Abstract
In this paper we are concerned with the stochastic Landau–Lifshitz–Slonczewski (LLS) equation that describes magnetisation of an infinite nanowire evolving under current-driven spin torque. The current brings into the system a multiplicative gradient noise that appears as a transport term in the equation. We prove the existence, uniqueness and regularity of pathwise solutions to the equation.
Avoid common mistakes on your manuscript.
1 Introduction
In this paper we are concerned with the existence, uniqueness and regularity of solutions to the stochastic Landau–Lifshitz–Slonczewski (LLS) equation considered on real line, see (2). This is a result on a system of stochastic PDEs that combines variational structure with the transport noise in the presence of geometric constraints, more precisely, with solutions taking values in a sphere.
Let us recall briefly the physical motivation for the LLS equation, see [9, Chapter 9] for more details. In a nanoscale ferromagnetic element, the interaction between an electric current and the magnetisation of the element can result in current-induced magnetisation switching and spin wave emission. This observation gave rise to the rapidly developing field of spintronics, including racetrack memory [31] and microwave oscillators [24]. It is expected that a good understanding of the current-induced magnetisation switching and the stability of the domain walls subject to random perturbations will allow us to develop new types of current-controlled magnetic memories.
Slonczewski [38] first modelled the magnetisation dynamics in a magnetic multilayer structure with the current flowing perpendicular to the layers. The first (fixed) layer is assumed to have a uniform magnetisation m that polarises the spins of the incoming current, and the spin accumulation \( \textbf{s} \) entering the second (free) layer is uniform and parallel to m. This results in a spin transfer torque that affects the dynamics in the free layer. Viewing these layers as cross sections of a nanowire, and taking a continuous limit of the spin torque between infinitely thin cross sections [26, 40], we arrive at the LLS equation (1) with \( \gamma = -\alpha \).
Zhang et al. [41] later proposed another model involving directly the time and spatially varying magnetisation m(t, x) and spin accumulation \( \textbf{s}(t,x) \). The vector \( \textbf{s} \) is still roughly parallel to m since the magnetisation varies at a much larger length scale than the Fermi wavelength of electron spins [23, 25], but the deviation of \( \textbf{s} \) from m is non-trivial. Deriving from the generalised spin continuity equation, \( \textbf{s} \) satisfies a diffusion equation with terms depending on m, and enters the classic Landau–Lifshitz–Gilbert equation as an additional source of effective field. This fully coupled system in dimension 1 is formulated in [35, 41] and later reduced to (1) in [42] by viewing \( \textbf{s} \) as a first-order approximation of m. In particular, the final effective spin torque in [42] consists of (the continuous limit of) Slonczewski’s torque and an effective field contribution \( m \times \partial _x m \), both weighed by the current velocity. After rescaling, this leads to the LLS equation (1) with \( \gamma = \frac{\beta -\alpha }{1+\alpha \beta } \) for a constant \( \beta \in {\mathbb {R}}\) that depends on the exchange coupling and spin relaxation times.
Mathematical theory of the LLS equation is at an early stage. For the coupled system in [41], the existence and uniqueness of solutions were analysed in [16, 19, 32], and a related optimal control problem was investigated in [5]. We mention that in these papers, the estimates of (approximations of) \( \textbf{s} \) and m could be decoupled, with bounds depending only on the current density. In comparison, the LLS equation (1) has the gradient of m in place of \( \textbf{s} \), which leads to differences in proving estimates, such as the way to control the length of approximations of m. The case, when the ferromagnetic material fills in a three-dimensional domain, was studied in an important paper [26], where the existence and uniqueness of solutions were proved and their regularity was studied. A physically important case of a nanowire is a subject of ongoing intense research in physics.
Mathematical analysis of general travelling domain walls and their stability was only recently initiated in [27, 33, 36].
The necessity to include random fluctuations (such as thermal noise) into the dynamics of magnetisation has been conjectured by physicists for many years (see for example [9, 10, 29]. In spintronics control of stability of the current-driven domain walls is the main obstacle for developing new generation of magnetic memories. The existence and uniqueness of solutions and numerical analysis for the Landau–Lifshitz equation without the Slonczewski term but including random thermal fluctuations were intensely studied in recent years, see, for instance, [2, 6, 7, 11, 20,21,22]. Some numerical schemes for the stochastic equation are closely related to their deterministic counterparts, which also attracted a lot of attention in the last two decades, see [3, 4, 8, 14] and references therein. In this paper, we will consider the LLS equation perturbed by transport instead of thermal noise. In the literature, transport noise has been studied for various models, for example, linear gradient noise in stochastic Navier–Stokes equation [28] which has a regularising effect [18], and locally Lipschitz gradient noise in stochastic geometric wave equation [13, 30] which satisfies the corresponding geometric constraint by residing in the tangent space to the target manifold. We refer the reader to [1, 34] and references therein for more examples, where global well-posedness is achieved for equations with locally Lipschitz or locally monotone coefficients under a coercivity (and sometimes stochastic parabolicity) condition on the leading order terms. In comparison, Eq. (2) lacks coercivity, and it has a nonlinearity \( m \times \partial _{xx}m \) (the Schrödinger map) which is absent in the drift of wave equations.
To describe the problem in more detail, let us first recall briefly the deterministic LLS equation. We will identify an infinite nanowire made of ferromagnetic material with a real line \({\mathbb {R}}\) and will denote by \(m(t,x)\in {\mathbb {R}}^3\) the magnetisation vector at a time \(t\ge 0\) and at a point \(x\in {\mathbb {R}}\). For temperatures below the Curie point the length |m(t, x)| of this vector is constant in (t, x) [10] and hence can be assumed equal to 1:
The LLS equation proposed in [38] describes the dynamics of the magnetisation vector subject to the spin-velocity field (electric current):
It takes the form
with \(\alpha >0\) and \(\gamma \in {\mathbb {R}}\).
The term \( v \partial _x m \) is known as the adiabatic term and the non-adiabatic term is given by \( \gamma m \times (v \partial _x m) \).
We will consider a version of Eq. (1) with the spin-velocity field perturbed by noise:
where W is an infinite-dimensional Wiener process taking values in an appropriate function space.
We emphasise that noise arises in Eq. (2) in a way very different from the way it appears in stochastic Landau–Lifshitz equations studied in [11, 12, 20, 21]. While in the aforementioned papers it is a thermal noise arising inside the magnetic domain and has bounded diffusion coefficient, in (2) it is a transport noise brought into the system by the electric current and has the gradient of the solution as a diffusion coefficient. Another noise term corresponding to the thermal noise might be included in the model below without any substantial difficulties. In order to simplify rather complicated technical arguments, we consider in this paper the transport noise only. Another novelty of our work is the presence of a “quadratic” diffusion coefficient in the noise term in (2). Therefore, we are able to study the physically important case when both adiabatic and non-adiabatic random transport terms are included. We note stochastic Landau–Lifshitz equations studied in recent years usually do not include the quadratic diffusion term despite the fact it naturally arises from formally including the noise term in the energy functional, and when included [12], a cutoff function is required for the estimates even in the case without gradient in the noise. Due to those more natural assumptions, analysis of Eq. (2) requires much more delicate arguments.
In this paper, we will show that for every initial condition \(m_0\) with
there exists a unique pathwise and strong in PDEs sense solution to (2). We will use the observation made in [26] that under the constraint \( |m(t,x)|=1 \) for all \( (t,x) \in (0,\infty ) \times {\mathbb {R}} \), we have
hence, Eq. (2) can be written in the form
with \( m(0) = m_0 \) satisfying (3). Once we prove the existence of the sphere-valued solution m, Eq. (4) reduces to (2). We will assume that W is a Wiener process taking values in \(H^2({\mathbb {R}})\) and will prove the existence and uniqueness of pathwise solutions to this equation, see Theorem 2.5 for details. Due to the presence of gradient noise of multiplicative type, we can prove this theorem only for the Wiener process with the covariance small enough. Smallness of the noise is usually necessary for stochastic PDEs with gradient noise, even in the case of linear equations, see, for example, Chapter 6 of [15]. For our equation, this is also evident from (13) when \( \gamma =0 \).
Let us comment on the proofs of Theorems 2.5 and 2.6. Following the approach of [26] we use lattice approximations that are well-known in physics literature. In [39] they were used to derive the equation for Schrödinger maps as limits of equations for discrete systems of interacting spins modelling the Heisenberg ferromagnet. This approach allows us to construct approximating solutions that satisfy the sphere constraint.
Then we obtain a set of uniform estimates for the approximate solutions. This step requires using quadratic interpolations and requires many new estimates.
Next, we follow compactness type argument to prove the existence of a limiting point that is a strong in PDEs sense solution to stochastic LLS equation. Finally, in Theorem 2.6 we show the uniqueness of pathwise solutions and use the Yamada–Watanabe theorem in the same way as in [11].
2 Semi-discrete scheme and the main result
2.1 Notation
2.1.1 Function spaces
Let \( \rho _w(x) = (1+x^2)^{-w} \) for \( w \ge 0 \). Clearly,
and for \( w > \frac{1}{2} \), \( \int _{\mathbb {R}}\rho _w(x) \ \textrm{d}x<\infty \). For \(p\in [1,\infty )\), define the weighted Lebesgue space \( {\mathbb {L}}^p_w \) by
If \(w=0\), then we will write \( {\mathbb {L}}^p\) instead of \( {\mathbb {L}}^p_0\). We will denote by \( {\mathbb {H}}_w^1\) the Hilbert space
Let \(0<w_1 < w_2 \). Then \( \rho _{w_1} > \rho _{w_2} \) and the embeddings \( {\mathbb {L}}^2 \hookrightarrow {\mathbb {L}}^2_{w_1} \hookrightarrow {\mathbb {L}}^2_{w_2} \) are continuous with
Moreover, the embeddings
are compact, where \(\mathring{\mathbb H}^2\) stands for a standard homogeneous Sobolev space of functions \(f:{\mathbb {R}}\rightarrow {\mathbb {R}}^3\) with weak derivatives \( Df, D^2 f \in {\mathbb {L}}^2 \). The Laplace operator \(\Delta \) considered in \(\mathbb L^2_w\) with the domain \(\mathbb H^2_w\) is variational, and the operator \( A_1 = I-\Delta \) is invertible. For \( \beta > 0 \), let \( \mathbb H^\beta _w \) denote the domain of \( A_1^{\beta /2} \) endowed with the norm \( |\cdot |_{\mathbb H^{\beta }_w}:= |A_1^{\beta /2} \cdot |_{{\mathbb {L}}^2_w} \) and with dual space \( \mathbb H^{-\beta }_w \).
