Measurement & Collapse

Looking at a qubit forces it to pick a single 0 or 1 — and the carefully tuned blend is gone for good.

Quantum Computing · Lesson 3 · 8 min read

You spend real effort tuning a qubit into a precise blend — say 73% toward 0, 27% toward 1. Now you want to “read” that blend. So you measure it… and out pops a single 0. Puzzled, you measure the same qubit again — 0 again. And again. Where did your carefully set 73/27 go? You can never read it back off that qubit. This is measurement and collapse, and it’s the rule that decides what a quantum computer can and can’t actually give you. Hold the question: if measuring destroys the blend, how is a qubit ever useful?

Measurement forces one definite answer

A qubit can hold a rich blend (L02), but the instant you measure it, that blend collapses to a single, definite 0 or 1 — never “both,” never the blend itself. Which one you get is random, with the odds set by the weights you tuned: a blend leaning toward 0 makes 0 the likely readout. You don’t choose the outcome; you only shaped the odds beforehand. One qubit measured gives you exactly one bit of ordinary, classical information.

Collapse is one-way and destructive

Collapse can’t be undone. After the qubit lands on, say, 0, it really is 0 now — measure it again and you’ll get 0 every time, because the blend is already gone. You can’t peek at the blend and then put it back; the act of looking is what destroyed it. So you never get to “read the dial.” You get one frozen snapshot, and the continuous blend that produced it is unrecoverable from that single look.

Your 73/27 qubit, measured: most likely you read 0 (≈73% of the time). Measure that same qubit again — 0, 0, 0. It’s stuck on its collapsed value. Nothing in those repeats reveals “73%”; the first measurement already erased the blend and replaced it with a plain 0.

So how do you ever learn the blend? Statistics

If one qubit only yields one bit, the blend is learned the way pollsters learn opinions: across many identical samples. Prepare the same blend on a thousand fresh qubits, measure each once, and tally — the frequencies reveal the weights (≈730 zeros, ≈270 ones ⇒ a 73/27 blend). This is why a qubit is not free storage: any single run hands back just one outcome. A real quantum algorithm’s whole job is to arrange the blend so the useful answer is the one most likely to survive that single measurement.

Reading a blend, the only way you actually can:
1. You have ONE qubit tuned to 60/40 (favoring 0). Measure it → you get a single 0 or 1 (more likely 0). The blend is now destroyed.
2. Measure that same qubit again → you get the same value as step 1, forever. No new information; it collapsed already.
3. So instead, prepare 1000 fresh qubits in the same 60/40 blend and measure each once.
4. Tally: about 600 read 0, about 400 read 1. Those frequencies are how you infer “60/40.”
5. Lesson: the blend is a real thing (L02), but it only shows up in the pattern across many measurements — never in a single qubit’s single readout. That constraint is exactly what quantum algorithms must design around.