The early Cenozoic experienced at least three short but major hyperthermals associated with disruptions of the global carbon cycle. The largest among those, the Paleocene-Eocene thermal maximum, was associated with a negative carbon isotopic excursion of ~ 2.5‰ that appears to be best explained by the thermal dissociation of methane hydrates due to an initial period of warming. The cause of the initial warming has been attributed to a massive injection of carbon (CO2 and/or CH4) into the atmosphere; however, the source of the carbon is as yet unknown. The emplacement of a large cluster of kimberlite pipes at ~56 Ma in the Lac de Gras region of northern Canada may have provided the carbon that triggered early warming in the form of exsolved magmatic CO2. Our calculations indicate that the estimated 900–1100 Pg of carbon required for the initial ~3°C of ocean water warming associated with the Paleocene-Eocene thermal maximum could have been released during the emplacement of a large kimberlite cluster. The coeval ages of two other kimberlite clusters in the Lac de Gras field and two other early Cenozoic hyperthermals indicate that CO2 degassing during kimberlite emplacement is a plausible source of the CO2 responsible for these sudden global warming events.
 The Paleocene-Eocene thermal maximum (PETM) is an early Cenozoic hyperthermal that occurred both on land and in the ocean. It is associated with a global disturbance in carbon isotopes and a global dissolution of CaCO3 sediments. On the basis of these and other environmental changes, the PETM has been related to an enormous release of CO2 and/or CH4 [Dickens et al., 1997]. Although the sequence of the events during the PETM are difficult to piece together, the available evidence indicates early warming, followed by a more drastic warming coincident with a negative shift in δ13C [Carozza et al., 2011; Leon-Rodriguez and Dickens, 2010; Secord et al., 2010; Sluijs et al., 2007; D. J. Thomas et al., 2002]. Although the negative carbon isotopic excursion (CIE) associated with the PETM has been attributed to carbon released from methane hydrates on the sea floor, the source of carbon that produced the early warming that destabilized these hydrates is yet unknown [Bowen et al., 2004; Bralower et al., 1997; Carozza et al., 2011; Dickens, 2000; Dickens et al., 1995, 1997; Hancock and Dickens, 2005; Kent, 2003; Lourens et al., 2005; Moore and Kurtz, 2008; Panchuk et al., 2008; Sluijs et al., 2007; D. J. Thomas et al., 2002; E. Thomas, 2003; E. Thomas and Shackleton, 1996; E. Thomas and Zachos, 2000; Tripati and Elderfield, 2005; Zachos, 2003, 2004; Zachos et al., 1993, 2001, 2005, 2007; Zeebe et al., 2009]. In this paper, we propose that the initial warming that occurred during the early stages of the PETM (and two other early Cenozoic hyperthermals) was caused by the release of CO2 during the eruption of a large cluster of kimberlite pipes in the Lac de Gras kimberlite field of the Slave Province, Canada.
 Several proposals have been offered to explain the PETM including: comet impact [Kent, 2003], global wild fires [Kurtz et al., 2003], volcanic activity [Svensen, 2004], and methane hydrate dissociation [Dickens et al., 1997]. The lack of extra-terrestrial tracer material in CIE deposits, such as iridium anomalies, and the discovery that magnetotactic bacteria is the cause of magnetically anomalous clay layers renders the impact hypothesis unlikely [Kopp et al., 2007]. The relative absence of anomalous graphitic black carbon in the PETM sediments argues against the possibility that global wildfires were the cause of the PETM [Moore and Kurtz, 2008]. Typical basaltic volcanic activity is also an unlikely source of this initial CO2 because of the low CO2 contents of basalts (< 1 wt %) associated with large igneous provinces and their slow eruption rates [Blundy et al., 2010; Caldeira and Rampino, 1991; Gerlach et al., 2002; Keeling et al., 1995; Sobolev et al., 2011]. Furthermore, volcanic SO2 released during large igneous province eruptions forms considerable amounts of atmospheric sulfate aerosols, which would cause global cooling [Self et al., 2006]. The release of thermogenic methane produced by igneous intrusions into organic-rich sediments would require that the CIE precedes the ocean water warming event, contrary to the observed isotopic data. Although methane from hydrate dissociation is thought to have played a pivotal role during the PETM, because it is highly depleted in 13C and a strong greenhouse gas, its role is most consistent with having caused the CIE and not the initial warming.
