Computing Green’s functions and improving ground state energy estimation on quantum computers with Liouvillian recursion
J. Leblanc, O. Nahman-Lévesque, J. Forget, T. Lepage-Lévesque, S. Verret, A. Foley · 2026
Read Computing Green’s functions and improving ground state energy estimation on quantum computers with Liouvillian recursion by J. Leblanc, O. Nahman-Lévesque, J. Forget, T. Lepage-Lévesque, S. Verret, A. Foley (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-05349.txt.
Introduction
Jérôme Leblanc and team present hybrid Liouvillian recursion for Green's functions on a quantum computer. From a ground state circuit, the algorithm makes local and inter-site Green's functions. They show it on a superconducting four-site Hubbard processor. Computed Green's functions feed the Galitskii-Migdal formula, beating Hamiltonian expectation for energy figures. Results show exponential settling and polynomial hardness in Green's-function accuracy. Results also show toughness to noise and preparation inaccuracies.
Catalog record
| CREATOR | J. Leblanc, O. Nahman-Lévesque, J. Forget, T. Lepage-Lévesque, S. Verret, A. Foley |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 29293 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2603.05349
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Computing Green’s functions and improving ground state energy estimation on quantum computers with Liouvillian recursion J. Leblanc, O. Nahman-Lévesque, J. Forget, T. Lepage-Lévesque, S. Verret, A. Foley Abstract We present a quantum-classical hybrid implementation of the Liouvillian recursion method to compute many-body Green’s functions using a quantum computer. From an approximate ground state preparation circuit, this algorithm produces the local ( r = r ′ r=r^{\prime} ) and inter-site ( r ≠ r ′ r\neq r^{\prime} ) Green’s functions G r r ′ ( ω ) G_{rr^{\prime}}(\omega) by measuring observables generated recursively. We demonstrate the approach on a superconducting quantum processor for the open-boundary four-site Hubbard model. We then use the computed Green’s functions as input to the Galitskii-Migdal formula to produce better ground state energy estimation than the expectation value of the Hamiltonian for the approximate circuit. Empirical results indicate exponential convergence in the number of iterations, yielding a computational complexity polynomial in the Green’s-function accuracy, as measured with the Wasserstein distance. Our results also indicate significant robustness to noise and to inaccuracies of the ground state preparation, providing evidence that Liouvillian recursion is well adapted to the constraints of near-term quantum computing. One of the most promising applications of quantum computing is the simulation of strongly correlated electron systems 21 ; 29 ; 2 ; 4 . Candidates for early quantum advantage include small molecules and cluster impurity models, which underpin dynamical mean-field theory (DMFT) approaches to correlated materials. While determining the ground state of such systems is of intrinsic interest, extracting physical observables and leveraging DMFT typically requires response functions like the single-particle Green’s function. Many quantum algorithms have been proposed to compute the Green’s functions. To date, all such algorithms require some form of ground state preparation, and most of them rely on additional costly operations, such as multi-qubits control operations 8 ; 9 ; 3 ; 30 ; 5 or time evolution 17 ; 13 ; 20 ; 23 ; 28 . This leads to quantum circuits of depth prohibitive for near-term quantum computers. A few algorithms, however, only require query access to the ground state, circumventing the need for additional operations 12 ; 26 ; 14 ; 16 . They can be divided in two different approaches. The first projects the Hamiltonian in a subspace and builds the Green’s function in the Lehmann representation through subspace diagonalization. Examples include the works of 12 , who use non-orthogonal basis and highlight opportunities for noise filtering in building a low-excitation subspace, and 26 , who showcase how to use such method as an impurity solver in cluster extensions to DMFT. The second approach uses recursion relations or moment expansions to build the Green’s function in a continued-fraction representation. These approaches are related to the Lanczos method 7 , and more systematic than other subspace methods. They fit the broader framework of the recursion method 33 , with various formulations in the Hamiltonian formalism and the Liouvillian formalism. Examples include the works of 14 , who showcase a DMFT impurity solver where the Green’s function is built by measuring moments of the Hamiltonian, and 16 , who use the projection operator formalism to compute Green’s functions and prove the exponential convergence of the Liouvillian recursion with respect to the number of iterations. They also prove stability to noise and ground state imperfections; a particularly important feature when considering that ground state preparation can be hard even for a quantum computer 18 . In this paper, we present a new quantum algorithm based on the Liouvillian recursion approach, and we show that, given