betaFeedback

WorksPhysics and Astronomy

Stokes flow of an evolving fluid film with arbitrary shape and topology

Abstract

The dynamics of evolving fluid films in the viscous Stokes limit is relevant to various applications, such as the modeling of lipid bilayers in cells. While the governing equations were formulated by Scriven in 1960, solving for the flow of a deformable viscous surface with arbitrary shape and topology has remained a challenge. In this study, we present a straightforward discrete model based on variational principles to address this long-standing problem. We replace the classical equations, which are expressed with tensor calculus in local coordinates, with a simple coordinate-free, differential-geometric formulation. The formulation provides a fundamental understanding of the underlying mechanics and directly translates to discretization. We construct a discrete analogue of the system using the Onsager variational principle, which, in a smooth context, governs the flow of a viscous medium. In the discrete setting, instead of term-wise discretizing the coordinate-based Stokes equations, we construct a discrete Rayleighian for the system and derive the discrete Stokes equations via the variational principle. This approach results in a stable, structure-preserving variational integrator that solves the system on general manifolds.

Introduction

When surface dissipation dominates bulk dissipation (i.e., when the Saffman–Delbrück length far exceeds the system size), the viscous flow of evolving fluid films can be modeled without accounting for the bulk fluid ([3, 19]). The governing equations, known as the evolving Stokes equations, have not only been of theoretical interest, but also found practical applications in engineering, such as the study of foam ([7]). Their use also has a long-standing history in biophysical contexts, addressing fundamental problems across scales—from modeling subcellular structures like lipid bilayers to tissue-level phenomena such as epithelial monolayers ([1]).

Historically, the governing equations were formulated by Scriven [21] to study foam instability. When coupled with bending elasticity, the evolving Stokes equations serve as a common model for lipid membranes ([3]). Recently, there has been substantial interest in modeling cell and tissue growth by coupling the viscous layer with additional anisotropy and active processes, such as those seen in the cellular cortex and active nematic fluids ([2, 14, 23]). The evolving Stokes equations, along with extensions to the full Navier–Stokes equations, have been independently derived through various principles ([21, 3, 23, 11, 18, 20, 16, 2]) and are compared in Brander et al. (2022). Our approach is closest to the formulation of [3] via the Onsager variational principle for dissipative systems. To computationally solve the evolving Stokes equations, most existing methods tackle the partial differential equations by explicitly splitting them into normal and tangential components. This involves a vector-valued equation for the tangent velocity and a scalar equation for the normal velocity on a manifold. When viewed this way, the system of equations is challenging to solve. First, the tangent equation is tensor-valued on an evolving Riemannian manifold, necessitating specialized techniques for covariant differentiation [17, 8, 10, 22, 24]. Some methods avoid the tensor-valued equation of tangent velocity through streamfunction formulations, although this approach is limited to simply-connected domains [23] unless additional topological techniques are employed [25]. Second, the equations in the decomposed form are explicitly coupled with surface curvatures, which, in classic numerical approximations, require representations using high-order basis functions [20, 12].

In this study, instead of directly tackling the classic equations, we return to the fundamental governing kinematics and principles of a viscous surface film. We replace the system of equations expressed in tangent-normal splitting with a simple, coordinate-free differential-geometric formulation. We show that our resulting equations agree with earlier formulations by Scriven [21] and Arroyo & DeSimone [3] but in a more elegant form. Our approach also directly carries over to the discrete setting. The abstraction directly leads to a discretization for the strain rate tensor on discrete meshes. We then construct a discrete Rayleighian for the system and derive the discrete Stokes equations via the Onsager variational principle, leading to a simple linear system. We provide a self-contained exposition of both the continuous and discrete models, as well as the numerical methods used to solve the system of equations on arbitrary geometries and topologies.

Theory

We derive the equations of motion for an evolving surface driven by viscous Stokes flow using Onsager’s variational principle. Specifically, a dissipation functional, called the Rayleighian, is assigned to each position and velocity state of a deformable surface. The equations of motion are then obtained by finding the velocity that minimizes this functional.

We adopt a differential geometric approach, similar to Marsden & Hughes [13], to describe continuum mechanics. Tensors are expressed using fiber bundle notation for clarity. Readers unfamiliar with this terminology can refer to § 2.1 for a brief overview. We first define relevant tensors and differential operators in § 2.2, followed by the definition of the Rayleighian for the system in § 2.3. For simplicity in presenting our smooth theory, all functions and tensors are assumed to be of C∞C^{\infty} class.

Terminology and notation

On a manifold MM, the tangent space TpMT_pM is a vector space that provides a local linear approximation of the manifold at p∈Mp \in M. The disjoint union of all tangent spaces forms the tangent bundle TM=⋃pTpMTM = \bigcup_p T_pM, where each TpMT_pM is also called the fiber of the tangent bundle TMTM at the base point p∈Mp \in M on its base manifold MM. The dual bundle of TMTM is the cotangent bundle, T∗MT^*M, whose fibers are the dual spaces of the tangent spaces. The (m,n)(m,n)-typed tensor bundle is written as TM⊗m,n=TM⊗m⊗T∗M⊗nTM^{\otimes m,n} = TM^{\otimes m} \otimes T^*M^{\otimes n}. A tensor field Ψ\boldsymbol{\Psi} on MM is a section of the tensor bundle, denoted as Ψ∈Γ(TM⊗m,n)\boldsymbol{\Psi} \in\Gamma(TM^{\otimes m,n}), which assigns a tensor Ψ∣p∈TpM⊗m,n\boldsymbol{\Psi}|_p \in T_pM^{\otimes m,n} to each point p∈Mp \in M. When a basis section is chosen, its components, Ψi1…inj1…jm\Psi^{j_1\ldots j_m}_{i_1\ldots i_n}, have mm contravariant indices and nn covariant indices.

For a map φ:M→W\varphi: M \to W, where WW is another manifold, the pullback bundle TφW=φ∗TWT_\varphi W = \varphi^*TW is defined with MM as its base manifold, while the fiber at p∈Mp \in M, (TφW)∣p=Tφ(p)W(T_\varphi W)|_p = T_{\varphi(p)}W, is the tangent space to WW at φ(p)∈W\varphi(p) \in W.

Note that tensors can be identified with linear maps; for example, Ψ∣p∈TpM⊗m,n\Psi|_p \in T_pM^{\otimes m,n} acts as a linear map Ψ∣p:TpM→TpM⊗m,n−1\Psi|_p : T_pM \to T_pM^{\otimes m,n-1}.

