⚛️ quantum physics

A Longitudinal Analysis of the CEC Single-Objective Competitions (2010-2024) and Implications for Variational Quantum Optimization

This paper analyzes the evolution of IEEE CEC single-objective optimization competitions from 2010 to 2024, highlighting how the introduction of non-separable benchmarks shifted dominance to rotation-invariant Differential Evolution variants and hybrid optimizers, ultimately suggesting that these evolved solvers possess the adaptive capabilities necessary for Variational Quantum Algorithms.

Original authors: Vojtěch Novák, Tomáš Bezděk, Ivan Zelinka, Swagatam Das, Martin Beseda

Published 2026-03-26

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Vojtěch Novák, Tomáš Bezděk, Ivan Zelinka, Swagatam Das, Martin Beseda

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

The Big Picture: A 15-Year Race for the "Smartest Solver"

Imagine a massive, annual Olympic-style competition called CEC. Instead of athletes running on a track, computer algorithms compete to solve the hardest math puzzles (optimization problems) imaginable. The goal is to find the single best answer (the "global optimum") in a landscape full of hills, valleys, and traps.

This paper looks back at the last 15 years of this race (2010–2024) to answer two questions:

How did the winners change?
Can these winning strategies help us build the next generation of computers: Quantum Computers?

Part 1: The Evolution of the Race (The Three Eras)

The paper argues that the race didn't just get harder; the rules of the terrain changed, forcing the runners to evolve.

Era 1: The "Specialists" (2010–2013)

The Terrain: The mountains were tall, but the paths were straight. You could walk north, then east, then south to find the bottom.
The Winners: A mix of different runners. Some used "Swarm Intelligence" (like a flock of birds), others used "Genetic Algorithms" (like breeding dogs for speed).
The Analogy: Imagine a maze where the walls are straight. If you just keep walking in a straight line, you'll eventually hit the exit. Different strategies worked fine here.

Era 2: The "Spin Cycle" (2014–2019)

The Terrain: The organizers decided to make the game unfair. They took the entire maze and rotated it. Now, "North" isn't a straight line anymore; it's a diagonal path that twists and turns.
The Problem: The old runners (like the "Swarm" or "Genetic" ones) were used to walking in straight lines (North/South/East/West). When they tried to walk North, they hit a wall because the maze was tilted. They got stuck.
The New Winner: A new runner called Differential Evolution (specifically L-SHADE) took over.
The Analogy: Imagine you are trying to find a hidden treasure in a foggy field.

Old Runners: They take steps only North, South, East, or West. If the treasure is in a diagonal valley, they keep hitting the sides.
L-SHADE: This runner doesn't care about North or South. It looks at two other runners, measures the distance between them, and jumps in that direction. Because it jumps based on the relationship between points, it naturally follows the diagonal valleys, no matter how the map is rotated. It learned to "dance" with the terrain rather than fight it.

Era 3: The "Swiss Army Knife" (2020–2024)

The Terrain: The maze got even weirder. It wasn't just one tilted valley; it was a patchwork quilt of different terrains. Some parts were smooth, some were jagged, some were tilted, and some were flat.
The Winners: The pure runners weren't enough. The winners became Hybrids. They were like a team of specialists working together.
The Analogy: Imagine a rescue team.

One member is a Climber (good for steep, twisted cliffs).
One is a Swimmer (good for flat, wide lakes).
One is a Navigator (good for finding the general direction).
The Winner: A single algorithm that can switch between being a climber, a swimmer, and a navigator depending on what part of the maze it's in. They also started using "Reinforcement Learning" (like a video game AI) to learn which tool to use at the right time.

Part 2: The Twist – Why This Matters for Quantum Computers

The authors realized something fascinating: The math puzzles in the CEC competition look exactly like the problems Quantum Computers face.

The Quantum Problem

Quantum computers are amazing but very fragile. When you try to program them, you are navigating a "landscape" to find the best settings. But this landscape has three scary features:

It's Rotated: Just like the 2014 CEC puzzles, the variables in a quantum computer are "entangled." Changing one knob affects everything else in a twisted, diagonal way.
It's Noisy: Quantum computers are like trying to hear a whisper in a hurricane. The data is full of static (noise).
It's Flat: Sometimes the landscape is so flat (a "Barren Plateau") that you can't tell which way is down.

