The property that appears in none of the equations

Search this site for the word ductility in a formula and there is nothing to find. The plastic hinge is computed from the yield stress and the section’s geometry. The collapse load comes out of a work equation containing moments and rotations. The redundant beam’s moments come from stiffnesses. Every number quoted anywhere is produced without ever asking how far the material can stretch before it comes apart.

And yet almost every one of those methods is silently conditional on the answer. Ductility is not a term in the equations; it is a term in the derivations, and it drops out of them by being assumed sufficient. Which means the failure mode it guards against is not a larger number in a familiar calculation. It is the calculation ceasing to describe the structure.

The comparison is worth taking apart, because on the scale a designer usually works at the two materials do not look different at all.

What the methods are actually assuming #

The plastic hinge assumes rotation. A propped cantilever collapses when two hinges have formed, and the second one forms only after the first has rotated far enough for the moment at the other critical section to catch up. If the first hinge fractures at the rotation the second needs, the mechanism never assembles and collapse arrives at the load that formed the first hinge — which for that beam is 0.686 of the load the analysis predicts.

The work equation that produces the collapse load multiplies each plastic moment by a rotation, and nothing in it asks whether the material can supply the rotation. A mechanism is drawn with a kink at each hinge, and a kink is a point; a point can turn through any angle at all. A real hinge is a region a few section depths long, and how far it can turn before something tears is a quantity that appears nowhere in the calculation it governs.

Moment redistribution assumes rotation too. The whole benefit of continuity — the moment over the support falling as the mid-span moment rises — is a statement that once the support section has yielded, further load goes elsewhere. It goes elsewhere by the support section rotating at constant moment. A section that cannot rotate holds its elastic share of the load right up to the moment it breaks, and the redistribution never happens.

The lower-bound theorem assumes it everywhere at once. The theorem says that any set of internal forces in equilibrium with the load and nowhere exceeding the section capacity is a safe estimate of collapse. It is the most useful result in plastic theory and the licence for an enormous amount of practical design, including every masonry arch on this site. Its proof requires that the structure be able to reach the assumed state, which means every section that has to yield can yield and stay yielded while the others catch up. Without that, a lower-bound solution is not a bound at all.

A bolt group assumes it. A line of bolts in a lap joint does not share the load equally: the end bolts take much more, because the plates stretch between them. The design assumption that all the bolts carry the same force is only true once the end ones have yielded and shed their excess to their neighbours. A group of brittle high-strength bolts fails when the first one reaches capacity, and the joint carries a fraction of what the sum of its bolt capacities suggests.

Redundancy itself assumes it. The reason one support too many is worth having is that a structure with two routes to the ground survives losing one. It survives by the remaining route taking up the load, which requires the failed route to deform without shedding what it was already carrying, and requires the remaining one to accept a substantial overload while it does. Both are ductility demands, and a brittle redundant structure is a structure with two routes that both fail on the same afternoon.

The picture of what it costs #

The measurable version of ductility for a section in bending is its rotation capacity: how far past the curvature at first yield it can be bent while still carrying its plastic moment. That is a quantity the moment-curvature curve answers directly, and it varies enormously between shapes carrying the same load.

That last sentence is the trade that a shape factor by itself conceals, and it has an essay of its own. The relevant point here is that a demand for rotation is generated by the analysis and supplied by the section, and nothing in the analysis checks that the second covers the first.

Which free body produced the number #

The rotation figures above come from the section itself, taken as a free body under a curvature. Slice the cross-section into fibres, give each fibre the mild steel curve, impose a straight-line strain distribution of slope $\kappa$, and require the stresses to integrate to zero axial force. That fixes the position of the neutral axis; integrating the same stresses against distance gives the moment.

Sweeping $\kappa$ upward gives the whole curve, and the numbers quoted are read off it: the curvature at which the first fibre reaches 0.131% strain, and the curvature at which the moment reaches 98% of the rigid-plastic value computed independently from the equal-area axis. For the rectangle those are in the ratio 4.10.

Two independent checks make that trustworthy. At small curvature the computed moment divided by $EI\kappa$ is 1.000000, which it has to be if the fibre integration reproduces elastic beam theory. And the rigid-plastic moment for the rectangle comes out as $275 \times 100 \times 200^2/4$ to within one part in a million of the closed form, computed from areas with no stress-strain law involved at all.

