Exact solutions of fluid equations on a sphere Sun-Chul Kim, Habin Yim Abstract Exact solutions of both the Navier-Stokes and Euler equations are found on the surface of a sphere. Under the assumption of a vanishing convection term, the flow of two oppositely rotating point vortices at the poles turns out to be the unique common solution. 2020 Mathematics Subject Classification : 76D05 Keywords . fluid equations, exact solutions, flow on a sphere, vanishing convection 1 Introduction Fluid flows on the surface of a sphere are directly applicable to atmospheric or ocean circulations on a macroscopic scale where the curvature effect is not negligible. Moreover, due to the compactness of the sphere, the total flow displays different properties compared to the planar case. Many efforts have been made to understand and propose possible flow models, taking into account these aspects, which differ from two-dimensional planar fluid flow. One such attempt is to find exact solutions that provide an accurate mathematical model of real physical flows and can also be used to improve the accuracy and reliability of numerical simulations. The main difficulty here is the high nonlinearity of fluid equations (Navier-Stokes, Euler), which permits no closed-form general solutions. However, various types of exact solutions have been found for two- or three-dimensional cases [ 3 , 6 , 9 , 1 , 7 ] , with or without certain symmetries. For flows on the surface of a sphere, the situation is quite similar. The geometric and topological characteristics of a sphere, such as positive curvature and compactness, as well as nonlinearity, make it more difficult to find exact solutions in general. To the best of the author’s knowledge, there are only a few known exact solutions for flows on the surface of a sphere [ 2 , 5 , 4 ] . In connection with this, the first author specified and classified certain exact solutions of both the Navier-Stokes and Euler equations in two dimensions, as well as for the axisymmetric case [ 8 ] . These are characterized by the vanishing nonlinear convection term in the equation and are possibly called basic solutions , as they include well-known fundamental solutions, such as Poiseuille and Couette flows, rigid rotation, etc., in fluid dynamics. In this note, using a similar method, we attempt to find and classify analogous basic solutions of fluid flows on the surface of a sphere. More specifically, we consider two-dimensional stationary solutions of the Navier-Stokes (and Euler) equations on a sphere of radius R . In coordinate free form, the equations are \displaystyle({\bf u}\cdot\nabla){\bf u} \displaystyle= \displaystyle\nu\triangle{\bf u}-\frac{1}{\rho}\nabla p, (1) \displaystyle{\rm div}\,{\bf u} \displaystyle= \displaystyle 0. (2) Here {\bf u} denotes the velocity field, p the pressure. \nu and \rho are positive constants representing the kinematic viscosity and the mass density, respectively. If we confine the flow on the surface of sphere \Sigma , we define the stream function \psi=\psi(\theta,\phi) in the spherical coordinate (r,\theta,\phi),\,r>0,0<\theta<\pi,0<\phi\leq 2\pi and represent the velocity {\bf u} by, {\bf u}=\nabla\psi\times{\bf e_{r}}=(0,u_{\theta},u_{\phi})=\left(0,\frac{1}{R\sin\theta}\frac{{\partial}\psi}{{\partial}\phi},-\frac{1}{R}\frac{{\partial}\psi}{{\partial}\theta}\right). Taking curl to the equation ( 1 ) we introduce the vorticity {\bf\omega}=\nabla\times{\bf u}=\omega{\bf e_{r}}=(\omega,0,0). For convenience, we assume R=1 hereafter. By these functions \psi and {\omega} , whose existence as single-valued functions is assumed, the equations ( 1 ) and ( 2 ) are rewritten as follows: \displaystyle\frac{1}{R^{2}\sin\theta}\left(\frac{\partial\psi}{\partial\phi}\frac{\partial\omega}{\partial\theta}-\frac{\partial\psi}{\partial\theta}\frac{\partial\omega}{\partial\phi}\right) \displaystyle= \displaystyle\nu\tilde{\nabla}^{2}\omega, (3) \displaystyle-\tilde{\nabla}^{2}\psi \displaystyle= \displaystyle\omega. (4) where we define the Laplace-Beltrami operator \tilde{\nabla}^{2} on the sphere by \tilde{\nabla}^{2}\omega=\frac{1}{\sin\theta}\frac{{\partial}}{{\partial}\theta}\left(\sin\theta\frac{{\partial}\omega}{{\partial}\theta}\right)+\frac{1}{\sin^{2}\theta}\frac{{\partial}^{2}\omega}{{\partial}\phi^{2}}. (5) We are interested in the exact solutions and their physical implications here. Before doing so, we list one important integral constraint of the solution. As the sphere is a closed (and compact) surface, the total vorticity should vanish by the Stokes’ theorem: \int\!