# Stinespring dilation theorem

The **Stinespring dilation theorem** is a structure theorem in operator algebra theory stating that every completely positive map from a unital C\(^*\)-algebra into the bounded operators on a Hilbert space can be written as a compression of a \(*\)-homomorphism: \( \phi(a) = V^{*}\pi(a)V \) for a representation \( \pi \) on a larger [Hilbert space](https://www.edgepedia.org/hilbert-space) and a bounded operator \( V \). It was proved by W. Forrest Stinespring in the 1955 paper *Positive Functions on C\(^*\)-Algebras*, published in the Proceedings of the American Mathematical Society (volume 6, issue 2, pages 211–216)<sup>[1](https://cir.nii.ac.jp/crid/1363670319426467968)</sup>. That paper also introduced the term "completely positive"<sup>[2](https://www.cmat.edu.uy/~lessa/tesis/Positive%20Definite%20Matrices.pdf)</sup>.

| Key fact | Detail |
|---|---|
| Statement | For a unital C\(^*\)-algebra \( A \), a Hilbert space \( H \), and a completely positive map \( \phi : A \to B(H) \), there exist a Hilbert space \( K \), a \(*\)-homomorphism \( \pi : A \to B(K) \), and a bounded operator \( V : H \to K \) with \( \phi(a) = V^{*}\pi(a)V \) for all \( a \in A \)<sup>[3](https://repository.dl.itc.u-tokyo.ac.jp/record/2013159/files/A41773.pdf)</sup> |
| Norm relation | \( \|V\|^{2} = \|\phi(1)\| \); when \( \phi \) is unital, \( V \) is an isometry<sup>[4](https://web.math.ku.dk/rordam/students/Kirchberg-thm-MSC.pdf)</sup><sup> • </sup><sup>[5](https://www.math.ucsd.edu/~helton/247A10/completePosIntro.pdf)</sup> |
| Normal case | If \( A \) is a von Neumann algebra and \( \phi \) is normal, \( \pi \) can be taken normal<sup>[3](https://repository.dl.itc.u-tokyo.ac.jp/record/2013159/files/A41773.pdf)</sup> |
| Uniqueness | A minimal dilation (smallest possible \( K \)) is unique up to unitary equivalence<sup>[2](https://www.cmat.edu.uy/~lessa/tesis/Positive%20Definite%20Matrices.pdf)</sup> |
| Origin | W. Forrest Stinespring, *Positive Functions on C\(^*\)-Algebras*, Proc. Amer. Math. Soc. 6(2), 211–216 (April 1955)<sup>[1](https://cir.nii.ac.jp/crid/1363670319426467968)</sup> |
| Quantum form | Every quantum channel (completely positive trace-preserving map) \( E : \mathrm{End}(H_A) \to \mathrm{End}(H_B) \) has an isometry \( U : H_A \to H_B \otimes H_R \) with \( E(S) = \mathrm{Tr}_R(U S U^{*}) \), and \( \dim H_R \le d_A d_B \)<sup>[6](https://edu.itp.phys.ethz.ch/hs15/QIT/renes_lecture_notes14.pdf)</sup> |
| Predecessor | Generalizes Naimark's dilation theorem, replacing the commutative algebra \( C(X) \) by a unital C\(^*\)-algebra<sup>[2](https://www.cmat.edu.uy/~lessa/tesis/Positive%20Definite%20Matrices.pdf)</sup> |

## Statement and meaning

A linear map \( \phi : A \to B \) between C\(^*\)-algebras is completely positive when the amplified maps \( \phi_n : M_n(A) \to M_n(B) \), defined entrywise by \( \phi_n([a_{i,j}]) = [\phi(a_{i,j})] \), are positive for every \( n \in \mathbb{N} \); it is unital completely positive when additionally \( \phi(1_A) = 1_B \)<sup>[3](https://repository.dl.itc.u-tokyo.ac.jp/record/2013159/files/A41773.pdf)</sup>. The theorem says that such a map is exactly a compression of a \(*\)-homomorphism: there is a Hilbert space \( K \), a representation \( \pi : A \to B(K) \), and a bounded operator \( V : H \to K \) such that \( \phi(a) = V^{*}\pi(a)V \) for all \( a \in A \)<sup>[3](https://repository.dl.itc.u-tokyo.ac.jp/record/2013159/files/A41773.pdf)</sup>. The converse direction is immediate: if \( \pi \) is a unital \(*\)-homomorphism and \( V \in B(H,K) \), then \( a \mapsto V^{*}\pi(a)V \) is completely positive, so the theorem characterizes the completely positive maps rather than merely producing a representation for them<sup>[5](https://www.math.ucsd.edu/~helton/247A10/completePosIntro.pdf)</sup>.