2.1.2 Assumption and notation
Let W be an \(H^2({\mathbb {R}})\)-valued Wiener process with the covariance operator Q. Then there exists a complete orthonormal sequence \(\left\{ f_j;\,j\ge 1\right\} \) of \(H^2({\mathbb {R}})\) made of eigenvectors of Q, that is
and then we have
The following is a standing assumption for the rest of the paper, and it will not be enunciated again.
Assumption 2.1
Let \( m_0 \) satisfy (3) with \( |m_0|_{{\mathbb {L}}^\infty \cap \mathring{{\mathbb {H}}}^1} \le K_0 \) for some constant \( K_0 >0 \). Let
and \( v \in {\mathcal {C}}([0,T]; H^1({\mathbb {R}})) \) with
Define a function
Remark 2.2
(a) Assumption 2.1 yields
(b) Every \(\mathbb H^2\)-valued finite-dimensional Wiener process satisfies (6) provided \(f_j^{\prime \prime } \in \mathbb L^\infty \) for \(j\ge 1\).
The following notations will be used throughout the paper. Let \( G: {\mathbb {L}}^\infty \cap \mathring{{\mathbb {H}}}^1\rightarrow {\mathbb {L}}^2\) be defined as
Let \( {\mathcal {G}}(m):= G^\prime (m) (G(m)) \),
which can be expressed as
In the rest of the paper, we consider Eq. (4) in its Itô form:
Here,
and the Stratonovich correction term S(m) takes the form
2.2 Semi-discrete scheme
2.2.1 Discrete operators and discrete spaces
Let \( {\mathbb {Z}}_h = \{ x = kh: k \in {\mathbb {Z}}\} \) denote a discretisation of the real line of mesh size \( h > 0 \). For \( u: {\mathbb {Z}}_h \rightarrow {\mathbb {R}}^3 \), we write \( u^{\pm }(x) \) for \( u(x \pm h) \), and we introduce discrete gradient and discrete Laplace operators:
Let \( {\mathbb {L}}_h^\infty \), \( {\mathbb {L}}_h^p \), \( {\mathbb {H}}_h^1\) and \( {\mathbb {E}}_h:= {\mathbb {L}}_h^\infty \cap \mathring{\mathbb {H}}_h^1 \) be discrete spaces equipped with respective norms:
where \(p\in [1,\infty )\).
We will say that \(u:[0,T]\times \Omega \times \mathbb Z_h\rightarrow {\mathbb {R}}^3\) is an \(\mathbb E_h\)-valued progressively measurable process if for every \(x\in \mathbb Z_h\) the process \(u(\cdot ,x)\) is progressively measurable and for every \(t \in [0,T]\),
In particular, the process \(\left\{ |u(t)|_{\mathbb E_h};\,t\ge 0\right\} \) is progressively measurable.
Let \( {\mathcal {E}}_h \) denote the space of \( {\mathbb {E}}_h \)-valued progressively measurable processes, with norm
2.2.2 Discrete equation
For \( u \in {\mathbb {E}}_h \), we define
where
Fix a terminal time \( T \in (0,\infty ) \); we describe the semi-discrete scheme for (8) as a stochastic differential equation in the space \({\mathbb {E}}_h\):
with initial condition \( m^h(0) = m_0^h:=m_0 \vert _{{\mathbb {Z}}_h} \in {\mathbb {E}}_h \), where \( \sup _{h>0} |m_0^h|_{{\mathbb {E}}_h} \) is bounded under Assumption 2.1. In (10) and (11), \( \kappa ^2, v(t) \) and W(t) are similarly the restrictions of the corresponding functions to \( {\mathbb {Z}}_h \) for every \( t \in [0,T] \). The term \( S^h(m^h) \) is a discretised and a modified version of the Stratonovich correction S(m) . It is chosen to simplify the proof in Sect. 3.2 without affecting the limit. For example, \( G_3^h(m^h) \) with the constant 2 does not match with the term \( \gamma \langle m, \partial _x m \rangle m \times \partial _x m \) in \( {\mathcal {G}}(u) \), but if \( \langle m^h, \partial ^h m^{h-} \rangle \) converges to 0 in a suitable space, then the constant will be irrelevant to the final equation.
Remark 2.3
The semi-discrete scheme (11) is introduced in this paper as a tool for proving the existence, uniqueness and regularity of solutions. The proof can be easily adapted to the case of bounded interval with Neumann boundary conditions without requiring weighted Sobolev spaces in the convergence of approximations. A genuine numerical scheme for solving Eq. (8) will require discretisation in time such as using implicit Euler method or Crank Nicolson method. The fully discrete schemes need uniform bounds of time translate estimates to conclude convergence. Fully discrete schemes and their error estimates will be a subject of further research.
2.3 Main result
Definition 2.1
We say that a progressively measurable process m defined on \(\left( \Omega ,{\mathcal {F}},\right. \left. \left( {\mathcal {F}}_t\right) ,{\mathbb {P}}, W\right) \) where W is a Wiener process, is a solution to Eq. (8) if
-
(a)
\(|m(t,x)|=1\) (t, x)-a.e.
-
(b)
for every \(T \in (0,\infty )\),
$$\begin{aligned}{\mathbb {E}}\left[ \sup _{t\in [0,T]} \left| \partial _x m\right| ^2_{\mathbb L^2}+\int _0^T\left| \partial _{xx}m\right| ^2_{\mathbb L^2}\,\textrm{d}t \right] <\infty ,\end{aligned}$$ -
(c)
for every \(t \in [0,T]\) the following equality holds in \(\mathbb L^2\) :
$$\begin{aligned} m(t)-m_0 = \int _0^t\left( F(m(s))+\frac{1}{2}S(m(s))\right) \,\textrm{d}s+\int _0^tG(m(s))\,\textrm{d}W(s),\quad {\mathbb {P}}\text {-a.s.} \nonumber \\ \end{aligned}$$(12)
Note that (a)–(d) above and Assumption 2.1 yield
hence the Bochner integral and the Itô integral in (12) are well defined in \(\mathbb L^2\).
Lemma 2.4
There exists a unique solution \( m^h \) of the semi-discrete scheme (11) in \( {\mathcal {E}}_h \) satisfying \( |m^h(t,x)|=1 \), \( {\mathbb {P}}\)-a.s. for all \( t \in [0,T] \) and \( x \in {\mathbb {Z}}_h \).
Theorem 2.5
There exists a solution \( (\Omega , {\mathcal {F}}, \left( {\mathcal {F}}_t\right) , {\mathbb {P}}, W, m) \) of (8) in the sense of Definition 2.1, such that for \( p \in [1,\infty ) \),
provided that \( C_\kappa \) is sufficiently small. Then for every \(T> 0\) and \(\alpha \in \left( 0,\frac{1}{2}\right) \),
Moreover, there exists a convergent subsequence \( \{m_h\} \) defined on \( (\Omega , {\mathcal {F}}, {\mathbb {P}}) \) such that \( m_h \) has the same law as a quadratic interpolation of \( m^h \) for every \( h > 0 \), and m is the \( {\mathbb {P}}\)-a.s. limit of \( \{m_h\} \) in \( {\mathcal {C}}([0,T]; {\mathbb {H}}^{-1}_{w}) \) for some \( w \ge 1 \).
For the existence (\( p=1 \)) and \( L^p(\Omega ) \)-regularity in Theorem 2.5, it is sufficient to assume that
where \( b_p \) is the constant in Burkholder–Davis–Gundy inequality. This gives an example of the required smallness of the noise, see (21) (or Remark 3.6) for a more precise condition (or bound) on \( C_\kappa \).
Theorem 2.6
The solution m of (8) is pathwise unique and therefore unique in law.
3 Discretisation
From the definition of discrete operators and the discrete \({\mathbb {L}}_h^p\) norm, we deduce following results.
Remark 3.1
For \( u: {\mathbb {Z}}_h \rightarrow {\mathbb {R}}^3 \),
-
(a)
\(\partial ^{h} (\partial ^{h} u) = \frac{1}{h} (\partial ^{h}u^+- \partial ^{h} u) = \Delta ^h u^+\),
-
(b)
for any \(p\in [1,\infty ]\), \(|u|_{{\mathbb {L}}_h^p} = |u^+|_{{\mathbb {L}}_h^p} = |u^-|_{{\mathbb {L}}_h^p}\), which implies \( |\partial ^{h}u|_{{\mathbb {L}}_h^p} = |\partial ^{h} u^+|_{{\mathbb {L}}_h^p} = |\partial ^{h} u^-|_{{\mathbb {L}}_h^p}\), and hence,
$$\begin{aligned} |\partial ^h u|_{\mathbb L_h^p}^2 \le \frac{4}{h^2} | u|_{\mathbb L_h^p}^2, \quad |\Delta ^h u|_{\mathbb L_h^p}^2 \le \frac{4}{h^2} |\partial ^hu|_{\mathbb L_h^p}^2, \end{aligned}$$ -
(c)
Lemma A.1 indicates \(|u|_{\mathbb L_h^\infty }\le C |u|_{\mathbb H^1_h}\) for any \(u\in \mathbb H_h^1 \,\).
3.1 Existence of a unique solution of the semi-discrete scheme
Lemma 3.2
For every \( h >0 \), if \( f,g: {\mathbb {E}}_h \rightarrow {\mathbb {E}}_h \) are locally Lipschitz and satisfy \( f(0) = g(0) = 0 \), then \( f \times g \), \( \langle f, g \rangle \) and \( \partial ^h f \) are also locally Lipschitz on \( {\mathbb {E}}_h \).
The result in Lemma 3.2 is clear, and we omit the proof here. Then we check that the coefficients in (11) are locally Lipschitz on \( {\mathbb {E}}_h \).
Lemma 3.3
For every \( h > 0 \), \( F^h, G^h, S^h: {\mathbb {E}}_h \rightarrow {\mathbb {E}}_h \) are locally Lipschitz on \( {\mathbb {E}}_h \).
Proof
Let \( u,w \in {\mathbb {E}}_h \). It follows from Remark 3.1(b) that
and
By Lemma 3.2 and (7), \( F^h, G^h\) and \(S^h \) are locally Lipschitz. \(\square \)
Proof of Lemma 2.4
For each \( n \in {\mathbb {N}} \) and \( r^h =F^h, S^h \) and \( G^h \), define
Then \( F_{n}^h \), \( S_n^h \) and \( G_n^h \) are Lipschitz on \( {\mathbb {E}}_h \).
