Why we need the density matrix

By the end of chapter 9 we held a complete quantum-information toolkit: qubits, entanglement, Bell inequalities, teleportation. But that chapter ran on an assumption that was never said out loud — every system is isolated, evolution is always unitary, states are always pure.

Open the door of any real laboratory and that assumption shatters on the spot.

The phenomenon first: why a qubit in the lab “goes bad”

A superconducting qubit is fabricated, initialised into the superposition $\frac{1}{\sqrt2}(\ket{0}+\ket{1})$, and then left alone. Measure it again a few tens of microseconds later and its ability to interfere has vanished — it behaves exactly like a classical coin that is “50% $\ket{0}$, 50% $\ket{1}$”.

Nobody measured it, and the Hamiltonian did not evolve it anywhere else. By the rules of chapter 3, unitary evolution preserves inner products and preserves superpositions; $\ket{\psi}$ should forever remain some definite $\ket{\psi(t)}$. And yet it went bad.

The reason is not hard to guess: the qubit is not isolated. It soaks in an environment of electromagnetic noise, substrate phonons, and stray two-level defects. To describe a system that “is entangled with its environment while we can only see the system itself”, the language of state vectors is congenitally inadequate — which is exactly why section 3.10 introduced the density operator.

What this chapter does is turn that section’s static tool into dynamics: how open systems evolve (10.3, 10.4), how coherence disappears (10.5), and how the classical world emerges from all of it (10.6). This section first re-inspects the foundations, then pushes three steps forward.

Re-inspecting the foundations: ρ at a glance

The essentials of section 3.10, in four sentences:

$$\hat\rho=\sum_i p_i\ket{\psi_i}\bra{\psi_i},\qquad \langle A\rangle=\operatorname{Tr}(\hat\rho\hat A)\tag{10.1.1}$$

1. Definition: an ensemble that is in the pure state $\ket{\psi_i}$ with classical probability $p_i$ is described by the formula above; a pure state is the special case with a single term.
2. Three properties: Hermitian, $\operatorname{Tr}\hat\rho=1$, positive semi-definite — and conversely, any operator satisfying these three is a legitimate quantum state.
3. Purity: $\operatorname{Tr}(\hat\rho^2)=1$ if and only if the state is pure; the maximally mixed state in $d$ dimensions gives the lowest value, $1/d$.
4. Reduction: a part of an entangled state is described by $\hat\rho_A=\operatorname{Tr}_B\hat\rho_{AB}$, which is necessarily mixed.

If any one of these four sentences makes you hesitate, go back and work through section 3.10 first — every page of this chapter is built on it.

Two kinds of randomness, one table

Quantum mechanics contains two completely different kinds of “uncertain”, and the density matrix files them into different places within one and the same object.

The two matrices have the same diagonal, so a measurement along $z$ gives identical statistics; the entire difference lives in the off-diagonal elements, and measuring along any other direction separates them at once (section 3.10 worked the 50/50 version). A real state generally carries both kinds of randomness: the diagonal gives “the probability of what you measure”, the off-diagonal records “how much interference capability is left”. The main plot of this chapter is how the environment devours specifically the latter.

Step one forward: the ensemble interpretation and “infinitely many recipes”

Reading $\hat\rho=0.7\ket{\uparrow}\bra{\uparrow}+0.3\ket{\downarrow}\bra{\downarrow}$ as “an ensemble, 70% up and 30% down” comes naturally. But the warning at the end of section 3.10 must now be promoted to an official position: the recipe is not unique, and the recipes are indistinguishable in principle.

Two completely different recipes, one and the same ρbasic~5 minexpand

Recipe one: mix $\ket{\uparrow},\ket{\downarrow}$ in proportions $0.7/0.3$.

Recipe two: prepare, with probability $\tfrac12$ each, the two pure states

$$\ket{a}=\sqrt{0.7}\,\ket{\uparrow}+\sqrt{0.3}\,\ket{\downarrow},\qquad \ket{b}=\sqrt{0.7}\,\ket{\uparrow}-\sqrt{0.3}\,\ket{\downarrow}\tag{10.1.4}$$

Compute recipe two’s density matrix. The two projectors are

$$\ket{a}\bra{a}=\begin{pmatrix}0.7&\sqrt{0.21}\\\sqrt{0.21}&0.3\end{pmatrix},\qquad \ket{b}\bra{b}=\begin{pmatrix}0.7&-\sqrt{0.21}\\-\sqrt{0.21}&0.3\end{pmatrix}\tag{10.1.5}$$

Multiply each by $\tfrac12$ and add; the off-diagonal elements cancel in pairs:

$$\hat\rho=\begin{pmatrix}0.7&0\\0&0.3\end{pmatrix}\tag{10.1.6}$$

Element for element identical to recipe one. Since every measurable prediction enters only through $\operatorname{Tr}(\hat\rho\hat A)$, the two recipes give the same results for any measurement and any subsequent evolution — not “hard to tell apart”, but no fact of the matter, in principle.

The general result (the Hughston–Jozsa–Wootters theorem, stated here without proof): two ensembles give the same $\hat\rho$ if and only if their lists of states are related by a unitary matrix. One and the same $\hat\rho$ has infinitely many legitimate recipes.

