Quantum simulation of a qubit with non-Hermitian Hamiltonian
Anastashia Jebraeilli, Michael R. Geller · 2025
Read Quantum simulation of a qubit with non-Hermitian Hamiltonian by Anastashia Jebraeilli, Michael R. Geller (2025) 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-2502-13910.txt.
Introduction
Anastashia Jebraeilli and Michael R Geller simulate a qubit with unusual physics in a 2025 paper. The models at issue use Hamiltonians that are non-Hermitian, meaning energy can leak, as in open systems and certain symmetry studies. Simulating them on quantum hardware has been held back by methods whose success chance shrinks fast.
The authors instead use a fixed-depth variational circuit, a short tunable quantum program, to run the simulation. On IBM superconducting qubits they pushed deep into the regime where the key symmetry is broken, around a special point where states merge. They argue such circuits, joined with machine learning, can widen what near-term quantum machines can explore.
Catalog record
| CREATOR | Anastashia Jebraeilli, Michael R. Geller |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 23 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2502-13910
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Quantum simulation of a qubit with non-Hermitian Hamiltonian Anastashia Jebraeilli, Michael R. Geller Abstract Modeling non-Hermitian Hamiltonians is increasingly important in classical and quantum domains, especially when studying open systems, P T PT symmetry, and resonances. However, the quantum simulation of these models has been limited by the extensive resources necessary in iterative methods with exponentially small postselection success probability. Here we employ a fixed-depth variational circuit to circumvent these limitations, enabling simulation deep into the P T PT -broken regime surrounding an exceptional point. Quantum simulations are carried out using IBM superconducting qubits. The results underscore the potential for variational quantum circuits and machine learning to push the boundaries of quantum simulation, offering new methods for exploring quantum phenomena with near-term intermediate-scale quantum technology. I Introduction Open quantum systems are not restricted to Hermitian Hamiltonian dynamics. A diverse array of systems—including superconducting qubits, optical waveguides, oscillators, polarized photons, ultracold Fermi gases, and atomic or molecular configurations—exhibit behavior that can be described by non-Hermitian Hamiltonians [ 1 , 2 , 3 , 4 , 5 , 6 , 7 , 8 , 9 , 10 , 11 , 12 , 13 ] . Despite their prevalence, non-Hermitian Hamiltonians have yet to be simulated on near-term intermediate-scale quantum (NISQ) devices. The simulation of non-Hermitian Hamiltonians is not only interesting from a quantum computing perspective but also can help unlock a deeper understanding of quantum mechanics itself [ 14 , 15 , 16 , 17 , 18 , 19 , 20 ] . Figure 1: Eigenvalues of H_{\rm eff} for |\omega|>\Gamma (green points) and |\omega|<\Gamma (blue points). At the exceptional points \omega=\pm\Gamma , the eigenvalues vanish (red point). Non-Hermicity results in the inclusion of phenomena that are absent in closed Hermitian systems. Exceptional points (EPs), where the eigenvectors of a system coalesce, are examples of such non-Hermitian behavior, along with PT symmetry breaking. Recently, a method using postselection has been proposed [ 21 ] to simulate the non-Hermitian Hamiltonian