What are the solution existence and uniqueness conditions for Stokes' flow?

$$\begin{gathered} \nabla p = \mu \Delta \vec{u} + \vec{f} \\ \nabla \cdot \vec{u} = 0 \end{gathered}$$

Maybe you could also provide some articles or books about the topic? Most physics books seem not to care about these details.

  • $\begingroup$ @ColinMcFaul, is it ok to reask it on math.SE or I should flag and hope for a migration? $\endgroup$ – Džuris May 18 '13 at 20:12
  • $\begingroup$ The etiquette is to flag for a migration. But I think the mods will migrate only if several people think it's off-topic here. $\endgroup$ – Colin McFaul May 18 '13 at 22:10
  • $\begingroup$ The second condition should be $\nabla \cdot u= 0$, i.e., incompressible? $\endgroup$ – Shuhao Cao Jun 6 '13 at 4:16
  • $\begingroup$ @ShuhaoCao, adding the dot changes anything there? $\endgroup$ – Džuris Jun 11 '13 at 14:43
  • $\begingroup$ The dot means dot product $\nabla \cdot $ is the divergence operator: en.wikipedia.org/wiki/Divergence What you wrote was gradient operator without that dot. The Stokesian flow we are learning is most of the time *incompressible*, using math terms, should be divergence free, i.e., $\nabla \cdot \vec{u} = 0$. $\endgroup$ – Shuhao Cao Jun 11 '13 at 15:13

Michiel's answer is more from the aspects of physics. Here is the pde style answer.

Short answer: The Stokes flow's variational problem is well-posed (uniqueness and existence) in certain Hilbert spaces pair which relies on inf-sup condition.

Functional equations:$\newcommand{\b}{\boldsymbol}$ Suppose $\mu=1$, the Stokesian flow can be written in the following way (I believe this is called the Pressure-Velocity formulation): $$\left\{ \begin{aligned}&-\Delta \b{u} + \nabla p =\b{f}, \\ &\nabla\cdot \b{u} =0.\end{aligned}\right.\tag{1} $$ In operator form this can be written as the following abstract problem: $$ \begin{pmatrix}A & B' \\ B& 0\end{pmatrix}\begin{pmatrix}\b{u}\\p \end{pmatrix} = \begin{pmatrix}\b{f}\\0 \end{pmatrix}, $$ where $A = -\Delta$ is the vector Laplacian, and $B = -\nabla\cdot$ with $B' = \nabla$, $\b{u}\in X$, and $p\in Y$, the operators: $$ A: X\to X',\quad B:X\to Y', \quad \text{and}\quad B': Y\to X'. $$

Stokes problem is well-posed when the follpwing operator is an isomorphism: $$ \mathscr{S}:(\b{v},q)\in X\times Y \mapsto (A\b{v}+B'q,B\b{v})\in X'\times Y'. $$

Normally, the isomorphism is either proved using Lax-Milgram through coercivity to pin down a fixed point, or using Fredholm alternative.

The sufficient condition for this is a weak version of coercivity (you can view it as invertibility of an operator):

$B: \mathrm{ker}(B)^{\perp}\subset X \to Y'$ is an isomorphism and $\|\b{v}\|_X \leq \beta \|B\b{v}\|_{Y'}$.

$B': Y \to (\mathrm{ker}(B)^{\perp})'$ is an isomorphism and $\|q\|_Y \leq \beta \|B'q\|_{X'}$.

Then by closed range theorem, Babuska proved an equivalence of these conditions with the inf-sup condition(in that pdf link 1.1). Whenever that condition holds for certain Hilbert spaces pair $X\times Y$, (1)'s variational problem has a unique solution.

Weak formulation:

The weak formulation for the abstract version of (1) is then:

$$\left\{ \begin{aligned}\langle A \b{u},\b{v}\rangle + \langle p,B\b{v}\rangle =\langle\b{f},\b{v}\rangle,\quad \forall \b{v} \in X&, \\ \langle q,B\b{u}\rangle =0,\;\;\;\quad\quad \forall q \in Y.&\end{aligned}\right.\tag{2} $$ Possible pairs of Hilbert spaces $X\times Y$ mentioned above are: $$\begin{gathered} H^1_0(\Omega)\times \{q\in L^2(\Omega):\int_{\Omega}q=0\}, \\ H(\mathrm{div})= \{\b{v}\in L^2(\Omega):\nabla\cdot \b{v}\in L^2(\Omega)\}\times L^2(\Omega). \end{gathered}$$ Using integration by parts for (2) leads to: $$\left\{ \begin{aligned}\int_{\Omega} \mathrm{tr}\big((\nabla \b{u})^T \nabla \b{v}\big) +\int_{\Omega}p(\nabla \cdot \b{v}) =\int_{\Omega}\b{f}\cdot\b{v},\quad \forall \b{v} \in X&, \\ \int_{\Omega}q(\nabla \cdot \b{u}) =0,\quad\quad\quad \forall q \in Y.&\end{aligned}\right.\tag{3} $$

