---
title: Time-Independent Operator-Valued Riccati Equation
url: https://www.emergentmind.com/topics/time-independent-operator-valued-riccati-equation
type: topic
---

# Time-Independent Operator-Valued Riccati Equation

Searching arXiv for recent and foundational papers on time-independent operator-valued Riccati equations to ground the article in relevant literature.
arXiv search query: "time-independent operator-valued Riccati equation"
The time-independent operator-valued Riccati equation is a nonlinear operator equation on a Hilbert space in which the unknown is itself an operator, typically denoted \(X\) or \(P\), and the equation contains both linear and quadratic operator terms. In the literature represented here, it appears in several algebraic forms, including
\[
A_1X-XA_0-XVX+V^*=0,
\]
\[
XBX+XA-CX-B^\dagger=0,
\]
and
\[
A^*P+PA-PGP+F=0.
\]
These formulations arise in block operator matrix theory, finite-rank self-adjoint perturbations, infinite-horizon linear-quadratic control, and open quantum systems. Across these settings, the equation is tied to invariant graph subspaces, block diagonalization, spectral analysis, optimal-cost operators, feedback synthesis, and exact reduced dynamics [1403.5527] [2308.10130] [2012.05670] [1001.3541] [1006.1931].

## 1. Principal formulations and solution concepts

A common structural setting is a Hilbert space decomposition into two components. In the finite-rank perturbation framework one assumes
\[
\mathcal H=\mathcal H_0\oplus\mathcal H_1,
\]
with \(\mathcal H_1\) finite-dimensional, \(A\) self-adjoint and reducing \(\mathcal H_0,\mathcal H_1\), and an off-diagonal perturbation
\[
B=A+\begin{pmatrix}0&V^*\\V&0\end{pmatrix}
=\begin{pmatrix}A_0&V^*\\V&A_1\end{pmatrix}.
\]
The associated time-independent operator Riccati equation for an a priori unbounded operator
\[
X:\operatorname{Dom}(X)\subset\mathcal H_0\to\mathcal H_1
\]
is
\[
A_1X-XA_0-XVX+V^*=0.
\]
A densely defined operator \(X\) is called a strong solution if
\[
\operatorname{Ran}(A_0+VX)\big|_{\operatorname{Dom}(X)}\subset \operatorname{Dom}(X),
\]
and
\[
A_1Xx-XA_0x-XVXx+V^*x=0,\qquad x\in\operatorname{Dom}(X).
\]
This notion is adapted to potentially unbounded solutions constructed from spectral data [1403.5527].

A second formulation uses a Hermitian \(2\times2\) block operator matrix
\[
R=\begin{pmatrix}A&B\\B^\dagger&C\end{pmatrix},
\]
and seeks \(X:\mathcal H_E\to\mathcal H_E\) satisfying
\[
XBX+XA-CX-B^\dagger=0.
\]
This is the form used in the qubit-environment decoherence model, where the operator equation is the algebraic condition underlying block diagonalization of the total Hamiltonian [1001.3541].

A third formulation is the infinite-horizon algebraic Riccati equation on a Hilbert space \(H\),
\[
A^*P+PA-PBR^{-1}B^*P+Q=0,
\]
or, with \(F:=Q\) and \(G:=BR^{-1}B^*\),
\[
A^*P+PA-PGP+F=0.
\]
In this setting one seeks \(P\in\mathcal L(H)\), typically with \(P=P^*\ge0\), and the equation is often interpreted in weak or variational form on \(D(A^*)\) or \(D(A)\) [2308.10130] [2012.05670].

| Form | Unknown | Typical setting |
|---|---|---|
| \(A_1X-XA_0-XVX+V^*=0\) | \(X:\mathcal H_0\to\mathcal H_1\) | Off-diagonal self-adjoint perturbations |
| \(XBX+XA-CX-B^\dagger=0\) | \(X:\mathcal H_E\to\mathcal H_E\) | Block operator matrices in decoherence |
| \(A^*P+PA-PGP+F=0\) | \(P\in\mathcal L(H)\) | Infinite-horizon LQR and ARE theory |

These formulations are not interchangeable at the level of notation, but they encode the same basic phenomenon: a quadratic operator equation attached to a \(2\times2\) block structure or to the stationary limit of a control problem.

