Qunova HI-VQE Chemical Accuracy Explained: 80-Qubit Iron-Sulfur Result Matches the Best Classical Calculation

Image: Wikimedia Commons / Ayacop (public domain) — the 4Fe-4S iron-sulfur cluster in the enzyme aconitase; Qunova's test molecule is a [4Fe-4S] cubane cluster of the same type.

TL;DR: Qunova HI-VQE chemical accuracy explained: on 12 November 2026, South Korean firm Qunova said its HI-VQE algorithm reached chemical accuracy on an 80-qubit iron-sulfur calculation, matching the best classical result within 1.1 millihartree. It is a genuine milestone — but the headline number is an extrapolation, not a raw measurement, and it matches classical computing rather than beating it.

Key Highlights

• Qunova announced on 12 November 2026 that its HI-VQE algorithm reached chemical accuracy at 80 qubits on a [4Fe-4S] iron-sulfur cluster — a molecule class classical computers struggle with.
• The extrapolated ground-state energy is −327.2482 ± 0.0007 Hartree, against the best classical reference (DMRG) of −327.2471 Hartree — a gap of 1.1 millihartree, inside the 1.6 millihartree chemical-accuracy threshold.
• The run used IBM's 156-qubit Heron r2 processor (ibm_kobe) at RIKEN in Japan, with RIKEN's Fugaku supercomputer doing the classical half of the calculation.
• At iteration 2, HI-VQE already beat the previous best quantum result for this molecule while using about 300 times fewer Fugaku node-hours (2,518).
• The honest caveat: the underlying paper reports a roughly 64 millihartree gap to DMRG for the raw variational result — the chemical-accuracy claim rests on an extrapolation, and the result matches the best classical method rather than beating it.

What happened?

On 12 November 2026, Qunova Computing — a quantum software company based in Daejeon, South Korea — announced that its hybrid quantum-classical algorithm, HI-VQE, had reached chemical accuracy on an 80-qubit simulation of an iron-sulfur cluster. The announcement, carried as a press release, says HI-VQE is the first quantum algorithm to reach that standard at this scale, on a molecule that is one of the hardest test cases in chemistry.

The calculation was a three-way effort. The quantum half ran on ibm_kobe, IBM's 156-qubit Heron r2 processor installed at RIKEN in Kobe, Japan. The classical half ran on Fugaku, RIKEN's flagship supercomputer. The work was supported by Japan's NEDO agency, and the research team spans Qunova and RIKEN. IBM Research director Jay Gambetta called it an "impressive 'quantum plus HPC' workflow" that showed "promising efficiency compared with selected-CI approaches" — the classical methods it is measured against.

The result is built on a research paper, "Extending the Handover-Iterative VQE to Challenging Strongly Correlated Systems: N₂ and Fe-S Cluster" (arXiv:2601.06935), first posted in January 2026 and revised in September. Today's news is the announcement of the 80-qubit milestone itself — including the extrapolated energy figure, which does not appear in the preprint. That distinction matters, and this post keeps it visible throughout.

Qunova HI-VQE chemical accuracy explained: what the numbers actually say

First, the vocabulary. "Chemical accuracy" means a calculated energy within 1.6 millihartree (mHa) of the true value — roughly 1 kilocalorie per mole, the precision chemists need before a calculation is useful for real design work. A Hartree is the atomic unit of energy; a millihartree is a thousandth of one. Getting within 1.6 mHa on a strongly correlated molecule is hard, because the electrons in these systems interact in ways that defeat the usual approximations.

Here is the result, as announced:

Figure	Value	What it means
HI-VQE extrapolated energy	−327.2482 ± 0.0007 Hartree	Qunova's best estimate of the [4Fe-4S] ground-state energy, after extrapolation
Best classical reference (DMRG)	−327.2471 Hartree	Density Matrix Renormalization Group — the strongest classical method for this molecule
Gap between them	1.1 millihartree	Inside the 1.6 mHa chemical-accuracy threshold
Qubits used	80 physical (72 ansatz + 8 connectors)	72 qubits encode the molecule; 8 are connector qubits for the chip layout
Circuit size	10,786 CZ gates, depth 2,125	A deep circuit for today's noisy hardware — depth is the hard part
Intermediate step (iteration 2)	−326.9821 Ha, 2,518 Fugaku node-hours	Already beat the previous SQD result with ~300× fewer supercomputer hours

The arithmetic behind the headline is simple: −327.2482 minus −327.2471 is 0.0011 Hartree, or 1.1 millihartree — under the 1.6 mHa bar. Qunova's estimate also sits slightly below the classical reference, which chief executive Kevin Rhee flagged in the announcement as a hint that the true ground state may be a little lower than the classical estimate. That is a suggestion, not a proof — and it is exactly the kind of claim that needs independent checking before anyone builds on it.

