New RVE concept and FFT methods in micromechanics of composites subjected to body force with compact support

arXiv:2601.00822 · physics.comp-ph, cond-mat.mtrl-sci · Submitted 2025-12-23 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Physics. The day in general physics: instruments, methods and results across the field.

Kai: I'm Kai.

Mira: And I'm Mira, here with Lev.

Kai: We're looking at the paper: "New RVE concept and FFT methods in micromechanics of composites subjected to body force with compact support".

Mira: Here's the gist of the paper. New RVE concept and FFT methods in micromechanics of composites subjected to body force with compact support enables a new RVE concept that reduces the infinite periodic medium to a finite domain without boundary artifacts <ref:2601.00822#pg2>.

Kai: This functionally reduced RVE is used for translated averaging of direct numerical simulations (DNS) results, efficiently computed via a newly developed FFT-based solver for BFCS loading <ref:2601.00822#pg6>.

Mira: The resulting dataset captures localized field responses and is used to train machine learning (ML) and neural networks (NN) models to learn effective nonlocal surrogate operators <ref:2601.00822#pg5>.

Kai: By accounting for field localization while simultaneously eliminating influences from finite sample size and boundary effects, it provides a physically grounded and data-driven framework for constructing accurate surrogate models for the homogenization of complex materials <ref:2601.00822#pg5>.

Paper summary: Kai: So we talked about how they set up this whole process, using the new RVE concept and then feeding that data into machine learning models to learn these nonlocal operators. The main thesis is that this new framework allows them to bypass the computational cost of traditional methods.

Mira: They claim the core idea is shifting away from relying on a fixed constitutive behavior or a specific analytical form of surrogate operators, instead basing the RVE on field concentration factors within each phase to characterize it. That's a big conceptual shift for modeling composites.

Lev: So they are essentially saying that whatever material you put in this framework, the RVE setup is flexible enough to adapt to whatever nonlocal operator you need to predict, which accelerates the modeling process.

Kai: They achieve this flexibility by using body force fields with compact support, which effectively reduces the infinite periodic medium down to a finite domain without introducing those boundary artifacts that usually plague these simulations.

Mira: And they use this finite domain to do translated averaging of direct numerical simulations, and they make that simulation fast by employing an FFT-based solver tailored for BFCS loading. That’s where the computational efficiency comes from.

Lev: The paper points out that this FFT-based technique lets them reduce complexity from something like O of n squared down to O of n log n via spectral solvers leveraging the discrete Fourier transform. That's a measurable speedup.

Kai: And because they get this dataset, Dp(three point seven) or Dd(three point eight), they can train neural networks and machine learning models to create these effective nonlocal surrogate operators that predict the material response under any macroscopic loading conditions we might encounter in the real world.

Mira: The implication is that instead of running a full, expensive simulation every time we want to know how a composite reacts nonlocally, we use these learned models which are much faster and capture the necessary microstructural details.

Lev: It's about making it practical for things like real hardware testing where you need fast feedback loops. The overhead landed by using these pre-trained surrogate forms is what matters for running things at scale.

Kai: So to wrap up this summary, the paper is proposing a new RVE concept and FFT methods in micromechanics of composites subjected to body force with compact support as a way to generate data that trains machine learning models for effective nonlocal surrogate operators.

Mira: And it’s important because it provides a physically grounded and data-driven framework for constructing these accurate surrogate models for the homogenization of complex materials.

Lev: That’s the big picture—moving from computationally expensive nested finite element discretizations to faster spectral solvers fed by localized simulation data.

Conclusion: Kai: So looking at the whole paper, "New RVE concept and FFT methods in micromechanics of composites subjected to body force with compact support," what we've seen is a sophisticated method for generating simulation data that directly feeds into machine learning for material prediction.

Mira: They’ve established this new RVE concept and the use of FFT methods with BFCS loading to create a finite domain from an infinite periodic medium, which then gets translated into averaged direct numerical simulations.

