Find one marked item among N with no clues. A classical search checks them one by one — about N/2 tries. Grover does it in about √N, by slowly amplifying the right answer’s amplitude while the rest shrink.
Picture every possible answer humming at the same volume — a flat row of equal bars. You can’t hear which is the one you want. Grover applies a repeated two-step move that nudges the right bar a little louder and every other bar a little quieter, over and over. After about √N rounds the marked answer is practically shouting and you simply measure it.
Imagine a roomful of identical candles, one secretly the one you need — indistinguishable by eye. Grover's move: mark that candle's flame to point downward, then fold every flame across the room's average height. Folding leaves the marked one a little taller than before. Repeat the mark-and-fold a handful of times and it towers over the rest.
Click a bar to mark the secret answer, then step the algorithm. Each step flips the marked amplitude, then reflects every bar about the average. Watch the marked bar climb — and notice that past the optimal point it starts falling again. More is not better.
Collapse the whole problem onto a 2D plane: one axis is the marked state, the other is everything else. The starting superposition sits at a tiny angle θ off the “wrong” axis, where sinθ = 1/√N. Each Grover step is a rotation by 2θ toward the marked axis:
That fixed rotation is also why overshooting hurts: keep going past 90° and you rotate away from the answer, and success probability falls. The marked bar in the lab rising then sinking is this rotation sweeping past its target.
“Grover checks all N answers at once, so it’s an exponential leap.” It is not. The speedup is quadratic — √N versus N — real and useful, but a world away from Shor’s exponential jump. A billion items still takes tens of thousands of steps, not one.
And it isn’t parallel evaluation. There’s a single amplitude vector being slowly rotated by repeated reflections. The proof that it’s genuinely quantum is the overshoot: a classical search never gets worse the longer you run it.