H_{\text{eff}}=\frac{\omega}{2}\sigma_{x}+\frac{i\Gamma}{2}\sigma_{z}=\begin{pmatrix}\frac{i\Gamma}{2}&\frac{\omega}{2}\\ \frac{\omega}{2}&-\frac{i\Gamma}{2}\end{pmatrix},\ \ \ {\rm EPs}\!:\omega=\pm\Gamma, (1) where \sigma_{x} and \sigma_{z} are Pauli matrices. Here \Gamma\geq 0 , while \omega\in{\mathbb{R}} is arbitrary. The eigenvalues and eigenvectors of H_{\text{eff}} respectively are \displaystyle\lambda_{\pm} \displaystyle=\pm\frac{\sqrt{\omega^{2}-\Gamma^{2}}}{2}, (2) \displaystyle|\phi_{\pm}\rangle \displaystyle=c_{\pm}\begin{pmatrix}\omega\\ -i\Gamma\pm\sqrt{\omega^{2}-\Gamma^{2}}\end{pmatrix}\!, (3) where c_{\pm}=(\omega^{2}+|\!-i\Gamma\pm\sqrt{\omega^{2}-\Gamma^{2}}\,|^{2})^{-\frac{1}{2}} is real and positive. In the special (and trivial) case where \Gamma=0 , standard Hamiltonian simulation is recovered; henceforth, we will assume \Gamma>0. If |\omega|>\Gamma , both eigenvalues are real, and if |\omega|<\Gamma the eigenvalues become a purely imaginary complex conjugate pair, as illustrated in Fig. 1 . An exceptional point (EP) in parameter space occurs if the Hamiltonian becomes defective (lacking two linearly independent eigenvectors). The model ( 1 ) has two EPs, at \omega=\pm\Gamma . At an EP, the single eigenvector is \displaystyle|\Phi_{\omega=\Gamma}\rangle=\frac{1}{\sqrt{2}}\begin{pmatrix}1\\ -i\end{pmatrix}\ \ {\rm or}\ \ |\Phi_{\omega=-\Gamma}\rangle=\frac{1}{\sqrt{2}}\begin{pmatrix}1\\ i\end{pmatrix}, (4) a pure state on the equator of the Bloch sphere. We see that, in contrast to a Hermitian Hamiltonian, H_{\rm eff} can have complex eigenvalues. A standard framework for understanding this is through PT symmetry, where P is parity (satisfying P{\vec{r}}P^{-1}=-{\vec{r}} ) and T is time reversal (satisfying TzT^{-1}=z^{*} where z\in{\mathbb{C}} ). A non-Hermitian Hamiltonian that commutes with PT is defined as PT -invariant. A PT -invariant Hamiltonian with a real spectrum is classified as PT -symmetric, whereas PT symmetry is broken if the eigenvalues have imaginary parts. To make use of PT symmetry in the qubit model ( 1 ), we need to define a two-dimensional unitary parity operator P . A natural choice is P=\sigma^{x} , which satisfies P\sigma^{z}P^{-1}=-\sigma^{z} and renders the model PT -invariant [ 22 ] . If \omega<-\Gamma or \omega>\Gamma , the model is in a PT -symmetric regime; if -\Gamma<\omega<\Gamma the PT symmetry is broken. The EPs at \omega=\pm\Gamma separate these regimes. Suppose a Hamiltonian H is PT -invariant, and let |\phi\rangle be an eigenfunction with eigenvalue \lambda . Then, from [H,PT]=0 , we see that |\phi^{\prime}\rangle:=PT|\phi\rangle is also an H eigenfunction with eigenvalue \lambda , potentially leading to twofold degeneracy. If |\phi^{\prime}\rangle=e^{i\gamma}|\phi\rangle with \gamma\in{\mathbb{R}} , then |\phi^{\prime}\rangle is not distinct from |\phi\rangle and there is no energy degeneracy resulting from the PT -invariance, but |\phi\rangle is simultaneously an eigenfunction of PT with eigenvalue e^{i\gamma} : \displaystyle H|\phi\rangle=\lambda|\phi\rangle,\ \ PT|\phi\rangle=e^{i\gamma}|\phi\rangle. (5) However, if there is no \gamma\in{\mathbb{R}} such that |\phi^{\prime}\rangle=e^{i\gamma}|\phi\rangle , then |\phi\rangle and PT|\phi\rangle are distinct, and the PT invariance leads to twofold energy degeneracy. This has an important consequence for the PT -invariant model ( 1 ) at an EP. Namely, the eigenvectors ( 4 ) are PT eigenfunctions: \displaystyle PT|\Phi_{\omega=\pm\Gamma}\rangle=|\Phi_{\omega=\pm\Gamma}\rangle. (6) A circuit