The Solution

The paper suggests that the CEC winners (the L-SHADE and Hybrid algorithms) are the perfect tools for Quantum Computers.

Because they are rotationally invariant, they don't get confused by the twisted, entangled nature of quantum variables.
Because they are robust, they can handle the "noise" (static) better than traditional math methods.
Because they are hybrids, they can explore the flat, confusing areas without getting stuck.

The Final Takeaway

Think of the CEC competition as a training ground. For 15 years, computer scientists have been building "super-solvers" to beat increasingly difficult, twisted, and noisy puzzles.

The paper concludes that we have accidentally built the perfect remote control for Quantum Computers. Instead of inventing new math from scratch, we should take these evolved, battle-tested algorithms from the CEC competitions and use them to steer our new quantum machines.

In short: The algorithms that learned to win the hardest math games on Earth are the same ones we need to unlock the power of the quantum universe.

1. Problem Statement

The paper addresses two interconnected challenges:

Evolutionary Computation (EC) Trajectory: There is a lack of comprehensive longitudinal analysis regarding how the design of winning algorithms in the IEEE CEC Single-Objective Optimization competitions has evolved over the last 15 years. While annual reports exist, a holistic view of how benchmark function designs (specifically the introduction of non-separability and composition functions) forced specific algorithmic adaptations is missing.
Variational Quantum Algorithm (VQA) Optimization: VQAs, used for quantum chemistry and machine learning, face optimization landscapes characterized by extreme noise, non-separability (due to quantum entanglement), and "Barren Plateaus" (regions where gradients vanish). Current gradient-free optimizers often struggle with these specific topological features. The paper posits that the solvers evolved to win modern CEC competitions may possess the specific adaptive capabilities required to solve VQA optimization problems.

2. Methodology

The authors employed a systematic, longitudinal analysis of the CEC Single-Objective Optimization competition results from 2010 to 2024.

Data Collection: Rankings and algorithmic descriptions were collated from official repositories and technical reports of top-performing entries (Ranks 1–3).
Classification Framework: Algorithms were categorized not just by name, but by their primary search engine and structural mechanisms. The study tracked the dominance of specific families: Differential Evolution (DE), Evolution Strategies (ES/CMA-ES), Swarm Intelligence (PSO), and Genetic Algorithms (GA).
Mechanistic Analysis: The authors dissected the internal mechanisms of winning algorithms (e.g., Success-History Adaptation, Population Size Reduction, Eigenvector Crossover) to understand how they responded to specific benchmark changes, such as the introduction of dense rotation matrices.
Quantum Simulation: To validate the transferability of these solvers, the authors conducted a proof-of-concept simulation of a 10-qubit Variational Quantum Eigensolver (VQE) for a 1D Ising model. They tested various CEC-winning algorithms against both exact statevector simulations and noisy simulations (simulating finite shot noise) to evaluate robustness against local minima and stochastic noise.

3. Key Contributions and Findings

A. Three Distinct Evolutionary Eras of CEC Solvers

The analysis identifies three phases in the evolution of optimization algorithms:

