Liouvillian Superoperators

A Liouvillian is a superoperator that generates time evolution of operators, most often density operators. For a closed system,

$$\frac{d\rho}{dt} = \mathcal L_H(\rho), \qquad \mathcal L_H(\rho) = -\frac{i}{\hbar}[H,\rho].$$

The equation is the same Liouville–von Neumann Equation written as a linear differential equation on operator space. This page explains the superoperator viewpoint: what space the Liouvillian acts on, how its spectrum is interpreted, how vectorized matrix forms are built, and how the closed-system generator sits inside open-system master equations.

Operators as Vectors

In a finite-dimensional Hilbert space $\mathcal H$ with $\dim\mathcal H=d$, the set of linear operators on $\mathcal H$ is a $d^2$-dimensional vector space. A density operator is a special point in this vector space: positive and trace one.

It is often useful to temporarily forget positivity and trace normalization and treat an operator $X$ as a vector. The Hilbert–Schmidt inner product is

$$\langle A,B\rangle_{\rm HS} = \operatorname{Tr}(A^\dagger B).$$

With this inner product, an operator basis such as

$$E_{mn} = \lvert m\rangle\langle n\rvert$$

can be used like a basis of vectors. A superoperator is then a linear map from operators to operators:

$$\mathcal S: X\longmapsto \mathcal S(X).$$

The physical state space is not the whole vector space, but the whole vector space is the natural domain for linear algebra. This is why eigenoperators, kernels, left modes, and matrix representations appear in master-equation work.

In infinite-dimensional problems, one must specify domains and operator classes. Density operators are trace-class, while Hilbert–Schmidt operators form a useful Hilbert space of operators. For the functional-analytic background, see Trace-Class and Hilbert-Schmidt Operators.

Superoperator Adjoint

For the Hermiticity-preserving maps used here, the Hilbert–Schmidt adjoint also gives the dual map between states and observables. Its trace-pairing identity is

$$\operatorname{Tr}\!\left[ (\mathcal S\rho)A \right] = \operatorname{Tr}\!\left[ \rho\,\mathcal S^\dagger(A) \right].$$

For the Hamiltonian Liouvillian,

$$\mathcal L_H(\rho) = -\frac{i}{\hbar}[H,\rho],$$

the adjoint acts on observables as

$$\mathcal L_H^\dagger(A) = \frac{i}{\hbar}[H,A].$$

This is the Heisenberg generator. The sign difference between states and observables is exactly what keeps expectation values consistent:

$$\frac{d}{dt}\operatorname{Tr}(\rho A) = \operatorname{Tr}\!\left[ \mathcal L_H(\rho)A \right] = \operatorname{Tr}\!\left[ \rho\,\mathcal L_H^\dagger(A) \right].$$

For the observable-side interpretation of commutators as generators, see Commutator Dynamics.

Hamiltonian Liouvillian

Introduce left and right multiplication maps,

$$\mathsf L_A(X)=AX, \qquad \mathsf R_A(X)=XA.$$

Then the closed-system Hamiltonian Liouvillian is

$$\mathcal L_H = -\frac{i}{\hbar} \left( \mathsf L_H-\mathsf R_H \right).$$

This notation makes clear that $\mathcal L_H$ is not the Hamiltonian. It is a map built from two ways of multiplying by the Hamiltonian.

For any operator $X$,

$$\operatorname{Tr}\mathcal L_H(X) = -\frac{i}{\hbar}\operatorname{Tr}(HX-XH) = 0.$$

Thus the generator has zero trace on every input; its exponential, rather than the generator itself, is trace preserving. If $X=X^\dagger$, then $\mathcal L_H(X)$ is also Hermitian as a tangent vector to the Hermitian operator space. Under the finite-dimensional Hilbert–Schmidt inner product, $\mathcal L_H$ is skew-adjoint:

$$\langle A,\mathcal L_H(B)\rangle_{\rm HS} = -\langle \mathcal L_H(A),B\rangle_{\rm HS}.$$

Consequently its eigenvalues are purely imaginary when it is diagonalized as a closed-system generator.

