Skip to content

Purification

This is the canonical treatment of purification, including constructions, nonuniqueness, ensemble meaning, dilation links, and infinite-dimensional qualifications. For a density-operator-motivated first encounter, use Purification Overview.

Purification expresses every density operator as the reduced state of a pure state on a larger Hilbert space. If system AA has state ρA\rho_A, one may introduce a reference system RR and find a normalized vector ∣Ψ⟩AR\lvert\Psi\rangle_{AR} such that

ρA=Tr⁡R(∣Ψ⟩⟨Ψ∣AR).\rho_A = \operatorname{Tr}_R \bigl( \lvert\Psi\rangle\langle\Psi\rvert_{AR} \bigr).

This representation connects mixed states, entanglement, ensemble decompositions, and open-system models. It is a mathematical existence statement, not evidence that ρA\rho_A has one privileged hidden environment or one uniquely determined preparation history.

Let ρA\rho_A be a density operator on a finite-dimensional Hilbert space HA\mathcal H_A. A purification of ρA\rho_A consists of

  1. an auxiliary Hilbert space HR\mathcal H_R, and
  2. a normalized vector ∣Ψ⟩AR∈HA⊗HR\lvert\Psi\rangle_{AR}\in\mathcal H_A\otimes\mathcal H_R

for which

Tr⁡R(∣Ψ⟩⟨Ψ∣AR)=ρA.\operatorname{Tr}_R \bigl( \lvert\Psi\rangle\langle\Psi\rvert_{AR} \bigr) = \rho_A.

The condition says more than merely matching eigenvalues. It guarantees that every observable MAM_A acting locally on AA has the same expectation value in the two descriptions:

Tr⁡A(ρAMA)=Tr⁡AR(∣Ψ⟩⟨Ψ∣AR×(MA⊗IR))=⟨Ψ∣(MA⊗IR)∣Ψ⟩.\begin{aligned} \operatorname{Tr}_A(\rho_A M_A) &= \operatorname{Tr}_{AR} \bigl( \lvert\Psi\rangle\langle\Psi\rvert_{AR} \\ &\qquad{}\times (M_A\otimes I_R) \bigr) \\ &= \langle\Psi\rvert (M_A\otimes I_R) \lvert\Psi\rangle. \end{aligned}

No experiment confined to AA can distinguish ρA\rho_A from its description as part of the larger pure state. Access to RR, or to correlations between AA and RR, supplies additional information.

The label RR is deliberately neutral. Depending on the problem it may denote a mathematical reference, a controllable ancilla, an inaccessible environment, or a genuine second subsystem. Purification alone does not choose among these interpretations.

Spectral purification construction showing a mixed state lifted to a pure state on a larger system and reduced back to both marginals

The spectral construction assigns one orthogonal reference label to each nonzero eigenvalue. The smallest purifying space has dimension rank⁡ρA\operatorname{rank}\rho_A.

Purification theorem. Every finite-dimensional density operator has a purification. If

r=rank⁡ρA,r=\operatorname{rank}\rho_A,

then a reference space of dimension rr is sufficient and no smaller reference space can purify ρA\rho_A.

To prove existence, use the spectral decomposition on the support of ρA\rho_A:

ρA=∑k=1rpk∣k⟩A⟨k∣A,pk>0,∑k=1rpk=1.\begin{aligned} \rho_A &= \sum_{k=1}^{r} p_k \lvert k\rangle_A\langle k\rvert_A, \\ p_k&>0, \qquad \sum_{k=1}^{r}p_k=1. \end{aligned}

Choose orthonormal vectors {∣k⟩R}k=1r\{\lvert k\rangle_R\}_{k=1}^r and define

∣Ψρ⟩AR=∑k=1rpk ∣k⟩A∣k⟩R.\lvert\Psi_\rho\rangle_{AR} = \sum_{k=1}^{r} \sqrt{p_k}\, \lvert k\rangle_A\lvert k\rangle_R.

Normalization follows immediately:

⟨Ψρ∣Ψρ⟩=∑j,k=1rpjpk ⟨j∣k⟩A×⟨j∣k⟩R=∑k=1rpk=1.\begin{aligned} \langle\Psi_\rho\vert\Psi_\rho\rangle &= \sum_{j,k=1}^{r} \sqrt{p_jp_k}\, \langle j\vert k\rangle_A \\ &\qquad{}\times \langle j\vert k\rangle_R \\ &= \sum_{k=1}^{r}p_k =1. \end{aligned}

For the reduced state, expand the projector and trace the reference factor:

