Subsolidus convection in the mantle of Earth is the driving mechanism behind plate tectonics and provides a central framework linking geophysical, geochemical, petrological, hydrological, and biological processes within the system. Seismic observations have revealed mantle heterogeneities in wide-ranging scales from less than tens of to thousands of kilometers. Understanding the origins and dynamics of these anomalies is critical to advance our knowledge on how mantle convection operates and coevolves with the surface system. This dissertation attempts to constrain the past, present and future of mantle dynamics with lines of evidence from seismology, geodynamics, petrology, geochemistry, and astrophysics. Above Earth’s core, two continent-sized large low shear velocity provinces (LLSVPs) beneath Africa and the Pacific Ocean were seismically detected decades ago. Yet their origin, composition, detailed morphology and influence over mantle convection remain elusive. First, I propose the two LLSVPs may represent the mantle remnants of the Moon-forming impactor Theia. I show that the mantle of Theia is intrinsically denser than Earth’s mantle and would have sunk and accumulated into LLSVP-like structures in the deepest mantle after 4.5 billion years. Second, I examined the maximum height of the two LLSVPs and determined that the African LLSVP is ~1,000 km higher than the Pacific counterpart. Using geodynamic simulations, I find the height of a stable LLSVP is mainly controlled by its density and the ambient mantle viscosity. With ~1,000 numerical experiments, I conclude that the origin of the great height difference between the LLSVPs is that the African LLSVP is less dense, and thus less stable than the Pacific LLSVP. Next, I numerically identified another dynamic scenario accounting for the vastly different height of the two LLSVPs, which is caused by catastrophic sinking of accumulated subducted slabs at the 660-km boundary. Last, targeting one ancient carbonatite above the African LLSVP, I show that lithium isotopes in humite measured by nanoscale secondary ion mass spectrometry was able to uncover the signature of a subducted oceanic crust in its magma source, which may return from the interior to the surface by mantle plumes.