The P-wave Boundary of the Large-Low Shear Velocity Province beneath the Paciﬁc

The Large Low Shear Velocity Provinces (LLSVPs) in the lower mantle represent volumetrically 2 signiﬁcant thermal or chemical or thermo-chemical heterogeneities. Their structure and boundaries 3 have been widely studied, mainly using S-waves, but much less is known about their signature in 4 the P-waveﬁeld. We use an extensive dataset recorded at USArray to create, for the ﬁrst time, a 5 high-resolution map of the location, shape, sharpness, and extent of the boundary of the Paciﬁc 6 LLSVP using P(P di ﬀ )-waves. We ﬁnd that the northern edge of the Paciﬁc LLSVP is shallow 7 dipping (26 to 32 ◦ relative to the horizontal) and di ﬀ use ( ∼ 120 km wide transition zone) whereas 8 the eastern edge is steeper dipping (60 ◦ ) and apparently sharp ( ∼ 60 km wide). We trace the LLSVP 9 boundary up to ∼ 500 km above the CMB in most areas, and 700 km between 120 and 90 ◦ W at 10 the eastern extent of the boundary. Apparent P-wave velocity drops are ∼ 1-3 % relative to PREM, 11 indicating a strong inﬂuence of LLSVPs on P-wave velocity, at least in the high-frequency waveﬁeld, 12 in contrast to previous studies. Localised patches with greater velocity drops are detected, deﬁned 13 by high travel-time gradients. We identify these as a likely location of an Ultra-Low Velocity Zones 14 (ULVZs), matching the location of a previously detected ULVZ in this area. The boundary of a 15 separate low velocity anomaly, of a similar height to the LLSVP, is detected in the north-west 16 Paciﬁc, matching tomographic images. This outlier appears to be connected to the main LLSVP 17 through a narrow channel close to the CMB and may be in the process of joining or splitting 18 from the main LLSVP. We also see strong velocity increases in the lower mantle to the east of 19 the LLSVP, boundary is similar to that determined in high-resolution S-wave studies and follows the -0.4 % 21 ∆ V S iso-velocity contour in the S40RTS tomography model. Additionally, the LLSVP boundary 22 roughly matches the shape of the -0.4 % ∆ V P iso-velocity contour but deﬁnes an area more similar 23 to that deﬁned by the 0.0 % V P iso-velocity contour of the P-wave model GyPSuM. High resolution 24 P-wave velocity determination allows for estimation of the ratio of P- and S-wave velocity anomalies 25 ( R S,P ) which can be used to indicate dominantly thermal or chemical control of seismic velocities. 26 Our observations of the Paciﬁc LLSVP are consistent with a thermo-chemical anomaly whose shape 27 and boundary sharpness are controlled by proximity to active and past subduction. 28