Differential geometry of an evolving surface

We describe a deformable surface as follows. Let MM be a 2-dimensional closed manifold (∂M=∅\partial M = \varnothing) with arbitrary genus, representing the material/Lagrangian space. Let WW be a 3-dimensional Riemannian manifold representing the world/Eulerian space. Let g∈Γ(T∗W⊗symmT∗W)g \in\Gamma(T^*W \otimes_{\mathrm{symm}} T^*W) (i.e., a symmetric (0,2)(0,2) tensor field) denote the Riemannian metric tensor on WW. The common setup is to set W=R3W = \mathbb{R}^3, the Euclidean 3-space, with g[V,W]=V⋅Wg[V,W] = V \cdot W being the Euclidean inner product for any V,W∈R3V,W \in\mathbb{R}^3. However, our theory is not limited to this assumption. The position of a deformable surface is an embedding function φ:M↪W\varphi: M \hookrightarrow W from the material space to the world space. A time-dependent deformable surface, or an evolving surface, is a time-dependent embedding φ(t)=φt:M↪W\varphi(t) = \varphi_t : M \hookrightarrow W. The material velocity of an evolving surface φ(t)\varphi(t) is given by φ˙t:=∂tφt\dot{\varphi}_t := \partial_t\varphi_t, which is a pullback vector field φ˙t∈Γ(TφtW)\dot{\varphi}_t \in\Gamma(T_{\varphi_t}W) with its base point in the material frame MM (cf. § 2.1).1

Each material velocity φ˙t∈Γ(TφtW)\dot{\varphi}_t \in\Gamma(T_{\varphi_t}W) can be extended to a spatial velocity over WW, denoted by Ut∈Γ(TW)U_t \in\Gamma(TW), such that φ˙t=Ut∘φt\dot{\varphi}_t = U_t \circ\varphi_t, or equivalently, φ˙t∣p=Ut∣φt(p)\dot{\varphi}_t|_p = U_t|_{\varphi_t(p)} for φt(p)∈W\varphi_t(p) \in W. Note that this extension is not unique, as the assignment of UtU_t at locations away from the surface φt(M)⊂W\varphi_t(M) \subset W is arbitrary.

Deformation gradient and Cauchy–Green tensor

The differential of the positioning φ:M↪W\varphi: M \hookrightarrow W of a surface is denoted by F:=dφ∈Γ(TφW⊗T∗M)F := d\varphi\in\Gamma(T_{\varphi}W \otimes T^*M). This differential is also known as the pushforward map, the tangent map, or the deformation gradient. At each point p∈Mp \in M, the two-point tensor F∣pF|_p can be identified with a linear map (cf. § 2.1) F∣p=dφ∣p:TpM→Tφ(p)WF|_p = d\varphi|_p : T_pM \to T_{\varphi(p)}W2 that realizes a material tangent vector as a 3D world vector, as expected for the deformation gradient.

The embedding φ:M↪W\varphi: M \hookrightarrow W induces a Riemannian metric (the first fundamental form) I∈Γ(T∗M⊗symmT∗M)I \in\Gamma(T^*M \otimes_{\mathrm{symm}} T^*M) on MM from the Riemannian structure gg on WW. This induced metric is defined by I[v,w]:=g[Fv,Fw]=v⊤F⊤gFwI[v,w] := g[Fv,Fw] = v^\top F^\top gFw for each v,w∈TpMv,w \in T_pM, p∈Mp \in M. In continuum mechanics, I=F⊤gFI = F^\top gF is known as the right Cauchy–Green tensor.3

For brevity, we denote the induced metric as ⟨⋅,⋅⟩:=I[⋅,⋅]\langle\cdot,\cdot\rangle:= I[\cdot,\cdot] and the L2L_2 inner product as ⟨⟨⋅,⋅⟩⟩:=∫M⟨⋅,⋅⟩ dA\langle\langle\cdot,\cdot\rangle\rangle:= \int_M \langle\cdot,\cdot\rangle\,dA where dAdA is the area form from the metric II. These inner products define the vector norm ∥⋅∥2\|\cdot\|^2 and the L2L_2 norm ∥⋅∥2\|\cdot\|^2. Dual pairings between primal (contravariant) and dual (covariant) tensors are denoted by ⟨⋅∣⋅⟩\langle\cdot|\cdot\rangle and ⟨⟨⋅∣⋅⟩⟩=∫M⟨⋅∣⋅⟩dA\langle\langle\cdot|\cdot\rangle\rangle= \int_M \langle\cdot|\cdot\rangle dA. The L2L_2 inner product between two scalar functions ff and gg is denoted by ⟨f,g⟩=∫Mfg dA\langle f,g\rangle= \int_M fg\,dA.

Strain rate tensor

For an evolving surface φt\varphi_t, the induced metric It∈Γ(T∗M⊗symmT∗M)I_t \in\Gamma(T^*M \otimes_{\mathrm{symm}} T^*M) varies with time tt. The strain rate tensor Et∈Γ(T∗M⊗symmT∗M)E_t \in\Gamma(T^*M \otimes_{\mathrm{symm}} T^*M) is defined as half the rate of change of the induced metric: 2Et:=I˙t2E_t := \dot{I}_t.

To compute the time derivative of It=FtTgFtI_t = F_t^{\mathsf{T}} g F_t we first calculate the time derivative of FtF_t. Let Ut∈Γ(TW)U_t \in\Gamma(TW) be any extended velocity field such that φ˙t=Ut∘φt\dot{\varphi}_t = U_t \circ\varphi_t. Then,

F˙=dφ˙t=d(Ut∘φt)=∇(Ut∘φt)∘dφt≡(∇Ut)F.(1)\dot{F} = d\dot{\varphi}_t = d(U_t \circ\varphi_t) = \nabla(U_t \circ\varphi_t) \circ d\varphi_t \equiv(\nabla U_t)F. \tag*{(1)}

where ∇:Γ(TW)→Γ(TW⊗T∗W)\nabla: \Gamma(TW) \to\Gamma(TW \otimes T^*W) is the Levi-Civita covariant derivative compatible with the metric gg. When applied to a vector v∈Γ(TM)v \in\Gamma(TM), F˙tv=∇FtvUt\dot{F}_t v = \nabla_{F_t v}U_t, which is a directional derivative of UtU_t tangential to the surface. In particular, (1) depends only on the values of Ut∘φtU_t \circ\varphi_t at the surface and not on the choice of its extension UtU_t over WW.

With a time-invariant metric gg and (1), the strain rate tensor is given by

2Et=I˙t=F˙tTgFt+FtTgF˙t=FtT((∇Ut)Tg+g∇Ut)Ft.(2)2E_t = \dot{I}_t = \dot{F}_t^{\mathsf{T}}gF_t + F_t^{\mathsf{T}}g\dot{F}_t = F_t^{\mathsf{T}}\left((\nabla U_t)^{\mathsf{T}}g + g\nabla U_t\right)F_t. \tag*{(2)}

When applied to vectors, 2Et[v,w]p=(∇FtvUt)⋅(Ftw)+(∇FtwUt)⋅(Ftv)2E_t[v,w]_p = (\nabla_{F_t v}U_t)\cdot(F_t w) + (\nabla_{F_t w}U_t)\cdot(F_t v) for v,w∈Γ(TpM)v,w \in\Gamma(T_pM). Again, EtE_t depends only on the values of Ut∘φtU_t \circ\varphi_t and not on the choice of its extension UtU_t.