Tr⁡R(∣Ψρ⟩⟨Ψρ∣)=∑j,k=1rpjpk ∣j⟩A⟨k∣A×Tr⁡(∣j⟩R⟨k∣R)=∑j,k=1rpjpk δjk∣j⟩A⟨k∣A=∑k=1rpk∣k⟩A⟨k∣A=ρA.\begin{aligned} \operatorname{Tr}_R \bigl( \lvert\Psi_\rho\rangle \langle\Psi_\rho\rvert \bigr) &= \sum_{j,k=1}^{r} \sqrt{p_jp_k}\, \lvert j\rangle_A\langle k\rvert_A \\ &\quad{}\times \operatorname{Tr} \bigl( \lvert j\rangle_R\langle k\rvert_R \bigr) \\ &= \sum_{j,k=1}^{r} \sqrt{p_jp_k}\, \delta_{jk} \lvert j\rangle_A\langle k\rvert_A \\ &= \sum_{k=1}^{r} p_k \lvert k\rangle_A\langle k\rvert_A = \rho_A. \end{aligned}

This proves existence. It also displays the nonzero eigenvalues of the other marginal:

ρR=Tr⁡A(∣Ψρ⟩⟨Ψρ∣)=∑k=1rpk∣k⟩R⟨k∣R.\rho_R = \operatorname{Tr}_A \bigl( \lvert\Psi_\rho\rangle \langle\Psi_\rho\rvert \bigr) = \sum_{k=1}^{r} p_k \lvert k\rangle_R\langle k\rvert_R.

The rank bound follows from the Schmidt decomposition. Any pure state on ARA R has at most dim⁡HR\dim\mathcal H_R nonzero Schmidt coefficients, while the rank of its reduced state on AA equals its Schmidt rank. Therefore

dim⁡HR≥rank⁡ρA.\dim\mathcal H_R \ge \operatorname{rank}\rho_A.

The spectral construction saturates this inequality.

There is a basis-dependent formula that avoids diagonalizing ρA\rho_A. Let HR\mathcal H_R be a copy of the dd-dimensional space HA\mathcal H_A, choose corresponding orthonormal bases, and introduce the unnormalized maximally correlated vector

∣Ω⟩AR=∑j=1d∣j⟩A∣j⟩R.\lvert\Omega\rangle_{AR} = \sum_{j=1}^{d} \lvert j\rangle_A\lvert j\rangle_R.

Then

∣Ψρ⟩AR=(ρA⊗IR)∣Ω⟩AR\lvert\Psi_\rho\rangle_{AR} = \bigl( \sqrt{\rho_A}\otimes I_R \bigr) \lvert\Omega\rangle_{AR}

is a purification. Its norm is

⟨Ψρ∣Ψρ⟩=⟨Ω∣(ρA⊗IR)∣Ω⟩=Tr⁡ρA=1.\begin{aligned} \langle\Psi_\rho\vert\Psi_\rho\rangle &= \langle\Omega\rvert (\rho_A\otimes I_R) \lvert\Omega\rangle \\ &= \operatorname{Tr}\rho_A =1. \end{aligned}

Writing (ρA)jk=⟨j∣ρA∣k⟩(\sqrt{\rho_A})_{jk}=\langle j\vert\sqrt{\rho_A}\vert k\rangle gives

∣Ψρ⟩=∑j,k=1d(ρA)jk∣j⟩A∣k⟩R.\lvert\Psi_\rho\rangle = \sum_{j,k=1}^{d} (\sqrt{\rho_A})_{jk} \lvert j\rangle_A\lvert k\rangle_R.

Tracing RR contracts the shared index kk:

Tr⁡R(∣Ψρ⟩⟨Ψρ∣)=∑j,j′,k(ρA)jk(ρA)j′k∗×∣j⟩⟨j′∣A=ρA ρA†=ρA.\begin{aligned} \operatorname{Tr}_R \bigl( \lvert\Psi_\rho\rangle \langle\Psi_\rho\rvert \bigr) &= \sum_{j,j',k} (\sqrt{\rho_A})_{jk} (\sqrt{\rho_A})_{j'k}^{*} \\ &\quad{}\times \lvert j\rangle\langle j'\rvert_A \\ &= \sqrt{\rho_A}\, \sqrt{\rho_A}^{\dagger} = \rho_A. \end{aligned}

Because ρA\sqrt{\rho_A} is positive and therefore Hermitian, the last product is exactly ρA\rho_A. This construction is useful in matrix calculations and anticipates state–operator correspondences, but it depends on the chosen identification of bases in AA and RR.

Purification does not require an eigenstate ensemble. Suppose

ρA=∑j=1mqj∣ψj⟩⟨ψj∣,qj≥0,∑j=1mqj=1.\begin{aligned} \rho_A &= \sum_{j=1}^{m} q_j \lvert\psi_j\rangle \langle\psi_j\rvert, \\ q_j&\ge0, \qquad \sum_{j=1}^{m}q_j=1. \end{aligned}

where the normalized states ∣ψj⟩\lvert\psi_j\rangle need not be orthogonal. Introduce orthonormal labels {∣j⟩R}\{\lvert j\rangle_R\} and set

∣Φ⟩AR=∑j=1mqj ∣ψj⟩A∣j⟩R.\lvert\Phi\rangle_{AR} = \sum_{j=1}^{m} \sqrt{q_j}\, \lvert\psi_j\rangle_A \lvert j\rangle_R.

