betaFeedback

WorksMathematics

Modeling and simulation of inductionless magnetohydrodynamic free surface problems with unmatched densities

Abstract

We propose a new diffuse interface model for simulating an inductionless magnetohydrodynamic (MHD) free surface problem. By using the Onsager's variational principle and the laws of thermodynamics, we derive a thermodynamically consistent system that couples the Cahn–Hilliard equation modeling phase separation, the Navier–Stokes equations governing fluid motion, and a generalized Darcy's law accounting for electromagnetic effects. In contrast to existing diffuse interface MHD models, the proposed model can handle general material properties in practical engineering applications. Furthermore, through asymptotic arguments, we investigate the sharp interface limit, and then demonstrate that the classical sharp interface model can be recovered as the interface thickness approaches zero, theoretically validating the proposed diffuse interface model as an approximate approach. An efficient decoupled, linear, and charge-conservative finite element scheme is designed, and it significantly facilitates the large-scale and accurate numerical simulations involving large parameter ratios. Finally, we present several three-dimensional numerical experiments of magnetic damping effects on bubble dynamics for the demonstration of the capability of the proposed model and method in capturing complex MHD phenomena.

Introduction

When the magnetic fields are applied to the electrically conducting and nonmagnetic fluids (such as the liquid metals and strong electrolytes), the induced magnetohydrodynamic (MHD) effects will significantly influence the fluid dynamics, resulting in a fundamentally distinct flow behavior. Considering the ubiquitous presence of such a scenario in industrial processes, numerous studies have been carried out to explore the underlying flow mechanism, such as vortex dynamics, heat transfer, and turbulence, see, for example, [5, 7, 16]. While these flows exhibit extremely complex dynamics in practical engineering applications, free surface problems are usually encountered, yielding a more intricate fluid motion. Representative examples include continuous casting and refining of metals [11, 28], magnetic destabilization of free surfaces in aluminum reduction cells [9, 15], liquid metal batteries [29], and plasma-facing components [20], among others.

In this work, we consider the inductionless MHD free surface problems in a fixed bounded domain Ω⊂R3\Omega\subset\mathbb{R}^{3}, which contains two incompressible and immiscible fluids occupying two open, time-dependent subdomains Ω±(t)\Omega_{\pm}(t), respectively. Characterized by the fact that the free surfaces are codimension-1 manifolds from a macroscopic perspective [12], the two subdomains Ω±(t)\Omega_{\pm}(t) are separated by a sharp interface Γ(t)=Ω+(t)‾∩Ω−(t)‾\Gamma(t)=\overline{\Omega_{+}(t)}\cap\overline{\Omega_{-}(t)}. Consequently, the flow dynamic behavior can be governed by the following classical sharp interface model:

ρ±(∂tu+u⋅∇u)=∇⋅(2η±D(u)−pI)+J×B+ρ±g,in Ω±(t),(1)\rho_{\pm}\left(\partial_t u+u\cdot\nabla u\right)=\nabla\cdot\left(2\eta_{\pm}\mathbb{D}(u)-p\mathbb{I}\right)+\mathcal{J}\times B+\rho_{\pm}g,\qquad\text{in }\Omega_{\pm}(t), \tag*{(1)}
∇⋅u=0,in Ω±(t),(2)\nabla\cdot u=0,\qquad\text{in }\Omega_{\pm}(t), \tag*{(2)}
J=σ±(E+u×B),in Ω±(t),(3)\mathcal{J}=\sigma_{\pm}(E+u\times B),\qquad\text{in }\Omega_{\pm}(t), \tag*{(3)}
E=−∇V,in Ω±(t),(4)E=-\nabla V,\qquad\text{in }\Omega_{\pm}(t), \tag*{(4)}
∇⋅J=0,in Ω±(t),(5)\nabla\cdot\mathcal{J}=0,\qquad\text{in }\Omega_{\pm}(t), \tag*{(5)}
⟦2ηD(u)−pI⟧nΓ=−λκnΓ(6)\llbracket2\eta\mathbf{D}(\mathbf{u})-p\mathbf{I}\rrbracket\mathbf{n}_{\Gamma}=-\lambda\kappa\mathbf{n}_{\Gamma} \tag*{(6)}
⟦u⟧=0(7)\llbracket\mathbf{u}\rrbracket=0 \tag*{(7)}
⟦J⋅nΓ⟧=0(8)\llbracket\mathbf{J}\cdot\mathbf{n}_{\Gamma}\rrbracket=0 \tag*{(8)}
⟦V⟧=0(9)\llbracket V\rrbracket=0 \tag*{(9)}
VΓ=u⋅nΓ(10)V_{\Gamma}=\mathbf{u}\cdot\mathbf{n}_{\Gamma} \tag*{(10)}

In this system, the unknowns are the fluid velocity u\mathbf{u}, the pressure pp, the current density J\mathbf{J}, the electric field E\mathbf{E} and the electrostatic potential VV. In the above equations, D(u)=12(∇u+(∇u)T)\mathbf{D}(\mathbf{u})=\frac{1}{2}(\nabla\mathbf{u}+(\nabla\mathbf{u})^{\mathsf{T}}) and I\mathbf{I} stand for the deformation and the identity tensor, respectively. g=(0,0,−g)T\mathbf{g}=(0,0,-g)^{\mathsf{T}} represents the gravitational acceleration with gg being a positive constant. B\mathbf{B} is the uniform external magnetic field with the magnitude being denoted by BrB_r. The material properties ρ±\rho_{\pm}, η±\eta_{\pm} and σ±\sigma_{\pm} are the constant densities, the constant dynamic viscosities, and the constant electrical conductivities of the two fluids, respectively. Moreover, we denote by λ\lambda the surface tension, κ\kappa the mean curvature, nΓ\mathbf{n}_{\Gamma} the unit normal on Γ(t)\Gamma(t) pointing to Ω+(t)\Omega_{+}(t), and VΓV_{\Gamma} the normal velocity of Γ(t)\Gamma(t). We use the notation η=η+XΩ+(t)+η−XΩ−(t)\eta=\eta_{+}\mathcal{X}_{\Omega_{+}(t)}+\eta_{-}\mathcal{X}_{\Omega_{-}(t)} with XΩ±(t)\mathcal{X}_{\Omega_{\pm}(t)} being the characteristic functions corresponding to Ω±(t)\Omega_{\pm}(t). Generally, ⟦χ⟧=χ+−χ−\llbracket\chi\rrbracket=\chi_{+}-\chi_{-} denotes the jump of quantity χ\chi across Γ(t)\Gamma(t) from Ω+(t)\Omega_{+}(t) to Ω−(t)\Omega_{-}(t). Equations (1)–(5) stand for the inductionless MHD equations, which can be employed to simulate the mutual interaction between the electromagnetic fields and the fluid flows in most laboratory experiments and industrial processes; see [5] for a physical justification. Meanwhile, equations (6)–(9) are the jump conditions across the interface Γ(t)\Gamma(t). Equation (10) models the movement of Γ(t)\Gamma(t), accounting for the immiscibility assumption.

Numerically solving the system (1) typically presents a significant challenge primarily due to the presence of the free surfaces and the jump conditions. To characterize the evolving interfaces of MHD flows, substantial numerical methodologies have been developed in the past decades. The level-set method [8, 20, 21] was one of the earliest and extensively utilized methods for MHD free surface simulations. In this approach, the interface is represented by a signed distance function, which attains a zero value on the interface. Several numerical techniques are employed to achieve a stable simulation, such as the re-initialization (or re-distancing) method and the continuum surface force model; see [12] for more details of these techniques. Particularly, in [8] the level-set algorithm only serves as an indicator of the interface, and an interpolation method was employed to construct a fitted mesh. Moreover, Zhang and Ni [36, 37, 38] employed the volume-of-fluid method to characterize the free surface. In this method, a volume fraction function is exploited to describe the interface, and some interface reconstruction techniques are required to maintain interfacial sharpness. Furthermore, moving mesh methods have been extensively applied in simulating MHD free surface problems. Gerbeau et al. [9] utilized an arbitrary Lagrangian–Eulerian approach for the interface motion and developed a structure-preserving scheme. Samulyak et al. [25] employed a front-tracking technique for capturing the interface. Pan et al. [22] exploited the immersed boundary method to accurately resolve the moving boundary. Despite the universal and successful application of the above algorithms in MHD free surface simulations, extra numerical efforts are usually required to obtain the stable simulation results, including the high-resolution interface [12].

In addition to the aforementioned approaches, the diffuse interface method [2] emerges as another effective tool for free surface modeling. This methodology regards the interface as a significantly thin transition layer where the fluid properties vary smoothly. The derived models inherently incorporate the thermodynamic structure and the energy stability property. Meanwhile, this approach naturally handles the topological changes of the interface. Correspondingly, it requires solving a fourth-order nonlinear partial differential equation, i.e., the Cahn–Hilliard equation. This numerical challenge can be efficiently addressed by the mixed finite element methods [31] and some linearizing techniques [26]. Recently, Yang et al. [35] proposed a diffuse interface model to simulate MHD problems and established the existence of the weak solutions. Subsequently, many studies have been devoted to the convergence analysis of numerical schemes for this model; see, e.g., [24, 27, 30, 34]. Additionally, Chen and Zhang [4] proposed another diffuse interface MHD model, with its existence of the weak solutions being established by Wang et al. [32]. These models possess two features: (i) they are not derived from the thermodynamic principles rather than a direct coupling of the Cahn–Hilliard equation with different MHD equations; and (ii) they are not serving as the approximations of the corresponding sharp interface MHD models such as the system (1). Therefore, these features limit their application in the MHD simulations for general industrial processes.

This paper firstly aims to propose a novel diffuse interface model which are equivalent to the system (1) under proper assumptions. The proposed model is rigorously derived from the thermodynamic principles [1] via the Onsager’s variational principle [23]. As a result, the incompressibility constraint remains and an extra flux is incorporated into the momentum equation to guarantee the thermodynamic consistency, in turn leading to an energy law that is not an obvious approximation of the sharp interface counterpart, see Remark 1 in § 2. Utilizing an effective parameter scaling law proposed by Magaletti et al. [19], and the method of formally matched asymptotic expansions [1, 33], we rigorously demonstrate that the system (1) can be recovered as the interface thickness tends to zero. This result validates the reliability of the diffuse interface model in practical MHD simulations on a theoretical level. Secondly, we develop an efficient backward differential formula of order two (BDF2) finite element scheme, incorporating the linearizing technique [26] and the pressure stabilization method [14] to achieve a decoupled and linear structure, to perform the numerical simulations. Meanwhile, such a scheme is charge-conservative (the current density is exactly solenoidal on the discrete level), in turn enabling an accurate MHD simulation [36]. Focusing on applications in metallurgy processes, we compute several three-dimensional numerical examples to explore the magnetic damping effects on bubble dynamics, demonstrating the robustness and applicability of the proposed model and method in realistic MHD simulations.

The remainder of this paper is organized as follows. In § 2 we present the diffuse interface model and discuss the sharp interface asymptotics. Then, in § 3, the discrete scheme is presented, and several numerical results are presented. Conclusions and future research directions are discussed in § 4.