## 2. Block operator matrices, graph subspaces, and diagonalization

The Riccati equation is closely tied to invariant graph subspaces of block operator matrices. In the block-matrix setting, a solution \(X\) is used to construct a transformation that decouples the off-diagonal interaction. For the qubit-environment model, if one finds a bounded solution \(X\), then one builds
\[
U_X=
\begin{pmatrix}
1_E&-X^\dagger\\
X&1_E
\end{pmatrix},
\]
and obtains
\[
U_X^\dagger H U_X=
\begin{pmatrix}
A+BX&0\\
0&C-XB
\end{pmatrix}.
\]
Exponentiating the diagonal blocks then yields the full effective evolution in closed form as a \(2\times2\) block operator matrix [1001.3541].

The same general mechanism underlies the perturbative self-adjoint theory. There, solutions correspond to graph subspaces of the perturbed operator \(B\), and the existence of multiple graph subspaces leads to the possibility of multiple solutions. The finite-rank perturbation analysis explicitly notes that uniqueness is not addressed in full generality and that multiple bounded or unbounded solutions can occur corresponding to different invariant graph subspaces of \(B\) [1403.5527].

A common misconception is that any algebraic solution immediately furnishes a standard similarity reduction of the Schrödinger equation. The symmetry-based analysis of decoherence shows that this is false in the presence of an antilinear solution. If \(X=T\) is an antilinear involution, then
\[
S_T=
\begin{pmatrix}
I_E&-T\\
T&I_E
\end{pmatrix}
\]
is neither linear nor purely antilinear on \(\mathbb C^2\otimes H_E\). Consequently one cannot simply transform \(i\dot\Psi=H\Psi\) into a standard linear evolution with diagonal generator by setting \(\Phi(t)=S_T\Psi(t)\) [1006.1931].

These results suggest that the Riccati equation should be viewed less as an isolated nonlinear identity than as a structural criterion for decoupling a block dynamics, with the precise meaning of “solution” depending on whether boundedness, closability, linearity of the transform, or control-theoretic optimality is required.

## 3. Spectral characterization and finite-rank perturbations

For off-diagonal finite-rank perturbations, the Riccati equation is linked to detailed spectral data of the perturbed self-adjoint operator. Under the standing assumptions that \(\dim\mathcal H_1=n<\infty\) and \(\operatorname{Ran}V\) is a cyclic generating subspace for \(A_0\),
\[
\overline{\operatorname{span}}\{A_0^k v:k\in\mathbb N_0,\ v\in\operatorname{Ran}V\}=\mathcal H_0,
\]
one proves that \(\mathcal H_1\) is cyclic for \(B\), so \(\operatorname{mult}\operatorname{spec}(B)\le n\). The spectral analysis is encoded in the \(n\times n\)-valued Herglotz function
\[
M(z)=P_{\mathcal H_1}(B-z)^{-1}P_{\mathcal H_1}^*
=\bigl[(A_1-z)-V^*(A_0-z)^{-1}V\bigr]^{-1},\qquad z\in\mathbb C^+,
\]
with Herglotz representation
\[
M(z)=\int_{\mathbb R}\frac{d\Omega(t)}{t-z}.
\]
Its trace
\[
m(z)=\operatorname{tr}M(z)=\int_{\mathbb R}\frac{dw(t)}{t-z}
\]
defines a scalar measure \(w=\operatorname{tr}\Omega\) with the same null-sets as \(E_B\). Minimal supports of the singular and pure-point parts are described by
\[
S_{\mathrm{sing}}=\left\{\lambda\in\mathbb R:
\lim_{\varepsilon\to0+}\operatorname{tr}\Im M(\lambda+i\varepsilon)=+\infty\right\},
\]
\[
S_{pp}=\left\{\lambda\in\mathbb R:
\lim_{\varepsilon\to0+}\varepsilon\,\operatorname{tr}M(\lambda+i\varepsilon)>0\right\},
\]
and
\[
S_{sc}=S_{\mathrm{sing}}\setminus S_{pp}.
\]
Here \(S_{\mathrm{sing}}\) is a minimal support of the singular part of \(w\), \(S_{pp}\) supports the pure-point part, and \(S_{sc}\) supports the singularly continuous part [1403.5527].

