Lavas erupted at the Society hotspot during the Matuyama-Brunhes (M-B) polarity reversal record transitional field behavior containing two tight paleodirectional groups that when averaged are antipodal at the 95% confidence level, and thus correlate to antipodal clustered virtual geomagnetic poles (VGPs). The occurrence of these observations––data obtained from two published records of the M-B transition from distinct sections of a succession of flows on Tahiti––is associated with a time when the strength of the axial dipole was significantly reduced. One of the clusters was recorded by lavas that were not erupted in succession, suggesting that significant time had passed during this volcanic activity. Time spent during the formation of the antipodal cluster is unknown, although it resides in the same location as VGP clusters associated with four other transitional events obtained from Society hotspot lavas. Calculated VGPs at the Society hotspot for both “polarities” of the averaged historic field––less the axial dipole term––are found in the cluster locations. These findings offer strong evidence for a two-tiered dynamo process in which nearly the entire axial dipole component undergoes both demise and regeneration quasi-independently from that of the remainder of the field, the pattern of which is tied to long-held physical conditions of the lower-most mantle.
 Ever since the first account of geomagnetic field behavior during a polarity transition was reported by van Zijl et al. , a half-century ago, numerous paleomagnetic records associated with several distinct reversals have been made available from sites about the globe. Yet, by far the largest number of available records is associated with the last reversal, the Matuyama-Brunhes (M-B), and for this reason several attempts have been made to model global field behavior during this transition [e.g.,Hoffman, 1981, 2000; Love and Mazaud, 1997; Shao et al., 1999; Singer et al., 2005; Leonhardt and Fabian, 2007]. Notwithstanding the significant advances that have been provided through these analyses, the near impossibility of determining temporally synchronous behavior among globally scattered recording sites is one of many obstacles to a complete understanding of the dynamo process during this reversal. With this in mind we focus our attention on the published M-B paleomagnetic recordings obtained from one site, the Society hotspot, from where there are now available two distinct sections of a succession of lava flows exposed in the Punaruu Valley on the island of Tahiti, French Polynesia [Chauvin et al., 1990; Mochizuki et al., 2011].
 A number of Plio-Pleistocene records of transitional field behavior have come from Society hotspot lavas.Roperch and Duncan  and Chauvin et al. , working on Huahine and Tahiti, respectively, presented in total four such records spanning from ∼3 Ma through the M-B reversal.Hoffman and Singer  added to this dataset a more youthful Tahitian lava record of a geomagnetic event, the Big Lost, dated at 579 ± 6 ka, some 200 kyr after the M-B. In that paper the authors noted a feature common to all five available transitional field records––namely a cluster of virtual geomagnetic poles (VGPs) located west of Australia (Figure 1)––a finding that supports the contention that long-lived heterogeneities in the lower-most mantle control the pattern of magnetic flux emanating from the outer core [e.g.,Laj et al., 1991; Hoffman, 1992; Constable, 1992].
Hoffman and Singer also showed that the 20th Century modern-day field, given a severely weakened axial dipole, would at the site of the Society hotspot be associated with a vector direction corresponding to a VGP at the same cluster locality (seeFigure 1). Such a correlation offers evidence that significant features of the flux pattern emanating from the outer core remained stationary from the Plio-Pleistocene through to the present day. Apparent similarities in the location of two primary flux concentrations in the Southern Hemisphere observed in models at Earth's surface given removal of the axial dipole term (i.e., the complex, so-called non-axial dipole (NAD) field)––models deduced from both direct measurements of the present-day field and the time-averaged historic field [seeConstable, 2007], as well as from paleomagnetic field measurements over the entire Brunhes Chron [Shao et al., 1999]––attest to the longevity of such a stasis.
Chauvin et al.'s record of the Matuyama-Brunhes reversal contains a solitary cluster of virtual poles (shown in green inFigure 1) west of Australia, data obtained from a section of lavas exposed on the south face of the Punaruu Valley. Recently, Mochizuki et al. reported a far more detailed account of field behavior during the M-B reversal, including paleointensity variation, obtained from a distinct section of lavas exposed on the north face of the same valley.