Diffuse interface model

Governing equations

Considering the approximation of the sharp interface Γ(t)\Gamma(t) in the system (1) as an artificial interface with thickness ϵ\epsilon, we propose the following diffuse interface model:

∂tϕ+∇⋅(ϕu)=∇⋅(mϵ(ϕ)∇μ),(11)\partial_t\phi+\nabla\cdot(\phi\boldsymbol{u})=\nabla\cdot\left(m_\epsilon(\phi)\nabla\mu\right), \tag*{(11)}
μ=−λ^ϵΔϕ+λ^ϵ−1(ϕ3−ϕ),(12)\mu=-\widehat{\lambda}\epsilon\Delta\phi+\widehat{\lambda}\epsilon^{-1}\left(\phi^3-\phi\right), \tag*{(12)}
ρ(ϕ)∂tu+(ρ(ϕ)u−ρdmϵ(ϕ)∇μ)⋅∇u=∇⋅(2η(ϕ)D(u)−pI)+μ∇ϕ+J×B+ρg,(13)\rho(\phi)\partial_t\boldsymbol{u}+\left(\rho(\phi)\boldsymbol{u}-\rho_d m_\epsilon(\phi)\nabla\mu\right)\cdot\nabla\boldsymbol{u} =\nabla\cdot\left(2\eta(\phi)\mathbb{D}(\boldsymbol{u})-p\mathbb{I}\right)+\mu\nabla\phi+\boldsymbol{J}\times\boldsymbol{B}+\rho\boldsymbol{g}, \tag*{(13)}
∇⋅u=0,(14)\nabla\cdot\boldsymbol{u}=0, \tag*{(14)}
J=σ(ϕ)(E+u×B),(15)\boldsymbol{J}=\sigma(\phi)\left(\boldsymbol{E}+\boldsymbol{u}\times\boldsymbol{B}\right), \tag*{(15)}
E=−∇V,(16)\boldsymbol{E}=-\nabla V, \tag*{(16)}
∇⋅J=0.(17)\nabla\cdot\boldsymbol{J}=0. \tag*{(17)}

In the above system, ϕ\phi is the order parameter that labels the two fluids such that

ϕ(x,t)={1,x∈Ω+(t),−1,x∈Ω−(t).\phi(\boldsymbol{x},t)= \begin{cases} 1, & \boldsymbol{x}\in\Omega_+(t),\\ -1, & \boldsymbol{x}\in\Omega_-(t). \end{cases}

Additionally, μ\mu is the chemical potential serving as an auxiliary variable, mϵ(ϕ)>0m_\epsilon(\phi)>0 represents the mobility which plays the role of diffusivity for the diffuse interface, and λ^=3λ22\widehat{\lambda}=\frac{3\lambda}{2\sqrt{2}} stands for the scaled surface tension. Moreover, the density ρ\rho, the dynamic viscosity η\eta, and the electrical conductivity σ\sigma of the mixture are all linearly dependent on ϕ\phi such that (exemplarily for ρ\rho)

ρ=ρ+−ρ−2ϕ+ρ++ρ−2.\rho=\frac{\rho_+-\rho_-}{2}\phi+\frac{\rho_++\rho_-}{2}.

We define ρd:=∂ρ/∂ϕ=(ρ+−ρ−)/2\rho_d:=\partial\rho/\partial\phi=(\rho_+-\rho_-)/2 as the density difference between the two fluids. Equations (11)–(12) represent the widely recognized Cahn–Hilliard equation which describes the process of phase separation. It is necessary to point out that equation (16) represents merely a constitutive relation, and E\boldsymbol{E} will not be used in both the asymptotic analysis and the numerical computations.

To finalize the system, we impose some appropriate initial and boundary conditions. The initial data is set as

u(0)=u0andϕ(0)=ϕ0.\boldsymbol{u}(0)=\boldsymbol{u}^0 \qquad\text{and} \qquad\phi(0)=\phi^0.

We only consider the enclosed flows in this context, and therefore, no-slip walls are imposed for u\boldsymbol{u}, yielding

u=0on ∂Ω,\boldsymbol{u}=0 \qquad\text{on }\partial\Omega,

and insulating walls are applied for J\boldsymbol{J} and VV, leading to

J⋅n=0on ∂Ω,\boldsymbol{J}\cdot\boldsymbol{n}=0 \qquad\text{on }\partial\Omega,

where n\boldsymbol{n} is the unit outward normal to Ω\Omega. Mass conservation gives no-flux walls for ϕ\phi and μ\mu

∇ϕ⋅n=0on ∂Ω,\nabla\phi\cdot\boldsymbol{n}=0 \quad\text{on }\partial\Omega,
∇μ⋅n=0on ∂Ω.\nabla\mu\cdot\boldsymbol{n}=0 \quad\text{on }\partial\Omega.

Remark 1. It can be readily confirmed that the sharp interface model (1) satisfies the individual mass conservation and the energy dissipation law

ddt∫Ω±(t)ρ± dx=0,\frac{\mathrm{d}}{\mathrm{d}t}\int_{\Omega_{\pm}(t)}\rho_{\pm}\,\mathrm{d}x=0,
ddt(∫Ω12ρ∣u∣2 dx+∫Γ(t)λ ds+∫Ωρgz dx)=−∫Ω2η∣D(u)∣2 dx−∫Ωσ−1∣J∣2 dx≤0,\frac{\mathrm{d}}{\mathrm{d}t}\left(\int_{\Omega}\frac{1}{2}\rho|\boldsymbol{u}|^{2}\,\mathrm{d}x+\int_{\Gamma(t)}\lambda\,\mathrm{d}s+\int_{\Omega}\rho gz\,\mathrm{d}x\right)=-\int_{\Omega}2\eta|\mathbb{D}(\boldsymbol{u})|^{2}\,\mathrm{d}x-\int_{\Omega}\sigma^{-1}|\boldsymbol{J}|^{2}\,\mathrm{d}x\leq0,

with x=(x,y,z)T\boldsymbol{x}=(x,y,z)^{\mathsf{T}} and ρ=ρ+XΩ++ρ−XΩ−\rho=\rho_{+}\mathcal{X}_{\Omega_{+}}+\rho_{-}\mathcal{X}_{\Omega_{-}}. Nevertheless, for the proposed diffuse interface model (2), only the global mass conservation and the subsequent energy identity are valid,

ddt∫Ωϕ dx=0,\frac{\mathrm{d}}{\mathrm{d}t}\int_{\Omega}\phi\,\mathrm{d}x=0,
ddt(∫Ω12ρ∣u∣2 dx+∫Ωλ^ϵ2∣∇ϕ∣2 dx+∫Ωλ^ϵ−14(ϕ2−1)2 dx)=−∫Ωmϵ∣∇μ∣2 dx−∫Ω2η∣D(u)∣2 dx−∫Ωσ−1∣J∣2 dx+∫Ωρg⋅u dx.\frac{\mathrm{d}}{\mathrm{d}t}\left(\int_{\Omega}\frac{1}{2}\rho|\boldsymbol{u}|^{2}\,\mathrm{d}x+\int_{\Omega}\frac{\widehat{\lambda}\epsilon}{2}|\nabla\phi|^{2}\,\mathrm{d}x+\int_{\Omega}\frac{\widehat{\lambda}\epsilon^{-1}}{4}(\phi^{2}-1)^{2}\,\mathrm{d}x\right)=-\int_{\Omega}m_{\epsilon}|\nabla\mu|^{2}\,\mathrm{d}x-\int_{\Omega}2\eta|\mathbb{D}(\boldsymbol{u})|^{2}\,\mathrm{d}x-\int_{\Omega}\sigma^{-1}|\boldsymbol{J}|^{2}\,\mathrm{d}x+\int_{\Omega}\rho\boldsymbol{g}\cdot\boldsymbol{u}\,\mathrm{d}x.

The practical applications usually involve the gravity, and the energy of the diffuse interface model (2) is no longer dissipative as that of the sharp interface counterpart. This unexpected energy law fundamentally attributes to the global incompressibility constraint (14). Simultaneously, in order to guarantee the thermodynamic consistency at a theoretical level, an extra flux ρdmϵ∇μ\rho_{\mathrm{d}}m_{\epsilon}\nabla\mu must be incorporated into the momentum equation (13). In turn, this embarrasses the convergence analysis and the design of efficient structure-preserving schemes (a numerical demonstration on the influence of this flux can be found in [13]). However, the proposed model is capable of simulating MHD effects in general engineering processes, and remains a solenoidal velocity as well. In a contrast, interested readers are referred to [18] for another modeling approach leading to a much more complex quasi-incompressible diffuse interface model.

It is highly desirable to ensure the correct physical behavior for the simulation of diffuse interface models, compared to that of the sharp interface system (1). In other words, whether the standard sharp interface formulation (1) can be recovered from the diffuse interface model (2) as the interface thickness ϵ\epsilon approaches zero becomes a critical issue. In addition, it is known that the scaling between ϵ\epsilon and the mobility mϵm_{\epsilon} is also critical during the simulations. In this study, we take a constant mobility with a scaling law mϵ=O(ϵ2)m_{\epsilon}=O(\epsilon^{2}) employed in [19, 17, 31]. This scaling law facilitates the asymptotic analysis by making the extra flux term mϵ∇μm_{\epsilon}\nabla\mu in the momentum equation a higher-order contribution with respect to ϵ\epsilon. Consequently, this term does not affect the leading-order dynamics, ensuring that the correct physical behavior is preserved in the sharp interface asymptotics.

Theorem 1. Assuming the scaling law mϵ=O(ϵ2)m_{\epsilon}=O(\epsilon^{2}), the sharp interface limit of model (2) is system (1).

The derivation of the diffuse interface model (2), and the corresponding asymptotic analysis on the relationship between the system (2) and (1) will be given in the following two subsections, respectively.

Model derivation

In this subsection, we present the derivation of the system (2) by means of the Onsager’s variational principle and the laws of thermodynamics. First, we assume a partial mixing of the two fluids in a narrow interfacial zone of thickness ϵ\epsilon with the incompressibility on the volume. Let V(t)⊂ΩV(t)\subset\Omega be an arbitrary control volume with its mass and density being given by MM and ρ\rho. Then, denoted by M±M_{\pm} and ρ±\rho_{\pm} the masses and densities of the two fluids in VV, it holds that

ρ±=M±Vandρ=MV=M++M−V=ρ++ρ−.\rho_{\pm}=\frac{M_{\pm}}{V}\quad\text{and}\quad\rho=\frac{M}{V}=\frac{M_{+}+M_{-}}{V}=\rho_{+}+\rho_{-}.

In turn, the volume incompressibility assumption leads to

1=M+ρ+V+M−ρ−V=ρ+ρ++ρ−ρ−.1 = \frac{M_{+}}{\rho_{+}V}+\frac{M_{-}}{\rho_{-}V}=\frac{\rho_{+}}{\rho_{+}}+\frac{\rho_{-}}{\rho_{-}}.