Fix \( n \in {\mathbb {N}} \). Let \( A_n: {\mathcal {E}}_h \rightarrow {\mathcal {E}}_h \) be given by
We first verify that \( A_n(u) \in {\mathcal {E}}_h \) for \( u \in {\mathcal {E}}_h \). Note that \( F_n^h \) and \( S_n^h \) are bounded on \( {\mathbb {E}}_h \), with
for some constant \( C_1 \) that depends on h and n. For \( J_n(t) \), we have
where the last inequality holds by Tonelli’s theorem. This together with Remark 3.1(b) implies
Then by the definition of \( G_n^h \), the assumption (6) and Fubini’s theorem,
for \( C_2(h,n,\kappa ) = 2C_\kappa ^2 (n^4 + \gamma ^2 n^2) (1+\frac{4}{h^2}) T\). Thus, \( J_n \) is a \( {\mathbb {H}}_h^1 \)-valued continuous martingale. By Lemma A.1 (or Remark 3.1(c)), there exists a constant \( C>0 \) such that
From [15, Corollary 4.29],
It follows from (15) and (16) that
Thus, \( A_n(u) \in {\mathcal {E}}_h \) for \( u \in {\mathcal {E}}_h \).
For \( u, \nu \in {\mathcal {E}}_h \), there exists a constant \( C >0 \) such that
Similarly, \( M_n(t):= \int _0^t \left( G_n^h(\nu (s)) - G_n^h(u(s)) \right) \textrm{d}W(s) \) is a \( {\mathbb {H}}_h^1 \)-valued continuous martingale, and replacing \( J_n \) by \( M_n \) in (15) and (16), we have
By construction, if \( |\nu (s)|_{{\mathbb {E}}_h}, |u(s)|_{{\mathbb {E}}_h} \le n \), then
If \( |\nu (s)|_{{\mathbb {E}}_h}, |u(s)|_{{\mathbb {E}}_h} > n \), then let \( n_s^\nu = n |\nu (s)|_{{\mathbb {E}}_h}^{-1} \) and \( n_s^u = n |u(s)|_{{\mathbb {E}}_h}^{-1} \), we have
and
If \( |\nu (s)|_{{\mathbb {E}}_h} > n \) and \( |u(s)|_{{\mathbb {E}}_h} \le n \), then
implying that \( |n_s^\nu - 1| \ |u|_{{\mathbb {E}}_h} \le |u(s)- \nu (s)|_{{\mathbb {E}}_h} \) and
Similar result follows for \( \partial ^h (G_n^h(\nu )-G_n^h(u)) \) using Remark 3.1(b). Then, by (18), Lemma 3.3 and Hölder’s inequality, there exist constants \( L_1(h,n,T) \) and \( L_2(h,n,T) \) such that
Consider the discrete equation
with \( m^h_n(0) = m_0^h \in {\mathbb {E}}_h \), on intervals \( [(k-1) {\tilde{T}},k{\tilde{T}}] \) for \( k \ge 1 \), where \( {\tilde{T}} \) satisfies \( L_2(h,n,{\tilde{T}}) < 1 \). By the Banach fixed point theorem, there exists a unique solution \( m_n^h \in {\mathcal {E}}_h \) of (19) on [0, T] .
Define the stopping times
Let \( \tau = \tau _n \wedge \tau _n' \). Then \( A_n(m_{n+1}^h) = A_{n+1}(m_{n+1}^h) \) on \( [0,\tau ) \), and by (18),
which implies
by Grönwall’s lemma. Thus, \( m_{n+1}^h(\cdot \wedge \tau ) = m_n^h(\cdot \wedge \tau ) \) and \( \tau = \tau _n \), \( {\mathbb {P}} \)-a.s. and the discrete equation (11) admits a local solution \( m^h(t) = m_n^h(t) \) for \( t \in [0,\tau _n] \).
Recall that \( m_0^h \in {\mathbb {E}}_h \) and \( |m_0^h(x)| =1 \) for all \( x \in {\mathbb {Z}}_h \). Applying Itô’s lemma to \( \frac{1}{2}|m^h(t,x)|^2 \),
By \( \langle a, a \times b \rangle =0 \), for any \( t \in [0,\tau _n] \) and \( x \in {\mathbb {Z}}_h \),
Therefore, \( |m^h(t,x)| = |m_n^h(t,x)| = |m_0^h(x)| = 1 \) for any \( t \in [0,\tau _n] \) and \( x \in {\mathbb {Z}}_h \).
For any fixed h and n, the unique solution \( m_n^h \) of (19) satisfies
We apply Itô’s lemma to \( \frac{1}{2}|\partial ^{h} m_n^h(t \wedge \tau _n)|^2_{{\mathbb {L}}_h^2} \),
Since \( |m_n^h(t,x)|=1 \) for \( (t,x) \in [0,\tau _n] \times {\mathbb {Z}}_h \), and
there exist constants \( \beta _1 \) and \( \beta _2 \) that depend on h (not n) such that
and
Then, by the boundedness of \( \kappa ^2 \), the stochastic integral \( \int _0^{t \wedge \tau _n} \langle \Delta ^h m_n^h(s), G^h(m_n^h(s)) \textrm{d}W(s) \rangle _{{\mathbb {L}}_h^2} \) is a square integrable continuous martingale for \( m_n^h \in {\mathcal {E}}_h \), for every \( h>0 \). Now, we have
where the stochastic integral part vanishes after taking expectation. We obtain
By Grönwall’s lemma,
By the definition of \( \tau _n \), the left-hand side of (20) is greater than \( n^2 {\mathbb {P}}(\tau _n \in [0,t]) \), thus
In other words, \( \tau _n \rightarrow \infty \), \( {\mathbb {P}} \)-a.s, as \( n \rightarrow \infty \). Thus, the process \( m^h(t) = \lim _{n \rightarrow \infty } m_n^h(t \wedge \tau _n) \) is the unique solution of the semi-discrete scheme (11). \(\square \)
3.2 Uniform estimates for the solution \( m^h \) of the discrete SDE
For every \( h>0 \), let
In the following lemma, we deduce an upper bound of the stochastic integral \( M^h(t) \) which is used in the proof of Lemma 3.5 to obtain uniform estimates for \( m^h \).
Lemma 3.4
For any \( p \in (0,\infty ) \), there exists a constant \( b_p \) independent of h such that
Proof
We observe that for every \( j \ge 1 \),
which implies
Then as in the proof of Lemma 2.4, for every fixed h,
implying that \( M^h(t) \) is a continuous martingale. By the Burkholder–Davis–Gundy inequality, for \( p \in (0,\infty ) \), there exists a constant \( b_p \) such that
Taking the supremum over t for \( |\partial ^{h} m^h(t)|_{{\mathbb {L}}_h^2}^2 \),
\(\square \)
Lemma 3.5
For any \( p \in [1,\infty ) \), assume that
\(\{(q_j,f_j)\}_{j\ge 1}\) satisfies
for some small \( \delta > 0 \), where \( b_p \) is the constant in Lemma 3.4. Let \( \sup _{h>0}|m_0^h|_{{\mathbb {E}}_h} \le K_0 \). Then, there exist constants \( K_{1,p} \) and \( K_{2,p} \) that are independent of h, such that
for all \( h> 0 \).
Proof
As in Lemma 2.4, let \( \phi (u) = \frac{1}{2} | \partial ^{h} u|_{{\mathbb {L}}_h^2}^2 \) for \( u \in {\mathbb {E}}_h \). Then, for \( \nu ,w \in \mathring{{\mathbb {H}}}_h^1 \),
By Itô’s lemma,
where \( M^h(t) \) is already estimated in Lemma 3.4.
\(\underline{\text { An estimate on }\,\,T_1:}\)
The second term on the right-hand side of (25) is estimated using (7) and the fact that \( |m^h|=1 \) \( {\mathbb {P}}\)-a.s., as follows
for arbitrary \( \varepsilon >0 \). An estimate of \(T_1\) is obtained from (25) and (26)
\(\underline{\text { An estimate on }\,\,T_2:}\)
Using \( |m^h|=1 \), \( {\mathbb {P}}\)-a.s., we estimate \(T_{21}\):
where \( |\kappa \kappa '|_{{\mathbb {L}}_h^\infty }^2 \le C_\kappa ^4 \) by (6).
To estimate \(T_{22}\), we first note that for any \(u:{\mathbb {Z}}_h \rightarrow {\mathbb S}^2 \),
By (98), we deduce
It is clear that
where the second term on the right-hand side will cancel with parts of \( T_3 \).
To estimate \( T_{22b} \), we observe that for any \( u: {\mathbb {Z}}_h \rightarrow {\mathbb S}^2 \), using (98),
where
If \( |\partial ^h u(x)| \le |\partial ^h u^-(x)| \) at some \( x \in {\mathbb {Z}}_h \), then
and by (31),
If \( |\partial ^h u(x)| \ge |\partial ^h u^-(x)| \), then we can show that the term given by (31) is bounded by \( |u \times \Delta ^h u|^2(x) \). Explicitly, by (32),
where the last inequality holds by \( h^2 |\partial ^h u|^2 \le 4 \) and \( |\partial ^h u(x)| \ge |\partial ^h u^-(x)| \). Combining the two cases and replacing u by \( m^h(s) \), we have
We will see later in the proof that \( T_{22c} \) and \( T_{22d} \) also cancel with parts of \( T_3 \).
\(\underline{\text { An estimate on } \,\,T_3:}\)
We first estimate \( T_{31}(s) \). For \( u: {\mathbb {Z}}_h \rightarrow {\mathbb S}^2 \), we have for every \( j \ge 1 \),
where
Hence,
For the square \( \frac{1}{8} |A_0|^2(x) \):
where the second inequality holds by applying the Mean Value Theorem to \( f_j \) on the interval \( [x,x+h] \) for every \( j \ge 1 \), such that there exists some \( \xi _h \in (x,x+h) \) satisfying
and \( |\sum _j q_j^2 (f_j')^2|_{{\mathbb {L}}^\infty } \le C_\kappa ^2 \) by assumption (6).
For the squares \( \frac{1}{4}|A_1|^2(x) \) and \( \frac{1}{4}|A_2|^2(x) \):
where the right-hand side cancels with \( T_{22c}(s) \) in (29) when u is replaced with \( m^h(s) \).
For the squares \( \frac{1}{4}|B_1|^2(x) \) and \( \frac{1}{4}|B_2|^2(x) \), we first observe that
Then,
This implies that
where the right-hand side cancels with a part of the estimate for \( T_{22a} \) in (30) when \( u = m^h(s) \) as aforementioned.
