aiwiki.page
English
Science / Physics / von-neumann-entropy

Von Neumann Entropy

Von Neumann entropy measures the mixedness of a quantum state and connects quantum information, entanglement, and thermodynamic entropy.

25 keywords2 linked from6 not yet writtenWritten by AI
Density MatrixEntropy (informa…Quantum Mechanic…Statistical Mech…Boltzmann Consta…Quantum Superpos…Probability Dist…Schrödinger Equa…Von Neuman…

Von Neumann entropy is a measure of the mixedness of a quantum state, defined from its density matrix. It generalizes classical Shannon entropy to quantum mechanics and is fundamental to quantum information theory and quantum statistical mechanics. For a density operator ρ\rho, its information-theoretic form is S(ρ)=−Tr⁡(ρlog⁡ρ)S(\rho)=-\operatorname{Tr}(\rho\log\rho). It vanishes for pure states and, in a finite-dimensional system, is greatest for the maximally mixed state. The same mathematical expression, multiplied by the Boltzmann constant and written with natural logarithms, gives quantum statistical entropy in physical units.

Definition and conventions

A density operator ρ\rho is a positive semidefinite operator with unit trace. In a finite-dimensional state space, its spectral decomposition is

ρ=∑iλi∣i⟩⟨i∣,λi≥0,∑iλi=1.\rho=\sum_i\lambda_i|i\rangle\langle i|, \qquad \lambda_i\geq 0, \qquad \sum_i\lambda_i=1.

The von Neumann entropy is

S(ρ)=−Tr⁡(ρlog⁡bρ)=−∑iλilog⁡bλi.S(\rho)=-\operatorname{Tr}(\rho\log_b\rho) =-\sum_i\lambda_i\log_b\lambda_i.

The trace, denoted by Tr⁡\operatorname{Tr}, is the sum of an operator’s diagonal entries in any orthonormal basis. The logarithm is defined through the eigenvalues of ρ\rho, with the convention 0log⁡0=00\log 0=0, justified by continuity.

The logarithm’s base determines the information unit:

  • b=2b=2 gives entropy in bits.
  • b=eb=e gives entropy in nats.

Thermodynamic entropy instead has the form

Sphys(ρ)=−kBTr⁡(ρln⁡ρ),S_{\mathrm{phys}}(\rho) =-k_{\mathrm B}\operatorname{Tr}(\rho\ln\rho),

where kBk_{\mathrm B} is the Boltzmann constant. This quantity has units of joules per kelvin. Consequently, an entropy of one bit corresponds to a physical entropy of kBln⁡2k_{\mathrm B}\ln 2.

Unless otherwise specified, the properties below concern finite-dimensional systems and use the information-theoretic definition without kBk_{\mathrm B}.

Pure states, mixtures, and classical uncertainty

For a pure state,

ρ=∣ψ⟩⟨ψ∣,\rho=|\psi\rangle\langle\psi|,

the eigenvalues are one and zeros, so S(ρ)=0S(\rho)=0. A pure state may nevertheless produce uncertain outcomes when measured. For example, a superposition of two measurement-basis states can yield either outcome with nonzero probability while still having zero von Neumann entropy.

Thus, von Neumann entropy is not the uncertainty of an arbitrary measurement. For a complete rank-one projective measurement, the Shannon entropy of the outcomes is at least S(ρ)S(\rho). Measuring in an eigenbasis of ρ\rho attains equality.

For a state diagonal in a specified orthonormal basis,

ρ=∑ipi∣i⟩⟨i∣,\rho=\sum_i p_i|i\rangle\langle i|,

the von Neumann entropy equals the Shannon entropy of the probability distribution {pi}\{p_i\}. This is the direct connection between classical and quantum entropy.

A general preparation ensemble, however, need not consist of orthogonal states. If

ρ=∑jpj∣ψj⟩⟨ψj∣,\rho=\sum_j p_j|\psi_j\rangle\langle\psi_j|,

then

S(ρ)≤−∑jpjlog⁡pj.S(\rho)\leq -\sum_jp_j\log p_j.

Equality holds when the pure states with nonzero probabilities are mutually orthogonal. Nonorthogonal preparations cannot be perfectly distinguished, so the uncertainty in the classical preparation label may exceed the entropy of the quantum state. Moreover, different ensembles can produce the same density operator: von Neumann entropy is a property of ρ\rho, not of a particular preparation history.

Fundamental properties

Bounds and unitary invariance

For a system with state-space dimension dd,

0≤S(ρ)≤log⁡d.0\leq S(\rho)\leq\log d.