The construction carries a quantitative handle on the dilation space: \( \|V\|^{2} = \|\phi(1)\| \)<sup>[4](https://web.math.ku.dk/rordam/students/Kirchberg-thm-MSC.pdf)</sup>. When \( \phi \) is unital this forces \( V^{*}V = I \), so \( V \) is an isometry, and \( K \) can be identified with \( H \oplus V(H)^{\perp} \); the map then takes the form \( \phi(a) = P_H \pi(a)|_H \), a compression of a unital \(*\)-homomorphism to the original space<sup>[5](https://www.math.ucsd.edu/~helton/247A10/completePosIntro.pdf)</sup>. Size of the dilation follows the data: if \( A \) and \( H \) are separable, \( K \) may be taken separable, and if \( A \) and \( H \) are finite dimensional, \( K \) may be taken finite dimensional<sup>[5](https://www.math.ucsd.edu/~helton/247A10/completePosIntro.pdf)</sup>.

## Minimality and uniqueness

The triple \( (\pi, V, K) \) is called a Stinespring representation of \( \phi \)<sup>[5](https://www.math.ucsd.edu/~helton/247A10/completePosIntro.pdf)</sup>. It is minimal when the closed linear span of \( \pi(A)VH \) is all of \( K \); any representation can be reduced to a minimal one by restricting to this subspace, on which \( \pi \) remains a \(*\)-homomorphism<sup>[5](https://www.math.ucsd.edu/~helton/247A10/completePosIntro.pdf)</sup>. A minimal dilation is unique up to unitary equivalence<sup>[2](https://www.cmat.edu.uy/~lessa/tesis/Positive%20Definite%20Matrices.pdf)</sup>. More generally, for two Stinespring representations \( (K,\pi,V) \) and \( (K_1,\pi_1,V_1) \) of the same map there is a partial isometry \( W : K \to K_1 \) satisfying the intertwining relations \( WV = V_1 \), \( W^{*}V_1 = V \), and \( W\pi(a) = \pi_1(a)W \); when both representations are minimal, \( W \) is unitary<sup>[7](https://d-nb.info/987826840/34)</sup>.

Nonminimal representations are genuinely nonunique, and in quantum information this nonuniqueness is physically meaningful: many different environment models and unitaries produce identical reduced dynamics for the principal system<sup>[8](https://doi.org/10.1103/PRXQuantum.6.030202)</sup>.

## Proof idea and mathematical context

The proof is a direct generalization of the GNS construction for states<sup>[9](https://users.math.msu.edu/users/banelson/conferences/GOALS/notes/Cstar_notes.pdf)</sup>. Given a completely positive \( \Phi : A \to B(H) \), one defines a [sesquilinear form](https://www.edgepedia.org/sesquilinear-form) on algebraic tensors \( x = \sum_i a_i \otimes u_i \) by
\[
\langle x, y \rangle = \sum_{i,j} \langle u_i, \Phi(a_i^{*} b_j) v_j \rangle,
\]
which is nonnegative by complete positivity; quotienting by the null vectors and completing yields the Hilbert space \( K \), and \( \pi \) is defined by left multiplication \( \pi_0(a)(\sum_i a_i \otimes u_i) = \sum_i (a a_i) \otimes u_i \)<sup>[10](https://www.mat.uc.cl/~rolando.rebolledo/pub70-2002/pdf94-02/qp94-02/re00b.pdf)</sup>.

The theorem generalizes Naimark's dilation theorem, in which the commutative algebra \( C(X) \) is replaced by an arbitrary unital C\(^*\)-algebra<sup>[2](https://www.cmat.edu.uy/~lessa/tesis/Positive%20Definite%20Matrices.pdf)</sup>. Consistently, complete positivity coincides with plain positivity when the domain algebra is commutative (a result of Stinespring) or when the codomain is commutative (Arveson)<sup>[10](https://www.mat.uc.cl/~rolando.rebolledo/pub70-2002/pdf94-02/qp94-02/re00b.pdf)</sup>. The theory of positive and completely positive maps was vastly expanded in W. B. Arveson's papers *Subalgebras of C\(^*\)-algebras, I, II* (Acta Mathematica 123 (1969) 141–224 and 128 (1972) 271–308)<sup>[2](https://www.cmat.edu.uy/~lessa/tesis/Positive%20Definite%20Matrices.pdf)</sup>. Kasparov's extension of the construction to C\(^*\)-modules is central in bivariant K-homology theory<sup>[11](https://www.sciencedirect.com/topics/mathematics/completely-positive-maps)</sup>, and analogues hold for Hilbert modules over pro-C\(^*\)-algebras<sup>[12](https://arxiv.org/abs/1611.04759)</sup>.

