Skip to content

Entanglement is defined after choosing a state and a subsystem partition. The Many-Body Entanglement Overview explains how the definitions below connect to correlations and numerical representations. For a tensor-product bipartition H=HA⊗HAˉ\mathcal H=\mathcal H_A\otimes\mathcal H_{\bar A},

ρA=Tr⁡Aˉρ.\rho_A=\operatorname{Tr}_{\bar A}\rho.

Gauge constraints, identical particles, mode partitions, and algebraic subsystems may not admit this naive factorization and require their own conventions.

Term Compact meaning
Pure global state ρ=∣ψ⟩⟨ψ∣\rho=\lvert\psi\rangle\langle\psi\rvert; then S(A)=S(Aˉ)S(A)=S(\bar A)
Mixed global state Classical uncertainty and quantum entanglement can both contribute to subsystem entropy
Schmidt decomposition ∣ψ⟩=∑isi∣iA⟩∣iAˉ⟩\lvert\psi\rangle=\sum_i s_i\lvert i_A\rangle\lvert i_{\bar A}\rangle, si≥0s_i\ge0, ∑isi2=1\sum_i s_i^2=1
Schmidt rank Number of nonzero sis_i; partition dependent
Spatial cut Degrees of freedom assigned by region; geometry and boundary conditions matter
Mode or orbital cut Partition by single-particle labels rather than real-space locality
Algebraic subsystem A chosen observable subalgebra when tensor factorization is absent or unnatural
Object Definition or use
Purity Tr⁡ρA2\operatorname{Tr}\rho_A^2; equals one only for a pure reduced state
von Neumann entropy S(A)=−Tr⁡(ρAln⁡ρA)S(A)=-\operatorname{Tr}(\rho_A\ln\rho_A)
Rényi entropy Sn(A)=(1−n)−1ln⁡Tr⁡ρAnS_n(A)=(1-n)^{-1}\ln\operatorname{Tr}\rho_A^n, n>0n>0, n≠1n\ne1; the n→1n\to1 limit gives S(A)S(A)
Mutual information I(A:B)=S(A)+S(B)−S(AB)I(A:B)=S(A)+S(B)-S(AB); includes total correlations
Negativity N=(∥ρTB∥1−1)/2\mathcal N=(\lVert\rho^{T_B}\rVert_1-1)/2; a positive value detects entanglement, while zero does not exclude all entangled states
Entanglement spectrum Eigenvalues of ρA\rho_A, or levels ϵi=−ln⁡λi\epsilon_i=-\ln\lambda_i
Discarded weight w=∑i>χsi2w=\sum_{i>\chi}s_i^2; keeping and renormalizing the largest χ\chi coefficients gives squared overlap 1−w1-w with the original pure state
Operator entanglement Entanglement of a normalized vectorized operator under a specified doubled-space cut; not generally its entangling power

Entropy of a subsystem of a mixed global state is not, by itself, an entanglement measure. Logarithm base sets the units: natural logarithms give nats; base two gives bits.

Term Meaning and qualification
Area law Leading entropy scales with the boundary size; subleading terms may carry essential physics
Logarithmic violation A multiplicative or additive logarithm modifies boundary scaling, as at Fermi surfaces or one-dimensional critical points
Volume law Entropy scales with subsystem volume for a specified state family and subsystem fraction
Topological constant A universal subleading combination after boundary terms are cancelled; not a raw single-fit intercept
Bond dimension χ\chi Tensor-network virtual dimension; a cut crossing one MPS bond has S≤ln⁡χS\le\ln\chi; a periodic interval usually crosses two, giving S≤2ln⁡χS\le2\ln\chi
Finite-entanglement scaling A finite χ\chi introduces an effective correlation length and can round critical behavior

What subsystem entropy does and does not measure

Section titled “What subsystem entropy does and does not measure”

The Bell state (∣00⟩+∣11⟩)/2(\lvert00\rangle+\lvert11\rangle)/\sqrt2 and the separable mixture (∣00⟩⟨00∣+∣11⟩⟨11∣)/2(\lvert00\rangle\langle00\rvert+\lvert11\rangle\langle11\rvert)/2 both have ρA=I/2\rho_A=I/2 and S(A)=ln⁡2S(A)=\ln2. Their mutual information differs: 2ln⁡22\ln2 for the Bell state, ln⁡2\ln2 for the mixture. Their negativities are respectively 1/21/2 and zero. A reduced entropy alone therefore cannot distinguish these two kinds of correlation when the global state may be mixed.

Scaling also depends on the state family and the cut. An area law does not by itself establish an efficient simulation algorithm, and a volume law does not prove chaotic dynamics. See Area Laws, Entanglement Spectrum, and Matrix Product States for those distinctions.