💻 computer science

Level set-based inverse homogenisation of three-dimensional piezoelectric materials

This paper presents an open-source, memory-distributed level set-based topology optimization framework for designing three-dimensional piezoelectric metamaterials with enhanced stiffness and piezoelectric properties, identifying an approximate Schur complement preconditioned GMRES solver as the most scalable method for solving the underlying homogenization equations.

Original authors: Zachary J. Wegert, Anthony P. Roberts, Vivien J. Challis

Published 2026-01-26

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Zachary J. Wegert, Anthony P. Roberts, Vivien J. Challis

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 an architect trying to design a new kind of building material. But instead of bricks and mortar, you are designing a "smart" material that can turn squeezing it into electricity, or using electricity to make it move. This is the world of piezoelectric materials, used in things like sensors, underwater microphones (hydrophones), and actuators.

The problem is that the natural materials we have today are just "okay." They work, but they aren't very strong or efficient. The authors of this paper wanted to invent a metamaterial—a material whose properties come not from what it is made of, but from its internal shape. Think of it like the difference between a solid block of wood and a complex honeycomb structure; the honeycomb is lighter but can be just as strong, or even stronger in specific ways.

Here is how they did it, explained simply:

1. The "Shape Shifter" Tool (Level-Set Method)

To design these materials, the team used a computer program that acts like a digital sculptor. They call this the Level-Set Method.

Imagine you have a block of clay. In older computer methods, the sculptor would try to turn the clay into a solid shape by slowly turning the "mud" in the middle into either "solid rock" or "empty air." This often left messy, half-solid, half-air regions that didn't make sense in the real world.

The new method used in this paper is like a magical boundary line. It draws a perfect, smooth line between the "solid" part and the "empty" part. It doesn't leave any messy middle ground. This ensures the final design is a clean, manufacturable structure with clear solid parts and clear holes.

2. The "Super-Computer" Challenge

Designing these shapes in 3D is incredibly hard. It's like trying to solve a giant, 3D jigsaw puzzle where every piece changes the picture, and you have to do it millions of times to find the perfect one.

If you tried to do this on a single laptop, it would take years. The authors realized they needed to split the work up, like a team of chefs in a massive kitchen, each chopping vegetables on their own counter but working on the same giant soup. They used a distributed computing approach, spreading the math across hundreds of computer processors at once.

They tested different ways to get the computers to talk to each other efficiently. They found that a specific mathematical trick called the "Approximate Schur Complement" was the best way to speed things up. It's like finding the secret shortcut in a maze that lets the whole team finish the puzzle in a fraction of the time.

3. The Two Design Challenges

The team tackled two specific goals, like two different design contests:

Contest A: The "Stiff but Sensitive" Material.
They wanted a material that was very stiff (hard to squish) but also very good at turning pressure into electricity. They asked the computer: "Make the stiffest possible material, but it must also have this specific amount of electrical sensitivity."
- The Result: They found some amazing shapes. Some looked like thick, sturdy plates connected by thin pillars. These designs were 9 times better at sensing pressure than the base material they started with.
Contest B: The "Super-Sensitive" Material.
They wanted a material that was the most sensitive to pressure possible, but it had to be strong enough in a specific direction to hold up.
- The Result: They found structures that were incredibly efficient at converting movement to electricity, far better than anything previously designed in 3D.

4. The "Goldilocks" Discovery

In their search, they found something interesting about the shapes.

When they pushed for the absolute maximum sensitivity, the computer designed structures with very thin, delicate threads. While these worked great on paper, they would be very hard to manufacture in real life (like trying to build a bridge out of spider silk).
However, they also found a "Goldilocks" zone. They identified two specific designs that had no tiny, fragile threads. These structures were robust, easy to imagine manufacturing, and still offered massive improvements (several times better) over standard materials.

5. What They Didn't Find (The Reality Check)

The paper is very careful to say what they didn't do. They didn't build a physical prototype in a lab yet. They didn't test these materials in a real underwater hydrophone or a medical device. They didn't claim these materials will cure diseases or power cities.

Their claim is strictly about computational design: "We used math and supercomputers to prove that these specific shapes should work better than anything we have today, and we found a way to calculate them fast enough to be useful."

Summary

Think of this paper as the blueprint phase of a revolutionary new building. The authors didn't pour the concrete, but they used a powerful new digital tool to draw up plans for a structure that is stronger and smarter than anything currently built. They also figured out how to get a team of computers to draw these plans quickly, and they highlighted two specific designs that are strong, smart, and actually possible to build without breaking.