Problem (3) can be viewed as a constraint minimization problem for the following conjugate functionals also (viewing the pressure $p$ as a Lagrange multiplier): denote $E(\b{v}) = (\nabla \b{v}^T +\nabla \b{v})/2$ (symmetric part of the Jacobian), the stationary strain tensor, then $\mathrm{tr}\big((\nabla \b{v})^T \nabla \b{v}\big)= |E(\b{v})|^2 $ (a notation usually used in elasticity PDEs). Let $$ \mathcal{L}(\b{v},q) = \int_{\Omega}|E(\b{v})|^2 - \int_{\Omega} \b{f}\cdot \b{v} - \int_{\Omega} q\nabla\cdot \b{v}, $$ and $$\mathcal{J}(\b{v}) = \sup_{q\in Y}\mathcal{L}(\b{v},q) ,$$ then our goal is to minimize $\mathcal{J}$ in $X$ (like looking for a saddle point).

  • $\begingroup$ Thanks, this is quite the answer that I wanted, but I don't understand everything. The main problem is that I don't understand how could I check if particular problem is solvable or not - where do I plug in the boundary conditions? Maybe you could show an example? Could you please use this and show that the problem in the Stokes paradox is unsolvable? $\endgroup$ – Džuris Jun 11 '13 at 14:47
  • $\begingroup$ @Juris Hi, Juris. If you could show your particular problem, then please. For Stokes' paradox, the idea is that for an infinite cylinder, a sensible global solution can't be established for small Reynolds number. However, the good thing is, if we are dealing with bounded piecewise smooth domain, all those problems are gone. $\endgroup$ – Shuhao Cao Jun 11 '13 at 15:21
  • $\begingroup$ OK, how about this one: a box of shear flow with given velocity $\vec{u}=u_0z\vec{e_x}$ for $x=0$,$x=L_x$,$y=0$,$y=L_y$,$z=0$,$z=L_z$ and stationary ($\vec{u}=\vec{0}$) sphere of radius $r<L_z<L_y<L_x$ at $\frac{1}{2}(L_x,L_y,L_z)$. $\endgroup$ – Džuris Jun 11 '13 at 19:38
  • $\begingroup$ @Juris In this problem, I am not sure what to do though, for the solution $\vec{u}$ is already given. Do you wanna check if the flow satisfying the Stokes equation? I suggest you post a new question about this. Based on your problem, guess my answer doesn't really suit your instant need, but the point is, in a bounded smooth domain, the Stokes equation is well posed with Dirichlet boundary condition. Also the boundary condition is incorporated into the function spaces. $\endgroup$ – Shuhao Cao Jun 11 '13 at 20:06

There is one famous example in which there is no solution for the Stokes' flow case: Stokes flow around a cylinder, which is approriately named Stokes' paradox. In this case it is impossible to match the boundary conditions both at infinity (uniform flow) and at the cylinder surface (no-slip) with Stokes flow dynamics. See e.g. paragraph 6.4 of the Fluid Dynamics Lecture Notes of Jacques Lewalle. The breakdown of a solution means that there will always be some inertial effect, regardless how small the Reynolds number is.

An approximate solution to this issue was first proposed by Oseen who introduced a linearized inertial term to account for inertial contributions in the far field.

Later, Proudman and Pearson calculated a more precise solution through asymptotic expansions and matching of the far field solution and the near-cylinder solution.

So to answer your question: existence for Stokes' flow is not guaranteed even though the criterion $Re << 1$ is satisfied. A pretty good explanation for the precise reasons of non-existence is given in chapter 7 of the Fluid Dynamics I lecture notes by Prof. Childress. In the same document they also show (in paragraph 7.2) that Stokes' flow does exhibit uniqueness for non-trivial cases (i.e. $\textbf{u}\neq0$).

For more details on conditions for solvability: there is quite a lot of mathematical fluid dynamics literature on the solvability of Stokes' flow. Nazarov and Pileckas reference a number of them in their paper with the telling title "On the Solvability of the Stokes and Navier-Stokes Problems in the Domains That Are Layer-Like at Infinity"

  • 1
    $\begingroup$ Thanks, I know of Stokes' paradox, but has anyone specified conditions to distinguish solvable and unsolvable situations? $\endgroup$ – Džuris May 18 '13 at 12:37
  • $\begingroup$ Don't pin me down on it, but I believe the situation is only unsolvable if the far-field is unbounded $\endgroup$ – Michiel May 18 '13 at 12:43
  • 4
    $\begingroup$ Dear @michielm: Please mention properly the author, title, etc, of the lecture notes (as opposed to just saying 'this document'). This serves at least 2 purposes: i) I'm sure the author appreciates the proper recognition, and ii) in case of future link rot, we can reconstruct your answer. $\endgroup$ – Qmechanic May 18 '13 at 13:50
  • $\begingroup$ @Qmechanic Ok! I have corrected the referencing. $\endgroup$ – Michiel May 18 '13 at 15:23

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.