Quantum Simulation of Molecules Scales with Compressed State Representation.

A new hybrid quantum-classical method, Lossy-QSCI, efficiently simulates molecular systems. It compresses quantum state representations using fermionic number conservation – reducing qubit requirements to O(N log M) – and employs a neural network to decode compressed states. Demonstrations on C₂ and LiH achieved chemical accuracy with reduced computational resources.

The accurate modelling of molecular systems remains a significant computational challenge, demanding resources that quickly exceed the capabilities of classical computers. Researchers are now exploring quantum computing as a potential solution, but current quantum devices are limited in qubit number and susceptible to noise. A new approach, detailed in a paper by Yu-Cheng Chen (Hon Hai Research Institute), Ronin Wu (QunaSys Europe), M. H. Cheng (Blackett Laboratory, Imperial College London & Fraunhofer Institute for Industrial Mathematics), and Min-Hsiu Hsieh (Hon Hai Research Institute), presents a method for compressing the quantum states required for these simulations. Their work, entitled ‘Neural Network Assisted Fermionic Compression Encoding: A Lossy-QSCI Framework for Scalable Quantum Chemistry Simulations’, introduces a ‘Lossy-QSCI’ framework that combines a chemically-inspired compression technique with a neural network to efficiently decode the resulting compressed states, potentially enabling more complex molecular simulations with fewer qubits.

Lossy Selected Configuration Interaction: A Novel Approach to Quantum Molecular Simulation

Quantum simulation holds considerable promise for advancing fields such as chemistry and materials science, yet current quantum hardware presents significant limitations. The number of qubits available and their susceptibility to noise restrict the complexity of systems that can be accurately modelled. Recent research details a new framework, Lossy Selected Configuration Interaction (Lossy-SCI), designed to mitigate these challenges.

The primary obstacle lies in the exponential scaling of qubit requirements with system size. Accurately representing the electronic structure of even moderately sized molecules demands a substantial number of qubits, exceeding the capacity of near-term devices. Lossy-SCI addresses this by reducing the computational burden without substantial loss of accuracy.

The method builds upon Selected Configuration Interaction (SCI), a technique that approximates solutions to the molecular Schrödinger equation by focusing on the most chemically relevant electronic configurations. This reduces the computational space compared to a full configuration interaction calculation, but still requires a significant number of qubits. Lossy-SCI further refines this approach through a novel compression and reconstruction strategy.

Central to this strategy is Chemical-RLE (Random Linear Encoder). This encoder exploits the principle of fermionic number conservation – a fundamental property of electrons dictating that the number of electrons with spin up must equal the number with spin down – to compress the quantum state. This compression reduces the number of qubits required to represent the state to O(N log M), where N and M are parameters related to the size of the molecular system. This represents a significant reduction in qubit demand compared to traditional methods.

However, compressing the quantum state introduces information loss. To counteract this, the researchers developed a Neural Network-assisted Fermionic Expectation Decoder (NN-FED). This decoder efficiently reconstructs the necessary information from the compressed state, enabling accurate calculation of molecular properties despite the initial compression. The NN-FED overcomes challenges associated with measuring the full wavefunction, focusing instead on expectation values of key operators.

Testing on the diatomic molecules C₂ and LiH demonstrated the framework’s efficacy. Results indicate that Lossy-SCI achieves ‘chemical accuracy’ – a benchmark requiring energy calculations to be accurate to within 1 kcal/mol (approximately 4.18 kJ/mol) – while significantly reducing qubit requirements.

The framework’s design prioritises scalability, suggesting potential for simulating larger and more complex molecular systems than currently feasible. By combining efficient compression with intelligent reconstruction, Lossy-SCI offers a promising pathway towards overcoming the limitations of current quantum hardware and unlocking the full potential of quantum molecular simulations.