ are critical points of the reduced Hamiltonian ${\bf H}_{v,0}$:

$$D_{\bf A}{\bf H}_{v,0}({\bf S}_{v,0})=0,\qquad D_{\bm{\Pi}}{\bf H}_{v,0}({\bf S}_{v,0})=0,\qquad D_{\pi}{\bf H}_{v,0}({\bf S}_{v,0})=0.$$ (8.1)

Indeed, the first and second identities hold due to the first two equations of (6.4). The last identity follows from (5.4) and (6.2):

$$D_{\pi}{\bf H}_{v,\omega}({\bf S}_{v,\omega})=\frac{1}{I}[\pi_{v,\omega}-\langle y{\rm\wedge}{\bf A}_{v,\omega}(y),\rho(y)\rangle]=\omega.$$ (8.2)

So, in the case $\omega=0$ we can consider the Hamiltonian as the Lyapunov function. Let us prove the lower bound (7.1) for $\Lambda={\bf H}_{v,0}$.

Lemma 8.1.

Let the conditions (1.2) hold and let $|v|<1$. Then

$$\delta{\bf H}_{v,0}:={\bf H}_{v,0}({\bf S}_{v,0}+\delta{\bf Z})-{\bf H}_{v,0}({\bf S}_{v,0})\geq{\varkappa}\|\delta{\bf Z}\|_{{\mathbb{Z}}}^{2},\quad\delta{\bf Z}=(\alpha(y),\beta(y),\gamma)\in{\mathbb{Z}},\quad\|\delta{\bf Z}\|_{\mathbb{Z}}\ll 1$$ (8.3)

with some ${\varkappa}>0$.

Proof.

Due to (7.5)–(7.6) with $\omega=0$, it suffices to check that

$$\frac{1-|v|}{2}\int|\nabla\alpha(y)|^{2}\,dy+\frac{1}{2I}[\gamma-\langle y{\rm\wedge}\alpha(y),\rho(y)\rangle]^{2}\geq{\varkappa}(\|\nabla\alpha\|_{L^{2}}^{2}+\gamma^{2}).$$ (8.4)

Denote $\mu:=\gamma-\langle y{\rm\wedge}\alpha(y),\rho(y)\rangle$. Then (8.4) can be rewritten equivalently as

$$\|\nabla\alpha\|_{L^{2}}^{2}+[\mu+\langle y{\rm\wedge}\alpha(y),\rho(y)\rangle]^{2}\leq C(v,I)(\int|\nabla\alpha(y)|^{2}\,dy+\mu^{2}).$$

It remains to note that the Cauchy–Schwarz inequality gives

$$|\langle y{\rm\wedge}\alpha(y),\rho(y)\rangle|=|\langle\nabla\hat{\rho}(k){\rm\wedge}\hat{\alpha}(k)\rangle|=|\langle\frac{\nabla\hat{\rho}(k)}{|k|}{\rm\wedge}|k|\hat{\alpha}(k)\rangle|\leq\|\nabla\alpha\|_{L^{2}}\Big{[}\int\frac{|\nabla\hat{\rho}(k)|^{2}}{k^{2}}dk\Big{]}^{1/2}.\qquad\quad\qed$$

The lemma implies the orbital stability of solitons with $\omega=0$:

Theorem 8.2.

Let all the conditions (1.2) hold and let $|v|<1$. Then the soliton $S_{v,0}$ is orbitally stable.

Proof.

The proof requires a modification of the general scheme of [5], [15] and [31]. Namely, the inequality (6.13) with sufficiently small $\varepsilon>0$ implies that the momentum $P$ of the solution $Y(t)$ is close to $P_{v,0}$:

$$|P-P_{v,0}|\to 0,\qquad\varepsilon\to 0.$$ (8.5)

It is important, that the map $v\mapsto P_{v,0}$ is a homeomorphism $\mathbb{R}^{3}\to\mathbb{R}^{3}$ by Lemma 6.4. Hence, $P=P_{v_{*},0}$, where

$$|v_{*}-v|\to 0,\qquad\varepsilon\to 0.$$ (8.6)

Combining (8.5) with (6.13), we obtain $\|{\bf Z}(0)-{\bf S}_{P}\|_{{\mathbb{Z}}}\to 0$ as $\varepsilon\to 0.$ Therefore,

$$0\leq H_{P}({\bf Z}(0))-H_{P}({\bf S}_{P})\leq\varepsilon_{1},\qquad\varepsilon_{1}\to 0\,\,\,{\rm as}\,\,\,\varepsilon\to 0.$$

Hence, the conservation of $H_{P}({\bf Z}(t))$ for the system (5.5) with the parameter $P$ implies that

$$0\leq H_{P}({\bf Z}(t))-H_{P}({\bf S}_{P})\leq\varepsilon_{1},\quad t\in\mathbb{R}.$$

Now we apply the lower bound (8.3) with $v=v_{*}$:

$${\bf H}_{P}({\bf S}_{P}+\delta{\bf Z})-{\bf H}_{P}({\bf S}_{P})\geq{\varkappa}\|\delta{\bf Z}\|_{{\mathbb{Z}}}^{2},\qquad\delta{\bf Z}\in{\mathbb{Z}},\qquad\|\delta{\bf Z}\|_{\mathbb{Z}}\ll 1.$$ (8.7)

Setting $\delta{\bf Z}={\bf Z}(0)-{\bf S}_{P}$, we obtain

$$\sup_{t\in\mathbb{R}}\|{\bf Z}(t)-{\bf S}_{P}\|_{{\mathbb{Z}}}\to 0,\qquad\varepsilon\to 0.$$

Finally, combining with (8.6), we get

$$\sup_{t\in\mathbb{R}}\|{\bf Z}(t)-{\bf S}_{v,0}\|_{{\mathbb{Z}}}\to 0,\qquad\varepsilon\to 0.$$ (8.8)

It remains to notice that the last terms on the right-hand side of (6.12) are small uniformly in time:

$$\sup_{t\in\mathbb{R}}[|\dot{q}(t)-v|+|\omega(t)|]\to 0,\qquad\varepsilon\to 0.$$ (8.9)

This follows from (5.6) together with (8.5) and (8.8). ∎

9 Angular ​momentum ​conservation ​for ​reduced ​system ​with ​$P=0$

For $\omega\neq 0$, the solitons ${\bf S}_{v,\omega}$ are not critical points of the reduced Hamiltonian because of (8.2). In this case we will consider the Lyapunov function which is a perturbation of the Hamiltonian ${\bf H}_{v,\omega}$ by a Casimir invariant:

$$\Lambda_{v,\omega}({\bf Z})={\bf H}_{v,\omega}({\bf Z})-\frac{\nu_{\rm eff}}{2}{\pi^{2}},\qquad\nu_{\rm eff}:=\frac{1}{I_{\rm eff}}.$$ (9.1)

