Open Quantum Systems and Reduced Dynamics

Reduced dynamics is the daily working language of every quantum-hardware laboratory. When a group reports a superconducting qubit with $T_1=300\ \mu\mathrm{s}$ and $T_2=200\ \mu\mathrm{s}$, they are quoting the amplitude-damping and phase-damping rates of a reduced channel obtained by tracing out the qubit's electromagnetic and two-level-fluctuator environments; device simulators integrate a Lindblad master equation with exactly these collapse operators to predict gate and readout fidelities before a chip is fabricated. Gate-set and process tomography reconstruct the full CPTP map a physical operation implements, using the Choi–Jamiołkowski isomorphism to encode the channel in a positive Choi matrix that is measured directly — a model-independent verdict on how badly the environment corrupts the intended unitary. Fault-tolerant error correction is built on this picture: the threshold theorem assumes noise is a CPTP channel of bounded strength, and the specific error operators ($X$, $Z$, amplitude damping) set which logical errors a code must catch. At the frontier, the field is actively pushing beyond the memoryless limit — engineered non-Markovian baths that feed coherence back into the qubit, dynamical decoupling sequences that refocus dephasing, and reservoir-engineering schemes that use a tailored dissipator $\mathcal{D}$ to stabilize entangled states as the steady state of an open system rather than fight decoherence head-on. The same partial trace that once explained radiation damping now underwrites the entire engineering discipline of keeping quantum information alive.

Picture $S$ as a coin and $E$ as a meddlesome dealer you can never watch. The joint state may be perfectly definite — the dealer knows exactly how the coin correlates with his hidden record — yet your local description of the coin must average over everything the dealer might be holding. If his record is correlated with your coin, that average washes out the coin's definiteness and it looks random to you. The partial trace is precisely the operation "average over the part you cannot access," and entanglement is what makes that average throw information away rather than merely coarse-grain it. Purity is the currency of openness. A genuinely closed system keeps $\mathrm{Tr}\,\rho^2=1$ forever; an open one bleeds purity into system–bath entanglement. The crucial subtlety is that the global state stays pure the whole time, so the lost purity is not annihilated — it is hidden inside correlations between $S$ and $E$. That is why decoherence is, in principle, reversible: if you could reach back and grab the entire environment, the purity would return. In practice the environment has $10^{20}$ degrees of freedom fleeing at the speed of light, so you never can. The first figure fixes the whole idea in one relation: a globally pure Bell pair, locally maximally mixed.

The same partial trace explains how a state decoheres in time, not just that it can. When the interaction is diagonal in a preferred basis — the pointer basis singled out by what the environment monitors — the populations along the diagonal of $\rho_S$ are frozen while the off-diagonal coherences are multiplied by a bath-overlap factor whose modulus shrinks toward zero. The result is einselection: the system is driven to behave classically in the pointer basis, superpositions of pointer states become unobservable, and the timescale of that decay is the coherence time $T_2$. The second figure is the working picture of decoherence a device physicist actually watches on an oscilloscope: populations steady, coherence bleeding away exponentially.

The product-state assumption is the small print that turns messy physics into clean mathematics. Demand that $S$ and $E$ start uncorrelated, $\rho_{SE}=\rho_S\otimes\rho_E$, and the induced map is automatically completely positive — it stays a valid state-to-state map even when the input $\rho_S$ is secretly half of an entangled pair shared with an untouched bystander qubit. That "complete" positivity, stronger than mere positivity, is exactly the property that guarantees Kraus operators and quantum channels exist, and it is the hinge on which this whole article turns.

The starting point is the definition of the accessible state. Among all operators on $\mathcal{H}_S$, exactly one reproduces every system expectation value, and the partial trace constructs it by summing the environment's diagonal blocks.

where: $\rho_{SE}$ — joint state on $\mathcal{H}_S\otimes\mathcal{H}_E$; $\mathrm{Tr}_E$ — partial trace over the environment; $\{|e_j\rangle\}$ — any orthonormal basis of $\mathcal{H}_E$ (the answer is basis-independent); $\rho_S$ — the unique operator satisfying $\mathrm{Tr}(\rho_S A)=\mathrm{Tr}(\rho_{SE}\,A\otimes I)$ for every system observable $A$. Read it as: throw away the environment, keep the local statistics. Now let the joint system evolve. The closed system runs by the unitary $U_{SE}=e^{-iH_{SE}t/\hbar}$, and the reduced state at the end is what you get by tracing out $E$ afterward. With the product initial state this defines a channel $\mathcal{E}$ on $S$ alone — the Stinespring "dilate, evolve, project" form.

