The crack's demand meets the material's limit
In the last guide we replaced the useless 'stress concentration factor' of an infinitely sharp crack with one clean number — the stress intensity factor K, given by K = Y times σ times sqrt(π times a). Here σ is the far-field applied stress, a is the crack size (half-length for an internal crack, full length for an edge crack), and Y is a near-1 shape factor for the geometry. K measures how hard the crack tip is being driven: it rises with the load and with the square root of the crack, so a longer crack under the same stress shouts louder. That is the crack's demand.
Every material answers that demand with a supply: a maximum K it can withstand before the crack goes unstable and runs across the part in milliseconds. That critical value is the fracture toughness, written K_IC — the I means 'mode I', the simple opening (pulling-apart) mode, and the C means critical. The design rule is beautifully blunt: while K < K_IC the crack sits still and you are safe; the instant K climbs to meet K_IC it fractures. All of fracture-safe design is keeping the driving K below this one material property.
K carries the strange-looking unit MPa·sqrt(m) — megapascals times the square root of a metre — because it is a stress times sqrt(length), neither a pure stress nor an energy. That odd unit is exactly what lets K fold two things into one: a long crack under gentle load and a short crack under fierce load can share the same K, meaning identical stress fields at their tips and identical nearness to fracture. K_IC wears the very same unit, so you can compare demand and supply directly — the crack's K against the material's K_IC.
Griffith's equation and why metals are tough
Where does K_IC come from physically? From an energy bargain that A. A. Griffith struck in 1920 to explain a scandal: a perfect crystal snapping all its bonds at once should have a strength of roughly E/10 — for steel with a Young's modulus near 200 GPa that is about 20 GPa — yet real steel breaks near 0.2 to 2 GPa, often ten to a hundred times weaker. The measured 'strength' is never the strength of the bonds; it is the stress at which the worst pre-existing flaw begins to run. Flaws, not bonds, set the limit.
Griffith balanced the books. Extending a crack a sliver costs energy — you create two fresh surfaces, and surface costs a surface energy γ_s — but it also releases stored elastic energy from the material that unloads around the crack. The release grows with a^2 while the cost grows only with a, so past a critical size the release overwhelms the cost and the crack accelerates away, catastrophic and self-feeding. Balancing the two gives the Griffith criterion: σ_f = sqrt(2 times E times γ_s / (π times a)). Strength falls as 1/sqrt(a) — double the flaw and you drop the strength by sqrt(2). This is the same physics as K seen from the energy side, tied together by K = sqrt(E times G), where G is the energy released per unit of crack advance; fracture strikes when G reaches a critical G_c, equivalently when K reaches K_IC.
Pure Griffith works for truly brittle solids like glass, but it badly under-predicts metals, and the reason is the deepest lesson here. At a crack tip in a metal the material yields plastically first, and that plastic work dwarfs the surface energy. Orowan fixed the equation by replacing γ_s with (γ_s + γ_p), where the plastic term γ_p can be a thousand times larger. That plastic zone blunting the crack tip and soaking up energy is precisely why a ductile steel is tough and a glass is not — even though their atomic bonds are not wildly different in strength. Toughness lives in the plastic zone, not the bonds.
The critical crack size: the number that saves the part
Now the payoff, and it is worth memorising. Set the driving K equal to the material's K_IC and solve for the crack: a_c = (1/π) times (K_IC / (Y times σ))^2. This a_c is the critical crack size — the largest flaw a part can carry at a given stress before it fractures. It is the most useful equation in the whole rung, because it turns an abstract toughness into a concrete, answerable question: at my working stress, how big a crack can I actually live with?
A worked example, then its cautionary twin. Take an aluminium alloy with K_IC = 30 MPa·sqrt(m), stressed to σ = 200 MPa, with an internal through-crack so Y is about 1. Then a_c = (1/π) times (30 / 200)^2 = (1/π) times (0.15)^2 = 0.0072 m, roughly 7 mm — so a whole internal crack about 2a = 14 mm across. That is big enough to spot by eye or by easy inspection: comfortable. Now a high-strength steel with K_IC = 50 MPa·sqrt(m) run hard at σ = 1500 MPa gives a_c = (1/π) times (50 / 1500)^2 = 0.00035 m, just 0.35 mm — a crack far too small to reliably find. Same equation, opposite fate.