For the cross terms \( \frac{1}{2} \left\langle A_1,B_1 \right\rangle (x) \) and \( \frac{1}{2} \left\langle A_2,B_2 \right\rangle (x) \):
and similarly,
Then,
By (36), the left term in the inner product (38) can be simplified as
where the second equality holds by observing \( \langle u+u^+, \partial ^{h}u \rangle (x) = 0 \) due to \( |u(x)| = 1 \) for all x. Recall that the right term in the inner product (38) is \( \partial ^{h}(u \times \partial ^{h}u) = u^+ \times \Delta ^hu^+ \) by (36). Then,
Taking the sum over \( x \in {\mathbb {Z}}_h \),
For the cross term \( \frac{1}{4}\left\langle A_0, \ A_1 + A_2 + B_1 + B_2\right\rangle (x) \):
and
which imply
Then, using again the Mean Value Theorem for \( \Delta ^h (f_j)^2 \),
Therefore, by (34), (35), (37), (39) and (40),
Next, we estimate \( T_{32}(s) \). Using (36),
Finally, we estimate \( T_{33}(s) \). We note that for \( u = u(x) \) with \( |u(x)|=1 \) for all x and for all \( j \ge 1 \),
Since \( \sum _j q_j^2 |f_j| |\partial ^h f_j|(x) \le C_\kappa ^2 \) for all \( x \in {\mathbb {Z}}_h \), we have
where \( T_{22d}(s) \) is given in (29).
\(\underline{\text { An estimate on } \,\,T_1 + T_2 + T_3:}\)
and by (43),
Then, by (27), (28), (33) and (42),
where
\(\underline{\text { Uniform estimate of } \,\, \partial ^h m^h }:\)
Taking a sufficiently small \( \varepsilon \) such that \( \frac{1}{2}\varepsilon ^2 \left( 3 + 2|\gamma | C_\kappa ^2 \right) < \delta \), we have from (21):
Then, for \( p \ge 1 \) and \( q = \frac{p}{p-1} \),
where the second inequality holds by (46) and the last inequality holds by Lemma 3.4. Then, by the definitions of \( N_{1,p} \) and \( N_{2,p} \),
By Fubini’s theorem and Grönwall’s inequality,
where \( K_{1,p} \) depends on \( p, C_v, C_\kappa , \varepsilon , T \) and \( K_0 \), but not on h, proving (22).
Finally, by (21), (48) and (50),
where \( K_{2,p} \) depends on \( p, C_v, C_\kappa , \varepsilon , T \) and \( K_0 \), but not on h, proving (23). \(\square \)
Remark 3.6
Fix \( p \in [1,\infty ) \), if
then the assumption (21) of Lemma 3.5 is satisfied.
Lemma 3.7
For any \( p \in [1, \infty ) \), under the conditions of Lemma 3.5, there exists a constant \( K_{3,p} \) independent of h such that
Proof
Since \( |m^h(s,x)|=1 \) for all \( (s,x) \in [0,T] \times {\mathbb {Z}}_h \), we have
Then,
where
Applying Lemma A.1 on \( \partial ^{h} m^h(s) \),
where \( |\partial ^{h}(\partial ^{h} m^h(s)) |_{{\mathbb {L}}_h^2} = |\Delta ^h m^h(s)|_{{\mathbb {L}}_h^2} \). Thus,
We have
Then, by Lemma 3.5,
proving (51). \(\square \)
4 Quadratic interpolation for the solution \(m^h\) of (11)
4.1 Interpolations
For any fixed \(h>0\), let \( x_k = kh \in {\mathbb {Z}}_h \), for \( k \in {\mathbb {Z}}\). We introduce interpolations of discrete functions defined on \({\mathbb {Z}}_h\) to functions defined on \( {\mathbb {R}}\).
Given \( u: {\mathbb {Z}}_h \rightarrow {\mathbb {R}}^3 \), let \( \overline{u}: {\mathbb {R}}\rightarrow {\mathbb {R}}^3 \) denote a quadratic interpolation of u, given by
for \( x \in [x_k,x_{k+1}) \), \( k \in {\mathbb {Z}}\), where \( \overline{u} \) is continuously differentiable with
Let \( \widehat{u}: {\mathbb {R}}\rightarrow {\mathbb {R}}^3 \) denote the piecewise constant interpolation of u, given by
for any \( k \in {\mathbb {Z}}\). In terms of \( \widehat{u} \), we can express \( \overline{u} \) as
for \( x \in [x_k,x_{k+1}) \), \( k \in {\mathbb {Z}}\).
We collect estimates of \(\overline{u}\) and \(\widehat{u}\) in terms of u in the following remark.
Remark 4.1
Let \( u: {\mathbb {Z}}_h \rightarrow {\mathbb {R}}^3 \). Then
and
4.2 An equation and estimates for \( \overline{m}^h \)
4.2.1 Equation for \( \overline{m}^h \)
Since \( m^h \) is the solution of the semi-discrete scheme (11), the piecewise constant interpolation \( \widehat{m}^h \) satisfies
where \( F^h_{\widehat{v}} \) and \( S^h_{\widehat{\kappa }} \) are defined as in (10) but with \( \widehat{v} \), \( \widehat{\kappa } \) and \( \widehat{\kappa \kappa '} \) in place of v, \( \kappa \) and \( \kappa \kappa ' \), respectively, and
In particular, for every fixed \( h>0 \), \( m^h \in {\mathcal {C}}([0,T]; {\mathbb {E}}_h) \) and \( \widehat{m}^h \in {\mathcal {C}}([0,T]; {\mathbb {L}}^2_w \cap \mathring{{\mathbb {H}}}^1) \) for \( w \ge 1 \).
In order to obtain an equation for \( \overline{m}^h \), by using (57) we note that
where for \( u: {\mathbb {R}}\rightarrow {\mathbb {R}}^3 \) with well-defined weak derivatives,
and
for \( x \in [x_k,x_{k+1}) \), \( k \in {\mathbb {Z}}\).
Moreover, for \( u: [0,T] \times {\mathbb {R}}\rightarrow {\mathbb {R}}^3 \), define
By (59), we arrive at the equation of \( \overline{m}^h \). For \( t \in [0,T] \),
4.2.2 Estimates for \( \overline{m}^h \)
For \( p \in [1, \infty ) \) and \( w \ge 1 \), we deduce from Lemmata 3.5, 3.7 and Remark 4.1:
and
for some constant C(p, T, w) .
Results of convergence of \(R_0 \overline{m}^h\), \(R_1 \overline{m}^h\), \(P_1 \overline{m}^h,\ldots , P_5 \overline{m}^h\) and \(Q_1 \overline{m}^h\), \(Q_2 \overline{m}^h\) are proved in the following lemma.
Lemma 4.2
For \( f = R_1 \), \( P_1 \) or \( P_2 \),
Moreover, for \( f=P_3 \), \( P_4 \), \( P_5 \), \( Q_1 \) or \( Q_2 \), for any measurable process \( \varphi \in L^4(\Omega ; L^4(0,T; {\mathbb {L}}^4)) \),
Proof
By construction, \( |x-x_k| < h \) for \( x \in [x_k, x_{k+1}) \), \( k \in {\mathbb {Z}}\), and \( \sup _{t \in [0,T]} |\overline{m}^h(t)|_{{\mathbb {L}}^\infty } \le 5 \), \( {\mathbb {P}}\)-a.s. Thus, for \( p \in [1, \infty ) \), there is a constant \( C_{R_0} \) independent of h such that
Using (55), we can often rewrite \( R_0, \ldots , Q_2 \) in terms of \( \widehat{m}^h \) to simplify the estimates.
\(\underline{\text { An estimate on } \,\, R_0\overline{m}^h:}\)
which implies
The expectation on the right-hand side of (66) is bounded by Lemma 3.5; thus, the left-hand side converges to 0 as \( h \rightarrow 0 \). As a result,
\(\underline{\text { An estimate on } \,\, R_1\overline{m}^h:}\)
implying \( R_1 \overline{m}^h \rightarrow 0 \) in \( L^2(\Omega ; L^2(0,T; {\mathbb {L}}^2)) \) by (65).
\(\underline{\text { An estimate on } \,\, P_1\overline{m}^h:}\)
where \( D\overline{m}^h \in L^4(\Omega ; L^4(0,T; {\mathbb {L}}^4)) \) for all \( h > 0 \) by (65). Then, by the \( L^4 \)-convergence of \( R_0\overline{m}^h \) in (67) and the \( L^2 \)-convergence of \( R_1\overline{m}^h \), we have \( P_1 \overline{m}^h \rightarrow 0 \) in \( L^2(\Omega ; L^2(0,T; {\mathbb {L}}^2)) \).
\(\underline{\text { An estimate on } \,\, P_2\overline{m}^h:}\)
which implies that \( P_2 \overline{m}^h \rightarrow 0 \) in \( L^2(\Omega ; L^2(0,T; {\mathbb {L}}^2)) \) by (64) and (65) together with the convergences of \( P_1\overline{m}^h \) and \( R_0\overline{m}^h \).
\(\underline{\text { An estimate on } \,\, P_3 \overline{m}^h:}\)
Then for \( \varphi \in L^4(\Omega ; L^4(0,T; {\mathbb {L}}^4)) \),
and
By Lemmata 3.5 and 3.7, (64), (65) and the property of \( \varphi \), the expectations on the right-hand side of the two inequalities above are finite. Then by the convergences of \( R_0\overline{m}^h \) and \( P_1\overline{m}^h \), we obtain the weak convergence of \( P_3\overline{m}^h \) as desired.
\(\underline{\text { An estimate on } \,\, P_4\overline{m}^h:}\)
We have
and
Using (65), (53) and the convergences of \( R_1 \overline{m}^h \) and \( P_1 \overline{m}^h \), the right-hand side of each of the two inequalities above converges to 0 as \( h \rightarrow 0 \).
\(\underline{\text { An estimate on } \,\, P_5\overline{m}^h:}\)
By Lemma 2.4 and (64), we have
and
Similarly, by Lemma 3.5, (65) and the convergences of \( R_0\overline{m}^h \) and \( P_1\overline{m}^h \), the right-hand side of each of the inequalities above converges to 0 as \( h \rightarrow 0 \).
\(\underline{\text { An estimate on } \,\, Q_1\overline{m}^h:}\)
Thus,
where the three expectation terms on the right-hand side are finite, proving that the right-hand side converges to 0 as \( h \rightarrow 0 \).
\(\underline{\text { An estimate on } \,\, Q_2\overline{m}^h:}\)
where on the right-hand side, the first term converges to 0 as \( h \rightarrow 0 \) by the argument for \( Q_1 \overline{m}^h \) with \( \varphi \times \overline{m}^h \in L^4(\Omega ; L^4(0,T; {\mathbb {L}}^4)) \), and the second term converges to 0 by (67). \(\square \)
We also obtain uniform bounds for \(\overline{m}^h\) in weighted spaces.
