Ground-state energies of Ising models calculated using the samples from a quantum computer that simulates short-time evolution

This paper demonstrates the calculation of ground-state energies for homogeneous and random-coupling Ising models on up to 63 qubits using the Cascaded Variational Quantum Eigensolver with a Guided-Sampling Ansatz, establishing the error boundaries and performance insights for near-term quantum utility on heavy-hex lattice architectures.

The Gist

Imagine you are trying to find the absolute lowest point in a vast, foggy mountain range. This "lowest point" represents the most stable, calm state of a complex system (in this case, a magnetic material called an Ising model). On a classical computer, trying to map every single valley and peak in a huge mountain range is like trying to count every grain of sand on a beach; it takes too long and is practically impossible.

This paper describes an experiment where researchers used a quantum computer to help find that lowest point, but with a twist: they didn't wait for the computer to be perfect. Instead, they used a "good enough" approach that works even when the computer is a bit noisy and error-prone.

Here is a breakdown of their method and findings using everyday analogies:

The Problem: The Foggy Mountain

The researchers are studying a specific type of magnetic system (the Ising model). They want to know its "ground-state energy," which is just a fancy way of saying: What is the most relaxed, lowest-energy arrangement of these magnetic spins?

For large systems, classical computers get lost in the fog. They can't calculate the answer because there are too many possibilities.

The Solution: A Guided Hike (The CVQE Algorithm)

Instead of trying to solve the whole mountain at once, the researchers used a method called the Cascaded Variational Quantum Eigensolver (CVQE) with a Guided-Sampling Ansatz (GSA).

Think of it like this:

1. The Short Hike: Imagine you are blindfolded and dropped on a mountain. You can't see the bottom. So, you take a very short, quick walk (short-time evolution) in a specific direction. You don't reach the bottom, but you end up in a valley that is lower than where you started.
2. The Sample: You take a snapshot of where you ended up. You do this many times (1,000 times, in their experiment).
3. The Map: You give all these snapshots to a classical computer (a regular laptop). The laptop looks at all the spots you visited and says, "Okay, if we combine all these specific locations, we can build a small, detailed map of the most promising valleys."
4. The Calculation: The classical computer solves the math for just that small map to find the true lowest point.

The "Guided-Sampling" part is the key. The quantum computer doesn't just guess randomly; it takes that short, quick walk to "guide" the search toward the right area, filtering out the useless parts of the mountain.

The Experiment: IBM's "Heavy-Hex" Playground

The researchers used an IBM quantum computer named Torino. This computer has a specific layout of qubits (the quantum bits) that looks like a heavy-hex lattice (a pattern of connected hexagons). They mapped their magnetic problem directly onto this shape so the computer could handle it efficiently.

They tested two types of magnetic systems:

1. Homogeneous: Where all the magnets interact with each other in the exact same way (like a perfectly uniform forest).
2. Random-Coupling: Where the interactions are random and messy (like a forest where some trees are tangled, some are far apart, and the wind blows differently everywhere). This is harder to solve and is similar to a "spin glass."

They tested systems with up to 63 qubits (spins).

The Results: When Does the Fog Win?

The researchers found that this method works well, but there is a limit.

• The Sweet Spot: For smaller systems and weaker magnetic interactions, the quantum computer successfully guided them to a very accurate answer.
• The Breaking Point: As the system got larger (more qubits) or the magnetic interactions got stronger, the "noise" (errors) in the quantum computer started to drown out the signal. It's like trying to hear a whisper in a hurricane; eventually, you can't tell if you're in a valley or just on a windy ridge.

They discovered a "boundary" where the quantum computer stops being useful. If the system is too big or too complex, the errors make the answer unreliable.

The "Information Score" (How do we know it's right?)

One of the cleverest parts of the paper is how they checked if their answer was good without knowing the answer beforehand.

They created an "Information Ratio":

• Imagine the quantum computer is a detective gathering clues (samples).
• Imagine the classical computer is the detective trying to solve the case using those clues.
• If the detective gathers more clues than are actually needed to solve the case, they are confident they have the right answer.
• If the detective gathers fewer clues than needed, they are just guessing.

They found that when their "Information Ratio" was positive, the answer was likely correct. When it dropped below zero, the quantum errors had taken over, and the result was unreliable.

The Big Takeaway

The paper concludes that Ising models are a perfect candidate for today's "noisy" quantum computers.

Even though the quantum computers aren't perfect yet, the math behind these magnetic models is simple enough that the "short hike" method works. The number of clues (samples) needed to find the answer doesn't explode exponentially as the system gets bigger; it grows slowly. This suggests that we can use current, imperfect quantum computers to solve physics problems that are impossible for classical computers, specifically for understanding magnetic materials and spin glasses.

In short: They taught a noisy quantum computer to take a few quick steps to find the right neighborhood, then used a regular computer to find the exact house. It works great for medium-sized problems, but if the neighborhood gets too huge, the noise gets too loud to hear the directions.

Technical Summary

1. Problem Statement

The paper addresses the challenge of calculating ground-state energies for the Transverse-Field Ising Model on near-term quantum computers (NISQ devices).

• The Challenge: Many quantum algorithms require long coherence times and deep circuits, which are currently infeasible due to noise and decoherence.
• The Specific Goal: To determine if short-time evolution simulations can be effectively used to construct an ansatz for finding ground-state energies, specifically for systems up to 63 qubits.
• The Context: The authors aim to operate in the "quantum utility" regime, where quantum computers can solve problems that are classically intractable (or difficult to approximate without specific heuristics) but do not yet require full fault tolerance. They focus on both homogeneous and random-coupling (spin-glass) models on a heavy-hex lattice, matching the architecture of IBM's quantum processors.