Differential operators

The operator that takes each velocity field φ˙=U∘φ\dot{\varphi}=U\circ\varphi at the surface to its induced strain rate tensor ((2)) is called the Killing operator:

K:Γ(TφW)→Γ(T∗M⊗symmT∗M),Kφ˙≔E=FT(∇U)Tg+g∇U2F.(3)\mathcal{K} : \Gamma(T_\varphi W) \to\Gamma(T^*M \mathbin{\otimes_{\mathrm{symm}}} T^*M), \qquad\mathcal{K}\dot{\varphi} \coloneqq E = F^{\mathsf{T}}\frac{(\nabla U)^{\mathsf{T}}g + g\nabla U}{2}F. \tag*{(3)}

The trace of the strain rate tensor defines the divergence operator

DIV⁡:Γ(TφW)→C∞(M),DIV⁡φ˙≔tr⁡(Kφ˙),(4)\operatorname{DIV} : \Gamma(T_\varphi W) \to C^\infty(M), \qquad\operatorname{DIV}\dot{\varphi} \coloneqq\operatorname{tr}(\mathcal{K}\dot{\varphi}), \tag*{(4)}

which measures the rate of change of area induced by the given velocity field. Define the gradient operator as the negative adjoint of the divergence operator:

GRAD⁡=−DIV⁡∗:C∞(M)→Γ(Tφ∗W),⟨ ⁣⟨GRAD⁡f∣φ˙⟩ ⁣⟩=−⟨ ⁣⟨f,DIV⁡φ˙⟩ ⁣⟩(5)\operatorname{GRAD} = -\operatorname{DIV}^* : C^\infty(M) \to\Gamma(T^*_\varphi W), \qquad\langle\!\langle\operatorname{GRAD} f \mid\dot{\varphi}\rangle\!\rangle= -\langle\!\langle f,\operatorname{DIV}\dot{\varphi}\rangle\!\rangle \tag*{(5)}

for all f∈C∞(M)f \in C^\infty(M). The covariant divergence is the negative adjoint of the Killing operator:

DIV⁡∇=−K∗:Γ(TM⊗symmTM)→Γ(Tφ∗W),⟨ ⁣⟨DIV⁡∇T∣φ˙⟩ ⁣⟩=−⟨ ⁣⟨T∣Kφ˙⟩ ⁣⟩(6)\operatorname{DIV}^{\nabla} = -\mathcal{K}^* : \Gamma(TM \mathbin{\otimes_{\mathrm{symm}}} TM) \to\Gamma(T^*_\varphi W), \qquad\langle\!\langle\operatorname{DIV}^{\nabla}T \mid\dot{\varphi}\rangle\!\rangle= -\langle\!\langle T \mid\mathcal{K}\dot{\varphi}\rangle\!\rangle \tag*{(6)}

for all symmetric tensor fields T∈Γ(TM⊗symmTM)T \in\Gamma(TM \mathbin{\otimes_{\mathrm{symm}}} TM). Here the dual pairing ⟨T∣E⟩\langle T\mid E\rangle between (0,2)(0,2) and (2,0)(2,0) tensors is ⟨T∣E⟩=tr⁡(TTE)=∑ijTijEij\langle T\mid E\rangle= \operatorname{tr}(T^{\mathsf{T}}E) = \sum_{ij}T^{ij}E_{ij}.

Notably, the operators K\mathcal{K}, DIV⁡\operatorname{DIV}, GRAD⁡\operatorname{GRAD}, and DIV⁡∇\operatorname{DIV}^{\nabla}, derived from the strain rate tensor EE of the evolving surface, differ from the commonly used intrinsic operators K\mathcal{K}, div⁡\operatorname{div}, grad⁡\operatorname{grad}, and div⁡∇\operatorname{div}^{\nabla}. As shown through the normal-tangent decomposition in Appendix § 6, these differences arise from additional contributions related to curvature (cf. A 3–A 5).

Onsager's variational principle for an evolving viscous fluid film

The relaxation dynamics of an evolving viscous fluid film is governed by the Onsager’s variational principle, which states that Stokes flow follows the dynamics of the least energy dissipation [6, 3]. The dissipation functional, or Rayleighian, captures the system’s dissipation rate, and the governing flow is the stationary condition of this functional subject to the incompressibility constraint.

Consider a viscous dynamical system where the dissipation functional D\mathcal{D} comprises the rate of viscous dissipation, and power inputs from an external force B∈Γ(T∗W)B \in\Gamma(T^*W) and a shape-dependent potential energy VφV_{\varphi}:

D(U)=12⟨⟨E∣T⟩⟩−⟨⟨B∣U⟩⟩+V˙φ=12∫M⟨E∣T⟩ dA−∫φ(M)⟨B∣U⟩ dA+∫M⟨δφVφ∣φ˙⟩ dA.(7)\begin{aligned} \mathcal{D}(U) = \frac{1}{2}\langle\langle E|T\rangle\rangle- \langle\langle B|U\rangle\rangle+ \dot{V}_{\varphi} \\ &= \frac{1}{2}\int_{M}\langle E|T\rangle\,\mathrm{d}A - \int_{\varphi(M)}\langle B|U\rangle\,\mathrm{d}A + \int_{M}\langle\delta_{\varphi}V_{\varphi}|\dot{\varphi}\rangle\,\mathrm{d}A. \tag*{(7)} \end{aligned}

Here, T=2μE∈Γ(TM⊗symmTM)T = 2\mu E \in\Gamma(TM \otimes_{\mathrm{symm}} TM) is the viscous stress tensor (assumed Newtonian with constant viscosity μ\mu), δφVφ\delta_{\varphi}V_{\varphi} is the conservative force obtained by taking the variation of VφV_{\varphi}, and φ˙=U∘φ\dot{\varphi} = U \circ\varphi is the material velocity.

