The formula
Bessel functions are the two independent solutions of Bessel's equation, the differential equation you get when you separate the wave or heat equation in cylindrical coordinates:
x²·y'' + x·y' + (x² − ν²)·y = 0
Reading it: it looks like a spring equation y'' + y = 0 (which gives sine and cosine), but the x out front on the first two terms and the ν²/x² buried in the last term bend the solution so its amplitude decays and its wavelength stretches as x grows. The number ν (the order) is set by how the field wraps around the circle — mode number, azimuthal index, sideband count.
The solution that stays finite at the center is the Bessel function of the first kind:
J_ν(x) = Σ (−1)^k / ( k!·Γ(ν+k+1) ) · (x/2)^(2·k+ν) k = 0,1,2,...
Reading it: a power series like the one for cosine, but the factorials in the denominator grow faster, so it converges everywhere and the humps shrink as you go out. J_0(0) = 1; every other J_ν(0) = 0.
The second solution, singular at the center, is the Bessel function of the second kind (Weber's function):
Y_ν(x) = ( J_ν(x)·cos(ν·π) − J_(−ν)(x) ) / sin(ν·π)
Reading it: Y_ν blows up to −∞ as x → 0, so you keep it only when the origin is not part of your domain — a coaxial line, an annulus, a pipe with a hole down the middle. One trap in the formula itself: for integer ν it reads 0/0 — both sin(ν·π) and the numerator vanish — so you take the limit ν → n, and that limit is how Y_0, Y_1, ... (the orders you actually use on coax and annulus problems) are defined [3].
The two Hankel functions package J and Y into traveling waves instead of standing ones:
H_ν^(1)(x) = J_ν(x) + i·Y_ν(x)
H_ν^(2)(x) = J_ν(x) − i·Y_ν(x)
Reading it: same trick as e^(±ix) = cos x ± i·sin x. But which one radiates outward depends entirely on your time convention. With time dependence e^(−iωt) — the physics convention — H^(1) is the outgoing wave and H^(2) the incoming. With the engineering e^(+jωt) convention used in most antenna and waveguide texts (Harrington, Balanis), the roles swap and H^(2) is the wave radiating out from your antenna or scatterer. Check the convention on page one of whatever book you're working from before you reach for either.
For sphere problems (radar cross-section, acoustic scattering, a quantum particle rattling around a spherical cavity, the partial-wave expansions of scattering theory) you use spherical Bessel functions, which are half-integer-order J dressed up in closed form:
j_n(x) = √(π/(2·x))·J_(n+1/2)(x) e.g. j_0(x) = sin(x)/x
Where you meet it
- FM sidebands on the bench. Frequency-modulate a carrier and the spectrum isn't one line — it's a carrier plus sidebands whose amplitudes are
J_n(β), whereβis the modulation index. Crankβup to 2.40483 andJ_0hits zero: the carrier vanishes entirely. That "Bessel null" is how you calibrate an FM deviation meter without trusting the dial — you sweep deviation until the carrier disappears on the spectrum analyzer, then back out the exact deviation from a known audio tone. - Waveguide and fiber cutoffs at the RF design review. In circular waveguide the field pattern across the bore is a Bessel function, and each mode can only propagate above a cutoff frequency set by a Bessel zero. The TM01 mode cutoff rides on the first zero of
J_0(2.40483); the dominant TE11 mode rides on the first zero ofJ_1'(1.84118). Optical-fiber mode counts (the V-number) fall out of the same functions. If somebody asks why a 3.6 mm bore waveguide dies below a certain frequency, the answer is a Bessel zero. - Circular membrane and disk modes on the shaker table. A drumhead, a pressure-sensor diaphragm, a circular hatch cover, a spinning turbine disk — the mode shapes are
J_ν(k·r)·cos(ν·θ), and the natural frequencies come from forcingJ_νto zero at the clamped rim. Unlike a string, the overtones are not harmonic (2.405, 5.520, 8.654 for the axisymmetric family), which is exactly why a drum sounds like a thud and not a note. - Transient heat in a cylinder. Quench a round bar, shaft, or fuel rod and the temperature decays as a sum of
J_0modes in radius times exponentials in time. The thermal-soak time constant of the slowest mode again traces back to that firstJ_0zero.
