Solution (source code)

= Solution

The equilibrium <isothermal equation of state> is $p=c_s^2\rho$. <Hydrostatic equilibrium> gives $p_z=-\rho\Omega^2z$, hence
$$
\frac{d\ln\rho}{dz}=-\frac{\Omega^2z}{c_s^2},\qquad
\boxed{\rho(z)=\rho_0e^{-z^2/(2H^2)},\quad H=\frac{c_s}{|\Omega|}.}
$$
Here $\Omega\ne0$ is assumed. This is the same local harmonic-gravity balance as <isothermal vertical structure in stellar gravity>.

For $\gamma>1$, this equilibrium is stably stratified under <adiabatic fluid perturbations>. Its squared <buoyancy frequency> is
$$
N^2=g\left(\frac1\gamma\frac{p_z}{p}-\frac{\rho_z}{\rho}\right)
=\frac{\gamma-1}{\gamma}\frac{g^2}{c_s^2}\ge0.
$$
A general perturbation can therefore support <acoustic waves> restored by compression and <internal gravity waves> restored by <buoyancy>, with nonoscillatory <entropy> or vortical disturbances also possible. The prescribed <velocity> is purely vertical and horizontally uniform. It describes vertical <acoustic waves>; it does not contain the horizontally varying circulation required for an oscillatory internal-gravity branch. For zero horizontal <wavenumber>, that branch has zero <frequency>. An isothermal equilibrium must not be confused with an isothermal perturbation: the latter would remove the buoyancy restoring force.

Introduce the <fluid Lagrangian displacement> $\xi(z)e^{i\sigma t}$, so $w=i\sigma\xi$. Use $\delta\rho,\delta p$ for <fluid Eulerian perturbations> and $\Delta\rho,\Delta p$ for <fluid Lagrangian perturbations>. The <continuity equation> and the adiabatic relation are
$$
\delta\rho=-(\rho\xi)_z,\qquad
\Delta\rho=-\rho\xi_z,\qquad
\Delta p=\gamma p\frac{\Delta\rho}{\rho},\qquad
\delta p=-\gamma p\xi_z-\xi p_z.
$$
The fixed external <gravitational potential> is not perturbed. The vertical <linearized Euler equations> consequently give
$$
-\sigma^2\rho\xi=-(\delta p)_z-g\,\delta\rho.
$$
Substituting the preceding relations, and using $p_z=-\rho g$ and $p_{zz}=-\rho_zg-\rho g_z$, reduces its right-hand side to
$$
\gamma p\xi_{zz}-\gamma\rho g\xi_z-\rho g_z\xi.
$$
Since $g=\Omega^2z$ and $g_z=\Omega^2$, the same equation holds for $w$:
$$
\boxed{w_{zz}-\frac z{H^2}w_z+
\frac1\gamma\left(\frac{\sigma^2}{c_s^2}-\frac1{H^2}\right)w=0.}
$$
The rigid fixed base supplies the essential condition \b[$w(0)=0$]. At infinity the condition is finite <kinetic energy>, not bounded <velocity> amplitude; a <polynomial> can grow while the Gaussian <mass density> makes its energy finite.

Put $x=z/H$ and $\lambda=(\sigma^2/\Omega^2-1)/\gamma$. The equation becomes the <Hermite differential equation> in its probabilists' convention,
$$
w_{xx}-xw_x+\lambda w=0.
$$
For $w=\sum_{j\ge0}a_jx^j$, its recurrence is
$$
\boxed{a_{j+2}=\frac{j-\lambda}{(j+2)(j+1)}a_j.}
$$
The base condition makes $a_0=0$ and therefore every even coefficient zero. An odd series terminates at degree $2n+1$ when $\lambda=2n+1$. It remains to prove that a nonterminating odd solution cannot have finite energy; termination alone would not establish the requested exhaustion of modes.

For that purpose, set $y=e^{-x^2/4}w$. The finite-energy condition is $y\in L^2(0,\infty)$, and its equation is
$$
Ly\equiv\left(-\frac{d^2}{dx^2}+\frac{x^2}{4}\right)y
=\left(\lambda+\frac12\right)y,\qquad y(0)=0.
$$
Because the coefficients are even and the solution is odd at the base, extend $y$ oddly to the whole real line. Define lowering and raising operators
$$
a=\frac d{dx}+\frac x2,\qquad
a^\dagger=-\frac d{dx}+\frac x2,\qquad
L=a^\dagger a+\frac12,\qquad [L,a]=-a.
$$
Integration by parts gives $\|ay\|^2=\lambda\|y\|^2$, so $\lambda\ge0$. These identities apply to the square-integrable solution: multiplying its equation by a compactly cut off conjugate and taking larger cutoffs bounds $y'$ and $xy$ in $L^2$, and removes the endpoint terms. The same argument applies to each lowered <eigenfunction>.

If $\lambda$ is not a nonnegative integer, repeatedly apply $a$. Each application lowers $\lambda$ by one and remains nonzero whenever its current value is positive, by the norm identity. After finitely many applications it would yield a nonzero square-integrable <eigenfunction> with negative $\lambda$, contradicting that identity. Thus $\lambda=N$ is a nonnegative integer. At $N=0$, $ay=0$ gives $y\propto e^{-x^2/4}$, which is even. Lowering reverses parity, so an initially odd <eigenfunction> can reach this even ground state only if $N$ is odd. Consequently $N=2n+1$. The recurrence then terminates and uniquely determines the odd solution from $a_1$.

The complete <adiabatic vertical modes of a Gaussian atmosphere> are therefore
$$
\boxed{w_n(z)=A_n\mathrm{He}_{2n+1}(z/H),\qquad
\sigma_n^2=\Omega^2[1+\gamma(2n+1)],\quad n=0,1,2,\ldots.}
$$
Here $\mathrm{He}_j$ is the <Probabilists' Hermite polynomial>. For example the first shapes are $x$, $x^3-3x$, and $x^5-10x^3+15x$, with <frequencies> squared $\Omega^2(1+\gamma)$, $\Omega^2(1+3\gamma)$, and $\Omega^2(1+5\gamma)$. Either sign of each real <frequency> gives a time-harmonic solution. Gaussian moments make every displayed <polynomial> have finite <kinetic energy>. Without the fixed-base condition, finite energy alone would also admit even <polynomials>, which is why that condition is necessary.