This function is an invariant for the reduced system (5.5), and the solitons ${\bf S}_{v,\omega}$ are its critical points since

$${D_{\pi}\Lambda({\bf S}_{v,\omega})=\omega-\nu_{\rm eff}I_{\rm eff}\omega=0}$$

by (8.2) and (6.9). Now (7.7) leads to

$$\delta\Lambda_{v,\omega}(\delta{\bf Z})\!\! \!\!\!:=\!\!\! \!\!\Lambda({\bf S}_{v,\omega}+\delta{\bf Z})-\Lambda({\bf S}_{v,\omega})=\delta{\bf H}_{v,\omega}-\frac{\nu_{\rm eff}}{2}((\pi_{v,\omega}+\gamma)^{2}-\pi_{v,\omega}^{2})$$ (9.2)

$$\!\!\!=\!\!\! \!\!\frac{1}{2}\!\int\!(|\beta|^{2}\!\!+\!|\nabla\alpha|^{2})dy\!+\!\langle\beta,v\!\cdot\!\nabla\alpha\rangle\!+\!\frac{(\delta p)^{2}}{2m}\!+\!\frac{\nu}{2}[\gamma\!-\!\langle y{\rm\wedge}\alpha,\rho(y)\rangle]^{2}\!-\!\frac{\nu_{\rm eff}}{2}\gamma^{2},\quad\nu\!:=\!\frac{1}{I}.\qquad$$

The presence of the last “negative” term makes problematic a lower bound of type (7.1), and perturbations of ${\bf H}_{v,\omega}$ by other Casimir invariants do not help; see Appendix A.

That is why we need to involve an additional invariant of our system: the angular momentum. However, the expression (2.13) is a functional on the phase space ${\mathbb{Z}}$ of the reduced system only in the case of zero total momentum $P=0$. This is why we consider below solutions and solitons of the system (3.1) with the total momentum $P=0$. In this case, the reduced system (5.5) reads

$$\left\{\begin{array}[]{rcl}\dot{\bf A}(y,t)&\!\!=&\!\!{\bm{\Pi}}(y,t)\!+\!(v(t)\cdot\nabla){\bf A}(y,t)\\ \dot{\bm{\Pi}}(y,t)&\!\!=&\!\!\Delta{\bf A}(y,t)+(v(t)\cdot\nabla){\bm{\Pi}}(y,t)+{\bf j}(y,t)\\ \dot{\pi}(t)&\!\!=&\!\!-\pi(t){\rm\wedge}\omega(t)\end{array}\right|,$$ (9.3)

where $v(t)$ and $\omega(t)$ are given by (5.6) with $P=0$. The angular momentum conservation (2.13) with $P=0$ becomes

$$J({\bf Z}(t)):=\langle{\bf A}(y,t){\rm\wedge}{\bm{\Pi}}(y,t)\rangle-\langle(y{\rm\wedge}\nabla)_{*}{\bf A}(y,t),{\bm{\Pi}}(y,t)\rangle+\pi(t)=\mathop{\rm const}\nolimits,\quad t\in\mathbb{R}.$$ (9.4)

Stationary solutions of this system are the solitons ${\bf S}_{v,\omega}$ with $v=0$. This follows from the formula $P_{v,\omega}=m_{\rm eff}(v,\omega)v$ established in [27, (5.17)]. We denote the solutions as ${\bf S}_{\omega}:={\bf S}_{0,\omega}$. Due to the first equation of (6.4) with $v=0$,

$${\bf S}_{\omega}=({\bf A}_{\omega},0,\pi_{\omega})=({\bf A}_{0,\omega},0,\pi_{0,\omega}).$$ (9.5)

Finally, the Hamiltonian (5.4) with $P=0$ becomes

$${\bf H}_{0}({\bf Z}) = \frac{1}{2}\!\int[|{\bm{\Pi}}(y)|^{2}+|{\rm curl{\hskip 1.42262pt}}{\bf A}(y)|^{2}]dy$$ (9.6)

$$+\frac{1}{2m}[\langle{\bm{\Pi}},\nabla_{*}{\bf A}\rangle-\langle{\bf A},\rho\rangle]^{2}+\frac{1}{2I}[\pi-\langle y{\rm\wedge}{\bf A}(y),\rho(y)\rangle]^{2},\quad{\bf Z}=({\bf A},{\bm{\Pi}},\pi).$$

The set ${\mathbb{Z}}_{0}^{\infty}={\mathbb{Z}}\cap[C_{0}^{\infty}\oplus C_{0}^{\infty}\oplus\mathbb{R}^{3}]$ is dense in ${\mathbb{Z}}$ by Remark 2.2. Proposition 2.3 iii) implies the following lemma.

Lemma 9.1.

For solutions to (9.3) with initial state $({\bf A}(y,0),{\bm{\Pi}}(y,0),\pi(0))\in{\mathbb{Z}}_{0}^{\infty}$, we have for any $T>0$, $l=0,1,2$ and $|\alpha|+l\leq 2$, that $\partial_{y}^{\alpha}\partial_{t}^{l}{\bf A}(x,t)\in C^{2-(|\alpha|+l)}(\mathbb{R}^{3}\times[0,T])\!\otimes\!\mathbb{R}^{3}$, and

$$\left\{\begin{array}[]{rcll}|\partial_{t}^{l}{\bf A}(y,t)|&\leq&C(1+|y|)^{-2},&l\leq 2\\ \\ |\partial_{y}^{\alpha}\partial_{t}^{l}{\bf A}(y,t)|&\leq&C(1+|y|)^{-3},&\alpha\neq 0,\,\,|\alpha|+l\leq 2\end{array}\right|,\qquad y\in\mathbb{R}^{3},\,\,\,|t|\leq T,$$ (9.7)

and the angular momentum conservation (9.4) holds.

The conservation (9.4) suggests that the stability of solitons ${\bf S}_{\omega}$ for the nonlinear system (9.3) follows from a lower bound of type (7.1) on the manifold

$${\mathbb{M}}_{\omega}:=\{{\bf Z}\in{\mathbb{Z}}:J({\bf Z})=J({\bf S}_{\omega})\}$$ (9.8)

for the function (9.2) with $v=0$. However, the expression (9.4) is not well-defined on the phase space ${\mathbb{Z}}$, so the manifold ${\mathbb{M}}_{\omega}$ is also not well-defined. On the other hand, the “tangent space” ${\mathbb{T}}_{\omega}=T_{{\bf S}_{\omega}}{\mathbb{M}}_{\omega}$ can be defined as

$${\mathbb{T}}_{\omega}:=\{\delta{\bf Z}\in{\mathbb{Z}}:(E_{\omega,n},\delta{\bf Z})=0,\,\,\,n=1,2,3\},\qquad E_{\omega,n}:=dJ_{n}({\bf S}_{\omega})=(0,b_{\omega,n}(x),1)\in{\mathbb{Z}}^{0},$$ (9.9)