The eigenvalues of \(B\) are characterized by solvability conditions in \(\mathcal H_1\). A point \(\lambda\in\operatorname{spec}_{pp}(B)\) of multiplicity \(k\) is characterized by the existence of \(k\) linearly independent \(y_j\in\mathcal H_1\) with \(Vy_j\in\operatorname{Ran}(A_0-\lambda)\) and one of three cases:
\[
(A_1-\lambda)y_j=V^*(A_0-\lambda)^{-1}Vy_j
\]
if \(\lambda\notin\operatorname{spec}_p(A_0)\),
\[
(A_1-\lambda)y_j=\lim_{\varepsilon\to0+}V^*(A_0-\lambda-i\varepsilon)^{-1}Vy_j
\]
if \(\lambda\in\operatorname{spec}_p(A_0)\), and a third case involving an eigenvector \(x\in\ker(A_0-\lambda)\),
\[
(A_1-\lambda)y_j=
\lim_{\varepsilon\to0+}V^*(A_0-\lambda-i\varepsilon)^{-1}Vy_j-V^*x.
\]

The existence theory is correspondingly spectral. If there is an eigenvalue \(\lambda\in\operatorname{spec}_p(B)\) of multiplicity \(n\), then there exists a bounded solution \(X\in\mathcal B(\mathcal H_0,\mathcal H_1)\) of the Riccati equation. More generally, if \(B\) has at least \(n\) eigenvalues \(\lambda_1,\dots,\lambda_n\), counted with multiplicity and all lying outside \(\operatorname{spec}_p(A_0)\), and if corresponding eigenvectors have projections onto \(\mathcal H_1\) that span \(\mathcal H_1\), then the Riccati equation has a bounded solution. In finite dimension, if \(\operatorname{spec}(B)\cap\operatorname{spec}(A_0)=\emptyset\), then the equation admits at least one bounded solution [1403.5527].

The paper also gives an explicit construction. Let
\[
K_{pp}\subset S_{pp},\qquad K_{sc}\subset S_{sc}
\]
be the spectral points for which the resolvent formula
\[
(A_1-\lambda)y=\lim_{\varepsilon\to0+}V^*(A_0-\lambda-i\varepsilon)^{-1}Vy
\]
admits a nonzero \(y\in\mathcal H_1\), with an additional divergence condition in the singularly continuous case. Choosing linearly independent pairs \((\lambda_j,y_j)\), one defines projections \(P_j\) onto \(\operatorname{span}\{y_j\}\) along the span of the remaining \(y_k\), and then sets
\[
\operatorname{Dom}(X)=
\left\{
x\in\mathcal H_0:
\lim_{\varepsilon\to0+}
\sum_{j=1}^n
P_jV^*(A_0-\lambda_j+i\varepsilon)^{-1}x
\text{ exists in }\mathcal H_1
\right\},
\]
\[
Xx=
\lim_{\varepsilon\to0+}
\sum_{j=1}^n
P_jV^*(A_0-\lambda_j+i\varepsilon)^{-1}x.
\]
Under finite-dimensional hypotheses this \(X\) is a strong solution. If any \(\lambda_j\in K_{sc}\), then \(X\) is unbounded and non-closable; if all \(\lambda_j\in K_{pp}\), then \(X\) is bounded, in fact finite-rank. The four-dimensional example with
\[
A_0=\begin{pmatrix}0&1\\1&0\end{pmatrix},\quad
A_1=\begin{pmatrix}1&1\\1&0\end{pmatrix},\quad
V=\begin{pmatrix}1&1\\1&0\end{pmatrix}
\]
illustrates the simultaneous appearance of bounded and unbounded constructions.