Then

Tr⁡R(∣Φ⟩⟨Φ∣)=∑j,kqjqk ∣ψj⟩⟨ψk∣⟨k∣j⟩R=∑jqj∣ψj⟩⟨ψj∣=ρA.\begin{aligned} \operatorname{Tr}_R \bigl( \lvert\Phi\rangle \langle\Phi\rvert \bigr) &= \sum_{j,k} \sqrt{q_jq_k}\, \lvert\psi_j\rangle \langle\psi_k\rvert \langle k\vert j\rangle_R \\ &= \sum_j q_j \lvert\psi_j\rangle \langle\psi_j\rvert = \rho_A. \end{aligned}

Unlike the spectral construction, this expression is generally not a Schmidt decomposition because the ∣ψj⟩\lvert\psi_j\rangle need not be orthogonal.

The finite-dimensional Hughston–Jozsa–Wootters theorem sharpens this observation: every pure-state ensemble decomposition of ρA\rho_A can be generated by a suitable measurement on a purifying system, with the measurement outcome retained as a classical label. If that outcome is ignored, the state of AA remains ρA\rho_A. This is why an ensemble records a preparation procedure rather than an additional intrinsic decomposition of the density operator. See Ensembles and Preparation Procedures and Conditional States.

Write a qubit density operator in Bloch form:

ρA=12(I+r⋅σ),r=∥r∥,0≤r≤1.\begin{aligned} \rho_A &= \frac12 \bigl( I+\boldsymbol r\mathbin{\cdot}\boldsymbol\sigma \bigr), \\ r&=\lVert\boldsymbol r\rVert, \qquad 0\le r\le1. \end{aligned}

For r>0r>0, let ∣n,+⟩\lvert\boldsymbol n,+\rangle and ∣n,−⟩\lvert\boldsymbol n,-\rangle be eigenstates of n⋅σ\boldsymbol n\mathbin{\cdot}\boldsymbol\sigma, where n=r/r\boldsymbol n=\boldsymbol r/r. The eigenvalues are

λ±=1±r2,\lambda_\pm = \frac{1\pm r}{2},

so a two-qubit purification is

∣Ψr⟩AR=1+r2 ∣n,+⟩A∣0⟩R+1−r2 ∣n,−⟩A∣1⟩R.\begin{aligned} \lvert\Psi_{\boldsymbol r}\rangle_{AR} &= \sqrt{\frac{1+r}{2}}\, \lvert\boldsymbol n,+\rangle_A \lvert0\rangle_R \\ &\quad+ \sqrt{\frac{1-r}{2}}\, \lvert\boldsymbol n,-\rangle_A \lvert1\rangle_R. \end{aligned}

At r=1r=1, one eigenvalue vanishes and the purification is a product state after discarding the unused reference direction. At r=0r=0, the qubit is maximally mixed and the direction n\boldsymbol n is irrelevant; every minimal purification is maximally entangled. One choice is

∣Φ+⟩AR=∣00⟩+∣11⟩2,Tr⁡R(∣Φ+⟩⟨Φ+∣)=I2.\begin{aligned} \lvert\Phi^+\rangle_{AR} &= \frac{ \lvert00\rangle+\lvert11\rangle }{ \sqrt2 }, \\ \operatorname{Tr}_R \bigl( \lvert\Phi^+\rangle \langle\Phi^+\rvert \bigr) &= \frac I2. \end{aligned}

Thus the Bloch radius measures both local purity and, for a purifying two-qubit pure state, the imbalance of the Schmidt coefficients:

Tr⁡(ρA2)=λ+2+λ−2=1+r22.\operatorname{Tr}(\rho_A^2) = \lambda_+^2+\lambda_-^2 = \frac{1+r^2}{2}.

The geometry of r\boldsymbol r itself is developed in Bloch Sphere.

