If Elon Musk’s Starship works – and there’s no reason it shouldn’t – it will expand access to space to a degree that will seem ludicrous. At present, getting into low earth orbit costs in the neighborhood of a thousand dollars per kilogram. Starship could lower that to tens of dollars per kilogram. At present, the number of launches to orbit is in the double digits per year, with the larger launchers delivering on the order of ten tons. Starship, with both stages being fully reusable, could do more flights even just using a single spacecraft; and it is so large that each flight could deliver on the order of a hundred tons. And Elon is building not just one Starship or four Starships but a factory for Starships; we saw something of its production rate when Starships were being crashed on what seemed like a monthly basis during landing tests. Multiply all those factors together, and we’re talking perhaps a thousandfold increase in launch capacity.
But what is to be done with all that capacity? Elon has been saying that the idea is to colonize Mars, but how that would pay off is not clear. Mars has a mere wisp of an atmosphere (and that mostly CO2) and very little water. The soil is actively hostile to life (that being one of the few things wrong with the movie The Martian: he couldn’t have grown his crops). Living there would have to be done inside a pressurized bubble, and would require most supplies to be transported from Earth, at least until a very substantial local economy had emerged. Even living in the most hostile desert on Earth would be far easier. For a settlement to be viable, it would have to export something to pay for all those imports – and no, Halliburton isn’t going to drill for oil there, as some clowns were imagining when Bush Jr proclaimed his Mars initiative: there probably isn’t oil on Mars, and even if there were lakes of oil on Mars, it isn’t valuable enough to pay the return freight, even at Starship rates.
But there is a different Mars that pays very well: Mars the god of war – though he can be fickle.
In particular, that sort of lift capacity would make constructing a national missile defense entirely practical.
There were a lot of arguments against Reagan’s SDI initiative, but many were intellectually puerile. Among the arguments from experts, the computer scientists’ was perhaps the worst: they proclaimed that such a system would require a million lines of code (or something like that), and thus could never be debugged properly and would be doomed. Of course that number was made up; it wasn’t like they even came close to an accounting of what all those lines of code would be doing. The Apollo moon landing was done with a computer with 36,864 words of program memory, each word being 16 bits, and was a vaguely similar task, even involving radar-guided rendezvous of two spacecraft (albeit not head-on at high closing rates).
Still, with today’s software practices (which are more lavish of lines of code than they were even in the 1980s, let alone the 1960s), it would be probable that far more than a million lines of code would be used in a missile defense system, if one were to count every single line of code that the system used. Probably lots of computers in the system would be running the Linux kernel, and that alone has 25 million lines of code (at last count, per cloc, counting only C and assembler code and not counting blank lines or comments). Now, the bulk of those 25 million lines is in device drivers (which aren’t used if you don’t have the devices they drive), so the actual number of lines of code that are used in any one computer is quite a bit smaller. Still, even the core “kernel” and “mm” (memory management) subdirectories total to about 400,000 lines of code. Adding in device drivers, filesystems, and such, the count of lines of code actually used in one computer could easily get to a million – and that just for the Linux kernel.
Yet despite using something like a million lines of code, the Linux kernel works. Yes, there are bugs lurking in it – more are continually found – but lurking is normally all they do; they only manifest themselves in rare circumstances. Similarly, it is commonly said that any serious compiler has bugs; yet the same people who say that will tell programmers that when there’s a problem with their program and they suspect it’s a compiler bug, it isn’t: it’s their bug. And there is nothing inconsistent in this: the probability of hitting a compiler bug is nonzero, but it’s vastly smaller than the probability of you making a mistake.
War doesn’t require perfection. There never was a weapon system that didn’t have bugs. Even sticks and stones can fail due to hidden cracks. Yes, a bug in a missile defense system can mean a city getting vaporized; but not even trying means they all get vaporized.
