How far can a long span skylight go on a flat roof?

Further than most people expect, but not by making the beams bigger. On the systems we build, a cable-supported flat skylight spans 10 to 45 metres — at Fanap, one roof of 20 × 20 m and two of 26 × 17 m. The reason the span is not simply a matter of deeper steel is that deflection rather than strength governs a glass roof, and deflection rises with the fourth power of the span: a 26-metre member deflects around forty-six times as much as a 10-metre one of the same section. Holding that back by depth alone would need roughly three and a half times the structural depth, which on a skylight is sky the occupant no longer sees.

What is a cable truss, and why use one in a roof?

A cable truss is a structural assembly in which a slim compression member is held against pre-tensioned cables, usually spread apart by struts, so that the pair works as a single deep element. The load is carried by the steel in compression and the cable in tension rather than by any member in bending. Its usefulness on a skylight is that the structural depth becomes the distance between the steel and the cable, not the depth of the steel itself — so the member can be slender while the assembly is deep. From underneath you see a line of steel, a strut and a cable rather than a truss.

Why are cables so much more efficient than steel beams?

Two reasons that compound. A cable works in pure tension, so every strand in it carries the same stress — where in a beam carrying bending the stress is highest at the top and bottom faces and falls to zero at the neutral axis, so most of the section does very little. And cable steel is drawn as high-tensile wire to several times the design strength of a structural section. Full use of the material and a far higher working stress, together, are why a stainless spiral strand of 12 to 36 millimetres can do the work of a member many times its own depth.

What does pre-tensioning a cable roof actually do?

Three things. It keeps every cable in tension under every load case, including wind uplift on a light flat roof, because a slack cable contributes nothing and leaves the steel above it carrying load at a size it was never designed for. It supplies stiffness, since a cable with no tension in it has almost none until it has pulled straight, so the roof's deflection behaviour is a property of the pre-tension as much as of the sections. And it sets the geometry — a pre-tensioned structure is only in its designed shape when the designed tension is in it.

Why is there a tensioning sequence?

Because tensioning one cable changes the force in every other cable in the system and moves the geometry of the whole assembly. There is no order-independent way to do it. The sequence is computed in advance, carried out in order, and the geometry is surveyed as it proceeds, so the structure arrives at its design shape at the end of the operation rather than at any point during it. It is the part of a cable project that cannot be compressed in the programme, and it is what separates a roof that performs as calculated from one that merely stands up.

Does temperature affect a pre-tensioned cable roof?

Yes, and it is the effect most often underestimated. Pre-tension is created by stretching a cable a known amount, and steel changes length with temperature — so the tension in the system varies with how warm it is. A 26-metre run of steel through a 50-degree annual swing changes length by around 16 millimetres on thermal movement alone, in a system whose working tension is set by a few tens of millimetres of stretch. The cable and the structure it is anchored to move together only in part, so pre-tension is specified as a range checked at both extremes — nothing slack at the warm end, nothing overstressed at the cold end — and set on site against the temperature the structure is actually at.

How much does a cable-supported glass roof deflect?

Considerably more than people expect, and deliberately. The criterion on this system is L/100 to L/200 under snow, which on a 20-metre span is between 100 and 200 millimetres, and on 26 metres between 130 and 260. The glazing, the gaskets and every joint are detailed to follow that movement rather than resist it, because glass has no capacity to follow a frame that moves more than its fixings allow. The drainage falls are also set on the deflected shape rather than the built one, so the roof as constructed is cambered against the deflection it is going to have.

Can you put a large skylight in a flat roof?

Yes, and the question is what carries it rather than whether it is possible. Below roughly eight metres a framed unit spans on its own. Beyond that, a flat roof needs a structure, and because deflection grows with the fourth power of span, that structure gets deep fast — which on a skylight defeats the purpose. Cable reinforcement is the usual answer where the roof must stay flat: it takes the load out of bending so the members stay slim. Where the architecture will accept curvature instead, a curved surface achieves the same thing without pre-tension or a tensioning sequence.

How does a glass roof meet a curtain wall?