where ${\mathbb{Z}}^{0}:={\cal F}^{0}\oplus{\cal F}^{0}\oplus\mathbb{R}^{3}$. Indeed, according to (9.4),

$$(dJ_{n}({\bf S}_{\omega}),\delta{\bf Z}) = \langle({\bf A}_{\omega}(y){\rm\wedge}\beta(y))_{n}\rangle-\langle(y{\rm\wedge}\nabla)_{n}{\bf A}_{\omega}(y),\beta(y)\rangle+\gamma_{n}$$ (9.10)

$$= \langle(\hat{\bf A}_{\omega}(k){\rm\wedge}\hat{\beta}(k)_{n}\rangle+\langle(\nabla{\rm\wedge}k)_{n}\hat{\bf A}_{\omega}(k),\hat{\beta}(k)\rangle+\gamma_{n},\quad\delta{\bf Z}\!=\!(\alpha,\beta,\gamma)\!\in\!{\mathbb{Z}},\qquad$$

where ${\bf A}_{\omega}(y)$ and $(y{\rm\wedge}\nabla)_{n}{\bf A}_{\omega}(y)$ belong to $L^{2}$ by (6.6) with $v=0$. Moreover,

$$E_{\omega,n}=(0,b_{\omega,n}(x),1)\in{\mathbb{Z}}^{0}\cap[H^{\infty}\oplus H^{\infty}\oplus\mathbb{R}^{3}].$$ (9.11)

Indeed, (1.2) and (6.6) with $v=0$ imply that

$$(1+|k|)^{s}\hat{\bf A}_{\omega}(k)\in L^{2},\quad(1+|k|)^{s}(\nabla{\rm\wedge}k)_{n}\hat{\bf A}_{\omega}(k)\in L^{2},\qquad\forall s>0.$$ (9.12)

Definition 9.2.

Denote by ${\mathbb{T}}_{\omega}^{0}$ the Hilbert space which is the closure of ${\mathbb{T}}_{\omega}\cap{\mathbb{Z}}^{0}$ in ${\mathbb{Z}}^{0}$ endowed with the norm of ${\mathbb{Z}}^{0}$.

The space ${\mathbb{T}}_{\omega}^{0}$ admits also the representation (9.9) with $\delta{\bf Z}\in{\mathbb{Z}}^{0}$. So, ${\mathbb{T}}_{\omega}^{0}$ is a closed linear subspace of codimension three in ${\mathbb{Z}}^{0}$. Denote by $P_{\omega}$ the orthogonal projection of ${\mathbb{Z}}^{0}$ onto ${\mathbb{T}}_{\omega}^{0}$:

$$P_{\omega}=1-\sum_{1}^{3}p_{n}E_{\omega,n}\otimes E_{\omega,n}^{*},$$ (9.13)

where $p_{n}$ are suitable normalization factors.

10 The stability of the linearized dynamics in the tangent space

Let us linearize the reduced system (9.3) at the soliton ${\bf S}_{\omega}$. We write the system as

$$\dot{\bf Z}(t)=F({\bf Z}(t)),\qquad t\in\mathbb{R};\qquad{\bf Z}(t)=({\bf A}(t),{\bm{\Pi}}(t),\gamma(t)).$$ (10.1)

The variational equation is obtained by replacing ${\bf Z}(t)$ with ${\bf Z}(t)+\xi(t)$ and keeping the terms linear in $\xi(t)$. Formally, the variational equation reads as

$$\dot{\bm{\xi}}(t)=dF({\bf Z}(t)){\bm{\xi}}(t).$$ (10.2)

Accordingly, the linearization of (10.1) at the soliton is

$$\dot{\bm{\xi}}(t)=dF({\bf S}_{\omega}){\bm{\xi}}(t).$$ (10.3)

Let us calculate the corresponding Hamilton–Poisson structure for the linearized dynamics. Substitute ${\bf Z}(t)={\bf S}_{\omega}+\xi(t)$ into (5.2). Then (5.3) implies that

$$\dot{\xi}(t)={\mathbb{J}}({\bf S}_{\omega})DH_{0}({\bf S}_{\omega}+\xi(t))+{\mathbb{J}}_{0}({\bf S}_{\omega}+\xi(t))DH_{0}({\bf S}_{\omega}+\xi(t)),\quad{\mathbb{J}}_{0}(\xi(t))\!=\!\!\left(\!\!\begin{array}[]{ccccc}0&0&0\\ 0&0&0\\ 0&0&-\gamma(t){\rm\wedge}\end{array}\!\!\right)\!.$$ (10.4)

We have to keep in the right-hand side of (10.4) only linear terms in $\xi$. Thus, it suffices to keep in $H_{0}({\bf S}_{\omega}+\xi(t))$ only linear and quadratic terms. Hence, we can replace $H_{0}({\bf S}_{\omega}+\xi)$ with $\xi=(\alpha,\beta,\gamma)\in{\mathbb{Z}}$ by its quadratic part: according to (7.7) and (7.4),

$$\tilde{H}_{\omega}(\xi)=\frac{1}{2}\!\int\!(|\beta|^{2}\!\!+\!|\nabla\alpha|^{2})dy\!+\frac{1}{2m}{(\langle\beta,\nabla_{*}{\bf A}_{0,\omega}\rangle\!-\!\langle\alpha(y),\rho(y)\rangle)^{2}}\!+\!\frac{\nu}{2}[\gamma\!-\!\langle y{\rm\wedge}\alpha,\rho(y)\rangle]^{2}+\omega\cdot\gamma.$$ (10.5)

Note that the last linear term contributes only to the last component of the differential,

$$D_{\gamma}\tilde{H}_{0}(\xi)=\nu[\gamma\!-\!\langle y{\rm\wedge}\alpha,\rho(y)\rangle]+\omega.$$

Moreover, due to the structures of the matrices ${\mathbb{J}}({\bf S}_{\omega})$ and ${\mathbb{J}}_{0}(\xi)$, this component contributes only to the equation for $\dot{\gamma}(t)$:

$$\dot{\gamma}(t) = -\pi_{\omega}{\rm\wedge}(\nu[\gamma\!-\!\langle y{\rm\wedge}\alpha,\rho(y)\rangle]+\omega)-\gamma{\rm\wedge}(\nu[\gamma\!-\!\langle y{\rm\wedge}\alpha,\rho(y)\rangle]+\omega)$$ (10.6)

$$= -\nu\pi_{\omega}{\rm\wedge}[\gamma\!-\!\langle y{\rm\wedge}\alpha,\rho(y)\rangle]-\gamma{\rm\wedge}(\nu[\!-\!\langle y{\rm\wedge}\alpha,\rho(y)\rangle]+\omega)$$

since $\pi_{\omega}{\rm\wedge}\omega=0$ by (6.9). The corresponding linearized equation is

$$\dot{\bm{\gamma}}(t)=-\nu\pi_{\omega}{\rm\wedge}[{\bm{\gamma}}\!-\!\langle y{\rm\wedge}{\bm{\alpha}},\rho(y)\rangle]-{\bm{\gamma}}{\rm\wedge}\omega.$$ (10.7)

