Physical seam · a question the internet answers badly · show-the-check
Why Spaghetti Won't Break in Two
Take a dry strand, bend it until it snaps, and count the pieces. Almost never two. Three, usually — sometimes four or five. Feynman spent an evening on it and ended up with broken pasta all over the kitchen and no answer. The answer, when it came, was that the first break is not the end of the story.
Hold a strand of spaghetti by both ends and bring your hands together. It bows into an arc. Somewhere — wherever the strand is weakest — it reaches the curvature it cannot take, and it cracks. Your intuition says: one crack, two pieces. That is what a snapped pencil does, a broken twig, a torn sheet. But spaghetti refuses. Film it at ten thousand frames a second and you see the truth: after the first crack, more cracks appear, in a fraction of a millisecond, clustered near the first one. The strand shatters into a little burst of fragments before the halves have even begun to fall apart.
The physicist Richard Feynman noticed this while cooking with the computer scientist Danny Hillis. As Hillis tells it1, they spent two hours on it — breaking strands underwater to deaden the sound, inventing theories, testing them — and gave up: "…we ended up at the end of a couple of hours with broken spaghetti all over the kitchen and no real good theory about why spaghetti breaks in three." It took until 2005 for two physicists in Paris to settle it2 — work that won an Ig Nobel Prize, and turns out to ride on the same special function that draws the edge of a shadow.
The mistake in the intuition
A break does not just remove material. It lets go.
Here is the thing the snapped-pencil picture leaves out. While the strand is bent, every cross-section is under a bending moment — an internal torque, supplied by the rest of the rod, that holds it curved. The instant a crack opens, that torque at the new free end vanishes. It does not fade; it disappears. The end, a microsecond ago held hard in a curve, is suddenly free — and it whips straight.
That sudden straightening is not a quiet, local event. It launches a wave — a bending wave that races back along the still-curved rod. And a bending wave does something a sound wave never would: as it passes through a region that is already curved, it can pile its own curvature on top of what is there, so that for a brief moment the rod is bent more sharply than it ever was before the break. More sharply than the curvature that broke it in the first place. So it breaks again. And the second break launches its own wave. An avalanche.
The first panel integrates the linear beam equation; its strand drawing is schematic. The second evaluates the Fresnel integral. The third is a hand-set threshold toy. The offline verifier checks selected numerical results (13/13). Watch the calculated curvature climb past the chosen breaking line.
Bend it, break it, watch the wave overshoot
Top: a schematic strand. Bottom: |curvature| along it, with the breaking line κ_c. The first crack opens at the apex; the freed end straightens and sends a bending wave inward — the hump that climbs above the line is the overshoot. Where it crosses, the rod breaks again.
Audoly and Neukirch's 1.428× is the curvature amplification in the intermediate self-similar regime, after a short release and before end reflections matter. Their ideal instantaneous release gives this same coefficient. The larger peaks in this finite, clamped-rod calculation depend on its boundaries and observation window; they do not establish a universal 2× ceiling.
On slow release, the hump stays below the chosen line during the displayed window, t ≤ 0.34. The release lasts until τ = 1.1. Extend the same calculation and it crosses the line. This short view cannot tell us that the strand will stay in two pieces.
The universal number
1.428 — and it comes from the theory of light
The remarkable part of Audoly and Neukirch's 2005 result is that the overshoot factor is universal. It does not depend on the pasta, the thickness, how hard you bent it, or where it first broke. In the moments after the release, the curvature settles into a shape that spreads as √t and has no free parameters at all and its maximum in that intermediate regime is a pure number:
The Fresnel sine integral is the function that describes how light leaks into the shadow behind a sharp edge — the faint bright-and-dark fringes at the boundary of a shadow are S(z) and its cosine twin, plotted. The same curve, to the same maximum, sets how hard a breaking rod re-bends itself. Its largest value sits exactly at z = √2; double it and you have 1.428. Drag along the curve and find the peak yourself.
The Fresnel sine integral, and its maximum
S(z) climbs to 0.71397 at z = √2 = 1.41421, then oscillates down toward ½. Twice that maximum, 1.4279, is the intermediate-regime curvature overshoot. Verified against the special function offline to 2×10⁻³.
How to cheat it
Make it break in two: twist, or bend slowly
In 2018 a group at MIT, following an undergraduate course project3, demonstrated two ways to obtain two pieces. The two protocols act differently.
Quench it. Here the quench speed is how fast the ends move together while bending the rod, before it cracks. Slower bending gives larger minimum fragments and fewer fracture sites; sufficiently slow bending can give two pieces. This is a different control from the first panel's release time at the newly freed end.