Phase 1: Specialization & Diversity (2010–2013):
- Context: Benchmarks focused on large-scale optimization (1000D) and real-world engineering problems.
- Dominant Algorithms: Memetic Algorithms (MA-SW-Chains) and Genetic Algorithms (GA-MPC) excelled.
- Key Insight: Algorithms were diverse; no single framework dominated. GA-MPC won in 2011 by handling discrete/constrained real-world problems where DE struggled.
Phase 2: The Rise of Adaptive DE (2014–2019):
- Catalyst: The 2014 introduction of dense rotation matrices destroyed variable separability, rendering coordinate-dependent methods (PSO, standard GA) ineffective.
- Dominant Algorithm: L-SHADE (Linear population size reduction with Success-History Adaptation) and its variants became the undisputed champions.
- Mechanism: DE's mutation operator is inherently rotationally invariant. Unlike CMA-ES, which incurs a quadratic ( $O(D^2)$ ) learning cost to model the covariance matrix, DE implicitly learns the rotation direction through difference vectors without the computational overhead. The Success-History mechanism allowed DE to adaptively tune crossover rates to near 1.0, preserving rotational invariance.
Phase 3: Structural Hybridization & Complexity (2020–2024):
- Context: Benchmarks introduced highly complex Composition Functions (merging multiple landscapes with bias and weights) and shifted/rotated topologies to prevent "center-bias" shortcuts.
- Dominant Algorithms: Highly complex Hybrid Ensembles (e.g., IMODE, AGSK, EA4eigN100, L-SRTDE).
- Mechanism:
  - Heterogeneous Populations: Algorithms split populations into "Junior" (exploration) and "Senior" (exploitation) groups.
  - Eigenvector Crossover: Recent winners (e.g., EA4eigN100) integrate PCA/CMA-ES logic to explicitly learn the rotation of valleys, combining DE's efficiency with CMA-ES's covariance learning.
  - Success Rate Adaptation: The 2024 winner (L-SRTDE) shifted from history-based adaptation to Success Rate-based adaptation, using real-time success metrics to tune step sizes, offering better noise filtering.

B. The "Complexity Inflation" Critique

The paper notes a concerning trend where algorithms accumulate redundant components (ensembles of 4+ engines) to ensure ranking stability rather than genuine efficiency. Ablation studies suggest that simplified versions of these complex winners often perform better, indicating that current benchmarks may be overfitting to specific algorithmic "portfolios" rather than testing fundamental optimization power.

C. Implications for Variational Quantum Algorithms (VQAs)

The paper establishes a strong theoretical and empirical link between CEC benchmarks and VQA landscapes:

Structural Similarity: VQA landscapes exhibit non-separability (due to entanglement), multimodality (local minima), and noise (shot noise), mirroring the rotated, composition-based CEC benchmarks.
Proof-of-Concept Results: In the 10-qubit Ising VQE simulation:
- Gradient-based methods (L-BFGS-B, SPSA) failed catastrophically under noise, getting trapped in local minima or failing to converge due to gradient corruption.
- Modern CEC Variants (L-SHADE, IMODE, L-SRTDE) successfully navigated the landscape, finding the global ground state ( $E = -9.0$ ) even under noisy conditions.
- Topological Analysis: Persistence homology revealed massive energy barriers separating local and global minima. Only the spatial diversity and non-local search mechanisms of modern DE variants could bypass these barriers.

4. Results

Algorithmic Dominance: Differential Evolution variants (specifically the SHADE lineage) have dominated CEC competitions since 2014, accounting for the vast majority of top rankings.
Mechanism Evolution: The field moved from static heuristics to adaptive mechanisms (Success-History) and finally to structural hybridization (combining DE with covariance learning and societal sharing).
Quantum Transferability: The study demonstrates that CEC-evolved solvers are uniquely suited for VQAs. They handle the "Barren Plateau" problem (flat landscapes) and shot noise better than traditional gradient-free methods because they rely on population diversity and rotationally invariant operators rather than local gradient estimation.

5. Significance

For Evolutionary Computation: The paper provides a critical retrospective on the field, highlighting that the dominance of DE is a direct result of benchmark design (rotation matrices) rather than inherent universal superiority. It warns against "complexity inflation" and calls for more rigorous benchmarking to distinguish genuine innovation from ensemble stacking.
For Quantum Computing: This is a pivotal contribution to the Quantum Machine Learning (QML) and VQA fields. It suggests that the classical optimizer is the bottleneck in achieving Quantum Advantage. By adopting the advanced, noise-robust, and rotationally invariant solvers developed for CEC competitions (like L-SHADE and its hybrids), researchers can significantly improve the training of quantum circuits, potentially overcoming Barren Plateaus and hardware noise that currently limit NISQ (Noisy Intermediate-Scale Quantum) devices.

In conclusion, the paper argues that the "arms race" between benchmark designers and algorithm developers in the classical domain has inadvertently produced the exact class of optimizers needed to unlock the potential of near-term quantum hardware.