Onsager’s variational principle determines the Stokes flow by solving the constrained optimization problem:

min⁡UD(U)subject toDIV⁡U=0.(8)\min_{U}\mathcal{D}(U) \quad\text{subject to} \quad\operatorname{DIV}U = 0. \tag*{(8)}

Using the fluid pressure p∈C∞(M)p \in C^{\infty}(M) as a Lagrange multiplier, this problem can be formulated as a minimax problem involving the augmented Rayleighian R\mathcal{R}:

min⁡Umax⁡pR(U,p),whereR(U,p)=D(U)+⟨⟨p,DIV⁡U⟩⟩.(9)\min_{U}\max_{p}\mathcal{R}(U,p), \quad\text{where} \quad\mathcal{R}(U,p) = \mathcal{D}(U) + \langle\langle p,\operatorname{DIV}U\rangle\rangle. \tag*{(9)}

On a closed surface, the stationary conditions yield the Stokes equations for an evolving surface:

2μDIV⁡∇KU−GRAD⁡p+B−δφVφ=0,DIV⁡U=0.(10)2\mu\operatorname{DIV}^{\nabla}\mathcal{K}U - \operatorname{GRAD}p + B - \delta_{\varphi}V_{\varphi} = 0, \qquad\operatorname{DIV}U = 0. \tag*{(10)}

The symmetric, negative-definite, Laplacian-like operator 2DIV⁡∇K:Γ(TW)→Γ(T∗W)2\operatorname{DIV}^{\nabla}\mathcal{K}:\Gamma(TW)\to\Gamma(T^{*}W) quantifies the viscous force, and is thus termed viscosity Laplacian. Although not the main focus of this paper, a common free energy in this context is the Helfrich Hamiltonian Vφ=∫WKH2 dAV_{\varphi}=\int_{W}KH^{2}\,\mathrm{d}A, which results in the bending force Bκ=−δφVφ=κN(ΔH+2H(H2−K))B_{\kappa}=-\delta_{\varphi}V_{\varphi}=\kappa N(\Delta H+2H(H^{2}-K)), where HH and KK are the mean and Gaussian curvatures, respectively, κ\kappa is the bending modulus, and NN is the surface normal [9]. This bending-driven relaxation dynamics is also used in the numerical examples of §4. The commonly used form of (10), expressed through normal-tangent decomposition, is derived in Appendix § 6.

Remarks on the evolving Stokes equations and their solvability

Note that the presence of a Killing vector field XX (KX=0\mathcal{K}X=0) can make the solution to (10) non-unique. In that case we select the least-norm solution, effectively projecting out rigid motions. This treatment aligns with scenarios where the surface is immersed with friction in a bulk fluid. A simple way to model friction is by augmenting the Rayleighian to Rα=R+α∥U−U0∥2/2\mathcal{R}^{\alpha}=\mathcal{R}+\alpha\lVert U-U_{0}\rVert^{2}/2, where U0U_{0} is the bulk velocity and α\alpha is the friction coefficient. This introduces a friction force Bα=α(U−U0)B_{\alpha}=\alpha(U-U_{0}) in the momentum equation. Such flow driven by the bulk fluid is used as a numerical example in § 4.3. When U0=0U_{0}=0, the solution approaches the least-norm solution as α→0\alpha\to0.

Methods

In this section, we present a simple numerical method to solve (10) on a triangle mesh with general geometry and topology. We begin by discretizing the strain rate tensor using finite elements, following the definition provided in (2). Based on the discretized strain rate, we construct the system’s Rayleighian. A nonlinearly stable, structure-preserving variational integrator is derived based on the Onsager’s variational principle. To keep the numerical scheme minimal, we demonstrate the method using first-order spatial and temporal discretizations.

We discretize the system on a closed triangular mesh MM of arbitrary genus, consisting of vertices, edges, and faces {V,E,F}\{\mathcal{V},\mathcal{E},\mathcal{F}\}. The vertex positions are given by the realization φ:M↪R3\varphi: M \hookrightarrow\mathbb{R}^{3}. MM has a tangent space TαMT_\alpha M and a surface normal NαN_\alpha at each face α∈F\alpha\in\mathcal{F} (cf. Figure 1).

Discretization of the strain rate tensor

Figure 1. Discretization of the strain rate tensor. The velocity UiU_i is defined at each vertex ii, with surface normals NαN_\alpha at each face α\alpha. The finite-element hat function Φj\Phi_j, shown by the color map, has local support around vertex jj. Differentiating UiU_i via Φj\Phi_j produces (∇U)β(\nabla U)_\beta, which is symmetrized and projected using Pβ=g−Nβ⊗NβP_\beta= g - N_\beta\otimes N_\beta to yield the strain rate tensor EβE_\beta on each face β\beta.

In the discrete setting, we express tensor-valued functions and operators using index notation, with upper indices denoting their components in Cartesian coordinates. Vertex and face elements are denoted with lower indices, where vertex indices are Roman letters and face indices are Greek letters. For example, we denote vertex positions as φji\varphi_j^i, face normals as NαiN_\alpha^i, face pressure as pαp_\alpha, for j∈Vj \in\mathcal{V}, α∈F\alpha\in\mathcal{F}, and i∈{1,2,3}i \in\{1,2,3\}.

Strain rate tensor and differential operators

To express the strain rate tensor EE on MM (cf. (2)) in Cartesian coordinate, we first realize it in R3\mathbb{R}^{3}. We denote this realization as 2E~=P((∇U)Tg+g∇U)P2\widetilde{E}=P((\nabla U)^{\mathsf{T}}g+g\nabla U)P, which satisfies E~⟦V,W⟧=E⟦PV,PW⟧\widetilde{E}\llbracket V,W\rrbracket=E\llbracket PV,PW\rrbracket for V,W∈Γ(R3)V,W\in\Gamma(\mathbb{R}^{3}). Here, the linear projector P=g−N⊗N:Γ(R3)→Γ(TM)P=g-N\otimes N:\Gamma(\mathbb{R}^{3})\to\Gamma(TM) projects R3\mathbb{R}^{3} vectors onto the surface. In the discrete setting (cf. Figure 1), the strain rate tensor E~αkj\widetilde{E}_\alpha^{kj} and Killing operator Kαlkji\mathcal{K}_{\alpha l}^{kji} are expressed in terms of the velocity UjiU_j^i, projector Pαij=δij−NαiNαjP_\alpha^{ij}=\delta^{ij}-N_\alpha^iN_\alpha^j, and R3\mathbb{R}^{3} component-wise surface derivative (∇P)αji(\nabla P)_{\alpha j}^{i} as:

2E~αkj=∑l,i2KαlkjiUli=∑l,i(φαik(∇P)αlj+φαij(∇P)αlk)Uli.(11)2\widetilde{E}_\alpha^{kj}=\sum_{l,i}2\mathcal{K}_{\alpha l}^{kji}U_l^i=\sum_{l,i}\left(\varphi_\alpha^{ik}(\nabla P)_{\alpha l}^{j}+\varphi_\alpha^{ij}(\nabla P)_{\alpha l}^{k}\right)U_l^i. \tag*{(11)}

Component-wise, (∇P)αji=∇iΦαj(\nabla P)_{\alpha j}^{i}=\nabla^{i}\Phi_{\alpha j} follows scalar finite element discretization, where Φαj\Phi_{\alpha j} is the hat function at vertex jj restricted to face α\alpha. The divergence (4) is given by the trace DIV⁡αli=∑jKαljji\operatorname{DIV}_{\alpha l}^{i}=\sum_j\mathcal{K}_{\alpha l}^{jji}. The gradient (5) and covariant divergence (6) can be obtained, up to a sign flip, by permuting the lower indices of the divergence and Killing operators, respectively.

