Cram sheet — Section 1: Perform quantum operations

One-screen review layer for this section: short background primers per objective, then the key facts — each backed by an official source or by an executed proof from the question bank.

Define Pauli Operators

A Pauli label reads right-to-left

A Pauli label is read RIGHT to LEFT against qubit indices: the rightmost character acts on qubit 0, the leftmost on the highest-index qubit.

from qiskit.quantum_info import Pauli
Pauli('IX')              # X on qubit 0, I on qubit 1
Pauli('XZ').to_matrix()  # == np.kron(X, Z): X on qubit 1, Z on qubit 0
Pauli('iXX')             # phase prefix: '', 'i', '-', '-i'

The object stores the I/X/Y/Z string plus a phase drawn from +1, i, −1, −i. Writing the phase into the label makes a distinct object, so a plain multi-letter label carries no phase and its matrix is the bare tensor product.

Two families of combination methods, and the whole objective lives on telling them apart.

Call	Result	Notes
a.tensor(b)	a ⊗ b	caller a goes to the higher-index subsystem, b to qubit 0
a.expand(b)	b ⊗ a	the mirror: a sits on subsystem 0
a.dot(b)	A·B	plain matrix product
a.compose(b)	B·A	circuit order — a acts first
a.compose(b, front=True)	A·B	same as dot

On Paulis the difference is one character: Pauli('X').dot(Pauli('Y')) is iZ, while Pauli('X').compose(Pauli('Y')) is -iZ. The class never silently drops the phase.

Sources: operators-overview · operator-class · quantum_info

SparsePauliOp, and commutation by counting

Observables are SparsePauliOp — a linear combination of Pauli strings — and that is what the Estimator primitive expects, not a hand-built Pauli.

from qiskit.quantum_info import SparsePauliOp
SparsePauliOp.from_list([("ZZ", 1), ("XX", 0.5)])
SparsePauliOp.from_sparse_list([("ZX", [1, 4], 1)], num_qubits=5)  # prints XIIZI

from_list takes (label, coefficient) pairs on full-width labels. from_sparse_list takes (label, qubit_indices, coefficient) plus an explicit num_qubits, and there the label characters pair with the index list left-to-right — the little-endian display rule read back at you. simplify() drops only terms whose coefficient is zero within tolerance.

Commutation is a counting exercise, not a matrix multiplication: two multi-qubit Paulis commute exactly when the number of positions at which their single-qubit factors anticommute is EVEN.

Pair	Anticommuting positions	a.commutes(b)
Pauli('XX') vs Pauli('ZZ') (X⊗X vs Z⊗Z)	2 (even)	True
Pauli('XI') vs Pauli('ZZ') (X⊗I vs Z⊗Z)	1 (odd)	False

An identity factor contributes zero to that count — it never buys commutation by itself.

Sources: operators-overview · quantum_info

Must know:

• Pauli('IX') is I ⊗ X — X on qubit 0, identity on qubit 1 — and indexing the object returns the qubit-0 factor first. — ⚙️ proven in s1-q012
• Pauli products keep their phase: X·Y = iZ, Y·Z = iX, Z·X = iY via .dot; reversing the operands conjugates it (Y·X = -iZ, Z·Y = -iX). — ⚙️ proven in s1-q014
• SparsePauliOp.simplify() SUMS duplicate terms' coefficients (Z with 1 and Z with 2 becomes Z with 3), never averaging them or keeping only the largest. — ⚙️ proven in s1-q018
• cx(0, 1); rz(θ, 1); cx(0, 1) is exactly the ZZ evolution PauliEvolutionGate(SparsePauliOp('ZZ'), time=θ/2) — rz(π/3) in the middle equals time=π/6, because RZ also carries the half-angle convention. — ⚙️ proven in s1-q065

Traps:

• Z on qubit 0 is Pauli('IZ'); Pauli('ZI') is the opposite operator, and .adjoint() does not reverse qubit order (Hermitian labels come back unchanged). — ⚙️ proven in s1-q020
• PauliEvolutionGate(op, time=t) is exp(−i·t·op) with NO factor ½, while RXXGate(θ) and RZZGate(θ) mean exp(−i·θ/2·P): XX evolution for t = π/6 on |00⟩ gives {'00': 0.75, '11': 0.25}, but RXXGate(π/6) only 0.933/0.067. So evolution time t matches an RXX/RZZ angle of 2t. — ⚙️ proven in s1-q064
• a @ b on SparsePauliOps is the operator PRODUCT, not a tensor product: (X + Z) @ (X + Z) has raw terms I, −iY, +iY, I, and .simplify() leaves 2·I because the anticommuting cross terms cancel. a.tensor(a) is the different two-qubit XX + XZ + ZX + ZZ. — ⚙️ proven in s1-q057

Apply quantum operations

Statevector: build it, evolve it, read it

Statevector is the simulator you carry in your head, and it runs UNITARY circuits only.

from qiskit.quantum_info import Statevector
Statevector.from_label('10')     # |q1 q0⟩ — amplitude 1 at index 2
Statevector.from_int(2, dims=4)  # same state; an int dims is the TOTAL dimension
sv = Statevector(qc)             # delegates to Statevector.from_instruction(qc)
sv2 = sv.evolve(op)              # returns a NEW Statevector; sv is untouched

from_label also accepts the eigenstate letters '+', '−', 'r', 'l'. Leave a measure in the circuit and both constructors raise QiskitError: Cannot apply instruction with classical bits: measure. evolve accepts a Pauli, an Operator, a gate or a circuit.