We define the order parameter for the mixture as

ϕ=ρ+ρ+−ρ−ρ−,\phi=\frac{\rho_{+}}{\rho_{+}}-\frac{\rho_{-}}{\rho_{-}},

which indicates that

ρ=ρ+−ρ−2ϕ+ρ++ρ−2.\rho=\frac{\rho_{+}-\rho_{-}}{2}\phi+\frac{\rho_{+}+\rho_{-}}{2}.

Supposing that the fluids move with different velocities u±\boldsymbol{u}_{\pm}, the continuity equations in the bulk read

∂tϱ±+∇⋅(ϱ±u±)=0.\partial_t\varrho_{\pm}+\nabla\cdot(\varrho_{\pm}\boldsymbol{u}_{\pm})=0.

Then we introduce the volume-averaged velocity and relative mass flux as

u=ρ+ρ++ρ−u++ρ−ρ++ρ−u−andj±=ϱ±(u±−u).\boldsymbol{u}=\frac{\rho_{+}}{\rho_{+}+\rho_{-}}\boldsymbol{u}_{+}+\frac{\rho_{-}}{\rho_{+}+\rho_{-}}\boldsymbol{u}_{-} \qquad\text{and}\qquad \boldsymbol{j}_{\pm}=\varrho_{\pm}(\boldsymbol{u}_{\pm}-\boldsymbol{u}).

As a consequence, we obtain

∇⋅u=∇⋅(ρ+ρ+u++ρ−ρ−u−)=−∂t(ρ+ρ++ρ−ρ−)=0,\nabla\cdot\boldsymbol{u}=\nabla\cdot\left(\frac{\rho_{+}}{\rho_{+}}\boldsymbol{u}_{+}+\frac{\rho_{-}}{\rho_{-}}\boldsymbol{u}_{-}\right)=-\partial_t\left(\frac{\rho_{+}}{\rho_{+}}+\frac{\rho_{-}}{\rho_{-}}\right)=0,

and

∂tϕ=−∇⋅(ρ+ρ+u+j+ρ+−ρ−ρ−u−j−ρ−)=−∇⋅(ϕu+jϕ),\partial_t\phi=-\nabla\cdot\left(\frac{\rho_{+}}{\rho_{+}}\boldsymbol{u}+\frac{\boldsymbol{j}_{+}}{\rho_{+}}-\frac{\rho_{-}}{\rho_{-}}\boldsymbol{u}-\frac{\boldsymbol{j}_{-}}{\rho_{-}}\right)=-\nabla\cdot(\phi\boldsymbol{u}+\boldsymbol{j}_{\phi}),

where jϕ=j+ρ+−j−ρ−\boldsymbol{j}_{\phi}=\frac{\boldsymbol{j}_{+}}{\rho_{+}}-\frac{\boldsymbol{j}_{-}}{\rho_{-}}. Correspondingly, this implies that the mass diffusion is permitted in our method. Hence, the momentum ρu\rho\boldsymbol{u} should be transported by ρu+ρdjϕ\rho\boldsymbol{u}+\rho_d\boldsymbol{j}_{\phi}, giving

ρ∂tu+(ρu+ρdjϕ)⋅∇u=∇⋅S−∇p+F.\rho\partial_t\boldsymbol{u}+(\rho\boldsymbol{u}+\rho_d\boldsymbol{j}_{\phi})\cdot\nabla\boldsymbol{u}=\nabla\cdot\boldsymbol{S}-\nabla p+\boldsymbol{F}.

where S\boldsymbol{S} is a symmetric second-order tensor, pp is the pressure serving as a Lagrange multiplier for the incompressibility constraint, and F\boldsymbol{F} is a force density. Invoking Ohm’s law with the charge conservation, we conclude

∂tϕ+∇⋅(ϕu+jϕ)=0,ρ∂tu+(ρu+ρdjϕ)⋅∇u=∇⋅S−∇p+F,∇⋅u=0,J=σ(ϕ)(E+u×B),E=−∇V,∇⋅J=0.\begin{aligned} \partial_t\phi+\nabla\cdot(\phi\boldsymbol{u}+\boldsymbol{j}_{\phi})&=0,\\ \rho\partial_t\boldsymbol{u}+(\rho\boldsymbol{u}+\rho_d\boldsymbol{j}_{\phi})\cdot\nabla\boldsymbol{u}&=\nabla\cdot\boldsymbol{S}-\nabla p+\boldsymbol{F},\\ \nabla\cdot\boldsymbol{u}&=0,\\ \boldsymbol{J}&=\sigma(\phi)(\boldsymbol{E}+\boldsymbol{u}\times\boldsymbol{B}),\\ \boldsymbol{E}&=-\nabla V,\\ \nabla\cdot\boldsymbol{J}&=0. \end{aligned}

where jϕ\boldsymbol{j}_{\phi}, S\boldsymbol{S}, and F\boldsymbol{F} are to be determined, and the definition of σ(ϕ)\sigma(\phi) will be given later. To derive a thermodynamically consistent model, we employ the Onsager’s variational principle as in [1]. By introducing the Rayleighian functional

R=dEdt+P,\mathcal{R}=\frac{\mathrm{d}\mathcal{E}}{\mathrm{d}t}+\mathcal{P},

where E\mathcal{E} and P\mathcal{P} are the free energy and dissipation functional, the Onsager’s variational principle states that the kinetic equation is equivalent to the minimum of R\mathcal{R} [23]. Then, we define

E=∫Ω(12ρ∣u∣2+f(ϕ,∇ϕ)) dx,\mathcal{E}=\int_{\Omega}\left(\frac{1}{2}\rho\lvert\boldsymbol{u}\rvert^2+f(\phi,\nabla\phi)\right)\,\mathrm{d}x,

where f(ϕ,∇ϕ)f(\phi,\nabla\phi) is a Helmholtz free energy density. Thus, it holds that

dEdt=∫Ω(−S:D(u)+u⋅F−μu⋅∇ϕ+jϕ⋅∇μ) dx,\frac{\mathrm{d}\mathcal{E}}{\mathrm{d}t}=\int_{\Omega}\left(-\boldsymbol{S}:\mathsf{D}(\boldsymbol{u})+\boldsymbol{u}\cdot\boldsymbol{F}-\mu\boldsymbol{u}\cdot\nabla\phi+\boldsymbol{j}_{\phi}\cdot\nabla\mu\right)\,\mathrm{d}x,

where μ=∂f∂ϕ−∇⋅∂f∂∇ϕ\mu= \frac{\partial f}{\partial\phi} - \nabla\cdot\frac{\partial f}{\partial\nabla\phi} is abbreviated to denote the chemical potential, and here we assume a vanishing normal component of ∂f∂∇ϕ\frac{\partial f}{\partial\nabla\phi} on ∂Ω\partial\Omega. Since the first law of thermodynamics reads

dEdt=−TdSdt+dWdt,\frac{\mathrm{d}\mathcal{E}}{\mathrm{d}t} = -\mathcal{T}\frac{\mathrm{d}S}{\mathrm{d}t} + \frac{\mathrm{d}\mathcal{W}}{\mathrm{d}t},

where T\mathcal{T}, SS, and W\mathcal{W} represent the temperature, entropy, and mechanical work, we are able to identify

dWdt=∫Ω(u⋅F−μu⋅∇ϕ) dx.\frac{\mathrm{d}\mathcal{W}}{\mathrm{d}t} = \int_{\Omega} \left(\boldsymbol{u} \cdot\boldsymbol{F} - \mu\boldsymbol{u} \cdot\nabla\phi\right)\,\mathrm{d}x.

Incorporating external gravitational and Lorentz forces, we can write F\boldsymbol{F} as

F=μ∇ϕ+J×B+ρg.\boldsymbol{F} = \mu\nabla\phi+ \boldsymbol{J} \times\boldsymbol{B} + \rho\boldsymbol{g}.

To further specify S\boldsymbol{S} and jϕj_{\phi}, we introduce the dissipation functional

P=∫Ω(∣S∣24η(ϕ)+∣jϕ∣22mϵ(ϕ)+∣J∣22σ(ϕ)) dx,\mathcal{P} = \int_{\Omega} \left(\frac{|\boldsymbol{S}|^{2}}{4\eta(\phi)} + \frac{|j_{\phi}|^{2}}{2m_{\epsilon}(\phi)} + \frac{|\boldsymbol{J}|^{2}}{2\sigma(\phi)}\right)\,\mathrm{d}x,

which accounts for the viscous dissipation, the diffusive dissipation, and the Ohmic dissipation, respectively. In accordance to the Onsager’s variational principle, we get

S=2ηD(u)andjϕ=−mϵ∇μ.\boldsymbol{S} = 2\eta\boldsymbol{D}(\boldsymbol{u}) \qquad\text{and} \qquad j_{\phi} = -m_{\epsilon}\nabla\mu.

In this work, we consider the Helmholtz free energy density in the Ginzburg-Landau form:

f(ϕ,∇ϕ)=λ^ϵ2∣∇ϕ∣2+λ^ϵ−14(ϕ2−1)2.f(\phi,\nabla\phi) = \frac{\widehat{\lambda}_{\epsilon}}{2}|\nabla\phi|^{2} + \frac{\widehat{\lambda}_{\epsilon}^{-1}}{4}(\phi^{2}-1)^{2}.

Moreover, we assume that the material properties vary linearly in the mixture, i.e.,

η=η+−η−2ϕ+η++η−2andσ=σ+−σ−2ϕ+σ++σ−2,\eta= \frac{\eta_{+}-\eta_{-}}{2}\phi+ \frac{\eta_{+}+\eta_{-}}{2} \qquad\text{and} \qquad \sigma= \frac{\sigma_{+}-\sigma_{-}}{2}\phi+ \frac{\sigma_{+}+\sigma_{-}}{2},

so that the model is compatible with the inductionless MHD equations in the single fluid regime. This finally leads to the system (2).

The proof of Theorem 1

We exploit the method of formally matched asymptotic expansions to identify the sharp interface limit of system (2). This argument depends on two critical assumptions: (i) with vanishing interface thickness the domain can be decomposed into two distinct subdomains separated by an interface, each containing only one fluid; and (ii) the solutions admit separate asymptotic expansions in terms of the interface thickness in the bulk and near the interface, which must be matched in the overlapping region.

Preliminaries

For a solution (ϕϵ,μϵ,uϵ,pϵ,Jϵ,Vϵ)(\phi_{\epsilon}, \mu_{\epsilon}, \boldsymbol{u}_{\epsilon}, p_{\epsilon}, \boldsymbol{J}_{\epsilon}, V_{\epsilon}) of system (2), we assume that it converges formally to a limit (ϕ,μ,u,p,J,V)(\phi,\mu,\boldsymbol{u},p,\boldsymbol{J},V) as the interface thickness ϵ→0\epsilon\to0, and further,

Ω+(t)={x∈Ω:ϕϵ(x)>0}andΩ−(t)={x∈Ω:ϕϵ(x)<0}.\Omega_{+}(t) = \{x \in\Omega: \phi_{\epsilon}(x) > 0\} \qquad\text{and} \qquad \Omega_{-}(t) = \{x \in\Omega: \phi_{\epsilon}(x) < 0\}.

