Abstract:Efficiently learning an unknown Hamiltonian given access to its dynamics is a problem of interest for quantum metrology, many-body physics and machine learning. A fundamental question is whether learning can be performed at the Heisenberg limit, where the Hamiltonian evolution time scales inversely with the error, $\varepsilon$, in the reconstructed parameters. The Heisenberg limit has previously been shown to be achievable for certain classes of qubit and bosonic Hamiltonians. Most recently, a Heisenberg-limited learning algorithm was proposed for a simplified class of fermionic Hubbard Hamiltonians restricted to real hopping amplitudes and zero chemical potential at all sites, along with on-site interactions. In this work, we provide an algorithm to learn a more general class of fermionic Hubbard Hamiltonians at the Heisenberg limit, allowing complex hopping amplitudes and nonzero chemical potentials in addition to the on-site interactions, thereby including several models of physical interest. The required evolution time across all experiments in our protocol is $\mathcal{O}(1/\varepsilon)$ and the number of experiments required to learn all the Hamiltonian parameters is $\mathcal{O}(\text{polylog}(1/\varepsilon))$, independent of system size as long as each fermionic mode interacts with $\mathcal{O}(1)$ other modes. Unlike prior algorithms for bosonic and fermionic Hamiltonians, to obey fermionic parity superselection constraints in our more general setting, our protocol utilizes $\mathcal{O}(N)$ ancillary fermionic modes, where $N$ is the system size. Each experiment involves preparing fermionic Gaussian states, interleaving time evolution with fermionic linear optics unitaries, and performing local occupation number measurements on the fermionic modes. The protocol is robust to a constant amount of state preparation and measurement error.
What problem does this paper attempt to address?
This paper aims to solve the problem of efficiently learning unknown fermionic Hamiltonians under the Heisenberg limit. Specifically, the researchers proposed an algorithm that can learn a wider range of fermionic Hubbard model Hamiltonians under the Heisenberg limit. These models allow for complex transition amplitudes and non - zero chemical potentials, as well as on - site interactions. This extends the previous learning algorithms for the simplified class of fermionic Hubbard models that were limited to real - valued transition amplitudes and zero chemical potentials.
### Research Background and Objectives
In fields such as quantum metrology, many - body physics, and machine learning, efficiently learning unknown Hamiltonians given access to dynamics is an important problem. The Heisenberg limit means that the Hamiltonian evolution time is inversely proportional to the reconstruction parameter error \( \varepsilon \), that is, \( t\sim O(1 / \varepsilon) \). Previous studies have proven that for certain types of spin and bosonic Hamiltonians, the Heisenberg limit can be reached. Recently, a Heisenberg - limit learning algorithm for a simplified class of fermionic Hubbard models was proposed, but it was limited to the case of real - valued transition amplitudes and zero chemical potentials.
### Main Contributions
The main contribution of this paper is to provide a new algorithm that can learn more general fermionic Hubbard models, including those with complex transition amplitudes and non - zero chemical potentials. Specifically:
1. **Algorithm Overview**:
- **Experimental Design**: This algorithm is implemented through a series of experiments. Each experiment includes preparing a two - mode fermionic Gaussian state, inserting time evolution and fermionic linear optics (FLO) unitary transformations, and performing local occupation number measurements.
- **Time Complexity**: The total evolution time of all experiments is \( O(1 / \varepsilon) \), and the number of experiments is \( O(\text{polylog}(1 / \varepsilon)) \).
- **Robustness**: This protocol is robust against constant errors in state preparation and measurement.
2. **Technical Details**:
- **Hamiltonian Form**: The form of the fermionic Hamiltonian considered is as follows:
\[
H=\sum_{\langle i,j\rangle}\sum_{\sigma\in\{\uparrow,\downarrow\}}h_{ij\sigma}a^{\dagger}_{i\sigma}a_{j\sigma}+\sum_{\sigma\in\{\uparrow,\downarrow\}}\omega_{i\sigma}n_{i\sigma}+\sum_{i}\xi_{i}n_{i\uparrow}n_{i\downarrow}
\]
where \( h_{ij\sigma} \) is the transition amplitude, \( \omega_{i\sigma} \) is the chemical potential, and \( \xi_{i} \) is the on - site interaction strength.
- **Discrete Quantum Control**: To abide by the fermionic conservation constraint, this algorithm uses \( O(N) \) auxiliary fermionic modes, where \( N \) is the system size. Each experiment involves preparing a fermionic Gaussian state, inserting time evolution and FLO unitary transformations, and performing local occupation number measurements.
- **Robust Phase Estimation**: The robust phase estimation (RPE) algorithm is used to estimate the Hamiltonian parameters and achieve Heisenberg - limit precision.
### Experimental Steps
1. **Single - Site Hamiltonian Learning**:
- Use the RPE algorithm to gradually learn the parameters \( \omega_1, \omega_2, \xi_{12} \) of the single - site Hamiltonian.
- Obtain the required measurement statistics by introducing auxiliary modes and selecting appropriate initial states and measurement bases.
2. **Two - Site Hamiltonian Learning**:
- First, learn the coefficients \( \omega \) and \( \xi \) of the number - conservation terms.
- Use a random unitary transformation to decouple the two spatial sites, and then learn these coefficients through RPE.
- Then learn the coefficient \( h \) of the transition term.