Iron Content‐Dependence of Ferropericlase Elastic Properties Across the Spin Crossover From Novel Experiments and Machine Learning

The iron spin crossover in (Mg1‐xFex)O ferropericlase causes changes to its physical properties that are expected to affect seismic velocities in Earth's lower mantle. We present new time‐resolved pressure‐volume measurements of iron‐rich ferropericlase (xFe = 0.40, 0.59) and combine the results with literature data with xFe = 0.04–0.6 to investigate the dependence of ferropericlase elastic properties on iron content. We infer the relationship between unit‐cell volume, pressure and iron content directly from the data by training Mixture Density Networks and derive bulk modulus, density and bulk sound velocity from the outputs. This allows us to constrain the effect of the spin crossover on these properties and estimate their uncertainties for different iron contents. Our findings indicate that the spin crossover may significantly alter the physical properties of ferropericlase in iron‐enriched regions in the lowermost mantle, with implications for the interpretation of seismic heterogeneities observed near the core‐mantle boundary. Plain Language Summary The most voluminous layer in Earth's interior is the lower mantle, the deepest layer above the core. About 18% consists of the mineral ferropericlase, an iron‐magnesium oxide. Under the extreme pressures in the lower mantle, the iron atoms of ferropericlase change at the electronic level, in a process called the spin crossover. This leads to changes in the material properties of ferropericlase that affect seismic wave speeds. Seismic waves are produced by earthquakes and can be used to investigate Earth's interior, but interpreting seismic observations requires constraints on the properties of mantle minerals. Here we investigate how the spin crossover depends on ferropericlase iron content. We conducted new high‐pressure experiments on samples containing 40% and 59% iron and combine our results with literature data for a range of ferropericlase compositions. Results from different studies show inconsistencies, caused by uncertainties inherent to experimental data. To overcome this, we use machine learning to infer the relationships between material properties, pressure and iron content. Our results show that the spin crossover would significantly alter properties of iron‐rich ferropericlase in the deepest parts of the lower mantle. This may help to explain the anomalous behavior of seismic waves in regions near the core‐mantle boundary. Key Points New time‐resolved compression data on (Mg0.60Fe0.40)O and