Fuzzy Neural Network Performance and Interpretability of Quantum Wavefunction Probability Predictions
Pedro H. M. Zanineli, Matheus Zaia Monteiro, Vinicius Francisco Wasques, Francielle Santo Pedro Simões, Gabriel R. Schleder · 2025
Read Fuzzy Neural Network Performance and Interpretability of Quantum Wavefunction Probability Predictions by Pedro H. M. Zanineli, Matheus Zaia Monteiro, Vinicius Francisco Wasques, Francielle Santo Pedro Simões, Gabriel R. Schleder (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2511-05261.txt.
Introduction
Pedro Zanineli and co-authors weigh accuracy against clarity in models that predict quantum wave probability maps, a core need in chemistry and materials work. They compare a plain neural network with a fuzzy neural system on data for a hydrogen molecular ion made by a physics-based net.
The plain network scored higher on fit, but used more than fifty times as many parameters. The fuzzy system, with Gaussian membership rules, stayed close in score and stays far easier to read. The trade-off matters when a model's reasons must be checked, not just its numbers.
Catalog record
| CREATOR | Pedro H. M. Zanineli, Matheus Zaia Monteiro, Vinicius Francisco Wasques, Francielle Santo Pedro Simões, Gabriel R. Schleder |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 31 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2511-05261
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Fuzzy Neural Network Performance and Interpretability of Quantum Wavefunction Probability Predictions Pedro H. M. Zanineli, Matheus Zaia Monteiro, Vinicius Francisco Wasques, Francielle Santo Pedro Simões, Gabriel R. Schleder Abstract Predicting quantum wavefunction probability distributions is crucial for computational chemistry and materials science, yet machine learning (ML) models often face a trade-off between accuracy and interpretability. This study compares Artificial Neural Networks (ANNs) and Adaptive Neuro-Fuzzy Inference Systems (ANFIS) in modeling quantum probability distributions for the H + 2 {}_{2}^{+} ion, leveraging data generated via Physics-Informed Neural Networks (PINNs). While ANN achieved superior accuracy (R 2 = 0.99 vs. ANFIS’s 0.95 with Gaussian membership functions), it required over 50 × \times more parameters (2,305 vs. 39–45). ANFIS, however, provided unique interpretability: its Gaussian membership functions encoded spatial electron localization near proton positions ( μ = ± 1.2 Å \mu=\pm 1.2\ \text{\AA} ), mirroring Born probability densities, while fuzzy rules reflected quantum superposition principles. Rules prioritizing the internuclear direction revealed the system’s 1D symmetry, aligning with Linear Combination of Atomic Orbitals theory—a novel data-driven perspective on orbital hybridization. Membership function variances ( σ \sigma ) further quantified electron delocalization trends, and peak prediction errors highlighted unresolved quantum cusps. The choice of functions critically impacted performance: Gaussian/Generalized Bell outperformed Sigmoid, with errors improving as training data increased, underscoring scalability. This study underscores the context-dependent value of ML: ANN for precision and ANFIS for interpretable, parameter-efficient approximations that link inputs to physical behavior. These findings advocate hybrid approaches in quantum simulations, balancing accuracy with explainability to accelerate discovery. Future work should extend ANFIS to multi-electron systems and integrate domain-specific constraints (e.g., kinetic energy terms), bridging data-driven models and fundamental physics. I Introduction In the framework of quantum mechanics, the wavefunction \Psi represents a fundamental concept, encapsulating the state of an isolated quantum system. This function assigns a complex value to every point in space, generating a probability amplitude that, when interpreted through the Born rule, translates into actual charge densities or probability distributions 5 . A crucial property of \Psi is that its squared modulus, |\Psi|^{2} , provides the probability density of finding a particle at a specific location. A well-known example is the hydrogen atom, where a single-electron wavefunction describes the likelihood of detecting the electron within a given spatial region 5 . Importantly, determining the wavefunction of a system does not only offer insight into spatial probability distributions but also provides a comprehensive description, including observables quantities such as energy and momentum obtained via their corresponding operators. The wavefunction is obtained by solving the Schrödinger equation: \hat{H}\Psi=E\Psi, where the Hamiltonian operator \hat{H}=\left[-\frac{\hbar^{2}}{2m}\nabla^{2}+V(\mathbf{r})\right] dictates the system’s total energy, with its eigenvalues corresponding to possible energy