Then, the corresponding interface separating Ω±(t)\Omega_{\pm}(t) is denoted by

Γ(t)={x∈Ω:ϕϵ(x)=0}.\Gamma(t) = \{x \in\Omega: \phi_{\epsilon}(x) = 0\}.

In outer regions far from Γ(t)\Gamma(t), there holds the outer expansions with respect to the interface thickness ϵ\epsilon of ϕϵ\phi_{\epsilon}, uϵ\boldsymbol{u}_{\epsilon}, pϵp_{\epsilon}, Jϵ\boldsymbol{J}_{\epsilon}, VϵV_{\epsilon} in the form (exemplarily for ϕϵ\phi_{\epsilon})

ϕϵ=ϕ0±+ϵϕ1±+ϵ2ϕ2±+⋯ .\phi_{\epsilon} = \phi_{0}^{\pm} + \epsilon\phi_{1}^{\pm} + \epsilon^{2}\phi_{2}^{\pm} + \cdots.

As an auxiliary variable, we have evidently for μ\mu:

μϵ=ϵ−1μ0±+μ1±+ϵμ2±+⋯\mu_{\epsilon} = \epsilon^{-1}\mu_{0}^{\pm} + \mu_{1}^{\pm} + \epsilon\mu_{2}^{\pm} + \cdots

with μ0±=λ^((ϕ0±)3−ϕ0±)\mu_0^{\pm}=\hat{\lambda}\left((\phi_0^{\pm})^3-\phi_0^{\pm}\right). In the transition layer near Γ(t)\Gamma(t), we introduce a new stretched coordinates to scale the variables. Let d(x)d(x) be the signed distance function to Γ(t)\Gamma(t) such that d(x)>0d(x)>0 if x∈Ω+x\in\Omega_+, then we have nΓ=∇d\mathbf{n}_{\Gamma}=\nabla d and VΓ=−∂tdV_{\Gamma}=-\partial_t d. Defining the scaled distance z=d/ϵz=d/\epsilon, the following change of variables holds [1]:

∂tχ=−ϵ−1VΓ∂zχ+O(1),(18)\partial_t\chi=-\epsilon^{-1}V_{\Gamma}\partial_z\chi+O(1), \tag*{(18)}
∇χ=ϵ−1∂zχnΓ+∇Γχ+O(ϵ),(19)\nabla\chi=\epsilon^{-1}\partial_z\chi\mathbf{n}_{\Gamma}+\nabla_{\Gamma}\chi+O(\epsilon), \tag*{(19)}
∇⋅χ=ϵ−1∂zχ⋅nΓ+∇Γ⋅χ+O(ϵ),(20)\nabla\cdot\boldsymbol{\chi}=\epsilon^{-1}\partial_z\boldsymbol{\chi}\cdot\mathbf{n}_{\Gamma}+\nabla_{\Gamma}\cdot\boldsymbol{\chi}+O(\epsilon), \tag*{(20)}
Δχ=ϵ−2∂zzχ−ϵ−1κ∂zχ+O(1),(21)\Delta\chi=\epsilon^{-2}\partial_{zz}\chi-\epsilon^{-1}\kappa\partial_z\chi+O(1), \tag*{(21)}

where χ\chi and χ\boldsymbol{\chi} represent the generic scalar-valued and vector-valued variables, respectively. Then in this region we have the inner expansions in terms of the interface thickness ϵ\epsilon for ϕϵ\phi_{\epsilon}, uϵ\mathbf{u}_{\epsilon}, pϵp_{\epsilon}, Jϵ\mathbf{J}_{\epsilon}, VϵV_{\epsilon} in the form (also exemplarily for ϕϵ\phi_{\epsilon})

ϕϵ=ϕ0i+ϵϕ1i+ϵ2ϕ2i+⋯ ,\phi_{\epsilon}=\phi_0^i+\epsilon\phi_1^i+\epsilon^2\phi_2^i+\cdots,

with μϵ=ϵ−1μ0i+μ1i+ϵμ2i+⋯\mu_{\epsilon}=\epsilon^{-1}\mu_0^i+\mu_1^i+\epsilon\mu_2^i+\cdots, where

μ0i=λ^(−∂zzϕ0i+(ϕ0i)3−ϕ0i)andμ1i=λ^(−∂zzϕ1i+κ∂zϕ0i+(3(ϕ0i)2−1)ϕ1i).\mu_0^i=\hat{\lambda}\left(-\partial_{zz}\phi_0^i+(\phi_0^i)^3-\phi_0^i\right) \qquad\text{and}\qquad \mu_1^i=\hat{\lambda}\left(-\partial_{zz}\phi_1^i+\kappa\partial_z\phi_0^i+\left(3(\phi_0^i)^2-1\right)\phi_1^i\right).

In the overlapping region, the outer and inner expansions must satisfy the matching conditions [1]

lim⁡z→±∞χ0i=χ0±,(22)\lim_{z\to\pm\infty}\chi_0^i=\chi_0^{\pm}, \tag*{(22)}
lim⁡z→±∞∂zχ0i=0,(23)\lim_{z\to\pm\infty}\partial_z\chi_0^i=0, \tag*{(23)}
lim⁡z→±∞∂zχ1i=∇χ0±⋅nΓ.(24)\lim_{z\to\pm\infty}\partial_z\chi_1^i=\nabla\chi_0^{\pm}\cdot\mathbf{n}_{\Gamma}. \tag*{(24)}

Next, we will establish that the limit solution (ϕ,μ,u,p,J,V)(\phi,\mu,\mathbf{u},p,\mathbf{J},V) exactly satisfies the sharp interface model (1) by following the procedure in [1, 33].

Bulk equations

Expansion of (13) at order O(ϵ−1)O(\epsilon^{-1}): Recalling the assumption mϵ:=mϵ2m_{\epsilon}:=m\epsilon^2, we naturally obtain that the leading order of the flux mϵ∇μm_{\epsilon}\nabla\mu is O(ϵ)O(\epsilon). Hence, we arrive at

μ0±∇ϕ0±=0,\mu_0^{\pm}\nabla\phi_0^{\pm}=0,

which is equivalent to

∇((ϕ0±)2−1)24=0.\nabla\frac{\left((\phi_0^{\pm})^2-1\right)^2}{4}=0.

This implies that ϕ0±=c±\phi_0^{\pm}=c_{\pm} with c±c_{\pm} two constants such that c+>0c_+>0 and c−<0c_-<0.

Expansion of (13)–(17) at order O(1)O(1): Using the fact that mϵ∇μm_{\epsilon}\nabla\mu is of order O(ϵ)O(\epsilon), we conclude

ρ0±(∂tu0±+u0±⋅∇u0±)=∇⋅(2η0±D(u0±)−∇p0±I)+μ0±∇ϕ1±+J0±×B+ρ0±g,∇⋅u0±=0,J0±=σ0±(E±+u0±×B),E±=−∇V0±,∇⋅J0±=0.(25)\begin{aligned} \rho_0^{\pm}\left(\partial_t\mathbf{u}_0^{\pm}+\mathbf{u}_0^{\pm}\cdot\nabla\mathbf{u}_0^{\pm}\right) &=\nabla\cdot\left(2\eta_0^{\pm}\mathbb{D}(\mathbf{u}_0^{\pm})-\nabla p_0^{\pm}\mathbb{I}\right)+\mu_0^{\pm}\nabla\phi_1^{\pm}+\mathbf{J}_0^{\pm}\times\mathbf{B}+\rho_0^{\pm}\mathbf{g},\\ \nabla\cdot\mathbf{u}_0^{\pm}&=0,\\ \mathbf{J}_0^{\pm}&=\sigma_0^{\pm}\left(\mathbf{E}^{\pm}+\mathbf{u}_0^{\pm}\times\mathbf{B}\right),\\ \mathbf{E}^{\pm}&=-\nabla V_0^{\pm},\\ \nabla\cdot\mathbf{J}_0^{\pm}&=0. \tag*{(25)} \end{aligned}

where we abbreviate χ0±=χ+−χ−2ϕ0±+χ++χ−2\chi_0^{\pm}=\frac{\chi_+-\chi_-}{2}\phi_0^{\pm}+\frac{\chi_++\chi_-}{2} for χ∈{ρ,η,σ}\chi\in\{\rho,\eta,\sigma\}.

Jump conditions

Expansion of (14) at order O(ϵ−1)O(\epsilon^{-1}): Invoking change of variables (4), we get

∂zu0i⋅nΓ=0,(26)\partial_z\mathbf{u}_0^i\cdot\mathbf{n}_{\Gamma}=0, \tag*{(26)}

which implies that u0i⋅nΓ\mathbf{u}_0^i\cdot\mathbf{n}_{\Gamma} is independent of $z.

Expansion of (2a) at order O(ϵ−1)O(\epsilon^{-1}): With a similar procedure, we obtain

−VΓ∂zϕ0i+∂z(ϕ0iu0i)⋅nΓ=m∂zzμ0i.-V_{\Gamma}\partial_z\phi_0^i+\partial_z\left(\phi_0^i\mathbf{u}_0^i\right)\cdot\mathbf{n}_{\Gamma}=m\partial_{zz}\mu_0^i.

Integrating this equation with respect to zz from −∞-\infty to ∞\infty yields

(−VΓ+u0i⋅nΓ)∫−∞∞∂zϕ0i dz=m∫−∞∞∂zzμ0i dz.\left(-V_{\Gamma}+\mathbf{u}_0^i\cdot\mathbf{n}_{\Gamma}\right)\int_{-\infty}^{\infty}\partial_z\phi_0^i\,\mathrm{d}z=m\int_{-\infty}^{\infty}\partial_{zz}\mu_0^i\,\mathrm{d}z.

Then matching conditions (22)-(23) requires

(−VΓ+u0i⋅nΓ)(ϕ0+−ϕ0−)=0,\left(-V_{\Gamma}+\mathbf{u}_0^i\cdot\mathbf{n}_{\Gamma}\right)\left(\phi_0^+-\phi_0^-\right)=0,

in turn giving

VΓ=u0i⋅nΓ.(27)V_{\Gamma}=\mathbf{u}_0^i\cdot\mathbf{n}_{\Gamma}. \tag*{(27)}

Additionally, we have

∂zzμ0i=0,\partial_{zz}\mu_0^i=0,

or equivalently, μ0i\mu_0^i linearly depends on zz. Required by matching condition (22), we conclude that μ0i\mu_0^i has to be a constant.