Now it is clear that the linearized equation (10.3) reads as

$$\dot{\bm{\xi}}(t)={\mathbb{J}}({\bf S}_{\omega})D\tilde{H}_{0}({\bm{\xi}}(t))-\left(\!\!\!\begin{array}[]{c}0\\ 0\\ {\bm{\gamma}}{\rm\wedge}\omega\end{array}\!\!\!\right),\qquad{\bm{\xi}}(t)=({\bm{\alpha}}(t),{\bm{\beta}}(t),{\bm{\gamma}}(t))\in{\mathbb{Z}}.$$ (10.8)

Now we are going to prove that the tangent space ${\mathbb{T}}_{\omega}$ is invariant with respect to the linearized dynamics (10.3). As the first step, we prove that the vector field $dF({\bf S}_{\omega}){\bm{\xi}}$ is tangent to ${\mathbb{T}}_{\omega}$.

Lemma 10.1.

For ${\bm{\xi}}\in{\mathbb{Z}}$, one has

$$(dJ_{n}({\bf S}_{\omega}),dF({\bf S}_{\omega}){\bm{\xi}})=0,\qquad n=1,2,3.$$ (10.9)

Proof.

Let us consider solutions ${\bf Z}(t)$ to (10.1) with initial states ${\bf Z}(0)\in{\mathbb{Z}}_{0}^{\infty}$. Then the angular momentum conservation (9.4) holds by Lemma 9.1. Differentiating (9.4) in time, we obtain by (9.11) that $(dJ_{n}({\bf Z}(t)),\dot{\bf Z}(t))=(dJ_{n}({\bf Z}(t)),F({\bf Z}(t)))=0$ for $t\in\mathbb{R}.$ Therefore,

$$(dJ_{n}({\bf Z}),F({\bf Z}))=0,\qquad{\bf Z}\in{\mathbb{Z}}_{0}.$$ (10.10)

Taking into account (9.11) and the density of ${\mathbb{Z}}_{0}^{\infty}$ in ${\mathbb{Z}}$, we conclude that the identity (10.10) holds for ${\bf Z}={\bf S}_{\omega}+\varepsilon{\bm{\xi}}$ with any vector ${\bm{\xi}}\in{\mathbb{Z}}$. Differentiating the obtained identity in $\varepsilon$ at $\varepsilon=0$, we get

$$([L_{{\bm{\xi}}}dJ_{n}]({\bf S}_{\omega}),F({\bf S}_{\omega}))+(dJ_{n}({\bf S}_{\omega}),[L_{{\bm{\xi}}}F]({\bf S}_{\omega}))=0,$$ (10.11)

where $L_{{\bm{\xi}}}$ stands for the operator of differentiation. However, $F({\bf S}_{\omega})=0$, so (10.9) follows since $dF({\bf S}_{\omega}){\bm{\xi}}:=[L_{{\bm{\xi}}}F]({\bf S}_{\omega})$. ∎

The following theorem on stability is the main result of the present paper.

Theorem 10.2.

Let conditions (1.2) hold. Then
i) For every ${\bm{\xi}}(0)\in{\mathbb{T}}_{\omega}$ there exists a unique solution ${\bm{\xi}}(t)\in C(\mathbb{R},{\mathbb{T}}_{\omega})$ to (10.8), and the map ${\bm{\xi}}(0)\mapsto{\bm{\xi}}(t)$ is continuous in ${\mathbb{T}}_{\omega}$ for $t\in\mathbb{R}$.
ii) There is the uniform bound

$$\|\xi(t)\|_{\mathbb{Z}}\leq C\|\xi(0)\|_{\mathbb{Z}},\qquad t\in\mathbb{R}.$$ (10.12)

Proof.

It suffices to prove (10.12) as an priori estimate; then the theorem will follow by the same methods as Proposition 2.3. The estimate holds due to the conservation of an appropriate “energy functional”. Let us show that the functional can be chosen as the quadratic part of the Lyapunov function (9.1) with $v=0$:

$$L_{\omega}({\bm{\xi}})=\tilde{H}_{0}({\bm{\xi}})-\frac{\nu_{\rm eff}}{2}{{\bm{\gamma}}^{2}}.$$ (10.13)

Indeed, (11.1) shows that the needed lower bound holds. It suffices to prove that the functional is conserved under the linearized dynamics (10.8). Assuming “a priori” the existence of the solution and differentiating $L_{\omega}(t)=L_{\omega}({\bm{\xi}}(t))$, we obtain by (10.8) and (10.7) that

$$\dot{L}_{\omega}(t) = (DL_{\omega}({\bm{\xi}}(t)),\dot{\bm{\xi}}(t))$$ (10.20)

$$= \Big{(}D\tilde{H}_{0}({\bm{\xi}}(t))-\left(\!\!\!\begin{array}[]{c}0\\ 0\\ \nu_{\rm eff}{\bm{\gamma}}\end{array}\!\!\!\right),{\mathbb{J}}({\bf S}_{\omega})D\tilde{H}_{0}({\bm{\xi}}(t))\Big{)}-\Big{(}DL_{\omega}({\bm{\xi}}(t)),\left(\!\!\!\begin{array}[]{c}0\\ 0\\ {\bm{\gamma}}{\rm\wedge}\omega\end{array}\!\!\!\right)\Big{)}$$

$$= -\nu_{\rm eff}{\bm{\gamma}}\cdot[-\pi_{\omega}{\rm\wedge}D_{\gamma}\tilde{H}_{0}({\bm{\xi}}(t))]-[D_{\gamma}\tilde{H}_{0}({\bm{\xi}}(t))-\nu_{\rm eff}\gamma]\cdot({\bm{\gamma}}{\rm\wedge}\omega)=0.\quad\quad\qed$$ (10.21)

Finally, the conservation of $L_{\omega}$ provides (10.12) by the following proposition.

Proposition 10.3.

The Lyapunov function $L_{\omega}$ admits a lower bound: for some ${\varkappa}>0$,

$$L_{\omega}({\bm{\xi}})\geq{\varkappa}\|{\bm{\xi}}\|_{\mathbb{Z}}^{2},\qquad{\bm{\xi}}\in{\mathbb{T}}_{\omega}.$$ (10.22)

This theorem is proved in the remaining part of the paper.

Remark 10.4.

The calculation (10.20) shows that the “linearized” Lyapunov function (10.13) is conserved, but the terms $\tilde{H}_{\omega}({\bm{\xi}})$ and $\frac{\nu_{\rm eff}}{2}{{\bm{\gamma}}^{2}}$ do not. This fact is contrary to the case of the “nonlinear” Lyapunov function (9.1), where both terms are conserved quantities. This fact is a peculiarity of the energy-Casimir method for the linearized system.

