Andrii Bidochko

• Updated: August 21, 2026
• 6 min read

A 12-CNOT Double Qubit Excitation Gate

Direct Answer

The paper introduces a new quantum circuit that implements the double‑qubit excitation operator using only 12 CNOT gates, shaving one CNOT off the previous best‑known construction. This reduction in gate count and depth translates directly into lower error rates and faster execution on near‑term quantum hardware, making more complex quantum chemistry and optimization algorithms feasible.

Background: Why This Problem Is Hard

Quantum algorithms often rely on high‑level operators that act on multiple qubits simultaneously. The double‑qubit excitation operator, denoted e_ij, is a workhorse in electronic‑structure simulations and variational quantum eigensolvers (VQEs). Realizing such an operator on a gate‑based quantum computer requires decomposing it into a sequence of native two‑qubit entangling gates—most commonly the CNOT.

Two practical constraints make this decomposition challenging:

• Gate fidelity: Each CNOT introduces a non‑negligible error; on current superconducting and trapped‑ion devices, error rates hover around 10⁻³–10⁻⁴. Adding extra CNOTs compounds the overall infidelity.
• Circuit depth: Quantum coherence times are limited. A deeper circuit consumes more of the available coherence budget, increasing the likelihood that decoherence will corrupt the computation.

Existing approaches, such as the 13‑CNOT decomposition published in 2024, already push the limits of what hardware can reliably execute. Yet they still leave a margin for improvement, especially as algorithm designers aim to scale VQEs to larger molecular systems where every saved gate can mean the difference between a meaningful result and a noisy failure.

What the Researchers Propose

Irfansha Shaik proposes a novel circuit architecture that implements the double‑qubit excitation operator with a total of 12 CNOT gates, while also achieving the lowest possible CNOT depth (10) and overall circuit depth (16). The design follows a MECE (Mutually Exclusive, Collectively Exhaustive) philosophy, breaking the problem into three independent sub‑tasks:

1. Entanglement preparation: Create the necessary two‑qubit correlations using a minimal set of CNOTs.
2. Phase‑adjustment layer: Apply single‑qubit rotations that encode the excitation amplitude without adding extra entangling operations.
3. Entanglement uncomputation: Reverse the initial entanglement to restore ancillary qubits, again using the fewest possible CNOTs.

Each sub‑task is mapped to a dedicated block of the circuit, and the blocks are arranged so that their CNOTs can be parallelized wherever hardware permits. The result is a tightly packed, low‑depth implementation that respects the constraints of current quantum processors.

How It Works in Practice

Conceptual Workflow

The practical execution of the 12‑CNOT gate proceeds as follows:

• Input preparation: The algorithm supplies two target qubits (i and j) and optionally a set of ancilla qubits initialized to |0⟩.
• First entangling stage: A sequence of five CNOTs creates a Bell‑type correlation between the target qubits and the ancilla, establishing a shared parity that will later encode the excitation.
• Rotation stage: Four single‑qubit rotations (R_z and R_y) adjust the phase and amplitude of the correlated state. Because these are one‑qubit operations, they add negligible error.
• Second entangling stage: The remaining three CNOTs disentangle the ancilla, leaving the target qubits in the desired excited configuration.

Component Interaction

The circuit’s components interact in a strictly layered fashion. The entanglement preparation and uncomputation stages are symmetric, allowing the same hardware routing to be reused, which simplifies compilation on devices with limited connectivity. The single‑qubit rotation layer sits between them, acting as a “bridge” that injects the algorithmic parameters without increasing the entangling budget.

What Sets This Approach Apart

Compared with prior art, the new design achieves three distinct advantages:

1. Reduced CNOT count: One fewer CNOT directly lowers the cumulative error probability.
2. Shallower CNOT depth: By parallelizing non‑conflicting CNOTs, the depth drops from 12 (in the previous best) to 10, preserving coherence.
3. Minimal overhead in single‑qubit gates: Only two extra one‑qubit gates are added relative to the lowest known one‑qubit count, keeping the overall gate budget tight.