It wasn’t as if the objecting computer scientists actually used the tools of their profession, such as proofs of complexity. They were just throwing up their hands and declaring it too hard – which for them it might well have been: to be a professor of computer science one doesn’t actually have to be good at writing programs, especially not large-scale programs. But there are people in the world who are good at it. And the problem of missile defense is not conceptually difficult: it’s basically see missile, shoot down missile. Even if considered in the widest sense, where you have a bunch of interceptors in various locations and want to assign each one to an incoming missile in a way that maximizes the total amount of interceptions done, it is not a particularly hard problem. If you wanted to run that algorithm in a distributed fashion with extreme fault tolerance, as in the Byzantine Generals Problem – okay, then you’ve got a problem difficult enough that a computer scientist could be proud of solving it. But we don’t demand that level of fault tolerance from any other military system: it’s enough that it works when commanded and that its communications links are resilient to attack. The system for launching nuclear weapons is engineered to that sort of standard; a system for shooting them down can be too. It might be good for either or both of those systems to be engineered to an even higher standard, but it would be an unprecedented advance.
This doesn’t mean the task of programming the system, considered in its entirety, would be easy: details of, for instance, a radar algorithm, can be quite gnarly. But radars can be tested thoroughly without having to fight a nuclear war to do so. Much of the code to be used can also be proven in other applications – or already has been, as in the case of the Linux kernel. The part of the code that could only be tested in an actual nuclear war would not be a complicated part.
One would still want to test that part, which would be done by rigging up a simulation of war: modeling the outside world to calculate what the missile defense system would see as its input, and checking that the code did the right thing when presented with that input. Such a simulation would probably require more coding than the warfighting code would, but it’d be worth it. It would even be best to have more than one such simulation, with teams competing against each other to find bugs. The question is just whether the whole thing would be so complicated that bugs in the simulation code (or codes) would interact with bugs in the warfighting code to produce an illusion of success: that’s the sense in which the level of complexity matters.
Physics-based arguments against SDI were often more reasonable, and in some cases (as against X-ray lasers) they seem sound: X-ray lasers would require detonating a nuclear weapon in space, something that would be quite damaging to satellites and would generate a destructive electromagnetic pulse on the Earth’s surface, and thus would be impractical to test in peacetime. An X-ray laser could be tested underground, but it’d be hard to test its ability to point at a target – a task that demands precision that it might not be able to deliver. Even in a real nuclear attack scenario, people would hesitate to use them due to all the collateral damage, and might hesitate too long. But the X-ray laser was never a necessary part of SDI: it seems to have been thrown in because some of the people involved in promoting SDI (such as Edward Teller) were from a nuclear weapons background, and that was their hammer in search of a nail. For the most part the interception of missiles is a matter of engineering rather than anything involving new physics.
Other arguments against SDI involved countermeasures: for instance warheads can eject balloon decoys, making it hard to tell which is the real warhead. Balloons are cheap, lightweight, and easy to deploy. But they only work only in space: even in very thin air they slow down much quicker than the real warhead. So a terminal defense (one protecting just a target city, and operating in the atmosphere) is not bothered by them. Even in space, it might be possible to precede an interceptor with a cloud of gas to distinguish between them, say by briefly firing one of its rocket engines forwards towards the targets and decoys: upon hitting that gas cloud the decoys would slow more. Hitting them with a brief high-power laser pulse could produce a similar effect. Heavier decoys can defeat those methods of discrimination, but their weight comes at a high cost because it’s in the final stage of the rocket: every increase in weight in the final stage means about a 20-fold increase in liftoff weight.
Another countermeasure missiles can use is to maneuver. The result is what currently is overhyped as “hypersonic missiles”, which as fielded today essentially means a ballistic missile that can do a bit of maneuvering (ballistic missiles always having been hypersonic). There are more sophisticated “hypersonic” technologies out there, such as scramjets – ramjets that burn fuel as it passes through supersonically – but they are research projects and have been for decades; it’s doubtful if they will ever become cost-effective compared to rocket propulsion.
Now, maneuvering a ballistic missile can be quite a useful military capability when coupled with a system to detect and home in on a moving target such as an aircraft carrier. But the idea that maneuvering makes a missile unstoppable is one of Putin’s sillier boasts. Surface-to-air missiles have been shooting down maneuvering aircraft ever since they were invented. There is a notion out there that an interceptor missile has to be faster than the missile it’s intercepting, but it seems entirely groundless: missile defense isn’t a tail-chase scenario, it’s a head-on meeting. The actual rule for countering maneuvering missiles is that the interceptor missile has to be able to pull more Gs than the missile it’s intercepting, which can be done even if it’s slower. (If an interceptor wants to start from the ground and meet a ballistic missile midway, as in the case of the present interceptors stationed in Alaska, then yes, it does have to be about as fast; but that’s not the only way to do things.)