 The negative shift of δ13C (CIE) during the PETM is thought to be related to a geologically rapid (<10 kyr) carbon release that was significantly depleted in 13C [Röhl et al., 2000; Zachos et al., 2007]. Prior to the light carbon injection, a brief period of oceanic [Sluijs et al., 2007; D. J. Thomas et al., 2002] and continental [Secord et al., 2010] warming, as well as dissolution of seafloor carbonates [Leon-Rodriguez and Dickens, 2010], is thought to have occurred, indicating that the carbon input that caused the CIE is unlikely to have produced the initial warming. Recent models of carbon emissions during the PETM indicate that it is characterized by an initial period of warming and carbonate dissolution requiring ~1000 Pg of carbon, which is then followed by a rapid input of depleted carbon that caused the CIE [Carozza et al., 2011]. There are few carbon reservoirs available that could contribute to the early warming, without resulting in large δ13C excursions, because organic matter and methane hydrates are strongly depleted in 13C.
 The Earth's mantle is a vast reservoir of carbon, likely equivalent to all other carbon reservoirs combined. Normally, mantle carbon leaks out slowly through mid-ocean ridges and volcanoes, at rates of 10–3 Pg/yr of carbon [Caldeira and Rampino, 1991], three orders of magnitude lower than that required for hyperthermals (~1 Pg/yr; [Zeebe et al., 2009]). However, in rare circumstances, direct conduits between the deep mantle and the atmosphere are provided by rapid explosive kimberlite eruptions. In contrast, the CO2 contents of basaltic magmas in large volcanic provinces (i.e., shield volcanoes or flood basalts) are estimated to be << 1 wt % [Blundy et al., 2010; Gerlach et al., 2002; Keeling et al., 1995; Sobolev et al., 2011] and although large volcanic provinces provide vast amounts of lava (105–106 km3), thus vast amounts of CO2, they are emplaced over timescales of millions of years [Wignall, 2001]. Kimberlite fields are found on every continent, and throughout at least the last 1 billion years of the geological record, and their emplacement is distinct in that they occur in discrete fields and during distinct episodes (Figure 1).
 Magmatic kimberlite rocks have very high carbonate contents (CO2 ~20 wt %; [Brooker et al., 2001; Francis and Patterson, 2009; Nielsen and Jensen, 2005; Patterson et al., 2009; Sparks et al., 2006]), yet volcanoclastic kimberlite rocks are relatively carbonate-poor (CO2 < 5 wt %) and characterized by the final crystallization of monticellite, phlogopite, and diopside at the expense of the typical carbonate-rich matrix that characterizes magmatic kimberlite [Skinner and Marsh, 2004; Sparks et al., 2006]. That is as CO2 degasses from the ascending magma, the aSiO2 of the melt increased to levels that favor crystallization of silicate minerals (i.e., monticellite, phlogopite and diopside) in the simple CaO-MgO-SiO2-CO2 system [Canil and Bellis, 2008; Franz and Wyllie, 1967; Ootto and Wyllie, 1993]. Thus, the transition from magmatic kimberlite to volcanoclastic kimberlite is associated with the release of CO2. Kimberlite magma is estimated to rise to the surface in a matter of hours and have eruption durations on the order of weeks, thus significant amounts of CO2 can be released at high rates.
 The ages of the three early Cenozoic hyperthermals (~59.2, 55.5, and 53.2 Ma; [Charles et al., 2011; Hancock and Dickens, 2005; Lourens et al., 2005; E. Thomas and Zachos, 2000]) are indistinguishable from three of the four ages of kimberlite clusters that comprise the Lac de Gras kimberlite field of northern Canada. There are over 270 kimberlites in the Lac de Gras kimberlite field that range in age from 74 to 45 Ma, with at least four distinct age and spatial clusters: 59.0±0.7, 55.5±0.7, 53.2±0.3, and 47.8±0.3 Ma [Creaser et al., 2004; Graham et al., 1999]. Reversals in magnetic polarity in the Lac de Gras kimberlites [Cande and Kent, 1995] provide substantially better temporal resolution than that obtained by radiometric dating [Lockhart et al., 2004]. Seismic data indicate that the three younger kimberlite clusters appear to have been emplaced along distinct corridors that represent deep crustal or lithospheric dyke swarms composed of numerous dykes clustered within zones kilometers in width [Snyder and Lockhart, 2005]. Although only 50 of the over 270 Lac de Gras kimberlite pipes have precise ages [Creaser et al., 2004; Graham et al., 1999; Heaman et al., 2003], their age data exhibit a clustered distribution (59.0±0.7, 55.5±0.7, 53.2±0.3, and 47.8±0.3 Ma).
