Multireference error mitigation for quantum computation of chemistry

This paper introduces Multireference-state Error Mitigation (MREM), an advanced quantum error mitigation technique that utilizes compact multireference states constructed via Givens rotations to significantly improve the accuracy of quantum chemistry calculations for strongly correlated molecular systems, overcoming the limitations of traditional Reference-state Error Mitigation.

The Gist

Imagine you are trying to take a perfect, high-definition photograph of a complex scene using a camera that is slightly broken. The lens is smudged, and the sensor has a bit of static. No matter how carefully you frame the shot, the resulting image will be blurry and distorted.

In the world of quantum computing, scientists are trying to "photograph" the behavior of molecules (like water or nitrogen) to understand chemistry. But the "cameras" they use today—called Noisy Intermediate-Scale Quantum (Q) devices—are very much like that broken camera. They are prone to "noise" (static and errors) that ruins the calculation, making the results unreliable.

This paper introduces a clever new trick to fix these blurry pictures without waiting for perfect, expensive cameras to be built. Here is how they did it, explained simply:

The Problem: The "Broken Camera"

When scientists use quantum computers to calculate the energy of a molecule, the noise in the machine makes the answer wrong. Usually, the answer is too high, like a scale that always adds a few extra pounds to your weight.

To fix this, they previously used a method called Reference-State Error Mitigation (REM).

• The Old Trick: Imagine you know exactly what a "perfect" photo of a simple object (like a plain white ball) should look like. You take a photo of that ball with your broken camera, see how blurry it is, and then use that "blur factor" to clean up the photo of the complex scene.
• The Limitation: This worked great for simple molecules (like a single ball). But for complex molecules with "strongly correlated" electrons (where electrons are dancing in a complicated, synchronized way), the "plain white ball" reference wasn't good enough. The reference was too simple to help fix the complex picture.

The New Solution: MREM (The "Smart Reference")

The authors, led by Hang Zou and colleagues, developed a new method called Multireference-State Error Mitigation (MREM).

Instead of using a simple "plain white ball" as a reference, they use a complex, pre-sketched blueprint that looks very similar to the actual molecule they are studying.

• The Analogy: If the old method was using a photo of a blank wall to fix a photo of a crowded city street, the new method uses a rough sketch of that same city street. Because the sketch already captures the complexity of the crowd, the "blur" on the sketch tells you exactly how to fix the blurry photo of the real street.

How They Build the Blueprint: Givens Rotations

To create these complex reference sketches on a quantum computer, they needed a special tool. They used something called Givens rotations.

• The Metaphor: Think of a quantum state as a stack of cards. A simple reference is just one card. A complex reference is a specific mix of a few cards shuffled together.
• The Tool: Givens rotations are like a very precise, magical shuffler. They allow the scientists to take a simple starting state and mix in just a few extra "cards" (quantum configurations) to create a reference that closely resembles the messy, complex reality of the molecule.
• Why it matters: They didn't try to mix every possible card (which would take too long and introduce too much noise). They picked the top 2 or 3 most important cards that mattered most. This kept the process fast and efficient while still being accurate enough to fix the errors.

The Results: Sharper Pictures

The team tested this new method on three molecules: Water ($H_2O$), Nitrogen ($N_2$), and Fluorine ($F_2$).

1. Water ($H_2O$): The new method cleaned up the noise significantly, giving a much clearer picture of the molecule's energy than the old method.
2. Nitrogen ($N_2$): This molecule is very tricky because its electrons are highly correlated. The old method struggled here, but the new "complex blueprint" approach managed to recover the correct physical behavior, especially when the molecule was being stretched.
3. Fluorine ($F_2$): This was the biggest success. The new method reduced the error by about 100 times compared to the raw noisy data, and by 10 times compared to the old method. It got the answer so close to the "perfect" theoretical value that it was almost indistinguishable from a noise-free calculation.

The Bottom Line

The paper claims that by using a slightly more complex "reference" (a mix of a few key quantum states) instead of a simple one, and by using a specific, efficient way to build that reference (Givens rotations), they can fix the errors in current quantum computers much better.

This allows scientists to get reliable, accurate results for difficult chemical problems today, even while the quantum computers themselves are still imperfect and noisy. It's like getting a crystal-clear photo from a broken camera by using a smarter way to correct the blur.