Even if a missile could maneuver more strongly than the interceptor, it would have to maneuver at the right time to dodge the interceptor. It almost certainly couldn’t be maneuvering strongly for its whole flight, since maneuvering is expensive. In space, maneuvering requires a rocket engine and rocket fuel; it’s impractical to keep that rocket firing the whole time the missile is on its way. In air, hypersonic lift-to-drag ratios are miserable, maxing out at about 5 or so, meaning that pulling 5 Gs to avoid an interceptor would slow it at 1 G; a missile that does that all the time will quickly cease being hypersonic. (And a real-world missile might easily have only half that lift-to-drag ratio.)
Dodging is a well-known game in the context of airplanes and surface-to-air missiles: fighter pilots can very often see the missile and maneuver at just the right time for it to miss – an illustration of how being faster doesn’t necessarily mean an ability to pull more Gs: the missile is faster, but the airplane has much larger wings and can in many cases maneuver more sharply. Still, evading in that fashion saps the plane’s energy and may leave it without enough to evade the next missile.
And it relies on seeing the missile. In air, hypersonic missiles have difficulty seeing anything in front of them, due to the heat produced at hypersonic speeds. The Space Shuttle had front-facing windows, but during reentry was pitched up so those windows were in the lee of the vehicle. The SR-71, cruising at Mach 3, had the outer temperature of its windscreen reach 600°F; accordingly, its windscreen was made of quartz, an optical material which can stand high temperatures. (They used it again for the outer layer of the Shuttle’s windows.) Hypersonic speeds are usually defined to start at Mach 5; and the temperature rise scales as the square of speed; scaling that SR-71 temperature to Mach 5 yields about 1700°F. ICBMs move at more than Mach 20, and the short-range Russian Iskander ballistic missile (much used against Ukraine, and said to have some maneuvering capability) hits Mach 6-7 (corresponding temperatures: 2400°F-3300°F). Against such temperatures, even protecting parts of the warhead that don’t have to be transparent is no easy task; it often is accomplished using ablatives (materials that burn and vaporize away in the heat but do so slowly enough that the missile can complete its mission). An optical window wouldn’t just have to survive at such temperatures but to preserve its optical clarity: it couldn’t even have surface deterioration. (Diamond oxidizes in hot air, so would be unsuitable.)
This doesn’t mean that a hypersonic missile can never see anything – indeed, the Iskander is said to have an optical seeker to guide it to its target. But it does mean that the seeker can only operate after the missile has slowed down to more ordinary supersonic speeds (as such missiles naturally do as they near their target, due to aerodynamic drag being ferocious at low altitudes). It can be hypersonic, and it can see to dodge, but not at the same time – at least not in atmosphere. In space, even the fastest missile can see fine; but an interceptor in space could be made unusually visually stealthy, through the use of mirrors.
A hypersonic missile in air could still use radar to detect interceptors, at least until about Mach 10, where the surrounding air becomes so hot that it turns into a plasma and becomes electrically conductive, trapping the radar waves inside. But radar can be jammed, to name just one of many electronic warfare methods. (And no, contrary to Russian propaganda, the plasma sheath won’t hide the missile: the missile is visible to radar because it’s conductive, so surrounding it with a conductive plasma sheath is not something that could hide it.)
That is just a taste of the games of countermeasures and counter-countermeasures that can be played; but it is perhaps a sufficient taste to show that they aren’t all that different from the games of countermeasures and counter-countermeasures that have been going on forever in other parts of war. Why be so pessimistic in this case?
There actually is a decent reason, though I’ve never seen it stated explicitly: nuclear war is thought of as one huge exchange of missiles, each side trying for immediate and complete destruction of the opponent – as Herman Kahn put it, a “wargasm”. Whether or not that is really the best way to fight a nuclear war, and whether or not it is the way a nuclear war would actually end up being fought, any missile defense should work in that scenario.
In past wars, people had time to learn how enemy weapons worked and develop and mass-produce countermeasures; there’s no time for that in a missile exchange that lasts 30 minutes. Yes, espionage can sometimes figure out what the enemy is doing; but it can never figure it out completely. Yes, enemy test launches can be monitored to see what sort of capabilities they are testing; but such monitoring can never give a complete picture either. (A recent trend is to label such gaps in knowledge “intelligence failures”, as if one could fix them completely and definitively with proper spying; but really espionage will always be a cat-and-mouse game: great if you can catch the mouse, but never to be relied on absolutely.) So, unlike in previous wars, a missile defense system has to work not just against the countermeasures the enemy is actually fielding but against the ones they might possibly be fielding. That can be done, but it’s expensive.