FRACTURE TOUGHNESS K_IC (MPa*sqrt(m), approx, mode I) soda-lime glass ........ 0.7 | brittle: a scratch is a fatal crack PMMA / acrylic ......... 1 - 2 | alumina (Al2O3) ........ 3 - 5 | silicon carbide ........ 3 - 4 v a_c only microns wide ------------------------------------------------------------ cast iron .............. 6 - 20 ^ aluminium alloys ....... 20 - 40| titanium alloys ........ 50 - 80| tough: a_c is mm to cm wide high-strength steel .... ~50 | mild / structural steel 100-200| you can SEE the crack coming higher K_IC -> larger tolerable flaw a_c = (1/pi)(K_IC / Y*sigma)^2
Designing against fracture: three levers
The a_c equation hands the engineer three independent levers, and real damage-tolerant design pulls all three at once. First, choose a tougher material — raise K_IC and a_c grows with its square. Second, drop the working stress: since a_c falls as 1/σ^2, a bigger factor of safety on stress buys a disproportionately bigger tolerable crack. Third, inspect better: with good non-destructive testing — ultrasonics, X-ray, dye penetrant — you can find smaller cracks, so you design the part so that a_c stays comfortably above your smallest detectable flaw.
A favourite trick that combines these is leak-before-break, prized in pressure vessels and pipelines. You deliberately size the wall so that the critical crack length is larger than the wall is thick. Then a growing crack punches all the way through and leaks — visibly, harmlessly, and noisily — before it can ever reach the length needed to run unstably around the vessel. The failure announces itself as a puddle or a hiss rather than a bang. It is a designed-in early-warning system that trades a small leak for the avoidance of a catastrophe.
- State the job: the working stress σ the part will really see, and the shape factor Y for the likely crack (edge, internal, surface) and geometry.
- Look up the material's K_IC — and use a plane-strain value measured on a thick enough specimen, at the coldest temperature the part will meet.
- Compute the critical crack size a_c = (1/π)(K_IC / (Y·σ))^2 — the biggest flaw that stress can tolerate before fracture.
- Compare a_c with your smallest reliably detectable flaw. If a_c is not comfortably larger, pull a lever: tougher material, lower stress, or better inspection — then re-check.
Honest limits: scatter, cold, and slow-growing cracks
Three honesty checks keep this powerful tool from becoming overconfidence. First, scatter. For a brittle material there is no single true 'strength', because failure starts at the worst flaw and flaws differ from piece to piece — a big component is likelier to hide a nasty one than a small test coupon, so bigger brittle parts are statistically weaker. This is captured by Weibull statistics, which honestly reports a probability of failure rather than one number. It is a real advantage of K_IC that, being a property of the material rather than of its worst flaw, it is a cleaner quantity to design with than a lone ceramic strength value.
Second, temperature. K_IC is not even constant for one material: many steels and other body-centred-cubic metals lose their toughness in the cold, snapping brittle below a ductile-to-brittle transition temperature they would easily survive when warm. The Titanic's hull plate and the WWII Liberty ships that cracked in half in frigid seas are the classic cautionary tales, teased apart afterwards by impact testing. Guide 1 met that transition; the lesson to carry into a_c is simple — feed the equation the low-temperature K_IC, because a part that is tough in summer can be brittle on a winter night.
Third, and most important, everything above assumed the crack was already there and the load was steady. But most real cracks are not born full-size — they grow. Under repeated cyclic loads a sub-critical crack inches forward a tiny bit each cycle until it finally reaches a_c and the part fractures with no warning: that is fatigue, the cause of most in-service failures and the whole subject of the next guide. At high temperature the material itself slowly stretches and internal voids link up over time — creep, guide 5. Fracture mechanics is the frame; fatigue and creep are the two clocks quietly ticking a small flaw up toward its critical size.