## 4. Infinite-horizon algebraic Riccati equations on Hilbert spaces

In control-theoretic Hilbert-space form, the time-independent operator-valued Riccati equation is the infinite-horizon algebraic Riccati equation associated with a stable semigroup and a quadratic performance index. One considers
\[
A^*P+PA-PBR^{-1}B^*P+Q=0,
\]
or equivalently
\[
A^*P+PA-PGP+F=0,
\]
where \(A:D(A)\subset H\to H\) generates an exponentially stable \(C_0\)-semigroup \(S(t)\), \(B:U\to H\) is bounded, \(R\in\mathcal L(U)\) is self-adjoint and positive-definite, and \(Q\in\mathcal L(H)\) is self-adjoint and positive semi-definite. In weak form, for all \(\phi,\psi\in D(A^*)\),
\[
\langle \psi,A^*P\phi\rangle+\langle \psi,PA\phi\rangle
-\langle \psi,PGP\phi\rangle+\langle \psi,F\phi\rangle=0.
\]
The abstract well-posedness result assumes:

- exponential stability,
\[
\|S(t)\|_{\mathcal L(H)}\le M e^{-\alpha t},\qquad t\ge0;
\]
- compactness, \(F,Q,G\in\mathcal K(H)\);
- self-adjointness and nonnegativity of \(F\) and \(G\).

Under these hypotheses there is a unique \(P=P^*\ge0\) in \(\mathcal K(H)\) solving the ARE, and it satisfies the fixed-point identity
\[
P=\int_0^\infty S(s)\,[F-PGP]\,S^*(s)\,ds.
\]
This casts the steady-state problem as a nonlinear Lyapunov-type equation driven by the semigroup [2308.10130].

The same work develops a Galerkin framework. For finite-dimensional subspaces \(H_h\subset H\) with orthogonal projector \(\pi_h\), injection \(\iota_h\), discrete generator \(A_h\), and semigroup \(S_h(t):=\iota_h\widetilde S_h(t)\pi_h\), one solves the discrete ARE
\[
A_h^*P_h+P_hA_h-P_hG_hP_h+F_h=0,
\]
with
\[
F_h=\pi_hF\iota_h,\qquad G_h=\pi_hG\iota_h.
\]
Using a Brezzi-Rappaz-Raviart argument, the error is reduced to the approximation error of the steady-state Lyapunov operator
\[
\Phi_\infty(X)=\int_0^\infty S(s)XS^*(s)\,ds,
\]
leading to the estimate
\[
\|P-P_h\|_{\mathcal L(H)}
\le C\,\|\Phi_\infty(F)-\Phi_{\infty,h}(F)\|_{\mathcal L(H)}
\le C\int_0^\infty e^{-\alpha s}\,
\|S(s)-S_h(s)\|_{\mathcal L(K,H)}\,ds.
\]
If
\[
\|S(s)-S_h(s)\|_{\mathcal L(K,H)}\le C_1 h^p
\]
uniformly in \(s\ge0\), then
\[
\|P-P_h\|_{\mathcal L(H)}\le C_2 h^p.
\]
Accordingly, the Riccati error inherits the order of convergence of the underlying semigroup approximation. The stated concrete rates are \(O(h^{k+1})\) for a parabolic generator under classical finite-element estimates and \(O(h^k)\) for a weakly damped wave operator; the corresponding functional gain
\[
\|K-K_h\|=\|R^{-1}B^*P-R^{-1}B_h^*P_h\|
\]
converges at the same predicted order [2308.10130].