\!\int_{\Sigma}\omega\;dA=0, (6) which is often called the “Gauss constraint”. 2 Solutions depending only on \theta We first seek basic solutions by assuming \psi (and \omega ) is a function of \theta only. From the equations (\ref{eq:eq:ns03}) and (\ref{eq:lap_bel}) , we derive \frac{d\omega}{d\theta}=\frac{K_{1}}{\sin\theta}, which then is integrated into \omega(\theta)=K_{1}\log\left(\tan\frac{\theta}{2}\right)+K_{2}, where K_{1},K_{2} are arbitrary constants. This solution has singularity at \theta=0,\pi and thus undefined at the north and south poles. However, for K_{2}=0 , it satisfies the integral relation (\ref{eq:eq:constr}) and the corresponding streamfunction and velocity are well defined everywhere. Let us verify this fact. The streamfunction satisfies \frac{{\partial}}{{\partial}\theta}\left(\sin\theta\frac{{\partial}\psi}{{\partial}\theta}\right)=-K_{1}\log\left(\tan\frac{\theta}{2}\right), which simplifies into \psi^{\prime\prime}+(\cot\theta)\psi^{\prime}=-K_{1}\log\left(\tan\frac{\theta}{2}\right). Solving for u_{\phi}=-\psi^{\prime} under the continuity on u as \theta\rightarrow 0,\pi , we obtain u_{\phi}(\theta)=-\frac{K_{1}}{\sin\theta}\int_{0}^{\theta}\sin\varphi\log\left(\tan\frac{\varphi}{2}\right)d\varphi. For convenience, putting K_{1}=1 , the velocity assumes its maximum value at \theta=\frac{\pi}{2} and monotonically decreases to zero at the south and north poles. The vorticity then satisfies \int_{0}^{2\pi}\!\!\int_{0}^{\frac{\pi}{2}}\omega\;d\theta d\phi=-\int_{0}^{2\pi}\!\!\int_{\frac{\pi}{2}}^{\pi}\omega\;d\theta d\phi=\log 2. This solution is symmetric with respect to the equator and is independent with the Coriolis effect. We call this the basic solution, which identifies with two point vortices at the north and south poles with opposite vortex strengths. 3 Solutions of \psi=f(\omega) and \omega=g(\psi) Next, we seek exact solutions which have a functional relation between \psi and \omega and vanishing convection term. We first suppose that the vector field given by a stream function \psi represents an Euler flow and, at the same time, a Navier-Stokes flow. This amounts to assuming that \frac{\partial\psi}{\partial\phi}\frac{\partial\omega}{\partial\theta}-\frac{\partial\psi}{\partial\theta}\frac{\partial\omega}{\partial\phi}=0 (7) and \tilde{\nabla}^{2}\omega=0. (8) In the two dimensional flow case, these equations provide the most fundamental solutions containing Poiseulle, Couette, rigid rotation, and shear flows. Analogous examples are studied and listed in the three dimensional axisymmetric case [ 8 ] . In fact, it is a slightly over-determined system and there appear some nontrivial solutions. However, in the case of a sphere surface, we state a negative result. Theorem 1 There is no nontrivial solution except the basic solution satisfying the functional relation \psi=F(\omega) with a smooth single-valued function F on a sphere. Proof. We first introduce a new variable \chi=\log(\tan\frac{\theta}{2}) and regard the vorticity as a function with \chi and \phi i.e. \omega=\omega(\chi,\phi). This simplifies the Laplace-Beltrami operator in the usual Laplacian form : \tilde{\nabla}^{2}=\frac{1}{\sin^{2}\theta}\left(\frac{{\partial}^{2}}{{\partial}\chi^{2}}+\frac{{\partial}^{2}}{{\partial}\phi^{2}}\right)=\frac{1}{\sin^{2}\theta}\triangle_{\chi,\phi}. We compute then by ( 8 ) -\sin^{2}\theta\;\omega=\triangle_{\chi,\phi}\psi=F^{\prime\prime}(\omega)|\nabla\omega|^{2}+F^{\prime}(\omega)\triangle_{\chi,\phi}\omega=F^{\prime\prime}(\omega)|\nabla\omega|^{2}. It therefore follows that we may define \Phi by \frac{1}{\sin^{2}\theta}|\nabla\omega|^{2}=\frac{-\omega}{F^{\prime\prime}(\omega)}\equiv\Phi(\omega). Since \omega is harmonic in \chi and \phi , there exists a complex analytic function W(z)=W(\chi+{\rm i}\phi) such that \omega={\rm Re}\left[W(z)\right] . Consequently, \log|\nabla\omega|^{2}=\log\Phi+2\log(\sin\theta)=2{\rm