Rethinking Parity Check Enhanced Symmetry-Preserving Ansatz
Authors: Ge Yan, Mengfei Ran, Ruocheng Wang, Kaisen Pan, Junchi Yan
NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our extensive experimental results on both simulators and superconducting quantum processors verify that the combination of HWP ansatz with parity check is among the most promising candidates to show quantum advantages in the NISQ era to solve more realistic problems. |
| Researcher Affiliation | Academia | Ge Yan, Mengfei Ran, Ruocheng Wang, Kaiseng Pan, Junchi Yan Dept. of CSE & School of AI & Zhiyuan College, Shanghai Jiao Tong University |
| Pseudocode | No | The paper includes circuit diagrams (Figure 1, Figure 2(c)) and describes procedures, but it does not present any formal pseudocode or algorithm blocks. |
| Open Source Code | No | We will make our code available after the publication. |
| Open Datasets | Yes | Dataset: We select three well-studied molecules, i.e. Hydrogen (H2), Lithium Hydride (Li H), and Water (H2O). The molecular Hamiltonian is obtained from the Python package Open Fermion [58]. The computational basis for all the molecules is STO-3G with Jordan-Wigner transformation. Dataset: We generated a dataset comprising random instances of QAPs, with 100 instances for each size m = {3, 4, 5, 6}. Each instance includes a m m distance matrix D and a flow matrix F, with elements Dij = Dji and Fij = Fji, drawn from a uniform distribution [0, 1]. |
| Dataset Splits | No | The paper mentions generating datasets and using them for experiments but does not specify distinct training, validation, and testing splits or a cross-validation setup. |
| Hardware Specification | Yes | All the numerical simulations are performed on a machine with 190GB memory, one physical CPU with 32 cores AMD Ryzen Threadripper 3970X CPU, and 5 GPUs (Nvidia Ge Force RTX 3090). ... In the experiment, we utilized a 12-qubit Superconducting Quantum Processor with an intercoupled topology as shown in the Fig. 4.Tab. 5 presents the performance information regarding Single-Qubit Operations, while Tab. 6 presents the fidelity of the controlled-Z gate between two qubits. |
| Software Dependencies | Yes | To simulate the circuit with noise on Qiskit, we utilize the Aer-simulator from Qiskit based on the density matrix which is time-consuming. Therefore, we freeze some of the inactive orbitals to reduce the problem size of the above molecules. The detailed molecular information is listed in Tab. 1 Baselines: To show the efficiency of the HWP ansatz and the superiority of combining HWP ansatz with parity check, we select the well-studied UCC ansatz as our baselines. To better illustrate that universal HWP ansatz is able to solve the state preparation problem without any truncation, we include single excitation (UCCS), double excitation (UCCSD), and triple excitation (UCCSDT). All the ansatze are implemented with Qiskit-Nature [24] and initialized with Hartree-Fock state. The optimizer is SLSQP. ... [24] The Qiskit Nature developers and contributors. Qiskit nature 0.7.2, 2023. |
| Experiment Setup | Yes | The optimizer is SLSQP. QAP on simulator: The layer of XYmixer-FC, NBS-FC, HEA, and QAOA are set to 2 m and XYmixer-NN, NBS-NN, XYmixer-hard, and NBS-hard have the same number of parameters as NBS-FC. Learning rate is set to 0.05 and the number of iterations is 1000. QAP on Superconducting Quantum Processors: Due to limited coherence time and noise, the parameter setting is a bit different from that on the simulator. We are only able to apply one layer of NBS-FC considering the number of CNOTs, so the rest of the methods are set with the same amount of parameters as one layer of NBS-FC. As QAOA has a parameter-sharing strategy, we set its number of layers as 2. For NBS-Hard, we set T = 2. Each circuit is measured with 20K shots. Here we provide the noise levels we used in the experiments. We set the depolarizing error for single qubit as 1.6 10 3 and for two-qubit gates as 6.4 10 3. The bit-flip and phase-flip error for each gate are set as 5 10 3. We also set a 1 10 2 readout error for obtaining |1 given |0 , and 5 10 2 vise versa. To simulate the computing way on quantum computers, we set the number of shots per Pauli string in the measurement stage at 5000, which makes at least 500,000 shots in total for Li H and H2O since they all have more than 100 Pauli strings. |