Lattice thermal conductivity (k_lat) of mantle minerals establishes a connection to the heat flux across the core-mantle boundary (CMB), playing an important role in Earth's energy budget and various facets of the planet's dynamic processes. Here, we trained a machine learning potential (neuroevolution potential, NEP) for CaSiO₃ perovskite (CaPv), the third most abundant mineral in Earth's lower mantle, to investigate the k_lat of pyrolitic lower mantle at the CMB. The calculated k_lat of CaPv is in agreement with the results from previous first-principles molecular dynamics simulations, but smaller than those of phonon quasiparticle method. We show that the k_lat of pyrolitic aggregates has increased by 7% and 5% with the inclusion of CaPv, demonstrating its significance in shaping the thermal profile of Earth's interior. Considering other mantle minerals and iron content, as well as the global distribution of temperature, we evaluated the heat flow across the CMB by incorporating the previously determined  of MgSiO₃ and periclase. The estimated value is 8.39±2.1 TW, in discrepancy with the value derived from the assessment of Fe-alloy, supporting the possible presence of a thermally or chemically stratified layer at the top of the outer core.

thermal conductivity,Machine learning potential,lower mantle conditions,CaSiO3 perovskite