Lemma 4.3
For any \( w \ge 1 \), the quadratic interpolation \( \overline{m}^h \) satisfies
-
(i)
\( \sup _h {\mathbb {E}}[ |\overline{m}^h|_{B_w}^2 ] < \infty \) for \( B_w:= L^2(0,T; {\mathbb {L}}^2_w \cap \mathring{{\mathbb {H}}}^2) \cap {\mathcal {C}}([0,T]; {\mathbb {L}}^2_w \cap \mathring{{\mathbb {H}}}^1) \),
-
(ii)
\( \sup _h {\mathbb {E}}[ |\overline{m}^h|^2_{W^{\alpha ,p}(0,T; {\mathbb {L}}^2_w)} ] < \infty \), for \( p \in [2,\infty ) \) and \( \alpha \in (0, \frac{1}{2}) \) such that \( \alpha - \frac{1}{p} < \frac{1}{2} \),
-
(iii)
\( |\overline{m}^h| \rightarrow 1 \) in \( L^2(\Omega ; L^2(0,T; {\mathbb {L}}^2)) \).
Proof
\(\underline{\text { Part (i).}}\) For every fixed \( h > 0 \), \( \widehat{m}^h \) is in \( {\mathcal {C}}([0,T]; {\mathbb {L}}^2_w \cap \mathring{{\mathbb {H}}}^1) \) and so does \( \overline{m}^h \). Then part (i) follows directly from the estimates in (64) and (65).
\(\underline{\text { Part (ii).}}\) Recall (59), we have from the definition (54) that
where for \( x \in [x_k,x_{k+1}) \), \( k \in {\mathbb {Z}}\),
By the \( {\mathbb {L}}^\infty \)-estimate of \( \overline{m}^h \) in (64),
For \( I_1 \), by Lemma 3.5, there exists a constant \( a_1 \) that may depend on \( C_v, C_\kappa , \alpha , \gamma , T, K_{1,1}, K_{1,3} \) and \( K_{2,1} \) such that
For \( I_2 \) and \( I_3 \), since \( |x-x_k| \le h \) and
there also exist constants \( a_2, a_3 \) such that
Similarly, for the stochastic integrals in \( I_4 \), we only need to verify that \( \int _0^T G^h(\widehat{m}^h(s)) \textrm{d}\widehat{W}^h(s) \) is bounded in \( L^p(\Omega ; W^{\alpha , p}(0,T; {\mathbb {L}}^2)) \). By [17, Lemma 2.1], there exist a constant C depending on \( \alpha , p, \gamma , T \) and a constant \( a_4 \) depending on \( C, C_\kappa \) and \( K_{1,p} \) such that for \( p \in [2,\infty ) \) and \( \alpha \in (0, \frac{1}{2}) \),
Since \( {\mathbb {L}}^2 \hookrightarrow {\mathbb {L}}^2_w \) for \( w \ge 1 \), the estimates above hold for the \( {\mathbb {L}}^2_w \)-norm. By Lemma A.4, the embedding \( W^{1,2}(0,T; {\mathbb {L}}^2_w) \hookrightarrow W^{\alpha ,p}(0,T; {\mathbb {L}}^2_w) \) is continuous for \( \alpha - \frac{1}{p} < \frac{1}{2} \). Thus,
\(\underline{\text { Part (iii).}}\) Since \( |m^h(t,x)|=1 \), \( {\mathbb {P}}\)-a.s. for all \( (t,x) \in [0,T] \times {\mathbb {Z}}_h \), we observe that
for \( (t,x) \in [0,T] \times [x_k,x_{k+1}) \), \( k \in {\mathbb {Z}}\). This implies
where the last inequality holds by Lemmata 3.5 and 3.7, and we obtain the convergence after taking \( h \rightarrow 0 \). \(\square \)
5 Existence of solution
In this section, we first show that the sequence \(\{(\overline{m}^h, W)\}_h\) is tight and then by using the Skorohod theorem we obtain its almost sure convergence, up to a change of probability space. Finally, we prove that the limit is a solution of the stochastic LLS equation (4) in the sense of Definition 2.1.
5.1 Tightness and construction of new probability space and processes
Fix \( w_1, w_2 \) such that \( w_2 > w_1 \ge 1 \). Define
Recall \( {\mathbb {L}}^2_{w_1} \cap \mathring{{\mathbb {H}}}^2 {\mathop {\hookrightarrow }\limits ^{\text {compact}}} {\mathbb {H}}^1_{w_2} \hookrightarrow {\mathbb {L}}^2_{w_2} \). By (99) and Lemma A.2,
where \( \overline{m}^h \in E_0 \), \( {\mathbb {P}}\)-a.s. by Lemma 4.3. Also, since the embeddings \( {\mathbb {H}}^1_{w_1} \hookrightarrow {\mathbb {L}}^2_{w_2} \hookrightarrow {\mathbb {H}}_{w_1}^{-1} \) are compact and \( 4 \alpha >1 \), it holds by Lemma A.3 that
In summary, \( E_0 \) is compactly embedded in E. For any \( r>0 \),
where \( \{ |\overline{m}^h|_{E_0} \le r \} \) is compact in E, and the right-hand-side converges to 0 as r tends to infinity. Therefore, the set of laws \( \{ {\mathcal {L}}(\overline{m}^h) \} \) on the Banach space E is tight, which implies the following convergence result.
Lemma 5.1
There exists a probability space \( (\Omega ^*, {\mathcal {F}}^*, {\mathbb {P}}^*) \) and there exists a sequence \( (m_h^*, W_h^*) \) of \( E \times {\mathcal {C}}([0,T]; H^2({\mathbb {R}})) \)-valued random variables defined on \( (\Omega ^*, {\mathcal {F}}^*, {\mathbb {P}}^*) \), such that the laws of \( (\overline{m}^h, W) \) and \( (m_h^*, W_h^*) \) on \( E \times {\mathcal {C}}([0,T]; H^2({\mathbb {R}})) \) are equal for every h, and there exists an \( E \times {\mathcal {C}}([0,T]; H^2({\mathbb {R}})) \)-valued random variable \( (m^*, W^*) \) defined on \( (\Omega ^*, {\mathcal {F}}^*, {\mathbb {P}}^*) \) such that
and
Proof
Since \( E \times {\mathcal {C}}([0,T]; H^2({\mathbb {R}})) \) is a separable metric space, the result holds by the Skorohod theorem. \(\square \)
Since the laws of \( (\overline{m}^h, W) \) and \( (m_h^*, W_h^*) \) on \( E \times {\mathcal {C}}([0,T]; H^2({\mathbb {R}})) \) are equal, due to the following remark we obtain the same estimates for \(m_h^*\).
Remark 5.2
By Kuratowski’s theorem, the Borel sets of
are Borel sets of \( E = L^2(0,T; {\mathbb {H}}^1_{w_2}) \cap {\mathcal {C}}([0,T]; {\mathbb {H}}^{-1}_{w_1}) \) for \( w_1 < w_2 \), where
We can assume that \( m_h^* \) takes values in B and the laws on B of \( \overline{m}^h \) and \( m_h^* \) are equal.
By Remark 5.2, the sequence \( (m_h^*)_h \) satisfies the same estimates as \( (\overline{m}^h)_h\) on B. By (65), for any \( p \in [1,\infty ) \),
Since \( |\rho _w'| \le w \rho \) for \( w >0 \), by Gagliardo–Nirenberg inequality,
which implies
Thus, by (70) – (73), for \( p \in [1,\infty ) \),
As in (53),
which implies
5.2 Identification of the limit \( (m^*, W^*) \)
5.2.1 Convergence of functions of \( m_h^* \)
For \( p \in [1,\infty ) \), by the pointwise convergence of \( m_h^* \) in (68) and the uniform integrability of \( m_h^* \) and \( D m_h^* \) in (70) – (71), we have
By (71), \( Dm_h^* \) also converges weakly to a measurable process X in \( L^{2p}(\Omega ^*; L^2(0,T; {\mathbb {L}}^2)) \), which implies that \( X = Dm^* \in L^{2p}(\Omega ^*; L^2(0,T; {\mathbb {L}}^2)) \) by the uniqueness of the limit of weak convergence in \( L^{2p}(\Omega ^*; L^2(0,T; {\mathbb {L}}^2_{w_2})) \). By (77) and integration-by-parts,
Similarly, by (72), \( D^2m_h^* \) converges weakly to a measurable process Y in \( L^{2p}(\Omega ^*; {\mathbb {L}}^2(0,T; {\mathbb {L}}^2)) \), thus \( Y = D^2 m^* \in L^{2p}(\Omega ^*; L^2(0,T; {\mathbb {L}}^2)) \) and
Lemma 5.3
We have
-
(i)
\( |m^*(t,x)| = 1 \), (t, x) -a.e. \( {\mathbb {P}}^* \)-a.s.
-
(ii)
\( m_h^* \rho _{w_2}^\frac{1}{2} \rightarrow m^* \rho _{w_2}^\frac{1}{2} \) in \( L^p(\Omega ^*; L^p(0,T; {\mathbb {L}}^p)) \), for \( p \in [2,\infty ) \),
-
(iii)
\( Dm^* \in L^4(\Omega ^*; L^4(0,T; {\mathbb {L}}^4)) \cap L^p(\Omega ^*; L^\infty (0,T;{\mathbb {L}}^2)) \), for \( p \in [2,\infty ) \).
Proof
\(\underline{\text { Part (i).}}\) Recall Lemma 4.3(iii); a similar argument holds for \( {\mathbb {L}}^2_{w_2} \) (in place of \( {\mathbb {L}}^2 \)). Then,
where the first expectation on the right-hand side converges to 0 since the laws of \( \overline{m}^h \) and \( m_h^* \) are the same on \( L^2(0,T;{\mathbb {L}}^2_{w_2}) \), and the second expectation converges to 0 by (76). Thus,
which implies \( |m^*(t,x)| = 1 \) a.e. on \( [0,T] \times {\mathbb {R}}\), \( {\mathbb {P}}^* \)-a.s. This also means
\(\underline{\text { Part (ii).}}\) For \( w_2 > w_1 \ge 1 \) and \( p \in [2,\infty ) \),
for some constant C that depends on p and T. Then, by the \( {\mathbb {L}}^\infty \)-estimate (73), Lemma 5.3(i), (5) and the strong convergence (76),
\(\underline{\text { Part (iii).}}\) From part (i), we have \( \frac{1}{2} D|m^*(t,x)|^2 = \langle m^*, Dm^* \rangle (t,x) = 0 \) and thus
for (t, x) -a.e. \( {\mathbb {P}}^* \)-a.s. Since \( D^2m^* \in L^2(\Omega ^*; L^2(0,T; {\mathbb {L}}^2)) \), we deduce
As in [11], we extend the definition of the \( {\mathbb {L}}^2_{w_2} \cap \mathring{{\mathbb {H}}}^1 \)-norm to \( {\mathbb {H}}^{-1}_{w_1} \) such that \( |u|_{{\mathbb {L}}^2_{w_2} \cap \mathring{{\mathbb {H}}}^1} = \infty \) if the function u is in \( {\mathbb {H}}^{-1}_{w_1} \) but not \( {\mathbb {L}}^2_{w_2} \cap \mathring{{\mathbb {H}}}^1 \), where the extended map
is lower semicontinuous. Then, by the pointwise convergence (68), Fatou’s lemma and (71),
for \( p \in [2,\infty ) \). \(\square \)
Lemma 5.4
We have the following strong convergences:
-
(i)
\( m_h^* \times Dm_h^* \rightarrow m^* \times Dm^* \) and \( \langle m_h^*, Dm_h^* \rangle \rightarrow 0 \) in \( L^2(\Omega ^*; L^2(0,T; {\mathbb {L}}^2_{w_2})) \),
-
(ii)
\( m_h^* \times \left( m_h^* \times D m_h^* \right) \rightarrow m^* \times (m^* \times Dm^*) \) in \( L^2(\Omega ^*; L^2(0,T; {\mathbb {L}}_{w_1+w_2}^2)) \).