The rotation capacity of a member is a harder quantity and the figure above does not claim it. A hinge occupies a length, and turning a curvature into a rotation requires knowing that length, which depends on the moment gradient and on how far the yielded zone spreads along the beam. What the section-level calculation gives is the ratio between shapes, and the ratio is what the argument needs.

What a brittle redundant structure does instead #

The cleanest way to see what is being assumed is to take a structure whose whole design rationale is redundancy and remove the ductility from it.

A propped cantilever under a uniform load has one redundancy. Elastically, the fixed end carries $wL^2/8$ and the span carries about $9wL^2/128$ — the support moment is 1.78 times the span moment, so the support section reaches its capacity first, at a load well below what the beam as a whole could carry.

Everything the plastic analysis subsequently gains comes from that imbalance being allowed to even out, and evening out has a physical meaning: the fixed end holds its moment while the beam bends further and the span moment climbs to meet it. The imbalance is a fact about the elastic distribution. Whether it is allowed to even out is a fact about the material.

Grant ductility and the story is the familiar one: a hinge forms at the fixed end, the beam becomes simply supported for further load, a second hinge forms in the span, and the collapse load is $11.66\,M_p/L^2$ — 46% above the load that first yielded the support. Withhold ductility and the beam fails when the support section reaches its moment, at $8\,M_p/L^2$, and the entire 46% is not merely unavailable but was never there.

What makes this worth stating carefully is that the elastic analysis of the brittle beam is perfectly correct. Nothing about the moment distribution above is wrong. What has gone is the step that follows it — the claim that reaching capacity somewhere is not reaching capacity everywhere — and that step is not in any equation. It is in the sentence between two equations.

What it is really underwriting: the analysis itself #

There is one more thing ductility is quietly paying for, and it is larger than any of the mechanisms above. It is what makes an analysis of a redundant structure safe to use when its stiffnesses are wrong.

Consider what actually goes into a frame model. The flexural rigidity of a reinforced concrete member is taken as some fraction of the uncracked value — a half, a third, whatever the code suggests — when the true value varies along the member and changes with the load. A steel joint modelled as rigid has a real stiffness somewhere on a curve. A composite beam’s stiffness depends on how much of the slab is participating, which is a number nobody measures. A support modelled as fixed is a foundation with a rotational flexibility.

Every one of those errors changes the distribution of the internal forces, and in a redundant structure the distribution is what the analysis is for. A model whose relative stiffnesses are wrong by a factor of two gives moments that are wrong by a substantial fraction, and no amount of care with the arithmetic recovers them.

And it does not matter, provided the structure is ductile. The forces the model produced are in equilibrium with the load — that much is guaranteed by the solver, whatever it assumed about stiffness — so they are a statically admissible set, and the lower-bound theorem says a structure proportioned for a statically admissible set will carry the load. The stiffnesses cancel out of the guarantee. What the real structure does is redistribute from wherever the model was wrong to wherever it had capacity in hand, which is exactly the operation ductility supplies.

That is why a profession that knows its stiffness inputs to a factor of two is content to design from outputs quoted to three significant figures. The figures are not a claim about what the structure will do; they are one admissible answer among many, and the theorem covers the rest.

The exception is the case where it does not hold, and it is worth naming because it is the one where analysis has to be taken seriously in a way steel frames never require. A brittle structure’s distribution is its answer. A masonry building, a glass assembly, a bolted joint whose fasteners cannot yield, a prestressed member designed to stay uncracked: in each of these the first section to reach its limit is the last, so being wrong about the stiffnesses is being wrong about the capacity. There the model has to be right, the sensitivity has to be tested, and the comfortable slack this essay has been describing is simply not there.

Where the model stops #

Rotation capacity is not a material property, and it is not a section property either. It is a property of the section, the material, the moment gradient, the axial load and the local buckling behaviour of the plates the section is made of, and the last of those usually decides. A section whose flanges buckle locally at the strain the analysis wants sheds its moment immediately, and its rotation capacity is nearly zero however ductile the steel is. That is the subject of section classification, and it means a ductile material can be assembled into a brittle member.

Ductility measured on a coupon is not ductility available in a joint. A welded connection is a region of altered metallurgy, residual stress and geometric discontinuity, and the strain it can accept is far below the 20% a tensile test reports. Nearly every structural collapse attributed to brittleness has occurred at a joint.

That 20% is worth distrusting on its own account, before any joint is involved, because it is not purely a property of the material.

The specimen’s own size enters the same way, and the size of the effect is larger than the difference between two steels.

