🔊 AcoustoSolver

Real-Time Acoustic Metamaterial Simulator with Physics-Informed Neural Networks

AcoustoSolver is a Python framework for designing, simulating, and optimizing acoustic metamaterials. It combines rigorous computational physics (Plane Wave Expansion, FDTD) with physics-informed neural surrogates for real-time band structure prediction and interactive design exploration.

Features • Quick Start • App • Architecture • Theory • Citation

---

✨ Features

Feature	Description
📊 Band Structure Solver	Plane Wave Expansion (PWE) method for 2D phononic crystals with automatic bandgap detection
🔗 SSH Model	Su-Schrieffer-Heeger chain with Zak phase computation for topological invariant classification
🌊 FDTD Simulator	2D Finite-Difference Time-Domain wave propagation with PML boundaries
🧠 Neural Surrogate	Residual MLP with physics-informed loss for real-time band prediction (100× faster than PWE)
🧱 Material Database	12+ built-in materials including GST phase-change materials, piezoelectrics, and polymers
🖥️ Interactive App	Streamlit-based GUI for real-time design exploration
✅ Test Suite	35+ unit tests with CI/CD via GitHub Actions

🚀 Quick Start

Installation

# Clone the repository
git clone https://github.com/Samarjithbiswas/AcoustoSolver.git
cd AcoustoSolver

# Install with all dependencies
pip install -e ".[all]"

# Or install minimal (core only)
pip install -e .

Basic Usage

from acoustosolver.core.materials import MaterialDatabase
from acoustosolver.core.unit_cell import SquareCell
from acoustosolver.core.band_structure import BandStructureSolver

# Initialize material database
db = MaterialDatabase()

# Create a phononic crystal unit cell
cell = SquareCell(
    lattice_constant=0.01,              # 10 mm
    host_material=db.get("epoxy"),      # Polymer matrix
    inclusion_material=db.get("steel"), # Steel inclusions
    filling_fraction=0.30,              # 30% filling
)

# Compute band structure
solver = BandStructureSolver(cell, n_bands=6, n_kpoints=100)
result = solver.solve()

# Analyze bandgaps
print(f"Found {len(result.bandgaps)} bandgaps:")
for gap in result.bandgaps:
    print(f"  {gap.lower_freq:.0f} - {gap.upper_freq:.0f} Hz "
          f"(relative width: {gap.relative_width:.1%})")

SSH Model — Topological Phase Analysis

# Compute SSH dispersion with topological phase
result = solver.solve_ssh_1d(t1=1.0, t2=1.5, n_kpoints=200)
zak_phase = solver.compute_zak_phase(t1=1.0, t2=1.5)

print(f"Zak phase: {zak_phase:.4f} rad")
print(f"Topological phase: {'YES (π)' if zak_phase > 1.5 else 'NO (0)'}")
print(f"Edge states: {'Protected' if zak_phase > 1.5 else 'None'}")

Wave Propagation

from acoustosolver.core.wave_propagation import (
    WavePropagationFDTD, SimulationConfig, SourceType
)

config = SimulationConfig(
    grid_size=(200, 200),
    domain_size=(0.5, 0.5),
    total_time=0.003,
    source_frequency=5000.0,
    source_type=SourceType.POINT,
)

sim = WavePropagationFDTD(config)
sim.create_phononic_crystal(n_cells_x=5, n_cells_y=5)
snapshots = sim.run(save_every=50)

Neural Surrogate Training

from acoustosolver.ml.surrogate_model import BandSurrogate
from acoustosolver.ml.dataset import DatasetGenerator, DatasetConfig

# Generate training data
gen = DatasetGenerator(DatasetConfig(n_samples=1000))
X, y = gen.generate()

# Train surrogate model
surrogate = BandSurrogate(n_bands=6, n_kpoints=50)
print(surrogate.summary())
history = surrogate.train(X, y, epochs=200)

# Real-time inference (100× faster than PWE)
predictions = surrogate.predict(X[:5])

🖥️ Interactive App

Launch the Streamlit-powered interactive simulator:

streamlit run app.py

App Features:

• 🔧 Unit cell parameter sliders (lattice constant, filling fraction, materials)
• 📊 Real-time band structure computation with interactive Plotly charts
• 🔗 SSH model with Zak phase classification (topological vs. trivial)
• 🌊 FDTD wave propagation with snapshot slider
• 🧱 12+ material presets with full property display