Proof
\(\underline{\text { Part (i).}}\) Note that
Then by Hölder’s inequality,
where the last line converges to 0 by Lemma 5.3(ii) and (75). Similarly, by Lemma 5.3(i),
where the right-hand side converges to 0 by (77). Therefore,
Since \( |m^*(t,x)|=1 \), we have \( \langle m^*, Dm^* \rangle (t,x)=0 \). By the same argument as above (replacing cross product with scalar product), \( \langle m_h^*, Dm_h^* \rangle \rightarrow \langle m^*, Dm^* \rangle = 0 \) in \( L^2(\Omega ^*; L^2(0,T; {\mathbb {L}}^2_{w_2})) \).
\(\underline{\text { Part (ii).}}\) Note that
Then, with \( \rho _{w_1 + w_2} = \rho _{w_1} \rho _{w_2} \),
where the first expectation in the last inequality converges to 0 by Lemma 5.3(ii) and the second and third expectations are finite by (73) and (75). Also,
converges to 0 Lemma 5.3(i) and part (i). Then, the strong convergence of \( m_h^* \times (m_h^* \times Dm_h^*) \) follows as desired. \(\square \)
Lemma 5.5
Assume that \( w_2 \ge 4w_1 \). For any measurable process \( \varphi \in L^4(\Omega ^*; {\mathbb {L}}^4(0,T; {\mathbb {L}}^4_{w_2})) \), we have the following weak convergences (with test function \( \varphi \)):
-
(i)
\( m_h^* \times (m_h^* \times D m_h^*) \rightharpoonup m^* \times (m^* \times Dm^*) \) in \( L^2(\Omega ^*; L^2(0,T; {\mathbb {L}}^2_{w_2})) \),
-
(ii)
\( |m_h^* \times Dm_h^*|^2 m_h^* \rightharpoonup |m^* \times Dm^*|^2m^* \) in \( L^2(\Omega ^*; L^2(0,T; {\mathbb {L}}^2_{w_2})) \),
-
(iii)
\( Dm_h^* \times (m_h^* \times Dm_h^*) \rightharpoonup Dm^* \times (m^* \times Dm^*) \) in \( L^2(\Omega ^*; L^2(0,T; {\mathbb {L}}^2_{w_2})) \),
-
(iv)
\( \langle m_h^*, Dm_h^* \rangle m_h^* \times Dm_h^* \rightharpoonup 0 \) in \( L^2(\Omega ^*; L^2(0,T; {\mathbb {L}}^2_{w_2})) \),
-
(v)
\( m_h^* \times D^2m_h^* \rightharpoonup m^* \times D^2m^* \) in \( L^2(\Omega ^*; L^2(0,T; {\mathbb {L}}^2_{w_2})) \),
-
(vi)
\( m_h^* \times (m_h^* \times D^2m_h^*) \rightharpoonup m^* \times (m^* \times D^2m^*) \) in \( L^2(\Omega ^*; L^2(0,T; {\mathbb {L}}^2_{w_2})) \).
Proof
\(\underline{\text { Part (i).}}\) As in lemma 5.4(ii), we first observe that
where the first expectation in the last line converges to 0 by Lemma 5.3(ii), the second and the third expectations are finite by (74) with \( w_2 \ge 2w_1 \) (equivalently, \( \rho _{w_2} \le \rho _{w_1}^2 \)) and \( \varphi \in L^4(\Omega ^*; L^4(0,T; {\mathbb {L}}^4_{w_2})) \). Since \( |m^*(t,x)| = 1 \) from Lemma 5.3(i), we have \( m^* \times \varphi \in L^2(\Omega ^*; L^2(0,T; {\mathbb {L}}^2_{w_2})) \). Then by Lemma 5.4(i),
Combining (81) and (82), we have the desired weak convergence for part (i).
\(\underline{\text { Part (ii).}}\)
Then, for the first term in the line above,
We show that the first expectation on the right-hand side of (83) is finite. Since \( w_2 \ge 4w_1 \), it holds that \( \rho _{w_2} \le \rho _{w_1}^2 \rho _{w_2}^\frac{1}{2} \) and
where three expectations in the last inequality are finite by (73), (75) and \( \varphi \in L^4(\Omega ^*; L^4(0,T; {\mathbb {L}}^4_{w_2})) \). Similarly, by Lemma 5.3(i) and (iii),
Hence, the left-hand side of (83) converges to 0 as \( h \rightarrow 0 \) by Lemma 5.4(i). Similarly, with \( |m^*(t,x)|=1 \), \( {\mathbb {P}}^* \)-a.s. we have
where the last line converges to 0 by Lemma 5.3(ii) – (iii). Combining (83) and (84), we have the desired weak convergence for part (ii).
\(\underline{\text { Part (iii).}}\) Note that
Then,
where the last line converges to 0 by Lemma 5.3(iii) and (77). Similarly,
which converges to 0 by (75) and Lemma 5.4(i). Together, we have
\(\underline{\text { Part (iv).}}\) Again, since \( w_2 \ge 4 w_1 \), we have \( \rho _{w_2}^\frac{1}{4} \le \rho _{w_1} \). Then,
where the right-hand side converges to 0 by (73) (with \( \rho _{w_1}^\frac{1}{2} \le 1 \)), (75) and the convergence of the scalar product in Lemma 5.4(i).
\(\underline{\text { Part (v).}}\)
Then, for the first term on the right-hand side,
where the first expectation in the last inequality converges to 0 by Lemma 5.3(ii), the second expectation is finite as \( \varphi \in L^4(\Omega ^*; {\mathbb {L}}^4(0,T; {\mathbb {L}}^4_{w_2})) \) and the final expectation is finite by (72). Thus, the left-hand side converges to 0 as \( h \rightarrow 0 \). Also, \( m^* \times \varphi \in L^2(\Omega ^*; L^2(0,T; {\mathbb {L}}^2_{w_2})) \) and then by the weak convergence (78),
Therefore,
\(\underline{\text { Part (vi).}}\) Similarly,
Then,
where the last line converges to 0 by Lemma 5.3(ii), \( \varphi \in L^4(\Omega ^*; {\mathbb {L}}^4(0,T; {\mathbb {L}}^4_{w_2})) \) and (74) with \( \rho _{w_1} \ge \rho _{w_2}^\frac{1}{4} \). Also, we have
where \( m^* \times \varphi \in L^4(\Omega ^*; L^4(0,T; {\mathbb {L}}^4_{w_2})) \) and thus by part (iii), the expectation in the last line above converges to 0. Therefore,
\(\square \)
Lemma 5.6
Assume that \( w_2 \ge 4w_1 \). Recall the definitions (62), we have
-
(i)
\( F_{\widehat{v}}(m_h^*) \rightharpoonup F(m^*) \) in \( L^2(\Omega ^*; L^2(0,T; {\mathbb {L}}^2_{w_2})) \),
-
(ii)
\( S_{\widehat{\kappa }}(m_h^*) \rightharpoonup S(m^*) \) in \( L^2(\Omega ^*; L^2(0,T; {\mathbb {L}}^2_{w_2})) \),
-
(iii)
\( \kappa G(m_h^*) \rightarrow \kappa G(m^*) \) (strongly) in \( L^2(\Omega ^*; L^2(0,T; {\mathbb {L}}^2_{w_1+w_2})) \).
Proof
As in Lemma 5.5, let \( \varphi \) be an arbitrary measurable process in \( L^4(\Omega ^*; L^4(0,T; {\mathbb {L}}^4_{w_2})) \). By (6) and (7), \( \kappa ^2, \kappa \kappa ' \in {\mathbb {L}}^\infty \cap {\mathbb {H}}^1 \) and \( v \in {\mathcal {C}}([0,T]; {\mathbb {L}}^\infty \cap {\mathbb {H}}^1) \). Then for \( y = \kappa ^2, (\kappa ^2)^-, \kappa \kappa ', v \), any piecewise constant approximation (in the x-variable) z of y satisfies
For example, the approximation z can be taken to be \( \widehat{y}^- \) or \( \widehat{y} \). Let u be a function such that \( u(m_h^*) \in L^2(\Omega ^*; L^2(0,T; {\mathbb {L}}^2_{w_2})) \). Then,
If \( u(m_h^*) \rightharpoonup u(m^*) \) in \( L^2(\Omega ^*; L^2(0,T; {\mathbb {L}}_{w_2}^2)) \), then the right-hand side of (86) converges to 0 by (85) and \( y \varphi \in L^4(\Omega ^*; L^4(0,T; {\mathbb {L}}^4_{w_2})) \).
\(\underline{\text { Part (i).}}\) Let \( u(m_h^*) = m_h^* \times (m_h^* \times Dm_h^*) \) and let \( y = v \). The result follows immediately from (86) and Lemma 5.5(v) and (vi).
\(\underline{\text { Part (ii).}}\) Since \( w_2 \ge 4w_1 \), we have \( \rho _{w_2} \le \rho _{w_1}^4 \), We observe that from (75) and (73) that
are in \( L^2(\Omega ^*; L^2(0,T; {\mathbb {L}}_{w_2}^2)) \).
Taking the following choices of u, y and z:
and using Lemma 5.5(ii) – (vi), we follow again the argument (86) to obtain weak convergence of \( S_{\widehat{\kappa }}(m_h^*) \) to \( S(m^*) \) in \( L^2(\Omega ^*; L^2(0,T; {\mathbb {L}}^2_{w_2})) \).