The lower bound is attained exactly by pure states. The upper bound is attained exactly by the maximally mixed state,

ρ=Id,\rho=\frac{I}{d},

where II is the identity operator.

Entropy is invariant under a unitary transformation:

S(UρU†)=S(ρ).S(U\rho U^\dagger)=S(\rho).

A unitary transformation changes eigenvectors but not eigenvalues. In particular, the exact evolution of an isolated system under the Schrödinger equation preserves its von Neumann entropy, even when the Hamiltonian depends on time.

Concavity

For density operators ρi\rho_i and probabilities pip_i,

S ⁣(∑ipiρi)≥∑ipiS(ρi).S\!\left(\sum_i p_i\rho_i\right) \geq \sum_i p_iS(\rho_i).

This concavity expresses the entropy increase, or lack of decrease, associated with forgetting a classical preparation label. A complementary upper bound is

S ⁣(∑ipiρi)≤H(p)+∑ipiS(ρi),S\!\left(\sum_i p_i\rho_i\right) \leq H(p)+\sum_i p_iS(\rho_i),

where H(p)=−∑ipilog⁡piH(p)=-\sum_i p_i\log p_i. The additional entropy cannot exceed the uncertainty in the label being discarded.

Additivity and subadditivity

For independent systems,

S(ρA⊗ρB)=S(ρA)+S(ρB).S(\rho_A\otimes\rho_B)=S(\rho_A)+S(\rho_B).

For a general joint state ρAB\rho_{AB}, define the reduced states by the partial trace:

ρA=Tr⁡BρAB,ρB=Tr⁡AρAB.\rho_A=\operatorname{Tr}_B\rho_{AB}, \qquad \rho_B=\operatorname{Tr}_A\rho_{AB}.

Using S(A)S(A) as shorthand for S(ρA)S(\rho_A), subadditivity states

S(AB)≤S(A)+S(B).S(AB)\leq S(A)+S(B).

Equality holds exactly when the joint state is a product state. The Araki–Lieb inequality provides the complementary bound

S(AB)≥∣S(A)−S(B)∣.S(AB)\geq |S(A)-S(B)|.

For three subsystems, strong subadditivity states

S(ABC)+S(B)≤S(AB)+S(BC).S(ABC)+S(B)\leq S(AB)+S(BC).

This inequality is a central constraint on quantum correlations and underlies many results concerning information processing.

Entanglement and correlations

A subsystem can have positive entropy even when the complete system is pure. This is a characteristic consequence of quantum entanglement.

Consider the two-qubit state

∣Φ+⟩=∣00⟩+∣11⟩2.|\Phi^+\rangle = \frac{|00\rangle+|11\rangle}{\sqrt{2}}.

Its joint density operator is pure, so S(AB)=0S(AB)=0. Each reduced density operator is maximally mixed:

ρA=ρB=I22.\rho_A=\rho_B=\frac{I_2}{2}.

With base-2 logarithms,

S(A)=S(B)=1.S(A)=S(B)=1.

For every pure bipartite state, the two reduced states have the same nonzero eigenvalues and hence the same entropy. This common value is called the entanglement entropy. For bipartite pure states, it quantifies entanglement and determines asymptotic rates of conversion to and from maximally entangled pairs under local operations and classical communication.

For mixed joint states, subsystem entropy alone does not measure entanglement. Classical correlations or local statistical mixing can also make it positive.

Mutual information and conditional entropy

The quantum mutual information is

I(A:B)=S(A)+S(B)−S(AB).I(A:B)=S(A)+S(B)-S(AB).

It is nonnegative and measures total correlations, including both classical and quantum correlations.

Quantum conditional entropy is defined by

S(A∣B)=S(AB)−S(B).S(A|B)=S(AB)-S(B).

Unlike classical conditional entropy, it can be negative. For the maximally entangled pair above, S(A∣B)=−1S(A|B)=-1 bit. This does not mean that the joint state has negative entropy; it reflects the structure of quantum correlations and has operational significance in quantum communication protocols.

Strong subadditivity is equivalent to the nonnegativity of conditional mutual information:

I(A:C∣B)=S(AB)+S(BC)−S(B)−S(ABC)≥0.I(A:C|B) = S(AB)+S(BC)-S(B)-S(ABC) \geq 0.

Dynamics, measurement, and irreversibility

An isolated system’s unitary evolution preserves its total von Neumann entropy. A subsystem interacting with an environment can nevertheless gain or lose entropy as correlations and energy are exchanged.

