Interference

Quantum “weights” can be negative, so two routes to the same answer can cancel — the trick that deletes wrong answers.

Quantum Computing · Lesson 6 · 9 min read

L04 left you with a precise demand: the right answer is buried among 2ⁿ combinations, and you must make it the likely one to read out. But how? With ordinary chances you’re stuck — probabilities only ever pile up; adding a second way to reach an outcome can only make it more likely, never less. So how could a machine ever remove probability from all the wrong answers? The answer is the strangest and most important idea in the whole field. Hold the question: how can adding a second route to an outcome make it less likely?

The weights are amplitudes — and they can be negative

Until now we called the blend’s weights just that — weights. The deeper truth: each possibility carries an amplitude, and an amplitude can be positive or negative (more generally it has a direction, called phase). The probability you actually measure is the amplitude’s size, squared — so the sign is invisible at the very end (squaring +0.5 and −0.5 both give 0.25). The sign seems pointless… until two amplitudes meet.

When routes combine, amplitudes add — and can cancel

If an outcome can be reached by more than one route, you add the routes’ amplitudes first, then square to get the probability. Two +0.5s add to +1.0 → probability 1.0: the routes reinforce (constructive interference). But +0.5 and −0.5 add to 0 → probability 0: the routes cancel (destructive interference). That cancellation is flatly impossible for classical probabilities, which are never negative and can only accumulate. Negative amplitudes are the whole difference.

Outcome Y is reachable two ways. Each route alone gives amplitude 0.5 (probability 0.25). Classically you’d say “two 25% routes → more likely.” Quantumly, if the second route’s amplitude is −0.5, then 0.5 + (−0.5) = 0, and 0² = 0. Adding a second way to reach Y made Y impossible. Nothing in dice or coins can do that.

Interference is the actual engine of quantum computing

Now the recipe from L04 has a method. Superposition spreads amplitude across all 2ⁿ combinations; a quantum algorithm is engineered so that, for every wrong answer, the routes to it cancel (destructive), while the routes to the right answer reinforce (constructive). After this sculpting, a single measurement is overwhelmingly likely to return the right answer. Superposition makes the many routes exist; interference decides which survive; measurement reads the winner. That is the heart of Grover’s and Shor’s algorithms (Module 3).

A toy machine with two internal routes, A and B, and two possible answers, RIGHT and WRONG:
1. Route A sends amplitude +0.5 to RIGHT and +0.5 to WRONG. Route B sends +0.5 to RIGHT and −0.5 to WRONG.
2. RIGHT’s total amplitude: 0.5 + 0.5 = +1.0 → probability 1.0² = 1.0. Reinforced.
3. WRONG’s total amplitude: 0.5 + (−0.5) = 0 → probability 0² = 0. Cancelled.
4. Measure → you get RIGHT essentially every time, in one shot — even though, route-by-route, WRONG looked just as likely as RIGHT.
5. That deliberate cancellation of the wrong answer is destructive interference — the move classical probability can never make, and the reason a quantum computer is more than a fancy coin-flipper.