Identification of optimal history variables and corresponding hereditary laws in linear viscoelasticity

This paper presents an operator-theoretic framework for linear viscoelasticity that utilizes Kolmogorov $N$-widths to identify optimal finite-rank internal-variable approximations, ensuring thermodynamic consistency, stability, and rigorous convergence bounds for reduced-order modeling.

The Gist

Imagine you are trying to predict how a piece of memory foam or a sticky polymer will react when you squeeze it. Unlike a simple spring that snaps back instantly, these materials are "viscoelastic." They remember how you squeezed them in the past, and that history affects how they push back right now.

In physics and engineering, we call this the hereditary law. It's a fancy way of saying: "The stress you feel today depends on every little bit of strain you applied yesterday, last week, and a million years ago."

The Problem: The "Infinite Memory" Burden

The paper starts with a massive headache for computer simulations. To be perfectly accurate, a computer needs to remember the entire history of the material's deformation.

Imagine trying to calculate the weather. If you had to remember every single raindrop that fell since the beginning of time to predict tomorrow's storm, your computer would crash instantly. It's too much data.

Engineers usually try to fix this by using "shortcuts" (like the Prony series), which are like guessing the material's memory based on a few standard patterns. But the authors ask: "What if our shortcuts aren't the best ones? What if there's a perfect, most efficient way to summarize this memory?"

The Solution: The "Best Summary" (N-Widths)

The authors use a branch of mathematics called Kolmogorov N-widths. Think of this as the ultimate "compression algorithm" for memory.

Here is the analogy:
Imagine you have a 10-hour movie (the material's full history).

• The Old Way: You try to summarize it by picking 5 random scenes. You might miss the plot entirely.
• The Old "Smart" Way: You pick 5 scenes based on what you think is important (like action scenes). It's better, but maybe you missed the emotional climax.
• The Authors' Way: They use math to find the 5 specific scenes that, if you watched them, would allow you to reconstruct the entire movie with the least amount of missing information possible.

In the paper, these "scenes" are called Optimal History Variables. They are the specific, mathematically perfect "notes" you need to take to remember the material's behavior without storing the whole encyclopedia.

How They Did It (The "Encoder/Decoder" Trick)

The paper treats the material's memory as a machine (an operator) that takes a history of squeezing (input) and outputs a history of resistance (output).

1. The Encoder: They figure out the best way to "compress" the input history into a small number of numbers (variables).
2. The Decoder: They figure out the best way to "rebuild" the output from those numbers.

They proved that if the material behaves nicely (which most real materials do), this machine is compact. In math-speak, this means it can be perfectly approximated by a smaller, simpler machine.

The "Gibbs Phenomenon" (The Glitch)

In their tests, they found something interesting. When they tried to simulate a sudden, sharp change (like slamming the material with a step-function force), their perfect summary had a little "glitch" at the very end. It looked like a ringing bell that wouldn't stop vibrating.

They call this the Gibbs phenomenon.

• Analogy: Imagine trying to draw a perfect square using only smooth, wavy sine waves. You can get close, but at the sharp corners, your waves will overshoot and wiggle.
• The Fix: The authors explain that this wiggle is a natural trade-off. If you want a super-smooth summary, you have to accept a little wiggle at the sharp edges. They showed how to manage this so it doesn't ruin the simulation.

Why This Matters (The "Polycrystal" Test)

To prove this isn't just theory, they tested it on a Representative Volume Element (RVE).

• The Analogy: Imagine a block of Swiss cheese where every hole is a different material, and every piece of cheese is a different material. It's a chaotic mess.
• The Result: They took this chaotic, complex material and used their "Optimal Summary" method. They found that they could replace the incredibly complex, slow-to-calculate model with a tiny, fast model that was just as accurate.

The Takeaway

This paper gives engineers a rulebook for efficiency.

1. Don't guess: You don't need to guess which history variables are important. The math tells you exactly which ones are the "best."
2. Save time: You can simulate complex materials (like new metamaterials or biological tissues) much faster because you only need to track a few "optimal" variables instead of the whole history.
3. Data-agnostic: It doesn't matter if you got your data from a lab experiment or a supercomputer; this method finds the best summary for any linear viscoelastic material.

In short: They turned the problem of "remembering everything" into "remembering the most important things," using a mathematical lens that guarantees you aren't missing anything crucial. It's the difference between carrying a library of books with you to answer a question, versus carrying a single, perfectly written cheat sheet.

Technical Summary

1. Problem Statement

The paper addresses the challenge of representing linear viscoelastic material behavior efficiently in computational mechanics. While modern experimental and computational methods (e.g., Dynamic Mechanical Analysis, RVE simulations) can generate vast amounts of material data, translating this data into predictive constitutive models remains difficult.

• The Core Issue: Traditional approaches often rely on a priori assumptions about the functional form of the hereditary law (e.g., Prony series or polynomial expansions). This raises the question: What is the optimal, finite-rank representation of a viscoelastic material's history dependence?
• The Gap: There is a lack of rigorous mathematical frameworks to determine the "best" set of internal (history) variables and the corresponding low-rank hereditary operators that minimize approximation error for a given class of materials, particularly when the exact functional form of the relaxation kernel is unknown or computationally expensive to derive (e.g., from multiscale homogenization).