It’s doable against North Korea, an economic wimp of a country; that’s largely what those interceptors in Alaska are about. But it relies on outspending them tenfold or more, something that would be impractical against Russia or especially China. Each interceptor has a cost similar to an ICBM: it is a rocket of about the same size as an ICBM and its payload requires similar levels of engineering. And interceptors are not perfectly reliable, the result being practices such as shooting two of them at each incoming missile to increase the chances of interception, doubling the cost.
But with reusable launchers, the economics change completely. One reusable launcher can make a hundred flights and preposition a hundred interceptors in orbit. You’ve bought one rocket; they’ve bought one rocket; but they only have one ICBM, while you have a hundred interceptors for that ICBM. (This is in contrast to the situation with single-use launchers, where prepositioning an interceptor in orbit makes no sense: better to keep it on the ground mated to its launcher and ready to go; it’s safer there and can readily be inspected and maintained.) You do still have to buy the hundred interceptors, but that’s where it helps to have a launcher that’s not just reusable but delivers low cost per ton to orbit. With today’s launch prices, costs of space hardware are hugely increased by the need to mass-reduce everything to the extreme; a cheap launcher removes that need.
It might not even be necessary to preposition interceptors in orbit: it might be enough to keep sensors and communications links in orbit and leave the actual interception to be done by short-range interceptors positioned near the potential targets (such as cities and military installations). Short-range interceptors are smaller and cheaper than long-range ones; and tracking by an orbital system of sensors can give enough advance notice for even a relatively slow interceptor to make it to an intercept point.
Work on that sort of sensor system has in fact already begun, with SpaceX (among others) having being hired to provide a Starlink-like world-girdling array of satellites with sensors for the purpose of countering “hypersonic missiles”. Those contracts are on a trial scale (hundreds of millions of dollars rather than billions), and intended for local defense rather than national defense. But translating that sort of work into a national missile defense system wouldn’t require any huge leaps of faith; it just would be a matter of putting more sensors in orbit, data-linking them together more thoroughly, and upgrading interceptors to handle faster threats.
That upgrading also is a process that’s been going on for a while: in the first Gulf War they boasted about the interception record of the Patriot air-defense missile against Iraqi Scud missiles, but internally must have been honest enough about its real performance to demand upgrades to do better, resulting in the development and fielding of the Patriot PAC-3, rated for intercepting ballistic missiles.
But despite the theoretical possibility of doing interception only at the last stage, the best security would come from a layered defense, which would try to intercept missiles at every stage. Indeed, in the presence of an enemy who was thought to be capable of taking out the orbital part of the system, interception might need to be done earlier to protect the orbiting hardware. Still, buildout of such interceptors could be done as the threat emerged.
In general, the system could evolve like any other military system, with overconfidence, budget cuts, and slowdowns under some administrations, and scares, budget increases, and speedups under others, all to a backdrop of constant squabbling about which parts of the system were worthwhile. At no stage of building a national missile defense system would there need to be any bombastic talk about “rendering nuclear weapons impotent and obsolete” – something that isn’t going to happen: delivery modes may change, but not the usefulness of nuclear weapons themselves. Anyone who renounces them will just be prey for those who don’t. What can be changed is the system of mutual assured destruction, in which both sides in a nuclear war are guaranteed to lose.
There is a school of thought that says this system is good, since it has prevented war, even conventional war, between superpowers. Proxy wars (such as in Vietnam or Ukraine) have not been prevented; but direct war has. But that deterrent effect depends on a basic level of sanity on both sides, and erodes it in the process. Under mutually assured destruction a madman has to be cooperated with, since the alternative is worse; this provides an incentive for madness. It’s a slippery slope: a country makes an arrogant move, gets away with it due to possessing nuclear weapons, and starts to think that such arrogance is right and proper. Power corrupts, and nuclear superpower corrupts quite badly. (And no, it’s not just Putin that it’s been happening to, nor even just the Russians.)