where: $U_{SE}$ — joint unitary generated by the full Hamiltonian $H_{SE}=H_S\otimes I+I\otimes H_E+H_{\mathrm{int}}$; $\rho_E$ — fixed initial bath state (the product assumption); $\mathcal{E}$ — the resulting quantum channel on $S$. Reading: the messy open dynamics is a clean unitary on a bigger space, glimpsed through a partial trace. Hidden inside equation (2) is a finite list of system operators. Expanding the partial trace in a bath basis peels the channel apart into an operator-sum, the representation used everywhere in quantum information.

where: $E_k=\langle e_k|U_{SE}|e_0\rangle$ — Kraus operators on $\mathcal{H}_S$, the bath matrix elements of the joint unitary; $\sum_k E_k^\dagger E_k=I$ — the completeness relation, equivalent to trace preservation. Reading: a channel is a weighted "branch" over environment outcomes we never read. The whole point of openness is that these channels can lower purity, so we need scalars that grade how mixed a state is. Two do the job — purity and von Neumann entropy.

where: $\mathcal{P}$ — purity, equal to $1$ iff $\rho_S$ is pure and bounded below by $1/d$ in dimension $d$; $S(\rho_S)$ — von Neumann entropy, $0$ for pure states and $\log d$ for the maximally mixed state. For a globally pure state the Schmidt decomposition forces $S(\rho_S)=S(\rho_E)$ — the entanglement entropy. The clean CPTP structure is not automatic; it is bought with the product initial condition. This is the single most important qualifier in the subject.

where: CP — completely positive, meaning $\mathcal{E}\otimes\mathrm{id}_R$ maps positive operators to positive operators for a reference $R$ of any dimension; TP — trace preserving, $\mathrm{Tr}\,\mathcal{E}(\rho)=\mathrm{Tr}\,\rho$. Reading: uncorrelated start in, valid channel out. For continuous time we want a differential law rather than a snapshot. When the bath is large, weak, and memoryless, the exact dynamics collapse to a generator with a coherent commutator plus a dissipator.

where: $H_S$ — effective (Lamb-shifted) system Hamiltonian; $\mathcal{D}$ — the Lindblad dissipator; $L_k$ — jump operators; $\gamma_k\ge0$ — rates fixed by the bath spectral density. Reading: the commutator rotates, the dissipator relaxes and dephases. Finally, the canonical single-qubit example: a two-level atom emitting one quantum into a zero-temperature bath is the amplitude-damping channel, with two Kraus operators.

where: $p$ — probability that the excited state $|1\rangle$ decays to $|0\rangle$ by spontaneous emission, related to the relaxation time by $p=1-e^{-t/T_1}$. Reading: $E_1$ is the "a photon left" branch; $E_0$ is the "no photon, but I now know that" branch that still damps the coherence.

Reduced dynamics interpolates between several familiar regimes as the coupling strength, bath size, and characteristic timescales are dialed:

1 · A Bell pair is maximally mixed (fully worked)

Take a system qubit $S$ and an environment qubit $E$ in the Bell state $|\Phi^+\rangle=(|00\rangle+|11\rangle)/\sqrt2$. The joint state is pure. Find $\rho_S$, its purity, and its von Neumann entropy, and interpret the result.

Step 1. Write the global density operator: $\rho_{SE}=|\Phi^+\rangle\langle\Phi^+|=\tfrac12(|00\rangle+|11\rangle)(\langle00|+\langle11|)$. It is pure, so $\mathrm{Tr}\,\rho_{SE}^2=1$ exactly — the joint description holds complete information.

Step 2. Trace over the bath qubit: $\rho_S=\sum_{j\in\{0,1\}}\langle j|_E\,\rho_{SE}\,|j\rangle_E$. The cross terms carry factors like $\langle j|1\rangle_E\langle0|j\rangle_E$, which vanish by orthogonality of the bath states, so only the "matched" terms survive.

Step 3. Collect them: $\rho_S=\tfrac12\big(|0\rangle\langle0|+|1\rangle\langle1|\big)=\tfrac12 I$. The populations survive and the coherences are gone — the reduced state is the maximally mixed qubit, indistinguishable from a fair classical coin.

Step 4. Read off the mixedness. Purity $\mathcal{P}=\mathrm{Tr}\,\rho_S^2=(\tfrac12)^2+(\tfrac12)^2=\tfrac12$, and entropy $S(\rho_S)=-2\cdot\tfrac12\log_2\tfrac12=\mathbf{1\ bit}$, while $S(\rho_E)=1$ bit likewise — equal, as the Schmidt decomposition demands. A globally pure state has a maximally mixed part ($\mathcal{P}=1/2$, $S=1$ bit): the information is not destroyed, it has migrated entirely into the S–E entanglement.

2 · Amplitude damping of a superposition

A qubit is prepared in $|+\rangle=(|0\rangle+|1\rangle)/\sqrt2$ and sent through the amplitude-damping channel of equation (7) with decay probability $p=0.5$ (that is, half of one $T_1$ has elapsed, $p=1-e^{-t/T_1}$). Compute the output state and its purity, and relate the coherence decay to $T_2$.