With a joint engineered to move and to drain, and a structural separation behind it. The two systems move independently and by different amounts — the roof deflects under snow and recovers, while the wall moves under wind, thermal expansion and inter-storey drift — so the junction has to absorb the sum of both while staying watertight. The roof's reactions, which on a cable roof can include a horizontal pull at the anchorages, cannot be carried by the curtain wall, because a curtain wall is a non-loadbearing skin. They pass the wall line into the primary structure behind it. And the same line is where the roof drains, which on a wall running several storeys below gives any water that gets past a long way to travel.

Why were eleven thousand components machined rather than fabricated?

Because in a cable structure the geometry is the sum of the lengths of its components. A cable run is a fork, a swaged socket, a length of strand, a tensioner, another socket, another fork and a pin at each end, and the position of the node it reaches is the accumulation of every one of those dimensions. Those errors are in series, so they do not average out: a consistent millimetre of error at every fitting across a roof produces a geometry that does not close, and in a pre-tensioned system that means the tension is wrong everywhere, because force in a cable is set by how far it has been stretched. Turning the parts to dimension is what makes the assembly arrive where it was calculated to arrive.

Project 02

Fanap Campus Complex

Best Special Steel Structure in Iran, 2022

Three flat glass roofs over the courtyards of a Tehran campus — one of 20 × 20 m and two of 26 × 17 m — built as pre-tensioned cable trusses, so that the steel the span needed did not take away the sky the skylight was for.

Fanap Campus Complex, Tehran — a pre-tensioned cable truss skylight meeting the curtain wall at the end of a courtyard

Location: Science and Technology Park, Tehran
Type: Technology campus
Recognition: Best Special Steel Structure in Iran, 2022
Client: Fanap Company
Architect: Boompad Consulting Engineers
Main contractor: Paya Takht-e Jamshid Holding
Systems: Cable-supported skylight · curtain wall
Envelope package: 4,000 m²
Skylights: 20 × 20 m and two of 26 × 17 m — ≈ 1,280 m²
Machined components: ≈ 11,000

What the architecture asked for

The campus building is organised around three courtyards, side by side, and the architectural design closed each of them with a flat glass roof.

One of 20 × 20 metres. Two of 26 × 17 metres. Flat — not curved, not pitched, not vaulted. And on two sides each roof met a curtain wall at ninety degrees, with that wall continuing down, uninterrupted, to the ground floor.

Design visualisation of a Fanap courtyard — a flat glass roof on a cable truss, meeting a curtain wall that runs down to the ground floor — Design visualisation A courtyard as designed: a flat glass roof, and under its edge a curtain wall that runs to the ground floor.

At first look that is not a buildable drawing. Twenty-six metres is a long way for a flat glass roof to go without curvature to help it, and the junction — a glass roof meeting a glass wall at a right angle, with the wall running the full height of the building below it — concentrates every difficult thing a façade has into one line.

It was buildable, though, and the architect had already solved it. The scheme carried very deep beams and a trussed structure through the roof plane, sized to take the flat span and hold its deflection. Structurally it worked.

What it cost was the reason for building a skylight at all.

The three Fanap courtyard roofs seen from directly above — one of 20 × 20 m between two of 26 × 17 m — CORTEX project The three roofs from directly above, glazed: the square one in the middle and a longer one either side.

Why a correct answer was still the wrong one

This is worth setting out properly, because it is the whole project.

A flat member spanning a distance carries its load in bending, and bending punishes span twice over.

The bending moment a beam has to resist rises with the square of the span. Double the span and the moment goes up four times.

But on a glass roof it is not strength that governs — it is deflection, and deflection rises with the fourth power of the span. Against a 10-metre roof of the same section under the same load, one of 20 metres deflects sixteen times as much. At 26 metres, the long side of the two larger roofs here, it is forty-six times.

Diagram What span does to a flat roof

One section under one load per metre, compared with a 10 m span. 17, 20 and 26 m are the plan dimensions of the three roofs.