Variational integrator by Onsager's variational principle

Given the vertex position φ(0)\varphi_{(0)} and pressure p(0)p_{(0)} at the current time t=0t=0, with a time step ϵ>0\epsilon>0, the velocity is discretized as U=φ˙/ϵ=(φ−φ(0))/ϵU=\dot{\varphi}/\epsilon=(\varphi-\varphi_{(0)})/\epsilon, and the power input as V˙φ=(Vφ−Vφ(0))/ϵ\dot{V}_{\varphi}=(V_{\varphi}-V_{\varphi_{(0)})/\epsilon}. State variables φ(ϵ)\varphi_{(\epsilon)} and p(ϵ)p_{(\epsilon)} at t=ϵt=\epsilon follow a time-incremental Onsager’s variational principle on a discrete Rayleighian (cf. § 2.3), φ(ϵ)\varphi(\epsilon), p(ϵ)=arg⁡min⁡φmax⁡pRp(\epsilon)=\arg\min_{\varphi}\max_{p}\mathcal{R}, where

R(φ,p;φ(0),p(0),ϵ)≔ϵ2[μ∥E~∥2+⟨⟨DIV⁡U,p⟩⟩+V˙φ−⟨⟨U∣B⟩⟩]≡μKλmkji(Aλγ)−1Kγnkjlφ˙ilφ˙jn+ϵ(φ˙ilDIV⁡αilpα+Vφ−φ˙ilBil)≡μφ˙TLφ˙+ϵ[(DIV⁡φ˙)Tp+Vφ−φ˙TB].(12)\begin{aligned} \mathcal{R}(\varphi,p;\varphi_{(0)},p_{(0)},\epsilon)&\coloneqq\epsilon^2\left[\mu\lVert\widetilde{E}\rVert^2+\langle\langle\operatorname{DIV}U,p\rangle\rangle+\dot{V}_{\varphi}-\langle\langle U|B\rangle\rangle\right]\\ &\equiv\mu K_{\lambda m}^{kji}(A_{\lambda\gamma})^{-1}K_{\gamma n}^{kjl}\dot{\varphi}_{i}^{l}\dot{\varphi}_{j}^{n}+\epsilon\left(\dot{\varphi}_{i}^{l}\operatorname{DIV}_{\alpha i}^{l}p_{\alpha}+V_{\varphi}-\dot{\varphi}_{i}^{l}B_{i}^{l}\right)\\ &\equiv\mu\dot{\varphi}^{\mathsf{T}}L\dot{\varphi}+\epsilon\left[(\operatorname{DIV}\dot{\varphi})^{\mathsf{T}}p+V_{\varphi}-\dot{\varphi}^{\mathsf{T}}B\right]. \tag*{(12)} \end{aligned}

Here, we assume Einstein summation and L=KTA−1KL=K^{\mathsf{T}}A^{-1}K is half of the discrete viscosity Laplacian (cf. § 2.3), where AλγA_{\lambda\gamma} is the area of the face γ\gamma for λ=γ\lambda=\gamma and zero otherwise.

Under any choice of VφV_{\varphi} (assuming B=0B=0), the variational integrator is unconditionally stable by design, as it preserves the system’s dissipative structure. The dynamics follow a gradient flow of VφV_{\varphi} in the metric space defined by the Stokes operator, ensuring a monotonic decrease in the Rayleighian: R(ϵ)≤R(0)\mathcal{R}(\epsilon)\leq\mathcal{R}(0). Therefore, the allowable time step size depends only on the solvability of the optimization in (12), which can be efficiently handled using standard numerical optimization methods. Here we use a simple gradient flow to solve this saddle-point optimization: Hφ˙=−δφR=−2μLφ˙−ϵ(DIV⁡Tp+δφVφ−B)H\dot{\varphi}=-\delta_{\varphi}\mathcal{R}=-2\mu L\dot{\varphi}-\epsilon(\operatorname{DIV}^{\mathsf{T}}p+\delta_{\varphi}V_{\varphi}-B) and hp˙=δpR=DIV⁡φ˙h\dot{p}=\delta_{p}\mathcal{R}=\operatorname{DIV}\dot{\varphi}, where HH and hh provide a general metric for the gradient flow 4. In § 4, we adopt the Helfrich Hamiltonian as VV and follow the discretization used in Zhu et al. [26].

As mentioned in § 2.4, equation (10) and its discrete version (12) have non-unique solution up to rigid body modes in R3\mathbb{R}^{3}. To ensure that we obtain continuous dynamics during time evolution, we consistently project out rigid body modes.

Results

In this section, we validate the discrete model, analyze its convergence properties, and demonstrate its applicability to manifolds with intricate geometries and topologies. Numerical validation against analytical solutions is presented in § 4.1, with further examples provided in § 4.2 and § 4.3 to showcase its effectiveness.

Validations

We validate the differential operators KK and GRAD on a spheroid with semi-axes aa and cc, parametrized by latitude β∈[−π/2,π/2]\beta\in[-\pi/2,\pi/2] and longitude θ∈(−π,π]\theta\in(-\pi,\pi], using the realization φ(β,θ)=[acos⁡βcos⁡θ,acos⁡βsin⁡θ,csin⁡β]\varphi(\beta,\theta)=[a\cos\beta\cos\theta,a\cos\beta\sin\theta,c\sin\beta].

In Figure 2(a), we test the operator KK by computing the lowest ten eigenvalues of L=KTA−1KL=K^{\mathsf{T}}A^{-1}K with LUi=λiAUiLU_i=\lambda_iAU_i (cf. (12)). As predicted by the continuous theory, there are six vanishing eigenvalues, forming a six dimensional kernel of KK corresponding to the space of rigid body motions in R3\mathbb{R}^{3}.

Validation plots and Helfrich–Stokes relaxation

Figure 2. Validation of our model and method: (a) Minimum eigenvalues λ1≤i≤10\lambda_{1\le i\le10} of LUi=λiAUiL\mathbf{U}_i=\lambda_i A\mathbf{U}_i; (b) Comparison of numerical GRAD 1 and analytical H(β)H(\beta) evaluations of the mean curvature on a spheroid; (c) Relative error εU\varepsilon_{\mathbf{U}} for the momentum equation and εp\varepsilon_p for the continuity equation as functions of mean edge length ll, with the inset showing the tangential forcing b\mathbf{b} constructed from a spherical harmonic; (d) Relative error in total area AA compared to the initial area A0A_0 during Helfrich-Stokes relaxation, with insets showing the initial frame (prolate spheroid) and the final frame (sphere) of the relaxation.

