Abstract: Numerous reports claim that quantum advantage, which should emerge as a direct consequence of the advent of quantum computers, will herald a new era of chemical research because it will enable scientists to perform the kinds of quantum chemical simulations that have not been possible before. Such simulations on quantum computers, promising a significantly greater accuracy and speed, are projected to exert a great impact on the way we can probe reality, predict the outcomes of chemical experiments, and even drive design of drugs, catalysts, and materials. In this work we review the current status of quantum hardware and algorithm theory and examine whether such popular claims about quantum advantage are really going to be transformative. We go over subtle complications of quantum chemical research that tend to be overlooked in discussions involving quantum computers. We estimate quantum computer resources that will be required for performing calculations on quantum computers with chemical accuracy for several types of molecules. In particular, we directly compare the resources and timings associated with classical and quantum computers for the molecules H$_2$ for increasing basis set sizes, and Cr$_2$ for a variety of complete active spaces (CAS) within the scope of the CASCI and CASSCF methods. The results obtained for the chromium dimer enable us to estimate the size of the active space at which computations of non-dynamic correlation on a quantum computer should take less time than analogous computations on a classical computer. Using this result, we speculate on the types of chemical applications for which the use of quantum computers would be both beneficial and relevant to industrial applications in the short term.
Introduction: The Reality of Quantum Advantage
This paper, published in late 2020, serves as a sobering but constructive "reality check" for the field of quantum computational chemistry. While the popular narrative often suggests that quantum computers (QCs) will soon revolutionize drug discovery and materials science, the authors—a team from Qu & Co and Schrödinger Inc.—argue that the path to true industrial relevance is narrower and more technically demanding than many realize.
The paper identifies the "sweet spot" for quantum advantage in chemistry: medium-sized inorganic catalysts where classical methods fail to handle "non-dynamic correlation." It argues that current quantum experiments often confuse matching a model (precision) with matching reality (accuracy) and estimates that a real advantage over classical supercomputers requires fault-tolerant systems handling active spaces of roughly 20–30 orbitals.
Defining the Accuracy Gap
The authors address a significant "disconnect" between the quantum computing and quantum chemistry communities. In the quantum computing world, "quantum advantage" is often claimed when a QC performs a calculation that is hard for a classical computer. However, in chemistry, a calculation is only useful if it achieves chemical accuracy (an error of less than 1 kcal/mol compared to experimental results).
Many "breakthrough" quantum experiments have achieved high precision (matching a simplified classical model) but low accuracy (the model itself was too simple to represent reality). This gap is visualized by comparing the number of spin-orbitals actually mapped to qubits in literature versus the number required to reach chemical accuracy for those same molecules.
Methods and Benchmarking
The researchers performed a rigorous resource estimation for two primary quantum algorithms: Trotterization and Qubitization (specifically the "sparse" variant). They compared these against state-of-the-art classical methods, including CCSD(T) (the "gold standard" for many chemical systems) and CASSCF/CASCI (used for "strongly correlated" systems).
The study used the chromium dimer ($Cr_2$) as a primary benchmark because its "sextuple bond" makes it notoriously difficult for classical computers to simulate accurately. They calculated the "wall-clock time" and physical qubit requirements needed for a QC to beat a classical high-performance computing (HPC) cluster.
Meta-Analysis of Quantum Chemistry Experiments
The paper evaluates specific molecular "reference problems" and provides a comprehensive meta-analysis of quantum chemistry experiments on QCs from 2013 to 2020, highlighting the platforms and methods used to date.
Reference problems include:
- Small molecules ($H_2$, $LiH$, $H_2O$): Used to show that classical computers are already so fast and accurate here that QCs offer no practical advantage.
- Chromium Dimer ($Cr_2$): The core benchmark for scaling.
- FeMo-co (Nitrogenase cofactor): A famous "holy grail" problem for quantum computing. The authors argue it is much harder than previously estimated, likely requiring over 100 orbitals for chemical accuracy.
Key Insights and Results
- The Moving Target: Classical algorithms like DLPNO-CCSD(T) can now handle systems with hundreds of atoms. QCs should target "strongly correlated" systems where these methods fail rather than competing on simple molecules.
- The "Sweet Spot": The authors argue that the best targets for early quantum advantage are homogeneous catalysts (e.g., ruthenium-based catalysts for $CO_2$ conversion) and biomimetics. These have active sites complex enough to break classical solvers but are small enough to potentially fit on early fault-tolerant QCs.
- Crossover Point: For the chromium dimer, the transition point where a QC (using surface code error correction) should outperform a classical computer occurs at an active space size of approximately $19 \le N \le 34$ orbitals.
- Algorithm Efficiency: Sparse Qubitization is significantly more efficient than Trotterization, reducing the estimated spacetime complexity by about four orders of magnitude.
Limitations and Future Outlook
Even for the "sweet spot" molecules, we likely need millions of physical qubits to implement the necessary error correction (surface codes). A neglected problem is the "data bottleneck": loading two-electron integrals for a 1,000-orbital system can reach ~1 Terabyte, creating a massive I/O bottleneck between classical and quantum components.
Furthermore, choosing which orbitals to include in the "active space" is currently a manual, expert-driven process. The authors suggest that automated, machine-learning-driven orbital selection will be crucial for industrial scaling. While QCs excel at "non-dynamic" correlation, "dynamic" correlation still requires huge basis sets, suggesting that hybrid quantum-classical methods may be the only viable path forward in the medium term.
Summary updated