 Here we analyze the available Tahitian lava M-B transitional field dataset, and argue that these two distinct records, when considered together and in light of the recurring field behavior shown inFigure 1, display remarkable features providing important insights into the dynamo process in Earth's fluid outer core. These features of field behavior are found during the most dramatic directional changes recorded during the M-B reversal.
2. Recorded Directional Rebounds
 The path of the virtual geomagnetic pole (VGP) [Mochizuki et al., 2011, Figure 12a], is redrawn here in our Figure 2 with the original authors' numbers indicating the chronological order of the transitional portion of the record. Of the 28 VGPs reported for this lava sequence, transitional paleodirections are seen to start with directional group (DG) 17 and complete with DG 24. We focus our discussion on results from these particular flows.
 Following the initial movement of the VGP to a transitional location (DG 17), one can see in Figure 2 a cluster of three transitional VGPs in eastern North America. Note, however, that the lavas involved (DGs 18, 21 and 23) are associated with flows that were not erupted in succession. We now consider this finding by Mochizuki et al.  in more detail.
 1. The VGPs corresponding to DGs 19 and 20 are found to the north of the virtual pole associated with DG 18 although they are not statistically distinguishable from DG 18 at the 95% confidence level. The next VGP, associated with DG 21, is then seen as a possible rebound to the DGs 18-21-23-cluster locality.
 2. DG 22, having reverse polarity, lies chronologically between DGs 21 and 23. Hence, this finding corresponds to a second directional rebound to the cluster locality. This rebound is similar to that reported for the much older Miocene reverse-to-normal Steens Mountain record [e.g.,Prévot et al., 1985; Jarboe et al., 2011] insofar as full polarity is reached before the rebound takes place. It differs in that prior to the rebound the recorded directional behavior in the Tahitian record involves a return to the initial (reverse) polarity while at Steens Mountain the record first shows an unsuccessful attempt to reach a stable final (normal) polarity.
 3. DG 24, erupted after DG 23, possesses a paleodirection associated with a VGP in western Australia, nearly antipodal to the North American cluster. Following DG 24 the VGP moves to high northern latitudes at which point the reversal is completed.
3. Antipodal Directional Clusters
 We now turn to the Chauvin et al. Matuyama-Brunhes directional data obtained from lavas sampled from the south face of the Punaruu Valley, that is, the five VGPs that cluster off the west coast of Australia (seeFigure 2). Remarkably, the average paleodirection for these flows (Dec = 229.8, Inc = 1.5, α95 = 6.3) is almost precisely antipodal to that for the Mochizuki et al.  rebound cluster flows (Dec = 43.3, Inc = 3.8, α95 = 9.1) (Figure 3).
 We argue that the following features seen in these Tahitian lava records are indicative of fundamental determinable aspects of the dynamo process responsible for field reversal: (1) the finding of two directional rebounds, in particular, the latter which spans a round-trip to reverse polarity; (2) the antipodal nature of the two transitional directional groupings, and associated VGP clusters, from distinct flows on each side of the Punaruu Valley; and (3) the nearly 180° swing in direction recorded in DG 24 relative to DG 23--corresponding to a swing of the VGP path to western Australia from eastern North America––just prior to the recorded completion of the reversal.
4. Evidence of a Two-Stage Reversal Process
 The transitional Matuyama-Brunhes directional dataset from Tahiti fully supports the magnetic flux separation model posited byHoffman and Singer . Specific to the model is the contention that the origin of the axial dipole (AD) field arises almost exclusively in the deeper portion of the core fluid. And the structure of the remainder of the field is both defined and sustained by long-held, mantle-controlled heterogeneities about the core-mantle boundary––i.e., variations in physical conditions that affect fluid motion at the top of the outer core. According to the model an attempt by the dynamo to reverse involves first the demise of the deeper-generated AD field, leaving the shallow-core-generated (SCOR) field––note that the SCOR-field can be approximated by the harmonic content of the NAD-field––to define the transitional global field at Earth's surface. A complete reversal then consists of two stages: (1) reversal of the AD-field and (2) reversal of the SCOR-field. Following demise of the axial dipole at the onset of the Matuyama-Brunhes transition [seeSinger et al., 2005], these two stages are required for a complete field reversal. It is likely that stage 1 and stage 2 occur at different times; if so, the order of occurrence is not presently understood.
