Why a Saw Needs an S to Sing
Abstract
An S-shaped blade traps a flexural vibration near its inflection point. The explanation connects ordinary shell mechanics to a carefully qualified topological mode.
Contents
Part of The Beauty of Physics, Mathematics, and Music (and Maybe Cats).
A steel saw blade can produce a sustained, nearly pure note. The trick is to bend it into an S: one part of the blade curves one way, another part the opposite way, and the curvature changes sign between them. A bow supplies energy, but the shape determines where a particular vibration can remain.
In a 2022 study, Suraj Shankar, Petur Bryde and L. Mahadevan investigated this familiar instrument using experiments and elastic-shell models. Their explanation has two layers. Curvature couples bending to stretching, making some vibrations costly away from the middle. The change of curvature sign also separates two mechanically distinct regions, with a localised mode between them. [[1]]
The second layer involves topology. Before reaching it, there is a useful question about an ordinary strip of metal: why should bending its resting shape change the way it vibrates?
Curvature Makes Bending Borrow the Stiffness of Stretching
A thin, flat sheet is comparatively easy to bend. Stretching it is harder because it requires changing distances within the sheet rather than moving an almost unchanged surface through space.
Curvature couples those motions. Push a curved surface along its normal and the geometry forces distances within the surface to change. The resulting vibration pays some of the cost of stretching. An arch is the everyday version of this effect, although an arch is a static load problem and a vibrating saw needs a shell calculation.
Let the blade have thickness \(h\), Young’s modulus \(E\), mass density \(\rho\), and Poisson’s ratio \(\nu\). For a homogeneous, isotropic, thin plate, its bending rigidity is
\[ D=\frac{Eh^3}{12(1-\nu^2)}. \]Here \(D\) has units of newton metres. The mass per unit area is \(\mu=\rho h\). In the linear, long-wavelength plate approximation, a plane flexural wave with wave-vector magnitude \(q\) has angular frequency
\[ \omega^2=\frac{D}{\mu}q^4. \]The fourth power is characteristic of bending waves. Their frequency is proportional to \(q^2\), rather than to \(q\) as for an ideal stretched string. Thickness enters strongly: since \(D/\mu\propto h^2\), a thicker plate vibrates at a higher frequency for the same wavelength.
This expression describes a locally flat plate. It does not yet include the saw’s curved shape or its finite edges.
The Curved Sections Restrict a Particular Family of Waves
Take \(x\) along the blade and \(y\) across it. In the shallow-shell description used in the paper, the resting surface curves along \(x\), with constant curvature \(b\) in a locally uniform section. Curvature has units of inverse metres; its sign tells us which way the surface bends.
For a plane wave, the reduced dispersion relation is [[1]]
\[ \omega^2 =\frac{D}{\rho h}q^4 +\frac{E}{\rho}b^2\left(\frac{q_y}{q}\right)^4, \qquad q^2=q_x^2+q_y^2. \]The assumptions matter: small oscillations, a thin and shallow shell, curvature in one direction, and in-plane motions that equilibrate rapidly compared with the flexural motion. The formula describes uniform regions; a spatially changing curvature requires a wave equation or a numerical shell model.
The first term is the flat-plate result. The second is the added contribution from curvature and stretching. At a fixed nonzero transverse wave number \(q_y\), the frequency cannot be made arbitrarily small by varying the longitudinal wave number \(q_x\): the curved region has a positive minimum frequency for that family.
There is no universal prohibition on low-frequency motion. If \(q_y=0\), the curvature term vanishes. The paper’s finite-strip calculations also contain extended low-frequency modes that do not acquire the relevant gap. The saw traps a particular flexural vibration while other motions remain possible.
A Numerical Example Separates the Mechanism From a Measurement
Consider an illustrative local model, rather than a fit to a particular musical saw. Choose steel-like properties
\[ E=200\ \mathrm{GPa},\quad \rho=7850\ \mathrm{kg\,m^{-3}},\quad \nu=0.30,\quad h=1.0\ \mathrm{mm}. \]These give \(D\approx18.3\ \mathrm{N\,m}\) and \(\mu=7.85\ \mathrm{kg\,m^{-2}}\). Choose \(q_x=0\), \(q_y=20\ \mathrm{m^{-1}}\), and \(b=0.60\ \mathrm{m^{-1}}\). The corresponding transverse wavelength is \(2\pi/q_y\approx0.314\ \mathrm m\). This is a chosen local wave, not the eigenmode selected by actual blade edges.
The flat contribution gives
\[ f_{\mathrm{flat}}= \frac{1}{2\pi}\sqrt{\frac{D}{\mu}}q_y^2 \approx97\ \mathrm{Hz}. \]The curvature contribution on its own gives
\[ f_{\mathrm{curvature}}= \frac{1}{2\pi}\sqrt{\frac{E}{\rho}}|b| \approx482\ \mathrm{Hz}. \]The two contributions add as squared angular frequencies, so the combined frequency is about \(492\ \mathrm{Hz}\), rather than their sum. Throughout, \(\omega\) is in radians per second and \(f=\omega/(2\pi)\) is in hertz.
The calculation shows how a gentle resting curvature can dominate the stiffness experienced by a flexural wave. It does not predict the study’s measured note, and it does not calculate the minimum of the entire dispersion relation: both would require further information about the mode and geometry.
The Inflection Point Provides an Interior Place to Vibrate
An S-shaped blade contains an inflection point, where its longitudinal curvature passes through zero and changes sign. Near that point, the curvature-induced contribution to the stiffness becomes small. Farther away, it grows again.
A convenient model profile is
\[ b(x)=b_0\tanh(x/\ell), \]where \(b_0>0\) is the curvature magnitude far from the middle and \(\ell\) sets the length over which it changes. This profile approaches \(-b_0\) on one side and \(+b_0\) on the other.
The localised vibration has its greatest displacement around the intervening region and diminishes along the blade away from it. Localisation matters because the hands, handle and supports can remove vibrational energy. A mode with little motion at those contacts couples less strongly to that loss route.
Thinking of a locally flexible region between stiffer regions gives a useful physical starting point. It leaves a further question, however: why does the sign change matter when the frequency formula contains \(b^2\), which cannot distinguish positive from negative curvature?
The Sign Change Appears in the Modes, Not Just Their Frequencies
The dispersion relation lists frequencies. It does not fully specify the displacement and stress fields associated with each frequency. Those mode structures retain information about the sign of the curvature.
Shankar and colleagues construct a topological difference between the two uniformly curved regions in the effective one-dimensional description. Writing the difference as \(\Delta\mathcal I\), their parity result is [[1]]
\[ (-1)^{\Delta\mathcal I} =\operatorname{sgn}(b_{\mathrm L}b_{\mathrm R}), \]where \(b_{\mathrm L}\) and \(b_{\mathrm R}\) are the curvatures on the left and right sides. Opposite signs give an odd difference and an interface mode within the relevant bulk gap.
This result relies on the model’s symmetries, including its uniaxial shell geometry and an emergent local reflection symmetry. It also compares regions made of the same material, so their short-distance behaviour can be matched. It is not a statement that any bent object has a protected note.
The distinction is useful: smoothly changing the shape while retaining the relevant sign change and model conditions need not remove the localised mode. Its frequency and position can still change. Protection concerns the mode’s existence within those conditions; it does not promise a fixed pitch, immunity to damage, or zero dissipation.
The supporting information also checks asymmetric end conditions. The localised mode remains concentrated in the interior while extended modes respond much more strongly to the changed boundary. That is the practical meaning of an interior interface here. No cut, joint or material boundary is needed. [[2]]
A Trapped Mode Still Loses Energy
A sustained note needs both a mode that stores energy and a mechanism that supplies it. Bowing can replenish the energy; striking the blade produces a ringdown. Localisation reduces losses at the supports, but the material and surrounding air can still dissipate energy.
The supporting experiments include an S-shaped saw struck with a mallet. The fitted amplitude-decay time was \(0.27\pm0.025\ \mathrm s\), with nearby spectral peaks around \(580\) and \(590\ \mathrm{Hz}\). Their 10 Hz separation corresponds to a beat period of about \(0.1\ \mathrm s\). The authors associate the modulation with the blade flexing and the inflection point moving after a strike. [[2]]
Those details keep the idealisation grounded. A performer does not hold a perfectly fixed shell, and an impulse can excite more than one motion. A localised mode can survive while its pitch and amplitude vary.
The S shape therefore has a precise mechanical role. It creates a change in curvature, concentrates a suitable vibration away from the supports, and makes that vibration comparatively reluctant to surrender its energy there. The mathematical boundary sits inside one continuous piece of steel.
Literature checked against the full paper and supporting information on 12 October 2026. Illustrative calculations were checked independently; they are not fits to the reported experiments.
References
[1] S. Shankar, P. Bryde and L. Mahadevan, “Geometric control of topological dynamics in a singing saw”, Proceedings of the National Academy of Sciences 119(17), e2117241119 (2022). doi:10.1073/pnas.2117241119. Author-hosted full text.
[2] S. Shankar, P. Bryde and L. Mahadevan, supporting information to “Geometric control of topological dynamics in a singing saw” (2022), especially sections 1, 3 and 4. Full paper and supporting information.
Changelog
- 2026-10-12: Research and calculation review. Curvature-induced localisation, its topological scope, and an independently checked local shell example.