Step two forward: where mixed states come from — purification

There is a deeper question still: are mixed states “fundamental”? Or does every mixed state hide a larger pure state behind it?

The answer is the latter, and the construction is surprisingly simple.

Purification: finding an entangled 'upstream' for any mixed stateadvanced~6 minexpand

Take any $\hat\rho_A$ and first diagonalise it (section 3.10: it is Hermitian, so this can always be done):

$$\hat\rho_A=\sum_k\lambda_k\ket{k}\bra{k},\qquad \lambda_k\ge0,\ \sum_k\lambda_k=1\tag{10.1.7}$$

Bring in an auxiliary system B whose dimension is at least the number of non-zero $\lambda_k$, and build the pure state of the composite system

$$\ket{\Psi}_{AB}=\sum_k\sqrt{\lambda_k}\,\ket{k}_A\ket{k}_B\tag{10.1.8}$$

(You should recognise this: it is exactly the Schmidt form from chapter 9.) Verify that it really is the upstream of $\hat\rho_A$ — trace out B:

$$\operatorname{Tr}_B\ket{\Psi}\bra{\Psi} =\sum_j\bra{j}_B\Big(\sum_{k,l}\sqrt{\lambda_k\lambda_l}\, \ket{k}_A\bra{l}_A\otimes\ket{k}_B\bra{l}_B\Big)\ket{j}_B\tag{10.1.9}$$

The inner products $\braket{j}{k}_B=\delta_{jk}$ and $\braket{l}{j}_B=\delta_{lj}$ collapse the double sum into a single one:

$$=\sum_k\lambda_k\ket{k}_A\bra{k}_A=\hat\rho_A\qquad\blacksquare\tag{10.1.10}$$

Example: one purification of $\hat\rho_A=0.7\ket{\uparrow}\bra{\uparrow}+0.3\ket{\downarrow}\bra{\downarrow}$ is

$$\ket{\Psi}_{AB}=\sqrt{0.7}\,\ket{\uparrow}_A\ket{\uparrow}_B+\sqrt{0.3}\,\ket{\downarrow}_A\ket{\downarrow}_B\tag{10.1.11}$$

— an entangled state. The more mixed the state, the more entanglement it takes; the purification of the maximally mixed state is precisely a Bell state.

∎Key formulas

• Density matrix and expectation values

  $$\hat\rho=\sum_ip_i\ket{\psi_i}\bra{\psi_i},\quad\langle A\rangle=\operatorname{Tr}(\hat\rho\hat A)$$

  The foundation from section 3.10; this whole chapter is built on it

• Where the two randomnesses sit

  $$\rho_{nn}\ \text{probabilities};\quad \rho_{mn}\,(m\ne n)\ \text{coherences}$$

  The environment devours specifically the off-diagonal — the main plot of this chapter

• Recipes are not unique

  $$\sum_ip_i\ket{\psi_i}\bra{\psi_i}=\sum_jq_j\ket{\phi_j}\bra{\phi_j}$$

  One ρ has infinitely many ensemble decompositions, indistinguishable in principle

• Purification

  $$\ket{\Psi}_{AB}=\sum_k\sqrt{\lambda_k}\,\ket{k}_A\ket{k}_B,\quad\operatorname{Tr}_B\ket{\Psi}\bra{\Psi}=\hat\rho_A$$

  Every mixed state is a local view of some entangled pure state

?Self-check4 questions

1. 1.

   How do the density matrices of "70% prepared in |↑⟩, 30% in |↓⟩" and of the pure state √0.7|↑⟩ + √0.3|↓⟩ differ?

   • Same diagonal, different off-diagonal: zero for the former, √0.21 for the latter
   • Different diagonal, same off-diagonal
   • The two matrices are completely identical
   • The two matrices are completely different, with no element in common

   Submit
2. 2.

   The ensemble is 70% |↑⟩, 30% |↓⟩. Find the purity Tr(ρ²).

   1% relative tolerance

   Submit
3. 3.

   Which statements about "different preparation recipes for the same ρ" are correct? (Select all that apply.)

   Select all that apply

   • One and the same ρ can be mixed from infinitely many different ensembles of pure states
   • With sufficiently precise instruments, one can always determine which recipe prepared the system
   • All predictions enter only through Tr(ρÂ), so a difference in recipe has no observable consequence
   • The security of quantum key distribution exploits exactly this indistinguishability

   Submit
4. 4.

   The purification theorem says "every mixed state is the reduction of some larger pure state". The correct understanding of a lab qubit "going mixed" is:

   • The qubit entangles with the environment; the information moves into system-environment correlations, and the system alone looks mixed
   • Some random force is literally throwing dice, swapping the pure state for other pure states at random
   • The Schrödinger equation fails for open systems, and quantum mechanics needs modification
   • The length of the qubit’s state vector slowly shrinks

   Submit

What comes next

A density matrix is a $2\times2$ complex matrix, which sounds abstract. But for a qubit it has an exact map — the Bloch sphere. Pure states live on the surface, mixed states live in the interior, and every “going bad” process in this chapter appears on the map as a visible contraction. The next section draws that map.