the lateral extent of USArray [Meltzer et al., 1999] we are able to map the precise location of 113 the Pacific LLSVP, especially in the north and east of the LLSVP. We use travel-times of lower 114 mantle turning and core-grazing P-waves to determine the LLSVP boundary location. We utilise 115 a wide range of epicentral distances and back azimuths to track the vertical and lateral extent 116 of the LLSVP, respectively. We correct for both upper mantle (down to depths of 1600 km) and 117 crustal structure in the receiver region using the combined P-wave geodynamic tomography model 118 GyPSuM [Simmons et al., 2010] and the crustal structure model Crust1.0 [Laske et al., 2012]. We 119 are able to resolve the P-wave boundary of the LLSVP as the transition from positive to negative 120 travel time anomalies. The observed boundary tracks the 0% contour of GyPSuM well, but only 121 partially agrees with the S-wave velocity structure. between 300 and 600 operating stations. Using this network configuration allows for a wide sampling 128 of the lowermost mantle, both laterally and vertically, due to the large distance and azimuth range 129 covered by the stations. 130 We search the Reviewed Events Bulletin (REB) catalogue for events with magnitudes of 5.0 131 and above and select those that have distances from ∼85 • to ∼95 • from the centre-point of the 132 array. We concentrate on events from the Indonesian Arc, Tonga Trench, south-eastern Pacific, 133 and South-American Trench. The great-circle paths of these events to USArray are best suited 134 to sample the northern and eastern edges of the Pacific LLSVP. Although events at any depth, 135 including crustal events, are used in areas with low seismicity, we preferentially use events with 136 depths ≥30 km due to their simpler source mechanisms and to reduce travel-time anomalies from  Table 1). Stations are shown as inverted triangles with colour indicating year of deployment. Plate boundaries (red lines) from NUVEL- 1 [DeMets et al., 1990] are shown along with the area covered by the Pacific LLSVP, as defined by the -0.4 % VP contour in GyPSuM [Simmons et al., 2010], shown as the purple contours and shaded areas. The LLSVP contours are drawn at 2350-2500 km depth, 2500-2650 km, and 2650-2900 km, as defined by the depth slices in the tomography model. [Span 2 columns] For each event, data are de-spiked, re-sampled at 40 samples/s, and bandpass filtered. We filter 140 between 0.5 and 1.6 Hz, order 2, as this was found to be best to extract P and P diff arrivals from 141 the noise, where the order controls the rate of decay of energy with frequencies outside of the pass-142 band. Noisier events, where the P-wave is less clear relative to the noise, are filtered with order 3 143 or 4, defining a sharper frequency cut-off. To retain as much waveform information as possible, we 144 use the lowest possible order filter that clearly reveals the first arrivals. We only consider traces 145 at distances between 60 and 120 • to observe energy turning in the lower mantle. We then visually 146 inspect each trace to decide whether to include it in further processing, based on the P-wave arrival 147 being obvious above the noise. 148 We apply an adaptive stacking routine [Rawlinson and Kennett, 2004] to find the best alignment 149 of an ensemble of network stations and to determine travel-time deviations from a 1-D Earth 150 model. The adaptive stacking first applies a move-out correction based on distance through PREM 151 [Dziewonski and Anderson, 1981] and iterates to minimise residual travel-times by maximising the 152 6 amplitude and coherence of a stack of all traces. We correct for crustal structure on both the 153 source and receiver side and topography on the receiver side by applying travel-time corrections as 154 determined by Crust1.0 [Laske et al., 2012], and for upper mantle structure from the underside of 155 the crust down to 1600 km depth (the shallowest turning depth in our collection) by ray-tracing 156 through the P-wave component of GyPSuM [Simmons et al., 2010] ( Supplementary Figures 1 and   157 2). All travel-time deviations are calculated relative to PREM. The source side correction applied 158 is static and is only used for events shallower than 24 km as this is the thickness of the crustal layer 159 in PREM. Using the crustal thickness and velocities from PREM would, particularly in oceanic 160 regions, be inappropriate for waves travelling through the lithosphere. The crustal and mantle 161 corrections allow us to attribute the remaining travel-time residual to structure at depths greater 162 than 1600 km. Travel-time residuals are plotted at the location and depth of the turning point of 163 the ray as this represents the region in which the ray spends the most time and so has the potential 164 to accumulate the most residual time (Figure 2).

165
The boundary of the LLSVP is defined by obvious travel-time residual trends. We distinguish 166 between cases where the transition can be clearly identified, i.e. where both positive and negative 167 residuals are separated by zero residual, and where a trend towards the transition is observed, i.e.

168
where there are decreasing or increasing residuals but no change in sign. As the boundary location 169 changes with height, we consider each event individually and separate turning points into a series 170 of 100 km thick radial bins from the CMB upwards. Events with too few turning-points in a height 171 bin to show either the boundary or a trend towards the boundary are discarded.

172
For each event, we trace the LLSVP boundary in each height bin, with extent controlled by the 173 ray coverage. For each height bin, we then take the boundaries of all events together and define a 174 single boundary which best fits the individual measurements ( Figure 3). As an additional measure 175 of the LLSVP boundary, we use the magnitude of the gradient of travel-time residuals. We bin data 176 (in 0.5x0.5 • bins) and calculate the average residual. In regions where data fill adjacent bins, we 177 calculate the gradient of the travel-time residuals and choose a boundary defined by a line of highest 178 gradient ( Figure 4). Although this method is more robust as it analyses only the pattern of residual 179 travel-times, rather than the absolute value which can be affected by source depth errors, it is only 180 applicable in regions of dense sampling. In comparison, the absolute travel-time residuals can be 181 used to locate the boundary when sampling is poor, but the location will be less well constrained.

182
In general, in well sampled regions the results of both methods agree well.  The two events are closely located and sample the same region of the lower mantle. LLSVP contours from GyPSuM are shown as purple lines, as defined in Figure 1. No source-side crustal correction is applied as both events occur below the crust. Inset shows source location as a yellow star, and ray turning points as black circles. [Span 2 columns] 4 Results

184
We find laterally varying delay times at all depths with some of the largest deviations being seen 185 in the lowermost 300 km of the mantle. We observe patterns of the delay times that are consistent boundary can also be followed in height above the CMB, dependent on ray coverage.