To pull that back you need stiffness, and stiffness in a beam comes almost entirely from depth: the second moment of area rises roughly with the cube of the depth. So holding the deflection where it was means roughly two and a half times the structural depth at 20 metres, and three and a half at 26.

That is where the deep beams and the trusses in the original scheme came from. They are the honest structural answer to a long flat span, and any engineer would have drawn something similar.

And on a skylight, structural depth is the one thing you cannot spend.

Every millimetre of member depth is sky the occupant does not see. A deep truss grid over a courtyard reads from below as a ceiling with glass between its members rather than as a roof you can see through, and it takes three things away at once: the daylight reaching the floor, the view of the sky, and the appearance of the roof the architecture was organised around.

A long flat rooflight built this way stops being a skylight and becomes a roof with some glass in it.

What we did instead

To our knowledge, the first cable-supported flat skylight built in Iran.

Instead of making the members deep enough to resist bending, we removed most of the bending from the members.

Pre-tensioned cables and tie rods run below the roof plane, held off the steel by struts. They pull upward against the load. The steel above them works predominantly in compression and tension along its length rather than in bending across its depth, and the two together behave as a structure far deeper than the member is.

The structural depth becomes the distance between the steel and the cable — not the depth of the steel. A slim member held off a cable can be as deep, structurally, as a heavy truss, and visually it is almost nothing: a line of steel, a strut, and a cable you can see past.

Diagram The same structural depth, in steel and in air

Depth made of steel A truss gets its stiffness from two chords held apart by a web. All of that depth is something to look at, and from a courtyard floor, seen at an angle, the webs close up.
Depth made of air A slim member, a strut and a cable give the same lever arm with almost nothing in it. What crosses the sky is one line of steel and a tension member thin enough to see past.

Sections, and a strip of each roof seen from below. Schematic: neither is drawn to the scheme's own sizes.

That assembly has a name — a cable truss — and a roof built from it is a cable-supported roof structure rather than a framed one.

The Fanap roof steel complete before glazing, the tie rods of the cable truss picked out against clear sky — CORTEX project The roof steel before glazing: a slim grid, a strut under each node, and the tie rods of the cable truss drawn across the sky.

Two pieces of physics make it work, and they compound.

A cable works in pure tension, so all of it works. In a beam carrying bending, stress is highest at the top and bottom faces and falls to zero at the neutral axis — most of the section is along for the ride. A cable in tension carries the same stress through every strand. Nothing is idle.

Diagram Why a cable uses all of its steel and a beam does not

Beam in bending Only the top and bottom fibres reach full stress. The middle of the section is carried along as weight.
Cable in tension Every wire carries the same stress, so none of the section is idle — and the wire itself is several times stronger than structural steel.

And cable steel is drawn to several times the strength of structural steel. A structural section is designed at a few hundred megapascals. Cable is manufactured as high-tensile wire, laid into stainless spiral strand of 12 to 36 mm on the systems we build, and works at a multiple of that figure. Higher strength and full use of the section, together, are why a strand a few centimetres across can do the work of a member many times its depth.

The result over the Fanap courtyards is a flat roof at the size the architecture asked for, with members slim enough that what you see from the courtyard floor is the sky.

The shadow of the Fanap cable truss roof on a courtyard wall — the steel grid, the struts and the tie rods drawn in sunlight — Still from the project film The roof drawn by the sun on a courtyard wall: the grid, the struts and the tie rods. Everything between the lines is daylight.

Project film · 1 min 44 s The project on film: the steel and its tie rods going up over the open courtyards, the glass being laid, and the finished roofs from the air and from the courtyard floor.

Three roofs, not one

The 20 × 20 and the two 26 × 17 are three separate structures, and that matters more than it sounds.

Diagram Three roofs, three structures, one frame

Three roofs, side by side One square and two rectangles, each over its own courtyard, with a wing of the building between one and the next.
Three structures, one frame Each roof is tensioned on its own and closes its own forces. What it cannot close lands on the frame — and the wings in the middle take a roof on either side.

Plan: the roofs are to scale with each other, the building outline is not. Section: schematic.