Spectrum in the Energy Basis

Let

$$H\lvert n\rangle = E_n\lvert n\rangle.$$

The operator basis element $\lvert m\rangle\langle n\rvert$ is an eigenoperator of the closed-system Liouvillian:

$$\begin{aligned} \mathcal L_H \left( \lvert m\rangle\langle n\rvert \right) &= -\frac{i}{\hbar} \left[ H,\lvert m\rangle\langle n\rvert \right] \\ &= -\frac{i}{\hbar} (E_m-E_n) \lvert m\rangle\langle n\rvert. \end{aligned}$$

Thus the Liouvillian eigenvalue is

$$\lambda_{mn} = -\frac{i}{\hbar}(E_m-E_n).$$

Diagonal populations have $\lambda_{nn}=0$. Off-diagonal coherences rotate at Bohr frequencies. If $E_m=E_n$ because of degeneracy, the corresponding coherence is also a zero mode of the closed Hamiltonian Liouvillian.

This spectrum is a compact way to restate the density-matrix element solution:

$$\rho_{mn}(t) = e^{-i(E_m-E_n)t/\hbar}\rho_{mn}(0).$$

For a closed system, zero modes indicate operators that commute with $H$. They do not imply attraction to a unique full density operator: unitary dynamics preserves distances between states. Subsystems or restricted observables can nevertheless equilibrate under a closed many-body Hamiltonian.

Vectorization

For numerical work, a superoperator can be represented as an ordinary matrix acting on a vectorized density matrix. This page uses column-stacking:

$$\operatorname{vec}(AXB) = (B^{\mathsf T}\otimes A)\operatorname{vec}(X).$$

With this convention,

$$\operatorname{vec}(H\rho) = (I\otimes H)\operatorname{vec}(\rho),$$

and

$$\operatorname{vec}(\rho H) = (H^{\mathsf T}\otimes I)\operatorname{vec}(\rho).$$

Therefore

$$\frac{d}{dt}\operatorname{vec}(\rho) = \mathbb L_H\,\operatorname{vec}(\rho),$$

where the matrix representation of the Liouvillian is

$$\mathbb L_H = -\frac{i}{\hbar} \left( I\otimes H - H^{\mathsf T}\otimes I \right).$$

The transpose appears because of the chosen vectorization convention. Row-stacking or a different Choi convention changes where transposes appear. For example, column-stacking a $2\times2$ matrix orders its entries as $(X_{11},X_{21},X_{12},X_{22})^{\mathsf T}$. This order must agree with the Kronecker factors in code.

A complex Hamiltonian checks the transpose

A real symmetric Hamiltonian cannot reveal a mistaken replacement of $H^{\mathsf T}$ by $H^\dagger$. Take instead $H=\hbar\Omega\sigma_y/2$, with real $\Omega$, so that $H^\dagger=H$ but $H^{\mathsf T}=-H$. In the stated column-stacking order,

$$\mathbb L_H=\frac{\Omega}{2} \begin{pmatrix} 0&-1&-1&0\\ 1&0&0&-1\\ 1&0&0&-1\\ 0&1&1&0 \end{pmatrix}.$$

For $\rho_0=\lvert0\rangle\langle0\rvert$, the first column gives $\operatorname{vec}(\dot\rho_0)=(0,\Omega/2,\Omega/2,0)^{\mathsf T}$. The two off-diagonal entries agree, as required for a Hermitian derivative. The exact spin rotation gives the further check

$$\rho(t)= \begin{pmatrix} \cos^2(\Omega t/2)&\tfrac12\sin\Omega t\\ \tfrac12\sin\Omega t&\sin^2(\Omega t/2) \end{pmatrix}.$$

Both the operator commutator and the vectorized matrix must produce this same evolution. The exercise below shows the error caused by inserting an adjoint where the vectorization identity requires a transpose.