that simulates H_{\rm eff} is given in Fig. I , as we explain below. By entangling system qubits with ancillary qubits (one per system qubit), the Zhang et al. method simulates the entanglement of a quantum system with its surroundings leading to ( 1 ). Figure 2: Single step of the quantum circuit for simulating H_{\text{eff}} , proposed by Zhang et al. [ 21 ] . Here R_{x}(\varphi)=e^{-\frac{i}{2}\varphi\sigma^{x}} is an x rotation and |\phi|\ll 1 . I.1 Simulating non-Hermicity We assume that the system qubit (qubit 1, the upper qubit in Fig. I ) is initially in a pure state \left|{\psi}\right\rangle=\psi_{0}\left|{0}\right\rangle+\psi_{1}\left|{1}\right\rangle with \|\left|{\psi}\right\rangle\|=1 . We want to implement the map \displaystyle\left|{\psi}\right\rangle\mapsto\frac{A_{t}\left|{\psi}\right\rangle}{\|A_{t}\left|{\psi}\right\rangle\!\|},\ \ A_{t}:=e^{-iH_{\rm eff}t},\ \ t\geq 0, (7) where A_{t} is a linear (but not necessarily unitary) operator. To simulate ( 1 ), let \tau=t/K , where K is a positive integer (number of Trotter steps), and note that \displaystyle A_{t}=e^{-i(-i\Gamma\,e_{11}+\frac{\omega}{2}\sigma^{x})t}=\lim_{K\rightarrow\infty}A_{\tau}^{K}, (8) where e_{11}=|1\rangle\langle 1| . For large K , \displaystyle A_{\tau}=e^{-\Gamma\tau\,e_{11}}R_{x}(\omega\tau). (9) In the approach of Zhang et al. [ 21 ] , the Trotter step A_{\tau} is implemented by the circuit shown in Fig. I with |\phi|\ll 1. Each step of the circuit applies the post-selected channel \displaystyle|{\rm in}\rangle|0\rangle\mapsto|{\rm out}\rangle|0\rangle, (10) with \displaystyle|{\rm out}\rangle=\frac{e^{-\frac{\phi^{2}}{8}\,e_{11}}\,R_{x}(\theta)\,|{\rm in}\rangle}{\|e^{-\frac{\phi^{2}}{8}\,e_{11}}\,R_{x}(\theta)\,|{\rm in}\rangle\|}=\frac{A_{\tau}\,|{\rm in}\rangle}{\|A_{\tau}\,|{\rm in}\rangle\|}, (11) where the second equality assumes circuit parameters \displaystyle\theta=\omega\tau\ {\rm and}\ \phi=\sqrt{8\Gamma\tau}. (12) With these parameters, the circuit simulates ( 9 ) with nonnegative \phi . The post-selection succeeds with probability \displaystyle p=\|e^{-\frac{\phi^{2}}{8}\,e_{11}}\,R_{x}(\theta)\,|{\rm in}\rangle\|^{2}. (13) However this is for one Trotter step. The probability of observing K successful post-selections is P=p_{1}p_{2}\cdots p_{K} , where p_{k} is the success probability for step k . Then P\leq(\max_{k}p_{k})^{K} , which is exponentially small in K (assuming \theta,\phi\neq 0 ). This makes the EPs \omega=\pm\Gamma especially difficult to simulate because both \theta and \phi have to be small, requiring more Trotter steps. These limitations are bypassed using a variational technique explained below. II Methods II.1 Machine learning method Figure 3: Scheme of the process to create the variational PQCs. A density matrix-based simulator simulates the target quantum process (outlined in Section I ) shown within a blue box. The variational PQC ansatz (boxed in red) is similarly simulated with random initial parameters. The outcomes of both circuits are used to calculate the cost function f(r) for the gradient-based optimization (shown in black). The parameters of the variational PQC are updated using the gradient, and the process of recalculating the output of the variational PQC and subsequent optimization is repeated as needed until a target fidelity is reached. New developments in variational quantum circuits and machine learning models for non-unitary operations have provided an opportunity to develop short-depth quantum circuits capable of