Each is its own self-equilibrating assembly: its own cable layout, its own pre-tension, its own anchorage. A cable system does not share its forces with its neighbours — it closes its own loop or it pushes what it cannot close into whatever it is attached to.

So the project carried three tensioning operations rather than one, three sets of geometry to verify, and three sets of reactions arriving into the same primary frame. The frame sees all three simultaneously, in every combination of load, which is a different analysis from checking each roof on its own.

The Fanap building from directly above during construction, the roof steel in place over each of the three courtyards — CORTEX project The building from directly above during construction, with the roof steel in place over each of the three courtyards.

Where a cable roof's forces go is the question that decides the project, and on this building it was asked at the start rather than discovered during erection.

Pre-tension, and why it is a sequence rather than a number

A cable can only pull. It has no stiffness in compression and no bending stiffness at all, which is exactly why it is efficient — and it is also the thing that has to be designed around.

Pre-tension is what keeps a cable working. The system starts with tension already in it, enough that no cable goes slack under any load case the roof will see — including wind uplift, which on a light flat roof is a suction that tries to lift it and take the tension out of its cables. A slack cable contributes nothing, and the moment one goes slack the steel above it is carrying the load alone, at a size never designed for that.

Diagram What pre-tension has to survive

At rest The cables carry their pre-tension and nothing else. The roof is already stiff before any load arrives.
Under snow The roof is pushed down, the cables stretch and their tension climbs. It has to stay inside what the strand and its fittings are designed for.
Under wind uplift Suction lifts a light roof and takes tension out of the cables. It must never reach zero: a slack cable has left the structure.

Deflections exaggerated. The gauge has no scale: the figures are each project's own.

Pre-tension also supplies stiffness. A cable with no tension in it has almost none until it has pulled straight; a pre-tensioned one is stiff from the first increment of load. The roof's deflection behaviour is a property of the pre-tension as much as of the sections.

And it cannot be applied cable by cable as an independent operation. Tensioning one cable changes the force in every other cable in the system and moves the geometry of the whole assembly. So there is a tensioning sequence — computed in advance, carried out in order, with the geometry surveyed as it proceeds — and the structure only arrives at its design shape at the end of it, not at any point during.

A Fanap courtyard roof under erection — the steel grid in place, a strut under each node and the tie rods between them — CORTEX project A courtyard roof under erection: the steel grid in place, a strut under each node and the tie rods threaded between them.

That is the part of a cable project that cannot be compressed in the programme, and it is the part that separates a cable roof that performs as calculated from one that merely stands up. How the stages are planned is drawn under the tensioning sequence on the system page.

The temperature problem nobody expects

Pre-tension is set by stretching a cable a known amount. Steel changes length with temperature. So the tension in the system depends on how warm it is.

A 26-metre run of steel through a 50-degree annual swing — a Tehran roof sees that comfortably between a winter night and a summer afternoon — changes length by about 16 millimetres on thermal movement alone, and by nearer 21 in the stainless steel a strand is made of. In a system whose working tension is set by a few tens of millimetres of stretch, that is not a rounding error.

What counts is the difference. The steel a cable is anchored to moves as well, and where the two warm together most of the movement cancels. It never cancels completely: the cable is a different steel from the frame, it is far thinner, and under a glass roof it sits in the sun while the structure around it lags behind.

So pre-tension is not one number. It is a range, checked at both ends — at the warm end no cable has gone slack, at the cold end none is overstressed — across every temperature the roof will see for the life of the building, and set, on the day, against the temperature the structure is actually at.

Diagram Pre-tension is a window, not a number

Schematic. The slope and both limits come out of the analysis of a given roof.

The roof is designed to move

The deflection criterion on this system is L/100 to L/200 under snow. On a 20-metre span that is between 100 and 200 millimetres; taken on 26 metres, between 130 and 260.

What L/100 to L/200 comes to at the dimensions of these roofs mm under snow

17 m
20 m
26 m

Arithmetic on the published criterion, not a measurement of these roofs.

Up to a quarter of a metre of movement, in a glass roof, by design.