Technical Summary

Technical Summary: Multireference Error Mitigation for Quantum Computation of Chemistry

Problem Statement
Quantum error mitigation (QEM) is critical for enhancing the reliability of quantum chemistry algorithms on Noisy Intermediate-Scale Quantum (NISQ) devices. While Reference-State Error Mitigation (REM) has proven effective for weakly correlated systems by using a single-reference Hartree-Fock (HF) state to quantify and correct noise-induced energy errors, it faces significant limitations in strongly correlated regimes (e.g., bond dissociation). In these scenarios, the target ground state is often a multireference (MR) state—a linear combination of multiple Slater determinants (SDs). A single HF determinant fails to provide sufficient overlap with the true ground state, rendering standard REM ineffective and leading to inaccurate error correction.

Methodology
The authors introduce Multireference-State Error Mitigation (MREM), an extension of REM designed to handle strong electron correlation. The core methodology involves:

1. Multireference Reference States: Instead of a single HF determinant, MREM utilizes truncated multireference wavefunctions (composed of 2–3 dominant SDs) derived from conventional quantum chemistry methods (e.g., CISD, CCSD, DMRG). These states are chosen to have substantial overlap with the target ground state.
2. Givens Rotations for State Preparation: To efficiently prepare these MR states on quantum hardware, the authors employ Givens rotations. These unitary transformations act on two selected elements of the Hilbert space, allowing for the controlled mixing of Slater determinants while preserving key physical symmetries, specifically particle number conservation and spin projection ($m_s$).
   • Single-excitation Givens rotations ($G(\theta)$) mix $|01\rangle$ and $|10\rangle$ states.
   • Double-excitation Givens rotations ($G^{(2)}(\theta)$) mix four-qubit states like $|0011\rangle$ and $|1100\rangle$.
   • Controlled Givens rotations ($CG(\theta)$) and Controlled-X ($CX$) gates are used to construct specific configurations, particularly when qubit tapering (symmetry reduction) disrupts direct spin-orbital mappings.
3. MREM Protocol:
   • Classical Step: A truncated MR state is selected based on classical calculations, and its exact (noiseless) energy is computed.
   • Quantum Step: The MR state is prepared on the quantum device using Givens-based circuits, and its noisy energy is measured to determine the error shift ($\Delta E_{MREM}$).
   • Correction: The target system is run via Variational Quantum Eigensolver (VQE), and the measured error shift is subtracted from the noisy target energy to yield the error-mitigated result.
4. Symmetry Enforcement: To address symmetry breaking caused by noise and hardware-efficient ansätze (HEA), a linear spin penalty term ($\lambda \cdot \hat{S}^2$) is added to the Hamiltonian to stabilize the singlet ground state.

Key Contributions

• Extension of REM: The paper proposes MREM, systematically extending the REM framework to strongly correlated systems by utilizing multireference states rather than single-reference HF states.
• Efficient Circuit Construction: The authors demonstrate the use of Givens rotations as a structured, symmetry-preserving method to construct compact multireference states on quantum hardware, balancing circuit expressivity with noise sensitivity.
• Truncation Strategy: The work employs truncated wavefunctions (2–3 SDs) to avoid excessive circuit depth, leveraging the fact that a few dominant configurations often suffice for effective error mitigation.
• Spin Penalty Integration: The implementation includes a linear spin penalty to mitigate spin contamination in noisy VQE simulations without introducing the high measurement overhead associated with squared deviation penalties.

Results
The authors validated MREM through noisy digital quantum simulations of three molecular systems: H$_2$O, N$_2$, and F$_2$, using the FakeSydneyV2 noise model.

• H$_2$O: MREM using 3 SDs significantly reduced energy errors compared to single-reference REM, recovering a more accurate potential energy surface (PES) despite the increased circuit complexity.
• N$_2$: Due to strong multireference character requiring double excitations, the unmitigated VQE with 3 SDs performed worse than HF-based VQE due to noise. However, MREM successfully recovered the physical behavior in the bond-stretching region, demonstrating that MR references combined with error mitigation outperform single-reference approaches even in high-noise regimes.
• F$_2$: MREM with 2 SDs achieved near-computational accuracy (errors reduced by ~2 orders of magnitude compared to noisy VQE and ~1 order compared to REM-HF). The MR state provided a better initialization, helping the VQE avoid local minima in regions where standard VQE failed.

Significance and Claims
The paper claims that MREM broadens the scope of error mitigation to encompass molecular systems with pronounced electron correlation, a regime where standard REM fails. By utilizing physically motivated multireference states constructed via Givens rotations, the method offers a cost-effective way to improve computational accuracy on NISQ devices. The authors emphasize that while MR states alone may not improve noisy VQE outcomes due to added circuit depth, they serve as superior, expressive references for the error mitigation step, enabling the extraction of meaningful physical information in noisy conditions. The work underscores the potential of combining classical chemical insight (truncated MR states) with efficient quantum circuit design to overcome hardware limitations.