194
By calculating the gradient of the travel-time residuals and considering both the magnitude 195 8 (trend of changing travel-time residuals) and its direction, we are better able to observe structure 196 within regions of predominantly fast or slow delay-times. As such, we observe sharp increases in : Best fitting P-wave LLSVP boundary determined using travel-time anomalies in the height bin from the CMB to 2800 km depth, as in Figure 3. The LLSVP boundary determined using S-waves [He and Wen, 2012] and the -0.4 % VP iso-velocity contour from the GyPSuM tomography model [Simmons et al., 2010] are shown as dark grey and purple lines, respectively. The region of a suspected ULVZ, determined from a high travel-time residual gradient in the height bin from the CMB to 2800 km depth, is shown by the dotted blue line and red ellipse.
We use two methods to define the P-wave LLSVP boundary: the waveforms further in this paper, we note that there is evidence for wavelet broadening along the 218 northern edge of the LLSVP which is not observed for events sampling the eastern edge. However, 219 12 this could be due to a smaller range of azimuths available for the eastern side, suggesting that the 220 boundary may not actually be sampled. This is unlikely since the boundary is clearly evident in 221 the travel-time residuals. Alternatively, this region of varying residual sign could be the result of 222 the ray geometry relative to the edge of the LLSVP causing rays to travel through both slower and 223 faster material away from the turning point of the ray. On the eastern side of the LLSVP rays are 224 more likely to travel either outside or inside the LLSVP, but due to the small Fresnel zone of the 225 data are unlikely to travel through both, and so do not show a broad variable region. Using residual travel-times determined at different heights, we trace the boundary of the LLSVP 227 from the CMB up to ∼500 km above the CMB and ∼700 km in some regions (Figures 3 and 4). 228 We observe variations in the steepness of the detected boundary between the east, and the north 229 and north-east sides of the LLSVP, which are the regions best resolved by our data. Using cross 230 sections through the turning points, we are able to visually define the boundary and estimate the 231 slope. We find that the eastern edge is steeper at ∼60 • (relative to the horizontal) dipping roughly 232 to the north-east, while the northern edge is shallower with slopes between 26 and 32 • , dipping 233 north-west (Figures 6c and 6d).  Figures 3 and 4). 256 We use the REB catalogue due to the high quality source locations reported (Supplementary There is an imperceptible variation in turning point locations in both circumstances. Additionally, 267 the effect on both turning point location and delay time is negligible for events deeper than 10 km.

268
Therefore, source depth uncertainty is only significant for events shallower than 10 km depth. The  The applied crustal corrections for the source are static. We believe that this is suitable, given 274 that the difference in the section of the crust sampled by two rays, even with vastly different take-off 275 angles and back-azimuths, is negligible when compared to the 1 • resolution of the crustal model 276 used [Laske et al., 2012]. Therefore, inaccuracies in the source depth will affect the delay-time for 277 all stations in the same way, increasing or decreasing all delay-times as a DC shift. The transition 278 from positive to negative delay times will be affected, and so will the point at which the boundary 279 is defined. However, the pattern of delays relative to each other will not change. In these situations, 280 therefore, the magnitude of the gradient is a better measure of the location of the LLSVP boundary.

281
Gradients can only be calculated where there are rays sampling adjacent locations. If sampling of 282 the lower mantle is sparse then gradients cannot be determined. Also, care must be taken not to 283 pick sharp changes in gradient resulting from lack of sampling as a boundary, a problem which can 284 be easily avoided when using travel-time residuals.
285 Body-waves have been shown to be sensitive to off-ray-path structure [Marquering et al., 1998[Marquering et al., , 286 1999]. However, this is only significant for intermediate and long period waves. Using high-frequency 287 P-waves (∼1 Hz) with the related small Fresnel zone means that this is irrelevant and ray theory 288 approximation is still valid. The first Fresnel zone for P-waves sampling the lower mantle with a 289 dominant frequency of 1 Hz is ∼100-140 km [Sato and Fehler, 2008], equivalent to the distance 290 between 3 stations. This indicates that multi-pathing may affect the exact location at which the 291 LLSVP boundary is defined, but the location will still be accurate to within 2 • at the turning point 292 of the ray.

293
The remaining delay-times, therefore, represent the deviation of the wave arrival time from 294 that predicted by a 3-D tomography model. Any further errors are due to tomography models 295 not sufficiently explaining Earth structure on the scales imaged here and are unavoidable in high 296 frequency studies [Thorne et al., 2013b].

