loading...
Press Play or choose a preset.
running...
The science
A free-free bar bends according to the Euler-Bernoulli beam equation rho A(x) d2w/dt2 + d2/dx2 ( E I(x) d2w/dx2 ) = 0 with A = b h(x) and I = b h(x)^3/12. For a uniform bar the mode frequencies are f_n = (beta_n L)^2 / (2 pi L^2) sqrt(E I/(rho A)) with beta_n L = 4.730, 7.853, 10.996, so the overtones sit at 2.76 and 5.40 times the fundamental: inharmonic, bell-like.
Carving an arch under the bar thins its middle. Thinner material lowers the stiffness of the middle, which drops mode 1 a lot (it bends mostly at the centre) and the higher modes less (their bending is spread along the length), so the ratios fall toward 1 : 4 : 10, a harmonic set that the ear fuses into one pitch. The model discretises the bar into Hermite beam elements with the thickness taken at each element, solves the generalised eigenproblem K phi = w^2 M phi (Cholesky reduction and Jacobi rotations) and reports the lowest six modes. The arch is a cosine-power profile with depth, width and shape parameters; Auto-tune runs a Nelder-Mead search on these to minimise the error of f2/f1 and f3/f1 from 4 and 10 (the search reaches roughly 4.1 and 9.7, as the three-parameter family cannot hit both exactly).
The mallet is a half-sine force pulse of contact time Tc: it excites mode i with amplitude J phi_i(x_m)/w_i times the pulse spectrum, so a hard mallet (short Tc) and the strike position change which modes are heard, and a strike at a node of a mode cannot excite it. Each mode decays as exp(-t/tau_i), tau_i = 1/(pi f_i eta_i) with a loss factor that rises with frequency. A closed tube hanging under the bar is a quarter-wave resonator with f_r = c/(4(L_r + 0.6 r)); when it matches f1 it boosts the fundamental by about 15 dB and drains energy faster (a shorter, louder note).
Honest simplifications. Euler-Bernoulli bending only (no shear or rotary inertia, which matters for thick bars at high modes), free-free ends and a rectangular section without the cord holes, typical material constants, loss factors chosen to give plausible decay times, a single-number resonator gain (Q = 12) and radiation weight instead of a coupled bar-air model, and an ideal half-sine contact pulse. The vibration is drawn magnified (real amplitudes are tens of micrometres).
In-page checks. For a uniform bar the finite-element f1 and the ratios f2/f1, f3/f1 are compared with the closed-form roots above; an undercut bar is solved with 20 and 80 elements to show mesh convergence; and the first two mode shapes are tested for mass-orthogonality.