To validate the operator GRAD (A4), we check that the numerical evaluation of A−1∣DIV⁡T1∣A^{-1}|\operatorname{DIV}^{\mathsf{T}}1| aligns with the analytical mean curvature H(β)H(\beta) of an oblate spheroid with a=1a=1 and c=0.5c=0.5, as shown in Figure 2(b). In fact, this evaluation of mean curvature is identical with the established cotan Laplacian discretization (Meyer et al. [15]).

We evaluate the convergence of our discretization of (10) by comparing it to an analytical solution on a unit sphere. The forcing term is analytically prescribed as B=b+bnN=N×grad⁡ϕ+bnNB=b+b_nN=N\times\operatorname{grad}\phi+b_nN, where ϕ\phi is a spherical harmonic satisfying Δϕ≔div⁡grad⁡ϕ=λϕ\Delta\phi\coloneqq\operatorname{div}\operatorname{grad}\phi=\lambda\phi, NN is the surface normal, and bnb_n is a constant. On the unit sphere, bb is an eigenfunction of the viscosity Laplacian with eigenvalue λ+2\lambda+2. The analytical solution U‾=−b/(λ+2)\overline{U}=-b/(\lambda+2) and p‾=−bn/2\overline{p}=-b_n/2 satisfies (10) (μ=1\mu=1, δφVφ=0\delta_{\varphi}V_{\varphi}=0) exactly (cf. A7). Numerically, the relative residual of the momentum equation, εU=∥2A−1LU‾−A−1DIV⁡Tp−B∥/∥B∥\varepsilon_{\mathbf{U}}=\lVert2A^{-1}L\overline{\mathbf{U}}-A^{-1}\operatorname{DIV}^{\mathsf{T}}p-\mathbf{B}\rVert/\lVert\mathbf{B}\rVert, and the residual of the continuity equation, εp=∥A−1DIV⁡U∥\varepsilon_p=\lVert A^{-1}\operatorname{DIV}\mathbf{U}\rVert, shown in Figure 2(c), decrease approximately linearly as the mesh is refined.

Lastly, Figure 2(d) illustrates the total area of the fluid film during the Stokes relaxation (μ=1\mu=1) from a prolate spheroid (a=1a=1, c=2c=2) to a sphere under Helfrich energy (κ=0.05\kappa=0.05), without mesh reparametrization. Local incompressibility conserves the global area within a relative error of 10−510^{-5} throughout the evolution.

Relaxation of a genus-6 torus

We model the evolution of a high-genus, non-analytical lipid membrane (μ=1\mu=1, κ=0.05\kappa=0.05) based on the Helfrich-Stokes relaxation in Figure 3(a). In theory, the elastic Helfrich Hamiltonian should be monotonically dissipated through viscosity, V+Eμ=∫κH2dA+∫0t⟨⟨T∣E⟩⟩/2∣τdτ=constV+E_\mu=\int\kappa H^2\mathrm{d}A+\int_0^t\langle\langle T\mid E\rangle\rangle/2\mid_\tau\mathrm{d}\tau=\mathrm{const}. Although our numerical results in Figure 3(b) do not exactly match this theoretical expectation, they show very good agreement.

Helfrich-Stokes relaxation of a genus-6 torus and elastic energy and cumulative dissipation

Figure 3. (a) Helfrich-Stokes relaxation of a genus-6 torus. Snapshots are taken at t=0t = 0, 2, 8, 14, 20, and 40, from left to right, top to bottom. The animated simulation is available at https://youtu.be/Llh0_N0hCPw and in Movie 1 of the Supplementary Material. (b) Elastic energy V=∫κH2dAV = \int\kappa H^2 \mathrm{d}A and cumulative dissipation Eμ=∫0t⟨⟨T∣E⟩⟩/2∣τdτE_\mu= \int_0^t \left.\langle\langle\mathbf{T}|\mathbf{E}\rangle\rangle/2\right|_\tau\mathrm{d}\tau, which theoretically sum to a constant value.

Deformation of a lipid membrane under bulk flow

We model the deformation of a spherical lipid vesicle (μ=1\mu= 1, κ=0.01\kappa= 0.01) under the friction (α=0.5\alpha= 0.5, cf. § 2.4) from a constant bulk shear-extension flow defined by U0(θ,r,z)=U0shear+U0ext=(5rz)e^θ+(−0.5re^r+ze^z)\mathbf{U}_0(\theta,r,z) = \mathbf{U}_0^{\mathrm{shear}} + \mathbf{U}_0^{\mathrm{ext}} = (5rz)\hat{\mathbf{e}}_\theta+ (-0.5r\hat{\mathbf{e}}_r + z\hat{\mathbf{e}}_z) in cylindrical coordinates. As illustrated in Figure 4(a), this bulk flow stretches and twists the initially spherical vesicle along the zz-direction. A snapthrough-like instability occurs around t=3.7t = 3.7 between frame (iii) and frame (iv) in Figure 4(a). While the bulk friction governs most of the dynamics, the Helfrich energy dominates during this buckling event, as indicated by the dip of the energy dissipation rate in Figure 4(b) around t=3.7t = 3.7.

Deformation of a lipid membrane under bulk flow and energy dissipation rates

Figure 4. (a) Deformation of a lipid membrane under bulk flow, with the tangent velocity u\mathbf{u} shown as streamlines and the normal velocity UNU_N displayed in a colormap ranging from −1-1 to 1 (cf. Appendix § 6 for the tangent-normal decomposition). Snapshots (i)-(iv) are taken at t=1.5t = 1.5, 2.5, 3.4, and 3.8, respectively. The animated simulation is available at https://youtu.be/M0WsLihzRJk and in Movie 2 of the Supplementary Material. (b) Energy dissipation rate E˙\dot{E} of the system due to viscosity E˙μ=⟨⟨T∣E⟩⟩/2\dot{E}_\mu= \langle\langle\mathbf{T}|\mathbf{E}\rangle\rangle/2, elasticity E˙κ=⟨⟨Bκ∣U⟩⟩\dot{E}_\kappa= \langle\langle\mathbf{B}_\kappa|\mathbf{U}\rangle\rangle, and bulk friction E˙α=⟨⟨Bα∣U⟩⟩/2\dot{E}_\alpha= \langle\langle\mathbf{B}_\alpha|\mathbf{U}\rangle\rangle/2.

Conclusion

Using Onsager’s variational principle, we derived and reformulated the evolving Stokes equations on a manifold. We replaced the classic system of equations expressed in tangent-normal splitting with a simple, coordinate-free differential-geometric formulation. This approach directly leads to a straightforward discrete model and a numerical scheme to solve this long-standing problem in its full geometric generality.

