Abstract
We study the boundary layer theory for slightly viscous stationary flows forced by an imposed slip velocity at the boundary. According to the theory of Prandtl (in: International mathematical congress, Heidelberg, 1904; see Gesammelte Abhandlungen II, 1961) and Batchelor (J Fluid Mech 1:177–190, 1956), any Euler solution arising in this limit and consisting of a single “eddy” must have constant vorticity. Feynman and Lagerstrom (in: Proceedings of IX international congress on applied mechanics, 1956) gave a procedure to select the value of this vorticity by demanding a necessary condition for the existence of a periodic Prandtl boundary layer description. In the case of the disc, the choice—known to Batchelor (1956) and Wood (J Fluid Mech 2:77–87, 1957)—is explicit in terms of the slip forcing. For domains with non-constant curvature, Feynman and Lagerstrom give an approximate formula for the choice which is in fact only implicitly defined and must be determined together with the boundary layer profile. We show that this condition is also sufficient for the existence of a periodic boundary layer described by the Prandtl equations. Due to the quasilinear coupling between the solution and the selected vorticity, we devise a delicate iteration scheme coupled with a high-order energy method that captures and controls the implicit selection mechanism.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
Let \(M\subset \mathbb {R}^2\) be a bounded, simply connected domain. Consider the Navier–Stokes equations
Motion is excited through the boundary, where stick boundary conditions are supplied
where \(\hat{n}\) is the unit outer normal vector field on the boundary and \(\hat{\tau }=\hat{n}^\perp \), the unit tangent field. In the above, there is a given autonomous slip velocity \(f:\partial M \rightarrow \mathbb {R}\), which should be thought of as being generated by motion of the boundary (via so-called “stick" or “no slip" boundary conditions—the fluid velocity on the boundary matches its speed), and is responsible for the generation of complex fluid motions in the bulk. If the viscosity is large relative to the forcing, then it is easy to see that all solutions converge to a unique steady state as \(t\rightarrow \infty \) [38]. However, as viscosity is decreased, one generally expects solutions to develop and retain non-trivial variation in time, perhaps even forever harboring turbulent behavior. There one general exception to this expectation in a special setting, proved in Sect. 3:
Theorem 1
(Absence of turbulence) Let \(M=\mathbb {D}\) be the disk of radius \(\textsf{R}\) and \(f = \frac{1}{2} {\omega _0}\textsf{R}\) for any given \({\omega _0}\in \mathbb {R}\) be a constant slip on the boundary. For any distributionally divergence-free \(u_0\in L^2\), the unique Leray-Hopf weak solution converges at long time to solid body rotation \(u_\textsf{sb}=\frac{1}{2}{\omega _0}x^\perp \) having vorticity \(\omega _0\). In fact,
where \(\lambda _1\) is the first positive eigenvalue of \(-\Delta \) with Dirichlet boundary conditions on \(\mathbb {D}\).
Remark
The forcing (slip velocity) in Theorem 1 can be arbitrarily large and yet for any viscosity Navier–Stokes has a one-point attractor. This is the analogue of Marchioro’s results on the absence of turbulence on the torus with ‘gravest mode’ body forcing [31, 32].
Theorem 1 highlights a peculiarity of solid body rotation on the disk: if you center a circular basin on fluid on a record player, all motion will eventually be solid body. A question arises:
Question 1
What if the imposed slip is non-constant, or the domain is not a disk?
As mentioned above, one might expect that if either of these conditions is violated, time dependence generally survives. However, if the boundary forcing is special this need not be the case. For instance, consider the velocity field with constant (unit) vorticity on any M
Any such velocity field satisfies both the Euler and Navier–Stokes equations in the bulk. As such, it is a stationary solution of Euler, and also of Navier–Stokes provided \(u_*\) is taken as initial data and it is forced consistently on the boundary: \( f_* = u_{*}\cdot \tau . \) Thus, for any domain there is a family of non-trivial time-independent solutions uniformly in \(\nu \ge 0\).
For force sufficiently close to that generated by a stationary Euler solution, asymptotic stability may occur but is a delicate issue. To begin to understanding these issues, we are interested in the question of the existence of sequence of stationary Navier–Stokes solutions approximating an Euler flow in this setting. The Prandtl–Batchelor theory [1, 35] provides a restriction on the type of stationary Euler solutions that can arises as inviscid limits. Namely, it stipulates that they have constant vorticity within closed streamlines, so-called “eddies". See also Childress [4, 5] and Kim [22, 23]. The result, proved in Sect. 4, is:
Theorem 2
(Prandtl–Batchelor Theorem) Let \(M\subset \mathbb {R}^2\) be a simply connected domain with smooth boundary. Let \(\psi :M\rightarrow \mathbb {R}\) be a \(W^{4,1+}(M)\) streamfunction of a steady, non-penetrating solution of Euler \(u_e\) having a single stagnation point which is non-degenereate in a sense that the period of revolution of a particle is a differentiable function of the streamline. Suppose \(\{u^\nu \}_{\nu >0}\) is a family satisfying (1), (2) together with
for all interior open subsets \(U\subset M\). Then \(u_e= {\omega _0}u_*\) for a constant \({\omega _0}\in \mathbb {R}\) and \(u_*\) is (5).
See Fig. 1. for a cartoon of this convergence. In the above theorem, M can be thought of as a streamline of an Euler solution occupying some larger spatial domain. If the limiting Euler solution consists of multiple eddies, the above shows that, within each eddy, the vorticity tends to become constant. The vorticity of the resulting solution would be a staircase landscape separated, perhaps, by vortex sheets. Such a picture is consistent with the general expectation of the emergence of weak solutions in the inviscid limit on bounded domains [7, 8, 12]. We remark that similar selection principles to Theorem 2 appear also in two-dimensional passive scalar problems [33, 36], and in steady heat distribution in three-dimensional integrable magnetic fields [10].
If the boundary data is a sufficiently small perturbation of the corresponding slip of an unit vorticity Euler flow \(u_*\) on whole vessel M,
then the inviscid limit of steady Navier–Stokes solutions might be expected to consist of just a single eddy having constant vorticity (5), that is, for some appropriate constant \({\omega _0}\in \mathbb {R}\),
See Conjecture 1. This naturally leads to the following question:
Question 2
Given boundary data (3), (4), (7), how is the limiting vorticity \({\omega _0}\) (8) selected?