Nor is a coupon’s ductility available where the material is restrained. A tensile specimen thins as it stretches, and the thinning is what lets it stretch. Material in the middle of a thick plate, or at the root of a notch, cannot thin — the surrounding material holds it — and it is therefore pulled in three directions at once. Triaxial tension of that kind suppresses yielding entirely while leaving the stress that drives fracture untouched, so the same steel that reports 20% elongation on a coupon can break at a fraction of a per cent in a restrained detail. This is why plate thickness appears in toughness requirements at all, and why it is the geometry of a joint rather than the specification of its steel that usually decides whether it behaves.

Temperature moves it discontinuously. Mild steel is ductile at 20°C and can be brittle at −20°C — not a little less ductile, but capable of failing at a nominal stress it passes every test at, with no plastic deformation whatever. The transition is sharp and the material’s tensile test at room temperature contains no hint of it.

Reinforcement yielding is not the only route, though, and the other one is the clearest demonstration on this page that ductility is bought with geometry rather than with material.

Nothing was done to the concrete. A hoop was put round it, and the material that had a usable strain of 0.0035 now has one of 0.028 — which is the difference between a column that spalls and drops its load and one that can be asked for a rotation. The third number, the toughness, is what the hoops are actually provided for, and it is the number that appears in none of the equations.

The generalisation #

The pattern is that an assumption that appears in a derivation and not in its result is invisible to everyone who uses the result.

This site has met the same shape before. Plane sections stay plane is nowhere in $\sigma = My/I$, and it is the reason that expression fails for a deep beam. The pin-jointed idealisation is nowhere in a truss’s member forces, and it is the reason those forces are missing a bending stress that can reach a quarter of them. Ductility is the same class of thing, and it is the most consequential member of the class because its absence does not degrade an answer gracefully — it removes the guarantee entirely.

The practical response the profession has settled on is not to compute ductility but to legislate the conditions under which it can be assumed: limits on plate slenderness so that local buckling cannot pre-empt yielding, limits on the amount of moment that may be redistributed, minimum reinforcement so that a concrete section cannot crack and break in the same instant, toughness requirements tied to thickness and service temperature, and rules on where welds may be placed. Every one of those is a rule about geometry or specification standing in for a calculation nobody performs.

How much of a section has to have given up #

There is one more thing worth measuring, because it explains why the ductility demand is so much larger than intuition suggests. Reaching the plastic moment does not mean the extreme fibre has just yielded. It means nearly the whole section has.

That is the geometric reason the demand is large. The plastic moment is collected by recruiting the material nearest the neutral axis, and material near the neutral axis is at low strain by construction — so bringing it to yield requires the material at the surface to go far past it. A section’s shape factor and its ductility demand are two readings of the same fact, and the second is the one that has to be paid.

A surprising place this turns up #

The most valuable thing ductility buys is not capacity. It is information.

A ductile structure approaching its limit deflects visibly, cracks audibly and holds. That interval — between the first sign and the collapse — is what makes it possible to evacuate a building, close a bridge, or notice a badly built structure before it is loaded. It is not in any calculation and it is worth more than most of the margins that are.

The mild steel plateau is the same observation at the scale of a specimen: what the plateau buys is a permanent set that somebody notices. Ductility is the property that converts a structural margin into a warning, and a structure that is exactly as safe but brittle offers the same margin with none of the notice.

Where the ladder goes next #

Later rungs on this anchor: rotation capacity computed properly, with hinge length and moment gradient, against the demands that redistribution generates. The upper- and lower-bound theorems proved, and exactly which step of each proof consumes ductility. Section classification as a rotation-capacity requirement rather than a table. Robustness and tying — the design of structures to survive the removal of a member, which is a ductility argument applied at the scale of a building and arrived in codes after Ronan Point in 1968. Ductility in reinforced concrete, where the whole design philosophy is arranged so that the steel yields before the concrete crushes, and where over-reinforcing a section makes it stronger and brittle at once. And capacity design, the seismic idea of choosing where the ductility will be demanded and making everything else stronger than the chosen place.

Historically, the recognition that ductility was doing hidden work came out of failure rather than theory. Plastic design was developed at Cambridge from the mid-1930s in part from the observation that steel-framed buildings damaged by bombing had survived deformations no elastic analysis could account for — the frames had used ductility nobody had asked them for. The Liberty ships going the other way, splitting in two at ordinary stresses in cold water, made the same point in reverse, and the investigation that followed established that a steel meeting every strength requirement could be structurally useless if it could not deform.