Although the theoretical framework is not limited, the current minimal implementation is restricted to closed manifolds and employs first-order discretization. Future work could incorporate boundary conditions and higher-order discretizations. For applications, we expect the minimal system presented here to serve as a foundation for integrating more complex models for specific biophysical problems. These could include additional global volumetric/areal constraints, heterogeneity in material properties, in-plane anisotropy, and surface activity, as exemplified in [26, 27].

Acknowledgements. The authors acknowledge funding from National Science Foundation Grant No. DMS-2153520 (D.S.) and National Science Foundation CAREER Award 2239062 (A.C.). Additional support was provided by SideFX software.

Author ORCIDs. C. Zhu, https://orcid.org/0000-0003-1373-3492; D. Saintillan, https://orcid.org/0000-0001-9948-708X; A. Chern, https://orcid.org/0000-0002-9802-3619.

Appendix A: Evolving Stokes equations through normal-tangent decomposition

Differential operators for an evolving surface

The second fundamental form (i.e., curvature tensor), II∈Γ(T∗M⊗symmT∗M)\mathbb{II} \in\Gamma(T^{*}M \otimes_{\mathrm{symm}} T^{*}M), is given by the derivative of the surface normal NN:

II[v,w]=(Fv)⋅(∇FwN)=⟨v,Sw⟩,(13)\mathbb{II}[\mathbf{v},\mathbf{w}] = (F\mathbf{v}) \cdot(\nabla_{F\mathbf{w}}N) = \langle\mathbf{v},S\mathbf{w}\rangle, \tag*{(13)}

where S:Γ(TM)→Γ(TM)S:\Gamma(TM)\to\Gamma(TM) is the shape operator.

With the induced metric II, there is an intrinsic Levi-Civita covariant derivative ∇:Γ(TM)→Γ(TM⊗T∗M)\nabla:\Gamma(TM)\to\Gamma(TM\otimes T^{*}M) on MM. The covariant derivative in WW used in (A 1) and (1) can be expressed in terms of the intrinsic covariant derivative and curvature tensor as:

∇FwFv=∇wv−II[w,v]N.(14)\nabla_{F\mathbf{w}}F\mathbf{v}=\nabla_{\mathbf{w}}\mathbf{v}-\mathbb{II}[\mathbf{w},\mathbf{v}]N. \tag*{(14)}

We can decompose the velocity into tangent and normal parts, U=Fu+UNNU=F\mathbf{u}+U_NN, where u∈Γ(TM)\mathbf{u}\in\Gamma(TM), UN∈C∞(W)U_N\in C^{\infty}(W). Since FuF\mathbf{u} effectively embeds the tangent vector u\mathbf{u} in R3\mathbb{R}^{3}, we will abbreviate FuF\mathbf{u} as u\mathbf{u} when unambiguous. Given v∈TpM\mathbf{v}\in T_pM, (1) can be expressed as ∇FvU=∇vu+UNSv+N(⟨grad⁡UN,v⟩−II[u,v])\nabla_{F\mathbf{v}}U=\nabla_{\mathbf{v}}\mathbf{u}+U_NS\mathbf{v}+N\left(\langle\operatorname{grad}U_N,\mathbf{v}\rangle-\mathbb{II}[\mathbf{u},\mathbf{v}]\right). Combining with (2) and (3), we have 2(KU)[v,w]=⟨∇vu,w⟩+⟨∇wu,v⟩+2UNII[v,w]2(KU)[\mathbf{v},\mathbf{w}]=\langle\nabla_{\mathbf{v}}\mathbf{u},\mathbf{w}\rangle+\langle\nabla_{\mathbf{w}}\mathbf{u},\mathbf{v}\rangle+2U_N\mathbb{II}[\mathbf{v},\mathbf{w}], or

2KU=∇u+(∇u)⊤+2UNII=2Ku+2UNII,(15)2KU=\nabla\mathbf{u}+(\nabla\mathbf{u})^{\top}+2U_N\mathbb{II}=2\mathcal{K}\mathbf{u}+2U_N\mathbb{II}, \tag*{(15)}

where K:Γ(TM)→Γ(T∗M⊗symmT∗M)\mathcal{K}:\Gamma(TM)\to\Gamma(T^{*}M\otimes_{\mathrm{symm}}T^{*}M) denotes the intrinsic Killing operator. Combining (4), (5) and (A 3), the divergence and gradient can be expressed as

DIV⁡U=div⁡u+2HUN,GRAD⁡(p)=grad⁡p−2pHN,(16)\operatorname{DIV}U=\operatorname{div}\mathbf{u}+2HU_N,\qquad\operatorname{GRAD}(p)=\operatorname{grad}p-2pHN, \tag*{(16)}

where div⁡=tr⁡(K)\operatorname{div}=\operatorname{tr}(\mathcal{K}) is the intrinsic divergence, grad⁡=−div⁡∗\operatorname{grad}=-\operatorname{div}^{*} is the intrinsic gradient, and H=tr⁡II/2H=\operatorname{tr}\mathbb{II}/2 is the mean curvature of the surface. From (6) and (A 3), we get:

DIV⁡∇T=div⁡∇T−⟨II∣T⟩N,(17)\operatorname{DIV}^{\nabla}T=\operatorname{div}^{\nabla}T-\langle\mathbb{II}|T\rangle N, \tag*{(17)}

where div⁡∇=−K∗:Γ(TM⊗TM)→Γ(T∗M)\operatorname{div}^{\nabla}=-\mathcal{K}^{*}:\Gamma(TM\otimes TM)\to\Gamma(T^{*}M) is the intrinsic covariant divergence.

Evolving Stokes equations

The viscosity Laplacian for an evolving surface can be explicitly decomposed as:

2DIV⁡∇KU=2[div⁡∇Ku+div⁡∇(UNII)−(⟨II,Ku⟩+UN∥II∥2)N]=(Δ+K+grad⁡∘div⁡)u+2S(grad⁡UN)+4UNgrad⁡H−2[⟨∇u,II⟩+UN(4H2−2K)]N,(18)\begin{aligned} 2\operatorname{DIV}^{\nabla}\mathcal{K}U=2\left[\operatorname{div}^{\nabla}\mathcal{K}u+\operatorname{div}^{\nabla}(U_N\mathit{II})-\left(\langle\mathit{II},\mathcal{K}u\rangle+U_N\lVert\mathit{II}\rVert^2\right)N\right] \\ &=\left(\Delta+K+\operatorname{grad}\circ\operatorname{div}\right)u+2S(\operatorname{grad}U_N)+4U_N\operatorname{grad}H -2\left[\langle\nabla u,\mathit{II}\rangle+U_N(4H^2-2K)\right]N, \tag*{(18)} \end{aligned}

where Δ=div⁡∇∇:Γ(TM)→Γ(T∗M)\Delta=\operatorname{div}^{\nabla}\nabla:\Gamma(TM)\to\Gamma(T^*M) is the intrinsic Bochner Laplacian, and K=det⁡IIK=\det\mathit{II} is the Gaussian curvature.