This question was discussed by Batchelor (1956) [1] and Wood (1957) [40] for disk domains, and the resulting prediction is called the Batchelor–Wood formula. This analysis was done independently by Feynman–Lagerstrom (1956) [14] who also generalized this formula to domains with non-constant curvature. See also [29, 30]. The idea is: Navier–Stokes with small viscosity should approximate Euler (constant vorticity) in the bulk of the domain, and interpolate to the given boundary conditions across a layer of width \(\sqrt{\nu }\). In this layer, predicted by Prandtl [35], the leading order behavior of the fluid is captured by a simpler boundary layer equation which is supplied with data at infinity from Euler and at zero (the boundary), from Navier–Stokes. The complication is that the Euler solution is known only up to a constant multiple. The specific vorticity value \({\omega _0}\) is then fixed by demanding the corresponding Prandtl equation for the boundary layer admits a periodic solution.
We note that the (Prandtl–Batchelor or Feynman–Lagerstrom) theory was developed to address the specific phenomenon which only arises for stationary flows on closed domains. For the much different setting of unsteady flows, the question of which Euler flow is achieved in the inviscid limit is essentially completely determined by the initial data. For stationary flows on non-closed domains, for example on \([0, L] \times \mathbb {R}_+\), there is an analogue of data prescribed on the sides \(\{x = 0\}, \{x = L\}\) which similarly fixes the inviscid Euler flow (see for instance, results of [15, 21] for results in this direction). In contrast, for closed domains, the only prescribed data is the slip boundary data, \(f(\theta )\). Therefore, the selection mechanism is less obvious to uncover, and historically motivated investigations of Prandtl–Batchelor or Feynman–Lagerstrom, and from a rigorous standpoint, the results in this paper.
On the disk \(M= \mathbb {D}\) of radius \(\textsf{R}\), this amounts to:
This picture has been rigorously justified by Kim [24, 25] for the boundary layer and recently by Fei, Gao, Lin and Tao [13] for Navier–Stokes. The latter constructs a sequence of steady Navier–Stokes solutions on the disk forced by (7) converging towards this predicted end state. In the case of a general domain, Feynman–Lagerstrom argued that selecting \({\omega _0}\) to ensure a certain periodicity is a necessary condition for the existence of such a layer and therefore for convergence, but did not speak to its sufficiency. We now review their theory.
Recall Prandtl’s boundary layer equations written in von Mises coordinates \((s,\overline{\psi })\), where s is the periodic coordinate on the boundary and \(\bar{\psi }:=\psi /\sqrt{\nu }\) is the rescaled streamline coordinate. From hereon, we denote \(\bar{\psi }=\psi \) and
where \(\gamma : [0,|\partial M|)\rightarrow \partial M\) is the arc-length parametrization of the boundary, is the tangential slip along the boundary of unit vorticity Euler solutions (5). The Prandtl equations—which determine an unknown function \(q:[0,L)\times \mathbb {R}^+\rightarrow \mathbb {R}\) which serves as an approximation of the tangential Navier–Stokes velocity \(u^\nu \cdot \hat{\tau }\) in an \(O(\sqrt{\nu })\) boundary layer—are (see [34]):
which is to be satisfied on \((\psi ,s)\in [0,\infty )\times [0,L)\) where \(L=|\partial M|\) is the length of the boundary. For completeness, we derive these equations in Sect. 2. The solution q must connect Navier–Stokes to Euler: at the boundary (\(\psi =0\)), the solution q takes the Navier–Stokes data and away from the boundary (\(\psi =\infty \)), the solution q assumes the Eulerian behavior:
which, for Q, translates to the data
The solution q (or, equivalently, Q) of Eq. (10) must be periodic in the s variable (so that the boundary layer closes). See Fig. 2. Feynman–Lagerstrom noted that this leads to a self-consistency condition on \({\omega _0}\). We will enforce this in the following way. First, we rewrite (10) as
Integrating the above equation over the boundary, we obtain
where the nonlinearity is explicitly
Then, for some scalars A and B, we obtain the identity
From the boundary conditions, we have that \(A=B=0\). We thus obtain the nonlinear condition that at each \(\psi \in \mathbb {R}_+\), we have
Evaluating at \(\psi =0\), we find a nonlinear, nonlocal condition determining the constant \(\omega _0\):
Remark
(Feynman–Lagerstrom formulae) Letting \(1- \omega _0^2 =:{\varepsilon }\overline{\omega }_0\) with \(\overline{\omega }_0=O(1)\), we anticipate \(Q = O({\varepsilon })\). Since \(\frac{1}{\sqrt{ \omega _0^2+x}}-\frac{1}{\sqrt{ \omega _0^2}} = - \frac{x}{2\omega _0^3} + O (x^2)\), we see that \(\mathcal {N}^{\omega _0}[Q](\psi )=O(Q^2) = O({\varepsilon }^2)\). Moreover \(\mathcal {N}^{\omega _0}[Q](y)\) is trivial in the case of the boundary having constant curvature \(\kappa :=\hat{\tau }\cdot \nabla \hat{n} \cdot \hat{\tau }\). Indeed, in this case of M being a disk, it is readily seen that the integrand in (11) is a total derivative in s and hence \(\mathcal {N}^{\omega _0}[Q](\psi )\equiv 0\)), see the next Remark. Thus, as pointed out by Feynman and Lagerstrom [14], to leading order, condition (12) is
This formula is exact (having \(\partial _s \kappa =0\)) when M is the disk, and generally only for the disk.Footnote 1 Recalling \( f(s) = q_e (s) + {\varepsilon }g(s)\) so that \(f^2(s) - \omega _0^2 q_e ^2(s)= (1- \omega _0^2 ) q_e ^2(s) + 2{\varepsilon }g(s)q_e (s) + {\varepsilon }^2\,g^2(s)\), to leading order in \({\varepsilon }\) (the deviation of NS data from unit vorticity Euler slip) we have
Remark
(Wood’s formula when \(M=\mathbb {D}\)) On the disk of radius \(\textsf{R}\), the constant vorticity solution (5) is solid body rotation \(u_*(x)=\frac{1}{2} x^\perp \), so that \(q_e =\frac{\textsf{R}}{2}\) is a constant. In fact, by Serrin’s theorem [37] constraining a domain admitting a solution of \(\Delta \psi = 1\) with constant Neumann and Dirichlet data, the disk is the unique domain for which the solid body Euler solution has constant boundary slip velocity \(q_e \). See also [26]. On the disk, (13) without an error term is exact and, with the circumference \(L=2\pi \textsf{R}\), agrees with (9) of Wood [40].
