NATURE GEOSCIENCE | ADVANCE ONLINE PUBLICATION | www.nature.com/naturegeoscience 1 The Earth’s

magnetic field exhibits internally driven variations on an extremely wide range of timescales1. As a highly nonlinear system, the geodynamo could produce all of these through a stochastic pro-cess without the need to invoke any external forcing mechanism1–3. For variations observed on the timescale of tens to hundreds of mil-lions of years (Myr)4,5, however, the similarity to mantle convection timescales suggests an alternative hypothesis, whereby changes in core–mantle boundary (CMB) heat flow play an important role in determining average geomagnetic behaviour. This forcing could combine with a further stochastic component of geodynamo behaviour, and is worthy of intense investigation because it could potentially allow geomagnetic behaviour to be used to constrain lowermost mantle processes occurring over Earth’s history. An overarching theory of interaction could then be developed between the two great engines of the Earth’s interior: the geodynamo and mantle convection incorporating plate tectonics. Based on the a priori acceptance of the mantle forcing hypothe-sis, numerous researchers have causally related events in the palae-omagnetic and geological records6–19, linking, for example, changes in magnetic polarity reversal frequency, through mantle plumes, to the emplacement of large igneous provinces. Such claims are somewhat speculative, but their general concept is plausible. Interpretations of seismic wave tomography using global plate reconstructions suggest that sinking lithospheric slabs and rising mantle plumes are indeed whole-mantle processes20,21 conceivably influencing both the geodynamo and the surface. Furthermore, numerical geodynamo models strongly support claims made by palaeomagnetists22–24, that persistent non-axial dipole features of the geomagnetic field observed over the last 10 kyr, and during individual excursions and reversals, reflect the influence of the Possible links between long-term geomagnetic variations and whole-mantle convection processe