To summarize, by substituting (A 3)-(A 6) into (10), we get the system of evolving Stokes equations subject to external force B=b+bnNB=b+b_nN and bending force Bκ=κN(ΔH+2H(H2−K))B_\kappa=\kappa N\left(\Delta H+2H(H^2-K)\right):

μ[(Δ+K)u+2Q(grad⁡UN)+2UNgrad⁡H]−grad⁡p+b=0,\mu\left[(\Delta+K)u+2Q(\operatorname{grad}U_N)+2U_N\operatorname{grad}H\right]-\operatorname{grad}p+b=0,
−2μ[⟨∇u,II⟩+UN(4H2−2K)]+2pH+bn+κ(ΔH+2H(H2−K))=0,-2\mu\left[\langle\nabla u,\mathit{II}\rangle+U_N(4H^2-2K)\right]+2pH+b_n+\kappa\left(\Delta H+2H(H^2-K)\right)=0,
div⁡u+2HUN=0.(19)\operatorname{div}u+2HU_N=0. \tag*{(19)}

Note that the Hopf differential Q=S−HIQ=S-HI is the traceless part of the shape operator SS that represents the deviatoric curvature. Formulation (A 7) agrees with earlier results by Scriven [21] and Arroyo & DeSimone [3].

References

  1. [1]Al-Izzi, Sami C. & Morris, Richard G. 2021 Active flows and deformable surfaces in development. Seminars in Cell & Developmental Biology 120, 44–52.arxiv.org/abs/2103.12264
  2. [2]Al-Izzi, S. C. & Morris, R. G. 2023 Morphodynamics of active nematic fluid surfaces. J. Fluid Mech. 957, A4.DOI
  3. [3]Arroyo, M. & DeSimone, A. 2009 Relaxation dynamics of fluid membranes. Phys. Rev. E 79, 031915.pubmed.ncbi.nlm.nih.gov/19391979
  4. [4]Bader, B. W. & Kolda, T. G. 2008 Efficient MATLAB computations with sparse and factored tensors. SIAM J. Sci. Comput. 30, 205–231.
  5. [6]Doi, M. 2011 Onsager’s variational principle in soft matter. J. Phys. Condens. Matter 23, 284118.DOI
  6. [7]Exerowa, D. & Kruglyakov, P. M. 1997 Foam and foam films: theory, experiment, application. Elsevier.
  7. [8]Gross, B. J. & Atzberger, P. J. 2018 Hydrodynamic flows on curved surfaces: Spectral numerical methods for radial manifold shapes. J. Comput. Phys. 371, 663–689.DOI
  8. [9]Helfrich, W. 1973 Elastic properties of lipid bilayers: theory and possible experiments. Z. Naturforsch. C 28, 693–703.pubmed.ncbi.nlm.nih.gov/4273690
  9. [10]Knöppel, F., Crane, K., Pinkall, U. & Schröder, P. 2013 Globally optimal direction fields. ACM Trans. on Graph. 32, 1–10.DOI
  10. [11]Koba, H., Liu, C. & Giga, Y. 2017 Energetic variational approaches for incompressible fluid systems on an evolving surface. Q. Appl. Math. 75, 359–389.DOI
  11. [12]Krause, V. & Voigt, A. 2023 A numerical approach for fluid deformable surfaces with conserved enclosed volume. J. Comput. Phys. 486, 112097.
  12. [13]Marsden, J. E. & Hughes, T. J. R. 1994 Mathematical Foundations of Elasticity. Dover.
  13. [14]Metselaar, L., Yeomans, J. M. & Doostmohammadi, A. 2019 Topology and morphology of self-deforming active shells. Phys. Rev. Lett. 123, 208001.DOI
  14. [15]Meyer, M., Desbrun, M., Schröder, P. & Barr, A. H. 2003 Discrete differential-geometry operators for triangulated 2-manifolds. In Visualization and Mathematics III, pp. 35–57. Springer.DOI
  15. [16]Miura, T.-H. 2018 On singular limit equations for incompressible fluids in moving thin domains. Q. Appl. Math. 76, 215–251.arxiv.org/abs/1703.09698
  16. [17]NESTLER, M., NITSCHKE, I. & VOIGT, A. 2019 A finite element approach for vector- and tensor-valued surface PDEs. J. Comput. Phys. 389, 48–61.DOI
  17. [18]REUTHER, S., NITSCHKE, I. & VOIGT, A. 2020 A numerical approach for fluid deformable surfaces. J. Fluid Mech. 900, R8.DOI
  18. [19]SAFFMAN, P. G. & DELBRÜCK, M. 1975 Brownian motion in biological membranes. Proc. Natl. Acad. Sci. USA 72, 3111–3113.pubmed.ncbi.nlm.nih.gov/1059096
  19. [20]SAHU, A., OMAR, Y. A. D., SAUER, R. A. & MANDADAPU, K. K. 2020 Arbitrary Lagrangian–Eulerian finite element method for curved and deforming surfaces: I. General theory and application to fluid interfaces. J. Comput. Phys. 407, 109253.DOI
  20. [21]SCRIVEN, L. E. 1960 Dynamics of a fluid interface equation of motion for Newtonian surface fluids. Chem. Eng. Sci. 12, 98–108.DOI
  21. [22]TORRES-SÁNCHEZ, A., SANTOS-OLIVÁN, D. & ARROYO, M. 2020 Approximation of tensor fields on surfaces of arbitrary topology based on local Monge parametrizations. J. Comput. Phys. 405, 109168.
  22. [23]TORRES-SÁNCHEZ, A., MILLÁN, D. & ARROYO, M. 2019 Modelling fluid deformable surfaces with an emphasis on biological interfaces. J. Fluid Mech. 872, 218–271.
  23. [24]VOIGT, A. 2019 Fluid deformable surfaces. J. Fluid Mech. 878, 1–4.DOI
  24. [25]YIN, H., NABIZADEH, M. S., WU, B., WANG, S. & CHERN, A. 2023 Fluid cohomology. ACM Trans. Graph. 42, 126.DOI
  25. [26]ZHU, C., LEE, C. T. & RANGAMANI, P. 2022 Mem3DG: Modeling membrane mechanochemical dynamics in 3D using discrete differential geometry. Biophys. Rep. 2, 100062–100062.
  26. [27]ZHU, C., SAINTILLAN, D. & CHERN, A. 2024 Active nematic fluids on Riemannian 2-manifolds. arXiv preprint arXiv:2405.06044 .arxiv.org/abs/2405.06044

Paper details

vlnc:W_cbcjrk8e
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

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

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents