THEORYMERIDIAN

An experiment in breaking

Why Spaghetti Keeps Breaking

Feynman’s spaghetti puzzle. One snap that turns into a chain reaction.

Interactive companion 02Watch the film (opens in a new tab)
Simulationreal time
drag the right-hand grip to bend the strand
Strand 3 · 24 cm long · 1.4 mm thick
first crack100 µs1 ms10 ms1.5 s
Straightbreaks near 52 mm
20 cm28 cm
1 mm2 mm
Flawless12 %
WeakerStronger
25 mm60 mm
Each number is one strand, with its own weak spots

Changing a setting starts again with a straight strand. The same strand number always breaks the same way.

Readouts · this strand

Pieces1
Time since the first crack—
Sharpest bend now0 per m
Breaking limit14.3 per m

Look a little closer

One snap, then a ripple.

The pieces are straightening out overall. The ripple is what pushes a few spots past the limit.

Plate I

How sharply each spot is bent

Bend nowBend at the first crackBreaking limit, with weak spotsCracks, in order

Sharpness here is curvature: one divided by the radius of the circle that fits the bend at that spot, in units of “per metre”. A larger number means a tighter bend. Close to a fresh end the bend can pass the limit without a new crack: that is the shortest-piece rule described in the notes below.

Plate II

Why the ripple bends it more

At the new end nothing holds the strand bent, so the bend there drops to zero at once. That sudden change travels away as a ripple. For an ideal strand that was evenly bent, the mathematics gives a first crest about 1.43 times as sharp as the bend before the break, and the crest moves away from the end as the square root of the time elapsed.

A strand that was already at its limit cannot take 1.43 times more. It breaks again, and the new ends send out ripples of their own.

For the curious

Model and limits, in the open.

What is simulated

This is a simulation, not high-speed footage and not a validated prediction of any particular strand. The strand is a thin elastic rod that bends in one plane. By default it is 24 cm long between the grips and 1.4 mm thick, with stiffness (Young’s modulus) 3.8 GPa, density 1.5 g/cm³ and breaking stress 19 MPa: the values Heisser and co-authors report for the dry spaghetti in their experiments.

Slow bending. Each end is held at a point by a grip that is free to pivot, so the grips push the ends together without forcing an angle on them. The shape is the classical elastica: the exact equilibrium of a thin inextensible rod, not a parabola or a circular arc. Bending is taken to be slow enough that the strand is always in balance until the first crack.

The break. A spot cracks when its curvature passes the local breaking limit. For the default strand that limit is 14.3 per metre, a bend that fits a circle of radius 7 cm. The crack takes 10 millionths of a second to open. After that the motion of every piece is computed from Newton’s laws for a bending rod (240 segments, time steps of about a fifth of a millionth of a second), with the held ends still held and the new ends free. Any spot pushed past its limit cracks in turn. This part runs for the first 20 thousandths of a second.

After 20 ms. No more cracks are looked for. Each free piece keeps the speed and spin the strand gave it, falls 26 mm to the board, bounces a little, then slides and spins to rest under friction along its whole length. A fast-spinning piece barely grips the board, so it can skate a long way; the view widens to follow the pieces, and one that reaches the edge of the 62 by 44 cm board drops off. The rail is solid: a piece can land on it or be stopped by its side. A piece still held by a grip swings only a few degrees, because the pivot has friction. Leftover bending rings at the piece’s own natural frequencies and fades. Friction, bounce and fading rates are plausible values, not measurements, and pieces do not touch each other or the grips.

What is assumed, and what is left out

Weak spots. Real pasta has small flaws. Here the breaking limit varies along the strand by a few percent (5 % by default), drawn at random from the strand number shown in the controls. The pattern is fixed for each number, so every run can be repeated.

Shortest piece. In the ideal theory the ripple’s first crest appears right next to the new end, so a strand would shatter into dust. Real strands do not: a crack needs a little time and room to form, and measured pieces are rarely shorter than about 3 cm. The simulator represents this with a shortest allowed piece (40 mm by default, the value used for the film). This choice largely decides where the later cracks fall and how many pieces you get. Move the slider to see how much.

Left out. Twisting, air resistance, gravity during the first 20 ms, contact between pieces, and any motion out of the plane. A light internal damping is included (a relaxation time of half a millionth of a second, chosen as plausible, not measured). How fast you bend makes no difference here, because bending is treated as slow.

Twisting first. Heisser and co-authors showed that twisting a strand strongly before bending it makes a clean break in two far more likely. Twist is outside this planar model, so there is no twist control.

What the numbers say. Across 400 numbered strands at the default settings the model gave 3 pieces for <1 %, 4 pieces for 84 %, 5 pieces for 16 %, and never a clean break in two. This agrees with the common observation that two pieces are uncommon, but the exact shares depend on the shortest-piece assumption. Real strands vary more than this model does.

Scientific checks

The calculation that runs on this page is the same code that was tested. The tests compare the bent shape with the exact elastica solution, reproduce the 1.43 overshoot and its square-root-of-time spreading for a suddenly released rod, check the speed of small bending waves against theory, confirm that energy is conserved without damping and only falls with it, and count pieces over hundreds of strands at three resolutions.

Measured: the bent shape, its sharpest bend and the end force agree with the exact solution to within 0.11 %. The overshoot came out as 1.466, 1.436, 1.430 at three successively finer resolutions, against 1.428 in theory, with the crest position growing as time to the power 0.502 (theory 0.5). Wave frequencies were within 1.5 % of theory for waves at least 15 segments long. Without damping the total energy stayed constant to 0.011 % over 20 thousandths of a second; with damping it never rose.

The computational model is available as readable source, and the measured results as a validation record.

Sources, credits and licences

B. Audoly and S. Neukirch, “Fragmentation of rods by cascading cracks: why spaghetti does not break in half,” Physical Review Letters 95, 095505 (2005), doi:10.1103/PhysRevLett.95.095505.

R. H. Heisser, V. P. Patil, N. Stoop, E. Villermaux and J. Dunkel, “Controlling fracture cascades through twisting and quenching,” Proceedings of the National Academy of Sciences 115, 8665–8670 (2018), doi:10.1073/pnas.1802831115. Source of the material values and of the form of the breaking rule.

Developed for THEORYMERIDIAN by Daniel Hashemi. Original procedural apparatus and explanatory graphics. Oak board texture: Poly Haven “oak_veneer_01”, CC0. Fraunces, Geist and Geist Mono are distributed under the SIL Open Font License.

Companion to the film Why Spaghetti Keeps Breaking | Feynman’s Puzzle (opens in a new tab). The simulation runs in your browser. Its code does not include account registration or analytics. Hosting requests and external links are explained in our Privacy notice.