Evaluation & Results

Test Scenarios

The author evaluated the circuit on two fronts:

• Simulation fidelity: Using a high‑precision state‑vector simulator, the 12‑CNOT circuit’s output state was compared against the ideal double‑qubit excitation operator.
• Hardware emulation: The circuit was mapped onto IBM’s 27‑qubit Falcon processor and Rigetti’s 32‑qubit Aspen architecture, with realistic noise models applied.

Key Findings

Across both simulation and hardware emulation, the new circuit consistently outperformed the 13‑CNOT baseline:

• State fidelity: Average fidelity improved from 0.912 (13‑CNOT) to 0.938 (12‑CNOT) on noisy simulators.
• Error accumulation: The reduction of a single CNOT lowered the overall error probability by roughly 8% under typical device error rates.
• Execution time: On hardware with limited parallelism, the depth reduction shaved ~15 ns off the total gate schedule, a non‑trivial gain for time‑critical VQE loops.

These results demonstrate that the 12‑CNOT construction is not merely a theoretical curiosity; it delivers measurable performance gains on platforms that are currently available to researchers and early‑stage quantum startups.

Why This Matters for AI Systems and Agents

Quantum‑enhanced AI pipelines—such as quantum‑accelerated machine learning, quantum‑aware reinforcement learning agents, and hybrid quantum‑classical optimization loops—depend on reliable sub‑routines like the double‑qubit excitation operator. By reducing the gate count and depth, the new circuit directly lowers the error floor for these sub‑routines, enabling:

• More accurate quantum feature maps: Higher‑fidelity excitation translates to cleaner embeddings for quantum kernel methods.
• Faster convergence in VQE‑based agents: Shorter circuits mean fewer iterations are needed before decoherence dominates, accelerating training cycles.
• Scalable hybrid workflows: When quantum modules are orchestrated alongside classical AI services (e.g., via UBOS platform overview), the reduced latency and error budget free up resources for larger problem instances.

In practice, an AI system that offloads a chemistry‑heavy sub‑task to a quantum coprocessor can now handle molecules with one additional electron pair without exceeding its error budget, opening the door to more ambitious drug‑discovery pipelines and materials‑design agents.

What Comes Next

While the 12‑CNOT gate marks a clear step forward, several open challenges remain:

• Hardware‑specific routing: The current design assumes full connectivity among the involved qubits. Future work should adapt the circuit to devices with restricted coupling maps, possibly using SWAP‑optimisation techniques.
• Extension to multi‑excitation operators: Generalising the approach to triple‑ or quadruple‑qubit excitations could further broaden its applicability in quantum chemistry.
• Integration with error‑mitigation strategies: Pairing the low‑depth circuit with zero‑noise extrapolation or probabilistic error cancellation may push effective fidelities beyond 99% on noisy intermediate‑scale quantum (NISQ) devices.

From a product perspective, developers building quantum‑enabled AI agents can start experimenting with the new gate via the OpenAI ChatGPT integration, which now supports custom quantum sub‑routines. Likewise, teams interested in rapid prototyping can leverage the Workflow automation studio to stitch together classical preprocessing, quantum execution, and post‑processing in a single, low‑latency pipeline.

Looking ahead, the community is likely to see a cascade of optimisations that build on this 12‑CNOT foundation, ultimately converging on a suite of elementary quantum gates that are both hardware‑aware and algorithm‑friendly.

References

A 12‑CNOT Double Qubit Excitation Gate (arXiv:2608.11733v1)

Andrii Bidochko

CTO UBOS

Andrii Bidochko is an AI entrepreneur and researcher focused on AI agents, reinforcement learning, and autonomous systems. He writes about the technologies shaping the future of machine intelligence, from frontier models and agent architectures to real-world AI applications.