To recoil against the whole enterprise of scientific knowledge that brought the world nuclear weapons is a natural human reaction; the ancient Greeks wrote of such reactions in their legend of Pandora’s Box. Yet the legend carried a warning: when Pandora slammed shut the lid of the box, the evils had already escaped and she trapped only hope inside. Today, such reactions could trap the hope of defending against nuclear-tipped missiles: they could keep it mere theory. But in Starship, one of the larger steps toward reducing it to practice is taking place.
It’s a long-anticipated step. Though the vast majority of the liftoff weight of an orbital launcher is fuel, the cost of fuel has never been even close to being the main cost driver: that’s been the cost of throwing away the rocket after each launch. The Space Shuttle was once supposed to fix this, but although NASA started with the idea of two reusable flyback stages, by being penny-wise and pound-foolish they ended up with a launcher where the biggest piece of structure (the external tank) was thrown away, and where the orbiter and solid rocket boosters needed weeks of work after each flight to get them back into condition to fly again – and thus where per-launch costs were actually higher than for expendable launchers.
The SDI people tried again: they knew they needed cheap access to space. They contracted with McDonnell Douglas to build the DC-X, a prototype for a single-stage-to-orbit reusable launcher – itself nowhere near able to get to orbit, but intended to try out the necessary technologies to do so. Using just one stage to get to orbit is a lot less efficient than using two or three stages: from the same size rocket, you get a lot less payload. But the thought was that reusability conferred such a big advantage that this wouldn’t matter, and that using just a single stage would be simpler. But when DC-X crashed and burned, it was not rebuilt, even though it was just an ordinary accident, not an indication of any fundamental problem; the impetus for SDI was already dying with the end of the Cold War.
A bit later, NASA started their own program for such a vehicle, contracting with Lockheed Martin for the X-33 (again just a suborbital technology demonstrator). They ended up demonstrating that the technology was harder then they’d thought: they couldn’t get the X-33’s liquid hydrogen tank to work. Ordinary liquid hydrogen tanks are quite doable, but the X-33’s tank was to be made in a complicated shape to fit into the stubby-winged spacecraft, and had to be extremely lightweight; they chose to make it of two layers of carbon fiber composite with a honeycomb structure between them. They had a problem with it cracking when tested down to liquid hydrogen temperatures, and gave up.
In general, the problem with single-stage-to-orbit is that the apparent simplicity of using just one stage comes at the cost of a huge struggle to minimize weight. Still, it was one of the more sensible of what might be called the clickbait ideas of the space enthusiast world: ideas that were not actually advertised under headlines such as “This one weird trick will get you to orbit”, but had something of that spirit.
There have been a lot of such clickbait ideas, but SpaceX doesn’t seem to have really fallen for any of them: their rockets have basically been conventional. They’ve been two-stage. They were initially expendable, with reusability phased in later. There was no dabbling in pressure-fed rockets or novel sorts of pumps: just turbopumps as per the industry norm. They haven’t gone for unconventional nozzles (such as aerospikes, which the X-33 was to use), but have stuck with the usual bell-shaped nozzles. (By moving to higher chamber pressures, they’re getting most of the altitude-compensation benefits that aerospikes promise.) The propellants they first chose were liquid oxygen and RP-1 (a high-grade kerosene), bog-standard in the business; and now they are moving to methane, still an everyday cheap fuel without the excitement (or hassles or expense) of liquid hydrogen. The basic structure of Starship is stainless steel, nothing exotic. Reentry protection is heat-resistant tiles, as in the Space Shuttle, though not the super-delicate ultra-high-performance tiles the Space Shuttle used. (The underlying stainless steel can take more heat than the Shuttle’s aluminum structure could; and as basically a big empty tank during reentry, Starship will not be nearly as dense as the Shuttle and thus will do more of its deceleration higher up in the atmosphere where heating loads are lighter.) So while they’ve innovated, it hasn’t been any huge dramatic innovation that a theorist would be proud of (and that might not actually work out in practice): they’ve been taking conventional approaches and evolving them to be more economical. Grabbing the returning spacecraft with Mechazilla will visually be quite a different spectacle from their previous landings (or anyone else’s previous landings), but in engineering terms is a modest step to eliminate the weight of the landing gear.
So there’s no reason to expect failure from Starship, though we can expect more entertaining kabooms on the way to success.
And then to, uh, Mars.