In this paper, we rigorously establish this prediction by constructing a periodic boundary layer verifying the Feynman–Lagerstrom condition if the constant vorticity Euler solution \(u_*\) on M defined by (5) has no stagnation points on the boundary. We have
Theorem 3
(Existence of a periodic Prandtl boundary layer) Let M be a simply connected domain with \(L=|\partial M|\). Denote \(\mathbb {T}_L= [0,L)\). Let \(q_e:\mathbb {T}_L\rightarrow \mathbb {R}\) be a smooth, non-vanishing function. Let \(f(s):= q_e (s)+ {\varepsilon }g(s)\) with smooth \(g: \mathbb {T}_L \rightarrow \mathbb {R}\). For all \({\varepsilon }\) sufficiently small, depending only the data \((q_e,g)\), there exists a unique constant \({\omega _0}\in \mathbb {R}\) and function \(q:\mathbb {T}_L\times \mathbb {R}^+\rightarrow \mathbb {R}\) such that the pair \((\omega _0, q)\) solves the Prandtl equations on \(\mathbb {T}_L\times \mathbb {R}^+\):
Moreover, the solution Q lies in the space \(X_{2,50}\) defined by (25) and enjoys \(\Vert Q\Vert _{X_{2,50}} \lesssim {\varepsilon }\). The selected vorticity \(\omega _0\) can be expressed as follows: there exists a constant \(C>0\) so that
The sign of \(\omega _0\) agrees with that of the background \(q_e \) which, in this case, is positive.
This theorem, proved in Sect. 5, provides the first rigorous confirmation of the Feynman–Lagerstrom formula, and justifies their claim that for \(|f-q_e |\ll |f|\) (translating to \({\varepsilon }\ll 1\)), the leading term in (13) serves as a good approximation for the selected vorticity.
The constant \({\omega _0}\) satisfies (12), and has an explicit component, determined by \(q_e (s)\) and f(s), as well as an implicit component \(\overline{\omega }_{\textrm{Err}}\) which is smaller amplitude and for which we obtain bounds. We emphasize that the \(\omega _0\) appearing in (15) is nonlinearity selected as soon as the domain, M, is no longer a disk (for example, M is an ellipse). This requires, at an analytical level, a delicate coupling between the choice of constant, \(\omega _0\), and the control of the solution Q in an appropriately chosen norm, which is the main innovation of our work.Footnote 2
We anticipate the Prandtl system, which we analyze in this paper, to be stable in the inviscid limit for the full Navier–Stokes system. Indeed, this is what is proved in [13] when M is a disk and \(u_e = x^\perp \). However, as discussed above, in that very special setting the constant \({\omega _0}\) can be explicitly determined (9). In general, this is not the case and \({\omega _0}\) is only implicitly determined by the condition (12) described above, making the inviscid limit more delicate. Nevertheless, we believe that the nonlinearly determined constant \(\omega _0\) will describe, to leading order in viscosity, the selection principle. That is, we believe that Navier–Stokes vorticity \(\omega ^{\nu }\) should obey an asymptotic expansion
in the interior of the domain. In fact, we issue the following
Conjecture 1
Let \(M\subset \mathbb {R}^2\) be any simply connected domain such that the constant vorticity Euler solution \(u_*\) on M defined by (5) has a single eddy (streamfunction has a single, non-degenerate, critical point). Suppose that Navier–Stokes is forced by a slip of the form (7), e.g. \( f = u_*\cdot \tau + {\varepsilon }g\) for some smooth function \(g:\partial M \rightarrow \mathbb {R}\). Then, there exists an \({\varepsilon }_*:={\varepsilon }_*(M, g)\) such that for all \({\varepsilon }<{\varepsilon }_*\) we have weak convergence in \(L^2(M)\)
along a sequence of steady Navier–Stokes solutions, where \(\omega _0:=\omega _0(M,g)\) is (16) of Thm 3.
Of course, stronger convergence can be expected, along with a boundary layer description such as that established by [13] on the disk. The fact that moving boundaries can stabilize the inviscid limit is a well known phenomenon from the work of Guo and Nguyen [16] and Iyer [19, 20]. Verifying the expansion (17) to prove the above conjecture will require substantially new ideas. In the context of elliptical domains M, this is work in preparation.
Finally we remark that the failure of a boundary layer to exist is indicative of the existence of multiple eddies: constant vorticity regions are separated by internal layers which can be thought of as free boundaries. This can happen either if the constant vorticity solution on that domain has multiple eddies, or if the given slip data is far from that of a constant vorticity slip (according to our Theorem 3). Kim [26] showed that if the Navier–Stokes boundary slip is only slightly negative in places, the Prandtl–Batchelor theory still applies to good approximation in the bulk. For the situation of being far from compatible slip data, see Kim and Childress [28] for an analytical investigation on a rectangle, Greengard and Kropinski [17] for a numerical investigation on disk domains, and Henderson, Lopez and Stewart [18] for laboratory experiments.
2 Derivation of the Prandtl Boundary Layer Equation
In this section, we derive the Prandtl equations for any simply connected domain M. Assume that \(s:[0,L]\rightarrow \partial M\) be the arc-length parametrization of the boundary \(\partial M\). For \(s\in \mathbb T_L\), let \(\tau (s)\) and n(s) be unit the tangential vector to the boundary \(\partial M\). There exists \(\delta >0\) such that for any \(x\in M\) such that \(\textrm{dist}(x,\partial M)<\delta \), there exists a unique \(s\in \mathbb T_L\) and \(x(s)\in \partial M\) such that
Moreover, one has the representation
where \(x(s)\in \partial M\) and \(z=\textrm{dist}(x,\partial M)\), see [3]. The map
is a diffeomorphism. We also define the following quantities for the domain M:
where \(\gamma \) represents the boundary curvature, and J is a Jacobian for a near-wall mapping used to derived the following form of Navier–Stokes, see [3] and Appendix 6.