 Although the ages of the clusters within the Lac de Gras kimberlite field and those of the early Cenozoic hyperthermals have been previously reported, their striking correspondence is first demonstrated here (Figure 2). The lack of age dates for most of the kimberlite pipes in the Lac de Gras field requires a reconstruction of an age distribution for the undated kimberlite pipes to estimate the total number of pipes within each cluster. We have approached this reconstruction by assuming that the unknown age data will have a similar, albeit smooth, distribution as the 50 known ages. Considering that three of the age clusters (55.5±0.7, 53.2±0.3, and 47.8±0.3 Ma) have been shown to be spatially clustered [i.e., Lockhart et al., 2004, Figure 3], 29 additional undated kimberlite pipes that lie within these intrusive corridors were assigned the mean age for their respective clusters (55.5 = 19; 53.2 = 10). Although age clusters have been demonstrated in the Lac de Gras kimberlite field, it may be reasonable to assume that the remaining kimberlites of unknown age are similarly clustered; we have chosen a simpler distribution approach (Figure 3). The remaining 191 kimberlites are distributed across 45–75 Ma with a similar distribution as the dated kimberlite which have a median age of 55.9. The three methods are then stacked in a histogram resulting with a prominent mode at 55.9 Ma. The modes corresponding to four kimberlite clusters are at 59.0 Ma (n= 12), 55.9 Ma (n= 50), 53.2 Ma (n= 33), and 47.8 Ma (n= 7). The three older kimberlite age clusters correspond within error to three early Cenozoic hyperthermals, one of which is the PETM at 55.5 Ma.
4 Volume Estimates
 Kimberlite pipes in the Lac de Gras field have a cone morphology similar to that of the classic South African model [Field and Scott Smith, 1997]; however, they tend to be shallow (~950 m; [Moss et al., 2008; Nowicki et al., 2008]), thus they are volumetrically smaller. Phanerozoic strata are missing in the Lac de Gras area, although crustal xenoliths preserved within the kimberlite deposits suggest an original succession of Cretaceous marine shales, terrigenous arenite, and organic peat capping basement rocks during the Early Cenozoic, with variable estimates of thickness ranging from 100 to 300 m [Nowicki et al., 2004; Pell, 1997; L. D. Stasiuk et al., 1999; Sweet et al., 2003]. Detailed studies of kimberlite pipes in the Lac de Gras area indicate that their present depths vary between 400–1100 m [Moss et al., 2008; Nowicki et al., 2008], and thus emplacement depths of ~600–1300 m. Pipe morphology are characterized by steeply dipping walls (~80°; [Nowicki et al., 2008]), yielding an average volume of 3 × 107 m3 of kimberlite per pipe. Although kimberlite pipes often deviate from a simplified cone-shaped body [Nowicki et al., 2004], such deviations tend to increase volume and thus the foregoing would be a minimum volume estimate. Furthermore, kimberlite pipes are typically excavated multiple times (3–5) by repeated eruption episodes [Scott Smith, 2008] and thus the volume of the erupted kimberlite is likely many times that of the observed pipes. The Lac de Gras kimberlite cluster that is coeval with the PETM comprises ~50 known pipes corresponding to a minimum volume of erupted magma 5–8 × 109 m3.
 The capacity of kimberlite magma to transport CO2 is difficult to constrain because a gaseous CO2 phase is likely to develop as the magma undergoes depressurization during ascent [Russell et al., 2012; Sparks et al., 2006]. Russell et al.  demonstrated that the chemical assimilation of orthopyroxene by proto kimberlite magma (carbonatite) would release large amounts of CO2 due to the decreasing solubility of CO2 in a melt as the aSiO2 increases, thus it is likely that an additional 25–30 wt % of free CO2 gas is accompanying kimberlite eruptions. The presence of this free gas phase was first demonstrated as a requirement for kimberlite dyke initiation and propagation to explain the ultrarapid ascent of kimberlite from such great depths [Wilson and Head, 2007]. Despite the foregoing, the volume of a free CO2 gas phase is difficult to quantify; however, a minimum constraint on CO2 in magmatic kimberlite is simply estimated by the stoichiometry of carbonate and the carbonate content of magmatic kimberlite, ~20% of CO2 by weight [Brooker et al., 2001; Nielsen and Sand, 2008; Patterson et al., 2009; Sparks et al., 2006]. The density of magmatic kimberlite is ~3.2 × 106 g/m3 (80% Fo90 and 20% calcite), thus volumes of 5–8 × 109 m3 yield 2–3 × 1016 g of kimberlite. Kimberlite pipes, however, are dominated by volcanoclastic kimberlite rocks, which form through explosive degassing that occurs during kimberlite pipe excavation and thus are CO2 poor compared to magmatic kimberlite, typically <5% of CO2 by weight. Calculating the mass of CO2 degassed during emplacement as a 15% net weight loss, results in 2–4 × 1015 g of CO2 or 6.0–10.0 × 1014 g of carbon per kimberlite cluster, using the observed pipe volume. Studies regarding kimberlite eruption dynamics suggest the range of magma supply rates is 500–10,000 m3/s [Sparks et al., 2006], thus a kimberlite cluster of 55 pipes would degas the above mass of carbon between 5 and 200 days. The eruption of a single kimberlite cluster could release carbon at a rate of ~2–40 Pg/yr. As stated earlier, kimberlite pipes are typically excavated by repeated eruption episodes [Field and Scott Smith, 1997], and erupt larger volumes of magma than what remains in the pipes, thus the calculated duration required to produce 1000 Pg of carbon is estimated to be 200–3000 days (Figure 4).