Finite-Time Propagator

A time-independent Liouvillian generates a superoperator exponential:

$$\rho(t) = e^{t\mathcal L_H}\rho(0).$$

For the Hamiltonian Liouvillian, this abstract exponential equals unitary conjugation:

$$e^{t\mathcal L_H}\rho = e^{-iHt/\hbar}\rho\,e^{iHt/\hbar}.$$

In vectorized form,

$$\operatorname{vec}(\rho(t)) = e^{t\mathbb L_H} \operatorname{vec}(\rho(0)).$$

The matrix exponential $e^{t\mathbb L_H}$ is not a state-vector unitary on $\mathcal H$. It is the matrix representation of a channel acting on operator space. For the closed generator it is also a unitary matrix on Hilbert–Schmidt operator space. A general superoperator exponential is a physical channel only when its generator satisfies the needed complete-positivity and trace conditions.

Closed Versus Open Liouvillians

The closed-system Liouvillian is only the Hamiltonian commutator:

$$\mathcal L_H(\rho) = -\frac{i}{\hbar}[H,\rho].$$

An open-system Markovian Liouvillian adds dissipative terms. In a common Lindblad–GKSL convention,

$$\mathcal L(\rho) = -\frac{i}{\hbar}[H,\rho] + \sum_k\gamma_k \left( L_k\rho L_k^\dagger - \frac12\{L_k^\dagger L_k,\rho\} \right), \qquad \gamma_k\ge0.$$

The first term is the Hamiltonian Liouvillian. The rest is not generated by a Hamiltonian on the system alone; it represents reduced dynamics after environmental or measurement degrees of freedom have been eliminated under additional assumptions.

With the column-stacking convention, one dissipator contributes the matrix

$$\mathbb D[L] = L^*\otimes L - \frac12 I\otimes L^\dagger L - \frac12 (L^\dagger L)^{\mathsf T}\otimes I.$$

This formula is a convention-dependent computational representation, not a replacement for the operator equation. The open-system canonical home is Lindblad–GKSL Equation.

Steady States and Left Modes

A stationary operator lies in the kernel of the Liouvillian:

$$\mathcal L(X)=0.$$

For a closed system, all operators commuting with $H$ are stationary. For an open Markovian system, positive trace-one elements of $\ker\mathcal L$ are steady states.

Trace preservation is expressed by the adjoint condition

$$\mathcal L^\dagger(I)=0.$$

Thus the identity is a left zero mode of any trace-preserving Liouvillian. Additional left zero modes correspond to conserved observables or sector labels. The open-system theory of kernels, gaps, metastability, and relaxation modes is developed in Steady States and Relaxation.

Common Mistakes

• Calling $\mathcal L$ a Hamiltonian. It acts on operators, not on state vectors.
• Mixing vectorized and unvectorized notation without declaring the conversion.
• Using a row-stacking formula with column-stacked code, or the reverse.
• Interpreting closed-system imaginary Liouvillian eigenvalues as decay rates.
• Treating every zero mode as a physical state; stationary states must still be positive and trace one.
• Assuming an open-system Liouvillian is valid without checking trace preservation, complete positivity, and approximation assumptions.
• Forgetting that infinite-dimensional Liouvillians require domain and operator-class care.

References

• L. E. Ballentine, Quantum Mechanics: A Modern Development, 2nd ed., World Scientific, 2014.
• H.-P. Breuer and F. Petruccione, The Theory of Open Quantum Systems, Oxford University Press, 2002.
• V. Gorini, A. Kossakowski, and E. C. G. Sudarshan, “Completely positive dynamical semigroups of N-level systems,” Journal of Mathematical Physics 17, 821-825, 1976.
• G. Lindblad, “On the generators of quantum dynamical semigroups,” Communications in Mathematical Physics 48, 119-130, 1976.
• M. A. Nielsen and I. L. Chuang, Quantum Computation and Quantum Information, Cambridge University Press, 2010.
• R. Shankar, Principles of Quantum Mechanics, 2nd ed., Springer, 1994.
• J. von Neumann, Mathematical Foundations of Quantum Mechanics, Princeton University Press, 1955.