If ∣Ψ⟩AR\lvert\Psi\rangle_{AR} purifies ρA\rho_A and URU_R is unitary, then

∣ΨU⟩AR=(IA⊗UR)∣Ψ⟩AR\lvert\Psi_U\rangle_{AR} = (I_A\otimes U_R) \lvert\Psi\rangle_{AR}

is another purification. Indeed,

Tr⁡R[(IA⊗UR)∣Ψ⟩⟨Ψ∣(IA⊗UR†)]=Tr⁡R(∣Ψ⟩⟨Ψ∣)=ρA.\begin{aligned} &\operatorname{Tr}_R \left[ (I_A\otimes U_R) \lvert\Psi\rangle\langle\Psi\rvert (I_A\otimes U_R^\dagger) \right] \\ &\qquad= \operatorname{Tr}_R \bigl( \lvert\Psi\rangle\langle\Psi\rvert \bigr) = \rho_A. \end{aligned}

Several distinctions matter:

  • Changing the reference basis changes the vector in the larger Hilbert space but not the local state on AA.
  • A degenerate spectral decomposition allows different eigenbases, but it does not create inequivalent local density operators.
  • Different physical preparation histories can yield the same ρA\rho_A without being identified by purification.
  • The density operator fixes all local statistics, not a unique joint state with a real environment.

For example,

I2=12(∣0⟩⟨0∣+∣1⟩⟨1∣)=12(∣+⟩⟨+∣+∣−⟩⟨−∣).\begin{aligned} \frac I2 &= \frac12 \bigl( \lvert0\rangle\langle0\rvert + \lvert1\rangle\langle1\rvert \bigr) \\ &= \frac12 \bigl( \lvert+\rangle\langle+\rvert + \lvert-\rangle\langle-\rvert \bigr). \end{aligned}

Measuring the reference qubit of ∣Φ+⟩\lvert\Phi^+\rangle in the computational basis realizes the first conditional ensemble; measuring it in the XX basis realizes the second. Discarding the outcome gives I/2I/2 in either case.

For a pure state on ARA R, the Schmidt decomposition gives

∣Ψ⟩AR=∑k=1rpk ∣ak⟩A∣rk⟩R.\lvert\Psi\rangle_{AR} = \sum_{k=1}^{r} \sqrt{p_k}\, \lvert a_k\rangle_A \lvert r_k\rangle_R.

Its reduced states have the same nonzero eigenvalues:

ρA=∑k=1rpk∣ak⟩⟨ak∣,ρR=∑k=1rpk∣rk⟩⟨rk∣.\begin{aligned} \rho_A &= \sum_{k=1}^{r} p_k \lvert a_k\rangle\langle a_k\rvert, \\ \rho_R &= \sum_{k=1}^{r} p_k \lvert r_k\rangle\langle r_k\rvert. \end{aligned}

Consequently:

  • ρA\rho_A is pure exactly when the Schmidt rank is one.
  • ρA\rho_A is mixed exactly when every purification is entangled between AA and RR.
  • The two marginals have equal purity, Tr⁡(ρA2)=Tr⁡(ρR2)\operatorname{Tr}(\rho_A^2)=\operatorname{Tr}(\rho_R^2).
  • Their von Neumann entropies agree whenever those entropies are finite.

For a globally pure bipartite state, the entropy of either marginal therefore quantifies entanglement across the bipartition. The same statement is not valid for a general mixed joint state, where local mixedness can reflect both classical uncertainty and quantum correlations. See Entropy Overview and Subsystem Entropy.

Purification is a statement about one state. A closely related but distinct theorem represents an entire quantum channel by reversible evolution on a larger space. In finite dimensions, a channel E\mathcal E admits an isometry

V:HA⟶HB⊗HE,V†V=IA,V: \mathcal H_A \longrightarrow \mathcal H_B\otimes\mathcal H_E, \qquad V^\dagger V=I_A,

such that

E(ρA)=Tr⁡E(VρAV†).\mathcal E(\rho_A) = \operatorname{Tr}_E \bigl( V\rho_A V^\dagger \bigr).

Equivalently, one may append an environment in a fixed pure state, apply a unitary on a sufficiently large joint space, and ignore part of the output. This Stinespring representation explains why the same pattern appears repeatedly:

  1. enlarge the Hilbert space,
  2. use pure-state or reversible structure there,
  3. take a partial trace to recover the accessible description.

The distinction is important. A purification reproduces one specified density operator; a channel dilation must reproduce the map for every input state and preserve all correlations with untouched reference systems.

The finite-dimensional proof extends to a positive trace-class operator on a separable Hilbert space. Such a density operator has a countable spectral resolution,

ρA=∑kpk∣k⟩⟨k∣,∑kpk=1,\rho_A = \sum_{k} p_k \lvert k\rangle\langle k\rvert, \qquad \sum_k p_k=1,

and the same series with coefficients pk\sqrt{p_k} defines a normalized purification on a suitable reference space. The required reference rank may now be infinite, and convergence is understood in the Hilbert-space and trace-class senses.

There are settings, especially algebraic formulations of quantum field theory, in which a state is defined as a positive normalized functional and need not be represented by a density operator on a preferred tensor factor. The elementary theorem should not be exported to those settings without checking the representation and subsystem assumptions.

