Seismological expression of the iron spin crossover in ferropericlase in the Earth’s lower mantle

The two most abundant minerals in the Earth’s lower mantle are bridgmanite and ferropericlase. The bulk modulus of ferropericlase (Fp) softens as iron d-electrons transition from a high-spin to low-spin state, affecting the seismic compressional velocity but not the shear velocity. Here, we identify a seismological expression of the iron spin crossover in fast regions associated with cold Fp-rich subducted oceanic lithosphere: the relative abundance of fast velocities in P- and S-wave tomography models diverges in the ~1,400-2,000 km depth range. This is consistent with a reduced temperature sensitivity of P-waves throughout the iron spin crossover. A similar signal is also found in seismically slow regions below ~1,800 km, consistent with broadening and deepening of the crossover at higher temperatures. The corresponding inflection in P-wave velocity is not yet observed in 1-D seismic profiles, suggesting that the lower mantle is composed of non-uniformly distributed thermochemical heterogeneities which dampen the global signature of the Fp spin crossover.

anomalies. Area fraction of fast ( >+0.75, >+1, >+1.25 σ) anomalies as a function of depth for the 8 tomography models individually. This is similar to Fig. 1b and 4a (and Supplementary Fig. 5 for HMSL3), but presented here for the individual tomography models (4 P-wave and 4 S-wave) prior to being summed into the vote map. There is high variability between and within the individual tomography models, including the joint HMSL models. While a decorrelation between P-and S-wave models is apparent between some model combinations, it is not in others, which is why a single pair of tomography models in isolation may not render a robust signal of the spin transition.
Other features in these models, such as the oscillatory behaviour in HMSL, are not well understood and can also distract from the broader trends in P-wave and S-wave velocity. Surface area coverage for individual models and for variable sigma contours, for slow anomalies. Area fractions of slow (<-0.75, <-1, <-1.25 σ) anomalies as a function of depth for the 8 tomography models. Similar to Fig. 1c, but here presented for the individual tomography models (4 P-wave and 4 S-wave) prior to being summed into the vote map. As with the fast anomalies (Supplementary Figure 2), there is high variability between the tomography models. Other features in these models, such as the oscillatory behaviour in HMSL, are not well understood and can also distract from the broader trends in P-wave and S-wave velocity.

Supplementary Figure 5 (previous page)
The influence of using fewer tomography models. Area coverage as a function of depth for the sequential addition of tomography models used in this study. P-wave models (solid) and S-wave models (dashed), top panels (a) show fast anomalies, and bottom (b) panels show slow anomalies. Models used are listed within each panel (the HMSL profiles (1 model) are the same as shown in Supplementary Figures 2 and 3). There is an apparent decorrelation of P-wave and S-wave profiles in all combinations. However, the signal becomes more apparent when the models are summed into the vote maps, which identify the most common features between tomography models. . While there is some variability between the combinations, the observed decorrelation between P and S-wave velocity models is consistent at ∼1400 km for fast velocities and ∼1800 km for the slow velocities. This suggests that, regardless of which three models are used (conversely, the one model which is excluded), the effects of the iron spin cross-over in ferropericlase can be observed.     Figure 2). A higher proportion of Fe in the ferropericlase (lower KD) may increase the crossover pressure6. We find that the FeO content in ferropericlase remains below 25 mol%, which is the threshold for observing substantial increases in the crossover transition pressure7 for the depth-dependent KD case. Thus, depth-dependent KD does not have a significant influence on the crossover depth/pressure range over which we observe the anomalous signal in compressional velocity in the tomographic models. -500 K +500 K Avg.

Supplementary Figure 9
Plots related to velocity, temperature and pressure/depth calculations. Top panel: Development of Figure 1d. PREM8 is shown in black circles and the black lines are the calculated velocities for pyrolite9. Figure 1d demonstrates the spin transition effect on Vp for the case in which predicted Vs matches PREM (grey lines). Since Vs for pyrolite does not fit PREM with an adiabatic temperature gradient10, 11, the temperature profile that shifts Vs to align with PREM (grey line right panel) undulates in the lowermost mantle. Bottom panel: The self-consistent geotherms from our pyrolite calculations9 for the elastic moduli and velocity profiles plotted in Figure 2. The calculations start by setting the temperature at the top of the lower mantle to 1373 K (blue, the -500 K case), 1873 K (black, the average case), and 2373 K (red, the +500 K case) and allowing the temperature to increase adiabatically as the calculations proceed to higher pressures across the lower mantle.  Depth (km)

Supplementary Figure 10
The results of our Gaussian-fitting procedure (see Methods) for all 8 tomography models used in this study. Analysis of velocity-frequency distributions of a variety of tomographic models reveals that they exhibit significant differences that confound intermodal comparisons12. These differences can be categorized as scale/amplitude (e.g., caused by variability in tomographic model data, design, regularization), shift/alignment (e.g., caused by reference to different 1-D global models), and shape of the distributions (variations in distribution morphology that remain even after accounting for linear shift and scale differences). By analyzing distributions we find that all models yield Gaussian-like variations in Vp and Vs in the depth range 1,000-2,000 km, however, there are particularly large discrepancies in amplitude between the different models12. These scale differences must be normalized to a reference standard in order to establish a useful definition for fast and slow anomalies that can be compared across the suite of models. We do this by combining each model from 1,000-2,200 km depth, and performing iterative Gaussian fitting to the central portion (i.e., within ±σ) of the resultant distribution as described in Methods. The value of σ obtained in this manner is then used to define what qualifies as fast and slow anomalies in the models). Surface area calculations for two additional tomography models. In addition to the eight models used in the paper, depth-dependent change in surface areas for the joint tomography models of SP12RTS13 and TX201914 are also included for reference. The trend between P-and S-wave models is somewhat variable between the individual model pairs of SP12RTS, TX2019, (as for HMSL3, Supplementary Figure 5) but do hint at a mid-mantle decorrelation similar to the models analyzed in the main text. However, aspects of their construction such as inclusion of subducting slabs in the starting model for TX2019 and the long-wavelength SP12RTS make them less ideal for the mid-mantle focus in this study. b.