Now for \(x=x(s,z)\), we denote \(\tau (s)\) and n(s) to be the tangential and normal vector at \(x(s)\in \partial M\) on the boundary. Consider the steady Navier–Stokes equations
written in the region \(\textrm{dist}(x,\partial M)<\delta \). We define
By direct calculation, provided in Appendix 6, the Navier–Stokes equations become
Remark
On the disk with the usual polar coordinates \((\theta ,r)\in \mathbb T\times [0,1]\), we have
Near the boundary, in a layer of width \(\sqrt{\nu }\), we anticipate that Navier–Stokes velocity field \((u_\tau , u_n)\) will look like a small boundary layer correction \((u_\tau ^P, v_n^P)\), to a constant vorticity \(({\omega }_0q_e,0)\) Euler flow, as discussed in the introduction. That is,
where \(\lim _{Z\rightarrow \infty }u_\tau ^P(s,Z)=0\). See discussion in Oleinik and Samokhin [34]. Plugging in this ansatz into the Navier–Stokes equations near the boundary, and using the approximation
we obtain the equations
along with the divergence free condition
Taking \(Z\rightarrow \infty \) in the Eq. (18), we obtain
Replacing the pressure by the above into the equation (18), we obtain the Prandtl equations:
Define the von Mises variables \((s,\psi )\) such that
Let \(q=q(s,\psi )={\omega }_0 q_e (s)+u_\tau ^P\), the Prandtl equation becomes
which reduces to
Letting \(Q=q^2-{\omega }_0^2 q_e ^2\), the above equation becomes (10), namely (See Fig. 3 for a visualization)
3 Proof of Theorem 1: Absence of Turbulence
Let \(v^\nu = u^\nu - u_e\) be the difference of solutions of Euler and Navier–Stokes, where Navier–Stokes is forced by Euler’s slip velocity. On general domains M, it satisfies
Whence the error energy (which holds for Leray-Hopf solutions in dimension two) satisfies
In general, we may bound
since \(v^\nu |_{\partial \mathbb {D}}=0\) so we may apply the Poincaré inequality \(\lambda _1 \Vert v^\nu \Vert _{L^2} \le \Vert \nabla v^\nu \Vert _{L^2}^2\) where \(\lambda _1\) is the first positive eigenvalue of \(-\Delta _D\) on M. We remark, using the results of [38, Chapter 7] (which establish uniform bounds on the steady states), a similar energy identity can be used to prove global attraction of the unique steady state for Navier–Stokes forced by imposed slip on any domain, provided viscosity is large enough.
On the disk \(M=\mathbb {D}\), if \(u_e=u_\textsf{sb}= {\omega _0}x^\perp \) so that \(\nabla u_\textsf{sb}= {\omega _0}\begin{pmatrix} 0&{} -1\\ 1&{} 0 \end{pmatrix}\) and \(\Delta u_e =0\), we have
On the disk of radius \(\textsf{R}\), this is \(\lambda _1= (j_0/\textsf{R})^2\) where \(j_0\) is the first zero of \(J_0\) the Bessel function of the first kind and order zero). We thus have the stated result.
Remark
On the ellipse \(u_\textsf{sb}= {\omega _0}(-y, \alpha x)\) so that \(\nabla u_\textsf{sb}= {\omega _0}\begin{pmatrix} 0&{} -1\\ \alpha &{} 0\end{pmatrix}\) and \(v \cdot \nabla u_\textsf{sb}\cdot v= {\omega _0}(\alpha -1)v_1v_2\). It follows that provided
then the solid body rotation solution is the global attractor. In particular, as the eccentricity of the elliptical domain goes to zero, \(\alpha \rightarrow 1\) and the critical viscosity \(\nu _*\) goes to zero. Curiously, all flows in this elliptical family are isochronal [41], meaning that the period of revolution of a particle does not depend on the particular streamline. As such, the form examples of cut points in group of area preserving diffeomorphisms of those domains, see discussion in [9, 11]. The lack of differential rotation in the Euler solution may have important consequences for the asymptotic stability and realizability in the inviscid limit.
4 Proof of Theorem 2: Prandtl–Batchelor Theory
First, by [6, Lemma 5], under the stated assumptions we have that
for some \(C^1\) function \(F:\mathbb {R}\rightarrow \mathbb {R}\) and constant \(c_*\in \mathbb {R}\). Suppose without loss of generality that \(\{\psi = 0\}\) is the unique critical point in M, so that \(\textrm{rang} (\psi ) = [0,c_*]\). By the assumption (6), we have the convergence \(\psi ^\nu \rightarrow \psi \) in \(H^{7/2+}(U)\) and thus in \(C^1(U)\) for all interior open subsets \(U\subset M\). It follows that we have convergence of the streamlines (level sets of \(\psi ^\nu \)). Specifically, for any \(c\in \textrm{rang} (\psi )\), the set \(\{\psi ^\nu = c\}\) is a closed streamline (at least for sufficiently small \(\nu :=\nu (c)\)) converging to \(\{\psi = c\}\). In what follows, for fixed c we assume \(\nu \) is sufficiently small for the above to hold.
Integrating the Navier–Stokes vorticity balance in the sublevel set \(\{\psi ^\nu \le c\}\)
where \(\hat{n}^\nu = \nabla \psi ^\nu /|\nabla \psi ^\nu |\) is the unit normal to streamlines \(\{\psi ^\nu = c\}\). Thus
where \(A \Delta B\) denotes the symmetric difference between two sets. Under our assumptions, there exists an open set \(O\subset M\) containing the streamline \(\{\psi ^\nu = c\}\) uniformly in \(\nu \). By the trace theorem,
Combined with the fact that \(\omega = F(\psi )\), we find
Thus, for any \(\delta >0\), we have the bound
Consequently, using (6) and taking the limit of the upper bound, we have
By our hypotheses that \(\psi \) has a single stagnation point \(\{\psi =0\}\) in M, the circulation \( \oint _{\{\psi = c\}} u \cdot \textrm{d}\ell \ne 0\) for all \(c\ne 0\). Thus, since \(F'\) is continuous, we must have that \(F'(c)=0\) for all \(c\in \textrm{rang} (\psi )\) so that \(F= {\omega _0}\) for some \({\omega _0}\in \mathbb {R}\).
