Here is the single trick that powers almost every quantum algorithm. Aim a controlled gate at a target that is already an eigenstate — a state the gate only multiplies by a phase — and that phase doesn’t stick to the target. It kicks back onto the control. You point the gate one way; the effect lands the other.
↩ before you start · keep these handy
·From QI Ch. 5: |−⟩ = (|0⟩−|1⟩)/√2, and X|−⟩ = −|−⟩ (so |−⟩ is an eigenstate of X with eigenvalue −1).
·From QC 02: a controlled-U applies U to the target only on the control-1 branch.
·From QI 8b: Deutsch’s algorithm secretly runs on this very move.
🔑 symbol decoder · every new mark, in plain words
eigenstatea special input that a gate doesn’t reshape — it only multiplies it by a number: U|u⟩ = (number)|u⟩.eᵢφ (eigenvalue)that number for a unitary is always a pure phase of size 1: a point on the unit circle at angle φ.φ (phi)the phase angle, in degrees or radians, that the eigenstate picks up.kickbackthe eigen-phase appearing on the control qubit instead of the target.phase oraclea yes/no function turned into a sign: |x⟩ → (−1)f(x)|x⟩, built by kickback.global vs relativea phase on the whole state is invisible; a relative phase between branches is real and steers interference.
feel
Push the target, the control recoils
Set the target to a state the gate can’t actually move — an eigenstate. The gate still “fires,” but with nowhere to push the target, its only effect is a phase. And because the gate is controlled, that phase attaches to the deciding qubit. The target comes out untouched; the control carries the news.
🧶 everyday picture
Fire a rifle and the bullet flies forward — but you feel the recoil in your shoulder. Phase kickback is the same conservation trick: you aim the gate at the target, yet the measurable kick lands on the control. The target (the “bullet” in its unmovable eigenstate) sails on unchanged, while the control (your “shoulder”) absorbs the phase.
recapPoint a controlled gate at an eigenstate target and its phase recoils onto the control.
play
The kickback dial
The control starts in |+⟩ — equal parts |0⟩ and |1⟩, drawn as a phasor. Choose a target eigenstate and dial the gate’s phase φ. The |1⟩ arm of the control swings by exactly φ, while the target sits perfectly still.
▸ kickback dialcontrol |+⟩ · controlled-phase(φ)
target eigenstate
gate phase φ = {{ phiDeg }}°
control → {{ controlKet }} target → {{ targetNote }}
recapOn an eigenstate target the control’s |1⟩ arm swings by exactly the eigen-phase; the target never moves.
math
Three lines, the whole engine
Let the target be an eigenstate: U|u⟩ = eᵢφ|u⟩. Put the control in |+⟩ and apply controlled-U:
CU · |+⟩|u⟩ = (|0⟩|u⟩ + |1⟩·U|u⟩)/√2
= (|0⟩|u⟩ + eᵢφ|1⟩|u⟩)/√2
= (|0⟩ + eᵢφ|1⟩)/√2 ⊗ |u⟩
The target |u⟩ factors right back out, untouched. The phase eᵢφ is now a relative phase on the control — readable by interference. Choose U = X, |u⟩ = |−⟩ (eigenvalue −1) and you get a clean −1 sign — the Grover phase oracle.
✎ worked example · turning a bit-oracle into a sign
1.oracle U f writes the answer: U f|x⟩|y⟩ = |x⟩|y⊕f(x)⟩
2.set the answer wire to |−⟩. Flipping |−⟩ gives −|−⟩: |y⊕1⟩ → − on |−⟩
3.so U f|x⟩|−⟩ = (−1)f(x)|x⟩|−⟩ — the flip became a sign
4.the |−⟩ wire is unchanged and factors out; f is now stamped as a phase on |x⟩. ✓
recapCU on |+⟩|u⟩ factors the eigenstate out and leaves eᵢφ as a relative phase on the control.
⚠ common misconceptions
“Kickback changes the target.” The opposite — it works because the target is an eigenstate the gate cannot change. The target comes out identical; only the control’s phase moves.
“A phase is invisible, so kickback is useless.” A global phase is invisible; this is a relative phase between the |0⟩ and |1⟩ branches. A Hadamard converts it straight into a measurable bit — which is what Deutsch, Grover and phase estimation all cash in.
“You must prepare a special eigenstate first.” Often it’s trivial: |−⟩ is one Hadamard away from |1⟩. And in phase estimation you don’t even need a pure eigenstate — any input is a blend of them, and each branch kicks back its own phase.