11 Lower bound for the Lyapunov function on the tangent space

In this section we prove the lower bound (10.22). We will prove a stronger bound than (10.22):

$${\cal Q}_{\rho}({\bm{\xi}}) := \frac{1}{2}\int[|\beta(y)|^{2}+|\nabla\alpha(y)|^{2}]dy+\frac{\nu}{2}[\gamma-\langle y{\rm\wedge}\alpha,\rho(y)\rangle]^{2}-\frac{\nu_{\rm eff}}{2}\gamma^{2}$$ (11.1)

$$\geq {\varkappa}\Big{[}\frac{1}{2}\int[|\beta(y)|^{2}+|\nabla\alpha(y)|^{2}]dy+\gamma^{2}\Big{]},\qquad{\bm{\xi}}=(\alpha,\beta,\gamma)\in{\mathbb{T}}_{\omega}.$$

The quadratic form ${\cal Q}_{\rho}$ is continuous on the Hilbert space ${\mathbb{T}}_{\omega}$ endowed with the norm of ${\mathbb{Z}}$. Hence, it suffices to prove the bound (11.1) on the space of smooth functions ${\mathbb{T}}_{\omega}^{\infty}:={\mathbb{T}}_{\omega}\cap{\mathbb{Z}}^{\infty}_{0}$ which is dense in ${\mathbb{T}}_{\omega}$. For the smooth functions, the bound (11.1) can be written as

$$({\bm{\xi}},Q_{\rho}{\bm{\xi}})\geq{\varkappa}({\bm{\xi}},Q_{0}{\bm{\xi}}),\qquad{\bm{\xi}}\in{\mathbb{T}}_{\omega}^{\infty},$$ (11.2)

where the brackets $(\cdot,\cdot)$ denote the inner product in the Hilbert space ${\mathbb{Z}}^{0}$, and

$$Q_{\rho}=\left(\begin{array}[]{ccc}-\Delta+\nu|m^{*}(x)\rho(x)\rangle\langle\rho(y)m(y)|&0&-\nu|m^{*}(x)\rho(x)\rangle\\ 0&1&0\\ -\nu\langle\rho(y)m(y)|&0&\delta\end{array}\right),\quad Q_{0}=\left(\!\!\!\begin{array}[]{ccc}-\Delta&0&0\\ 0&1&0\\ 0&0&\delta\end{array}\!\!\right).$$ (11.3)

Here $\delta:=\nu-\nu_{\rm eff}>0$ by (6.10), and $m(x)$ denotes the operator $\alpha\mapsto x{\rm\wedge}\alpha$, $\alpha\in\mathbb{R}^{3}$, with the corresponding skew-symmetric matrix

$$m(x)=\left(\begin{array}[]{ccc}0&-x_{3}&x_{2}\\ x_{3}&0&-x_{1}\\ -x_{2}&x_{1}&0\end{array}\right),\qquad x\in\mathbb{R}^{3}.$$

The operator $Q_{0}$ is positive-definite and selfadjoint in ${\mathbb{Z}}^{0}$ with domain ${\mathbb{D}}^{0}$. Hence, $Q_{\rho}$ is also a selfadjoint operator in ${\mathbb{Z}}^{0}$ with the same domain since the difference $Q_{\rho}-Q_{0}$ is a bounded finite rank operator in ${\mathbb{Z}}^{0}$ by (11.3) and (1.2). The common domain of both operators $Q_{\rho}$ and $Q_{0}$ contains the space ${\mathbb{Z}}_{0}^{\infty}={\mathbb{Z}}\cap[C_{0}^{\infty}\oplus C_{0}^{\infty}\oplus\mathbb{R}^{3}]\subset{\mathbb{Z}}^{0}.$ The space is dense in ${\mathbb{Z}}^{0}$ by Remark 2.2. In the next section, we will prove the following proposition.

Proposition 11.1.

Let conditions (1.2) hold. Then
i) the operator $Q_{\rho}$ is nonnegative:

$$({\bm{\xi}},Q_{\rho}{\bm{\xi}})\geq 0,\qquad{\bm{\xi}}\in{{\mathbb{Z}}_{0}^{\infty}.}$$ (11.4)

ii) $\dim\ker Q_{\rho}=3$ and $\ker Q_{\rho}\cap{\mathbb{T}}_{\omega}^{0}=0$.

Proof of Proposition 10.3. Consider the restrictions of the quadratic forms $({\bm{\xi}},Q_{\rho}{\bm{\xi}})$ and $({\bm{\xi}},Q_{0}{\bm{\xi}})$ onto ${\mathbb{T}}_{\omega}^{\infty}:={\mathbb{Z}}_{0}^{\infty}\cap{\mathbb{T}}_{\omega}$ and denote by $\tilde{Q}_{\rho}$ and $\tilde{Q}_{0}$ the corresponding nonnegative selfadjoint operators in ${\mathbb{T}}_{\omega}^{0}$:

$$({\bm{\xi}},\tilde{Q}_{\rho}{\bm{\xi}})=({\bm{\xi}},Q_{\rho}{\bm{\xi}}),\qquad({\bm{\xi}},\tilde{Q}_{0}{\bm{\xi}})=({\bm{\xi}},Q_{0}{\bm{\xi}}),\qquad{\bm{\xi}}\in{\mathbb{T}}_{\omega}^{\infty}.$$ (11.5)

The space ${\mathbb{T}}_{\omega}^{\infty}$ is dense in ${\mathbb{T}}_{\omega}^{0}$ and the operators can be expressed as

$$\tilde{Q}_{\rho}=P_{\omega}Q_{\rho},\qquad\tilde{Q}_{0}=P_{\omega}Q_{0},$$ (11.6)

where $P_{\omega}$ is the projection (9.13).

Lemma 11.2.

Let conditions (1.2) hold. Then $\tilde{Q}_{\rho}\gg\mu\tilde{Q}_{0}$ with sufficiently small $\mu>0$, i.e.

$$\|\tilde{Q}_{\rho}{\bm{\xi}}\|_{{\mathbb{Z}}^{0}}\geq\mu\|\tilde{Q}_{0}{\bm{\xi}}\|_{{\mathbb{Z}}^{0}},\qquad{\bm{\xi}}\in{\mathbb{T}}_{\omega}^{\infty}.$$ (11.7)

Proof.

We have $\tilde{Q}_{0}=\tilde{Q}_{0}\tilde{Q}_{\rho}^{-1}\tilde{Q}_{\rho}$. Hence, it suffices to prove that the operator $\tilde{Q}_{0}\tilde{Q}_{\rho}^{-1}:{\mathbb{T}}_{\omega}^{0}\to{\mathbb{T}}_{\omega}^{0}$ is bounded since then $\|\tilde{Q}_{0}{\bm{\xi}}\|_{{\mathbb{Z}}^{0}}\leq C\|\tilde{Q}_{\rho}{\bm{\xi}}\|_{{\mathbb{Z}}^{0}}$ for ${\bm{\xi}}\in{\mathbb{T}}_{\omega}^{\infty}$ which is equivalent to (11.7).