5 Proof of Theorem 3
5.1 Iteration and Bootstraps
Here we produce a unique solution \((Q,{\omega _0})\) of
on \((s,\psi )\in \mathbb {T} \times \mathbb {R}^+\), for arbitrary \(g:\mathbb {T}\rightarrow \mathbb {R}\) and sufficiently small \({\varepsilon }:= {\varepsilon }(g;q_e )\). Here, \({\omega _0}\) is to be determined together with Q, and we introduced \(\overline{{\omega _0}}\in \mathbb {R}\) (anticipated to be an O(1) quantity as it depends on \({\varepsilon }\)) defined by
To prove this result, it is convenient to rewrite (19) as
We will study of following iteration scheme
with \(Q_{-1} = 0\), \({\omega _0}_{-1}= 1\). Schematically, we think that \(({\omega _0}_{n-1}, Q_{n-1}) \mapsto {\omega _0}_n \mapsto Q_n\), that is \({\omega _0}_n\) is determined on the onset by a compatibility condition for the linear problem which depends on the prior iterate, and \(Q_n\) is subsequently solved for \({\omega _0}_{n-1}\). Let
In this system \({\omega _0}_n\) is chosen to enforce that
Conceptually, it is clearer to separate out the explicit component of \(\overline{{\omega _0}}_n\), which is O(1) and independent of n, and the smaller amplitude implicit component of \(\overline{{\omega _0}}_n\) as follows
With this, we can solve the above equation for \(Q_n\). For \(\psi \ge 0\), we define
For a function \(f=f(s,\psi )\) defined on \(\mathbb T_L\times \mathbb R_+\), we define
We will construct the unique solution of the Eq. (20) in the space \(X_{2,50}\). By the standard Sobolev embedding, we also have
Remark
(Exponential decay of Q for \(\psi \gg 1\)) In fact, one can prove existence in a space encoding exponential decay in \(\psi \), as should be expect for a (nonlinear) heat equation with data at \(\psi =0\). For simplicity of presentation, we prove only algebraic decay but the requisite modifications involving exponential weights are standard.
Indeed, we have the following result that tells us this iteration is well-defined.
Lemma 4
Let \(n \ge 0\). Assume that \(Q_{n-1} \in X_{2,50}\), and that (27)–(28) are valid until index \(n -1\). Assume further that \(\overline{{\omega _0}}_{n}\) is defined according to (23). Then, there exists a unique solution, \(Q_n\) to the system (21).
Proof
We first of all write the system (21) as follows
We introduce the variable
We notice that \(\frac{\textrm{d}t}{\textrm{d}s}\) is bounded above and below and hence determines an invertible transformation due to the fact that \({\omega _0}_{n-1} q_e (s) > 0\). We also note that by writing \({\omega _0}_{n-1} q_e (s) = \langle {\omega _0}_{n-1} q_e \rangle + ({\omega _0}_{n-1} q_e - \langle {\omega _0}_{n-1} q_e \rangle )\), we have
which maps \(\mathbb {T}_L\) into \(\mathbb {T}_{{\omega _0}_{n-1} \langle q_e \rangle L}\). We next introduce
This object satisfies the system
We expand the solution of \(V_n\) in a Fourier basis in the t variable as follows
The zero mode equation is exactly the Feynman-Lagerstrom formula, (23). For the k’th mode, where \(k \ne 0\), we write the explicit formula:
where \(\sqrt{ik}\) is the complex square root of ik with positive real part. We now observe that for \(G_{n-1} \in X_{2,50}\), the above integrals converge to zero as \(\psi \rightarrow \infty \) (when \(k \ne 0\)). This completes the proof.
We define the differences \(\triangle \overline{{\omega _0}}_n\) and \(\triangle Q_n\) to be
Differences in Q obey:
Remark
(Obtaining the sharper bounds stated in Theorem 3) In what follows, we will bootstrap bounds of \({\varepsilon }^{1-}\), specifically \({\varepsilon }^{0.97}\) for \(|\overline{{\omega _0}}_{\textrm{Err}, n}|\) and \({\varepsilon }^{0.99} \) for \( \Vert Q_{n} \Vert _{X_{2, 50}}\), although any power less than 1 would suffice by the same argument given below. This is not essential, it is to avoid keeping track of large constants for simplicity of the bootstrap argument. In fact, from these bounds one can deduce a posteriori sharper estimates of the form \( |\overline{{\omega _0}}_{\textrm{Err}, n}| \le C_1 {\varepsilon }\) and \( \Vert Q_{n} \Vert _{X_{2, 50}} \le C_2 {\varepsilon }\) for some, possibly large, constants \(C_1,C_2>0\) by taking the proved bounds on \({\omega _0}\) and Q, returning to the equation, and performing the estimate again.
We will establish the following bounds
which immediately imply the main result. The bounds (27)–(28) show that \((\overline{{\omega _0}}_{\textrm{Err},n}, Q_n) \in B_{{\varepsilon }^{1.99}, {\varepsilon }^{0.99}} \subset \mathbb {R} \times X_{2, 50}\), whereas the bounds (29)–(30) show that iteration converges to a unique fixed point. A standard fixed point result imply that these bootstrap bounds give the main theorem:
Proof
We insert the bound (29) into the second term on the right-hand side of (30) in order to get the following
Define \(\textbf{Y}_n:= (\Delta Q_n, {\varepsilon }\Delta \overline{{\omega _0}}_n) \in X_{2,50} \times \mathbb {R}\), endowed with the product norm. Then
It is therefore clear that
This then implies that \((Q_n, \overline{{\omega _0}}_{n})\) is a Cauchy sequence in \(X_{2,50} \times \mathbb {R}\), and hence converges to a limit, \((Q_{\infty }, \overline{{\omega _0}}_{\infty })\). We can therefore pass to the limit in Eq. (21) as well as in (23) to conclude that \((Q_\infty , \overline{{\omega _0}}_{\infty })\) satisfy the system (19).
We now prove uniqueness. We assume that \((Q_{1}, {\omega _0}_{,1})\) and \((Q_2, {\omega _0}_{,2})\) are two solutions to (19) in the space \(X_{2,50} \times \mathbb {R}\). We may therefore write an analogous Eq. (26) on \(Q_1 - Q_2\) (without the iteration), which reads:
as well as the analogue of expression (26) (again without the iteration)
Re-applying the a-priori estimates on these systems results in the following bounds:
which are the analogues of (29)–(30). The two bounds above clearly imply that \( \overline{{\omega _0}}_{,1} = \overline{{\omega _0}}_{,2}\) and \(Q_1 = Q_2\). This proves uniqueness.
5.2 \(\overline{{\omega _0}}_{\textrm{Err}, n}\) Estimates
Here, we will establish the bootstrap bound (27). Indeed,
Lemma 5
Assume (27)–(30) are valid until the index \(n - 1\). Then \(\overline{{\omega _0}}_{\textrm{Err},n}\) satisfies:
Proof
Recall the expression (24), after which we estimate as follows
where we have invoked the bootstrap bound (28) in the final step, as well as the \(L^\infty \) estimate
Above, we have used the following Sobolev inequality on \(\mathbb {T} \times \mathbb {R}_+\), which reads \(\Vert f \Vert _{L^\infty } \lesssim \Vert f \Vert _{X_{1,0}}\). This Sobolev embedding will be used repeatedly to estimate nonlinear terms.