304
In the north-west (Figures 3b and 4b) the LLSVP boundary slopes fairly shallowly to the north-305 west. However, ray coverage in this location does not allow the boundary to be traced to depths 306 less than ∼2600 km, 300 km above the CMB. At the CMB, the boundary is mapped further north 307 than either the high resolution S-wave study [He and Wen, 2012] (Figures 3 and 4), it seems that 347 the boundary is well defined. However, as the boundary close to South America can be traced to ∼900 km above the CMB (Figure 4), we conclude that this edge is likely not the eastern edge 349 of the Pacific LLSVP but the transition to some other velocity structure and might be related to 350 subduction structures in this region Lay, 2003, Thomas et al., 2004, Hutko et al., 351 2006, Thorne et al., 2007, Hutko et al., 2009.

352
Knowledge of the sharpness of the transition will help to resolve arguments about the degree to shown to also have laterally varying boundary steepness [Ni and Helmberger, 2003a]. The northern 363 edge of the African LLSVP is reported to be steeply overturned [Ni et al., 2002], although other 364 studies show that, despite the boundaries being steep, dipping at between 28 and 70 • , they are 365 not overturned [Wen, 2001, Wang and Wen, 2004, Helmberger et al., 2009].

366
In our study we too find boundaries dipping as steeply as that of the African LLSVP and also 367 with lateral variation in the steepness ( Figure 6). where there is no active or recent subduction. We observe that the transition from positive to 384 negative delay times is sharper on the eastern side of the LLSVP than on the northern side ( Figure   385 6), possibly owing to the closer proximity to an active subduction zone on the eastern side. The 386 observation of a steeper eastern edge than northern edge agrees with previous S-wave studies [He 387 and Wen, 2012], indicating that this is a robust observation.

388
Previously, some studies have reported the LLSVPs to show little to no P-wave velocity change 389 in the lower mantle [Masters et al., 2000 where 390 S-wave velocities change more significantly. However, we find substantial P-wave velocity variations:  Alternatively, a 500 km thick layer would have to have ∆V P of -2.2 to -2.9 %, relative to PREM.

401
However, we accept that this is a grossly simplified calculation and constraining the wavespeed 402 deviation with waveform modelling would be preferable. This consideration notwithstanding, these 403 values are similar to that observed in past studies using P diff passing through the African LLSVP 404 [Wen, 2001.

405
The relationship between the P-and S-wave boundaries can help determine the material proper- ing tomography models indicate that the seismic velocity of the mantle is dominantly controlled by 411 chemical variations Woodhouse, 1996a,b, Trampert et al., 2004, Della Mora et al., 412 2011]. However, the validity of this method has been disputed [Schuberth et al., 2009, Davies et al., 413 2012]. Nonetheless, comparing our 1-D velocity calculation, with ∆V S ∼-5 % calculated for the 414 Pacific LLSVP [He and Wen, 2012], translates to a V S /V P ratio of 1.7 to 3.3. The median value of 415 ∼2.4 is higher than other high-frequency lower mantle studies [Wysession et al., 1999, Sun et al., 416 2007], but agrees with large-scale studies Woodhouse, 1996b, Mosca et al., 2012].

417
Using this estimate indicates that the Pacific LLSVP, at least in this region, can be explained by 418 a combination of chemical and thermal anomalies [Karato, 2003]. It should be said that this is a 419 maximum estimate for the magnitude of ∆V P , hence a minimum value of R, and by using either We propose that the boundary width, or sharpness, and steepness is related to mantle dynam- Corrections are applied to travel-times calculated for rays using PREM to account for 3-D 468 mantle structure and crustal heterogeneity on the source and receiver sides. We use the GyPSuM 469 P-wave tomography model [Simmons et al., 2010] and CRUST1.0 model [Laske et al., 2012]  Black circles indicate sources deeper than 24 km to which no source-side crustal correction is applied. Triangles show corrections for crustal structure at each station in USArray, averaged over all events which use the station. Background shows corrections for mantle structure along the whole ray-path at the turning point of each ray, averaged over all rays. Crustal structure is determined from CRUST1.0 [Laske et al., 2012] and mantle structure is determined from the P-wave component of the GyPSuM model [Simmons et al., 2010]. The -0.4 % VP contour from GyPSuM is shown by the purple line. The boundary of the Pacific LLSVP determined with S-waves travel-times residuals is shown (grey line) [He and Wen, 2012], along with the -0.4 % and 0 % VP contours in S40-RTS (purple and black line, respectively) [Ritsema et al., 2011].

Highlights
• P-wave travel-time deviations caused by the Pacific LLSVP are detected at USArray • We create the first high-resolution map of the LLSVP edge detected with Pwaves • LLSVP boundary is seismically diffuse in the central Pacific, and sharp in the east • The edge shape and sharpness may be linked with dynamics and subduction history • P-wave LLSVP boundary roughly matches the S-wave boundary, except at the eastern edge *Highlights (for review)