The difference $Q_{\rho}-Q_{0}$ is a finite rank bounded operator in ${\mathbb{Z}}^{0}$. Hence, $\tilde{Q}_{\rho}-\tilde{Q}_{0}$ is a finite rank bounded operator in ${\mathbb{T}}_{\omega}^{0}$ by (11.6) and (9.13) since $E_{\omega,n}\in{\mathbb{Z}}^{0}$ by (9.9). Therefore, the selfadjoint operators $\tilde{Q}_{\rho}$ and $\tilde{Q}_{0}$ have common domain $\tilde{\mathbb{D}}^{0}$. Moreover, Proposition 11.1 ii) implies that $\tilde{Q}_{\rho}$ is injective, and hence it is a bijection of $\tilde{\mathbb{D}}^{0}$ onto ${\mathbb{T}}_{\omega}^{0}$. Therefore, the operator $\tilde{Q}_{0}\tilde{Q}_{\rho}^{-1}$ is well-defined on the entire Hilbert space ${\mathbb{T}}_{\omega}^{0}$, so the operator is bounded by the closed graph theorem. ∎

By Proposition 11.1, $\tilde{Q}_{\rho}$ is a selfadjoint nonnegative operator in ${\mathbb{T}}_{\omega}^{0}$, similarly to $\tilde{Q}_{0}$. Hence, (11.7) and the Heinz inequality [21, Theorem 2] imply that $\tilde{Q}_{\rho}^{s}\gg\mu^{s}\tilde{Q}^{s}_{0}$ for all $s\in[0,1]$, i.e.,

$$\|\tilde{Q}_{\rho}^{s}{\bm{\xi}}\|_{{\mathbb{Z}}^{0}}\geq\mu^{s}\|\tilde{Q}_{0}^{s}{\bm{\xi}}\|_{{\mathbb{Z}}^{0}},\qquad{\bm{\xi}}\in{\mathbb{T}}_{\omega}^{\infty}.$$ (11.8)

Finally, this bound with $s=1/2$ obviously implies (11.2) with ${\varkappa}=\mu$. $\hfill\Box$

Remark 11.3.

Proposition 11.1 eliminates the point spectrum at zero of the operator $\tilde{Q}_{\rho}$ in the tangent space ${\mathbb{T}}_{\omega}^{0}$. However, this fact alone does not imply the bound (11.1) because of presence of the continuous spectrum of $\tilde{Q}_{\rho}$ which extends up to the point zero. The application of the Heinz inequality allows us to establish the equivalence of continuous spectra of $\tilde{Q}_{\rho}$ and $\tilde{Q}_{0}$.

12 The positivity of the Lyapunov function on the tangent space

Here we prove Proposition 11.1. It suffices to prove the nonnegativity (11.4) for $\beta=0$:

$${\cal B}(Y)\!:=\!\!\int\!|\nabla\alpha(y)|^{2}dy\!+\!\nu[\gamma\!-\!\langle m(y)\alpha(y),\rho(y)\rangle]^{2}\!-\!\nu_{\rm eff}\gamma^{2}\geq 0,\quad Y\!=\!(\alpha,\gamma)\!\in\!{\mathbb{Y}}^{\infty}\!:=\![\dot{\cal F}^{1}\!\cap\!C_{0}^{\infty}]\oplus\mathbb{R}^{3}.$$

This form can be written as ${\cal B}(Y)=(Y,BY)$ for $Y\in{\mathbb{Y}}^{\infty},$ where the brackets $(\cdot,\cdot)$ now denote the inner product in the Hilbert space ${\mathbb{Y}}^{0}:={\cal F}^{0}\oplus\mathbb{R}^{3}$ and

$$B=\left(\begin{array}[]{cc}-\Delta+\nu|m^{*}\rho(x)\rangle\langle\rho(y)m|&-\nu|m^{*}\rho(x)\rangle\\ -\nu\langle\rho m(y)|&\delta\end{array}\right)$$ (12.1)

by (11.3). Proposition 11.1 i) follows from the next lemma.

Lemma 12.1.

Let conditions (1.2) hold. Then the operator $B$ is nonnegative:

$$(Y,BY)\geq 0,\qquad Y\in{\mathbb{Y}}^{\infty}.$$ (12.2)

Proof.

For $\alpha\in\dot{\cal F}^{1}\cap C_{0}^{\infty}$, the action of the Laplacian $\Delta$ in the sense of distributions coincide with the Friedrichs closure of $\Delta$ in the invariant space ${\cal F}^{0}$. Hence, $B$ in (12.2) can be considered as a selfadjoint operator in ${\mathbb{Y}}^{0}$ with domain $D(B)$. Therefore, it suffices to show that for $\lambda<0$ the resolvent $R(\lambda)=(B-\lambda)^{-1}:{\mathbb{Y}}^{0}\to{\mathbb{Y}}^{0}$ is a bounded operator. The operator $B-\lambda$ is selfadjoint in ${\mathbb{Y}}^{0}$. Hence, it remains to check that

$$\ker(B-\lambda)=0,\qquad\lambda<0.$$ (12.3)

The calculation of the kernel reduces to a finite-dimensional problem in a standard way. Indeed, the equation $(B-\lambda)Y=0$ with $Y=(\alpha,\gamma)\in D(B)$ reads as

$$\left\{\begin{array}[]{rcl}(-\Delta-\lambda)\alpha+|\nu m^{*}\rho\rangle\langle\rho m|\alpha\rangle-\nu m^{*}\rho\gamma&=&0\\ -\nu\langle\rho m|\alpha\rangle+(\delta-\lambda)\gamma&=&0\end{array}\right|.$$ (12.4)

Using the last equation, we rewrite the first one as

$$(-\Delta-\lambda)\alpha(x) = -m^{*}(x)\rho(x)(\delta-\lambda)\gamma+\nu m^{*}(x)\rho(x)\gamma=m^{*}(x)\rho(x)(\nu_{\rm eff}+\lambda)\gamma,\,\,\lambda\in\mathbb{R}.\qquad\qquad$$ (12.5)

Hence, on the Fourier transform side,

$$\hat{\alpha}(k)=(k^{2}-\lambda)^{-1}[m^{*}(-i\nabla)\hat{\rho}(k)(\nu_{\rm eff}+\lambda)\gamma],\qquad\lambda\leq 0.$$ (12.6)

Let us emphasise that the formula holds for $\lambda=0$ since $[m^{*}(-i\nabla)\hat{\rho}](0)=0$.

Remark 12.2.