Given a finite-dimensional ρA\rho_A:

  1. Check positivity and Tr⁡ρA=1\operatorname{Tr}\rho_A=1.
  2. Diagonalize ρA\rho_A if a Schmidt-form purification is wanted.
  3. Keep only nonzero eigenvalues to identify the minimal reference dimension.
  4. Attach orthonormal reference labels and use square-root amplitudes.
  5. Verify normalization.
  6. Trace the reference system explicitly.
  7. Interpret RR only after the mathematical construction is complete.

For numerical work, the square-root construction is often convenient. For conceptual work, the spectral construction makes rank, entanglement, and entropy transparent.

Thermal states also admit purifications. If

ρβ=e−βHZ=∑npn∣n⟩⟨n∣,pn=e−βEnZ,\rho_\beta = \frac{e^{-\beta H}}{Z} = \sum_n p_n \lvert n\rangle\langle n\rvert, \qquad p_n = \frac{e^{-\beta E_n}}{Z},

then a standard purification on two copies of the Hilbert space is

∣TFD(β)⟩=∑npn ∣n⟩L∣n⟩R=1Z∑ne−βEn/2∣n⟩L∣n⟩R.\lvert\mathrm{TFD}(\beta)\rangle = \sum_n \sqrt{p_n}\, \lvert n\rangle_L\lvert n\rangle_R = \frac{1}{\sqrt Z} \sum_n e^{-\beta E_n/2} \lvert n\rangle_L\lvert n\rangle_R.

Tracing over the right copy gives the thermal density operator on the left copy:

Tr⁡R(∣TFD(β)⟩⟨TFD(β)∣)=ρβ,L.\operatorname{Tr}_R \bigl( \lvert\mathrm{TFD}(\beta)\rangle \langle\mathrm{TFD}(\beta)\rvert \bigr) = \rho_{\beta,L}.

This thermofield-double construction is one bridge from ordinary density operators to finite-temperature many-body physics and QFT. The details depend on the Hamiltonian, spectrum, and field-theory setting; this page only gives the purification pattern.

  • Calling an ensemble decomposition itself a purification. A purification is one vector on a larger tensor-product space.
  • Using amplitudes pkp_k instead of pk\sqrt{p_k}.
  • Omitting orthogonality of the reference labels in an ensemble-labeled construction.
  • Assuming the purifying system must be a literal environment.
  • Treating a purification as unique.
  • Choosing dim⁡HR<rank⁡ρA\dim\mathcal H_R<\operatorname{rank}\rho_A.
  • Forgetting that a product purification exists only for a pure ρA\rho_A.
  • Inferring that a mixed subsystem makes the total state mixed.
  • Confusing state purification with a channel dilation.
  • Applying the finite-dimensional density-matrix theorem where no subsystem tensor factor or density operator has been specified.
  • M. A. Nielsen and I. L. Chuang, Quantum Computation and Quantum Information, 10th anniversary edition, Cambridge University Press, 2010, Chapters 2 and 8. Cambridge DOI.
  • J. Watrous, The Theory of Quantum Information, Cambridge University Press, 2018, Sections 2.1–2.2. Author’s book page and manuscript.
  • L. P. Hughston, R. Jozsa, and W. K. Wootters, “A complete classification of quantum ensembles having a given density matrix,” Physics Letters A 183, 14–18 (1993). DOI: 10.1016/0375-9601(93)90880-9.
  • W. F. Stinespring, “Positive functions on C*-algebras,” Proceedings of the American Mathematical Society 6, 211–216 (1955). DOI: 10.1090/S0002-9939-1955-0069403-4.
  • A. Uhlmann, “The Transition Probability in the State Space of a C*-Algebra,” Reports on Mathematical Physics 9, 273–279 (1976).
  • A. Peres, Quantum Theory: Concepts and Methods, Kluwer (1995).
  • H. Umezawa, Advanced Field Theory: Micro, Macro, and Thermal Physics, American Institute of Physics (1993).
  1. Let
ρA=∑k=1rpk∣k⟩⟨k∣\rho_A = \sum_{k=1}^{r} p_k \lvert k\rangle\langle k\rvert

with pk>0p_k>0. Verify both normalization and the reduced-state condition for the spectral purification.

Solution

Set

∣Ψ⟩AR=∑k=1rpk ∣k⟩A∣k⟩R.\lvert\Psi\rangle_{AR} = \sum_{k=1}^{r} \sqrt{p_k}\, \lvert k\rangle_A\lvert k\rangle_R.

Orthonormality in both factors gives

⟨Ψ∣Ψ⟩=∑j,kpjpk δjkδjk=∑kpk=1.\langle\Psi\vert\Psi\rangle = \sum_{j,k} \sqrt{p_jp_k}\, \delta_{jk}\delta_{jk} = \sum_k p_k =1.

For the partial trace,

Tr⁡R(∣Ψ⟩⟨Ψ∣)=∑j,kpjpk ∣j⟩⟨k∣A×⟨k∣j⟩R=∑kpk∣k⟩⟨k∣A=ρA.\begin{aligned} \operatorname{Tr}_R \bigl( \lvert\Psi\rangle\langle\Psi\rvert \bigr) &= \sum_{j,k} \sqrt{p_jp_k}\, \lvert j\rangle\langle k\rvert_A \\ &\quad{}\times \langle k\vert j\rangle_R \\ &= \sum_k p_k \lvert k\rangle\langle k\rvert_A = \rho_A. \end{aligned}
  1. Prove that a density operator of rank rr cannot be purified using a reference space of dimension smaller than rr.
Solution

Any vector on HA⊗HR\mathcal H_A\otimes\mathcal H_R has Schmidt rank at most

min⁡(dim⁡HA,dim⁡HR).\min \bigl( \dim\mathcal H_A, \dim\mathcal H_R \bigr).

For a pure bipartite state, the rank of either reduced density operator equals the Schmidt rank. If the reduced state on AA has rank rr, then the purifying vector has Schmidt rank rr, so

dim⁡HR≥r.\dim\mathcal H_R\ge r.

The spectral construction uses exactly rr orthogonal reference vectors, so the bound is attainable.

  1. Prove the square-root construction directly in a fixed basis.
Solution

With

∣Ω⟩=∑k∣k⟩A∣k⟩R,\lvert\Omega\rangle = \sum_k \lvert k\rangle_A\lvert k\rangle_R,

the proposed vector has coefficients

∣Ψρ⟩=∑j,k(ρ)jk∣j⟩A∣k⟩R.\lvert\Psi_\rho\rangle = \sum_{j,k} (\sqrt\rho)_{jk} \lvert j\rangle_A\lvert k\rangle_R.

Tracing RR gives

Tr⁡R(∣Ψρ⟩⟨Ψρ∣)=∑j,j′,k(ρ)jk(ρ)j′k∗×∣j⟩⟨j′∣=ρ ρ†=ρ.\begin{aligned} \operatorname{Tr}_R \bigl( \lvert\Psi_\rho\rangle \langle\Psi_\rho\rvert \bigr) &= \sum_{j,j',k} (\sqrt\rho)_{jk} (\sqrt\rho)_{j'k}^{*} \\ &\quad{}\times \lvert j\rangle\langle j'\rvert \\ &= \sqrt\rho\,\sqrt\rho^{\dagger} = \rho. \end{aligned}

Taking the trace of this equality gives ⟨Ψρ∣Ψρ⟩=Tr⁡ρ=1\langle\Psi_\rho\vert\Psi_\rho\rangle=\operatorname{Tr}\rho=1.

  1. The nonorthogonal ensemble
ρ=12∣0⟩⟨0∣+12∣+⟩⟨+∣\rho = \frac12 \lvert0\rangle\langle0\rvert + \frac12 \lvert+\rangle\langle+\rvert

is given. Construct an ensemble-labeled purification and verify it.

Solution

Choose orthogonal reference states ∣0⟩R\lvert0\rangle_R and ∣1⟩R\lvert1\rangle_R:

∣Φ⟩AR=12(∣0⟩A∣0⟩R+∣+⟩A∣1⟩R).\lvert\Phi\rangle_{AR} = \frac{1}{\sqrt2} \left( \lvert0\rangle_A\lvert0\rangle_R + \lvert+\rangle_A\lvert1\rangle_R \right).

The vector is normalized because the reference labels are orthogonal:

⟨Φ∣Φ⟩=12+12=1.\langle\Phi\vert\Phi\rangle = \frac12+\frac12 =1.

When RR is traced out, the cross terms contain ⟨0∣1⟩R\langle0\vert1\rangle_R or ⟨1∣0⟩R\langle1\vert0\rangle_R and vanish. The remaining terms are

Tr⁡R(∣Φ⟩⟨Φ∣)=12∣0⟩⟨0∣+12∣+⟩⟨+∣.\operatorname{Tr}_R \bigl( \lvert\Phi\rangle\langle\Phi\rvert \bigr) = \frac12 \lvert0\rangle\langle0\rvert + \frac12 \lvert+\rangle\langle+\rvert.

The two states on AA need not be orthogonal because orthogonality of the labels on RR performs the required bookkeeping.

  1. Show that a unitary on the reference system cannot change the reduced state on AA.
Solution

Choose any orthonormal basis {∣a⟩R}\{\lvert a\rangle_R\}. For an operator XARX_{AR},

Tr⁡RXAR=∑a(IA⊗⟨a∣)XAR×(IA⊗∣a⟩).\begin{aligned} \operatorname{Tr}_R X_{AR} &= \sum_a (I_A\otimes\langle a\rvert) X_{AR} \\ &\quad{}\times (I_A\otimes\lvert a\rangle). \end{aligned}

Apply this to

XAR=(IA⊗UR)∣Ψ⟩⟨Ψ∣(IA⊗UR†).X_{AR} = (I_A\otimes U_R) \lvert\Psi\rangle\langle\Psi\rvert (I_A\otimes U_R^\dagger).

The vectors UR†∣a⟩U_R^\dagger\lvert a\rangle also form an orthonormal basis, so the basis sum is exactly the original partial trace. Therefore

Tr⁡RXAR=Tr⁡R(∣Ψ⟩⟨Ψ∣)=ρA.\operatorname{Tr}_R X_{AR} = \operatorname{Tr}_R \bigl( \lvert\Psi\rangle\langle\Psi\rvert \bigr) = \rho_A.
  1. For a qubit with Bloch radius rr, determine the Schmidt coefficients of a minimal purification, its reduced purity, and the values of rr for which the purification is entangled.
Solution

The eigenvalues of the qubit state are

λ±=1±r2.\lambda_\pm = \frac{1\pm r}{2}.

Hence the Schmidt coefficients are

λ+=1+r2,λ−=1−r2.\sqrt{\lambda_+} = \sqrt{\frac{1+r}{2}}, \qquad \sqrt{\lambda_-} = \sqrt{\frac{1-r}{2}}.

The reduced purity is

Tr⁡(ρ2)=λ+2+λ−2=(1+r)2+(1−r)24=1+r22.\begin{aligned} \operatorname{Tr}(\rho^2) &= \lambda_+^2+\lambda_-^2 \\ &= \frac{(1+r)^2+(1-r)^2}{4} \\ &= \frac{1+r^2}{2}. \end{aligned}

Both Schmidt coefficients are nonzero when 0≤r<10\le r<1, so every minimal purification is entangled in that range. At r=1r=1, one coefficient vanishes and the purification is a product state.

  1. Starting from ∣Φ+⟩=(∣00⟩+∣11⟩)/2\lvert\Phi^+\rangle=(\lvert00\rangle+\lvert11\rangle)/\sqrt2, show how measurements of RR produce two different ensembles of I/2I/2.
Solution

In the computational basis,

∣Φ+⟩=∣0⟩A∣0⟩R+∣1⟩A∣1⟩R2.\lvert\Phi^+\rangle = \frac{ \lvert0\rangle_A\lvert0\rangle_R + \lvert1\rangle_A\lvert1\rangle_R }{ \sqrt2 }.

A measurement of RR in that basis prepares ∣0⟩A\lvert0\rangle_A or ∣1⟩A\lvert1\rangle_A, each with probability 1/21/2.

Using

∣Φ+⟩=∣+⟩A∣+⟩R+∣−⟩A∣−⟩R2,\lvert\Phi^+\rangle = \frac{ \lvert+\rangle_A\lvert+\rangle_R + \lvert-\rangle_A\lvert-\rangle_R }{ \sqrt2 },

a measurement of RR in the XX basis instead prepares ∣+⟩A\lvert+\rangle_A or ∣−⟩A\lvert-\rangle_A, again with probability 1/21/2. Ignoring the outcome gives

ρA=I2\rho_A = \frac I2

for either measurement. The conditional ensembles differ, while the unconditioned density operator does not.

  1. Let a channel have Kraus operators {Kα}\{K_\alpha\} satisfying
∑αKα†Kα=IA.\sum_\alpha K_\alpha^\dagger K_\alpha = I_A.

Define

V=∑αKα⊗∣α⟩E.V = \sum_\alpha K_\alpha\otimes \lvert\alpha\rangle_E.

Show that VV is an isometry and that tracing EE reproduces the Kraus form of the channel.

Solution

Using orthonormality of the environment labels,

V†V=∑α,βKα†Kβ⟨α∣β⟩E=∑αKα†Kα=IA.\begin{aligned} V^\dagger V &= \sum_{\alpha,\beta} K_\alpha^\dagger K_\beta \langle\alpha\vert\beta\rangle_E \\ &= \sum_\alpha K_\alpha^\dagger K_\alpha = I_A. \end{aligned}

Thus VV preserves inner products. For an input ρ\rho,

VρV†=∑α,βKαρKβ†⊗∣α⟩⟨β∣E.V\rho V^\dagger = \sum_{\alpha,\beta} K_\alpha\rho K_\beta^\dagger \otimes \lvert\alpha\rangle\langle\beta\rvert_E.

The partial trace sets α=β\alpha=\beta:

Tr⁡E(VρV†)=∑αKαρKα†=E(ρ).\operatorname{Tr}_E \bigl( V\rho V^\dagger \bigr) = \sum_\alpha K_\alpha\rho K_\alpha^\dagger = \mathcal E(\rho).

This is a dilation of a map on all inputs, whereas a state purification represents one chosen density operator.

Additional exercises retained from the earlier canonical treatment

Section titled “Additional exercises retained from the earlier canonical treatment”
  1. Verify the qubit purification formula.
Solution

Let

∣Ψ⟩AR=p ∣0⟩A∣0⟩R+1−p ∣1⟩A∣1⟩R.\lvert\Psi\rangle_{AR} = \sqrt p\,\lvert0\rangle_A\lvert0\rangle_R + \sqrt{1-p}\,\lvert1\rangle_A\lvert1\rangle_R.

The density operator contains diagonal terms and cross terms:

∣Ψ⟩⟨Ψ∣=p∣00⟩⟨00∣+p(1−p)∣00⟩⟨11∣+p(1−p)∣11⟩⟨00∣+(1−p)∣11⟩⟨11∣.\begin{aligned} \lvert\Psi\rangle\langle\Psi\rvert &= p\lvert00\rangle\langle00\rvert + \sqrt{p(1-p)} \lvert00\rangle\langle11\rvert\\ &\quad + \sqrt{p(1-p)} \lvert11\rangle\langle00\rvert + (1-p)\lvert11\rangle\langle11\rvert. \end{aligned}

Tracing over RR removes the cross terms because ⟨1∣0⟩=0\langle1\vert0\rangle=0, leaving

ρA=p∣0⟩⟨0∣+(1−p)∣1⟩⟨1∣.\rho_A = p\lvert0\rangle\langle0\rvert + (1-p)\lvert1\rangle\langle1\rvert.
  1. What is the minimal purifying dimension for a density operator of rank rr?
Solution

The minimal dimension is rr. The spectral construction uses rr orthonormal reference states. A smaller reference system cannot work because a pure state on ARA R has Schmidt rank at most dim⁡HR\dim\mathcal H_R, and the rank of the reduced state equals the Schmidt rank.

  1. Show that applying a unitary to the purifying system does not change ρA\rho_A.
Solution

Let

ρAR′=(IA⊗UR)ρAR(IA⊗UR†).\rho'_{AR} = (I_A\otimes U_R)\rho_{AR}(I_A\otimes U_R^\dagger).

For any operator MAM_A on AA,

Tr⁡A[(Tr⁡RρAR′)MA]=Tr⁡AR[ρAR′(MA⊗IR)]=Tr⁡AR[ρAR(MA⊗UR†IRUR)]=Tr⁡AR[ρAR(MA⊗IR)].\begin{aligned} \operatorname{Tr}_A \bigl[ (\operatorname{Tr}_R\rho'_{AR})M_A \bigr] &= \operatorname{Tr}_{AR} \bigl[ \rho'_{AR}(M_A\otimes I_R) \bigr]\\ &= \operatorname{Tr}_{AR} \bigl[ \rho_{AR}(M_A\otimes U_R^\dagger I_R U_R) \bigr]\\ &= \operatorname{Tr}_{AR} \bigl[ \rho_{AR}(M_A\otimes I_R) \bigr]. \end{aligned}

Since all local expectation values on AA are unchanged, the reduced state on AA is unchanged.

  1. Which Bell state purifies the maximally mixed qubit state I/2I/2?
Solution

Any Bell state does. For example,

∣Φ+⟩=12(∣00⟩+∣11⟩)\lvert\Phi^+\rangle = \frac{1}{\sqrt2} \bigl( \lvert00\rangle+\lvert11\rangle \bigr)

has

Tr⁡R(∣Φ+⟩⟨Φ+∣)=12I.\operatorname{Tr}_R \bigl( \lvert\Phi^+\rangle\langle\Phi^+\rvert \bigr) = \frac12 I.
  1. Show that the thermofield-double state purifies the thermal density operator.
Solution

For

∣TFD(β)⟩=∑npn ∣n⟩L∣n⟩R,\lvert\mathrm{TFD}(\beta)\rangle = \sum_n \sqrt{p_n}\, \lvert n\rangle_L\lvert n\rangle_R,

the density operator is

∑m,npmpn ∣m⟩L⟨n∣L⊗∣m⟩R⟨n∣R.\sum_{m,n} \sqrt{p_mp_n}\, \lvert m\rangle_L\langle n\rvert_L \otimes \lvert m\rangle_R\langle n\rvert_R.

Tracing over RR gives Tr⁡(∣m⟩R⟨n∣R)=δmn\operatorname{Tr}(\lvert m\rangle_R\langle n\rvert_R)=\delta_{mn}, so

Tr⁡R(∣TFD⟩⟨TFD∣)=∑npn∣n⟩L⟨n∣L=ρβ,L.\operatorname{Tr}_R \bigl( \lvert\mathrm{TFD}\rangle\langle\mathrm{TFD}\rvert \bigr) = \sum_n p_n \lvert n\rangle_L\langle n\rvert_L = \rho_{\beta,L}.