Technical Summary: Level Set-Based Inverse Homogenisation of Three-Dimensional Piezoelectric Materials

Problem Statement
This work addresses the computational design of three-dimensional periodic piezoelectric metamaterials with enhanced macroscopic properties. While recent additive manufacturing techniques have enabled the creation of heterogeneous piezoelectric structures, optimizing their microstructures remains challenging. Previous density-based topology optimisation methods often suffer from the generation of non-physical intermediate densities, which can artificially inflate objective functions. Furthermore, existing 3D optimisation studies have been limited by low mesh resolutions due to the computational intractability of solving piezoelectric homogenisation equations in serial implementations. The authors aim to overcome these limitations by developing a high-resolution, memory-distributed level-set topology optimisation framework capable of solving inverse homogenisation problems for piezoelectric materials.

Methodology
The paper employs a level-set-based topology optimisation approach to implicitly track the material boundary, thereby avoiding large regions of intermediate density. The evolution of the boundary is governed by a Hamilton-Jacobi equation, where the normal velocity is determined using a Hilbertian projection method to handle constraints.

Key methodological components include:

Numerical Implementation: The optimisation is implemented in the Julia programming language using the GridapTopOpt package, which facilitates memory-distributed computing across multiple processors via MPI. This allows for the solution of problems with millions of degrees of freedom.
Piezoelectric Homogenisation: The authors solve the linear piezoelectric state equations over a representative volume element to compute effective material tensors (stiffness, piezoelectric coupling, and dielectric permittivity).
Linear Solvers: A critical contribution of the methodology is the comparative analysis of iterative solvers for the resulting block-structured linear systems (which are either symmetric indefinite or non-symmetric positive definite). The authors evaluate a diagonal block preconditioned GMRES (DBP-GMRES), a tridiagonal conjugate gradient (TriCG), and an approximate Schur complement preconditioned GMRES (SCP-GMRES).
Optimisation Formulations: Two distinct inverse homogenisation problems are formulated:
1. Maximising the effective Voigt or Reuss bulk modulus subject to constraints on volume fraction and the effective hydrostatic strain coefficient ( $\bar{d}_h$ ).
2. Maximising the effective hydrostatic strain coefficient subject to a constraint on the effective stiffness in the poling direction ( $\bar{C}_{zzzz}$ ).
Material Model: The base material is defined as z-poled PZT-5A.

Key Contributions

Solver Performance: The study identifies the approximate Schur complement preconditioned GMRES (SCP-GMRES) as the most scalable and efficient solver for piezoelectric homogenisation equations. It demonstrates superior performance by reducing global iteration counts by approximately half compared to other methods and showing insensitivity to the relative tolerance of inner solvers.
High-Resolution 3D Optimisation: By leveraging distributed computing, the authors achieve computational resolutions far exceeding previous literature, enabling the design of high-fidelity 3D piezoelectric metamaterials.
Level-Set Application: This work represents the first application of level-set-based methods to piezoelectric microstructure optimisation, successfully eliminating the issue of intermediate density regions that plagues density-based approaches.
Open Source Framework: The authors provide an open-source implementation of their memory-distributed level-set framework for the community.

Results
The authors present optimised unit cells for both optimisation problems:

Bulk Modulus Maximisation: The study maps the trade-off between bulk modulus and hydrostatic strain coefficient. It reveals that the parameter space is not convex, contrary to previous hypotheses. Optimising for the Reuss bulk modulus yields microstructures with thicker, more robust members in the poling direction compared to those optimising for the Voigt bulk modulus, which often feature thin, fragile connections.
Hydrostatic Strain Coefficient Maximisation: Structures designed to maximise $\bar{d}_h$ under stiffness constraints exhibit complex geometric features and high electromechanical coupling.
Performance Metrics: The authors highlight two specific robust structures (free of fine-scale features) that exhibit piezoelectric properties several times larger than the base material. Specifically, these structures achieve approximately a 9-fold increase in the hydrostatic strain coefficient ( $d_h$ ), a 43–50 fold increase in the voltage constant ( $g_h$ ), and a 7-fold increase in the electromechanical coupling factor ( $k_h$ ), while maintaining a 50% volume fraction.
Boundary Effects: The authors note that boundary smoothing introduces errors, particularly in calculating the Reuss bulk modulus for structures with fine-scale features, suggesting a need for unfitted finite element methods (CutFEM) in future work.

Significance and Claims
The paper claims that level-set-based topology optimisation is well-suited for piezoelectric problems due to its ability to avoid non-physical intermediate densities. The developed techniques allow for the computational design of high-resolution metamaterials that offer new insights into material design for sensor, hydrophone, and actuator applications. The authors suggest that the identified robust structures, which lack fine-scale features yet exhibit significantly enhanced properties, are promising candidates for practical engineering applications. The work emphasizes that while extreme properties are achievable, manufacturability constraints (such as avoiding thin members) must be considered in future designs. The paper concludes that the proposed memory-distributed implementation and solver selection are essential for tackling the computational complexity of 3D piezoelectric inverse homogenisation.