Expansion of (2c) at order O(ϵ−2)O(\epsilon^{-2}): Denoting by D(A)=12(A+AT)D(A)=\frac{1}{2}(A+A^{\mathsf{T}}), we obtain after the change of variables (4),

∇⋅(2ηD(u))=ϵ−2∂z(2ηD(∂zu⊗nΓ)nΓ)+ϵ−1∂z(2ηD(∇Γu)nΓ)+ϵ−1∇Γ⋅(2ηD(∂zu⊗nΓ)+O(1).(28)\nabla\cdot(2\eta D(\mathbf{u}))=\epsilon^{-2}\partial_z(2\eta D(\partial_z\mathbf{u}\otimes\mathbf{n}_{\Gamma})\mathbf{n}_{\Gamma})+\epsilon^{-1}\partial_z(2\eta D(\nabla_{\Gamma}\mathbf{u})\mathbf{n}_{\Gamma})+\epsilon^{-1}\nabla_{\Gamma}\cdot(2\eta D(\partial_z\mathbf{u}\otimes\mathbf{n}_{\Gamma})+O(1). \tag*{(28)}

Notice that mϵ∇μm\epsilon\nabla\mu is of order O(ϵ)O(\epsilon) in the inner region since μ0i\mu_0^i is independent of zz, we conclude

∂z(η0i∂zu0i)+μ0i∂zϕ0inΓ=0,\partial_z\left(\eta_0^i\partial_z\mathbf{u}_0^i\right)+\mu_0^i\partial_z\phi_0^i\mathbf{n}_{\Gamma}=0,

where we have used

(nΓ⊗∂zu0i)nΓ=(∂zu0i⋅nΓ)nΓ=0.\left(\mathbf{n}_{\Gamma}\otimes\partial_z\mathbf{u}_0^i\right)\mathbf{n}_{\Gamma}=\left(\partial_z\mathbf{u}_0^i\cdot\mathbf{n}_{\Gamma}\right)\mathbf{n}_{\Gamma}=0.

By integrating this equation with respect to zz, we can obtain

η0i∂zu0i∣−∞∞+μ0iϕ0i∣−∞∞nΓ=0.\left.\eta_0^i\partial_z\mathbf{u}_0^i\right|_{-\infty}^{\infty}+\left.\mu_0^i\phi_0^i\right|_{-\infty}^{\infty}\mathbf{n}_{\Gamma}=0.

Through equation (26), taking an inner product with nΓ\mathbf{n}_{\Gamma} and matching condition (22) yield

μ0i(ϕ0+−ϕ0−)=0,\mu_0^i\left(\phi_0^+-\phi_0^-\right)=0,

in turn indicating μ0i=0\mu_0^i=0, or equivalently

−∂zzϕ0i+(ϕ0i)3−ϕ0i=0.-\partial_{zz}\phi_0^i+\left(\phi_0^i\right)^3-\phi_0^i=0.

The solvability of this equation [33] leads to

ϕ0±=lim⁡z→±∞ϕ0i=±1andϕ0i=tanh⁡z2.\phi_0^{\pm}=\lim_{z\to\pm\infty}\phi_0^i=\pm1\quad\text{and}\quad\phi_0^i=\tanh\frac{z}{\sqrt{2}}.

Then we have a second-order ordinary differential equation

∂z(η0i∂zu0i)=0,\partial_z\left(\eta_0^i\partial_z\mathbf{u}_0^i\right)=0,

which, by matching conditions (22)-(23) and positivity of η\eta, implies that u0i\mathbf{u}_0^i is independent of zz. Hence, we obtain

⟦u0⟧=0.\llbracket\mathbf{u}_0\rrbracket=0.

Expansion of (2c) at order O(ϵ−1)O(\epsilon^{-1}): The combination of matching conditions (22) and (24) yields

lim⁡z→±∞(∂zu1i⊗nΓ+∇Γu0i)=∇u0±.(29)\lim_{z\to\pm\infty}\left(\partial_z\mathbf{u}_1^i\otimes\mathbf{n}_{\Gamma}+\nabla_{\Gamma}\mathbf{u}_0^i\right)=\nabla\mathbf{u}_0^{\pm}. \tag*{(29)}

Notice that u0i\mathbf{u}_0^i and μ0i\mu_0^i are indeed constants, we can derive

∂z(2η0iD(∂zu1i⊗nΓ)nΓ)+∂z(2η0iD(∇Γu0i)nΓ)−∂zp0inΓ+μ1i∂zϕ0inΓ=0.\partial_z\left(2\eta_0^iD(\partial_z\mathbf{u}_1^i\otimes\mathbf{n}_{\Gamma})\mathbf{n}_{\Gamma}\right)+\partial_z\left(2\eta_0^iD(\nabla_{\Gamma}\mathbf{u}_0^i)\mathbf{n}_{\Gamma}\right)-\partial_zp_0^i\mathbf{n}_{\Gamma}+\mu_1^i\partial_z\phi_0^i\mathbf{n}_{\Gamma}=0.

Then integrating and using matching conditions (22) and (29) yield

⟦2η0±D(u0±)−p0±I⟧nΓ=−∫−∞∞μ1i∂zϕ0inΓ dz.\llbracket2\eta_0^{\pm}\mathbb{D}(u_0^{\pm})-p_0^{\pm}\mathbb{I}\rrbracket n_{\Gamma}=-\int_{-\infty}^{\infty}\mu_1^i\partial_z\phi_0^i n_{\Gamma}\,\mathrm{d}z.

For the right-hand side of this equation, integration by parts with matching conditions (22)–(23) leads to

∫−∞∞μ1i∂zϕ0i dz=λ^∫−∞∞(−∂zzϕ1i+κ∂zϕ0i+(3(ϕ0i)2−1)ϕ1i)∂zϕ0i dz,=λ^κ∫−∞∞(∂zϕ0i)2dz+λ^∫−∞∞(−∂zzϕ1i+(3(ϕ0i)2−1)ϕ1i)∂zϕ0i dz=22λ^κ3+λ^∫−∞∞(∂zzϕ0i−((ϕ0i)3+ϕ0i))∂zϕ1idz=λκ,\begin{aligned} \int_{-\infty}^{\infty}\mu_1^i\partial_z\phi_0^i\,\mathrm{d}z &=\hat{\lambda}\int_{-\infty}^{\infty}\left(-\partial_{zz}\phi_1^i+\kappa\partial_z\phi_0^i+\left(3(\phi_0^i)^2-1\right)\phi_1^i\right)\partial_z\phi_0^i\,\mathrm{d}z,\\ &=\hat{\lambda}\kappa\int_{-\infty}^{\infty}(\partial_z\phi_0^i)^2\mathrm{d}z+\hat{\lambda}\int_{-\infty}^{\infty}\left(-\partial_{zz}\phi_1^i+\left(3(\phi_0^i)^2-1\right)\phi_1^i\right)\partial_z\phi_0^i\,\mathrm{d}z\\ &=\frac{2\sqrt{2}\hat{\lambda}\kappa}{3}+\hat{\lambda}\int_{-\infty}^{\infty}\left(\partial_{zz}\phi_0^i-\left((\phi_0^i)^3+\phi_0^i\right)\right)\partial_z\phi_1^i\mathrm{d}z\\ &=\lambda\kappa, \end{aligned}

where

∫−∞∞(∂zϕ0i)2dz=12∫−∞∞sech⁡4z2 dz=223.\int_{-\infty}^{\infty}(\partial_z\phi_0^i)^2\mathrm{d}z=\frac{1}{2}\int_{-\infty}^{\infty}\operatorname{sech}^4\frac{z}{\sqrt{2}}\,\mathrm{d}z=\frac{2\sqrt{2}}{3}.

Then we conclude

⟦2η0±D(u0±)−p0±I⟧nΓ=−λκ.\llbracket2\eta_0^{\pm}\mathbb{D}(u_0^{\pm})-p_0^{\pm}\mathbb{I}\rrbracket n_{\Gamma}=-\lambda\kappa.

Expansion of (15)–(17) at order O(ϵ−1)\mathcal{O}(\epsilon^{-1}): A direct change of variables (4) gives

∂zϕ0inΓ=0,\partial_z\phi_0^i n_{\Gamma}=0,
∂zJ0i⋅nΓ=0.\partial_z J_0^i\cdot n_{\Gamma}=0.

With matching condition (22), we obtain

⟦V0±⟧=0,\llbracket V_0^{\pm}\rrbracket=0,
⟦J0±⋅nΓ⟧=0.\llbracket J_0^{\pm}\cdot n_{\Gamma}\rrbracket=0.

Summary of the sharp interface limit

We summarize the main results as follows. For the region far from Γ(t)\Gamma(t), we have ϕ0±=±1\phi_0^{\pm}=\pm1 and μ0±=0\mu_0^{\pm}=0. Then, the bulk equations (25) can be simplified as

ρ±(∂tu0±+u0±⋅∇u0±)=∇⋅(2η0±D(u0±)−p0±I)+J0±×B+ρ0±g,\rho_{\pm}\left(\partial_t u_0^{\pm}+u_0^{\pm}\cdot\nabla u_0^{\pm}\right)=\nabla\cdot\left(2\eta_0^{\pm}\mathbb{D}(u_0^{\pm})-p_0^{\pm}\mathbb{I}\right)+J_0^{\pm}\times B+\rho_0^{\pm}g,
∇⋅u0±=0,\nabla\cdot u_0^{\pm}=0,
J0±=σ±(E±+u0±×B),J_0^{\pm}=\sigma_{\pm}\left(E^{\pm}+u_0^{\pm}\times B\right),
E±=−∇V0±,E^{\pm}=-\nabla V_0^{\pm},
∇⋅J0±=0,\nabla\cdot J_0^{\pm}=0,

and the jump conditions are

⟦2η0±D(u0±)−p0±I⟧nΓ=−λκ,\llbracket2\eta_0^{\pm}\mathbb{D}(u_0^{\pm})-p_0^{\pm}\mathbb{I}\rrbracket n_{\Gamma}=-\lambda\kappa,
⟦u0±⟧=0,\llbracket u_0^{\pm}\rrbracket=0,
⟦J0±⋅nΓ⟧=0,\llbracket J_0^{\pm}\cdot n_{\Gamma}\rrbracket=0,
⟦V0±⟧=0,\llbracket V_0^{\pm}\rrbracket=0,
VΓ=u0i⋅nΓ.V_{\Gamma}=u_0^i\cdot n_{\Gamma}.

We can deduce that the limit system is the sharp interface model (1) as the interface thickness ϵ\epsilon approaches zero.

Numerical results

Non-dimensionalization and discrete scheme

Let LrL_r, uru_r, ρr\rho_r, ηr\eta_r, σr\sigma_r denote the characteristic quantities of length, velocity, density, dynamic viscosity, and electrical conductivity, respectively. We introduce the following scalings:

x←xLr,t←turLr,u←uur,ρ←ρρr,η←ηηr,p←pρrur2,x \leftarrow\frac{x}{L_r},\quad t \leftarrow\frac{t u_r}{L_r},\quad u \leftarrow\frac{u}{u_r},\quad\rho\leftarrow\frac{\rho}{\rho_r},\quad\eta\leftarrow\frac{\eta}{\eta_r},\quad p \leftarrow\frac{p}{\rho_r u_r^2},
μ←μϵλ^,V←VLrurBr,J←JσrurBr,σ←σσr,E←EurBr.\mu\leftarrow\frac{\mu\epsilon}{\widehat{\lambda}},\quad V \leftarrow\frac{V}{L_r u_r B_r},\quad J \leftarrow\frac{J}{\sigma_r u_r B_r},\quad\sigma\leftarrow\frac{\sigma}{\sigma_r},\quad E \leftarrow\frac{E}{u_r B_r}.