2. Methodology

The authors develop an operator-theoretic formulation grounded in the theory of Kolmogorov $N$-widths. The methodology proceeds through the following steps:

A. Functional Framework

• Hereditary Law: The stress-strain relationship is defined via a Volterra integral equation (Boltzmann superposition principle):
  $$\sigma(t) = C \epsilon(t) - \int_{-\infty}^{t} K(t-s)\epsilon(s) ds$$
• History Operator: The integral term is reformulated as a linear operator $S$ acting on the space of strain histories. The authors define weighted Hilbert spaces ($H$) for strain and stress histories to account for fading memory properties, utilizing a weighting function $w(\tau)$.
• Compactness: Under natural physical assumptions (boundedness of the relaxation kernel and finite time duration), the history operator $S$ is proven to be a compact Hilbert-Schmidt operator. This property is crucial as it guarantees the existence of optimal finite-rank approximations.

B. Optimal Approximation via $N$-widths

• Singular Value Decomposition (SVD): The authors utilize the spectral theory of compact operators. The optimal rank-$N$ approximation of the history operator $S$ is derived from the eigenfunctions of the self-adjoint operators $S^*S$ and $SS^*$.
• Optimal Variables: The "optimal history variables" are identified as the coordinates of the strain history projected onto the eigenfunctions (singular vectors) of $S^*S$.
• Error Bounds: The approximation error in the operator norm is exactly equal to the $(N+1)$-th singular value ($s_{N+1}(S)$) of the operator. This provides a rigorous, quantitative bound on the error introduced by reducing the memory dimension.

C. Numerical Implementation

• Truncation and Sampling: Since the exact operator $S$ is often unavailable (especially for complex microstructures), the authors propose a two-step numerical scheme:
  1. Sampling: Evaluate the material response to a finite set of $M$ basis strain histories (where $M$ is large).
  2. Projection: Construct a truncated matrix representation of the operator and compute its first $N$ singular vectors ($N \ll M$).
• Convergence Analysis: The paper provides a rigorous error analysis showing that the total error is bounded by the sum of the rank error (truncation of the singular value expansion) and the sampling error (truncation of the basis dimension).

3. Key Contributions

1. Rigorous Identification of Optimal Variables: The paper moves beyond heuristic choices of internal variables (like standard Prony terms) to mathematically prove that the eigenfunctions of the history operator constitute the optimal basis for representing viscoelastic memory.
2. Unification of Theories: It establishes a clear structural link between hereditary integral formulations and internal variable theories, showing that the latter are essentially finite-rank approximations of the former.
3. Thermodynamic Consistency: The resulting reduced models inherently inherit thermodynamic consistency, stability, and provable approximation bounds, unlike many machine-learning-based constitutive models which may lack physical guarantees.
4. Data-Driven Agnosticism: The framework is indifferent to the source of the data. It applies equally to experimental data, first-principles simulations, or multiscale homogenization (RVE) results, provided the material response can be queried for specific strain paths.

4. Results

The authors validate the theory through two numerical examples:

• Example 1: One-Dimensional Standard Linear Solid:

  • The method successfully recovered the exact relaxation function using a low-rank approximation.
  • Convergence: The error in the inelastic strain decreased rapidly as the rank $N$ increased, converging to the theoretical lower bound defined by the singular values.
  • Gibbs Phenomenon: The authors analyzed the behavior near sharp cutoffs in strain history, noting that while the $L^2$-norm error converges, the optimal basis (derived from Sturm-Liouville problems) exhibits oscillatory ringing (Gibbs phenomenon) near discontinuities. They provided estimates for the boundary layer width.

• Example 2: Viscoelastic Representative Volume Element (RVE):

  • A complex 3D polycrystal with 64 grains, each having random viscoelastic properties (Maxwell-Wiechert models with Gamma-distributed parameters), was simulated.
  • The effective macroscopic behavior was computed via Finite Element Analysis (FEA).
  • The optimal low-rank model was constructed from the FEA data. The results demonstrated that the reduced model could accurately reproduce the complex, heterogeneous macroscopic response with significantly fewer internal variables than a standard Prony series would require for the same accuracy.

5. Significance and Implications

• Computational Efficiency: The proposed method enables the creation of compact surrogates for complex rheological behaviors. This is critical for large-scale simulations where storing full history integrals is computationally prohibitive.
• Optimal Model Reduction: It provides a principled criterion for selecting the number of internal variables required to achieve a specific accuracy, eliminating the guesswork often associated with model calibration.
• Bridging Data and Physics: The work offers a rigorous mathematical bridge between "big material data" and continuum mechanics. It allows researchers to extract optimal constitutive laws directly from data without assuming a specific functional form, while maintaining the physical constraints of linear viscoelasticity.
• Future Directions: The authors suggest extending this framework to nonlinear viscoelasticity, thermo-viscoelasticity, and integrating it with uncertainty quantification for material parameter identification.

In summary, this paper provides a foundational mathematical toolset for reduced-order modeling in computational mechanics, ensuring that the simplification of viscoelastic memory is done in the most mathematically optimal way possible.