 If indeed the existence of a harmonically complex surface-core field is due to core-mantle interaction, the resulting SCOR (again, approximated by the NAD) during modern times may have been similar to that during the time of the Matuyama-Brunhes reversal. To test the Matuyama-Brunhes data from Tahiti we employ the gufm1 model [Jackson et al., 2000] from which the average NAD-field over 400+ years of historic time (from 1590 to 1995) can be calculated.Figure 4(left) shows the vertical component of the average NAD-field over this span of time at Earth's surface contoured in strength.Figure 4(right) is the same field following a change of sign, in essence the “reverse” NAD-field.
Figure 5displays the individual VGPs for both the rebound cluster reported from the north side of the Punaruu Valley, and the cluster reported from the south side of the Punaruu Valley. Also plotted are the VGPs associated with the average historic NAD-field calculated for the site from the gufm1 model for each case of sign. As can be seen, the averaged NAD-field having either sign correlates remarkably well with each of the two Matuyama-Brunhes Tahitian lava VGP clusters.
 The striking correlation seen in Figure 5further supports the compatibility of the Tahitian lava findings with the dual-source SCOR-field model. More specifically, these data display, for the first time, direct evidence of the two separate stages of the reversal process hypothesized byHoffman and Singer . What follows is a chronological interpretation of the Matuyama-Brunhes dynamo reversal process givenMochizuki et al.'s  findings.
4.1. Demise of the Axial Dipole Field
Mochizuki et al.  find a significant weakening of the paleofield starting with DG 14. Specifically, the field intensity is observed to decrease to 3–6 μT corresponding to virtual axial dipole moments (VADMs) of 0.8–1.4 × 1022 Am2.
4.2. SCOR-Field Domination
 By the time DG 18 erupted the ambient magnetic field on Tahiti was transitional, having a direction dictated by the dominating shallow-core SCOR-field. Specifically, due to the proximity of the Society hotspot to Australia, the site at that time was mostly affected by an outward-directed flux patch beneath western Australia, a feature still in existence in the historic field (Figure 4, left). Given a significantly weakened AD-field, the present strength of this patch would have a dominating effect over a vast area, mostly within the southern hemisphere, attracting either the north VGP or south VGP, depending on the sign of its flux [seeHoffman and Singer, 2004, 2008]. For the case when DG 18 acquired its magnetic remanence, the south VGP was drawn to the patch, and the north VGP to eastern North America. Between the time of eruption of DG 18 and DG 19 the dominating SCOR-related Australian flux patch and/or the AD-field may have varied in strength sufficiently to cause a minor change in the field direction experienced at the hotspot. Yet by the time of the eruption of DG 21 the SCOR flux patch may have regained its strength and/or the AD-field again may have nearly vanished so as to thoroughly control the Tahitian field (the case during the eruption of DG 18).
4.3. Predominance of the Axial Dipole
 Relative to the SCOR-field, the AD-field then regained sufficient strength to dominate the field direction, causing a return to reverse polarity, if only for a short time during which DG 22 erupted.
4.4. SCOR-Field Domination
 The axial dipole again loses its control over the global field to a more dominating SCOR-field. This is seen in DG 23 which records virtually the same field direction and associated VGP as was the case during the cooling of DG 18 and later DG 21.
4.5. Reversal of the SCOR-Field
 The SCOR-field (or, at least, a portion that contains the Australian flux patch), changes sign while the AD-field was weak. The patch now contains downward-directed flux (seeFigure 4, right) and again dominates the magnetic field at the hotspot site. At the time of eruption of DG 24 this “reversal” of the SCOR produces a transitional paleodirection antipodal to those seen for DGs 18, 21, and 23, and a north VGP in western Australia.
4.6. Reversal of the AD-Field and Conclusion of the Dynamo Process
 The axial dipole recovers in strength having changed its sign. DGs 25 through 28 record the completion of the polarity transition with increased values of paleointensity 14–21 μT yielding VADMs of 3.1–4.8 × 1022 Am2 [Mochizuki et al., 2011].