query access to an approximate ground state, we can compute accurate Green’s functions and improve this approximate ground state’s energy estimation. The paper is divided as follows. First, we explain how to obtain a continued fraction representation for the one-particle Green’s functions in the frequency domain by measuring a sequence of observables generated by Liouvillian recursion. Next, we demonstrate the viability of the approach on the IBM Quebec superconducting quantum processor, obtaining accurate Green’s functions for the four-site open boundary Fermi-Hubbard model. We then show how to estimate the true ground state energy from the calculated Green’s functions and we find that this estimate is more accurate than the energy expectation value of the approximately prepared ground state. Finally, we investigate the computational cost of our algorithm. We show that although the number of Pauli operators generated by the recursion grows exponentially with the number of iterations, the exponential convergence of the recursion method compensates this cost. This leads to an effective complexity that is polynomial in the Green’s function accuracy, where the accuracy is measured via the Wasserstein distance. Throughout this work, we highlight evidence of the algorithm’s robustness to noise and to inaccuracies in the ground state, suggesting a natural affinity between the Liouvilian recursion method and near-term quantum computation. Our algorithm is similar in spirit to that of 16 , but differs in two key aspects. Firstly, we obtain the full retarded Green’s function, rather than separate hole and electron contributions. Secondly, we obtain the off-diagonal elements using a polynomial hybrid formulation instead of with a continued fraction of matrices. Our approach was detailed by 10 for the classical case; here we present an implementation adapted to quantum computers. To begin, we consider the retarded Green’s function, G_{rr^{\prime}}(t)=-i\langle\{\mathbf{c}_{r}^{\phantom{\dagger}}(t),\mathbf{c}_{r^{\prime}}^{\dagger}\}\rangle\theta(t), (1) where \mathbf{c}_{r}(t) and \mathbf{c}_{r}^{\dagger}(t) are respectively the electronic annihilation and creation operators at time t and state r , a composite index that includes site i and spin \sigma , and \theta(t) is the Heaviside step function. The average \langle\cdots\rangle is with respect to the zero temperature ground state. We obtain the diagonal Green’s functions ( r^{\prime}=r ) as a continued fraction representation in the frequency \omega domain, G^{(k)}_{rr}(\omega)=\cfrac{1}{\omega-\alpha_{0,r}-\cfrac{\beta_{1,r}^{2}}{\omega-\alpha_{1,r}-\raisebox{-6.00006pt}{$\ddots\raisebox{-3.00003pt}{$-\dfrac{\beta_{k,r}^{2}}{\omega-\alpha_{k,r}}$}$}}}, (2) where the coefficients \alpha_{k,r} and \beta_{k,r} are computed recursively up to iteration k , using the following set of equations: \displaystyle\beta_{k+1,r}\mathbf{f}_{k+1,r} \displaystyle=[\mathbf{H},\mathbf{f}_{k,r}]-\alpha_{k,r}\mathbf{f}_{k,r}-\beta_{k,r}\mathbf{f}_{k-1,r}, (3) \displaystyle\beta^{2}_{k,r} \displaystyle=\langle\{\beta_{k,r}\mathbf{f}_{k,r}^{\dagger},\beta_{k,r}\mathbf{f}_{k,r}\}\rangle, (4) \displaystyle\alpha_{k,r} \displaystyle=\langle\{\mathbf{f}_{k,r}^{\dagger},[\mathbf{H},\mathbf{f}_{k,r}]\}\rangle, (5) \displaystyle\mathbf{f}_{0,r} \displaystyle=\mathbf{c}_{r}, (6) \displaystyle\mathbf{f}_{-1,r} \displaystyle=0. (7) This algorithm is equivalent to the Lanczos algorithm 7 , but as applied for a Hilbert space of operators instead of a Hilbert space of state vectors . The \mathbf{f}_{k,r} operators form an orthonormal basis set, and the expectation value of the anticommutator \langle\{\mathbf{A},\mathbf{B}\}\rangle is the inner product. In the above, we can see that, starting from \mathbf{c}_{r} , we obtain new directions in this operator Hilbert space through the action of the Liouvillian \mathcal{L}(\mathbf{f})=[\mathbf{H},\mathbf{f}] , where \mathbf{H} is the Hamiltonian. To obtain the off-diagonal elements of the Green’s functions ( r^{\prime}\neq r ), we use the diagonal Green’s functions in an orthonormal polynomial decomposition: G^{(k)}_{r^{\prime}r}(\omega)=\sum_{k^{\prime}=0}^{k}m_{k^{\prime},r^{\prime}r}\left(Q_{k^{\prime},r}(\omega)+L_{k^{\prime},r}(\omega)G^{(k)}_{rr}(\omega)\right). (8) The decomposition coefficients m_{k,r^{\prime}r} are obtained as m_{k,r^{\prime}r}=\langle\{\mathbf{f}_{k,r},\mathbf{c}^{\dagger}_{r^{\prime}}\}\rangle, (9) and Q_{k,r}(\omega) and L_{k,r}(\omega) are polynomials in \omega , both produced by the same three-term recursion relation, \beta_{k+1,r}X_{k+1,r}(\omega)=(\omega-\alpha_{k,r})X_{k,r}-\beta_{i}X_{k-1,r} (10) where X_{k,r}\in\{L_{k,r},Q_{k,r}\} and with initial conditions: \displaystyle L_{-1,r}(\omega) \displaystyle=0, (11) \displaystyle L_{0,r}(\omega) \displaystyle=1, (12) \displaystyle Q_{0,r}(\omega) \displaystyle=0, (13) \displaystyle Q_{1,r}(\omega) \displaystyle=-1/\beta_{1,r} (14) From this, we see that the method only requires one additional expectation value ( 9 ) per iteration, for each off-diagonal element desired. We use the quantum computer to obtain expectation values ( 4 ), ( 5 ), and ( 9 ) via repeated statistical sampling of observables in