In the special case \( A = B(H) \) with \( H \) separable, the dilation yields operators \( b_n \in B(H) \) with \( \phi(a) = \sum_{n=1}^{\infty} b_n^{*} a b_n \), the Kraus form<sup>[11](https://www.sciencedirect.com/topics/mathematics/completely-positive-maps)</sup>.

## Role in quantum information

For finite-dimensional quantum systems, a quantum channel is a completely positive trace-preserving map, and Stinespring's theorem is its basic structure theorem: every channel arises from a unitary evolution on a larger system<sup>[13](https://arxiv.org/abs/quant-ph/0605009)</sup>. Concretely, for a channel \( E_{A \to B} : \mathrm{End}(H_A) \to \mathrm{End}(H_B) \) there is an isometry \( U_{A \to BR} \in \mathrm{Hom}(H_A, H_B \otimes H_R) \) such that
\[
E_{A \to B}(S_A) = \mathrm{Tr}_R\bigl( U_{A \to BR} \, S_A \, U_{A \to BR}^{*} \bigr),
\]
with \( \dim H_R \) at most \( d_A d_B \)<sup>[6](https://edu.itp.phys.ethz.ch/hs15/QIT/renes_lecture_notes14.pdf)</sup>. The three operations involved are tensoring the input with an ancilla in a specified state, applying a unitary to the combined system, and reducing by a partial trace<sup>[13](https://arxiv.org/abs/quant-ph/0605009)</sup>. The ancilla has a natural interpretation as the environment: the channel's output arises from a unitary interaction with the environment followed by tracing over the environment's degrees of freedom, and interchanging the roles of output and environment gives the complementary channel<sup>[13](https://arxiv.org/abs/quant-ph/0605009)</sup>.

The Stinespring, Kraus, and Choi representations of a channel are equivalent: from a Kraus representation \( \Phi(\rho) = \sum_k A_k \rho A_k^{\dagger} \) one builds an isometry \( V = \sum_i |i\rangle \otimes E_i \) on an ancilla whose dimension equals the number of [Kraus operators](https://www.edgepedia.org/kraus-operators), and conversely a Stinespring isometry yields Kraus operators by choosing a basis of the environment<sup>[14](https://qubit.guide/9.5-stinesprings-dilation-and-krauss-ambiguity.html)</sup>. A channel of Kraus rank \( r \) admits a dilation with an auxiliary system of dimension \( r \)<sup>[15](https://www.physik.fu-berlin.de/en/einrichtungen/ag/ag-eisert/teaching/ws20-21/solution05.pdf)</sup>, and the minimal environment dimension is the Kraus rank<sup>[16](https://citeseerx.ist.psu.edu/document?doi=afa58291b0b8bd47504acb1ab8f553f0b37685cf&repid=rep1&type=pdf)</sup>. A unitary channel \( \Phi(\rho) = U\rho U^{\dagger} \) has a Stinespring isometry with a trivial, one-dimensional environment<sup>[17](https://homel.vsb.cz/~bil0127/subjects/koch/exercise-3/)</sup>.

The dilation also supports quantitative results. Kretschmann, Schlingemann, and Werner proved a continuity theorem: two quantum channels close in cb-norm admit dilating unitaries close in operator norm, with the two-sided estimate
\[
\inf_{V_1,V_2} \|V_1 - V_2\|_{\infty}^{2} \le \|T_1 - T_2\|_{cb} \le 2 \inf_{V_1,V_2} \|V_1 - V_2\|_{\infty},
\]
and the bounds are independent of the dimensions of the underlying Hilbert spaces; the proof generalizes Uhlmann's theorem from states to channels<sup>[13](https://arxiv.org/abs/quant-ph/0605009)</sup>. Since complementary channels share a common Stinespring representation, this continuity estimate relates the distance between two channels to the distance between their complementaries, yielding the information-disturbance tradeoff: information can be recovered from a channel's output by a decoding operation if and only if the channel releases almost no information to the environment<sup>[13](https://arxiv.org/abs/quant-ph/0605009)</sup>.

