⚛️ quantum physics

Exponential Scaling Barriers for Variational Quantum Eigensolvers

This paper demonstrates that the Variational Quantum Eigensolver (VQE) faces exponential scaling barriers in both adaptive iterations and circuit depth as system size increases, suggesting it is unlikely to efficiently simulate large molecular systems without exponential resource requirements.

Original authors: Manuel Hagelueken, David A. Kreplin, Florian Wieland, Marco F. Huber, Marco Roth

Published 2026-03-16

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Manuel Hagelueken, David A. Kreplin, Florian Wieland, Marco F. Huber, Marco Roth

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to find the lowest point in a vast, foggy mountain range. This is what scientists do when they try to calculate the energy of a molecule: they are looking for the "ground state," the most stable and comfortable position for the atoms.

For decades, classical computers have been good at this for small mountains. But when the mountains get huge and complex (like large molecules with many interacting electrons), the fog gets so thick that classical computers get lost. This is where Quantum Computers were supposed to save the day.

Enter VQE (Variational Quantum Eigensolver). Think of VQE as a smart, adaptive hiker. Instead of trying to map the whole mountain at once, the hiker takes a step, checks if they are lower, takes another step, and repeats until they find the bottom. The hope was that this hiker could solve problems that were impossible for classical computers.

But this new paper drops a cold splash of water on that hope.

Here is the story of what the researchers found, explained simply:

1. The "Magic" Metric: The Complexity Score

The researchers realized that not all molecules are equally hard to solve. Some are like a gentle hill; others are like a jagged, chaotic cliff.

To measure this difficulty, they invented a "Complexity Score" based on something called Rényi Entropy.

The Analogy: Imagine you are trying to describe a painting.
- If the painting is just a blue sky, you only need one word: "Blue." (Low complexity).
- If the painting is a chaotic storm with thousands of unique brushstrokes, you need a massive, detailed description. (High complexity).
The researchers found that this "Complexity Score" (Rényi Entropy) is a crystal ball. If they calculated this score using a standard classical computer, they could predict exactly how many steps the quantum hiker (VQE) would need to take to find the bottom.

2. The Bad News: The Steps Grow Exponentially

The most shocking discovery is how the number of steps grows as the molecule gets bigger.

The Analogy: Imagine you are climbing a staircase.
- If you add one more step to the staircase, a normal person might need one more step to climb it.
- But for these quantum molecules, adding just one more atom to the molecule doesn't just add one step to the climb. It doubles the number of steps required. Then it doubles again. And again.
This is called Exponential Scaling.
- A small molecule might take 10 steps.
- A medium one might take 1,000 steps.
- A large, chemically interesting one (like a drug molecule or a catalyst) might require millions or billions of steps.

3. The Hardware Reality Check

The researchers tested this on 21 different molecules, ranging from simple chains of hydrogen to complex iron-based clusters.

The Result: For the molecules that actually matter in real-world chemistry (like those involved in making fertilizer or batteries), the quantum computer would need to perform thousands of iterations.
The Cost: Each iteration requires the quantum computer to run a circuit (a sequence of operations). To do thousands of iterations, the circuit becomes incredibly deep and long.
The Problem: Current quantum computers are like a toddler trying to run a marathon. They are "noisy" and make mistakes. If the circuit is too long (too many steps), the noise drowns out the signal before the hiker ever reaches the bottom. The researchers estimate that for real-world problems, we would need a quantum computer with resources far beyond what exists today—perhaps thousands of times more powerful.

4. Why Common Tests Were Misleading

For years, scientists tested these algorithms on simple molecules like Hydrogen chains or Lithium Hydride.

The Analogy: It's like testing a Formula 1 car on a flat, empty parking lot. The car looks fast and efficient.
The Reality: The researchers showed that these test molecules are "easy mode." They have low complexity scores. They don't represent the messy, chaotic mountains of real chemistry. When they tested on harder, more realistic molecules, the algorithm struggled immensely.

