Part II. Mathematical Core

Density Matrices and Mixed States

The state $∣ + ⟩$ and an honest coin flip between $∣0 ⟩$ and $∣1 ⟩$ give identical measurement statistics — and behave completely differently from then on. The density matrix is the notation that can tell them apart, and the one real hardware makes unavoidable.

Listen to this chapter

A statevector asserts complete knowledge of an isolated system; real devices are neither fully known nor isolated, so the density matrix is the honest description of a qubit in the field.

By the end you will build a density matrix from a state vector, assemble a mixed state from classical weights, check the three rules every $ρ$ must obey, and read coherence directly off the off-diagonal entries.

Core concepts: pure versus mixed states, coherence, trace, noise channels.

Notice that only the pure state sits on the surface. Both ρ's return identical Z-basis counts, but the off-diagonal entries of the left one let it interfere later — the right one's never can.

When one vector is not enough

A pure statevector represents maximal knowledge of an ideal, isolated quantum state. A density matrix represents pure states and everything short of them. It is the right language whenever you face uncertainty, noise, entanglement with an environment, or partial access to a larger system — which is to say, whenever you face hardware.

For a pure state $∣ ψ ⟩$ , the density matrix is the outer product:

ρ = ∣ ψ ⟩ ⟨ ψ ∣

A mixed state is a classical weighted combination of such projectors:

ρ = i \sum p_{i} ∣ ψ_{i} ⟩ ⟨ ψ_{i} ∣, i \sum p_{i} = 1

Read the mixture as a recipe: with probability $p_{i}$ , the device actually holds $∣ ψ_{i} ⟩$ , and you do not know which.

The rules ρ must obey

Three conditions certify any candidate density matrix:

Hermitian: $ρ = ρ^{†}$ .
Unit trace: $tr (ρ) = 1$ — total probability is conserved.
Nonnegative eigenvalues — no outcome gets negative weight.

Two examples to hold in your head. For $∣0 ⟩$ :

ρ_{0} = ∣0 ⟩ ⟨ 0∣ = (1000)

And for $∣ + ⟩ = (∣0 ⟩ + ∣1 ⟩) / 2$ :

ρ_{+} = \frac{1}{2} (1111)

The off-diagonal entries carry coherence — the memory that the amplitudes had a definite phase relationship. Noise typically shrinks or scrambles exactly those entries, which is why experimentalists watch them.

Worked example: coherent versus classical fifty-fifty

Build the density matrix for $∣ + ⟩$ :

ρ = ∣ + ⟩ ⟨ + ∣ = \frac{1}{2} (11) (11) = \frac{1}{2} (1111)

Check the trace: $\frac{1}{2} + \frac{1}{2} = 1$ . Good. Now compare a fifty-fifty classical mixture of $∣0 ⟩$ and $∣1 ⟩$ :

ρ_{mixed} = \frac{1}{2} (1000) + \frac{1}{2} (0001) = (1/2 0 0 1/2)

Both matrices give $0$ and $1$ half the time in the computational basis. Only $ρ_{+}$ has off-diagonal terms. Apply H to each and the difference surfaces immediately: $H ∣ + ⟩ = ∣0 ⟩$ deterministically, while the mixture maps to an equal blend of $∣ + ⟩$ and $∣ - ⟩$ — still fifty-fifty, forever. One matrix remembered a phase; the other never had one.

Where the intuition fails

The first trap is believing equal measurement probabilities mean equal states. One basis's statistics are a projection of the state, not the state; the worked example's two matrices are proof that the projection can hide everything that matters.

The second trap is discussing noise with statevectors alone. A statevector can describe the wrong state, but not an unknown one. Real hardware forces you to track uncertainty, decoherence, and channels — and the density matrix is where all three live.

The engineering view

For a computer scientist, the density matrix is a richer state representation for systems with uncertainty and environment interaction. It costs more — an $n$ -qubit $ρ$ holds $4^{n}$ entries against the statevector's $2^{n}$ — and it buys realism: noise processes act on $ρ$ in ways a statevector cannot express. The depolarizing channel, for instance, steadily pushes any state toward the less informative mixture at the center of the sphere, erasing off-diagonal coherence as it goes.

This is load-bearing for simulator design, error mitigation, quantum error correction, and hardware diligence alike. It also separates two stories that look identical in one measurement basis: a coherent superposition can interfere later, while a classical mixture with the same immediate probabilities cannot. A model that cannot represent that difference will overstate what an ideal statevector demonstration says about hardware.

What this buys you in diligence

When a company claims useful noisy hardware, ask which noise model was used, how it acts on density matrices or observables, and whether the claim survives anything beyond ideal simulation. And when a software tool models only pure states, work out which hardware effects it structurally cannot represent — then decide whether your question depends on any of them.

Exercise

Build two density matrices and compare. Pencil and matrix arithmetic only.

Submit: $ρ_{+}$ , the equal mixture of $∣0 ⟩$ and $∣1 ⟩$ , and a trace check for each.
Check: compare the off-diagonal entries, then explain in one paragraph why equal computational-basis probabilities do not imply identical states — the H test from the worked example makes a good spine for the argument.
Repair: if mixed states keep collapsing into "ordinary ignorance" without a $ρ$ to track them, redo Chapter 10 (Inner Products, Orthogonality, and Projection) and rebuild the outer-product mechanics.

Check your understanding

Answer without notes: create a pure-state density matrix, verify its trace, and explain what changes for a mixed state.

A passing answer writes the outer product, checks all three rules, and locates the difference between pure and mixed in the off-diagonal entries rather than in the basis probabilities.

Oral defense: explain why $I /2$ is the most ignorant possible one-qubit state — and what happens to it under any unitary.

If you get stuck

If the trace, coherence, or the mixed-state interpretation is shaky, go back to Chapter 10 (Inner Products, Orthogonality, and Projection) for the outer-product and probability mechanics, then rebuild the worked example. Density matrices are that arithmetic, organized for uncertainty.