IBM Quantum Watson Team Simulates Molecular Reactions Using 124 Qubits

IBM Watson Team Pushes Molecular Simulations to 124 Qubits

A new algorithm from IBM’s Watson Research Center, variable‑geometry sample‑based quantum diagonalization (VGSQD), demonstrates that near‑term quantum processors can go beyond simple energy calculations. By reusing quantum samples from a reference molecular geometry when exploring nearby configurations, VGSQD reduces the number of full circuit executions required for each geometry change. The approach was validated on IBM’s Heron R2 processor (accessed via the ibm‑quebec node) using circuits that span up to 124 qubits.

---

How VGSQD Works

Traditional quantum chemistry workflows treat every new molecular geometry as a fresh problem, demanding a complete quantum circuit run for each configuration. VGSQD instead treats the set of geometries as a sample space: a single, high‑quality quantum state prepared for a reference structure is reused as a basis for estimating properties of nearby structures. The algorithm:

1. Prepares a quantum state that encodes the electronic Hamiltonian of a reference geometry.
2. Samples this state to obtain expectation values and overlaps.
3. Applies small geometric perturbations classically, then re‑weights the sampled data to approximate the Hamiltonian of the perturbed geometry.
4. Diagonalizes the resulting effective Hamiltonian to extract energies, optimized structures, and reaction pathways.

By avoiding a full re‑initialisation of the quantum circuit for each geometry, VGSQD cuts the total gate count and measurement overhead, which are the primary sources of noise on current hardware.

---

Demonstrated Results on Small Molecules

The research team applied VGSQD to three benchmark systems:

• Beryllium dimer (Be₂) – a weakly bound system that challenges classical methods. Using a cc‑pVDZ basis, the algorithm reproduced experimental bond lengths within the expected uncertainty and matched coupled‑cluster singles‑doubles‑triples (CCSD(T)) reference energies.
• Water (H₂O) – geometry optimisation with a larger cc‑pVTZ basis set yielded bond lengths accurate to 0.002 Å and bond angles within 0.5° of high‑level classical calculations.
• Hydrogen exchange reaction (H + H₂ → H₂ + H) – the minimum‑energy path was traced, and barrier heights were within 0.2 kcal mol⁻¹ of full configuration interaction (FCI) results, a level of precision previously unattainable on noisy intermediate‑scale quantum (NISQ) devices.

All three cases required circuits that scaled to 124 qubits, showcasing the algorithm’s ability to handle relatively large active spaces on existing hardware.

---

Practical Implications and Open Questions

VGSQD’s sample‑reuse strategy offers a concrete pathway toward chemistry‑relevant quantum simulations without demanding fault‑tolerant machines. Potential applications include:

• Drug discovery – rapid screening of conformational ensembles for binding‑site analysis.
• Catalyst design – mapping reaction pathways for transition‑state optimisation.
• Materials science – exploring structural relaxations in low‑dimensional systems.

However, several challenges remain:

• Scalability – the current demonstrations involve modest molecules; extending the method to larger, strongly correlated systems will test both qubit count and error‑mitigation techniques.
• Quantified speed‑up – while VGSQD reduces the number of circuit executions, a rigorous comparison against repeated subspace diagonalization or classical sampling methods is needed to establish concrete runtime advantages.
• Error mitigation – the algorithm’s reliance on accurate sample re‑weighting makes it sensitive to measurement noise; integrating advanced mitigation strategies will be essential for broader adoption.

---

Looking Ahead

The Watson team’s work illustrates that algorithmic innovation can extract more value from today’s quantum hardware. By focusing on data efficiency rather than raw qubit numbers alone, VGSQD opens a new design space for quantum chemistry workflows. Future research will likely explore hybrid schemes that combine VGSQD with variational quantum eigensolvers (VQEs) or quantum subspace expansion techniques, aiming to push the frontier toward medium‑size organic molecules and realistic catalytic cycles.

For practitioners, the key takeaway is that sample reuse can be a powerful lever: even with noisy devices, thoughtful algorithmic design can deliver chemically accurate predictions that were previously out of reach. As quantum processors continue to grow in qubit count and coherence, methods like VGSQD will be essential for turning raw hardware capability into actionable scientific insight.