Certainly, the dimensionless system is given by

∂tϕ+∇⋅(ϕu)=1Pe⁡Δμ,μ=−Cn⁡2Δϕ+ϕ3−ϕ,ρ∂tu+(ρu−ρd1Pe⁡∇μ)⋅∇u=1Re⁡∇⋅(2ηD(u))−∇p+1We⁡Cn⁡μ∇ϕ+NJ×B−1Fr⁡ρe3,∇⋅u=0,J=σ(E+u×B),E=−∇V,∇⋅J=0.\begin{aligned} \partial_t\phi+\nabla\cdot(\phi u)&=\frac{1}{\operatorname{Pe}}\Delta\mu,\\ \mu&=-\operatorname{Cn}^2\Delta\phi+\phi^3-\phi,\\ \rho\partial_tu+\left(\rho u-\rho_d\frac{1}{\operatorname{Pe}}\nabla\mu\right)\cdot\nabla u&=\frac{1}{\operatorname{Re}}\nabla\cdot(2\eta\mathbb{D}(u))-\nabla p+\frac{1}{\operatorname{We}\operatorname{Cn}}\mu\nabla\phi+N J\times B-\frac{1}{\operatorname{Fr}}\rho e_3,\\ \nabla\cdot u&=0,\\ J&=\sigma(E+u\times B),\\ E&=-\nabla V,\\ \nabla\cdot J&=0. \end{aligned}

where e3=(0,0,1)Te_3=(0,0,1)^{\mathsf{T}}. Here, the dimensionless numbers are

  • Cahn number Cn⁡=ϵLr\operatorname{Cn}=\frac{\epsilon}{L_r} (interface thickness);

  • Péclet number Pe⁡=ϵLrurλ^mϵ\operatorname{Pe}=\frac{\epsilon L_r u_r}{\widehat{\lambda}m_\epsilon} (ratio of interface thickness to mobility);

  • Reynolds number Re⁡=Lrρrurηr\operatorname{Re}=\frac{L_r\rho_r u_r}{\eta_r} (ratio of the inertial to viscous forces);

  • Weber number We⁡=Lrρrur2λ^\operatorname{We}=\frac{L_r\rho_r u_r^2}{\widehat{\lambda}} (ratio of inertial to interfacial forces);

  • Stuart number N=LrσrBr2ρrurN=\frac{L_r\sigma_r B_r^2}{\rho_r u_r} (ratio of electromagnetic to inertial forces);

  • Froude number Fr⁡=ur2gLr\operatorname{Fr}=\frac{u_r^2}{gL_r} (ratio of inertial to gravitational acceleration forces).

In this scaling, we apply the scaling law 1Pe⁡=3Cn⁡\frac{1}{\operatorname{Pe}}=3\operatorname{Cn} given by Magaletti et al. [19].

Let L2(Ω)L^2(\Omega) be the Lebesgue space of square integrable functions equipped with inner product (⋅,⋅)(\cdot,\cdot), and let H1(Ω)H^1(\Omega) and H(div⁡;Ω)H(\operatorname{div};\Omega) be its subspaces with square integrable gradients and square integrable divergences, respectively. We define their subspaces L02(Ω)L^2_0(\Omega), H01(Ω)H^1_0(\Omega), and H0(div⁡;Ω)H_0(\operatorname{div};\Omega) with vanishing mean values on Ω\Omega, vanishing traces, and vanishing normal traces on ∂Ω\partial\Omega. Then let Qh⊂H1(Ω)Q_h\subset H^1(\Omega), Xh⊂H01(Ω)3X_h\subset H^1_0(\Omega)^3, Mh⊂L02(Ω)∩H1(Ω)M_h\subset L^2_0(\Omega)\cap H^1(\Omega), and Sh⊂L02(Ω)S_h\subset L^2_0(\Omega) be the conforming Lagrange finite element spaces, and Dh⊂H0(div⁡;Ω)D_h\subset H_0(\operatorname{div};\Omega) be the H(div⁡)H(\operatorname{div})-conforming finite element space. It is assumed that the pair Dh×ShD_h\times S_h satisfies the corresponding LBB condition [6]: there exists a positive constant CC which is independent of hh, such that

inf⁡ζh∈Shsup⁡Kh∈Dh(∇⋅Kh,ζh)∥Kh∥H(div⁡;Ω)∥ζh∥L2(Ω)≥C.\inf_{\zeta_h\in S_h}\sup_{K_h\in D_h}\frac{(\nabla\cdot K_h,\zeta_h)}{\lVert K_h\rVert_{H(\operatorname{div};\Omega)}\lVert\zeta_h\rVert_{L^2(\Omega)}}\ge C.

We use a uniform partition of the time interval [0,T][0,T] with the time step size τ\tau. Then we define

δτχn+1=γ0χn+1−χ^τ,\delta_\tau\chi^{n+1}=\frac{\gamma_0\chi^{n+1}-\widehat{\chi}}{\tau},

where χ\chi is a generic variable, and

χ^={χn,k=1,2χn−12χn−1,k=2,γ0={1,k=1,3/2,k=2.\widehat{\chi} = \begin{cases} \chi^n, & k=1,\\ 2\chi^n-\frac{1}{2}\chi^{n-1}, & k=2, \end{cases} \qquad \gamma_0 = \begin{cases} 1, & k=1,\\ 3/2, & k=2. \end{cases}

We also recall the extrapolation formulas

χ~n+1={χn,k=1,2χn−χn−1,k=2.\widetilde{\chi}^{n+1} = \begin{cases} \chi^n, & k=1,\\ 2\chi^n-\chi^{n-1}, & k=2. \end{cases}

With these notations, the discrete BDF finite element scheme is given as follows:

Step 1: find (ϕhn+1,μhn+1)∈Qh×Qh(\phi_h^{n+1},\mu_h^{n+1})\in Q_h\times Q_h, such that for all (ψh,ωh)∈Qh×Qh(\psi_h,\omega_h)\in Q_h\times Q_h,

(δtϕhn+1,ωh)+1Pe(∇μhn+1,∇ωh)=(ϕ~hn+1u~hn+1,∇ωh),(μhn+1,ψh)−Cn2(∇ϕhn+1,∇ψh)−(ϕhn+1,ψh)=((ϕ~hn+1)3−2ϕ~hn+1,ψh).(30)\begin{aligned} (\delta_t\phi_h^{n+1},\omega_h)+\frac{1}{\mathrm{Pe}}(\nabla\mu_h^{n+1},\nabla\omega_h)&=(\widetilde{\phi}_h^{n+1}\widetilde{u}_h^{n+1},\nabla\omega_h),\\ (\mu_h^{n+1},\psi_h)-\mathrm{Cn}^2(\nabla\phi_h^{n+1},\nabla\psi_h)-(\phi_h^{n+1},\psi_h)&=((\widetilde{\phi}_h^{n+1})^3-2\widetilde{\phi}_h^{n+1},\psi_h). \tag*{(30)} \end{aligned}

Then we evaluate ρhn+1\rho_h^{n+1}, ηhn+1\eta_h^{n+1} and σhn+1\sigma_h^{n+1} by (exemplarily for ρhn+1\rho_h^{n+1})

ρhn+1=ρ+−ρ−2ρrC(ϕhn+1)+ρ++ρ−2ρr,\rho_h^{n+1}=\frac{\rho_+-\rho_-}{2\rho_r}\mathcal{C}(\phi_h^{n+1})+\frac{\rho_++\rho_-}{2\rho_r},

where we define

C(ϕhn+1)={ϕhn+1,if ∣ϕhn+1∣≤1,sign⁡(ϕhn+1),otherwise.\mathcal{C}(\phi_h^{n+1})= \begin{cases} \phi_h^{n+1}, & \text{if }|\phi_h^{n+1}|\leq1,\\ \operatorname{sign}(\phi_h^{n+1}), & \text{otherwise}. \end{cases}

Step 2: find uhn+1∈Xhu_h^{n+1}\in X_h, such that for all vh∈Xhv_h\in X_h,

(ρhn+1δtuhn+1,vh)+2Re(ηhn+1D(uhn+1),D(vh))=−(ρhn+1u~hn+1−ρd1Pe∇μhn+1)⋅∇u~hn+1,vh+(p~hn+1,∇⋅vh)+1WeCn(μhn+1∇ϕhn+1,vh)+N(J~hn+1×B,vh)−1Fr(ρhn+1e3,vh).(31)\begin{aligned} (\rho_h^{n+1}\delta_tu_h^{n+1},v_h)+\frac{2}{\mathrm{Re}}(\eta_h^{n+1}\mathbb{D}(u_h^{n+1}),\mathbb{D}(v_h)) ={}&-\left(\rho_h^{n+1}\widetilde{u}_h^{n+1}-\rho_d\frac{1}{\mathrm{Pe}}\nabla\mu_h^{n+1}\right)\cdot\nabla\widetilde{u}_h^{n+1},v_h\\ &+(\widetilde{p}_h^{n+1},\nabla\cdot v_h)+\frac{1}{\mathrm{We}\mathrm{Cn}}(\mu_h^{n+1}\nabla\phi_h^{n+1},v_h)\\ &+\mathrm{N}(\widetilde{J}_h^{n+1}\times B,v_h)-\frac{1}{\mathrm{Fr}}(\rho_h^{n+1}e_3,v_h). \tag*{(31)} \end{aligned}

Step 3: find phn+1∈Mhp_h^{n+1}\in M_h, such that for all qh∈Mhq_h\in M_h,

(∇(phn+1−phn),∇qh)=−γ0ϑτ(∇⋅uhn+1,qh).(32)(\nabla(p_h^{n+1}-p_h^n),\nabla q_h)=-\frac{\gamma_0\vartheta}{\tau}(\nabla\cdot u_h^{n+1},q_h). \tag*{(32)}

with ϑ=min⁡(ρ+,ρ−)/ρr\vartheta=\min(\rho_+,\rho_-)/\rho_r.

Step 4: find Vhn+1∈MhV_h^{n+1}\in M_h, such that for all Λh∈Mh\Lambda_h\in M_h

(σhn+1∇Vhn+1,∇Λh)=(σhn+1uhn+1×B,∇Λh).(33)(\sigma_h^{n+1}\nabla V_h^{n+1},\nabla\Lambda_h)=(\sigma_h^{n+1}u_h^{n+1}\times B,\nabla\Lambda_h). \tag*{(33)}

Step 5: find (Jhn+1,ξhn+1)∈Dh×Sh(J_h^{n+1},\xi_h^{n+1})\in D_h\times S_h, such that for all (Kh,ζh)∈Dh×Sh(K_h,\zeta_h)\in D_h\times S_h

(Jhn+1,Kh)−(ξhn+1,∇⋅Kh)=(σhn+1(−∇Vhn+1+uhn+1×B),Kh),−(∇⋅Jhn+1,ζh)=0.(34)\begin{aligned} (J_h^{n+1},K_h)-(\xi_h^{n+1},\nabla\cdot K_h)&=(\sigma_h^{n+1}(-\nabla V_h^{n+1}+u_h^{n+1}\times B),K_h),\\ -(\nabla\cdot J_h^{n+1},\zeta_h)&=0. \tag*{(34)} \end{aligned}

where ξhn+1\xi_h^{n+1} is a Lagrange multiplier.

