---
title: Kraus Representation Theorem in Quantum Channels
url: https://www.emergentmind.com/topics/kraus-representation-theorem
type: topic
---

# Kraus Representation Theorem in Quantum Channels

The Kraus Representation Theorem provides the canonical structure of completely positive, trace-preserving (CPTP) linear maps—known as quantum channels—on operator algebras. These evolution maps govern the dynamics of open quantum systems, parameterize quantum operations, and underpin both foundational theory and statistical learning frameworks in contexts from quantum information to classical-quantum hybrid models. The theorem asserts that any CPTP map can be represented by a countable family of bounded operators (Kraus operators), subject to completeness relations. Its generalizations accommodate infinite dimensions, hybrid probabilistic structures, and serve as the basis for efficient algorithmic parameterizations.

## 1. Structural Statement and Derivation

Consider Hilbert spaces $\mathcal{H}_S$ (system) and $\mathcal{H}_E$ (environment), with $\dim(\mathcal{H}_S)=d_S$, $\dim(\mathcal{H}_E)=d_E$. Suppose the joint initial state is uncorrelated: $\rho_{SE}(0)=\rho_S(0)\otimes|E_0\rangle\langle E_0|$, and evolves unitarily under $U(t)$:
\[
\rho_{SE}(t) = U(t)\left[\rho_S(0)\otimes|E_0\rangle\langle E_0|\right]U^\dagger(t)
\]
The reduced system state at time $t$ is:
\[
\rho_S(t) = \operatorname{Tr}_E\left[U(t)\left(\rho_S(0)\otimes|E_0\rangle\langle E_0|\right)U^\dagger(t)\right]
\]
The Kraus Representation Theorem asserts that $\rho_S(t)$ admits the operator-sum form
\[
\rho_S(t) = \sum_k K_k(t)\, \rho_S(0)\, K_k^\dagger(t)
\]
where $K_k(t)$: $\mathcal{H}_S\to\mathcal{H}_S$, constructed as $K_k(t)=\langle E_k|U(t)|E_0\rangle$ for any orthonormal basis $\{|E_k\rangle\}$ of $\mathcal{H}_E$. The completeness relation $\sum_k K_k^\dagger K_k = I_S$ guarantees trace preservation. This operator-sum form generalizes to any finite-dimensional CPTP map, and the minimal number of Kraus operators ("Kraus rank") is bounded above by $d_S^2$ [1510.09081].

## 2. Operator-Sum Structure and Complete Positivity

Let $A$ and $B$ be operator systems. A linear map $\Phi:A\rightarrow B$ is called completely positive (CP) if its ampliations $\Phi_n: M_n(A)\rightarrow M_n(B)$ are positive for all $n\ge 1$. The Kraus Representation Theorem for CP maps states:
\[
\Phi(\rho) = \sum_j K_j\rho K_j^\dagger
\]
for (possibly infinite) $K_j\in B(\mathcal{H})$. For trace-preservation, $\sum_j K_j^\dagger K_j = I_{\mathcal{H}}$ holds, with strong operator convergence in infinite dimensions [2301.05488, 2512.00256]. The theorem is both necessary and sufficient: any finite-sum of this form is CP, and every CP map admits a Kraus decomposition.

A table summarizing the core correspondence:

| Property                                      | Mathematical Formulation                    | Kraus Representation Criterion           |
|-----------------------------------------------|---------------------------------------------|-----------------------------------------|
| Complete positivity (CP)                      | $\Phi\otimes\mathrm{id}_n$ is positive      | Exists $K_j$ with $\Phi(\cdot)=\sum_j K_j(\cdot)K_j^\dagger$ |
| Trace preservation (TP)                       | $\operatorname{tr}\Phi(\rho)=\operatorname{tr}(\rho)$ | $\sum_j K_j^\dagger K_j=I$             |
| Minimal number of Kraus operators (rank)      | $r=\operatorname{rank}(C_\Phi)$, $C_\Phi$ Choi | $r \leq d_S^2$                          |

This characterization is foundational in quantum information theory and is exploited in statistical learning to parametrize the entire CP family by a finite set of matrix-valued parameters [2007.00935].

## 3. Construction Methods and Unitary Dilations

Multiple approaches yield the Kraus representation:

- **System–environment Unitary Picture:** Kraus operators are obtained by tracing out the environment after a joint unitary evolution, as above [1510.09081].
- **Stinespring Dilation:** CP maps are factorized as
  \[
  \Phi(\rho) = \operatorname{Tr}_E\big( U(\rho \otimes |\psi\rangle\langle\psi|)U^\dagger \big)
  \]
  where $U$ is a unitary on system + environment and $|\psi\rangle$ is a fixed pure state. Via orthonormal decomposition, projecting onto an environment basis recovers the operator-sum form [2301.05488].
- **Kernel-Theoretic Construction:** By constructing reproducing kernel Hilbert spaces (RKHS) from scalar positive-definite kernels $K((S,a), (T,b)) = \langle a, \Phi(S^*T)b\rangle$, the existence of a minimal Stinespring dilation $\pi(S)$ and operator $V$ yields the canonical Kraus decomposition through a tensor structure with the ancillary space [2512.00256].
- **Sz.-Nagy Dilation:** For infinite-dimensional systems, Sz.-Nagy’s dilation theorem extends partial isometries arising from Kraus families to explicit unitaries on an enlarged Hilbert space, yielding a fully explicit unitary realization of any CP map [2301.05488].