That single figure reorganises three other things.

The glazing has to stay sealed through it. Every pane, every gasket and every joint accommodates the movement of the structure under it rather than resisting it. Glass has no capacity to follow a frame that moves more than its fixings allow — it simply breaks at the corner where the load concentrated.

The drainage falls are set on the deflected shape, not the built one. A flat roof that deflects creates a low point exactly where it is already deflecting most, water runs to that low point, and the water is additional weight concentrated there. On a stiff roof that loop converges. On a flexible one, or one whose outlets cannot keep up, it does not. Falls are designed for the loaded shape — which means the roof as built is not level, it is cambered against the deflection it is going to have. The loop itself is drawn under ponding on the system page.

A Fanap roof part-glazed, finished panels behind the working edge and a glass lifter on the crane overhead — CORTEX project A roof part-glazed: finished panels behind the working edge, and the lifter for the next one on the crane.

And the junction with everything around it has to take the difference, which on this building is the hardest detail on the project.

The ninety-degree junction

Each skylight meets the curtain wall at a right angle on two sides, and that curtain wall runs down, continuous, to the ground floor.

A Fanap courtyard roof meeting the curtain wall at ninety degrees, the wall continuing down past the galleries to the ground floor — Still from the project film A courtyard roof meeting the curtain wall at its end. The wall carries on down past every gallery to the ground floor.

Three separate problems meet on that line.

The two systems move differently and by different amounts. The roof deflects under snow and recovers. The curtain wall moves under wind, under thermal expansion and with the frame it is fixed to. Neither movement has anything to do with the other, and the joint between them has to absorb the sum of both, in every direction, while staying watertight.

The roof's reactions cannot land on the wall. A cable roof arrives at its edge with the weight of the roof and everything on it, and with whatever part of the cable force the roof steel does not close within itself. A curtain wall is a non-loadbearing skin; it carries its own weight and the wind on its own face, and nothing else. The skylight reactions therefore pass the wall line and go into the primary structure behind it, which means a structural separation at exactly the point where the architecture wants a continuous glass line.

And it is the drainage line. A flat roof drains to its perimeter, and the perimeter here is the head of a glazed wall that runs the full height of the building. Water that gets past the junction has a long way to travel and plenty to find on the way.

Diagram One line, three problems

1 · Two movements, one joint The roof goes down under snow and comes back. The wall moves with wind, temperature and the frame. The joint between them takes the sum, in every direction, and stays sealed.
2 · One load path, and it is not the wall The roof lands on the building's structure behind the glass line. The curtain wall carries its own weight and the wind on its own face, and none of the roof.
3 · The drainage line A flat roof drains to its edge, and this edge is the head of a glazed wall that runs down to the ground floor.

Section, schematic: the principle, not the project's own detail.

The resolution is that the glass line is continuous and the structure behind it is not: the roof lands on the frame, the wall is carried on its own supports, and between them sits a joint engineered to move and to drain rather than to hold still.

Glazing being installed at the edge of a Fanap courtyard roof from a platform over the open courtyard, the roof steel landing on its own column — CORTEX project The same edge during construction: the roof steel landing on its own column, and glazing going in from a platform over the open courtyard.

The wall is a curtain wall façade, and the movements a wall of that kind has to take are set out on its own page.

Eleven thousand machined components

Roughly eleven thousand turned parts were fabricated to complete the cable systems — about nine components for every square metre of skylight.

Machined components: ≈ 11,000
Glass roof, three skylights: ≈ 1,280 m²
Components per square metre: ≈ 9
Best Special Steel Structure in Iran: 2022

That number sounds like a quantity. It is actually a tolerance argument.

In a cable structure the geometry is the sum of the lengths of its components. A cable run is not a single object: it is a fork, a swaged socket, a length of strand, a tensioner, another socket, another fork, and a pin at each end — and the position of the node it arrives at is the accumulation of every one of those dimensions.

Diagram Why the parts are machined

One run is a chain of parts The distance between the two nodes a run joins is every length in the chain added together, and that distance is what sets the force in it.
In series, errors add Parts that are each slightly long push every node further from where the model put it. On a pre-tensioned roof a geometry that does not close means the force is wrong as well.