## 5. Unbounded coefficients and uniqueness

Uniqueness becomes substantially subtler when the control operator is unbounded. In the infinite-horizon LQ framework with state space \(Y\), control space \(U\), and observation space \(Z\), one studies
\[
y'(t)=Ay(t)+Bu(t),\qquad y(0)=y_0\in Y,
\]
where \(A:D(A)\subset Y\to Y\) is closed, \(0\in\rho(A)\), and \(A\) generates an exponentially stable semigroup,
\[
\|e^{At}\|_{L(Y)}\le M e^{-\omega t},
\]
while
\[
B\in L(U,[D(A^*)]').
\]
The quadratic cost is
\[
J(u)=\int_0^\infty \bigl(\|Ry(t)\|_Z^2+\|u(t)\|_U^2\bigr)\,dt.
\]
The optimal-cost operator \(P\in L(Y)\) is defined by
\[
(Py_0,y_0)_Y=\inf_{u\in L^2(0,\infty;U)}J(u),
\]
and satisfies, in weak form on \(D(A)\),
\[
(A^*Px,z)_Y+(Px,Az)_Y-(B^*Px,B^*Pz)_U+(Rx,Rz)_Z=0.
\]
Because \(A\) is unbounded, the pairings involving \(A^*P\) and \(PA\) are interpreted in the appropriate dual spaces, and \(B^*P\) is only guaranteed to be densely defined on \(D(A)\) [2012.05670].

The uniqueness theorem depends on a refined decomposition of the adjoint control-semigroup map,
\[
B^*e^{A^*t}=F(t)+G(t),\qquad t>0,
\]
where \(F(t)\in L(Y,U)\) satisfies the singular estimate with exponential decay
\[
\|F(t)\|_{L(Y,U)}\le N t^{-\gamma}e^{-\omega_0 t},\qquad 0<\gamma<1,
\]
and \(G\) satisfies additional regularity, including
\[
B^*e^{A^*\cdot}A^{*-\varepsilon}\in L(Y,L^q(0,T;U))
\]
for some \(q\in(1,2)\). Together with the assumption
\[
R^*R\in L(D(A^\varepsilon),D(A^{*\varepsilon})),
\]
these hypotheses imply solvability of the LQ problem and \(B^*P\in L(D(A^\varepsilon),U)\).

Under these assumptions, the algebraic Riccati equation has exactly one solution in the class
\[
Q:=\{Q\in L(Y):Q=Q^*\ge0,\ B^*Q\in L(D(A^\varepsilon),U)\}.
\]
Moreover, that solution is precisely the optimal-cost operator. The proof is organized around the fundamental identity
\[
(Qy(t),y(t))_Y-(Qx,x)_Y
=
-\int_0^t
\left[
\|Ry(s)\|_Z^2+\|u(s)+B^*Qy(s)\|_U^2
\right]\,ds,
\]
valid for mild solutions
\[
y(t)=e^{At}x+\int_0^t e^{A(t-s)}Bu(s)\,ds,
\]
and the associated closed-loop equation
\[
y(t)=e^{At}x-\int_0^t e^{A(t-s)}BB^*Qy(s)\,ds.
\]
A comparison argument with the actual optimal control then yields \(Q=P\). This shows that uniqueness is not a generic feature of all operator Riccati equations, but a theorem that depends on the exact regularity class in which one searches for solutions [2012.05670].

## 6. Quantum decoherence, symmetry, and explicit solutions

In open quantum-system models, the time-independent operator-valued Riccati equation appears after converting a time-dependent Hamiltonian into a static block operator matrix. For a spin-\(\tfrac12\) system in a rotating magnetic field coupled to an environment,
\[
H(t,B)=H_Q(t,B)\otimes 1_E+1_Q\otimes H_E+H_{\rm int},
\]
with
\[
H_Q(t,B)=B_0\sigma_3+\alpha(\sigma_1\cos\omega t+\sigma_2\sin\omega t),
\qquad
H_{\rm int}=f(\sigma_3)\otimes V,
\]
one shows
\[
H(t,B)=e^{iKt}\,H(0,B)\,e^{-iKt},
\qquad
K=-\frac{\omega}{2}\sigma_3\otimes 1_E,
\]
and hence
\[
U(t,B)=e^{iKt}e^{-i[H(0,B)+K]t}.
\]
Writing \(B_{\rm eff}=B-\frac{\omega}{2}\), the effective time-independent Hamiltonian \(H(0,B_{\rm eff})\) has block form
\[
H=
\begin{pmatrix}
H_++\beta & \alpha\\
\alpha & H_- - \beta
\end{pmatrix},
\qquad
H_\pm=H_E\pm V,\quad \beta=B_{\rm eff}.
\]
The corresponding time-independent Riccati equation is
\[
\alpha X^2+X(H_++\beta)-(H_--\beta)X-\alpha 1_E=0.
\]
No special spectral gap assumption is needed to write down this equation, but the existence of a bounded solution generally requires suitable separation of the spectra of the diagonal blocks [1001.3541].