Scale is the other half of the story. HI-VQE has climbed a long ladder to reach 80 qubits:

System	Qubits	Error vs exact reference	Source
Ammonia (NH₃)	16	0.08 mHa	IBM Quantum docs benchmark
Lithium sulfide (Li₂S)	20	0.60 mHa	IBM Quantum docs benchmark
Nitrogen molecule (N₂)	40	1.20 mHa	IBM Quantum docs benchmark
Iron-porphyrin + NO (FeP-NO)	44	Within 1.6 mHa	IBM Quantum docs benchmark
Nitrogen dissociation (N₂)	56	Sub-millihartree	arXiv:2601.06935
[4Fe-4S] iron-sulfur cluster	80	1.1 mHa (extrapolated)	Qunova announcement, 12 Nov 2026

Image: Wikimedia Commons / Irekia, Eusko Jaurlaritza, CC BY 4.0 — an IBM Quantum System Two installation (illustrative image; the calculation ran on IBM's Heron r2 processor at RIKEN).

How HI-VQE works — the handover trick

HI-VQE stands for Handover Iterative Variational Quantum Eigensolver. The name describes the method: the quantum computer and the classical supercomputer hand work back and forth, each doing only what it is good at.

The problem is the ground-state energy of a molecule — the lowest energy its electrons can have. For a [4Fe-4S] cluster, the exact answer would require tracking between 10⁸ and 10¹⁵ electron configurations, far beyond exact solution. The standard quantum approach, VQE, prepares a trial state on the quantum chip, measures its energy, and lets a classical optimiser tune the circuit. That works at small scale, but the measurement bill grows steeply, hardware noise corrupts the energy estimate, and the optimisation landscape flattens as circuits grow.

HI-VQE reorganises the division of labour. The quantum processor is used only to propose candidate electron configurations — it samples them from a quantum state, which is what quantum hardware does naturally well. Those configurations are handed to the classical computer, which builds the Hamiltonian in the subspace they span and diagonalises it to get the energy. The resulting eigenvector then defines the target for re-optimising the circuit, and the loop repeats. Crucially, no energy is ever estimated from measurements on the hardware, so the method dodges the measurement overhead and noise sensitivity that limit ordinary VQE. Invalid configurations produced by hardware noise are repaired or discarded by a configuration-recovery step, and the subspace is screened each round so only the configurations that matter are kept.

That is why the same algorithm can run as a product, not just a paper: HI-VQE ships as a Qiskit Function on the IBM Quantum platform, and Qunova launched it on the AWS Marketplace for Amazon Braket in January 2026, running on trapped-ion, superconducting and neutral-atom hardware from IonQ, IQM, QuEra, AQT and Rigetti. The original method paper (arXiv:2503.06292) dates to March 2025; today's result is the largest system it has been run on.

The honest caveats

A milestone is only useful if its limits are stated plainly. Four apply here.

First, the headline number is an extrapolation. The preprint reports a residual gap of roughly 64 millihartree between the raw HI-VQE variational result for [4Fe-4S] and the DMRG reference — far outside chemical accuracy. The 1.1 mHa figure comes from extrapolating the sequence of variational results, a standard technique, but an extrapolation is a statistical inference, not a measurement. The press release does not publish the extrapolation procedure in detail, so independent researchers cannot yet reproduce that step from the public record. Treat the chemical-accuracy claim as "announced, pending verification" until the method is published or the data released.

Second, the efficiency headline is a step-level comparison, not a whole-run figure. The "300 times fewer Fugaku node-hours" compares one intermediate step (iteration 2, 2,518 node-hours) against a previously published SQD result. The paper itself reports that one full [4Fe-4S] calculation used approximately 0.66 million Fugaku node-hours in total. Both numbers are real; they describe different things, and the bigger one is the honest cost of the whole run.

Third, this is a preprint, and "first" is Qunova's word. The underlying paper has not been peer-reviewed, and the claim that HI-VQE is "the first quantum algorithm to reach chemical accuracy at 80 qubits" comes from the company, not from an independent benchmark. Other groups have reported chemical accuracy at similar or larger scales on different molecules and metrics, so the "first" needs the qualifier "on this molecule, by this measure".

Fourth, matching classical computing is not beating it. The result puts a quantum workflow on equal footing with the best classical method for one of the hardest molecules in chemistry. That is genuinely significant — it is the standard a quantum method must reach before industrial users can trust it. It is not quantum advantage, and Qunova does not claim it is. The next goal the company names is finding systems where HI-VQE consistently beats the best classical methods, starting with the full nitrogenase cofactor (FeMoco), which the paper estimates would need roughly 100–150 qubits.

Why it matters for Indian readers

No Indian reader will run a HI-VQE job tomorrow, and Qunova has announced no India deployment. But the molecule at the centre of this story is directly relevant to India. Iron-sulfur clusters are the catalytic heart of nitrogenase, the enzyme that fixes nitrogen from the air — the chemistry behind nitrogen fertiliser. India is one of the world's largest users of nitrogen fertiliser, and a better catalyst for nitrogen fixation is one of the oldest dreams in chemistry: it could cut the energy and cost of fertiliser production. A quantum method that can be trusted on iron-sulfur clusters is a small, real step toward being able to design such catalysts on a computer instead of finding them by trial and error.