Re}\left[\log\left(\frac{{\rm d}W}{{\rm d}z}\right)\right] is harmonic, too. This implies that 0=\triangle_{\chi,\phi}(\log\Phi(\omega)+2\log(\sin\theta))=\left(\frac{\Phi^{\prime}}{\Phi}\right)^{\prime}|\nabla\omega|^{2}-2\sin^{2}\theta, which gives us the differential equation for \Phi , \left(\frac{\Phi^{\prime}({\omega}(\theta,\phi)}{\Phi({\omega}(\theta,\phi)}\right)^{\prime}\Phi({\omega}(\theta,\phi)=2. (9) We will show that \omega is in fact a function of \theta only. (Strictly speaking, \omega={\rm constant} in some open set and ( 9 ) holds in the remaining region. But, for the moment, we proceed as if we have proved that \omega is constant throughout {\bf R}^{2} or ( 9 ) throughout {\bf R}^{2} .) The equation ( 9 ) can be easily integrated and we have \Phi(\omega)=Ae^{B\omega} , where A and B are real constants. This, in turn, implies that |\nabla\omega|^{2}=\left|\frac{{\rm d}W}{{\rm d}z}\right|^{2}/\sin^{2}\theta=Ae^{B\omega}. (10) Note that the case where \omega={\rm constant} is included in this equation as a special case that A=0 . Clearly A\geq 0 . Since A=0 implies the case where the vorticity is constant, we may assume that A>0 . Applying \log on ( 10 ), we have 2{\rm Re}\left[\log\left(\frac{{\rm d}W}{{\rm d}z}\right)\right]=B{\rm Re}\left[W(z)\right]+\log A+2\log(\sin\theta). To find such solution, since \sin\theta=\frac{2e^{\chi}}{1+e^{\chi}} , we should find an analytic function f(\chi+i\phi) satisfying the relation \log\frac{2e^{\chi}}{1+e^{\chi}}={\rm Re}f(\chi+i\phi), but there is no such function from the Cauchy-Riemann condition. We thus conclude \omega=\omega(\theta) and the solution then degenerates to the basic solution case which we mentioned earlier. The reverse case \omega=g(\psi) is similar. The problem is to find \psi and G such that \displaystyle-\triangle\psi \displaystyle= \displaystyle G(\psi), (11) \displaystyle\triangle^{2}\psi \displaystyle= \displaystyle 0. (12) Theorem 2 If G:{\bf R}\rightarrow{\bf R} is a single-valued smooth function, and ( 11 ) and ( 12 ) are satisfied, then \psi is again the basic solution. Proof. We have 0=\triangle G(\psi)=G^{\prime\prime}(\psi)|\nabla\psi|^{2}+G^{\prime}(\psi)\triangle\psi, which implies that |\nabla\psi|^{2}=\frac{G(\psi)G^{\prime}(\psi)}{G^{\prime\prime}(\psi)}. Since \nabla\omega=G^{\prime}(\psi)\nabla\psi , we obtain |\nabla\omega|^{2}=\frac{G(\psi)\left(G^{\prime}(\psi)\right)^{3}}{G^{\prime\prime}(\psi)}. Let the right hand side be denoted by J(\psi) . Then, the same procedure as in the previous case produces 0=\triangle(\log J(\psi)+2\log\sin\theta)=\frac{J^{\prime}}{J}\triangle\psi+\left(\frac{J^{\prime}}{J}\right)^{\prime}|\nabla\psi|^{2}-\sin^{2}\theta, which can be written as follows: -\frac{J^{\prime}}{J}G+\left(\frac{J^{\prime}}{J}\right)^{\prime}\frac{GG^{\prime}}{G^{\prime\prime}}=\sin^{2}\theta. Again, we proceed by a similar argument as in the case \psi=f({\omega}) . Namely, as the left side is a function of only \psi , let us write it into F(\psi(\theta,\phi))=\sin^{2}\theta. Differentiating on \phi gives, F^{\prime}(\psi(\theta,\phi))\frac{{\partial}\psi}{{\partial}\phi}=0, so either F^{\prime}=0 or \frac{{\partial}\psi}{{\partial}\phi}=0 . The first does not happen and it follows that \psi=\psi(\theta) , again returning to the basic case. We found no new solution for both cases. 4 Concluding remark There are various exact solutions of the Navier-Stokes and Euler equations. Among them, we may characterize and classify basic solutions such as Couette flow, Poiseuille flow, solid body rotation, etc., as those with a vanishing advection term. In other words, they are common solutions of the Navier-Stokes and Euler equations. Such solutions exist in both two and three dimensions [ 8 ] . However, on the surface of a sphere, we found only one solution involving two point vortices, which has singularities at two antipodal points. Aside from this, we analytically showed that no additional solutions exist. This is a negative result for the existence of exact solutions, imposing a rather strict condition for such common solutions. We suppose that this is likely due to the strong restrictions imposed by compactness and a positive curvature.