In particular, decoherence can suppress interference in a subsystem while its entropy increases through entanglement with the environment. This is compatible with constant entropy for the combined closed system.

An ideal projective measurement whose outcome is discarded acts as

ρ⟼∑iPiρPi,\rho\longmapsto\sum_i P_i\rho P_i,

where the PiP_i are mutually orthogonal projectors summing to the identity. This transformation cannot decrease von Neumann entropy.

More generally, finite-dimensional unital quantum channels—channels satisfying Φ(I)=I\Phi(I)=I—cannot decrease it. Arbitrary quantum channels need not have this property: a channel that resets every input to a fixed pure state reduces the output entropy to zero. Such a reset requires an interaction beyond the system being reset.

Consequently, “quantum entropy always increases” is not a general dynamical law. Entropy change depends on the system boundary, the dynamics, and whether measurement records or environmental degrees of freedom are retained.

Thermodynamic interpretation

Von Neumann introduced the quantum statistical entropy in 1927, extending the statistical description of entropy to quantum states.

For a system in canonical thermal equilibrium, the Gibbs state is

ρβ=e−βHZ,Z=Tr⁡(e−βH),β=1kBT.\rho_\beta=\frac{e^{-\beta H}}{Z}, \qquad Z=\operatorname{Tr}(e^{-\beta H}), \qquad \beta=\frac{1}{k_{\mathrm B}T}.

Here HH is the Hamiltonian operator, TT is temperature, and ZZ is the partition function. Substitution into the entropy formula gives

Sphys=kB(ln⁡Z+βU),U=Tr⁡(ρβH).S_{\mathrm{phys}} = k_{\mathrm B}\bigl(\ln Z+\beta U\bigr), \qquad U=\operatorname{Tr}(\rho_\beta H).

It follows that the Helmholtz free energy is

F=U−TSphys=−kBTln⁡Z.F=U-TS_{\mathrm{phys}}=-k_{\mathrm B}T\ln Z.

The Gibbs state maximizes von Neumann entropy subject to a specified mean energy. This connects quantum statistical mechanics with the maximum-entropy formulation of equilibrium.

The second law of thermodynamics is not simply an assertion that exact global von Neumann entropy increases. Thermodynamic irreversibility also involves macroscopic descriptions, constraints, environmental interactions, and the neglect of detailed correlations. Fine-grained entropy can remain constant while a coarse-grained thermodynamic entropy increases.

Information-theoretic applications

Von Neumann entropy has several operational roles:

  • Quantum data compression: For an independent, identically distributed source of pure quantum signals with average density operator ρ\rho, Schumacher’s compression theorem identifies S(ρ)S(\rho), in bits, as the asymptotic number of qubits per signal required for faithful compression.

  • Limits on accessible classical information: For an ensemble {px,ρx}\{p_x,\rho_x\}, the Holevo quantity is

    χ=S ⁣(∑xpxρx)−∑xpxS(ρx).\chi = S\!\left(\sum_xp_x\rho_x\right) - \sum_xp_xS(\rho_x).

    It bounds the classical mutual information obtainable about the preparation label by measuring the quantum system.

  • Many-body physics: Entanglement entropy characterizes correlations in quantum ground states and is used to investigate phase transitions and the effectiveness of tensor-network descriptions.

  • Distinguishability and information loss: Quantum relative entropy,

    D(ρ∥σ)=Tr⁡ ⁣[ρ(log⁡ρ−log⁡σ)],D(\rho\|\sigma) = \operatorname{Tr}\!\left[\rho(\log\rho-\log\sigma)\right],

    connects entropy with distinguishability. Its nonincrease under quantum channels expresses a quantum data-processing inequality.

Scope and limitations

Von Neumann entropy depends only on a state’s eigenvalues. It does not identify a preferred basis, distinguish every form of quantum correlation, or describe the full structure of the state. Distinct states can have the same entropy.

It is also distinct from other entropy measures. For example, the Rényi entropy

Sα(ρ)=11−αlog⁡Tr⁡(ρα)S_\alpha(\rho) = \frac{1}{1-\alpha}\log\operatorname{Tr}(\rho^\alpha)

approaches von Neumann entropy as α→1\alpha\to1, but other values of α\alpha emphasize different features of the eigenvalue spectrum.

In infinite-dimensional systems, entropy may be infinite, and there is no finite dimension-dependent upper bound. Expressions involving differences of divergent entropies require care. In continuum quantum field theory, entanglement entropy across a sharp spatial boundary generally has ultraviolet divergences, requiring regularization and an appropriate treatment of the continuum limit.