Comparative analysis between constructions (standard finite-dimensional, Hellwig-Kraus, Sz.-Nagy) demonstrates differences in auxiliary space dimension and technical details, but all instantiate the same structural theorem.

## 4. Minimality, Choi Matrix, and Variants

The Choi isomorphism gives a matrix bijectively associated to any linear map $\Phi:\mathcal{M}_p \to \mathcal{M}_q$ via $C_\Phi = \sum_{ij} E_{ij}\otimes\Phi(E_{ij})$. Complete positivity is equivalent to $C_\Phi \succeq 0$, and the minimal Kraus rank is $\operatorname{rank}(C_\Phi)$ [2007.00935, 2512.00256]. The family of Kraus operators is not unique: any set $\{K_j'\}$ related by $K_j' = \sum_n V_{jn} K_n$ for a unitary $V$ yields the same channel, reflecting the redundancy in the operator-sum representation.

In the kernel-theoretic formulation, the dimension of the ancillary RKHS sets the minimal Kraus index set, and any other minimal family is obtained by isometric rotation in the ancilla space [2512.00256].

## 5. Illustrative Example: Amplitude-Damping Channel

A standard application considers a two-level atom (qubit) undergoing amplitude damping via coupling to a vacuum field:
\[
U_{\text{CAA}} |S_0,E_0\rangle = |S_0,E_0\rangle,\qquad
U_{\text{CAA}} |S_1,E_0\rangle = \sqrt{1-p}|S_1,E_0\rangle + \sqrt{p}|S_0,E_1\rangle
\]
with $p=p(t)\in[0,1]$ depending on time. The Kraus operators are
\[
K_0 = \begin{pmatrix}1 & 0 \\ 0 & \sqrt{1-p}\end{pmatrix},\qquad
K_1 = \begin{pmatrix}0 & \sqrt{p} \\ 0 & 0\end{pmatrix}
\]
and the atomic state evolution becomes
\[
\rho_S(t) = K_0 \rho_S(0) K_0^\dagger + K_1 \rho_S(0) K_1^\dagger
\]
This yields an explicit illustration of decoherence: off-diagonal coherences decay as $\sqrt{1-p}$, and a basis-dependent coherence measure $C[\rho_S(t)] = \sqrt{1-p}C[\rho_S(0)]$ [1510.09081].

## 6. Generalizations: Hybrid Classical–Quantum Systems

For hybrid systems—jointly governed by classical and quantum degrees of freedom—Kraus-type theorems require lifting the notion of CP and state space. A hybrid state $w(A,E)$ (joint probability measure on classical events $A$ and quantum effects $E$) with a reference measure $\mu$ admits a density-operator-valued function $\omega(x)$. A hybrid operation $\mathcal{O}_h$ subject to continuity and a generalized CP condition (formulated with respect to classical partitions and quantum ancillas) is characterized by a Kraus-type expansion using hybrid kernels [2408.10882]:
\[
[\mathcal{O}_h(\check{w})](A,E) = \sum_{n=1}^N \sum_\alpha \operatorname{tr}\left[ K_\alpha(A, A_n) w_n K_\alpha(A, A_n)^\dagger E \right]
\]
In the purely quantum limit ($X$ a point), this reduces to the standard Kraus representation. Specialization to the discrete-classical case yields a finite sum over $x'$, $[\mathcal{O}_h(\omega)](x,E) = \sum_{x'}\sum_\alpha K_\alpha(x,x')\,\omega(x')\,K_\alpha(x,x')^\dagger E$ [2408.10882].

Continuity in trace-norm and separability conditions are required for existence and extension of this form to general (possibly infinite-dimensional) settings. Such hybrid generalized Kraus theorems extend the applicability of operator-sum structure to evolutions and operations in hybrid probabilistic contexts.

## 7. Applications, Learning, and Statistical Implications

The operator-sum (Kraus) parameterization underlies quantum dynamical maps, error modeling, decoherence, and quantum feedback protocols. In data-driven settings, parameterizing CP maps by low Kraus rank enables efficient learning and regularization of regression-like models mapping $X\mapsto Y$ between matrix domains. The Kraus decomposition is exploited algorithmically: empirical minimization over $r$ operator parameters balances accuracy and complexity, with risk bounds and pseudo-dimension scaling directly with $p, q$, and $r$ [2007.00935].

In kernel approaches, the entire Stinespring–and hence Kraus–dilation can be realized through manipulations in reproducing kernel Hilbert spaces, bypassing explicit representations and exploiting positivity at the scalar level [2512.00256].

The Kraus theorem is thus both structurally ubiquitous in the analysis of open quantum systems and central to the practical design and identification of quantum operations in a variety of theoretical and applied domains.

Source: https://www.emergentmind.com/topics/kraus-representation-theorem