Lev: The key takeaway for me, as someone focused on hardware, is that they are creating surrogate forms Dr or Dd which can be compressed further into De p and De d to cut down the computational cost significantly while preserving the essential micromechanical detail.

Kai: It’s about making the process more efficient, moving away from just running massive direct numerical simulations every time we need a result, to using these learned models for prediction.

Mira: The implication is that this data-driven framework can build accurate surrogate models for complex materials without needing the exact analytical form of those operators beforehand. It lets the microstructure speak for itself in training the machine learning part.

Lev: They do mention that the method relies on generating datasets from these specific body force fields, which means if you need to test a completely different type of loading scenario not covered by those BFCS fields, you'd need to generate a new dataset.

Kai: That’s the limitation they state—it’s highly dependent on the type of loading used for generating that initial dataset.

Mira: So while it offers a powerful path to creating these nonlocal surrogate operators, we have to be mindful that it requires starting with this specific data generation methodology.

Lev: That's fair. It’s a very focused approach, and it sets the stage for future work where they might try to expand the scope beyond those compact support fields.

Kai: So in short, this paper provides a new way to create physically grounded, data-driven surrogate models for composite materials using advanced simulation techniques and machine learning.

physics.comp-ph, cond-mat.mtrl-sci

Submitted: 2025-12-23

Updated: 2025-12-23

Comments: Submitted for publication. 33pp., 98 refs

Journal ref: Math. Mech. of Solids, v. 31, 2026

DOI: 10.1177/10812865261421526

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 81/100

The gist: New RVE concept and FFT methods in micromechanics of composites subjected to body force with compact support enables a new RVE concept that reduces the infinite periodic medium to a finite domain…

Key concepts

New RVE Concept
A novel approach where the RVE is defined by field concentration factors within each phase rather than relying on the constitutive behavior of individual phases or analytical surrogate forms. It leverages specifically designed body force fields with compact support to characterize microstructure effectively.
Body Force Fields with Compact Support (BFCS)
These are specially designed body force fields used to precisely control the excitation within a material sample. Applying BFCS allows the infinite periodic microstructure to be effectively reduced to a finite computational domain without introducing boundary artifacts or sacrificing microstructural response fidelity.
FFT-Based Solver Adaptation
A newly developed solver utilizing Fast Fourier Transform (FFT) methods is employed for efficient simulation of materials under BFCS loading. This adaptation enables fast, scalable, and accurate computation of local fields, which is crucial for generating the necessary datasets for machine learning training.

Terminology

Summary

New RVE concept and FFT methods in micromechanics of composites subjected to body force with compact support enables a new RVE concept that reduces the infinite periodic medium to a finite domain without boundary artifacts <ref:2601. p. 0

Where the paper says it

The core of the proposed methodology is the generation of a novel dataset using specially designed body force fields with compact support (BFCS), enabling a new RVE concept that reduces the infinite periodic medium to a finite domain without boundary artifacts. This functionally reduced RVE is used for translated averaging of direct numerical simulations (DNS) results, efficiently computed via a newly developed FFT-based solver for BFCS loading.

00822#pg2>. This functionally reduced RVE is used for translated averaging of direct numerical simulations (DNS) results, efficiently computed via a newly developed FFT-based solver for BFCS loading <ref:2601. p. 0
Where the paper says it

The core of the proposed methodology is the generation of a novel dataset using specially designed body force fields with compact support (BFCS), enabling a new RVE concept that reduces the infinite periodic medium to a finite domain without boundary artifacts. This functionally reduced RVE is used for translated averaging of direct numerical simulations (DNS) results, efficiently computed via a newly developed FFT-based solver for BFCS loading.

