The eigenvalue problem
A membrane's height u(x,y,t) obeys the 2-D wave equation
utt = c² ∇² u, with u = 0 on the clamped boundary.
Separating u = φ(x,y) cos(ω t) turns it into the Helmholtz eigenvalue problem:
-∇² φ = λ φ, φ|boundary = 0, ω = c √λ
Each eigenfunction φ is a mode; its eigenvalue λ fixes the frequency.
The set of all λ is the drum's spectrum — literally what Kac meant by “hearing” it.
Rectangle a × b
Separation gives φmn = sin(mπx/a) sin(nπy/b), so
λmn = π²[(m/a)² + (n/b)²] and therefore
fmn ∝ √( (m/a)² + (n/b)² )
The square-root-of-a-sum-of-squares is why the ratios are irrational and change as you drag the aspect slider.
Some ratios can even collide (degenerate modes) at special aspect ratios.
Circle radius R
In polar coordinates the radial factor solves Bessel's equation, giving
φmn = Jm(kmn r) cos(mθ). The clamp
Jm(k R) = 0 forces kmn R = jmn, the n-th positive zero of Jm:
fmn ∝ jmn
The fundamental is j0,1 ≈ 2.4048. The next modes sit at ratios
1.594, 2.136, 2.296, 2.653, ... — nowhere near 2, 3, 4.
That is why a drum is inharmonic, unlike a string whose modes are at exact integers 1, 2, 3, ...
Answer to Kac's question: NO (in general)
In 1992 Gordon, Webb & Wolpert built two differently-shaped polygonal drums with the
identical spectrum — isospectral but not congruent. So you cannot always
recover the shape from the sound. You can hear some things (area and perimeter, via Weyl's law and
heat-kernel asymptotics), just not the full shape.