arXiv Science⌕ Search

arXiv · 2611.18473

Machine learning of strength surfaces enables high regularization length phase-field fracture

Abstract

Classical phase-field models of brittle fracture capture toughness-controlled crack growth but not the material's strength surface, which governs crack nucleation. The theory of Kumar et al. (2020) incorporates the strength surface through a nucleation driving force, and Chockalingam (2025) derived this driving force for arbitrary strength surfaces described by $n$ material parameters, with which the phase-field strength surface (PFSS) converges to the material strength surface as the regularization length $\varepsilon \to 0$ and matches it exactly at $n$ chosen states for any $\varepsilon$. In practice, however, the strength surface must be supplied as an analytic form chosen and calibrated case by case, and for the widely used analytic surfaces ($n=2$) accurate strength reproduction requires $\varepsilon$ commensurate with the Irwin characteristic length $l_{ch}$, imposing prohibitive mesh requirements for large structures. We introduce a neural-network (NN) strength function that is learned directly from strength data, is physics-constrained by construction, and provides far more calibration parameters than analytic surfaces; the calibration is generalized to any number of strength states and to enforce openness in chosen directions. A single architecture captures synthetic data from the Mohr-Coulomb, Drucker-Prager, Mogi-Coulomb and 3D Hoek-Brown surfaces and experimental data for graphite, natural rubber and styrene-butadiene rubber. The resulting PFSS is accurate up to $\varepsilon = 10\,l_{ch}$, whereas that of the exact analytic surfaces performs comparatively poorly even at $\varepsilon = 0.5\,l_{ch}$. Finite-element simulations validate the theory across fracture regimes. By decoupling strength fidelity from the regularization length, the approach substantially reduces the mesh resolution required for phase-field fracture simulations of large structures.

Explore related subjects

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

S Chockalingam, Aditya Kumar, Adrian Buganza Tepole. 2026-11-13. Machine learning of strength surfaces enables high regularization length phase-field fracture. https://arxiv.science/abs/2611.18473

Citations retain the original authors and DOI and include this site’s record URL. If this discovery service contributed to your research, you can also cite arXiv Science by Indiciae: BibTeX · RIS.

KEEP EXPLORING

Related papers

Quantum Interband-Coherence-Driven Inverse Faraday Effect in Intrinsic Graphene

The inverse Faraday effect converts the angular momentum of circularly polarized light into a quasistatic magnetization, providing a fundamental route to ultrafast optical control of magnetic fields. We develop an analytical quantum theory of this effect in intrinsic graphene, where massless Dirac fermions, electron-hole symmetry, and weak spin-orbit coupling suppress conventional population-based interpretations. We show that in intrinsic graphene, the helicity-dependent orbital magnetization arises from the off-diagonal zero-harmonic component of the second-order density matrix, rather than from intraband carrier circulation or diagonal population imbalance. We obtain closed expressions for the induced temporal surface magnetization, its finite-pulse dynamics, and the nonlocal stray magnetic field generated by the two-dimensional (2D) source. Optical helicity, frequency, pulse envelope, and beam profile independently control the response. Our results identify intrinsic graphene as an ideal platform for coherent orbital optomagnetism in 2D Dirac matter.

cond-mat.other↗

Thermal radiation of nanowires and near-field interaction with substrates: exact computation and proximity approximation

Radiative heat transfer involving complex nano-objects is difficult to evaluate due to their sub-wavelength size, which induces deviation to macroscopic Planck's opaque surface-to-surface formalism, and due to the small distances involved, which lead to near-field radiative (photon tunneling) effects. As a result, thermal radiation has mostly been neglected in analyses of thermal transport in nanowires. Here, we report on calculations of thermal emission by single nanowires and on the radiative interaction between suspended silicon nanowires and flat substrates. We observe that metallic nanowires can be particularly emissive in comparison to wires made of other materials. While thermal emission from an isolated silicon nanowire is lower than Stefan-Boltzmann's prediction, the radiative exchange with a substrate can be larger by one order of magnitude. The considered configurations also allow benchmarking various levels of common or original approximations, including, in near field, the proximity approximation, the thickness-dependent PA, the electrostatic approach, and the effective emissivity in the far field.

cond-mat.other↗

Nonlinear optical response with relaxation in the velocity gauge

We develop a framework for the evaluation of the nonlinear optical conductivity of electrons in crystals, driven by homogeneous optical fields, taking into account dissipation effects within the relaxation time approximation (RTA). The present formalism is based on Hamiltonians with minimal light-matter coupling within the velocity gauge, where the electric field is expressed in terms of a vector potential. The nonlinear conductivities are obtained by perturbatively solving the equation of motion for the single-particle reduced density matrix, which includes dissipative effects via the relaxation time approximation. We present a perturbative method to evaluate the reference equilibrium density matrix, towards which dissipation relaxes the system, which in the velocity gauge adiabatically follows the applied vector potential, and is also broadened by dissipative effects. Comparing with previous formulations based on the length gauge, the present approach has the benefit of avoiding the cumbersome evaluation of derivatives of Bloch states with respect to momentum. We test the presented method against length-gauge results, and calculate second- and third-order responses in a graphene-like tight-binding model with a staggered sublattice potential and a periodic modulation. We find that the relaxation time approximation gives results that deviate both quantitatively and qualitatively from the simple inclusion of an imaginary damping term in the frequencies, referred to as adiabatic switching. At equal relaxation parameter, the leading above-gap injection contribution differs by a factor of two; below the gap, the numerical results indicate a finite imaginary dc response in the small-relaxation limit of the relaxation-time calculation. These results show that adiabatic switching does not generally reproduce relaxation through the equation of motion, even in the small-relaxation limit.

cond-mat.other↗