5. The "Spin" Twist

The paper also noticed something weird happening during the climb. To find the true bottom, the quantum hiker sometimes has to temporarily "break the rules" of physics (specifically, spin symmetry) to navigate a tricky part of the mountain, only to fix it later.

If you force the hiker to strictly follow the rules at every single step, they get stuck on a plateau and never find the bottom.
This suggests that the "messy" behavior of the algorithm is actually necessary, but it makes the path even harder to predict and control.

The Bottom Line

This paper is a reality check for the quantum computing community.

The Promise: Quantum computers will revolutionize chemistry.
The Barrier: The specific method (VQE) we are currently using to do this has a fundamental flaw. As molecules get bigger, the work required doesn't just grow; it explodes.
The Conclusion: Unless we invent a completely new way to climb these mountains, the current version of VQE is unlikely to solve the big, important chemistry problems we care about anytime soon. We are still a long way from having a quantum computer that can design new drugs or materials on its own.

In short: The map (Rényi entropy) tells us the journey is exponentially harder than we hoped, and our current vehicle (VQE) isn't built for the terrain.

1. Problem Statement

The Variational Quantum Eigensolver (VQE), particularly its adaptive variant ADAPT-VQE, is widely considered a promising near-term algorithm for calculating ground-state energies of quantum systems that are intractable for classical computers. The prevailing hope is that VQE can mitigate the exponential growth of the Hilbert space with system size.

However, a fundamental question remains unanswered: Do these methods converge reliably for large, strongly correlated molecular systems, and at what computational cost? While VQE has been successfully applied to small molecules (e.g., $H_4$ , $LiH$), it is unclear if these successes generalize to chemically relevant, strongly correlated systems (e.g., transition-metal clusters) without requiring exponential resources. This paper investigates whether the current form of VQE faces a fundamental scaling barrier that prevents it from simulating large molecular systems with high fidelity.

2. Methodology

The authors conducted a systematic benchmarking study using a curated dataset of 21 molecular systems (MolVQE-21) to analyze the scaling behavior of ADAPT-VQE.

Benchmark Set (MolVQE-21): The set includes 21 molecules spanning diverse chemical classes (hydrogen chains, organic molecules, transition-metal complexes) with active spaces ranging from 4 to 10 orbitals. It covers various spin states and basis sets.
Algorithm Variants: The study primarily employed CEO-ADAPT-VQE (Coupled Exchange Operator) with the TETRIS extension (which appends multiple non-overlapping operators per iteration to reduce iteration count). Comparisons were made against Qubit-ADAPT-VQE, QEB-ADAPT-VQE, and fixed-structure ansätze (pp-tUPS) with global optimization (BHPT-VQE).
Complexity Metric (Rényi Entropy): To predict computational cost, the authors introduced the Rényi entropy ( $h_\alpha$ $h_{α}$ ) derived from classical CASSCF (Complete Active Space Self-Consistent Field) calculations.
- The CI coefficients from CASSCF define a probability distribution $p_i = |C_i|^2$ .
- The Rényi entropy is defined as $h_\alpha = \frac{1}{1-\alpha} \ln (\sum p_i^\alpha)$ .
- The parameter $\alpha$ controls sensitivity to the distribution's tail. The authors found that $\alpha \approx 0.25$ maximizes the predictive power for chemical accuracy.
Simulation Environment: All simulations were performed as exact statevector simulations using Qulacs. Convergence was measured by the energy error $\epsilon = E_{VQE} - E_{CASSCF}$ , with chemical accuracy defined as $\epsilon_{chem} \approx 1.6 \times 10^{-3}$ Eh (1 kcal/mol).

3. Key Contributions

Discovery of Exponential Scaling: The paper demonstrates that the number of adaptive iterations ( $n_{ADAPT}$ ) required to reach chemical accuracy scales exponentially with the system size (number of active orbitals) and problem complexity.
Rényi Entropy as a Predictor: The authors establish that the Rényi entropy computed from purely classical CASSCF calculations is a highly accurate predictor ( $R^2 \approx 0.99$ ) of the number of VQE iterations required. This allows for resource estimation for systems too large for direct quantum simulation.
MolVQE-21 Benchmark Suite: The introduction of a comprehensive benchmark set that moves beyond simple test cases (like $H_4$ ) to include transition-metal complexes and stretched geometries, providing a more realistic assessment of algorithm performance.
Analysis of Convergence Dynamics: The study reveals that transient spin-symmetry breaking is functionally necessary for convergence in ADAPT-VQE, as spin-preserving operator pools fail to converge for certain systems.