Exercises

1. Show that the Hamiltonian Liouvillian preserves trace.

Solution

For any trace-class $X$ for which the commutator is defined,

$$\operatorname{Tr}\mathcal L_H(X) = -\frac{i}{\hbar} \operatorname{Tr}(HX-XH).$$

By cyclicity,

$$\operatorname{Tr}(HX) = \operatorname{Tr}(XH),$$

so $\operatorname{Tr}\mathcal L_H(X)=0$.

2. Let $H\lvert n\rangle=E_n\lvert n\rangle$. Verify that $\lvert m\rangle\langle n\rvert$ is a Liouvillian eigenoperator.

Solution

Compute

$$H\lvert m\rangle\langle n\rvert = E_m\lvert m\rangle\langle n\rvert,$$

and

$$\lvert m\rangle\langle n\rvert H = E_n\lvert m\rangle\langle n\rvert.$$

Therefore

$$\mathcal L_H \left( \lvert m\rangle\langle n\rvert \right) = -\frac{i}{\hbar}(E_m-E_n) \lvert m\rangle\langle n\rvert.$$

3. Using column-stacking vectorization, derive the matrix form of the Hamiltonian Liouvillian. For $H=\hbar\Omega\sigma_y/2$ with $\Omega\ne0$ and $\rho_0=\lvert0\rangle\langle0\rvert$, what derivative results if $H^{\mathsf T}$ is incorrectly replaced by $H^\dagger$?

Solution

With

$$\operatorname{vec}(AXB) = (B^{\mathsf T}\otimes A)\operatorname{vec}(X),$$

one has

$$\operatorname{vec}(H\rho) = (I\otimes H)\operatorname{vec}(\rho),$$

and

$$\operatorname{vec}(\rho H) = (H^{\mathsf T}\otimes I)\operatorname{vec}(\rho).$$

Since

$$\mathcal L_H(\rho) = -\frac{i}{\hbar}(H\rho-\rho H),$$

the vectorized equation is

$$\operatorname{vec}(\dot\rho) = -\frac{i}{\hbar} \left( I\otimes H - H^{\mathsf T}\otimes I \right) \operatorname{vec}(\rho).$$

The incorrect matrix would be $-i(I\otimes H-H^\dagger\otimes I)/\hbar$. On $\operatorname{vec}(\rho_0)=(1,0,0,0)^{\mathsf T}$ it gives $(0,\Omega/2,-\Omega/2,0)^{\mathsf T}$. Undoing column-stacking yields

$$\dot\rho_{\mathrm{wrong}}(0) =\frac{\Omega}{2} \begin{pmatrix}0&-1\\1&0\end{pmatrix}, \qquad \dot\rho_{\mathrm{correct}}(0) =\frac{\Omega}{2}\sigma_x.$$

The wrong derivative is anti-Hermitian and nonzero, so it immediately violates the Hermiticity required of a density operator.

4. Explain why $\mathcal L^\dagger(I)=0$ is the adjoint form of trace preservation.

Solution

Trace preservation at the generator level means

$$\operatorname{Tr}\mathcal L(\rho)=0$$

for all density operators, and by linearity for all operators in the relevant trace-class space. Using the adjoint definition,

$$\operatorname{Tr}\mathcal L(\rho) = \operatorname{Tr}\!\left[ \rho\,\mathcal L^\dagger(I) \right].$$

If this vanishes for all $\rho$, then $\mathcal L^\dagger(I)=0$. Conversely, if $\mathcal L^\dagger(I)=0$, then the trace derivative vanishes for every input.