00822#pg6>. The resulting dataset captures localized field responses and is used to train machine learning (ML) and neural networks (NN) models to learn effective nonlocal surrogate operators <ref:2601. p. 0
Where the paper says it

This functionally reduced RVE is used for translated averaging of direct numerical simulations (DNS) results, efficiently computed via a newly developed FFT-based solver for BFCS loading. The resulting dataset captures localized field responses and is used to train machine learning (ML) and neural networks (NN) models to learn effective nonlocal surrogate operators.

00822#pg5>. By accounting for field localization while simultaneously eliminating influences from finite sample size and boundary effects, it provides a physically grounded and data-driven framework for constructing accurate surrogate models for the homogenization of complex materials <ref:2601. p. 0
Where the paper says it

These operators accurately predict macroscopic responses while reflecting microstructural features and nonlocal interactions. By accounting for field localization while simultaneously eliminating influences from finite sample size and boundary effects, it provides a physically grounded and data-driven framework for constructing accurate surrogate models for the homogenization of complex materials.

00822#pg5>.

Novel RVE Framework

The paper introduces a novel Representative Volume Element (RVE) concept that is independent of the constitutive behavior of individual phases and the analytical form of surrogate operators, instead leveraging field concentration factors within each phase to characterize the microstructure effectively <ref:2601.00822#pg6>. This new RVE concept is derived through the application of specifically designed body force fields with compact support (BFCS), which allows for precise control of excitation within the material sample <ref:2601. p. 2

Where the paper says it

At the core of the methodology is the generation of a new type of dataset, derived through the application of specifically designed body force fields with compact support. These body forces are spatially localized, allowing for precise control of excitation within the material sample.

00822#pg6>. The resulting datasets, enriched by this micromechanically informed RVE, should be compatible with any ML or NN architecture for predicting nonlocal surrogate operators <ref:2601. p. 2
Where the paper says it

Instead, it leverages field concentration factors within each phase to characterize the microstructure effectively. The resulting datasets, enriched by this micromechanically informed RVE, should be compatible with any ML or NN architecture for predicting nonlocal surrogate operators.

00822#pg6>.

Data Generation Methodology

The generation of the new type of dataset is derived through the application of specifically designed body force fields with compact support <ref:2601.00822#pg6>. This feature enables the formulation of a new RVE concept, in which the infinite periodic microstructure is effectively reduced to a finite computational domain, without sacrificing the fidelity of microstructural response or introducing boundary-related artifacts <ref:2601. p. 2

Where the paper says it

These body forces are spatially localized, allowing for precise control of excitation within the material sample. This feature enables the formulation of a new RVE concept, in which the infinite periodic microstructure is effectively reduced to a finite computational domain, without sacrificing the fidelity of microstructural response or introducing boundary-related artifacts.

00822#pg6>. The dataset is built through translated averaging of direct numerical simulations (DNS) performed within this finite domain <ref:2601. p. 2
Where the paper says it

This feature enables the formulation of a new RVE concept, in which the infinite periodic microstructure is effectively reduced to a finite computational domain, without sacrificing the fidelity of microstructural response or introducing boundary-related artifacts. This RVE reduction is not merely geometric, but functional: the dataset is built through translated averaging of direct numerical simulations (DNS) performed within this finite domain.

00822#pg6>.

FFT-Based Solver Adaptation

A newly developed FFT-based solver tailored for CMs is employed to enable efficient simulation, leading to fast, scalable, and accurate computation of the local fields <ref:2601.00822#pg6>. This FFT solver is adapted to handle the response of materials subjected to body forces with compact support (BFCS), leading to fast, scalable, and accurate computation of the local fields <ref:2601. p. 2

Where the paper says it

To enable efficient simulation, a newly developed FFT-based solver tailored for CMs is employed. This FFT solver is adapted to handle the response of materials subjected to body forces with compact support, leading to fast, scalable, and accurate computation of the local fields.