Twist it. In the study, binary fracture dominates above about 250° of twist. Twist makes the rod reach its first fracture at a lower curvature. Torsional stress then dissipates before the bending stress peaks. Curvature still amplifies, but from that lower starting value the bending stress can stay below the level needed for another fracture.
A threshold toy: assumed amplification and piece count
This is a toy calculation with hand-set formulas and arbitrary dial units. The bar is an assumed amplification; the dashed line is a chosen threshold, κ_c = 1.12. The displayed piece count is assigned from that comparison. These knobs do not model physical twist or the 2018 quench protocol, and the verifier does not check this panel.
Record correction
What the internet tells you, and why it's wrong
Search the question and you will mostly get one of these. None is the mechanism.
| "It's random — flaws in the pasta." | Flaws decide where the first break is. They do not explain why extra breaks reliably appear, nor why they cluster near the first. The cascade is deterministic dynamics, not luck. |
| "The thickness is uneven." | It isn't, much — and the effect survives with perfectly uniform rods, and in simulation with no flaws at all. |
| "It's the vibrations." | Close, but it misses the point so completely it misleads. The slow, fundamental vibration keeps the curvature bounded by its starting value — Audoly and Neukirch say so explicitly. The overshoot lives in the fast, fine part of the bending wave, which a "just vibrations" answer never reaches. |
| "It breaks in the middle, then…" | The midpoint logic gets the first break right for a hand-bent strand. It says nothing about the secondary breaks — which is the entire puzzle. |
The real answer: a break is a release; a release is a wave; a bending wave amplifies curvature by 1.428× in the intermediate regime; and that re-breaks the rod before it can fall in two.
The check
The verifier checks the beam discretisation, Fresnel coefficient and selected finite-duration runs. Historical and experimental claims come from the sources below. research/why-spaghetti-wont-break-in-two/verify.mjs: 13/13:
- three beam modes follow ω = c·k²; the group/phase speed ratio is checked algebraically;
- energy drift rounds to 0.000% over 20 000 steps for the tested low modes;
- the intermediate-regime coefficient 2·max(Fresnel S) = 1.428, with the maximum at z = √2;
- the finite-rod runs reach about 1.425× for τ = 0.15 through t = 0.45 and 2.044× for a sharp step through t = 0.35; these are run results, not universal limits;
- the chosen symmetric static profile peaks at its centre; increasing release time lowers the peak within a fixed t = 0.45 window, but an extended τ = 1.1 run crosses the chosen 1.12 line.
The seam — what this models, and what it doesn't
The solver integrates the linear beam equation. Real fracture is nonlinear, three-dimensional, and involves an actual crack tearing through the pasta — none of which is here. What the linear model captures is the trigger: the curvature overshoot that re-crosses the breaking threshold. That is exactly the quantity Audoly and Neukirch isolate as the cause of the secondary breaks, and it is all this page claims. The first panel uses a chosen threshold and a schematic strand drawing. The third panel's formulas and piece counts are a threshold toy; they do not calculate either experimental route to binary fracture.
Sources
- The Feynman/Hillis anecdote: W. Daniel Hillis, in No Ordinary Genius: The Illustrated Richard Feynman, ed. Christopher Sykes (W. W. Norton, 1994). The wording is Hillis's recollection; we do not attribute it to Feynman directly. attribution solid · page not verified
- B. Audoly & S. Neukirch, "Fragmentation of Rods by Cascading Cracks: Why Spaghetti Does Not Break in Half," Phys. Rev. Lett. 95, 095505 (2005). DOI 10.1103/PhysRevLett.95.095505. Ig Nobel Prize in Physics, 2006. The 1.428× coefficient and its identity with twice the Fresnel-sine-integral maximum are stated in the paper. solid
- R. H. Heisser, V. P. Patil, N. Stoop, E. Villermaux & J. Dunkel, "Controlling fracture cascades through twisting and quenching," PNAS 115(35), 8665–8670 (2018). DOI 10.1073/pnas.1802831115. Binary fracture dominates above a twist of ≈250° (the ~270° figure widely quoted is MIT's press framing for a specific 10-inch strand). Dry-spaghetti diameter 1.4 mm and Young's modulus 3.8 GPa are from this paper. solid
- Beam-wave background: K. F. Graff, Wave Motion in Elastic Solids (Dover, 1991); L. D. Landau & E. M. Lifshitz, Theory of Elasticity (bending of rods). standard