How it works
Think of J_ν and Y_ν as cosine and sine that got tired. Far from the origin they behave almost exactly like a decaying sinusoid:
J_ν(x) ≈ √(2/(π·x))·cos( x − ν·π/2 − π/4 ) (large x)
Y_ν(x) ≈ √(2/(π·x))·sin( x − ν·π/2 − π/4 )
The amplitude falls off as 1/√x — geometry, not damping. A wave spreading out from a line source has to thin out as it covers more circumference, and the √x is that spreading. That asymptotic form is good to a couple percent by the time x reaches 20, and it's the reason waveguide and antenna work can lean on it so hard.
The zeros are the whole game, and they are not evenly spaced like π, 2π, 3π. The zeros of J_0 are 2.40483, 5.52008, 8.65373, 11.79153, ... — they only approach a π spacing far out. The single biggest mistake engineers make is reaching for n·π out of muscle memory. The first drum mode is at 2.405, not π; the second overtone is not double the first. Get this wrong on a diaphragm design and your predicted resonance is off by 30 percent at the low end.
Second gotcha: J versus Y. If your domain includes r = 0 (a solid rod, a filled waveguide), you drop Y because the field can't be infinite at the center. If it doesn't (a coax, an annular ring, a pipe), you keep both and the boundary conditions at two radii pin down the mix. Throwing away Y on an annulus gives you nonsense.
Third, on FM: because J_n(β) for large n stay small, the sidebands eventually die off, and roughly β + 1 significant sideband pairs carry the energy — that's the physical grounding of Carson's rule for FM bandwidth. And the power bookkeeping is exact: the squared amplitudes obey J_0² + 2·(J_1² + J_2² + J_3² + ...) = 1 for any β. FM doesn't create or destroy power; it just shovels it out of the carrier and into the sidebands. Verified numerically: at β = 5 the sum lands on 1.000 to fifteen digits.
Limits of validity: everything above assumes the separation actually worked — circular symmetry, a linear medium, boundaries that fall on constant-radius surfaces. Dent the waveguide, make the drumhead elliptical, or let the fiber index vary with angle and the clean Bessel picture degrades into a coupled-mode mess. The functions are still the right basis; the single-term answer just stops being enough.
History
Bessel functions were named for the wrong reason and by the wrong century — the usual story for anything useful. Daniel Bernoulli hit J_0 first, around 1732, chewing on the oscillations of a heavy chain hanging from one end and swinging free at the bottom [1][5][6]. Euler ran into the same functions attacking the vibrations of a stretched circular membrane [1] and worked out the series for integer orders [5][6]. Lagrange met them in planetary orbits, and Fourier used them in his 1822 treatise on heat conduction — the round-bar cooling problem that still carries them today [5][6].
The name got attached to Friedrich Wilhelm Bessel (1784–1846), a self-taught German astronomer who started as a merchant's apprentice and ended up director of the Königsberg observatory, famous separately for making the first solid measurement of a star's distance by parallax [2]. Around 1817 he was grinding through Kepler's problem — pinning down where a planet actually sits in its elliptical orbit — and needed the coefficients of a certain series expansion [1][2]. Those coefficients were the functions. He came back to them systematically in an 1824 paper on planetary perturbations, studied them as objects in their own right, and that treatment stuck his name on the whole family [1][2]. So the astronomer who never set out to invent a special function got the credit, while Bernoulli, Euler, and Fourier — who used them first — did not. Engineers should feel right at home with that.
Related tools
- /tools/frequency-wavelength
- /tools/vibration-natural-freq
- /tools/heat-conduction
- /tools/link-budget
- /tools/antenna-gain-beamwidth
- /tools/db-converter
Sources
- https://www.britannica.com/science/Bessel-function
- https://mathshistory.st-andrews.ac.uk/Biographies/Bessel/
- https://dlmf.nist.gov/10.2
- https://dlmf.nist.gov/10.21
- https://en.wikipedia.org/wiki/Bessel_function
- https://archive.org/details/treatiseontheory00watsuoft (G. N. Watson, *A Treatise on the Theory of Bessel Functions*, Ch. 1)