In the above scheme, we have exploited a linearizing technique in [26] and the pressure stabilization method developed by Guermond and Salgado [14]. Additionally, we solve the electrostatic potential VV through the electric potential Poisson equation [36] and the current density JJ is then projected into the H(div⁡)H(\operatorname{div})-conforming finite element spaces. These techniques greatly facilitate numerical computations. For time-independent coefficient matrix in equations (30) and (34), expensive block type preconditioners can be employed, and for Poisson-type equations in equations (31)–(33), robust multilevel preconditioners can be utilized. While this scheme offers a straightforward implementation, it does not hold a discrete energy law, which is particularly desirable in diffuse interface computations [26, 31]. Despite this restriction, one can readily verify that the scheme (30)–(34) admits a unique solution, and is globally mass conservative and charge conservative (we refer to [31] for some details),

∫Ωϕhn+1 dx=∫Ωϕhn dxand∇⋅Jh=0.\int_{\Omega}\phi_h^{n+1}\,\mathrm{d}x=\int_{\Omega}\phi_h^n\,\mathrm{d}x \qquad\text{and}\qquad \nabla\cdot J_h=0.

Remark 2. Via its definition, the order parameter ϕ\phi should physically belong to [−1,1][-1,1], yet the Ginzburg-Landau energy cannot guarantee this bound at a theoretical level, unlike the Flory-Huggins singular (logarithmic) energy [10]. In computations, we employ a cut-off approach to numerically preserve this physical constraint, which is a common strategy in diffuse interface simulations using Ginzburg-Landau energy [17, 26, 31]. This technique enables stable simulations with large parameter ratios.

Remark 3. Practical MHD engineering applications usually involves large parameter ratios, with a typical order O(103)O(10^3) for density, O(102)O(10^2) for dynamic viscosity, and O(107)O(10^7) for electrical conductivity; see [11, 36] and the references therein for more details in laboratory experiments. This results in extremely large Reynolds numbers and Stuart numbers, in turn demanding a highly robust solver to resolve the flow dynamics. In addition, many MHD processes exhibit multiscale features [15, 28, 29], requiring extremely small interface thicknesses to achieve a better simulation result. Undoubtedly, this will further consume a vast amount of computational resources. Under such a circumstance, developing an asymptotic-preserving and robust solver for realistic MHD simulations is essential and is of our future interest.

The magnetic damping effects on bubble dynamics

In this section, we present several three-dimensional numerical examples to demonstrate the magnetic damping effects on bubble dynamics. An illustration can be seen in Figure 1. This problem finds its applications in metallurgy processes [28], where injected bubbles serve to stir and homogenize molten metals, with magnetic fields providing a non-intrusive approach to control the motion of bubbles. Meanwhile, it has been extensively investigated both experimentally [11] and numerically [36, 37, 38]. Here, we focus on the scenario of a single bubble rising in a liquid column under either an external horizontal or vertical magnetic field. In this benchmark problem, a bubble of a lighter fluid with radius 0.5 m is centered at (0.5 m, 0.5 m, 0.5 m) in the domain Ω=(0,1 m)×(0,1 m)×(0,2 m)\Omega= (0, 1\,\mathrm{m}) \times(0, 1\,\mathrm{m}) \times(0, 2\,\mathrm{m}), corresponding to the following initial profile for ϕ\phi,

Setup of single rising bubble under magnetic fields

Figure 1. (a) Horizontal magnetic field (b) Vertical magnetic field. Setup of single rising bubble under magnetic fields.

ϕ0=tanh⁡((x−0.5)2+(y−0.5)2+(z−0.5)2−0.252ϵ).\phi^0 = \tanh\left(\frac{\sqrt{(x-0.5)^2+(y-0.5)^2+(z-0.5)^2}-0.25}{\sqrt{2}\epsilon}\right).

We will use the international system of units and ignore the units later if there is no confusion. The relevant physical parameters are

  • densities (kg⋅m−3\mathrm{kg}\cdot\mathrm{m}^{-3}): ρ+=1000\rho_+ = 1000, ρ−=1\rho_- = 1;

  • dynamic viscosities (kg⋅m−1⋅s−1\mathrm{kg}\cdot\mathrm{m}^{-1}\cdot\mathrm{s}^{-1}): η+=10\eta_+ = 10, η−=0.1\eta_- = 0.1;

  • electrical conductivities (S⋅m−1\mathrm{S}\cdot\mathrm{m}^{-1}): σ+=1000\sigma_+ = 1000, σ−=1\sigma_- = 1;

  • surface tension (N⋅m−1\mathrm{N}\cdot\mathrm{m}^{-1}): λ=1.96\lambda= 1.96;

  • gravitational acceleration (m⋅s−2\mathrm{m}\cdot\mathrm{s}^{-2}): g=0.98g = 0.98;

  • magnetic fields (T): a horizontal magnetic field B=(Br,0,0)T\mathbf{B} = (B_r, 0, 0)^\mathsf{T} and a vertical magnetic field B=(0,0,Br)T\mathbf{B} = (0, 0, B_r)^\mathsf{T}, with Br=3B_r = 3, 5, and 7.

Our numerical implementation relies on the deal.II finite element library [3] with ParaView postprocessing software (www.paraview.org). The proposed BDF2 scheme, initialized via BDF1, is utilized in simulation. Additionally, we discretize QhQ_h, XhX_h, and MhM_h with the continuous Q1_1 element, ShS_h with the discontinuous DGQ0_0 element, and DhD_h with the lowest order Raviart–Thomas RT0_0 element. To solve the resulting large-scale algebraic systems, we employ the flexible generalized minimum residual (FGMRES) algorithm with preconditioning: additive Schwarz methods for equations (30) and (34), and algebraic multigrid (AMG) methods for equations (31)–(33). Additionally, we use a iteration tolerance of ϵiter=10−10\epsilon_{iter}=10^{-10} with the initial guess set to the extrapolated value.

While adaptive mesh refinement (AMR) techniques lead to a poor mass conservation result [17] and present challenges in designing a desirable refinement strategy, we only adopt a fixed locally refined mesh to resolve the interfacial dynamics. Specifically, we set mesh size h=2−7h=2^{-7} in subdomain (0.15,0.85)×(0.15,0.85)×(0.2,1.8)(0.15,0.85)\times(0.15,0.85)\times(0.2,1.8) and h=2−6h=2^{-6} elsewhere; see Figure 2. This configuration generates 1,906,592 cells, resulting in 3,920,510, 5,880,765, 1,960,255, 1,960,255, and 7,701,408 degrees of freedom for equations (30)–(34), respectively. In terms of temporal discretization, we set the final dimensional time T∗=3T^*=3 with a fixed time step of 1×10−31\times10^{-3}, giving 3,000 total steps. With regard to characteristic quantities, we select ϵ=0.005\epsilon=0.005, Lr=1L_r=1, ur=gLru_r=\sqrt{gL_r}, ρr=ρ+\rho_r=\rho_+, ηr=η+\eta_r=\eta_+, and σr=σ+\sigma_r=\sigma_+. All results were obtained in a machine with AMD EPYC 9654 and 768 GB RAM.

Mesh, quarter of domain

Figure 2. Mesh, quarter of domain.

The cross-sections of the order parameter ϕ\phi at different dimensional time instants t∗t^* are presented in Figure 3 (in the absence of magnetic fields), Figure 4 (under horizontal magnetic fields), and Figure 5 (under vertical magnetic fields), and a comparison of final bubble shape can be found in Figure 6. The corresponding velocity streamlines, visualized via the “Surface LIC" configuration in ParaView, are also depicted in these figures.

Cross-sections of the order parameter at three time instants, with streamline distributions

Figure 3. Cross-sections of ϕhn\phi_h^n without magnetic fields, with streamline distributions of uhnu_h^n depicted on the left-hand side.

Cross-sections of $\phi_h^n$ under horizontal magnetic fields, with streamline distributions of $u_h^n$

Figure 4. Cross-sections of ϕhn\phi_h^n under horizontal magnetic fields, with streamline distributions of uhnu_h^n depicted on the left-hand side. Top row: Br=3B_r = 3; middle row: Br=5B_r = 5; bottom row: Br=7B_r = 7.

Cross-sections of $\phi_h^n$ under vertical magnetic fields, with streamline distributions of $\mathbf{u}_h^n$

Figure 5. Cross-sections of ϕhn\phi_h^n under vertical magnetic fields, with streamline distributions of uhn\mathbf{u}_h^n depicted on the left-hand side. Top row: Br=3B_r = 3; middle row: Br=5B_r = 5; bottom row: Br=7B_r = 7.

Final bubble shapes under horizontal magnetic fields

(a) Horizontal magnetic fields

Final bubble shapes under vertical magnetic fields

(b) Vertical magnetic fields

Figure 6. Final bubble shape. Left to right: Br=0B_r = 0, 33, 55, and 77.

In addition, Figure 7 shows the rise velocity, mass drift, and charge conservation, which are defined as

Derived benchmark quantities

Figure 7. Derived benchmark quantities.

∫ϕhn<0uhn⋅e3dx∫ϕhn<0dx,∫Ωϕhndx−∫Ωϕh0dx,and∥∇⋅Jhn∥L2(Ω),\frac{\int_{\phi_h^n<0}\mathbf{u}_h^n\cdot\mathbf{e}_3\mathrm{d}x}{\int_{\phi_h^n<0}\mathrm{d}x},\qquad\int_{\Omega}\phi_h^n\mathrm{d}x-\int_{\Omega}\phi_h^0\mathrm{d}x,\qquad\text{and}\qquad\lVert\nabla\cdot\mathbf{J}_h^n\rVert_{L^2(\Omega)},

respectively. For all the reported cases, it is clearly observable that the change in the total mass and the conservation of charge is of the order O(10−8)O(10^{-8}) and O(10−9)O(10^{-9}). This illustrates the good mass and charge conservation property of our numerical method in practical physical simulations.

Moreover, the following conclusions can be drawn:

  • velocity characteristics: The horizontal magnetic fields exert a significant damping effect on the bubble rise velocity. In contrast, the vertical fields display a non-monotonic influence. As the magnetic intensity increases, the velocity field becomes increasingly parallel with the magnetic fields, which is clearly demonstrated by the streamline distributions.

  • interface dynamics: The vertical magnetic fields preserve an axisymmetric bubble shape similar to non-magnetic scenarios. Conversely, the horizontal magnetic fields give rise to anisotropic deformations, compressing the bubble shape along the field direction. And in all cases, a stronger magnetic field leads to a progressive suppression of bubble motion.

These numerical observations are in qualitative agreement with previous results reported in [36, 37, 38], in turn validating the capacity of our model to capture the magnetic damping effects on bubble dynamics.

