Predicting stochastic fracture in hardened alite paste via a microstructure-informed multiscale reconstruction of mechanical fields

Microscale heterogeneity and interfacial gradients fundamentally govern the deformation and fracture of cementitious materials, yet these features remain inadequately captured by conventional continuum mechanics. While nanoindentation provides direct access to local mechanical properties at the nanometer resolution, standard mean-field homogenization methods inherently neglect spatial correlations and fail to resolve non-linear fracture and damage evolution. Here, we develop a two-dimensional, microstructure-informed multiscale framework to reconstruct high-fidelity, continuous mechanical fields from sparse nanomechanical data. Central to this spatial reconstruction is Gaussian Process Regression (GPR), combining a center-decay mean function with a composite Matern kernel. This mechanistically motivated prior enables super-resolution field reconstruction, capturing sharp phase-boundary gradients and plausible sub-grid variations below the original indentation spacing. Using synthetic alite (impure tricalcium silicate), a complex multi-phase composite as a model system, we demonstrate that incorporating the proposed spatial model outperforms classical Voigt, Reuss, Mori-Tanaka, and Hashin-Shtrikman estimates, yielding a predicted macroscopic elastic modulus in 95.6% agreement with the average measurements from in-situ compression tests. By further integrating indentation-derived fracture-resistance fields into a phase-field formulation, the model reveals stochastic microcrack initiation and morphology-dictated failure pathways within the heterogeneous low-density C-S-H matrix. This framework offers a potentially transferable route to convert sparse micromechanical data into spatially continuous fields for predicting nonlinear damage in cementitious materials.