For any $w\in\mathbb{R}^{3}$, the vector field $m^{*}(x)\rho(x)w$ is divergence-free by the spherical symmetry (1.2).

Substituting (12.6) into the second equation of (12.4), we obtain:

$$A(\lambda)\gamma=0,\quad A_{j,l}(\lambda):=-\nu\langle m(-i\nabla_{j})\hat{\rho}(k),(k^{2}-\lambda)^{-1}[m^{*}(-i\nabla_{l})\hat{\rho}(k)]\rangle(\nu_{\rm eff}+\lambda)+(\delta-\lambda).$$ (12.7)

The matrix $A(\lambda)$ is Hermitian for $\lambda\leq 0$, so it is defined uniquely by its quadratic form:

$$\gamma\cdot(A(\lambda)\gamma) = -\nu\langle\nabla\hat{\rho}(k){\rm\wedge}\gamma,(k^{2}-\lambda)^{-1}\nabla\hat{\rho}(k){\rm\wedge}\gamma\rangle(\nu_{\rm eff}+\lambda)+(\delta-\lambda)\gamma^{2}$$

$$= -\nu\gamma^{2}\int\frac{|\nabla\hat{\rho}(k)|^{2}\sin^{2}\widehat{(k,\gamma)}}{k^{2}-\lambda}dk(\nu_{\rm eff}+\lambda)+(\delta-\lambda)\gamma^{2}=-a_{-}(\lambda)\gamma^{2}+a_{+}(\lambda)\gamma^{2}.$$

Thus, the matrix $A(\lambda)$ is the scalar $-a_{-}(\lambda)+a_{+}(\lambda)$, and (12.3) is equivalent to

$$a_{-}(\lambda)\neq a_{+}(\lambda),\qquad\lambda<0.$$ (12.8)

We note that the function $a_{-}(\lambda)$ is strictly increasing while $a_{+}(\lambda)$ is strictly decreasing for $\lambda\leq 0$, and that both functions are continuous. Hence, (12.8) holds since $a_{-}(0)=a_{+}(0)$. Indeed, the last equality is equivalent to

$$\nu\nu_{\rm eff}\int\frac{|\nabla\hat{\rho}(k)|^{2}\sin^{2}\widehat{(k,\gamma)}}{k^{2}}dk=\delta=\nu-\nu_{\rm eff}=\nu\nu_{\rm eff}\delta I,$$ (12.9)

which holds since $\displaystyle\int\frac{|\nabla\hat{\rho}(k)|^{2}\sin^{2}\widehat{(k,\gamma)}}{k^{2}}dk=\delta I$ by (6.9) with $v=0$. ∎

Proof of Proposition 11.1. The nonnegativity (11.4) follows from Lemma 12.1. By (11.3), $\ker Q_{\rho}=\{(\alpha_{\gamma},0,\gamma)\}$, where $\gamma\in\mathbb{R}^{3}$, and $\alpha_{\gamma}$ is defined by (12.6) with $\lambda=0$:

$$\hat{\alpha}_{\gamma}(k)=\nu_{\rm eff}|k|^{-2}m^{*}(-i\nabla)\hat{\rho}(k)\gamma,$$ (12.10)

It is important that $\alpha_{\gamma}\in\dot{\cal F}^{1}$ since $[m^{*}(-i\nabla)\hat{\rho}](0)=0$. Hence, $\dim\ker Q_{\rho}=3$. Finally, (9.9) and (9.10) imply that $\gamma=0$ for $(\alpha_{\gamma},0,\gamma)\in\ker Q_{\rho}\cap{\mathbb{T}}_{\omega}^{0}$. Then also $\alpha_{\gamma}=0$ by (12.10). As a result, $\ker Q_{\rho}\cap{\mathbb{T}}_{\omega}^{0}=0$.

Appendix A Comments

I. The lower bound (11.1) for the quadratic form ${\cal Q}_{\rho}$ implies the same bound on the tangent space for the Lyapunov function $\Lambda_{0,\omega}$ defined by (9.1) since $\Lambda_{0,\omega}={\cal Q}_{\rho}+\frac{(\delta p)^{2}}{2m}$. However, this bound does not provide the stability for the nonlinear dynamics because the bound should hold on the manifold (9.8) which is not well-defined.
II. The bound (11.2) follows from the estimate (11.8) with $s=1/2$ for the square roots $\tilde{Q}^{1/2}$ and $\tilde{Q}_{0}^{1/2}$ which are pseudodifferential operators. Our method of proof of Theorem 10.2 reduces the problem to calculations with differential operators and allows us to avoid calculations of pseudodifferential ones.
III. The orbital stability of solitons $S_{v,\omega}$ with $\omega\neq 0$ and $|v|<1$ for the nonlinear dynamics remains an open question. We suppose that the solitary manifold ${\cal S}=\{S_{v,\omega}:(v,\omega)\in\Sigma\}$ is a global attractor of the system (1.1), and at least the submanifold ${\cal S}^{\|}$ of the solitons with $v\|\omega$ is asymptotically stable in a local convergence. We also expect that an adiabatic effective dynamics similar to [30, 32] holds for solutions close to ${\cal S}^{\|}$ in presence of a weak external magnetic field. We suppose that this dynamics explains the precession of angular momentum of soliton-like solutions in the magnetic field orthogonal to the velocity. Note that the precession of quantum spin was observed experimentally and shows that the directions of the spin and velocity coincide [39, p. 115], [10].

Appendix B On the soliton parameters

Here we prove Lemma 6.2.

ad i). Substituting (6.6) into (6.2), we obtain:

$$\pi_{v,\omega}\!\! \!\!=\!\! \!\!\omega-\!\langle{\bf A}_{v,\omega}(y){\rm\wedge}y\rho(y)\rangle=I\omega-\!\langle\frac{\omega{\rm\wedge}\nabla\hat{\rho}(k)}{k^{2}\!-\!(v\cdot k)^{2}}{\rm\wedge}\nabla\hat{\rho}(k)\rangle=I\omega-\!\int\frac{(\omega{\rm\wedge}k){\rm\wedge}k}{k^{2}-(v\cdot k)^{2}}\frac{|\nabla\rho(k)|^{2}\,dk}{k^{2}}$$ (B.1)

$$\!\!=\!\! \!\!\omega\Big{(}I+\!\int\frac{|\nabla\rho(k)|^{2}\,dk}{k^{2}-(v\cdot k)^{2}}\Big{)}-\!\int\frac{k(\omega\cdot k)|\nabla\rho(k)|^{2}}{k^{2}(k^{2}-(v\cdot k)^{2})}dk=\omega\Big{(}I+\!\int\frac{|\nabla\rho(k)|^{2}\,dk}{k^{2}-(v\cdot k)^{2}}\Big{)}-q_{v,\omega},$$

where we denote $\check{\rho}_{1}(|k|):=\hat{\rho}(k)$. We consider the two cases $\omega\|v$ and $\omega\bot v$ separately.
The case $\omega\|v$. We may assume that $\omega=(|\omega|,0,0)$ and $v=(\pm|v|,0,0)$. Then

$$q_{v,\omega}:=\int\frac{k(\omega\cdot k)|\nabla\rho(k)|^{2}}{k^{2}(k^{2}-(v\cdot k)^{2})}dk=|\omega|\int\frac{k_{1}k|\nabla\rho(k)|^{2}}{k^{2}(k^{2}-\!v^{2}k_{1}^{2})^{2})}dk=|\omega|e_{1}\int\frac{k_{1}^{2}|\nabla\rho(k)|^{2}}{k^{2}(k^{2}-v^{2}k_{1}^{2})}dk.$$ (B.2)

