= Solution
Let $p_1(z)$ denote the <Eulerian perturbation of a fluid variable> that is the <pressure> amplitude, and keep $\rho$ constant. The <linearized Euler equations> and <incompressible flow> constraint are
$$
ik(U-c)u+U'w=-\frac{ikp_1}{\rho},\qquad
ik(U-c)w=-\frac{p_1'}\rho,\qquad
iku+w'=0.
$$
The last equation gives $u=iw'/k$. Substituting it into the first yields
$$
\boxed{p_1=\frac{i\rho}{k}\left[(c-U)w'+U'w\right].}
$$
Differentiating this formula and using the vertical momentum equation produces the <Rayleigh equation for inviscid shear flow>:
$$
\boxed{(U-c)(w''-k^2w)-U''w=0.}
$$
For a growing <normal mode>, $c$ is nonreal, so $U-c$ never vanishes on the real $z$ axis. In each open region the base <velocity> is constant or linear and $U''=0$; therefore $w''-k^2w=0$. Decay away from the layer selects $e^{-kz}$ above and $e^{kz}$ below, while both exponentials are retained inside. Conveniently normalize these as
$$
w(z)=\begin{cases}
A e^{-k(z-d)},&z\ge d,\\
B e^{-k(z-d)}+C e^{k(z+d)},&-d<z<d,\\
D e^{k(z+d)},&z\le-d.
\end{cases}
$$
These are the spatially localized discrete modes relevant to instability, rather than all possible generalized neutral disturbances at a <critical layer in a shear flow>.
At each corner $U$ is continuous and $U'$ jumps. The two sides have the same normal <velocity> and <normal stress>: there is no membrane, density jump or singular external force supplying a <pressure> jump. Thus $w$ and $p_1$ must be continuous. The <pressure> formula proves the required continuity of \b[$(c-U)w'+U'w$]. This is <pressure matching at a piecewise-linear shear interface>; it does not require $w'$ to be continuous. Equivalently integrating the <Rayleigh equation for inviscid shear flow> through a corner gives $(U-c)[w']=[U']w$.
Write $\alpha=kd$, $y=c/U_0$, and $E=e^{2\alpha}$. Continuity of normal <velocity> gives
$$
A=B+EC,\qquad D=EB+C.
$$
At $z=d$, the outside <derivatives> are $w'_+=-kA$, $U'_+=0$, while the inside <derivatives> are $w'_-=-kB+kEC$, $U'_-=U_0/d$. Hence <pressure> matching reduces to
$$
B+E(1-2\alpha+2\alpha y)C=0.
$$
Similarly at $z=-d$, the outside <derivative> is $kD$ and the inside <derivative> is $-kEB+kC$. Matching gives
$$
E(1-2\alpha-2\alpha y)B+C=0.
$$
The two equations can be written as
$$
\begin{pmatrix}
1-2\alpha+2\alpha y&e^{-2\alpha}\\
e^{-2\alpha}&1-2\alpha-2\alpha y
\end{pmatrix}
\begin{pmatrix}C\\B\end{pmatrix}=0.
$$
A nonzero mode therefore requires
$$
(1-2\alpha)^2-4\alpha^2y^2-e^{-4\alpha}=0,
\qquad
\boxed{\frac{c^2}{U_0^2}=\frac{(2\alpha-1)^2-e^{-4\alpha}}{4\alpha^2}.}
$$
This is the <dispersion relation> for the <unbounded piecewise-linear shear layer>.
With the phase convention $e^{ik(x-ct)}$, positive $\operatorname{Im}c$ gives exponential temporal growth. Since the displayed $c^2$ is real, a growing/decaying pair occurs precisely when
$$
|2\alpha-1|<e^{-2\alpha}.
$$
For $0<\alpha\le1/2$, $e^{-2\alpha}>1-2\alpha$ by the strict exponential tangent inequality, so every such mode is unstable. For $\alpha\ge1/2$, define $f(\alpha)=2\alpha-1-e^{-2\alpha}$. Its <derivative> $2+2e^{-2\alpha}$ is positive, while $f(1/2)<0$ and $f(1)>0$. There is consequently exactly one root in that interval, and
$$
\boxed{0<kd<\alpha_s\text{ is unstable},\qquad
2\alpha_s-1=e^{-2\alpha_s},\qquad
\tfrac12<\alpha_s<1,quad\alpha_s\simeq0.63923227.}
$$
At the cutoff the discrete roots merge at $c=0$; above it the roots are real and have no exponential growth. The physical wavelengths in the unstable band satisfy $\lambda_x=2\pi/k>2\pi d/\alpha_s\simeq9.83d$.
This is a finite-thickness <Kelvin-Helmholtz instability>: long disturbances roll up the layer and draw <kinetic energy> from its shear. The off-diagonal $e^{-2kd}$ terms describe coupling of the two <vorticity> discontinuities. Short disturbances decay across the layer before strongly coupling its edges and are neutral. Indeed at $kd\to\infty$ the two phase speeds approach the <velocities> $\pm U_0$ of the separated edges. At $kd\to0$, expansion of the numerator gives $c^2/U_0^2\to-1$, recovering the <vortex sheet> result $c=\pm iU_0$; the growth rate then behaves as $k|U_0|$.
Back to article page