5.3 \(\Delta \overline{{\omega _0}}_{n}\) Estimates
Here we prove the following lemma.
Lemma 6
Assume (27)–(30) are valid until the index \(n - 1\). Assume (27) and (28) are valid until index n. Then the following bound holds
Proof
We use the expression
To estimate \(I_1\), we have
To estimate \(I_2\), we need to use the identity (22) to estimate
Clearly, using the inequality \(|1 - \sqrt{1-x}| \le |x|\) for \(x\le 1\), we have
and, using the inequality \(|\sqrt{1-x} - \sqrt{1-y}| \lesssim |x - y|\) for \(x, y \ll 1\), we have
Therefore, we have
Pairing these bounds together, we get the desired result.
5.4 Abstract Q Estimates
For future use, it turns out we will have a need to develop our estimates on a slightly more abstract system. Therefore, we consider
We develop a high-order energy method to treat equation (31). We commute \(\partial _s^k\) to obtain
where the commutator term
and where we adopt the short-hand \(f^{(k)}(s, \psi ):= \partial _s^k f(s, \psi )\) for an abstract function \(f(s, \psi )\).
Proposition 7
Assume that the boundary condition b(s) and the source term \(F + \partial _{\psi }^2\,G\) satisfy the Feynman–Lagerstrom compatibilty condition (23). Then the solution Q to (31) obeys the following inequality:
The first task is we lift the boundary condition b(s) by considering the lift function
and consequently
which satisfies the following system
where
We will need to work in higher order norms. Therefore, we present the equations upon commuting \(\partial _s^k\) to (34), which yield
Above, we define the commutator term as follows:
Lemma 8
For any \(\delta >0\) the following bounds hold (where the constant \(C_\delta \uparrow \infty \) as \(\delta \downarrow 0\)):
Proof
We multiply (36) by \(\mathring{Q}^{(k)}\langle \psi \rangle ^{2m}\) and integrate by parts to get the identity
We now integrate in \(s \in \mathbb {T}\), and the \(\partial _s\) term drops out due to periodicity. This implies
This implies
where \(\delta >0\) is small and \(C_\delta \sim \delta ^{-1}\). The result follows immediately, using the fact that
This concludes the proof of the lemma.
Lemma 9
Let \(k \ge 0, m \ge 0\). The solution \(Q^{(k)}\) to (32) satisfies the following estimate:
Proof
We multiply (36) by \(\frac{1}{q_e (s)} \partial _s \mathring{Q}^{(k)} \langle \psi \rangle ^{2\,m}\) and integrate by parts to produce
Above, we have used the homogeneous boundary condition for \(\mathring{Q}\) to integrate by parts. We now integrate over \(s \in \mathbb {T}_L\) and use periodicity to eliminate the second term on the right-hand side above, which results in
Recalling (37), (35) and absorbing the last term on the right-hand side to the left, we get
We conclude the proof of the lemma, upon using the fact that \( \mathring{Q}^{(k)}=Q^{(k)}-e^{-\psi }b^{(k)}(s).\)
We now need to estimate the zero mode of \(Q^{(k)}\). Clearly, this is nontrivial only for \(k = 0\) (when \(k\ge 1\) there is no zero mode).
Lemma 10
The zero mode, \(Q^{(=0)}\), to the solution of (31), satisfies the following bound:
Proof
We integrate Eq. (31) to generate the identity for each \(\psi \in \mathbb {R}_+\):
after which we integrate twice from \(\infty \) to get
We now separate out the left-hand side
This implies
We therefore obtain
Clearly, \(\mathcal {I}_3\) is majorized by the last term on the right-hand side of (40). We will estimate the first term above, which we call \(\mathcal {I}_1\). Using Hölder’s inequality, we get
To estimate \(\mathcal {I}_2\), we need to pay weights as follows using Cauchy-Schwartz:
which therefore implies that \(|\mathcal {I}_2| \lesssim \Vert F \langle \psi \rangle ^{m + 4} \Vert _{L^2}\). This completes the proof.
Lemma 11
Let \(k \ge 0, m \ge 0\). The solution \(Q^{(k)}\) to (32) satisfies the following estimate:
Proof
We simply rearrange Eq. (32) and apply \(L^2\) norm to both sides.
Proof of Proposition 7
Consolidating the bounds (38), (39), (40), (41), we proved (33).
5.5 \(Q_n\) Estimates
Lemma 12
Assume (27) is valid up to index n and (28) is valid up to index \(n-1\). Then
Proof
For this bound, motivated by Eq. (21), we set
According to (33), we fix \(k = 2, m = 50\), which results in
We therefore estimate the two quantities appearing on the right-hand side above. To make notation simpler, we define
then
By a direct calculation, we have the following identities
First we estimate \(\Vert F\langle \psi \rangle ^{54}\Vert _{L^2}\). We have
We now show that \(\left\Vert\partial _s F\langle \psi \rangle ^{54}\right\Vert_{L^2}\lesssim \epsilon ^{0.99}\epsilon ^{0.99}\). We have
We first bound \(A_1\). We have
We now estimate \(A_2\). We have
We now show that
By a direct calculation, we get
We first establish the following bounds on the auxiliary quantities \(U_{n-1}\). We have
Similarly, we have
We can now estimate of \(\left\Vert\partial _s^2 F\langle \psi \rangle ^{54}\right\Vert_{L^2}\). We first bound \(B_1\). We have
We next move to \(B_2\), for which we have
As for \(B_3\), we have
We finally conclude with an estimate on \(B_4\), for which we have
To conclude the proof of lemma, we need to estimate the \(H^3_s\) norm of b,
Therefore, according to (4260), the lemma is proven.
5.6 \(\Delta Q_n\) Estimates
Our main objective in this section is to close the final bootstrap bound, (30). We begin with a lemma which allows us to control our auxiliary quantity, \(U_{n-1}\), introduced in (43).
Lemma 13
Let \(0 \le m \le 50\). Assume (27) - (30) are valid until the index \(n - 1\). Assume (27) - (29) are valid until the index n. The quantities \(U_{n-1}, U_{n-2}\) satisfy:
Proof
Recalling (43), we have
where the coefficients are defined by
According to our bootstraps, we claim the following bounds. There exists a decomposition of \(\partial _s^2 \alpha = \alpha _A + \alpha _B\) such that
We will prove these bounds as follows. First, we define
after which the following identities are valid:
We will henceforth prove the following bounds. We claim there exists a decomposition of \(\partial _s^2 \alpha ^{(1)} = \alpha ^{(1)}_A + \alpha ^{(1)}_B\), where
upon which using (52), we obtain (47), (48), (49), (50), and (51).