Read-out splits into families that questions love to swap.

Call	Returns	Note
.data	complex amplitudes	in index order
.probabilities()	ndarray	squared moduli, index order
.probabilities_dict()	dict keyed by bitstring	keys run q_(n−1)…q_0
.probabilities([0])	ndarray of length 2 — not a dict	marginalizes every qubit not listed; a Bell state gives 0.5/0.5
.expectation_value(obs)	number	takes a Pauli or a SparsePauliOp

For |00⟩ against Z⊗Z + 0.5·(X⊗X) the answer is 1.0: |00⟩ is a +1 eigenstate of ZZ, while XX maps it to the orthogonal |11⟩.

Sources: quantum_info · operator-class

The phase family, and what equality means

The phase gates leave probabilities untouched and differ only in where the phase lands.

Gate	Matrix	Phase on the 1 amplitude
z	diag(1, −1)	−1
s	diag(1, i)	+i — the square root of Z, identical to p(π/2)
sdg	diag(1, −i)	−i, the conjugate of s
t	diag(1, e^(iπ/4))	the square root of S
tdg	diag(1, e^(−iπ/4))	the conjugate of t
p(θ)	diag(1, e^(iθ))	the whole angle
rz(θ)	diag(e^(−iθ/2), e^(+iθ/2))	split symmetrically over both amplitudes

So after h, s gives (|0⟩ + i|1⟩)/√2 and sdg gives (|0⟩ − i|1⟩)/√2. RZ and P of the same angle differ by a global phase, never by physics: rz(π/2) on |0⟩ records [0.7071−0.7071j, 0] while the probabilities stay [1.0, 0.0]. T on |1⟩ likewise records 0.7071+0.7071j — Qiskit keeps a global phase explicitly. That distinction is exactly what Operator.equiv and == separate. Setting qc.global_phase yourself is unobservable in the same way: exact probabilities and seeded counts come back bit-identical.

Two memorized formulas answer most probability questions without a simulator.

For RY(θ) on the zero state	Value
amplitudes	cos(θ/2) and sin(θ/2)
P(outcome 1)	sin²(θ/2)
⟨Z⟩	cos θ — 0.5 at θ = π/3, not sin or cos of the half angle

Sources: qiskit.circuit.library.SGate · operator-class

Must know:

• Operator(qc) for H on qubit 0 of a two-qubit circuit is I ⊗ H — qubit 0 sits on the RIGHT of the tensor product. — ⚙️ proven in s1-q037
• efficient_su2(n, reps=r) has 2·n·(r+1) free parameters — an ry and an rz per qubit, with one rotation layer MORE than reps: efficient_su2(4, reps=2) gives 24, and skip_final_rotation_layer=True drops it to 16. — ⚙️ proven in s1-q048
• real_amplitudes is the RY-only sibling of efficient_su2: its gate set is exactly ry + cx, so n·(reps+1) parameters (12 at the default reps=3 on 3 qubits). quantum_volume carries no free parameters at all. — ⚙️ proven in s1-q050
• QFTGate(n) takes only a qubit count (no do_swaps, no approximation_degree). One decompose() gives n h, n(n−1)/2 cp and ⌊n/2⌋ swap — for n = 4 count_ops() gives {'cp': 6, 'h': 4, 'swap': 2} (sorted by count), never cx. — ⚙️ proven in s1-q049
• Reading a Bloch drawing: h puts the arrow on +x and the phase gate then swings it inside the equator — s to +y, sdg to −y. An arrow left at a pole means that qubit was never rotated off the z axis. — ⚙️ proven in s1-q051
• A plot_state_city figure IS the density matrix: left panel Re(ρ), right panel Im(ρ). After h(0); cx(0,1); s(1) the two 0.5 populations stay real while the |00⟩⟨11| cell becomes −0.5i, so s is visible only in the imaginary panel. — ⚙️ proven in s1-q052

Traps:

• On a measured circuit, Statevector(qc) and from_instruction offer no shots/backend parameter, and appending a save instruction does not avoid the QiskitError. — ⚙️ proven in s1-q036
• X on qubit 0 of a two-qubit circuit gives the probabilities_dict() key '01' with probability 1.0 — never '10', never an integer index. — ⚙️ proven in s1-q031
• Operator.equiv ignores a global phase and == does not: Z versus RZ(π) is equiv-True but ==-False; T versus RZ(π/2) is False both ways (that pairing is S). — ⚙️ proven in s1-q033

Exam checklist

• Read every Pauli label right-to-left and name the operator on each qubit before answering.
• Say which of tensor, expand, dot or compose a question used, and which operand acts first.
• Count anticommuting positions to settle commutation: an even count commutes, an odd count does not.
• Delete every measure before handing a circuit to Statevector.
• Check whether the answer wants amplitudes, a probability ndarray, or a bitstring-keyed dict.
• Write out diag(1, i) for S and diag(e^(−iθ/2), e^(+iθ/2)) for RZ before comparing operators.
• Ask whether a difference is only a global phase: equiv ignores it, == does not.
• Compute 2·n·(reps+1) for efficient_su2 and n·(reps+1) for real_amplitudes.

---

Every fact above is sourced: 📖 links go to official documentation, ⚙️ marks facts observed by executing code against the pinned Qiskit stack (the linked section page shows the proof evidence on its practice questions).