Practically, the dilation underlies how noisy quantum operations are modeled and simulated: on a quantum computer, an arbitrary channel can be implemented by preparing clean ancilla qubits, running the circuit for the dilating unitary, and discarding the ancillas<sup>[18](https://www.nature.com/articles/s41534-022-00517-3)</sup>, and the Qiskit class `qiskit.quantum_info.Stinespring` stores a channel as a rectangular matrix \( A \) with \( E(\rho) = \mathrm{Tr}_2(A \rho A^{\dagger}) \)<sup>[19](https://docs.quantum.ibm.com/api/qiskit/qiskit.quantum_info.Stinespring)</sup>.

## References

1. [W. Forrest Stinespring (1955). Positive Functions on C*-Algebras. Proceedings of the American Mathematical Society 6(2), 211–216.](https://cir.nii.ac.jp/crid/1363670319426467968)
2. [Positive Definite Matrices (thesis chapter on completely positive maps and the Stinespring theorem)](https://www.cmat.edu.uy/~lessa/tesis/Positive%20Definite%20Matrices.pdf)
3. [3.3 Completely positive maps (University of Tokyo lecture notes)](https://repository.dl.itc.u-tokyo.ac.jp/record/2013159/files/A41773.pdf)
4. [Chapter 1. Preliminaries (Kirchberg theorem thesis, University of Copenhagen)](https://web.math.ku.dk/rordam/students/Kirchberg-thm-MSC.pdf)
5. [Complete positivity in operator theory, Chapter 5: Stinespring representation (UCSD notes)](https://www.math.ucsd.edu/~helton/247A10/completePosIntro.pdf)
6. [Quantum Information Theory, lecture notes by Renato Renner (ETH Zürich), Section 4.3.5](https://edu.itp.phys.ethz.ch/hs15/QIT/renes_lecture_notes14.pdf)
7. [Stinespring's representation theorem and uniqueness of the minimal dilation (dissertation, Deutsche Nationalbibliothek)](https://d-nb.info/987826840/34)
8. [Practical Introduction to Benchmarking and Characterization of Quantum Computers, PRX Quantum 6, 030202 (2025)](https://doi.org/10.1103/PRXQuantum.6.030202)
9. [Notes on C*-algebras (Michigan State University), Section 10](https://users.math.msu.edu/users/banelson/conferences/GOALS/notes/Cstar_notes.pdf)
10. [Complete positivity and open quantum systems (R. Rebolledo)](https://www.mat.uc.cl/~rolando.rebolledo/pub70-2002/pdf94-02/qp94-02/re00b.pdf)
11. [Completely positive maps – an overview, ScienceDirect Topics](https://www.sciencedirect.com/topics/mathematics/completely-positive-maps)
12. [Completely positive maps on Hilbert modules over pro-C*-algebras, arXiv:1611.04759](https://arxiv.org/abs/1611.04759)
13. [D. Kretschmann, D. Schlingemann, R. F. Werner (2006). The Information-Disturbance Tradeoff and the Continuity of Stinespring's Representation. arXiv:quant-ph/0605009](https://arxiv.org/abs/quant-ph/0605009)
14. [Stinespring's dilation and Kraus's ambiguity, Introduction to Quantum Information Science (qubit.guide)](https://qubit.guide/9.5-stinesprings-dilation-and-krauss-ambiguity.html)
15. [Exercise solutions, Quantum Information Theory (Freie Universität Berlin, WS 2020/21)](https://www.physik.fu-berlin.de/en/einrichtungen/ag/ag-eisert/teaching/ws20-21/solution05.pdf)
16. [Quantum Channels & Operations (lecture notes)](https://citeseerx.ist.psu.edu/document?doi=afa58291b0b8bd47504acb1ab8f553f0b37685cf&repid=rep1&type=pdf)
17. [Exercise Sheet 3: Quantum Channels (Adam Bílek)](https://homel.vsb.cz/~bil0127/subjects/koch/exercise-3/)
18. [Quasiprobability decompositions with reduced sampling overhead, npj Quantum Information (2022)](https://www.nature.com/articles/s41534-022-00517-3)
19. [qiskit.quantum_info.Stinespring, IBM Quantum Documentation](https://docs.quantum.ibm.com/api/qiskit/qiskit.quantum_info.Stinespring)

---
*Topic: Encyclopedia › Physical world and mathematics › Physics › Quantum physics › Quantum information science › Quantum communication and information theory › Quantum information theory › Quantum channels and capacity › Quantum channels: overview and formalism*

*Initially written Sep 17, 2026 · Reviewed: Sep 30, 2026; Oct 15, 2026 · Edited: Sep 30, 2026; Oct 15, 2026 · Last review: Oct 15, 2026*

*Copyright 2026 EdgeChat AI, a subsidiary of Biostate AI.*

License: Edgepedia Community License 1.0, https://www.edgepedia.org/edgepedia/license