Schematic. The overrun is exaggerated.

Get each fitting a millimetre long and the error does not average out, because they are in series. Across a roof with fittings at every node, a consistent small error becomes a geometry that does not close — and in a pre-tensioned system a geometry that does not close means the tension is wrong everywhere, because the force in a cable is set by how far it has been stretched.

So the components are machined rather than fabricated. Turned to a dimension on a lathe, measured, and made in families so that identical parts are genuinely identical. That is the difference between a cable roof that reaches its design geometry at the end of the tensioning sequence and one that is adjusted on site until it looks approximately right.

The end fittings, the tensioners, the strut heads and feet, the cable clamps at every crossing, the pins and bushes at every connection, and the glass fittings are all in that count.

The Fanap cable truss skylight from below — slim steel held against tie rods in tension, the sky visible through the grid — CORTEX project The finished roof at close range: fork ends on the tie rods, the fittings at the foot of each strut, the point fixings under the glass.

What this project demonstrates

Four things, and the first one generalises past this building:

On any long span skylight, deflection governs and depth is the price. Bending moment grows with the square of the span and deflection with the fourth power — so a flat roof that spans far enough will always end up deep, unless the load is taken out of bending altogether.
A cable truss buys structural depth without visual depth. The depth becomes the distance between the steel and the cable below it, so the member can be slim while the assembly is deep. On a skylight that is the entire point, because the structure is what people look at from underneath for the life of the building.
A pre-tensioned system is commissioned, not just installed. The tensioning sequence, surveyed as it proceeds, is what brings the structure to its design geometry — and the temperature on the day is part of the calculation.
The junction is harder than the span. A glass roof meeting a glass wall at ninety degrees, with two systems moving independently and the roof's reactions passing through to the frame behind, is where this project's engineering actually sits.

CORTEX's package on the building was the envelope, about 4,000 m²: the three skylights and the curtain wall they meet. The architect was Boompad Consulting Engineers, and the main contractor was Paya Takht-e Jamshid Holding, the group CORTEX belongs to.

The project was recognised with Best Special Steel Structure in Iran, 2022.

Where this system sits among every other kind of skylight, sorted by span, is in Skylight Types Explained.

Project gallery

Design visualisation of the Fanap building from outside, the flat glass roofs meeting the curtain walls at the ends of the courtyards

The Fanap campus from the air, the completed precast-clad block in front of the wider district and the hills beyond

The Fanap building from the air within the surrounding city fabric, a glazed courtyard end between the clad blocks

The Fanap glass roofs at closer range, the glazed bays running to the parapet with plant alongside

The Fanap campus under construction, tower cranes over the rising blocks

Design visualisation of a Fanap courtyard beneath its flat glass roof, galleries on every level

Systems used

Questions

Common questions

How far can a long span skylight go on a flat roof?

Further than most people expect, but not by making the beams bigger. On the systems we build, a cable-supported flat skylight spans 10 to 45 metres — at Fanap, one roof of 20 × 20 m and two of 26 × 17 m. The reason the span is not simply a matter of deeper steel is that deflection rather than strength governs a glass roof, and deflection rises with the fourth power of the span: a 26-metre member deflects around forty-six times as much as a 10-metre one of the same section. Holding that back by depth alone would need roughly three and a half times the structural depth, which on a skylight is sky the occupant no longer sees.
What is a cable truss, and why use one in a roof?

A cable truss is a structural assembly in which a slim compression member is held against pre-tensioned cables, usually spread apart by struts, so that the pair works as a single deep element. The load is carried by the steel in compression and the cable in tension rather than by any member in bending. Its usefulness on a skylight is that the structural depth becomes the distance between the steel and the cable, not the depth of the steel itself — so the member can be slender while the assembly is deep. From underneath you see a line of steel, a strut and a cable rather than a truss.
Why are cables so much more efficient than steel beams?