\(\underline{\text { Part (iii).}}\) The result follows from (6) and Lemma 5.4. \(\square \)
5.2.2 Wiener process
Define a sequence of processes \( \{ \overline{M}_h \}_{h > 0} \) on \( (\Omega ,{\mathcal {F}}, {\mathbb {P}}) \) by
Recall the equation of \( \overline{m}^h \), we have from (63) that
where the operator \( R^h \) is given by
Similarly, define a sequence of processes \( \{ M_h^* \}_{h > 0} \) on \( (\Omega ^*, {\mathcal {F}}^*, {\mathbb {P}}^*) \) by
Lemma 5.7
For each \( t \in (0,T] \), we have the following weak convergence in \( L^2(\Omega ^*; {\mathbb {H}}^{-1}_{w_1}) \):
Proof
Recall that \( {\mathbb {H}}_{w_1}^1 \) is compactly embedded in \( {\mathbb {L}}^2_{w_2} \). Let \( t \in (0,T] \) and \( \varphi \in L^2(\Omega ^*; {\mathbb {H}}_{w_1}^1) \). By Remark 5.2, the two sets of remainders
have the same laws for \( \overline{m}^h, m_h^* \in L^2(0,T; {\mathbb {L}}^2_{w_1} \cap \mathring{{\mathbb {H}}}^2) \cap {\mathcal {C}}([0,T]; {\mathbb {L}}^2_{w_1} \cap \mathring{{\mathbb {H}}}^1) \). Then, by Lemma 4.2,
By the pointwise convergence (68) of \( m_h^* \) in \( {\mathcal {C}}([0,T]; {\mathbb {H}}^{-1}_{w_1}) \), Lemma 5.6(i) – (ii) and the convergence of piecewise constant approximations \( \widehat{m}_0^h \rightarrow m_0 \) in \( {\mathbb {L}}^2_w \) for \( w>w_1 \) (due to \( \partial _x m_0 \in {\mathbb {L}}^2 \)), we have
\(\square \)
Lemma 5.8
The process \( W^* \) is a Q-Wiener process on \( (\Omega ^*, {\mathcal {F}}^*, {\mathbb {P}}^*) \), and \( W^*(t) - W^*(s) \) is independent of the \( \sigma \)-algebra generated by \( m^*(r) \) and \( W^*(r) \) for \( r \in [0,s] \).
Proof
See [11, Lemma 5.2(i)] (using Lemma 5.1). \(\square \)
Lemma 5.9
For each \( t \in [0,T] \),
Proof
Fix h and \( t \in (0,T] \). For each \( n \in {\mathbb N} \), define the partition \( \{ s_i^n = \frac{iT}{n}: i = 0, \ldots , n \} \). Define
where \( \widehat{W}_h^*(s) \) is the piecewise constant approximation of \( W_h^*(s) \) (as in (56)) for every \( s \in [0,T] \). As in (69), we also have
Consider the following two \( {\mathbb {L}}^2_{w_2} \)-valued random variables:
Following Remark 5.2, \( \overline{Y}_{h,n} \) and \( Y^*_{h,n} \) have the same distribution. As \( n \rightarrow \infty \),
This implies that \( Y^*_{h,n}(t) \) also converges to 0 in \( L^2(\Omega ^*; {\mathbb {L}}^2_{w_2}) \) as \( n \rightarrow \infty \). Thus,
We observe that
For \( J_0^h \) and \( J_2 \), let \( \varepsilon > 0 \) and choose \( n \in {\mathbb N} \) such that
Since \( \widehat{W}_h^* \) and \( \widehat{W}^h \) have the same laws on \( {\mathcal {C}}([0,T]; H^2({\mathbb {R}})) \), we have
Recall that \( {\mathbb {L}}^2_w \hookrightarrow {\mathbb {H}}^{-1}_{w_1} \) for all \( w > w_1 \). Let \( w = w_1+w_2 \). As \( h \rightarrow 0 \), the first and the third term on the right-hand side converges to 0 by Lemma 5.6(iii), the fourth term converges to 0 by Lemma 4.2 and (6), and the second term is less than \( \frac{\varepsilon }{2} \) by (88). Hence, for a sufficiently small h, we have
Similarly, \( |J_2|_{L^2(\Omega ^*; {\mathbb {H}}_{w_1}^{-1})} < \frac{\varepsilon }{2} \).
For \( J_1^h \), we have
Since \( W^* \) is a Q-Wiener process, the first term on the right-hand side converges to 0 by Lemma 5.6(iii), Also, the second term converges to 0 by the pointwise convergence (69) (or (87)) and the result \( G(m_h^*) \rho _{\frac{w}{2}} \in L^2(\Omega ^*; L^2(0,T;{\mathbb {L}}^\infty )) \), which can be deduced from the estimates (71), (72) and (73).
Therefore, for any sufficiently small h,
Using Lemma 5.7 and the uniqueness of weak limit, the proof is concluded. \(\square \)
We are ready to prove the main theorem.
5.2.3 Proof of Theorem 2.5
By Lemmata 5.7 and 5.9, \( m^* \) satisfies the (12) in \( {\mathbb {H}}^{-1}_{w_1} \). Moreover, using Lemma 5.3(i), we can simplify F, S and G:
and each of them is in \( L^2(\Omega ^*; L^2(0,T; {\mathbb {L}}^2)) \), hence the equality (12) holds in \( {\mathbb {L}}^2 \). Recall the properties of \( m^* \) shown previously in (79) and Lemma 5.3, we have now verified that \( m^* \) is a solution of (8) in the sense of Definition 2.1. It only remains to show that \( m-m_0 \in {\mathcal {C}}^\alpha ([0,T]; {\mathbb {L}}^2) \). For \( s,t \in [0,T] \) and \( p \in [1,\infty ) \), there exists a constant C that may depend on \( p,T,C_{\kappa } \) such that
where the expectation on the right-hand side is finite. Then by Kolmogorov’s continuity criterion, \( m^*(t)-m_0 \in {\mathcal {C}}^\alpha ([0,T]; {\mathbb {L}}^2) \), \( {\mathbb {P}}^* \)-a.s. for \( \alpha \in (0,\frac{1}{2}) \).
5.2.4 Proof of Theorem 2.6
Let \( (m_1,W) \) and \( (m_2,W) \) on \( (\Omega ,{\mathcal {F}},\left( {\mathcal {F}}_t\right) ,{\mathbb {P}}) \) be two solutions of (8) in the sense of Definition 2.1. Let \( u = m_1 - m_2 \) and \( w>0 \). Applying Itô’s lemma to \( \frac{1}{2}|u(t)|_{{\mathbb {L}}^2_w}^2 \),
\(\underline{\text { An estimate on } \,\, U_1:}\)
Then, for an arbitrary \( \varepsilon >0 \),
and
Similarly,
and
Also,
Hence,
for the process \( \psi _1 \) given by
For \( i=1,2 \), there exists a constant \( C>0 \) such that
which implies \( \int _0^T \psi _1(t) \ \textrm{d}t < \infty \), \( {\mathbb {P}}\)-a.s.
\(\underline{\text { An estimate on } \,\, U_2: }\)
Again, for \( \varepsilon >0 \),
and
Also,
For the remaining term in \( U_2 \), we use integration-by-parts as in (93):
Thus,
where
and by (96), \( \int _0^T \psi _2(t) \ \textrm{d}t < \infty \), \( {\mathbb {P}}\)-a.s.
\(\underline{\text { An estimate on } \,\, U_3:}\)
where for every \( j \ge 1 \),
Hence,
where the second term on the right-hand side cancels with the corresponding term in \( U_2(s) \) and \( \psi _3(s) = \gamma ^2 C_\kappa ^2 \left( \frac{1}{2} + \frac{1}{\varepsilon ^2}(1+\gamma ^2) C_\kappa ^2 \right) |Dm_1(s)|^2_{{\mathbb {L}}^\infty } \) is similarly integrable \( {\mathbb {P}}\)-a.s.
We have
We can choose a sufficiently small \( \varepsilon > 0 \) such that under the assumption (21),
which implies
Therefore, by (89),
Define the process Y by
Then,
Since \( |u(t)|_{{\mathbb {L}}^\infty } \le 2 \) \( {\mathbb {P}}\)-a.s. and there exists a constant C such that
the process
is a martingale, and then
By the definition of Y(t) , if \( Y(0) = m_1(0) - m_2(0) = 0 \), then
for \( t \in [0,T] \), proving pathwise uniqueness of the solution of (8). By the Yamada–Watanabe Theorem, the uniqueness in law follows.
Data availability
Data sharing is not applicable to this article as no datasets were generated or analysed during the current study.
References
Agresti A. and Veraar M. The critical variational setting for stochastic evolution equations. Probability Theory and Related Fields. 188(3) (2024). 957–1015.
Alouges F., De Bouard A. and Hocquet A. A semi-discrete scheme for the stochastic Landau-Lifshitz equation. Stochastic Partial Differential Equations: Analysis and Computations, 2(3) (2014). 281–315.
Alouges F. and Jaisson P. Convergence of a finite element discretization for the Landau-Lifshitz equations in micromagnetism. Mathematical Models and Methods in Applied Sciences, 16(02) (2006). 299–316.
An R., Gao H. and Sun W. Optimal error analysis of Euler and Crank-Nicolson projection finite difference schemes for Landau-Lifshitz equation. SIAM Journal on Numerical Analysis, 59(3) (2021). 1639–1662.
An X., Majee A.K., Prohl A. and Tran T. Optimal control for a coupled spin-polarized current and magnetization system. Advances in Computational Mathematics, 48(3), 2022, p.28.
Baňas L., Brzeźniak Z., Neklyudov M. and Prohl A. Stochastic ferromagnetism: analysis and numerics. Vol. 58. Walter de Gruyter, 2013.
Baňas L., Brzeźniak Z., Neklyudov M. and Prohl A. A convergent finite-element-based discretization of the stochastic Landau-Lifshitz-Gilbert equation. IMA Journal of Numerical Analysis, 34(2) (2014), 502–549.
Bartels S., Ko J. and Prohl A. Numerical analysis of an explicit approximation scheme for the Landau-Lifshitz-Gilbert equation. Mathematics of Computation, 77(262) (2008) 773–788.
Bertotti G., Mayergoyz I. and Serpico C. Nonlinear magnetization dynamics in nanosystems. Elsevier B. V., Amsterdam, 2009.
Brown W.F. Micromagnetics. New York: Robert E. Krieger Publishing Company, 1978.
Brzeźniak Z., Goldys B. and Jegaraj T. Weak solutions of a stochastic Landau-Lifshitz-Gilbert equation. Applied Mathematics Research eXpress. 1 (2012), 1–33.