The infrastructure angle matters too. This result was produced by a quantum–supercomputer hybrid: a quantum processor proposing configurations, a classical supercomputer doing the heavy diagonalisation. That is the architecture India's National Quantum Mission (₹6,003 crore, launched in 2023) is building toward, and Indian researchers at IISc and the IITs already work on exactly this class of hybrid algorithms. The skills this result rewards — quantum circuit design, error mitigation, classical–quantum workflow engineering — are the same ones Indian students and startups can practise today on cloud quantum hardware, without owning a quantum computer.

Finally, the caution in this story is itself useful. The gap between a press-release headline and the preprint's own caveats is where hype lives. Reading both — as this post has done — is the habit that keeps a fast-moving field honest.

Should you care?

✅ Follow it if: you work in quantum computing, computational chemistry or materials science — a quantum workflow matching DMRG on a [4Fe-4S] cluster is the credibility bar the field has been waiting for, and the next test, the full nitrogenase cofactor at 100–150 qubits, is the one to watch.

⏳ Wait if: you are looking for a drug, a fertiliser or a product — this is a benchmark result on one molecule, announced as a press release, with the key number resting on an extrapolation that has not yet been independently reproduced.

FAQs

What is chemical accuracy in quantum computing?
Chemical accuracy means a calculated molecular energy within 1.6 millihartree (about 1 kilocalorie per mole) of the true value. It is the precision chemists generally require before a calculation can guide real decisions, such as which catalyst to synthesise. Qunova's extrapolated [4Fe-4S] energy sits 1.1 millihartree from the best classical reference, inside that bar — but the figure is an extrapolation, not a direct measurement.

What is HI-VQE and how is it different from VQE?
HI-VQE is the Handover Iterative Variational Quantum Eigensolver, a hybrid algorithm from Qunova Computing. Ordinary VQE estimates a molecule's energy by measuring it on quantum hardware, which is slow and noise-sensitive. HI-VQE uses the quantum chip only to propose candidate electron configurations, then hands them to a classical computer that diagonalises the Hamiltonian in the subspace they span. The loop repeats, refining the circuit each round, and no energy is ever estimated from hardware measurements.

What is the 4Fe-4S iron-sulfur cluster and why test on it?
The [4Fe-4S] cubane cluster — four iron and four sulfur atoms in a cube — is the catalytic core of proteins including ferredoxins, hydrogenases and nitrogenase, the enzyme that fixes nitrogen from the air. Its electrons are strongly correlated, with many near-degenerate spin states, making it one of the most demanding test cases in quantum chemistry: the exact solution would need between 10⁸ and 10¹⁵ electron configurations. A method trusted on this cluster is a method that can plausibly be trusted on real catalysts and drug targets.

Did a quantum computer beat a classical supercomputer?
No. HI-VQE matched the best classical method (DMRG) on this molecule — it did not beat it. The result used both machines together: IBM's Heron r2 quantum processor proposed configurations while RIKEN's Fugaku supercomputer did the classical diagonalisation, using about 0.66 million node-hours for the full calculation. The achievement is reaching the classical standard with a quantum workflow, which is the prerequisite for industrial use — not a victory over classical computing.

Takeaway

Qunova's HI-VQE reached chemical accuracy on an 80-qubit iron-sulfur calculation — a real credibility milestone for quantum chemistry — but the headline figure is an extrapolation from a preprint, and the result matches the best classical method rather than beating it. Watch the independent verification, and the 100–150-qubit nitrogenase test next.

Read the paper itself — arXiv:2601.06935 — and compare its caveats with the press release; the gap between the two is where the real story is. Our IBM Quantum Nighthawk r3 explainer covers the hardware roadmap this kind of algorithm will run on.

Sources

The Quantum Insider — "Qunova's Hybrid Computing Method Matches Classical Benchmark for Iron-Sulfur Cluster" (Qunova press release, 12 November 2026)
arXiv:2601.06935 — Yoo et al., "Extending the Handover-Iterative VQE to Challenging Strongly Correlated Systems: N₂ and Fe-S Cluster" (preprint, v2 12 September 2026)
IBM Quantum Documentation — "HI-VQE Chemistry by Qunova Computing" (method description and benchmark table)
arXiv:2503.06292 — Pellow-Jarman et al., "HI-VQE: Handover Iterative Variational Quantum Eigensolver for Efficient Quantum Chemistry Calculations" (original method paper, March 2025)
TechByCountry — Google DeepMind AI Agents Quantum Programs Explained: 290x Cheaper Designs Proved in Lean
TechByCountry — Shallow Quantum Circuits Distribution Learning Advantage Explained