Ph · Physics

How Does the Replica Trick Calculate Entanglement Entropy?

Published By the 9to5Science editorial team

At a glance

  1. 1. Reduce to the region

    Trace out the degrees of freedom outside region A to obtain its normalized reduced density operator.

  2. 2. Compute integer moments

    For positive integers n, evaluate the replica quantity Tr(ρ_A^n), often by joining n copies cyclically.

  3. 3. Continue to n = 1

    Analytically continue the moments in n and differentiate at n = 1 to obtain the von Neumann entropy.

  4. 4. Check the setup

    Keep track of normalization, the logarithm base, and any ultraviolet regulator in a continuum theory.

The replica trick is a way to calculate the entanglement entropy of a subsystem when directly finding all eigenvalues of its reduced density operator is difficult. It replaces that task with calculations of powers of the density operator, followed by an analytic continuation in a parameter called the replica number.

For a subsystem A with reduced density operator ρA\rho_A, the von Neumann entropy is

SA=−Tr⁡(ρAln⁡ρA).S_A=-\operatorname{Tr}(\rho_A\ln\rho_A).

The central identity is that this entropy can be obtained from the moments Tr⁡(ρAn)\operatorname{Tr}(\rho_A^n) by differentiating with respect to nn at 1. The replica trick is especially useful in statistical mechanics and quantum field theory, where those moments can often be expressed as partition functions on a replicated geometry.

From moments to entropy

For real n>0n>0, n≠1n\ne1, whenever the trace exists, define the Rényi entropy

Sn(A)=11−nln⁡Tr⁡(ρAn).S_n(A)=\frac{1}{1-n}\ln\operatorname{Tr}(\rho_A^n).

The von Neumann entropy is the limit as nn approaches 1:

SA=lim⁡n→1Sn(A)=−∂∂nln⁡Tr⁡(ρAn)∣n=1.S_A=\lim_{n\to 1}S_n(A) =-\left.\frac{\partial}{\partial n}\ln\operatorname{Tr}(\rho_A^n)\right|_{n=1}.

To see why, diagonalize the density operator, with eigenvalues pip_i. They are nonnegative and sum to 1, so

Tr⁡(ρAn)=∑ipin.\operatorname{Tr}(\rho_A^n)=\sum_i p_i^n.

Differentiating gives ∑ipinln⁡pi\sum_i p_i^n\ln p_i. At n=1n=1, its negative is exactly −∑ipiln⁡pi-\sum_i p_i\ln p_i, the von Neumann entropy. Terms with pi=0p_i=0 contribute zero by continuity.

The integer values n=2,3,…n=2,3,\ldots are often accessible directly. The challenge is to infer the function near n=1n=1, where the entropy is defined. That continuation must be justified by the problem’s mathematics or physical framework; a finite set of integer values alone does not, in general, uniquely determine it.

What “replicas” mean in a path integral

A density operator can be represented by a path integral with a cut along a time slice. Taking a partial trace over everything outside A joins the fields on the two sides of the cut outside A, while leaving the indices in A open. To calculate Tr⁡(ρAn)\operatorname{Tr}(\rho_A^n), take nn copies of this construction and join the open indices cyclically across the cut in A.

This produces an nn-sheeted space, or replica geometry. If Zn(A)Z_n(A) denotes the path-integral partition function on that geometry and Z1Z_1 is the single-copy partition function, normalization gives, in this setup,

Tr⁡(ρAn)=Zn(A)Z1n.\operatorname{Tr}(\rho_A^n)=\frac{Z_n(A)}{Z_1^n}.

Thus one can calculate a family of partition functions for integer nn, continue the result to values near 1, and differentiate. In some quantum-field-theory settings the replicated space is described as having a conical singularity at the boundary of A. The geometry is a computational representation of the trace, not a claim that the physical system literally consists of multiple copies.

Worked example: one qubit of a Bell pair

Consider the two-qubit state

∣Ψ⟩=∣00⟩+∣11⟩2.|\Psi\rangle=\frac{|00\rangle+|11\rangle}{\sqrt{2}}.

Let A be the first qubit. Tracing out the second qubit gives

ρA=12(1001).\rho_A=\frac{1}{2}\begin{pmatrix}1&0\\0&1\end{pmatrix}.

Its replica moments are

Tr⁡(ρAn)=2(12)n=21−n.\operatorname{Tr}(\rho_A^n)=2\left(\frac12\right)^n=2^{1-n}.

This expression is already defined for real nn, so its continuation is immediate. Then

SA=−∂∂nln⁡(21−n)∣n=1=−∂∂n((1−n)ln⁡2)∣n=1=ln⁡2.S_A=-\left.\frac{\partial}{\partial n}\ln(2^{1-n})\right|_{n=1} =-\left.\frac{\partial}{\partial n}\bigl((1-n)\ln 2\bigr)\right|_{n=1} =\ln 2.

This is the entropy in natural-log units, called nats. If entropy is measured in bits, use logarithms base 2, or divide the result by ln⁡2\ln 2; the answer is 1 bit. The nonzero entropy reflects the entanglement between the two qubits, even though the joint Bell-pair state is pure.

Common mistakes and limitations

  • Using the full-system density operator: Entanglement entropy for A comes from ρA\rho_A, not from the density operator of the entire pure system. A pure full-system state has zero von Neumann entropy.
  • Dropping the normalization: The reduced density operator must have trace 1. In path-integral ratios, the factor Z1nZ_1^n accounts for normalization in the stated setup.
  • Treating integer replicas as the answer: Integer nn moments give Rényi entropies, not automatically the von Neumann entropy. The continuation to n=1n=1 is an essential step.
  • Ignoring units and regulators: Entropy is dimensionless, but its numerical value depends on the logarithm base. In continuum quantum field theories, entanglement entropy often has ultraviolet divergences; a regulator and a clear definition of the region’s boundary are needed before interpreting the result.

The replica method is therefore both an identity and a strategy: the identity relates entropy to a derivative at n=1n=1, while the strategy uses replicated traces or geometries to make that derivative calculable.

References

  1. Entanglement Entropy and Quantum Field Theory — Pasquale Calabrese and John Cardy
  2. Quantum Field Theory and Critical Phenomena — Jean Zinn-Justin

Spotted an error? Send a correction. Read how we write and check articles in our editorial policy.