Brzeźniak Z., Goldys B. and Jegaraj T. Large deviations and transitions between equilibria for stochastic Landau-Lifshitz-Gilbert equation. Arch. Ration. Mech. Anal. 226(2) (2017), 497–558.
Brzeźniak Z. and Ondreját M. Stochastic geometric wave equations with values in compact Riemannian homogeneous spaces. Ann. Probab. 41(38) (2013), 1938–1977.
Cimrák I. Error estimates for a semi-implicit numerical scheme solving the Landau-Lifshitz equation with an exchange field. IMA journal of numerical analysis, 25(3) (2005), 611–634.
Da Prato G. and Zabczyk J. Stochastic Equations in Infinite Dimensions. Cambridge University Press, 2014.
Di Fratta G., Jüngel A., Praetorius D. and Slastikov V. Spin-diffusion model for micromagnetics in the limit of long times. Journal of Differential Equations, 343 (2023), 467–494.
Flandoli F. and Gatarek D. Martingale and stationary solutions for stochastic Navier-Stokes equations. Probab. Theory Relat. Fields. 102 (1995), 367–391.
Flandoli F. and Luo D. High mode transport noise improves vorticity blow-up control in 3D Navier-Stokes equations. Probability Theory and Related Fields. 180 (2021), 309–363.
García-Cervera C.J. and Wang X.P. Spin-polarized transport: existence of weak solutions. Discrete Contin. Dyn. Syst. Ser. B 7.1 (2007), 87–100.
Goldys B., Le, K.-N. and Tran T. A finite element approximation for the stochastic Landau-Lifshitz-Gilbert equation. J. Differential Equations. 260 (2016), 937–970.
Goldys B., Grotowski, J.F. and Le K.-N. Weak martingale solutions to the stochastic Landau-Lifshitz-Gilbert equation with multi-dimensional noise via a convergent finite-element scheme. Stochastic Process. Appl. 130 (2020), 232–261.
Gussetti E. and Hocquet A. A pathwise stochastic Landau-Lifshitz-Gilbert equation with application to large deviations. Journal of Functional Analysis. 285(9) (2023), p.110094.
Hubert A. and Schäfer R. Magnetic domains: the analysis of magnetic microstructures. Springer Science & Business Media, 2008.
Kaka S., Pufall M.R., Rippard W.H., Silva T.J., Russek S.E. and Katine J.A. Mutual phase-locking of microwave spin torque nano-oscillators. Nature 437(7057) (2005), 389–392.
Li Z., and Zhang S. Domain-wall dynamics and spin-wave excitations with spin-transfer torques. Physical review letters 92(20) (2004), 207203.
Melcher C. and Ptashnyk M. Landau-Lifshitz-Slonczewski equations: global weak and classical solutions. SIAM J. Math. Anal. 45 (2013), 407–429.
Melcher C. and Rademacher J.D.M. Pattern formation in axially symmetric Landau-Lifshitz-Gilbert-Slonczewski equations. J. Nonlinear Sci. 27(5) (2017), 1551–1587.
Mikulevicius R. and Rozovskii B.L. Global \( L_2 \)-solutions of stochastic Navier-Stokes equations. Ann. Probab. 33(1) (2005), 137–176.
Néel L. Bases d’une nouvelle théorie générale du champ coercitif. Annales de l’université de Grenoble. 22 (1946), 299–343.
Ondreját M. Existence of global mild and strong solutions to stochastic hyperbolic evolution equations driven by a spatially homogeneous Wiener process J. Evol. Equ. 4 (2004), 169–191.
Parkin S., Hayashi M. and Thomas L. Magnetic domain–wall racetrack memory. Science. 320 (2008), no. 5873, 190–194.
Pu X. and Guo B. Global smooth solutions for the one-dimensional spin-polarized transport equation. Nonlinear Analysis: Theory, Methods & Applications, 72(3-4) (2010), 1481-1487.
Rademacher J.D.M. and Siemer L. Domain wall motion in axially symmetric spintronic nanowires. SIAM J. Appl. Dyn. Syst. 20(4) (2021), 2204–2235.
Röckner M., Shang S. and Zhang T. Well-posedness of stochastic partial differential equations with fully local monotone coefficients. Mathematische Annalen. (2024), 1–51.
Shpiro A., Levy P.M. and Zhang S. Self-consistent treatment of nonequilibrium spin torques in magnetic multilayers. Physical Review B, 67(10) (2003), 104430.
Siemer L., Ovsyannikov I. and Rademacher, J.D.M. Inhomogeneous domain walls in spintronic nanowires. Nonlinearity 33(6) (2020), 2905–2941.
Simon J. Sobolev, Besov and Nikolskii fractional spaces: imbeddings and comparisons for vector valued spaces on an interval. Annali di Matematica Pura ed Applicata. 157(1) (1990), 117–148.
Slonczewski J.C. Current-driven excitation of magnetic multilayers. J. Magn. Magn. Mater. 159 (1996), L1–L7.
Sulem P.-L., Sulem C. and Bardos C. On the continuous limit for a system of classical spins. Comm. Math. Phys. 107(3) (1986), 431–454.
Thiaville A., Nakatani Y., Miltat J. and Suzuki Y. Micromagnetic understanding of current-driven domain wall motion in patterned nanowires. Europhysics Letters 69(6) (2005), 990.
Zhang S., Levy P.M. and Fert A. Mechanisms of spin-polarized current-driven magnetization switching. Physical review letters 88.23 (2002), 236601.
Zhang S. and Li Z. Roles of nonequilibrium conduction electrons on the magnetization dynamics of ferromagnets. Physical review letters, 93(12) (2004), 127204.
Zhou Y. Applications of Discrete Functional Analysis to the Finite Difference Method. International Academic Publishers, 1990.
Funding
Open Access funding enabled and organized by CAUL and its Member Institutions.
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflict of interest
The authors have no conflict of interest to declare.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
This work was supported by the Australian Research Council Project DP200101866.
APPENDIX A
APPENDIX A
1.1 A.1 Some calculations in discrete spaces
-
(a)
discrete integration-by-parts:
$$\begin{aligned} \sum _{x \in {\mathbb {Z}}_h} \langle u(x), \partial ^{h} w(x) \rangle = - \sum _{x \in {\mathbb {Z}}_h} \langle \partial ^{h} u^-(x), w(x) \rangle , \end{aligned}$$for \( u,w \in {\mathbb {H}}_h^1 = \{ v \in {\mathbb {L}}_h^2: |\partial ^h v|_{{\mathbb {L}}_h^2} < \infty \} \) with appropriate decay properties. In particular,
$$\begin{aligned} \left\langle \partial ^h u, \partial ^h w \right\rangle _{{\mathbb {L}}_h^2} = - \left\langle \Delta ^h u, w \right\rangle _{{\mathbb {L}}_h^2}. \end{aligned}$$ -
(b)
discrete expansion of \( \langle u, \Delta ^h u \rangle \) and \( \langle u, \partial ^h u \rangle \): for any u satisfying that \(|u(x)|=1\) for all \(x\in {\mathbb {Z}}_h\),
$$\begin{aligned} \begin{aligned} \langle u(x), \Delta ^h u (x) \rangle&= - \frac{1}{2} \left( |\partial ^{h} u(x)|^2 + |\partial ^{h} u^-(x)|^2 \right) \le 0, \\ \langle u(x), \partial ^h u(x) \rangle&= -\frac{h}{2} |\partial ^h u(x)|^2 \le 0. \end{aligned} \end{aligned}$$(98) -
(c)
product rule:
$$\begin{aligned} \partial ^{h} (fu)&= (\partial ^{h} f) u(x) + f^+ \partial ^{h} u = (\partial ^{h} f) u^+ + f \partial ^{h} u(x) \end{aligned}$$for f scalar-valued and u vector-valued; similarly for f and u both scalar-valued, and for \( \langle f, u \rangle \) and \( u \times u \) when f, u are vector-valued.
-
(d)
\( L_h^2 \)-norm of \( \Delta ^h u \):
$$\begin{aligned} |\Delta ^h u|_{{\mathbb {L}}_h^2} = |\partial ^{h} (\partial ^{h} u)^-|_{{\mathbb {L}}_h^2} = |\partial ^{h}(\partial ^{h} u)|_{{\mathbb {L}}_h^2}. \end{aligned}$$
Lemma A.1
[43, Chapter 1, Theorem 3] For \( u^h: {\mathbb {Z}}_h \rightarrow {\mathbb {R}} \),
for \( p \in [2,\infty ] \), \( k \in [0,n) \) and C is a constant independent of \( u^h \).
1.2 A.2 Some tightness results
Lemma A.2
[17, Theorem 2.1] Let \( B_0 \subset B \subset B_1 \) be Banach spaces, \( B_0 \) and \( B_1 \) reflexive, with compact embedding of \( B_0 \) in B. Let \( p \in (1,\infty ) \) and \( \alpha \in (0,1) \) be given. Let X be the space
endowed with the natural norm. Then the embedding of X in \( L^p(0,T; B) \) is compact.
Lemma A.3
[17, Theorem 2.2] If \( B_1 \subset {\widetilde{B}} \) are two Banach spaces with compact embedding, and the real numbers \( \alpha \in (0,1) \), \( p>1 \) satisfy \( \alpha p > 1 \), then the space \( W^{\alpha ,p}(0,T; B_1) \) is compactly embedded into \( {\mathcal {C}}([0,T]; {\widetilde{B}}) \).
Lemma A.4
[37, Corollary 19] Let I be an either bounded or unbounded interval of \( {\mathbb {R}}\). Let E be a Banach space. Suppose \( s \ge r \), \( p \le q \) and \( s- \frac{1}{p} \ge r - \frac{1}{q} \) for \( 0< r \le s <1 \) and \( 1 \le p \le q \le \infty \). Then,
In addition, we verify the continuous embedding
Indeed, for \( u \in W^{\alpha ,4}(0,T; {\mathbb {L}}^2_w) \),
where the second inequality holds by \( \sqrt{a} + \sqrt{b} \le \sqrt{2a + 2b} \) for \( a,b \ge 0 \).
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Goldys, B., Jiao, C. & Le, K.N. Existence, uniqueness and regularity of solutions to the stochastic Landau–Lifshitz–Slonczewski equation. J. Evol. Equ. 24, 79 (2024). https://doi.org/10.1007/s00028-024-01011-3
Accepted:
Published:
DOI: https://doi.org/10.1007/s00028-024-01011-3
Keywords
- Landau–Lifshitz equations
- Spin-transfer torque
- Stochastic partial differential equations
- Gradient noise