Proof of (53)
Clearly, we have
Next, we have the identity \(\partial _s \alpha ^{(1)} = \frac{\partial _s D_{n-1}}{D_{n-1}^2}\). Since we have already established a lower bound on \(D_{n-1}\), it suffices to estimate \(\partial _s D_{n-1}\):
where we have invoked the bootstraps (27) and (28). This proves the bound (53).
Proof of (54)
For this bound, we differentiate once more to find the identity
We estimate
and
This proves the bound (54).
Proof of (55)
We turn now to the definition of \(\alpha ^{(2)}\). We will use freely the bounds \(|D_{n-1}| + |D_{n-2}| > rsim 1\) and \(\Vert \partial _s D_{n-1} \Vert _{L^\infty } + \Vert \partial _s D_{n-2} \Vert _{L^\infty } \lesssim 1\), which have already been established. First, we have
Next, we have the identity
from which we obtain
This proves the bound (55).
Proof of (56)
We differentiate (58) again to obtain the identity
after which we obtain the bound
This proves the bound (56).
Proof of (57)
We have
and upon using (58), we have
We have therefore established (53)–(57) and hence (47)–(51). From here, the desired estimates, (44)–(45), follow from an application of the product rule applied to the identity (46). Indeed, we have:
where we have used the bounds (47) and (49). In \(L^2\), we similarly have
where we have used the bounds (47) and (51).
Next, we have upon differentiating (46), the identity
after which we have the following \(L^\infty \) bound:
where we have invoked (47) and (49). In \(L^2\), we similarly have
where we have used the bounds (47) and (51).
Differentiating (46) twice in s, we obtain the identity
where we use the decomposition \(\partial _s^2 \alpha = \alpha _A + \alpha _B\). We now estimate the \(L^2\) norm as follows:
where we have used the bounds (47)–(50). We have therefore established the bounds (44)–(45), and this concludes the proof of the lemma.
Lemma 14
Assume (27)–(30) are valid until the index \(n - 1\). Assume (27)–(29) are valid until the index n. Then
Proof
For this estimate, motivated by (26), we set
According to (33), we fix \(k = 2, m = 50\), which results in
We therefore estimate the two quantities appearing on the right-hand side above. We first address the term F, which we rewrite as follows
An identical calculation to the estimate of the forcing, F, in Lemma 12 results in the bound
We develop the following identities
We estimate \(\partial _s F_2\) as follows. First,
Next, to estimate \(C_2\), we have
Finally, to estimate \(C_3\), we have
We now move to the second derivative, \(\partial _s^2 F_2\), which we will treat as follows:
We have
First, we estimate
Next, to estimate \(C_{1,2}\), we have
Next, to estimate \(C_{1,3}\), we have
We next move to the \(\partial _s C_2\) contributions, for which we record the identity
We first estimate \(C_{2,1}\) for which we have
Next, we have
Finally, we have the \(C_{2,3}\) contribution for which we estimate
We next compute \(\partial _s C_3\), which results in the following identity,
We estimate first \(C_{3,1}\) as follows
Next, we estimate \(C_{3,2}\) as follows
Next, we estimate \(C_{3,3}\) as follows
Finally, we estimate \(C_{3,4}\) as follows
Now, upon invoking (44)–(45), the above estimates give
Next, we clearly have
Finally, we have the boundary condition
Consolidating all the above bounds with estimate (59) concludes the proof of the lemma.
Data Availability Statement
No data was generated for the purposes of this research.
Notes
Among domains with smooth boundary. For Lipschitz domains, it holds also for regular polygons [39].
In this respect, the selection mechanism is similar to another arising in fluid dynamics: inviscid damping [2]. There, perturbations to certain stable shear flows return to equilibrium in a weak sense, but the which equilibrium they converge to must determined together with the entire time history of the solution.
References
Batchelor, G.K.: On steady laminar flow with closed streamlines at large Reynolds number. J. Fluid Mech. 1, 177–190, 1956
Bedrossian, J., Masmoudi, N.: Inviscid damping and the asymptotic stability of planar shear flows in the 2D Euler equations. Publications mathématiques de l’IHÉS 122(1), 195–300, 2015
Bardos, C., Nguyen, T.T., Nguyen, T.T., Titi, E.S.: The inviscid limit for the \(2 d \) Navier–Stokes equations in bounded domains. arXiv preprint arXiv:2111.14782 (2021)
Childress, S.: An Introduction to Theoretical Fluid Mechanics, vol. 19. American Mathematical Society, Providence (2009)
Childress, S.: Topological fluid dynamics for fluid dynamicists. Lecture Notes (2004)
Constantin, P., Drivas, T.D., Ginsberg, D.: Flexibility and rigidity of free boundary MHD equilibria. Nonlinearity 35(5), 2363, 2022
Constantin, P., Lopes Filho, M.C., Nussenzveig Lopes, H.J., Vicol, V.: Vorticity measures and the inviscid limit. Arch. Ration. Mech. Anal. 234, 575–593, 2019
Constantin, P., Vicol, V.: Remarks on high Reynolds numbers hydrodynamics and the inviscid limit. J. Nonlinear Sci. 28, 711–724, 2018
Drivas, T.D., Elgindi, T.M.: Singularity formation in the incompressible Euler equation in finite and infinite time. arXiv preprint arXiv:2203.17221 (2022).