Substituting into (B.1), we obtain:

$$\pi_{v,\omega}=I_{\rm eff}^{\|}\omega,\qquad I_{\rm eff}^{\|}=I_{\rm eff}^{\|}(v)=I+\langle\frac{k_{2}^{2}+k_{3}^{2}}{k^{2}(k^{2}\!-\!v^{2}k_{1}^{2})},|\nabla\rho(k)|^{2}\rangle.$$ (B.3)

The case $\omega\bot v$. We may assume that $\omega=(|\omega|,0,0)$ and $v=(0,|v|,0)$. In this case,

$$q_{v,\omega}=|\omega|e_{1}\int\frac{k_{1}^{2}|\nabla\rho(k)|^{2}}{k^{2}(k^{2}-v^{2}k_{2}^{2})}dk,\quad\pi_{v,\omega}=I_{\rm eff}^{\bot}\omega,\quad I_{\rm eff}^{\bot}=I+\langle\frac{k_{2}^{2}+k_{3}^{2}}{k^{2}(k^{2}\!-\!v^{2}k_{2}^{2})},|\nabla\rho(k)|^{2}\rangle$$ (B.4)

by (B.1). Now Lemma 6.2 i) is proved.
ad ii). Now we consider the case $(v,\omega)\not\in\Sigma$. We may assume that $v=(|v|,0,0)$ with $|v|\neq 0$ and $\omega=(\omega_{1},\omega_{2},0)$, where $\omega_{1}\neq 0$ and $\omega_{2}\neq 0$. In this case,

$$q_{v,\omega}=\int\begin{pmatrix}k_{1}^{2}\omega_{1}\\ k_{2}^{2}\omega_{2}\\ 0\end{pmatrix}\frac{|\nabla\rho(k)|^{2}dk}{k^{2}(k^{2}-v^{2}k_{1}^{2})}=\begin{pmatrix}\omega_{1}\alpha_{1}\\ \omega_{2}\alpha_{2}\\ 0\end{pmatrix},\quad\alpha_{j}=\int\frac{k_{j}^{2}|\nabla\rho(k)|^{2}dk}{k^{2}(k^{2}-v^{2}k_{1}^{2})},~{}~{}j=1,2.$$ (B.11)

It remains to show that $\alpha_{1}>\alpha_{2}$ since then $q_{v,\omega}\nparallel\omega$, so (6.8) breaks down by (B.1). Calculating in spherical coordinates, we obtain:

$$\alpha_{1}=\frac{2\pi C_{\rho}}{|v|^{3}}\Big{(}\ln\frac{1+\!|v|}{1-\!v}-2|v|\Big{)},\qquad\alpha_{2}=\frac{\pi C_{\rho}}{|v|^{3}}\Big{(}(v^{2}-1)\ln\frac{1+\!|v|}{1-\!|v|}+2|v|\Big{)},$$

where $C_{\rho}>0$ due to the last condition of (1.2). Hence, for $|v|<1$, we obtain:

$$\alpha_{1}-\alpha_{2}=\frac{\pi C_{\rho}}{|v|^{3}}\Big{(}(3-v^{2})\ln\frac{1+|v|}{1-|v|}-6|v|\Big{)}=\pi C_{\rho}\sum\limits_{k=2}^{\infty}\frac{8(k-1)}{(2k+1)(2k-1)}v^{2(k-2)}>0.$$

Remark B.1.

The above inequality and the formulas (B.3)-(B.4) imply that $I_{\rm eff}^{\bot}>I_{\rm eff}^{\|}$ for $\rho(x)\not\equiv 0$.

Appendix C Density of $C_{0}^{\infty}$ in the space of divergence-free vector fields

Here we prove that $C_{0}^{\infty}$ is dense in the Hilbert space ${\cal F}^{0}\!:=\!\{A\!\in\!L^{2}\!:{\rm div{\hskip 1.42262pt}}A(x)\!\equiv\!0\}$. For any $A\in{\cal F}^{0}$, its Fourier transform satisfies $\hat{A}(k)\bot k$ for $k\in\mathbb{R}^{3}$. Hence, $\hat{A}(k)=k{\rm\wedge}\hat{a}(k)$ with $\hat{a}(k)=-\frac{k{\rm\wedge}\hat{A}(k)}{k^{2}}$. For $m\in\mathbb{N}$, let us denote

$$\hat{B}_{m}(k)=\left\{\begin{array}[]{ll}\hat{A}(k),&|k|>\frac{1}{m}\\ 0,&|k|<\frac{1}{m}\end{array}\right|,\qquad\hat{b}_{m}(k)=-\frac{k{\rm\wedge}\hat{B}_{m}(k)}{k^{2}}.$$

Then

$$\hat{B}_{m}(k)=k{\rm\wedge}\hat{b}_{m}(k)\stackrel{{\scriptstyle L^{2}}}{{\longrightarrow}}\hat{A}(k),\qquad m\to\infty.$$ (C.1)

Note that $(1+|k|)\hat{b}_{m}(k)\in L^{2}$. Hence, $b_{m}\in H^{1}$, so there exist $b_{m,n}\in C_{0}^{\infty}$ such that $b_{m,n}\stackrel{{\scriptstyle H^{1}}}{{\longrightarrow}}b_{m}$ as $n\to\infty$ for every $m$. Therefore,

$$A_{m,n}:={\rm curl{\hskip 1.42262pt}}b_{m,n}\stackrel{{\scriptstyle L^{2}}}{{\longrightarrow}}{\rm curl{\hskip 1.42262pt}}b_{m},\quad n\to\infty\qquad{\rm and}\quad{\rm curl{\hskip 1.42262pt}}b_{m}=B_{m}\stackrel{{\scriptstyle L^{2}}}{{\longrightarrow}}A,\quad m\to\infty.$$

Hence, there exists a subsequence $A_{m,n(m)}\stackrel{{\scriptstyle L^{2}}}{{\longrightarrow}}A$ as $m\to\infty$. It remains to note that $A_{m,n}\in C_{0}^{\infty}$ and ${\rm div{\hskip 1.42262pt}}A_{m,n}(x)\equiv 0$.