Two reasons that compound. A cable works in pure tension, so every strand in it carries the same stress — where in a beam carrying bending the stress is highest at the top and bottom faces and falls to zero at the neutral axis, so most of the section does very little. And cable steel is drawn as high-tensile wire to several times the design strength of a structural section. Full use of the material and a far higher working stress, together, are why a stainless spiral strand of 12 to 36 millimetres can do the work of a member many times its own depth.
What does pre-tensioning a cable roof actually do?

Three things. It keeps every cable in tension under every load case, including wind uplift on a light flat roof, because a slack cable contributes nothing and leaves the steel above it carrying load at a size it was never designed for. It supplies stiffness, since a cable with no tension in it has almost none until it has pulled straight, so the roof's deflection behaviour is a property of the pre-tension as much as of the sections. And it sets the geometry — a pre-tensioned structure is only in its designed shape when the designed tension is in it.
Why is there a tensioning sequence?

Because tensioning one cable changes the force in every other cable in the system and moves the geometry of the whole assembly. There is no order-independent way to do it. The sequence is computed in advance, carried out in order, and the geometry is surveyed as it proceeds, so the structure arrives at its design shape at the end of the operation rather than at any point during it. It is the part of a cable project that cannot be compressed in the programme, and it is what separates a roof that performs as calculated from one that merely stands up.
Does temperature affect a pre-tensioned cable roof?

Yes, and it is the effect most often underestimated. Pre-tension is created by stretching a cable a known amount, and steel changes length with temperature — so the tension in the system varies with how warm it is. A 26-metre run of steel through a 50-degree annual swing changes length by around 16 millimetres on thermal movement alone, in a system whose working tension is set by a few tens of millimetres of stretch. The cable and the structure it is anchored to move together only in part, so pre-tension is specified as a range checked at both extremes — nothing slack at the warm end, nothing overstressed at the cold end — and set on site against the temperature the structure is actually at.
How much does a cable-supported glass roof deflect?

Considerably more than people expect, and deliberately. The criterion on this system is L/100 to L/200 under snow, which on a 20-metre span is between 100 and 200 millimetres, and on 26 metres between 130 and 260. The glazing, the gaskets and every joint are detailed to follow that movement rather than resist it, because glass has no capacity to follow a frame that moves more than its fixings allow. The drainage falls are also set on the deflected shape rather than the built one, so the roof as constructed is cambered against the deflection it is going to have.
Can you put a large skylight in a flat roof?

Yes, and the question is what carries it rather than whether it is possible. Below roughly eight metres a framed unit spans on its own. Beyond that, a flat roof needs a structure, and because deflection grows with the fourth power of span, that structure gets deep fast — which on a skylight defeats the purpose. Cable reinforcement is the usual answer where the roof must stay flat: it takes the load out of bending so the members stay slim. Where the architecture will accept curvature instead, a curved surface achieves the same thing without pre-tension or a tensioning sequence.
How does a glass roof meet a curtain wall?

With a joint engineered to move and to drain, and a structural separation behind it. The two systems move independently and by different amounts — the roof deflects under snow and recovers, while the wall moves under wind, thermal expansion and inter-storey drift — so the junction has to absorb the sum of both while staying watertight. The roof's reactions, which on a cable roof can include a horizontal pull at the anchorages, cannot be carried by the curtain wall, because a curtain wall is a non-loadbearing skin. They pass the wall line into the primary structure behind it. And the same line is where the roof drains, which on a wall running several storeys below gives any water that gets past a long way to travel.
Why were eleven thousand components machined rather than fabricated?

Because in a cable structure the geometry is the sum of the lengths of its components. A cable run is a fork, a swaged socket, a length of strand, a tensioner, another socket, another fork and a pin at each end, and the position of the node it reaches is the accumulation of every one of those dimensions. Those errors are in series, so they do not average out: a consistent millimetre of error at every fitting across a roof produces a geometry that does not close, and in a pre-tensioned system that means the tension is wrong everywhere, because force in a cable is set by how far it has been stretched. Turning the parts to dimension is what makes the assembly arrive where it was calculated to arrive.