Step 1. The input is $\rho=\tfrac12\begin{pmatrix}1&1\\1&1\end{pmatrix}$, pure with $\mathcal{P}=1$. Apply the two Kraus operators. The "no decay" branch gives $E_0\rho E_0^\dagger=\tfrac12\begin{pmatrix}1&\sqrt{1-p}\\\sqrt{1-p}&1-p\end{pmatrix}$ and the "decay" branch gives $E_1\rho E_1^\dagger=\tfrac12\begin{pmatrix}p&0\\0&0\end{pmatrix}$.

Step 2. Sum the branches: $\rho'=\tfrac12\begin{pmatrix}1+p&\sqrt{1-p}\\\sqrt{1-p}&1-p\end{pmatrix}$. At $p=0.5$ this is $\rho'=\begin{pmatrix}0.75&0.354\\0.354&0.25\end{pmatrix}$, with $\mathrm{Tr}\,\rho'=1$ as required. Population has flowed toward the ground state ($\rho_{00}$ rose from $0.5$ to $\mathbf{0.75}$) and the coherence shrank from $0.5$ to $\sqrt{1-p}/2=\mathbf{0.354}$.

Step 3. Purity: using $\mathcal{P}=\tfrac{2-p+p^2}{2}$ for this input, at $p=0.5$ we get $\mathcal{P}=(2-0.5+0.25)/2=\mathbf{0.875}$ — the state is no longer pure. Note the coherence decays as $\sqrt{1-p}=e^{-t/2T_1}$ while the population offset decays as $1-p=e^{-t/T_1}$: the coherence envelope falls at half the rate, which is the microscopic origin of the famous $T_2=2T_1$ ceiling for a purely radiatively damped qubit.

Step 4. Sanity check the endpoints. At $p\to0$, $\rho'\to\rho$ and $\mathcal{P}\to1$ (nothing happened). At $p\to1$, $\rho'\to|0\rangle\langle0|$ and $\mathcal{P}\to1$ again — the qubit has fully relaxed to the ground state, a pure state, so purity dips to a minimum of $0.875$ at $p=0.5$ and then recovers. Decoherence is not monotone in purity: amplitude damping erases coherence on the way down but re-purifies the qubit once it has surely decayed.

The reduced density operator was not born from quantum information theory — it predates it by half a century. It was invented by a nineteen-year-old solving a problem in radiation damping, and it took decades of work by mathematicians and physicists to turn that object into the CPTP-map machinery of today.

Texts. Nielsen and Chuang, Quantum Computation and Quantum Information, Ch. 8 (quantum noise and quantum operations) · Preskill, Lecture Notes on Quantum Computation, Ch. 3 (measurement and evolution) · Watrous, The Theory of Quantum Information, Ch. 2 and 4 · Wilde, Quantum Information Theory, Ch. 4 (the noisy quantum theory) · Breuer and Petruccione, The Theory of Open Quantum Systems, Ch. 3 · Kitaev, Shen, Vyalyi, Classical and Quantum Computation, §11. The key results in this article are: (1) the operator-sum derivation gives $E_k=\langle e_k|U_{SE}|e_0\rangle$ with $\sum_k E_k^\dagger E_k=I$; (2) the Bell-state reduced state is $\rho_S=I/2$, with purity $1/2$ and entropy $1$ bit; (3) the amplitude-damping output is $\rho'=\tfrac12\begin{pmatrix}1+p&\sqrt{1-p}\\\sqrt{1-p}&1-p\end{pmatrix}$, giving purity $\tfrac{2-p+p^2}{2}=0.875$ at $p=0.5$, and $T_2=2T_1$ for pure radiative damping; (4) the phase-damping decoherence factor is $e^{-\Gamma}=2\lambda-1$ (off-diagonal ratio $0.6$ at $\lambda=0.8$); (5) complete positivity follows from the Stinespring dilation and the product initial condition; (6) the Schmidt decomposition enforces $S(\rho_S)=S(\rho_E)$. The $300\ \mu\mathrm{s}$/$200\ \mu\mathrm{s}$ figures quoted in the applications section are illustrative of the current scale, not cited experimental data.

Prerequisites. Density Operators and Mixed States · Partial Trace · Composite Systems and the Tensor Product · Unitary Evolution and the Schrödinger Equation. Leads to. Quantum Operations and CPTP Maps · Kraus Representation of Quantum Channels · Decoherence · Lindblad / GKSL Master Equation · Markovian vs Non-Markovian Dynamics · Process Tomography and the Choi–Jamiołkowski Isomorphism.

Ask the tutor: Curious how the abstract $\rho_S=\mathrm{Tr}_E\,\rho_{SE}$ becomes the concrete $T_1$ and $T_2$ on a real chip — or why a Bell pair looks like a fair coin? Ask me to walk you through any Kraus channel, the decoherence factor, or the road from partial trace to the Lindblad equation.