Abstract:Machine learning-based inverse materials discovery has attracted enormous attention recently due to its flexibility in dealing with black box models. Yet, many metaheuristic algorithms are not as widely applied to materials discovery applications as machine learning methods. There are ongoing challenges in applying different optimization algorithms to discover materials with single- or multi-elemental compositions and how these algorithms differ in mining the ideal materials. We comprehensively compare 11 different optimization algorithms for the design of single- and multi-elemental crystals with targeted properties. By maximizing the bulk modulus and minimizing the Fermi energy through perturbing the parameterized elemental composition representations, we estimated the unique counts of elemental compositions, mean density scan of the objectives space, mean objectives, and frequency distributed over the materials’ representations and objectives. We found that nature-inspired algorithms contain more uncertainties in the defined elemental composition design tasks, which correspond to their dependency on multiple hyperparameters. Runge–Kutta optimization (RUN) exhibits higher mean objectives, whereas Bayesian optimization (BO) displayed low mean objectives compared with other methods. Combined with materials count and density scan, we propose that BO strives to approximate a more accurate surrogate of the design space by sampling more elemental compositions and hence have lower mean objectives, yet RUN will repeatedly sample the targeted elemental compositions with higher objective values. Our work sheds light on the automated digital design of materials with single- and multi-elemental compositions and is expected to elicit future studies on materials optimization, such as composite and alloy design based on specific desired properties.
materials science, multidisciplinary,physics, applied,nanoscience & nanotechnology
What problem does this paper attempt to address?
### What problems does this paper attempt to solve?
This paper aims to explore the performance differences of various optimization algorithms in material design through benchmarking. Specifically, the authors hope to solve the following key problems:
1. **Selection and application of optimization algorithms**:
- In material design, choosing the appropriate optimization algorithm is crucial. Different types of optimization algorithms (such as machine - learning - based optimization and meta - heuristic optimization) exhibit different performances when dealing with multi - element material design. Therefore, researchers need to understand which algorithm is most suitable for a specific application scenario.
- The paper evaluates the performance of 11 different optimization algorithms, including Bayesian Optimization (BO), Genetic Algorithm (GA), Particle Swarm Optimization (PSO), etc., in finding single - element or multi - element crystals with target properties by comparing them.
2. **Performance evaluation of optimization algorithms**:
- The authors hope to evaluate the performance of these optimization algorithms in material design through systematic benchmarking. In particular, they focus on how to maximize the bulk modulus and minimize the Fermi energy, which are two important physical properties.
- Through quantitative analysis of these algorithms, researchers can better understand the advantages and limitations of each algorithm, thus providing guidance for future research.
3. **Construction of an automated material design framework**:
- The researchers have developed a flexible optimization framework that can automate the implementation of multiple advanced optimization methods. This framework not only simplifies the optimization process but also improves the efficiency of material design.
- By combining Graph Neural Network (GNN) surrogate models and optimization algorithms, researchers can quickly predict material properties and optimize without relying on traditional numerical methods (such as molecular dynamics or density functional theory).
4. **Exploration of new methods for multi - element material design**:
- The focus of the research is on how to effectively design multi - element materials. Traditional experimental synthesis and trial - and - error methods are both expensive and time - consuming, while the progress of computational engineering makes material design more efficient.
- By parameterizing the graph structure and adjusting design variables (such as the number of elements, atom types, etc.), researchers can explore a wider design space and find material combinations with ideal properties.
### Formula representation
To describe the optimization problem more clearly, the authors define an objective function \( \mathcal{J} \) to measure the bulk modulus \( K \) and Fermi energy \( E_{\text{Fermi}} \) of the material:
\[ \mathcal{J}=K - E_{\text{Fermi}} \]
The optimization problem can be expressed as:
\[ \arg\max_{n_{\text{atom}}, \xi_n, \eta} \mathcal{J}=K - E_{\text{Fermi}} \]
where,
\[ K, E_{\text{Fermi}}=\text{MEGNet}(\mathcal{G}_{\Theta}) \]
\[ \mathcal{G}=\Omega(n_{\text{atom}}, \xi_n, \eta) \]
\[ \Theta = [n_{\text{atom}}, \xi_n, \eta] \]
The constraints are:
\[ n_{\text{atom}}\in[1, 4]\quad\text{or}\quad\equiv 1 \]
\[ \xi_n\in[0, 100] \]
\[ \eta\in[0, 100] \]
### Conclusion
Through this research, the authors hope to provide valuable references for researchers in the field of materials science, helping them choose the most appropriate optimization algorithm when designing new materials. In addition, they also propose a general framework that can be applied.