| partial trace | |
|---|---|
| Name | Partial trace |
| Caption | Schematic of tracing out a subsystem |
| Field | Quantum mechanics |
| Introduced | 20th century |
| Used in | Quantum information theory, Quantum statistical mechanics |
partial trace
The partial trace is a linear operation on operators in a composite Hilbert space that eliminates (or "traces out") degrees of freedom associated with a subsystem to produce an operator on the remaining subsystem. It is central in the description of mixed states, reduced density matrixs, and the dynamics of open quantum systems because it formalizes how one obtains observable predictions for a subsystem when the rest of the system is ignored or inaccessible. The partial trace underpins practically all treatments of decoherence and many measures of quantum entanglement.
Given a bipartite system with Hilbert space H = H_A ⊗ H_B, the partial trace over subsystem B is the unique linear map Tr_B: L(H_A ⊗ H_B) → L(H_A) satisfying Tr[ (X_A ⊗ I_B) ρ ] = Tr[ X_A Tr_B(ρ) ] for all operators X_A ∈ L(H_A) and density operators ρ ∈ L(H). For an orthonormal basis {|i⟩} of H_B, the partial trace can be written explicitly as Tr_B(ρ) = ∑_i (I_A ⊗ ⟨i|) ρ (I_A ⊗ |i⟩). Similarly Tr_A traces out subsystem A. For finite-dimensional spaces this reduces to sums over matrix blocks; in infinite dimensions one uses trace-class operator theory as in von Neumann algebra contexts. The operation is linear, completely positive, and trace-preserving when combined with tensoring by identity on the traced subsystem.
Physically, taking a partial trace corresponds to discarding or being ignorant of the degrees of freedom of an environment or measurement apparatus, leaving the effective state of the subsystem of interest. In the theory of open quantum systems this formalizes how a system coupled to an environment yields non-unitary dynamics for the system alone. The reduced density matrix ρ_A = Tr_B(ρ_{AB}) encodes expectation values of all observables acting on subsystem A, and it is the operational quantum state used in predicting measurement outcomes when the environment is unobserved. Models of dissipation and noise in platforms such as superconducting qubits, ion traps, and quantum dots use partial traces to derive master equations (e.g., the Lindblad equation) by tracing out bath degrees of freedom modeled by reservoirs like the Caldeira–Leggett model.
Important algebraic properties include - Linearity: Tr_B(a ρ + b σ) = a Tr_B(ρ) + b Tr_B(σ). - Positivity preservation: if ρ ≥ 0 then Tr_B(ρ) ≥ 0. - Compatibility with full trace: Tr_A[Tr_B(ρ)] = Tr(ρ). - Invariance under local unitaries on the traced subsystem: Tr_B[(I_A ⊗ U_B) ρ (I_A ⊗ U_B†)] = Tr_B(ρ). - Cyclicity holds only for the full trace, not for partial trace, but one has Tr_A[(X_A ⊗ I_B)ρ] = X_A Tr_B(ρ) in expectation values. Partial trace is a special case of a quantum operation (completely positive trace-preserving map) when combined with preparation of a maximally mixed input or discarding subsystems. In operator-sum representations it appears in Stinespring dilations where environment degrees of freedom are traced out after unitary evolution.
A canonical example is a two-qubit pure entangled state, e.g., the Bell state |Φ+⟩ = (|00⟩+|11⟩)/√2. Tracing out one qubit yields the maximally mixed state ρ_A = I/2, demonstrating how entanglement produces mixed reduced states. In quantum thermodynamics and quantum statistical mechanics, partial trace is used to derive canonical ensembles by tracing out bath modes. In models of decoherence, system-environment entanglement followed by partial trace produces suppression of off-diagonal elements in preferred bases (pointer basis), explaining emergence of classical probabilities; this is central to work by researchers such as Wojciech Zurek on environment-induced decoherence. Partial trace is also used to compute operational quantities: expectation values, conditional states after generalized measurement (POVM), and channel output states in quantum channel theory.
Practical computation uses basis expansions, matrix reshaping, and tensor contractions. For finite-dimensional systems represented as Kronecker-product matrices, one can implement Tr_B by reshaping the matrix into a four-index tensor ρ_{iα,jβ} and summing over the environment indices α=β. Numerical libraries in Python (programming language) (e.g., QuTiP), MATLAB, and packages from research groups at institutions like IBM Quantum and Google Quantum AI provide optimized routines. For large many-body systems, tensor network methods such as matrix product states (MPS) and density matrix renormalization group (DMRG) compute reduced density matrices efficiently by exploiting low entanglement. Monte Carlo approaches and stochastic unravelings can approximate partial traces for high-dimensional baths; analytic approximations include weak-coupling Born and Markov approximations leading to master equations.
Partial trace interfaces with the formalism of quantum channels: any CPTP map can be represented by a unitary on a larger Hilbert space followed by a partial trace (Stinespring dilation). Entanglement measures such as entanglement entropy (von Neumann entropy of the reduced density matrix S(ρ_A) = −Tr ρ_A log ρ_A), concurrence, and negativity are computed from reduced states produced by partial trace. Operational tasks in quantum information theory—quantum teleportation, entanglement distillation, and state merging—rely on properties of reduced density matrices and hence on the partial trace. The mathematical and computational role of partial trace therefore links foundational questions about nonlocal correlations to practical characterization of quantum devices developed by groups at MIT, University of Cambridge, and industrial labs.