2. Methodology

The authors employ a hybrid quantum-classical approach combining the Cascaded Variational Quantum Eigensolver (CVQE) with a Guided-Sampling Ansatz (GSA).

A. Short-Time Evolution (Diabatic Preparation)

Instead of slow adiabatic evolution (which requires long coherence times), the authors use short-time evolution ($T \sim \Delta t$).

• Hamiltonian: They define a time-dependent Hamiltonian $\hat{H}(t)$ where the coupling strength $\Omega_{ij}(t)$ ramps from 0 to $J_{ij}$ over time $T$.
• Trotter-Suzuki Decomposition: The evolution operator is approximated using a single-step (or few-step) Trotter decomposition.
• Mechanism: Even if the evolution is not perfectly adiabatic, the short-time evolution acts as a filter. It exponentially suppresses high-energy states outside a window $\Delta E \sim 1/\Delta t$, guiding the system toward a subspace containing the ground state.

B. Guided-Sampling Ansatz (GSA)

The quantum computer is used to generate a "guiding state" rather than the final solution directly.

1. Sampling: The quantum computer runs the short-time evolution circuit and performs $N_S$ shots to measure basis states $|\Phi_s\rangle$.
2. Subspace Construction: These measured states form a set $B$. The authors then construct a larger subspace $B_1$ by applying the Hamiltonian $\hat{H}$ to each state in $B$ once (i.e., $B_1 = \{ \hat{H}|\Phi_s\rangle \}$).
3. Classical Diagonalization: The Hamiltonian is projected onto this subspace $B_1$ to form a matrix $\hat{H}^{(1)}$. This matrix is diagonalized on a classical computer to find the lowest eigenvalue (the effective ground-state energy).
4. Efficiency: The method relies on the fact that for Ising models, the size of the subspace $|B_1|$ grows sub-exponentially with system size, making the classical diagonalization feasible.

C. Entropic Analysis (Error Metric)

To assess the reliability of the results without knowing the exact ground state, the authors introduce an information-theoretic metric:

• Shot Entropy ($S_B$): Measures the information gained from quantum measurements.
• Distribution Entropy ($S_\Psi$): Measures the information required to describe the constructed ground state.
• Information Ratio ($R$): Defined as $R = \log_2 \left( \frac{I_B}{K I_\Psi} \right)$, where $I_B$ is information from the quantum computer, $I_\Psi$ is information needed for the ansatz, and $K$ is a coupling factor.
  • $R > 0$: Indicates the quantum computer provided sufficient information to build an accurate state (successful).
  • $R < 0$: Indicates the quantum computer provided insufficient information, likely due to noise (failure).

3. Key Contributions

• Demonstration of Quantum Utility: Successfully calculated ground-state energies for Ising models with up to 63 qubits on the IBM Torino (Heron r1) processor, a scale where brute-force classical simulation is impossible.
• Novel Error Metric: Developed the Information Ratio ($R$) to quantitatively determine the boundary of acceptable quantum errors. This allows researchers to distinguish between physical deviations and hardware-induced errors without prior knowledge of the exact solution.
• Subspace Analysis: Provided evidence that the number of basis states required to approximate the ground state for Ising models grows polynomially (linear to cubic) rather than exponentially in the regime where $|J| \sim |B|$. This validates the efficiency of the CVQE/GSA approach for these specific problems.
• Random Coupling (Spin Glass): Extended the method to random-coupling models (spin glasses), a class of problems notoriously difficult for classical algorithms, showing that the method remains robust up to certain error thresholds.

4. Results

• Homogeneous Model:
  • The algorithm successfully tracked the ground-state energy and average spin as a function of coupling strength ($J_0$) and system size ($N_Q$).
  • Error Boundary: Deviations from smooth physical curves correlated with $R < 0$. As $N_Q$ and $|J_0|$ increased, quantum noise (gate errors and readout errors) degraded the results, eventually causing the method to fail for the largest systems at high coupling.
• Random-Coupling Model:
  • The method handled the disorder of spin glasses effectively.
  • The Information Ratio ($R$) successfully identified regions where quantum errors dominated the results (bottom-right of the parameter space), while $R > 0$ in other regions indicated reliable results.
• Subspace Scaling:
  • Analysis of the basis state structure showed that the number of significant basis states (those with non-negligible probability) scales sub-exponentially (roughly linear to cubic) with the number of qubits for both homogeneous and random models.
  • This confirms that the "effective" Hilbert space for the ground state is small enough for classical post-processing.

5. Significance and Conclusion

• Feasibility of NISQ Algorithms: The paper demonstrates that short-time evolution combined with guided sampling is a viable strategy for near-term quantum computers. It bypasses the need for deep, error-corrected circuits.
• Physical Insights: The ability to simulate random-coupling Ising models (spin glasses) on quantum hardware opens the door to discovering new physical properties of disordered magnetic systems that are currently unknown.
• Future Outlook: The authors conclude that while current hardware limits the accuracy for the largest systems at high coupling, the information requirements for the ground state do not increase significantly with system size. As quantum hardware improves (lower gate errors), this method is expected to scale further, potentially solving problems that are strictly intractable for classical computers.

In summary, this work provides a concrete pathway for utilizing current quantum hardware to solve complex many-body physics problems by leveraging short-time dynamics and hybrid quantum-classical subspace expansion, validated by a novel information-theoretic error metric.