magic state

In plain terms

The Gottesman-Knill theorem says Clifford circuits are classically simulable -- so where does quantum advantage hide? In one extra gate, the T gate, an eighth-turn phase twist. Add T to the Clifford set and you can approximate anything; leave it out and a laptop keeps up.

What makes T genuinely different is provable: push a Pauli operation through any Clifford gate and out comes another Pauli operation -- that closure is the whole simulability mechanism. Push a Pauli through T and what comes out is NOT a Pauli, and never can be.

The "magic state" is T's resource form: the specific single-qubit state T would produce from a plain superposition. Feed copies of it into Clifford-only circuits and the T gate happens for free -- which is why fault-tolerant architectures spend most of their budget manufacturing (distilling) high-quality magic states.

In CSD

The precise complement of the corpus's Gottesman-Knill layer, landed the same day the GK plan closed. Three theorems draw the boundary: T squared IS the Clifford phase gate (the hierarchy descends); conjugating the Pauli X by T gives exactly (X + i XZ)/sqrt 2 -- out of the Pauli family but into its two-term span, the level-3 hierarchy identity as a closed operator equation; and the no-go -- no scalar multiple of any Pauli equals T X T-dagger, proved by pinning two basis columns until the arithmetic demands 1 = +-i.

The magic state itself is exhibited with closed coordinates and unit norm. Honest scope: distillation protocols (Bravyi-Kitaev 15-to-1), universality of Clifford+T, and the injection circuit are named residues with reasons -- the module proves what magic IS, not how it is farmed. The standing disclaimer applies: the algorithm layer consumes the unitary pillar and does not support foundations.

Mathematically

With tPhase = e^(i pi/4): tPhase squared is i, so T^2 = S on the register. The conjugation identity T X T-dagger = (X + i XZ)/sqrt 2 holds coordinatewise via e^(+-i pi/4) = (1 +- i)/sqrt 2. The no-go: if T X T-dagger = c X^a Z^b then reading the |0> and |1> columns forces c = e^(i pi/4) and e^(-i pi/4) = c times a sign, hence e^(-i pi/2) in {+1, -1} -- but e^(-i pi/2) = -i. The magic state T H |0> has coordinates (1, e^(i pi/4))/sqrt 2.

The name

The term "magic state" is Sergey Bravyi and Alexei Kitaev's, from their 2005 paper showing that noisy copies of certain single-qubit states can be distilled by ideal Clifford operations into fuel for universal computation -- the 15-to-1 protocol that still anchors fault-tolerant cost estimates. The gate-injection idea goes back to Gottesman and Chuang's 1999 teleportation construction, which organised gates into the Clifford hierarchy: Paulis at level one, Clifford at level two, T at level three.

The economics of quantum computing largely reduce to magic-state throughput, which is why "magic" -- Bravyi and Kitaev's deliberately playful word -- became sober technical vocabulary.