 Again, the presence of the flux patch seen in the historic field beneath Australia and its influence at the Society hotspot during transitional times (when the axial dipole has severely weakened) is well documented in the Plio-Pleistocene paleomagnetic record (seeFigure 1). However, its influence is also observed in records of older Cenozoic reversals obtained from Australian lavas [Hoffman, 1986; Hoffman et al., 2008]. With regard to the Matuyama-Brunhes, domination of the transitional field by the Australian flux patch extended over a considerable portion of Earth's surface [seeHoffman, 2000], as its apparent effect has been observed in Chilean lavas [Brown et al., 1994, 2004], Japanese sediments [e.g., Niitsuma, 1971; Tsunakawa et al., 1999], and Chinese loess [Zhu et al., 1994].
 The M-B transitional flows from the south side of the Punaruu Valley studied byChauvin et al.  have a 40Ar/39Ar determined age of 794.8 ± 6.8 ka [Singer et al., 2005], and hence, erupted near the onset of the reversal process during the so-called M-B precursor [e.g.,Hartl and Tauxe, 1996]. For the 30 flows recording the M-B transition exposed on the north side of the valley, a weighted mean age of 771 ± 16 (2σ) ka was obtained from five 40Ar/39Ar ages [see Mochizuki et al., 2011]. It is yet to be determined whether all of the 30 flows were erupted during the main stage of the Matuyama-Brunhes reversal, or whether some were erupted during the supposed precursor. Regardless, the scenario above at the very least is consistent with a time-lag during the reversal process between the change in polarity of the axial dipole and the change in sign for the remainder of the field about the globe.
 Available Matuyama-Brunhes transitional field records obtained from Tahitian lavas display remarkable features of the dynamo process. Beginning with the demise of the axial dipole field at the reversal onset, the magnetic flux pattern due to the remaining field at the Society hotspot produced at different times VGP clusters that were antipodal [Chauvin et al., 1990; Mochizuki et al., 2011]. There is ample evidence that a significant period of time elapsed while the ambient transitional field at the hotspot site retained either of the corresponding antipodal directions. Both the time-averaged historic non-axial dipole (NAD) field, as well as its reverse state, if present at the time of the Matuyama-Brunhes reversal, would have produced at the hotspot site vector directions associated with VGPs within the same two paleo-clusters deduced from the Punaruu Valley lavas. Not only do these findings provide further evidence that the vast majority of the NAD-field is tied to anomalies about the core-mantle boundary controlled by long-held physical conditions of the lowermost mantle, but also that they are indicative of a dual-stage reversal process. This dynamo process is one in which the axial dipole field, primarily generated at depth within the outer core, acts for the most part independently from the rest of the field. The remainder of the field is primarily, but not precisely, the non-axial dipole field. Rather, having its apparent origin nearer to the surface of the outer core, and containing a non-zero fraction of the axial dipole term, it has been more accurately termed the shallow-core-generated SCOR-field [Hoffman and Singer, 2008].
 The observable presence of the SCOR-field during reversals and other transitional events as well as evidence of its two “polarity” states, necessitates a physical mechanism. Separating sources of magnetic flux for the AD and SCOR fields [Hoffman and Singer, 2008], perhaps through stratification within the fluid outer core due to compositional variation [e.g., Buffett and Seagle, 2010; Helffrich and Kaneshima, 2010], a phase transition [Ozawa et al., 2011], or inner-outer, dual-convection structure in the fluid core [Miyagoshi et al., 2010], is clearly a possible explanation for the paleomagnetic observations. Regardless of the active mechanism, the paleomagnetic behavior recorded in lava flows exposed in the Punaruu Valley offers support for a deterministic view of dynamo activity during a change in polarity––a process that may have significant relevance for prediction of field behavior over Earth's surface during the next reversal.
 KAH wishes to thank senior project Cal Poly students Nate Padilla and Joseph Dierkhising for their assistance. We thank Mike Fuller and Nicholas Jarboe for helpful reviews. This work was supported by the National Science Foundation through grant EAR-1015360.
 The Editor thanks Nicholas Jarboe and an anonymous reviewer for their assistance in evaluating this paper.