Drivas, T.D., Ginsberg, D., Grayer II, H.: On the distribution of heat in fibered magnetic fields. arXiv preprint arXiv:2210.09968 (2022)
Drivas, T.D., Misiołek, G., Shi, B., Yoneda, T.: Conjugate and cut points in ideal fluid motion. Annales mathématiques du Québec 1–19 (2022)
Drivas, T.D., Nguyen, H.Q.: Remarks on the emergence of weak Euler solutions in the vanishing viscosity limit. J. Nonlinear Sci. 29, 709–721, 2019
Fei, M., Gao, C., Lin, Z., Tao, T.: Prandtl–Batchelor flows on a disk. Commun. Math. Phys. 397, 1103–1161, 2023
Feynman, R.P., Lagerstrom, P.A.: Remarks on high Reynolds number flows in finite domains. In: Proceedings of IX International Congress on Applied Mechanics, vol. 3, pp. 342–343
Guo, Y., Iyer, S.: Validity of steady Prandtl layer expansions. Commun. Pure Appl. Math. 76(11), 3150–3232, 2023
Guo, Y., Nguyen, T.T.: Prandtl boundary layer expansions of steady Navier–Stokes flows over a moving plate. Ann. PDE 3, 1–58, 2017
Greengard, L., Kropinski, M.C.: An integral equation approach to the incompressible Navier–Stokes equations in two dimensions. SIAM J. Sci. Comput. 20(1), 318–336, 1998
Henderson, D.M., Lopez, J.M., Stewart, D.L.: Vortex evolution in non-axisymmetric impulsive spin-up from rest. J. Fluid Mech. 324, 109–134, 1996
Iyer, S.: Steady Prandtl boundary layer expansions over a rotating disk. Arch. Ration. Mech. Anal. 224, 421–469, 2017
Iyer, S.: Steady Prandtl layers over a moving boundary: nonshear Euler flows. SIAM J. Math. Anal. 51(3), 1657–1695, 2019
Iyer, S., Masmoudi, N.: Boundary layer expansions for the stationary Navier–Stokes equations. arXiv preprint arXiv:2103.09170 (2021).
Kim, S.-C.: Asymptotic study of Navier–Stokes flows. Trends Math. Inf. Cent. Math. Sci. 6, 29–33, 2003
Kim, S.-C.: On Prandtl–Batchelor theory of steady flow at large Reynolds number. Ph.D. thesis, New York University (1996)
Kim, S.-C.: On Prandtl–Batchelor theory of a cylindrical eddy: asymptotic study. SIAM J. Appl. Math. 58, 1394–1413, 1998
Kim, S.-C.: On Prandtl–Batchelor theory of a cylindrical eddy: existence and uniqueness. Zeitschrift für angewandte Mathematik und Physik ZAMP 51(5), 674–686, 2000
Kim, S.-C.: A free-boundary problem for Euler flows with constant vorticity. Appl. Math. Lett. 12(4), 101–104, 1999
Kim, S.-C.: Batchelor–Wood formula for negative wall velocity. Phys. Fluids 11, 1685–1687, 1999
Kim, S.-C., Childress, S.: Vorticity selection with multiple eddies in two-dimensional steady flow at high Reynolds number. SIAM J. Appl. Math. 61(5), 1605–1617, 2001
Lagerstrom, P.A.: Solutions of the Navier–Stokes equation at large Reynolds number. SIAM J. Appl. Math. 28(1), 202–214, 1975
Lagerstrom, P.A., Casten, R.G.: Basic concepts underlying singular perturbation techniques. SIAM Rev. 14(1), 63–120, 1972
Marchioro, C.: An example of absence of turbulence for any Reynolds number. Commun. Math. Phys. 105(1), 99–106, 1986
Marchioro, C.: An example of absence of turbulence for any Reynolds number: II. Commun. Math. Phys. 108(4), 647–651, 1987
Novikov, A., Papanicolaou, G., Ryzhik, L.: Boundary layers for cellular flows at high Péclet numbers. Commun. Pure Appl. Math. 58(7), 867–922, 2005
Oleinik, O.A., Samokhin, V.N.: Mathematical Models in Boundary Layer Theory, vol. 15. CRC Press, Boca Raton (1999)
Prandtl, L.: Uber flussigkeitsbewegung bei sehr kleiner reibung. In: International Mathematical Congress, Heidelberg, pp. 484–491 (1904); see Gesammelte Abhandlungen II, pp. 575–584 (1961)
Rhines, P.B., Young, W.R.: How rapidly is a passive scalar mixed within closed streamlines? J. Fluid Mech. 133, 133–145, 1983
Serrin, J.: A symmetry problem in potential theory. Arch. Ration. Mech. Anal. 43, 304–318, 1971
Tsai, T.-P.: Lectures on Navier–Stokes Equations, vol. 192. American Mathematical Society, Providence (2018)
van Wijngaarden, L.: Prandtl–Batchelor flows revisited. Fluid Dyn. Res. 39(1–3), 267, 2007
Wood, W.W.: Boundary layers whose streamlines are closed. J. Fluid Mech. 2, 77–87, 1957
Yudovich, V.I.: On the loss of smoothness of the solutions of the Euler equations and the inherent instability of flows of an ideal fluid. Chaos Interdiscip. J. Nonlinear Sci. 10(3), 705–719, 2000
Acknowledgements
The authors are grateful to the referees for comments improving the presentation of the article. The research of TDD was partially supported by the NSF DMS-2106233 Grant and NSF CAREER Award #2235395, as well as an Alfred P. Sloan Fellowship. The research of SI was partially supported by NSF DMS-2306528 and a UC Davis Hellman Foundation Fellowship. The research of TN is partially supported by the AMS-Simons Travel Grant Award. SI gratefully acknowledges support by NSF award DMS2306528, and a UC Davis Society of Hellman Fellowship award.
Author information
Authors and Affiliations
Corresponding authors
Ethics declarations
Conflict of interest
The authors declare the are no conflict of interest.
Additional information
Communicated by V. Vicol.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendix A. Derivation of Near-Boundary Navier–Stokes Equations
Appendix A. Derivation of Near-Boundary Navier–Stokes Equations
In this section, we give the detailed calculations for the Navier–Stokes equations claimed in Sect. 2. We recall the standard identities, which will be used in the next lemmas:
We recall that the map
is a diffeomorphism. In this transformation, we have
For a vector field \(u:M\rightarrow \mathbb R^2\), we also have
Lemma 15
For any vector field \(u:M\rightarrow \mathbb R^2\) and scalar function \(f:M\rightarrow \mathbb R\) supported near the boundary \(\partial M\) there holds
In particular, by choosing \(u=\tau \) and \(u=n\) respectively, there hold
Proof
This follows by direct calculation. We have
Lemma 16
The following identities holds for any given vector field \(u:M\rightarrow \mathbb R^2\):
Proof
We check the first, the third and the fifth identities only, and the proofs for other identities are similar. We have
We note that
Combining the above with the previous calculation, we obtain
Now we show the third identity. We have
For incompressibility, we find
The proof is complete.
Rights and permissions
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
About this article
Cite this article
Drivas, T.D., Iyer, S. & Nguyen, T.T. The Feynman–Lagerstrom Criterion for Boundary Layers. Arch Rational Mech Anal 248, 55 (2024). https://doi.org/10.1007/s00205-024-01991-z
Received:
Accepted:
Published:
DOI: https://doi.org/10.1007/s00205-024-01991-z