levels and eigenvectors representing the associated quantum states. However, solving this equation is a significant challenge, particularly for many-body or complex systems 11 ; 20 . Analytical solutions exist solely for a limited number of scenarios, including a free particle, the particle in a box, the quantum harmonic oscillator, and the hydrogen atom 5 ; 4 . Thus, different approaches of machine learning (ML) have been used for finding the equation’s solution, such as neural networks 6 , generative models 7 , genetic algorithms 21 , and reinforcement learning 18 . Due to these limitations, alternative approaches such as Physics Informed Neural Networks (PINNs) have been explored to enhance our ability to find approximate solutions for more intricate quantum systems 10 ; 9 ; 16 . In particular, PINNs have been utilized to determine the solutions for the hydrogen molecular ion \text{H}_{2}^{+} 15 . By leveraging fundamental physical principles, this method identifies parametric eigenvalues and eigenfunction surfaces of quantum systems. Notably, this approach is capable of enabling ab initio calculations for simple molecular structures. Considering the importance of uncovering the probability density of a wavefunction, creating a machine learning model from the PINN-generated data can bring advantages for different reasons. Firstly, knowing that solving the Schrödinger equation is computationally costly, using ML would be interesting for finding an approximate solution with less time 14 . Secondly, if the ML is well-suited, it can be generalized for different systems, which can be used for different molecules and materials. Finally, according to the chosen model, it can also be important from the perspective of interpretability and physical analysis. In this last case, while on one hand classical Artificial Neural Networks (ANN) are known for hindering the result’s interpretability 3 , on the other hand, Adaptive Neuro-Fuzzy Inference System (ANFIS) might be suitable for better comprehending the pattern found from the neural network 12 ; 2 . Yet, it is fundamental to find a balance between the model’s efficiency and interpretability. Thus, in the light of the exposed, the present work aims to establish a comparison between classical Artificial Neural Networks and the Adaptive Neuro-Fuzzy Inference System in the wavefunction prediction of a dihydrogen ion. II Some preliminaries This section provides some mathematical tools to better understand how ANFIS works. II.1 Fuzzy Sets and Fuzzy Logic Fuzzy sets extend the concept of classical sets by introducing a membership function \varphi:U\to[0,1] , which assigns each element of the universe U to a value within the interval [0,1] . In other words, considering a subset A in the universe U , the value \varphi_{A}(x) means the degree of association of the element x to the set A , in the following sense, the higher the degree, the higher is the association. This generalization enables the representation of vague or imprecise relationships that are common in natural language but lack precise mathematical definitions. Fuzzy sets address this issue by allowing elements to belong to a set with varying degrees of membership, rather than the strict binary classification of classical sets 1 . Fuzzy logic expands classical logic by redefining logical operators such as conjunctions and disjunctions through t -norms and t -conorms (or also called s -norms), respectively. A t -norm is a function t:[0,1]^{2}\to[0,1] that satisfies the properties of commutativity, monotonicity, and associativity. Examples of t -norms include the minimum and the algebraic product functions. Also, t -conorms satisfy similar properties, and they are connected by the following relation s(x,y)=1-t(1-x,1-y) , where t(\cdot,\cdot) represents a t -norm and s(\cdot,\cdot) a t -conorm. Common examples of t -conorms include the maximum operator and the algebraic sum 13 . Fuzzy logic introduces degrees of truth rather than the traditional binary true or false statements. Using fuzzy sets, fuzzy propositions take the form: “IF antecedent , THEN consequent ”, where both antecedent and consequent are represented by fuzzy sets. A typical fuzzy rule can be expressed as: “IF x_{1} is A_{1} AND x_{2} is A_{2} , THEN y is B ”, where A_{1} , A_{2} , and B are fuzzy sets defined over the respective domains of x_{1} , x_{2} , and y , and the AND operator is modeled by a t -norm. A fuzzy proposition can be interpreted as a function P:U_{1}\times U_{2}\to[0,1] , where the codomain represents the degree of truth of the proposition. II.2 Fuzzy Rule-Based Systems (FRBS) Fuzzy Rule-Based Systems (FRBSs) employ fuzzy propositions to replicate human decision-making. The system begins by mapping numerical inputs onto fuzzy sets in a process known as fuzzification. Thus, it applies an inference system