4. Key Results

Correlation between Entropy and Iterations:
- A linear relationship was found between the Rényi entropy ( $h_{\alpha^*}$ with $\alpha^* \approx 0.25$ ) and the logarithm of the required iterations ( $\log(n_{ADAPT})$ ).
- This implies an exponential relationship: $n_{ADAPT} \propto \exp(h_\alpha)$ .
- The correlation holds across diverse molecules (organic, inorganic, transition metals) with $R^2 = 0.948$ at chemical accuracy.
System Size Scaling (Hydrogen Chains):
- Using hydrogen chains ( $H_4$ to $H_{10}$ ) as a model system, the authors confirmed that $n_{ADAPT}$ scales exponentially with the number of orbitals ( $N$ ).
- The fit yielded $R^2 = 0.98$ , with the relationship $\ln(n_{ADAPT}) \approx 0.75N - 1.83$ .
- Since circuit depth and parameter count scale linearly with $n_{ADAPT}$ , the total quantum resources also scale exponentially.
Extrapolation to Large Systems:
- Using the established scaling laws, the authors extrapolated resource requirements for chemically relevant, classically challenging systems:
  - $Cr_2$ (Chromium dimer): ~700 iterations.
  - $[2Fe-2S]$ (Iron-sulfur cluster): ~1,000+ iterations.
  - $[Cu_2O_2(NH_3)_6]^{2+}$ : ~5,000+ iterations.
- For a system like $H_{15}$ , reaching chemical accuracy would require approximately 3,600 iterations, translating to ~64,000 parameters and ~270,000 CNOT gates.
Geometric Dependence:
- Equilibrium Geometries: Show monotonic, predictable convergence.
- Stretched Geometries: Exhibit step-wise, non-monotonic convergence. While they reach chemical accuracy quickly (due to a low-energy, symmetry-broken state), they fail to converge to the true spin-pure ground state without significantly more iterations. This highlights that equilibrium geometries are more reliable for establishing scaling laws.
Comparison of Variants:
- CEO-ADAPT-VQE with TETRIS required the fewest iterations and CNOT gates compared to other variants.
- Fixed-structure ansätze (pp-tUPS) with global optimization (BHPT-VQE) failed to reach chemical accuracy for challenging systems like $FeO_4^{2-}$ within practical limits, suffering from local minima and barren plateaus.

5. Significance and Implications

Fundamental Barrier: The paper concludes that VQE in its current form is unlikely to simulate large molecular systems with high fidelity without exponential resource requirements. The exponential scaling of $n_{ADAPT}$ with problem complexity (quantified by Rényi entropy) presents a fundamental barrier.
Re-evaluation of Benchmarks: Commonly used test molecules ( $H_4$ , $LiH$, $BeH_2$ ) have low Rényi entropy and require few iterations, providing a false sense of scalability. The MolVQE-21 set is proposed as a more rigorous standard for future algorithm development.
Resource Constraints: The estimated circuit depths (e.g., $10^5$ CNOT gates) and parameter counts far exceed the capabilities of current Noisy Intermediate-Scale Quantum (NISQ) hardware, even with error mitigation.
Role of Classical Pre-computation: The strong correlation between classical CASSCF entropy and VQE cost suggests that for systems where classical methods can compute the CI coefficients, the quantum advantage may be limited. However, the study does not rule out VQE as a subroutine for more sophisticated algorithms (like Quantum Phase Estimation) or for state preparation.

In summary, this work provides a rigorous, data-driven argument that the "exponential advantage" of VQE is currently limited by the exponential scaling of the adaptive iterations required to capture strong static correlation, challenging the notion that VQE alone can solve classically intractable chemical problems in the near term.