📐 Architecture

AcoustoSolver/
├── acoustosolver/
│   ├── core/                    # Computational engines
│   │   ├── materials.py         # Material database (12+ materials)
│   │   ├── unit_cell.py         # Square & hexagonal lattice cells
│   │   ├── band_structure.py    # PWE solver + SSH model + Zak phase
│   │   └── wave_propagation.py  # 2D FDTD with PML boundaries
│   ├── ml/                      # Machine learning
│   │   ├── surrogate_model.py   # Residual MLP + physics-informed loss
│   │   └── dataset.py           # Synthetic dataset generation
│   ├── viz/                     # Visualization
│   │   └── plotting.py          # matplotlib + plotly figures
│   └── utils/
│       └── physics_constants.py # Physical constants & conversions
├── app.py                       # Streamlit interactive app
├── tests/
│   └── test_core.py             # 35+ unit tests
├── examples/
│   └── quick_start.py           # End-to-end demonstration
├── .github/workflows/ci.yml     # GitHub Actions CI/CD
├── pyproject.toml               # Modern Python packaging
└── README.md

📖 Theory

Plane Wave Expansion (PWE)

The acoustic wave equation in a periodic medium:

$$\nabla \cdot \left[\frac{1}{\rho(\mathbf{r})} \nabla p(\mathbf{r})\right] = -\frac{\omega^2}{\kappa(\mathbf{r})} p(\mathbf{r})$$

is solved by expanding pressure and material properties in Fourier series, yielding an eigenvalue problem at each wave vector k.

SSH Model

The Su-Schrieffer-Heeger (SSH) model with alternating couplings $t_1$ and $t_2$:

$$E(k) = \pm|t_1 + t_2 e^{ika}|$$

The Zak phase (Berry phase across the Brillouin zone) is the topological invariant:

• $\gamma = 0$ → Trivial phase ($t_1 > t_2$)
• $\gamma = \pi$ → Topological phase ($t_2 > t_1$), supports protected edge states

Physics-Informed Loss

The neural surrogate is trained with a composite loss:

$$\mathcal{L} = \mathcal{L}_{\text{MSE}} + \lambda_s \mathcal{L}_{\text{symmetry}} + \lambda_o \mathcal{L}_{\text{ordering}} + \lambda_r \mathcal{L}_{\text{smoothness}}$$

where:

• Symmetry: $\omega(\mathbf{k}) = \omega(-\mathbf{k})$ (time-reversal invariance)
• Ordering: $\omega_{n+1}(\mathbf{k}) \geq \omega_n(\mathbf{k})$ (no band crossings)
• Smoothness: Penalizes discontinuous frequency variations

🏗️ Built With

Technology	Purpose
NumPy / SciPy	Core numerical computation
PyTorch	Neural surrogate models
Matplotlib / Plotly	Static & interactive visualization
Streamlit	Web-based interactive app
pytest	Testing framework
GitHub Actions	CI/CD pipeline

📝 Citation

If you use AcoustoSolver in your research, please cite:

@software{biswas2026acoustosolver,
  author = {Biswas, Samarjith},
  title = {AcoustoSolver: Real-Time Acoustic Metamaterial Simulator with Physics-Informed Neural Networks},
  year = {2026},
  publisher = {GitHub},
  url = {https://github.com/Samarjithbiswas/AcoustoSolver}
}

Related Publications

• Biswas, S., Krawczyk, Z., & Manimala, J.M. "Multifunctional Characterization of a Thermoacoustic Meta-Structure." Int. J. Thermofluids (2025).
• Biswas, S., Kresl, W., & Manimala, J.M. "On the relationship between acoustic absorption and temperature gradient in a thermoacoustic liner." Int. J. Aeroacoustics, 24(1-2), 42-67 (2025).
• Manimala, J. & Biswas, S. "Thermoacoustic Meta-Structure." US Patent WO 2025/128,348 A1 (2025).

👤 Author

Samarjith Biswas, PhD
Research Scientist III · University of Arizona
New Frontiers of Sound (NewFoS) Science & Technology Center

• 🌐 samarjithbiswas.com
• 🎓 Google Scholar
• 📺 YouTube
• 💼 LinkedIn

📄 License

This project is licensed under the MIT License — see LICENSE for details.