Several exact solution mechanisms are known. In the pure-dephasing case, with
\[
A=H_E+m_{11}V,\qquad C=H_E+m_{22}V,\qquad B=m_{12}V,
\]
the ansatz \(X=x1_E\) reduces the operator equation to the scalar quadratic
\[
m_{12}x^2+(m_{11}-m_{22})x-m_{12}=0,
\]
whose roots are explicit. By contrast, for the spin-boson coupling
\[
H_E=\int dw\,\omega\,a^\dagger(w)a(w),\qquad
V=\int dw\,[g^*(w)a(w)+g(w)a^\dagger(w)],
\]
no closed-form \(X\) is known [1001.3541].

A different exact mechanism uses symmetry. If there exists an antilinear involution \(T\) on \(H_E\) such that
\[
[H_E,T]=0,\qquad [V,T]=0,
\]
then the antilinear operator \(X=T\) is an exact solution of the Riccati equation. This result does not produce a standard linear similarity transform, because the matrix built from \(T\) is neither linear nor purely antilinear on \(\mathbb C^2\otimes H_E\) [1006.1931].

When the environment operators commute,
\[
[H_E,V]=0,
\]
there is an orthonormal basis \(\{\ket n\}\) with
\[
H_E\ket n=E_n\ket n,\qquad V\ket n=V_n\ket n,
\]
and the Riccati equation becomes a scalar quadratic for each component:
\[
\alpha x_n^2+2\beta V_n x_n-\alpha=0.
\]
This yields the positive-definite solution
\[
x_n=
\frac{\sqrt{V_n^2\beta^2+\alpha^2}-\beta V_n}{\alpha},
\]
or, in operator form,
\[
X=\sum_n
\frac{\sqrt{(V_n+\beta)^2+\alpha^2}-(V_n+\beta)}{\alpha}
\,\ket n\!\bra n
=f(V+\beta I_E).
\]

The Riccati solution feeds directly into reduced dynamics. For factorized initial states,
\[
\rho_{QE}(0)=\rho_Q(0)\otimes \rho_E,
\]
the reduced qubit evolution has Kraus form,
\[
\rho_Q(t)=\sum_n K_n(t)\rho_Q(0)K_n(t)^\dagger,
\qquad
K_n(t)=\sqrt{p_n}\,U_n(t).
\]
If the initial state is correlated, a single Kraus sum valid for all correlations is not available in general; a necessary and sufficient condition for that would be factorization of the total propagator as \(U_t=U_Q(t)\otimes U_E(t)\), which does not hold here. The reduced dynamics can nevertheless be written in the manageable operator form
\[
\rho_Q(t)=\sum_{n,i,j}\epsilon_{n,ij}\,
U_n(t)\rho_Q(0)U_n(t)^\dagger,
\qquad
\epsilon_{n,ij}=\langle i,n|\rho_{QE}(0)|j,n\rangle.
\]
This quantum-mechanical line of work emphasizes that the time-independent operator-valued Riccati equation is not merely a control-theoretic object: it is also an exact solvability condition for block Hamiltonians, and its bounded, unbounded, linear, or antilinear solutions have distinct dynamical consequences [1001.3541] [1006.1931].

Source: https://www.emergentmind.com/topics/time-independent-operator-valued-riccati-equation