00822#pg6>. The DFT's properties are used in the formulation of Eq. (5.3) to present the discretized modified L-S equation in a form suitable for iterative schemes <ref:2601. p. 1
Where the paper says it

Lippmann–Schwinger-type solvers: These include polarization-based iterative schemes and Krylov-subspace methods (e. , conjugate gradient, BiCG) for solving the linearized form of the L–S equation14–18.

00822#pg17>.

Surrogate Operator Learning

These models are trained to learn the effective surrogate nonlocal operators that can predict the material response under arbitrary macroscopic loading conditions <ref:2601.00822#pg5>. The resulting dataset Dp(3. p. 0

Where the paper says it

This functionally reduced RVE is used for translated averaging of direct numerical simulations (DNS) results, efficiently computed via a newly developed FFT-based solver for BFCS loading. The resulting dataset captures localized field responses and is used to train machine learning (ML) and neural networks (NN) models to learn effective nonlocal surrogate operators.

7) or Dd(3.8) is used as input to Block 3 (Optimization), replacing the more data-intensive DDNS <ref:2601. p. 2
Where the paper says it

Initial efforts by Silling39 and You et al. 40,41 demonstrated how DNS data could be used to construct surrogate integral operators for complex materials.

00822#pg27>. This substitution significantly reduces dataset size and improves efficiency, requiring only minor adjustments between Blocks 2 and 3 <ref:2601.00822#pg6>. The resulting datasets Dr or Dd are compressed into surrogate forms De p and De d to reduce computational cost while preserving essential micromechanical detail <ref:2601. p. 2
Where the paper says it

While ML and NN methods have advanced material modeling significantly, they frequently miss key micromechanical considerations—such as scale separation, boundary effects, and the RVE–which are essential for accurate predictions in both linear and nonlinear regimes. To bridge this gap, the proposed approach constructs new types of compressed datasets for complex microstructures (random or periodic), based on an innovative RVE framework.

00822#pg6>.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Novel RVE concept enables construction of datasets Dp and Dd that are independent of the constitutive behavior of individual phases and the analytical form of surrogate operators. This allows ML/NN models to predict effective nonlocal surrogate operators without needing to rebuild micromechanical models for every new architecture.

  2. FFT-based solvers are adapted to handle body forces with compact support, leading to a dataset generation process that is physically grounded and data-driven for constructing accurate surrogate models for the homogenization of complex materials. This allows the system to capture localized field responses efficiently.

  3. The CAMNN approach creates a modular, robust pipeline that generalizes across local/nonlocal, linear/nonlinear, and multi-physics CMs. The system can now be used to predict a priori undefined surrogate operators, removing the constraint of needing predefined effective nonlocal operators.

  4. The methodology eliminates issues related to size effects, boundary layers, and edge effects by defining a RVE that stabilizes outside a prescribed region where fields vanish. This ensures that the resulting predictions are free from boundary-layer artifacts.

  5. The system can integrate any numerical method for generating datasets Dp or Dd (e.g., FFT-based solvers) without requiring explicit knowledge of the final surrogate model structure. This makes the computational model entirely agnostic to both the composite’s microstructure and the numerical method used to obtain the dataset.

Abstract

We consider static linear elastic composite materials (CMs) with periodic structure. The core of the proposed methodology is the generation of a novel dataset using specially designed body force fields with compact support (BFCS), enabling a new RVE concept that reduces the infinite periodic medium to a finite domain without boundary artifacts. This functionally reduced RVE is used for translated averaging of direct numerical simulations (DNS) results, efficiently computed via a newly developed FFT-based solver for BFCS loading. The resulting dataset captures localized field responses and is used to train machine learning (ML) and neural networks (NN) models to learn effective nonlocal surrogate operators. These operators accurately predict macroscopic responses while reflecting microstructural features and nonlocal interactions. By accounting for field localization while simultaneously eliminating influences from finite sample size and boundary effects, it provides a physically grounded and data-driven framework for constructing accurate surrogate models for the homogenization of complex materials.

Sources

Related papers