MODULE 1 · LESSON 3
Free — no login requiredSign in to track progress, save quiz attempts and enrol in the full course.
Sign in to track progress / enrolMeasurement, and the Catch
The previous lesson ended on the constraint that shapes everything. Here it is properly.
Measurement destroys the state
When you measure a qubit, you get one classical outcome, 0 or 1. The amplitudes that described it are gone, and the qubit is now simply in the state you observed.
Read that again, because it is the whole difficulty. A system of 300 qubits is described by an unimaginably large collection of amplitudes. You measure it, and you receive 300 classical bits. That is all. One outcome, drawn at random according to those amplitudes, and the rest of the information is destroyed in the act of looking.
You cannot peek. You cannot measure gently. You cannot ask the machine which answer had the largest amplitude. Measurement is not a window onto the state, it is a transaction that consumes it.
This is why "tries all answers at once" is not merely imprecise but backwards. Even granting that the machine somehow holds every possibility, the reading step hands you one at random. Without careful construction, that is worth nothing.
Which is why algorithms are hard to find
The consequence is that a useful quantum algorithm must do something quite specific. It must arrange, before you measure, for the amplitude of the desired answer to be large and the amplitudes of the wrong answers to be small.
Notice what this requires. You need a problem with enough mathematical structure that a sequence of operations can systematically cancel the wrong answers. Most problems do not have that structure.
This is the honest reason the list of known quantum algorithms with real speedups is short, and why it has grown slowly since the 1990s. It is not a lack of effort or funding. It is that the requirement is genuinely restrictive.
A machine you must program by arranging interference so the right answer survives is not a general purpose accelerator. It is a specialised instrument for problems with a particular shape, which is why Module 3 covers a small number of algorithm families rather than a catalogue.
No cloning
A second consequence, and one with immediate practical effects.
You cannot copy an unknown quantum state. This is a theorem, not an engineering limitation. There is no procedure that takes an unknown qubit and produces two of it.
Three things follow.
Error correction had to be reinvented. Classical error correction leans on copying: store a bit three times, take a majority vote. That approach is unavailable, which is why quantum error correction is an entire discipline and the subject of Module 4.
Debugging is unlike anything in software. You cannot inspect an intermediate state without destroying it. Diagnosis means running the whole computation many times with different measurements at the end and reconstructing what must have happened statistically.
Eavesdropping becomes detectable. An interceptor cannot copy a qubit in transit and pass the original along undisturbed. Their measurement leaves evidence. This is the basis of quantum key distribution, covered in Module 5, which is a genuinely different technology from quantum computing and frequently confused with it.
They will not replace classical computers
A conclusion worth stating plainly, because it is widely misunderstood and it shapes how any organisation should plan.
Quantum computers will not run your spreadsheets, your database, your website or your payroll. Not because they are early, but because they are the wrong kind of machine, in the way a wind tunnel is the wrong tool for writing a letter.
For the overwhelming majority of computing tasks, a quantum computer is worse. Slower, vastly more expensive, error prone, and requiring elaborate physical conditions. It offers an advantage only where a problem's structure permits the interference trick, and that is a narrow class.
The realistic picture, already how the machines are used, is a specialised coprocessor. A classical computer runs the workload and hands off a specific subproblem to quantum hardware, then takes the result back and carries on. Rather like a graphics card, which is excellent at a narrow class of work and useless as a general processor.
So the question for any organisation is never whether to replace its computing with quantum computing. It is whether any problem you face falls into that narrow class, and Module 6 gives you a way to answer it.
A persistent claim holds that quantum computers will crack the hardest optimisation problems: routing, scheduling, packing, the travelling salesman. It appears in vendor material regularly. The consensus among researchers is that it is false, and the reasoning is worth following.
These problems share a property. Verifying a proposed answer is easy, while finding one appears to require searching an enormous space. That is roughly what NP-complete means, and thousands of practical problems fall into the class.
The tempting argument is that a quantum computer can hold all candidate solutions in superposition, check them all, and return the right one. The first half is true and the second half is where it fails. Holding a superposition over all candidates is easy. Getting the good one out is the problem, and it requires the amplitudes of the wrong answers to cancel, which requires structure in the problem for the operations to exploit.
Unstructured search is precisely the case where no such structure exists. The best available quantum result is Grover's algorithm, covered in Module 3, and it delivers a quadratic speedup: roughly the square root of the number of candidates rather than the number itself. Meaningful, and nothing like the exponential improvement the claim requires. Moreover, this is provably the best possible for unstructured search, so it is not awaiting a cleverer algorithm.
A quadratic speedup does not tame exponential growth. If a problem takes 2 to the 100 steps classically, the square root is 2 to the 50, which remains hopeless. Adding a handful of variables restores the original difficulty.
Where quantum methods may genuinely help on optimisation is more modest: heuristics that find good rather than optimal solutions on particular problem families, possibly faster than classical heuristics. That is a real research area and worth watching. It is not the same claim, and the gap between them is where a great deal of marketing lives.
Why does the ability to place a quantum computer in a superposition over many candidate answers not, by itself, solve hard search problems?
Measurement
Click to flipExtracting a classical outcome from a quantum state. Returns one result chosen according to the amplitudes and destroys the remaining information.
Click to flip backMeasuring 300 qubits yields 300 classical bits and destroys everything else, so the machine hands you one outcome at random unless the algorithm has already made the right answer overwhelmingly likely. That extraction problem, not hardware, is why the list of useful quantum algorithms is short: it requires mathematical structure most problems lack. Unknown states also cannot be copied, which forced error correction to be reinvented and makes eavesdropping detectable. Quantum machines will not replace classical ones; they are specialised coprocessors for a narrow class of problems.