Guide for Quantum Information

For a quantum information reader, rigorous quantum mechanics extends familiar density matrices and channels into settings where domains, trace ideals, continuity, and representation choices matter. The most useful reading habit is to identify the space on which a map acts before manipulating its formula. A transformation of states and its dual transformation of observables have different inputs, different normalization conditions, and opposite directions.

The examples assume familiarity with finite-dimensional density matrices, effects, tensor products, and Kraus operators. Definitions of the infinite-dimensional objects are developed in the linked treatments.

Identify the state space and the observable space

For a Hilbert space $\mathcal H$, bounded observables belong to $\mathcal B(\mathcal H)$, while density operators belong to the trace class $\mathcal T_1(\mathcal H)$. Their pairing is

$$(\rho,A)\longmapsto\operatorname{Tr}(\rho A).$$

In finite dimension these spaces contain the same matrices, which makes it easy to overlook their distinct roles. In infinite dimension, $I$ is bounded but not trace class, and an arbitrary bounded positive operator cannot be normalized into a density operator.

A density operator defines a normal state on $\mathcal B(\mathcal H)$. General algebraic states are positive normalized functionals and need not have density-operator representatives in that representation. Whenever a theorem says “state,” determine which class it quantifies over. Normal States explains the precise correspondence; Algebraic States explains the broader formulation.

An unbounded Hamiltonian is not an element of $\mathcal B(\mathcal H)$. Its bounded spectral projections still define measurement probabilities for every density operator, even when an energy moment diverges. Thus “this state has a well-defined energy distribution” is weaker than “this state has finite mean energy.”

Read a channel in both directions

In the Schrödinger picture, a channel has the direction

$$\Phi:\mathcal T_1(\mathcal H_{\rm in}) \longrightarrow\mathcal T_1(\mathcal H_{\rm out}).$$

It is completely positive and trace preserving. Its bounded-observable dual has the opposite direction:

$$\Phi^*:\mathcal B(\mathcal H_{\rm out}) \longrightarrow\mathcal B(\mathcal H_{\rm in}),$$

and is characterized by

$$\operatorname{Tr}(\Phi(\rho)A) =\operatorname{Tr}(\rho\,\Phi^*(A)).$$

The dual is completely positive, unital, and normal. Conversely, a normal unital completely positive map on the bounded-observable spaces has a trace-preserving predual. Normality is the relevant infinite-dimensional continuity requirement; finite-dimensional matrix calculations do not reveal why it is needed.

An output effect $0\le A\le I$ therefore pulls back to the input effect $\Phi^*(A)$. This gives a concrete interpretation of the duality: propagate the preparation forward or propagate the question asked by the detector backward. Both descriptions give the same probability.

Worked reading example: trace preservation is not unitality

For a qubit amplitude-damping channel, let $0<\gamma<1$ and

$$K_0=\begin{pmatrix}1&0\\0&\sqrt{1-\gamma}\end{pmatrix}, \qquad K_1=\begin{pmatrix}0&\sqrt{\gamma}\\0&0\end{pmatrix}.$$

The state map is $\Phi(\rho)=\sum_jK_j\rho K_j^*$. Its normalization test is

$$\sum_jK_j^*K_j=I,$$

so $\Phi$ is trace preserving and $\Phi^*(I)=I$. However,

$$\Phi(I)=\sum_jK_jK_j^* =\begin{pmatrix}1+\gamma&0\\0&1-\gamma\end{pmatrix}\ne I.$$

There is no contradiction: the two sums refer to different maps. The excited-state detector $E_1=|1\rangle\langle1|$ pulls back to $\Phi^*(E_1)=(1-\gamma)E_1$, expressing the probability that an excitation survives.

When a paper calls a channel “unital,” first determine its picture. In the Heisenberg picture unitality is the normalization condition. In the Schrödinger picture it is an additional property, when the expression is defined; in infinite dimension $I$ is not an allowed trace-class input.

Check what survives an infinite-dimensional limit

Finite-dimensional proofs often use compactness of the state space, finiteness of entropy, or equivalence of matrix norms without naming them. These features cannot be carried over indiscriminately. A sequence of normal states may escape every fixed finite-dimensional subspace; entropies may be infinite; convergence in one operator topology may fail to control operational state error.

For density operators, trace distance controls the worst difference in probabilities over all effects. For channels, distinguish convergence on each fixed input from a uniform statement allowing an arbitrary reference system; those are different tasks. Energy constraints can make otherwise unrealistic uniform comparisons physically meaningful, but they must be stated in the theorem rather than inferred from a numerical cutoff.

Use Trace Norm and Operator Norm, Quantum Channels in Infinite Dimensions, and Tracking Topologies to identify the relevant tests. The aim is to retain the operational question while checking the analytic assumptions that support its answer.

Exercise: reconstruct the input question

For the damping channel above, the input state is $\rho$ and the output is tested for its ground-state outcome $E_0=I-E_1$. Find the input effect that gives the same probability.

Solution

Unitality of the dual gives

$$\Phi^*(E_0)=I-(1-\gamma)E_1 =E_0+\gamma E_1.$$

The ground-state detection probability is therefore $\rho_{00}+\gamma\rho_{11}$. The result combines preparations already in the ground state with excited preparations that decay. It is an effect but is not a projection for $0<\gamma<1$, illustrating how a sharp output question becomes an unsharp input question.

References

• A. S. Holevo, Quantum Systems, Channels, Information: A Mathematical Introduction, 2nd ed., De Gruyter, 2019 — normal states, channel duality, and infinite-dimensional systems.
• J. Watrous, The Theory of Quantum Information, Cambridge University Press, 2018, Chapters 2–3 — finite-dimensional channels, dual maps, and operational distances.