Concluding remarks

In this paper, we developed a novel diffuse interface model to simulate inductionless MHD free surface flows. By means of Onsager’s variational principle and the laws of thermodynamics, we derived a nonlinear system that couples the Cahn-Hilliard equation with the inductionless MHD equations. This model is thermodynamically consistent and satisfies fundamental conservation laws. Meanwhile, compared to existing diffuse interface MHD models, it is capable of handling general material properties in practical engineering processes. We carried out an asymptotic analysis via the method of formally matched asymptotic expansions. Provided a scaling law for the mobility from [19], we deduced that the classical sharp interface system (1) can be recovered in the vanishing interface thickness limit, which illustrates the robustness of our proposed diffuse interface model as an approximate approach. Regarding numerical implementation, we design an efficient decoupled, linear, and charge-conservative finite element algorithm based on a dimensionless form of the diffuse interface model. Employing this algorithm, we investigated the magnetic damping effects through a single rising bubble benchmark. The observed flow mechanisms have been observed to be consistent with the results reported in [36, 37, 38], in turn validating the proposed diffuse interface method.

Although the developed method is capable of effectively capturing MHD phenomena in free surface flows, such as magnetic damping effects, our discussions are still relatively circumscribed. We identify several directions for future research. First, it is essential to formulate a general system that incorporates the complete MHD equations. Meanwhile, taking thermocapillary effects into account is also indispensable, in view of practical industrial applications [28, 36]. Another vital aspect for future investigation lies in the exploration of other MHD phenomena within the proposed method, for instance, metal pad rolling in multilayer liquid metal systems [9, 15, 29], which further involves moving contact line dynamics and exhibits multiscale features. Finally, it is highly desirable to develop structure- and asymptotic-preserving numerical schemes that can preserve fundamental physical properties and obtain more accurate simulation results through convenient formulations.

Acknowledgments

The authors would like to thank Prof. Wenyu Lei at UESTC for the fruitful discussions on numerical implementations using deal.II. The research of MJL is supported in part by the NSFC 12271082, 62231016. The research of ZYX is supported in part by the NSFC 12471371. The research of LWX is supported in part by the NSFC 12431015, 62231016.

References

  1. [1]H. Abels, H. Garcke, and G. Grün. Thermodynamically consistent, frame indifferent diffuse interface models for incompressible two-phase flows with different densities. Math. Models Methods Appl. Sci., 22(3):1150013, 40, 2012.arxiv.org/abs/1104.1336
  2. [2]D. M. Anderson, G. B. McFadden, and A. A. Wheeler. Diffuse-interface methods in fluid mechanics. Annu. Rev. Fluid Mech., 30:139–165, 1998.
  3. [3]D. Arndt, W. Bangerth, M. Bergbauer, M. Feder, M. Fehling, J. Heinz, T. Heister, L. Heltai, M. Kronbichler, M. Maier, P. Munch, J.-P. Pelteret, B. Turcksin, D. Wells, and S. Zampini. The deal.II library, version 9.5. J. Numer. Math., 31(3):231–246, 2023.
  4. [4]R. Chen and H. Zhang. Second-order energy stable schemes for the new model of the Cahn-Hilliard-MHD equations. Adv. Comput. Math., 46(6):Paper No. 79, 28, 2020.DOI
  5. [5]P. A. Davidson. Introduction to magnetohydrodynamics. Cambridge Texts in Applied Mathematics. Cambridge University Press, Cambridge, second edition, 2017.
  6. [6]A. Ern and J.-L. Guermond. Finite elements II—Galerkin approximation, elliptic and mixed PDEs, volume 73 of Texts in Applied Mathematics. Springer, Cham, 2021.
  7. [7]Y.-C. Fan, L. Chen, and M.-J. Ni. Large-eddy simulation for decaying magnetohydrodynamic turbulence at low magnetic Reynolds number. J. Fluid Mech., 1012:Paper No. A19, 28, 2025.DOI
  8. [8]M. Flueck, A. Janka, C. Laurent, M. Picasso, J. Rappaz, and G. Steiner. Some mathematical and numerical aspects in aluminum production. J. Sci. Comput., 43(3):313–325, 2010.
  9. [9]J.-F. Gerbeau, T. Lelièvre, and C. Le Bris. Simulations of MHD flows with moving interfaces. J. Comput. Phys., 184(1):163–191, 2003.
  10. [10]A. Giorgini, A. Miranville, and R. Temam. Uniqueness and regularity for the Navier-Stokes-Cahn-Hilliard system. SIAM J. Math. Anal., 51(3):2535–2574, 2019.
  11. [11]H.-Y. Gou, M.-J. Ni, and Z.-H. Yao. Experimental study of a single bubble’s motion in a liquid metal under a horizontal magnetic field. J. Fluid Mech., 988:Paper No. A21, 25, 2024.DOI
  12. [12]S. Gross and A. Reusken. Numerical methods for two-phase incompressible flows, volume 40 of Springer Series in Computational Mathematics. Springer-Verlag, Berlin, 2011.
  13. [13]G. Grün and F. Klingbeil. Two-phase flow with mass density contrast: stable schemes for a thermodynamic consistent and frame-indifferent diffuse-interface model. J. Comput. Phys., 257:708–725, 2014.arxiv.org/abs/1210.5088
  14. [14]J.-L. Guermond and A. Salgado. A splitting method for incompressible flows with variable density based on a pressure Poisson equation. J. Comput. Phys., 228(8):2834–2846, 2009.DOI
  15. [15]W. Herreman, C. Nore, J.-L. Guermond, L. C. Pappanera, N. Weber, and G. M. Hortsmann. Perturbation theory for metal pad roll instability in cylindrical reduction cells. J. Fluid Mech., 878:598–646, 2019.DOI
  16. [16]R. Hiptmair, L. Li, S. Mao, and W. Zheng. A fully divergence-free finite element method for magnetohydrodynamic equations. Math. Models Methods Appl. Sci., 28(4):659–695, 2018.
  17. [17]M. A. Khanwale, K. Saurabh, M. Ishii, H. Sundar, J. A. Rossmanith, and B. Ganapathysubramanian. A projection-based, semi-implicit time-stepping approach for the Cahn-Hilliard Navier-Stokes equations on adaptive octree meshes. J. Comput. Phys., 475:Paper No. 111874, 35, 2023.DOI
  18. [18]J. Lowengrub and L. Truskinovsky. Quasi-incompressible Cahn-Hilliard fluids and topological transitions. R. Soc. Lond. Proc. Ser. A Math. Phys. Eng. Sci., 454(1978):2617–2654, 1998.DOI
  19. [19]F. Magaletti, F. Picano, M. Chinappi, L. Marino, and C. M. Casciola. The sharp-interface limit of the Cahn-Hilliard/Navier-Stokes model for binary fluids. J. Fluid Mech., 714:95–126, 2013.DOI
  20. [20]N. Morley, S. Smolentsev, R. Munipalli, M.-J. Ni, D. Gao, and M. Abdou. Progress on the modeling of liquid metal, free surface, MHD flows for fusion liquid walls. Fusion Eng. Des., 72(1):3–34, 2004.
  21. [21]D. Munger and A. Vincent. A level set approach to simulate magnetohydrodynamic-instabilities in aluminum reduction cells. J. Comput. Phys., 217(2):295–311, 2006.DOI
  22. [22]J.-H. Pan, M.-J. Ni, and N.-M. Zhang. A consistent and conservative immersed boundary method for MHD flows and moving boundary problems. J. Comput. Phys., 373:425–445, 2018.DOI
  23. [23]T. Qian, X.-P. Wang, and P. Sheng. A variational approach to moving contact line hydrodynamics. J. Fluid Mech., 564:333–360, 2006.DOI
  24. [24]H. Qiu. Error analysis of fully discrete scheme for the Cahn-Hilliard-magneto-hydrodynamics problem. J. Sci. Comput., 95(1):Paper No. 16, 27, 2023.
  25. [25]R. Samulyak, J. Du, J. Glimm, and Z. Xu. A numerical algorithm for MHD of free surface flows at low magnetic Reynolds numbers. J. Comput. Phys., 226(2):1532–1549, 2007.
  26. [26]J. Shen and X. Yang. Decoupled, energy stable schemes for phase-field models of two-phase incompressible flows. SIAM J. Numer. Anal., 53(1):279–296, 2015.
  27. [27]H. Su, J. Wang, Z. Xia, and K. Zhang. Optimal L² error estimates for 2D/3D incompressible Cahn–Hilliard–magnetohydrodynamic equations. arXiv e-prints, 2025.arxiv.org/abs/2506.13080
  28. [28]B. G. Thomas. Review on Modeling and Simulation of Continuous Casting. Steel Res. Int., 89(1):1700312, 2018.
  29. [29]A. Tucs, V. Bojarevics, and K. Pericleous. Magnetohydrodynamic stability of large scale liquid metal batteries. J. Fluid Mech., 852:453–483, 2018.DOI
  30. [30]C. Wang, J. Wang, S. M. Wise, Z. Xia, and L. Xu. Convergence analysis of a temporally second-order accurate finite element scheme for the Cahn-Hilliard-magnetohydrodynamics system of equations. J. Comput. Appl. Math., 436:Paper No. 115409, 17, 2024.
  31. [31]J. Wang, M. Li, and C. Wang. Efficient finite element schemes for a phase field model of two-phase incompressible flows with different densities. J. Comput. Phys., 518:Paper No. 113331, 28, 2024.
  32. [32]X. Wang, X. Mao, S. Mao, and X. He. A mass and charge conservative fully discrete scheme for a 3D diffuse interface model of the two-phase inductionless MHD flows. Comput. Math. Appl., 182:139–162, 2025.
  33. [33]X. Xu, Y. Di, and H. Yu. Sharp-interface limits of a phase-field model with a generalized Navier slip boundary condition for moving contact lines. J. Fluid Mech., 849:805–833, 2018.
  34. [34]J. Yang, S. Mao, and X. He. Unconditionally optimal convergent zero-energy-contribution scheme for two phase MHD model. J. Sci. Comput., 102(2):Paper No. 55, 72, 2025.
  35. [35]J. Yang, S. Mao, X. He, X. Yang, and Y. He. A diffuse interface model and semi-implicit energy stable finite element method for two-phase magnetohydrodynamic flows. Comput. Methods Appl. Mech. Engrg., 356:435–464, 2019.
  36. [36]J. Zhang and M.-J. Ni. Direct simulation of multi-phase MHD flows on an unstructured Cartesian adaptive system. J. Comput. Phys., 270:345–365, 2014.
  37. [37]J. Zhang and M.-J. Ni. Direct simulation of single bubble motion under vertical magnetic field: Paths and wakes. Phys. Fluids, 26(10), 2014.
  38. [38]J. Zhang, M.-J. Ni, and R. Moreau. Rising motion of a single bubble through a liquid metal in the presence of a horizontal magnetic field. Phys. Fluids, 28(3):032101, 2016.

Paper details

vlnc:W_89h5tmtw
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 1

downloads · 0

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

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents