Bipartite Entanglement

Bipartite entanglement is nonseparability across a specified split into two subsystems, $A$ and $B$. For pure states, a single canonical construction—the Schmidt decomposition—solves the structural problem. For mixed states, separability is a convex problem and no equally simple universal diagnostic exists.

That difference determines the chapter’s organization:

$$\begin{aligned} \text{pure state} &\longrightarrow \text{Schmidt data}, \\ \text{mixed state} &\longrightarrow \text{separability criteria}. \end{aligned}$$

The first question should therefore be whether the joint state is pure or mixed. The second should be whether the task is to detect, quantify, or operationally use entanglement. Those are different tasks and may require different tools.

Chapter Map

Task	Canonical page	Main output
put a pure state in canonical bipartite form	Schmidt Decomposition	Schmidt bases, coefficients, and reduced spectra
classify pure-state factorization	Schmidt Rank	product rank one versus entangled rank greater than one
quantify pure-state bipartite entanglement	Entanglement Entropy	von Neumann entropy of either reduced state
compare spectral moments	Rényi Entropies	a family of reduced-state entropy measures
quantify total correlation	Mutual Information	classical plus quantum correlation, not entanglement alone
use the special two-qubit mixed-state formula	Concurrence for Two Qubits	two-qubit concurrence and its limits
test partial-transpose positivity	Negativity and the PPT Criterion	NPT detection, negativity, and PPT limitations
detect entanglement through observables	Entanglement Witnesses	one-sided, experimentally accessible certification
interpret entanglement as a resource	LOCC Preview	local operations, classical communication, and monotonicity

Fix the Bipartition First

Let

$$\mathcal H_{AB} = \mathcal H_A\otimes\mathcal H_B.$$

The factors must have a physical meaning: two particles, two modes, two spatial regions, two registers, or another operationally defined pair. Entanglement is invariant under local basis changes inside $A$ and $B$, but it can change if the physical factorization itself changes.

For a pure state $\lvert\psi\rangle_{AB}$,

$$\lvert\psi\rangle\text{ is entangled} \iff \lvert\psi\rangle \neq \lvert a\rangle_A\otimes\lvert b\rangle_B.$$

For a mixed state $\rho_{AB}$,

$$\begin{gathered} \rho_{AB}\text{ is entangled} \\ \iff \rho_{AB}\text{ is not separable} \\ \iff \rho_{AB} \neq \sum_k p_k\, \rho_A^{(k)}\otimes\rho_B^{(k)}. \end{gathered}$$

for every probability distribution $\{p_k\}$ and every collection of subsystem density operators. The inequality means that no separable decomposition exists; failing to find one is not, by itself, a proof.

Pure States: Schmidt Data Solve the Problem

Every normalized pure state of a finite-dimensional bipartite system admits a Schmidt decomposition

$$\lvert\psi\rangle_{AB} = \sum_{r=1}^{R} \sqrt{\lambda_r}\, \lvert u_r\rangle_A \lvert v_r\rangle_B,$$

where

$$\lambda_r>0, \qquad \sum_{r=1}^{R}\lambda_r=1,$$

and the two Schmidt families are orthonormal. The positive integer $R$ is the Schmidt rank.

The reduced states are diagonal in the Schmidt bases:

$$\begin{aligned} \rho_A &= \sum_{r=1}^{R}\lambda_r \lvert u_r\rangle\langle u_r\rvert, \\ \rho_B &= \sum_{r=1}^{R}\lambda_r \lvert v_r\rangle\langle v_r\rvert. \end{aligned}$$

Thus the nonzero spectra of $\rho_A$ and $\rho_B$ agree. The Schmidt coefficients determine all pure-state bipartite entanglement properties that are invariant under local unitaries.

Equivalent pure-state tests

For a finite-dimensional pure state, the following statements are equivalent:

$$\begin{aligned} \lvert\psi\rangle\text{ is product} &\iff R=1, \\ &\iff \operatorname{rank}\rho_A=1, \\ &\iff \operatorname{Tr}(\rho_A^2)=1, \\ &\iff S(\rho_A)=0. \end{aligned}$$

If the state is written in product bases as

$$\lvert\psi\rangle = \sum_{i,j} C_{ij} \lvert i\rangle_A\lvert j\rangle_B,$$

then $R=\operatorname{rank}C$. Numerically, the singular values of $C$ are $\sqrt{\lambda_r}$.

Pure-State Quantifiers

Entanglement entropy

For a bipartite pure state, the entanglement entropy is

$$\begin{aligned} E(\psi) &= S(\rho_A) = S(\rho_B) \\ &= -\sum_r\lambda_r\log\lambda_r. \end{aligned}$$

It is zero for product states and reaches $\log d$ when the Schmidt spectrum is uniform over $d=\min(d_A,d_B)$ nonzero coefficients. The logarithm base sets the unit: base two gives bits and base $e$ gives nats.

Rényi entropies

For $\alpha>0$ with $\alpha\neq1$,

$$\begin{aligned} S_\alpha(\rho_A) &= \frac{1}{1-\alpha} \log\operatorname{Tr}(\rho_A^\alpha) \\ &= \frac{1}{1-\alpha} \log\sum_r\lambda_r^\alpha. \end{aligned}$$

Different orders weight the Schmidt spectrum differently. In particular,

$$S_2(\rho_A) = -\log\operatorname{Tr}(\rho_A^2).$$

The limit $\alpha\to1$ gives the von Neumann entropy under the usual continuity conditions. Rényi entropies are useful in many-body numerics, experiments, and replica constructions, but different orders are not interchangeable summaries.

A One-Parameter Pure-State Example

Consider

$$\begin{aligned} \lvert\psi(\theta)\rangle &= \cos\theta\,\lvert00\rangle + \sin\theta\,\lvert11\rangle, \\ 0&\leq\theta\leq\frac{\pi}{4}. \end{aligned}$$

This is already in Schmidt form, with

$$\lambda_1=\cos^2\theta, \qquad \lambda_2=\sin^2\theta.$$

Its main diagnostics are

$$\begin{aligned} R&= \begin{cases} 1,&\theta=0,\\ 2,&0<\theta\leq\pi/4, \end{cases} \\ E(\theta) &= -\cos^2\theta\log(\cos^2\theta) \\ &\quad- \sin^2\theta\log(\sin^2\theta), \\ C(\theta)&=\sin(2\theta), \end{aligned}$$

where $C$ is the pure two-qubit concurrence. At $\theta=0$ the state is product. At $\theta=\pi/4$ it is a Bell state with a uniform Schmidt spectrum and maximal two-qubit entanglement.

This family is useful because rank, entropy, and concurrence agree on the endpoints while encoding different kinds of information between them.

Mixed States: Separate Detection from Quantification

For mixed states, local mixedness is inconclusive. A reduced state may be mixed because of entanglement, classical correlation, preparation uncertainty, or combinations of these.

No elementary analogue of Schmidt rank classifies arbitrary mixed states. Use a claim whose strength and scope match the available criterion.

Tool	Output	Correct scope
explicit separable decomposition	proves separability	sufficient in every finite bipartite dimension
negative partial transpose	proves entanglement	sufficient in every finite bipartite dimension
positive partial transpose	proves separability only in low dimensions	necessary in general; sufficient for two-by-two and two-by-three systems
negative witness expectation	proves entanglement	detects states separated by that witness
concurrence	detects and quantifies in the specified two-qubit setting	not a general high-dimensional measure
mutual information	quantifies total correlation	includes both classical and quantum correlation

The absence of a positive detection result does not usually prove separability. For example, a PPT state in higher dimension may still be entangled.

Partial Transpose and Negativity

Choose a basis $\{\lvert j\rangle_B\}$ and partially transpose subsystem $B$:

$$\left( \lvert i\rangle\langle k\rvert_A \otimes \lvert j\rangle\langle \ell\rvert_B \right)^{T_B} = \lvert i\rangle\langle k\rvert_A \otimes \lvert \ell\rangle\langle j\rvert_B.$$

Every separable state has positive partial transpose:

$$\rho_{AB}\text{ separable} \quad\Longrightarrow\quad \rho_{AB}^{T_B}\geq0.$$

Therefore a negative eigenvalue of $\rho_{AB}^{T_B}$ certifies entanglement. The negativity packages the negative spectrum into

$$\mathcal N(\rho) = \frac{ \lVert\rho^{T_B}\rVert_1-1 }{2}.$$

If $\mathcal N>0$, the state is entangled. If $\mathcal N=0$, the state is PPT; that proves separability only in the special dimensions stated above. The partial transpose is a mathematical diagnostic, not a physically implementable quantum channel on one side of an unknown state.

Entanglement Witnesses

Write $\operatorname{Sep}(A{:}B)$ for the set of separable states across the chosen split. An entanglement witness is a Hermitian operator $W$ satisfying

$$\begin{aligned} \operatorname{Tr}(W\sigma) &\geq0 \quad \forall\sigma\in\operatorname{Sep}(A{:}B), \\ \operatorname{Tr}(W\rho) &<0 \quad \text{for some }\rho. \end{aligned}$$

The witness defines a separating hyperplane between a target entangled state and the convex set of separable states. It can often be decomposed into locally measurable observables, making it experimentally useful.

A witness is one-sided. A negative value certifies entanglement; a nonnegative value means only that this witness did not detect the state. Witness design, noise tolerance, and measurement cost belong to the detailed Entanglement Witnesses page.

Concurrence Is a Special Two-Qubit Tool

For a pure two-qubit state, concurrence may be written

$$C(\psi) = 2\sqrt{\det\rho_A}.$$

It ranges from zero for product states to one for maximally entangled two-qubit states. A closed formula also exists for mixed two-qubit states through the spin-flipped density operator.

That formula is powerful precisely because it is specialized. It should not be applied to arbitrary local dimensions or treated as the unique meaning of entanglement. The construction and its relation to entanglement of formation are developed in Concurrence for Two Qubits.

Mutual Information Measures Total Correlation

For any bipartite state,

$$I(A{:}B) = S(\rho_A)+S(\rho_B)-S(\rho_{AB}).$$

It is nonnegative and vanishes exactly for product states in finite dimensions. For a pure bipartite state,

$$I(A{:}B) = 2S(\rho_A),$$

so it is twice the entanglement entropy. For a mixed state, it includes classical and quantum correlations and is not an entanglement measure.

The distinction is visible in two states with identical marginals. Using base-two logarithms,

Joint state	$S(A)$	$S(B)$	$S(AB)$	$I(A{:}B)$
diagonal mixture of matched bits	$1$	$1$	$1$	$1$
Bell state	$1$	$1$	$0$	$2$

The diagonal mixture is separable; the Bell state is entangled. Mutual information correctly distinguishes their total correlation but does not, in general, isolate the entangled part.

Local Operations and Classical Communication

LOCC protocols allow local quantum operations on $A$ and $B$, together with classical messages that can influence later local operations. They cannot create entanglement from a separable input.

For a separable state

$$\rho_{AB} = \sum_k p_k\, \rho_A^{(k)}\otimes\rho_B^{(k)},$$

each local branch maps product operators to product operators, and classical mixing preserves convex combinations. Consequently, every LOCC output obtained from a separable input remains separable.

This monotonicity motivates the resource viewpoint: entanglement is something joint preparation may supply and LOCC cannot freely manufacture. Exact state-conversion criteria, asymptotic rates, catalytic effects, and protocol design belong to Quantum Information; LOCC Preview establishes the boundary used here.

A Reliable Analysis Workflow

1. Specify the state and split

Record $\rho_{AB}$, the dimensions $d_A$ and $d_B$, and the physical meaning of the partition. Verify positivity and unit trace for a density operator.

2. Decide pure or mixed

Check whether $\operatorname{Tr}(\rho_{AB}^2)=1$. Pure states admit exact Schmidt analysis. Mixed states require separability reasoning.

3. Match the tool to the question

• For pure-state classification, use Schmidt rank or reduced-state purity.
• For pure-state quantification, use the Schmidt spectrum and a stated entropy.
• For total correlation, use mutual information.
• For low-dimensional mixed-state detection, consider PPT, negativity, concurrence, or a witness with its scope stated.
• For operational conversion, specify the allowed class of operations, such as LOCC.

4. Report what the result proves

Prefer a scoped conclusion such as “the state is NPT and therefore entangled across $A$ and $B$” over “the entanglement test succeeded.” State whether the calculation detects, quantifies, or only bounds entanglement.

5. Cross-check limiting cases

Test product limits, maximally entangled limits, local-unitary invariance, trace normalization, and continuity under small perturbations where appropriate.

Common Mistakes

• Using a mixed marginal as a general entanglement test. This works for a pure global state, not for an arbitrary mixed one.
• Applying Schmidt rank to density matrices. Schmidt rank classifies pure bipartite vectors; mixed-state generalizations are different and more difficult.
• Calling mutual information entanglement. It measures total correlation.
• Treating PPT as universally sufficient. Positive partial transpose does not exclude bound entanglement in higher dimensions.
• Interpreting a failed witness as separability. A witness detects only part of the entangled set.
• Exporting concurrence beyond two qubits. The closed formula and normalization are setting-specific.
• Comparing entropy values with different logarithm bases. State the units.
• Forgetting the partition. Entanglement is a relation between specified subsystems.
• Assuming all entanglement measures impose the same ordering. Different measures answer different operational or spectral questions.

Reading Paths

Pure-state core: Schmidt Decomposition → Schmidt Rank → Entanglement Entropy → Rényi Entropies.

Correlation versus entanglement: Mutual Information → Classical Correlation versus Entanglement → Subsystem Entropy.

Mixed-state diagnostics: Negativity and the PPT Criterion → Entanglement Witnesses → Concurrence for Two Qubits.

Resource boundary: LOCC Preview → Entanglement and Quantum Information → Entanglement Diagnostic Table.

References

• E. Schmidt, “Zur Theorie der linearen und nichtlinearen Integralgleichungen,” Mathematische Annalen 63, 433–476, 1907.
• A. Peres, “Separability Criterion for Density Matrices,” Physical Review Letters 77, 1413–1415, 1996.
• M. Horodecki, P. Horodecki, and R. Horodecki, “Separability of Mixed States: Necessary and Sufficient Conditions,” Physics Letters A 223, 1–8, 1996.
• W. K. Wootters, “Entanglement of Formation of an Arbitrary State of Two Qubits,” Physical Review Letters 80, 2245–2248, 1998.
• G. Vidal and R. F. Werner, “Computable Measure of Entanglement,” Physical Review A 65, 032314, 2002.
• M. A. Nielsen and I. L. Chuang, Quantum Computation and Quantum Information, 10th anniversary ed., Cambridge University Press, 2010.
• R. Horodecki, P. Horodecki, M. Horodecki, and K. Horodecki, “Quantum Entanglement,” Reviews of Modern Physics 81, 865–942, 2009.
• J. Watrous, The Theory of Quantum Information, Cambridge University Press, 2018.

Exercises

Exercise 1: Read Schmidt data directly

For

$$\lvert\psi\rangle = \sqrt{\frac12}\lvert00\rangle + \sqrt{\frac13}\lvert11\rangle + \sqrt{\frac16}\lvert22\rangle,$$

find the Schmidt rank, the reduced-state spectrum, and the entanglement entropy.

Solution

The state is already in Schmidt form. All three coefficients are nonzero, so $R=3$. Both reduced states have spectrum

$$\left\{ \frac12,\frac13,\frac16 \right\}.$$

Therefore

$$E(\psi) = -\frac12\log\frac12 -\frac13\log\frac13 -\frac16\log\frac16.$$

The logarithm base determines whether the answer is reported in bits or nats.

Exercise 2: Coefficient-matrix test

Determine for which complex values of $a,b,c,d$ the normalized two-qubit vector

$$\lvert\psi\rangle = a\lvert00\rangle +b\lvert01\rangle +c\lvert10\rangle +d\lvert11\rangle$$

is a product state.

Solution

The coefficient matrix is

$$C = \begin{pmatrix} a & b \\ c & d \end{pmatrix}.$$

A nonzero normalized state is product exactly when $C$ has rank one. For a $2\times2$ matrix this is equivalent to

$$\det C = ad-bc =0.$$

If $ad-bc\neq0$, the Schmidt rank is two and the state is entangled.

Exercise 3: Partial transpose of a Bell state

Find the eigenvalues of the partial transpose of

$$\rho_{\Phi^+} = \lvert\Phi^+\rangle\langle\Phi^+\rvert, \qquad \lvert\Phi^+\rangle = \frac{\lvert00\rangle+\lvert11\rangle}{\sqrt2}.$$

Use them to compute the negativity.

Solution

In the computational basis,

$$\rho_{\Phi^+}^{T_B} = \frac12 \begin{pmatrix} 1&0&0&0 \\ 0&0&1&0 \\ 0&1&0&0 \\ 0&0&0&1 \end{pmatrix}.$$

Its eigenvalues are

$$\left\{ \frac12,\frac12,\frac12,-\frac12 \right\}.$$

The negative eigenvalue proves entanglement. The trace norm is $2$, so

$$\mathcal N = \frac{2-1}{2} = \frac12.$$

Exercise 4: Mutual information is not entanglement

Using base-two logarithms, compute $I(A{:}B)$ for

$$\rho_{\mathrm{cc}} = \frac12 \left( \lvert00\rangle\langle00\rvert + \lvert11\rangle\langle11\rvert \right).$$

Explain why the result does not imply entanglement.

Solution

Both marginals are $I/2$, so $S(A)=S(B)=1$. The joint state has two equal nonzero eigenvalues, so $S(AB)=1$. Hence

$$I(A{:}B) = 1+1-1 =1\text{ bit}.$$

The state is nevertheless separable because its displayed form is a convex mixture of the product projectors $\lvert00\rangle\langle00\rvert$ and $\lvert11\rangle\langle11\rvert$. Mutual information measures its classical correlation as well as any quantum correlation.

Exercise 5: Local-unitary invariance of Schmidt data

Let

$$\lvert\psi'\rangle = (U_A\otimes U_B)\lvert\psi\rangle.$$

Show that $\lvert\psi\rangle$ and $\lvert\psi'\rangle$ have the same Schmidt coefficients.

Solution

If

$$\lvert\psi\rangle = \sum_r\sqrt{\lambda_r}\, \lvert u_r\rangle_A\lvert v_r\rangle_B,$$

then

$$\lvert\psi'\rangle = \sum_r\sqrt{\lambda_r}\, (U_A\lvert u_r\rangle_A) (U_B\lvert v_r\rangle_B).$$

Unitary maps preserve inner products, so the transformed local families remain orthonormal. This is a Schmidt decomposition with the same $\{\lambda_r\}$. Consequently Schmidt rank, entanglement entropy, and every function only of the Schmidt spectrum are local-unitary invariants.