 Our calculations suggest that kimberlite eruptions release carbon at a rate that far exceeds 1 Pg/yr. Estimates of minimum kimberlite eruption volumes that range from 106 to 108 m3 based on observable pipe volumes suggest durations of days to weeks dependent on the magma supply rate (500–10,000 m3/s; Sparks et al. ). These durations are, however, likely to be much shorter than the total duration of the eruptions: (1) the volume estimates are the lower bounds because they do not include the amount of magma ejected out of the kimberlite pipe, and (2) driving pressure in the source region may decline as an eruption proceeds, thus leading to an exponential decrease of magma supply rate, which results in lower estimated supply rates [Sparks et al., 2006; M. V. Stasiuk et al., 1993]. The inferred flow rate (500–10,000 m3/s) approaches those typical of Plinian eruption columns that reach 10 to 35 km into the stratosphere [Sparks et al., 1997, 2006]. The volume of kimberlite trapped in the pipe is only a small fraction of the total erupted magma and thus the eruptions are likely to be on the order of weeks to months in duration. Calculating the eruption duration if the magma supply rate is increased to 5000 m3/s, the duration required to produce 900–1100 Pg of carbon is reduced to 300–200 days (Figure 4). Furthermore, even these durations are considered longer than required due to the almost certain presence of a gaseous CO2 phase that develops during magma ascent.
 In addition to the much studied PETM, there are two other hyperthermals with dissolution horizons observed during the late Paleocene to mid-Eocene period that are similar to that associated with the PETM (~55.6 Ma); during the mid-Paleocene at ~59.1 Ma [Hancock and Dickens, 2005] and the early-Eocene at ~53.2 Ma [Lourens et al., 2005]. These horizons are similar to the horizon associated with the PETM and are characterized by the presence of a distinct reddish clay layer, an abrupt drop in carbonate content, and a pronounced peak in magnetic susceptibility that is thought to reflect increased clay content [Bernaola et al., 2007]. These two dissolution horizons are characterized by decreased oxygen isotope (δ18O) and carbon isotope (δ13C) values similar to the anomalies associated with the PETM, although of smaller magnitudes [Hancock and Dickens, 2005; Lourens et al., 2005]. The similarities between the three hyperthermals have led most workers to consider all three to be sedimentary responses to abrupt climatic change attributed to disruption of the carbon cycle.
 The correspondence between the ages of three early Cenozoic hyperthermals and three kimberlite clusters in the Lac de Gras kimberlite field suggests that kimberlite eruption may have provided the CO2 responsible for the initial global warming that destabilized marine methane hydrates. Although the analytical uncertainties associated with ages of the Lac de Gras kimberlites are relatively small (~1–2 Ma), it is difficult to discern whether individual kimberlite pipes in a cluster erupt concurrently or sporadically over longer time spans. The spatial trends of the clusters indicate that they belong to common dyke systems which, coupled with the high estimated ascent rates, suggest that the magma of individual kimberlite clusters is delivered to the kimberlite pipes simultaneously. Furthermore, detailed facies studies of re-sedimented kimberlite in the A154 kimberlite pipe (coeval with the PETM) indicate that material from adjacent kimberlite pipes was deposited in the still open pipe [Moss et al., 2008]. It is therefore likely that numerous kimberlite pipes composing a cluster erupt within relatively short time periods. Thus, our best estimates suggest that the eruption of a large kimberlite cluster is capable of delivering to the atmosphere the CO2 required to produce the early initial warming associated with three early Cenozoic hyperthermals. We propose that the eruption of kimberlite clusters within the Lac de Gras kimberlite field produced the CO2 responsible for the initial surface water warming associated with the early Cenozoic hyperthermals and that the transfer of those warm surface waters to intermediate depths led to thermal dissociation of sea floor methane hydrates and the resultant CIEs.
 The striking correspondence between the emplacement ages of three kimberlite clusters in the Lac de Gras field of the Slave Province and the three hyperthermals in the early Cenozoic suggests that they are genetically linked. Our calculations indicate that the eruption of three kimberlite clusters of the Lac de Gras field could have provided the CO2 required for the initial warming of ocean water of the Paleocene-Eocene thermal maximum and two other early Cenozoic hyperthermals. As the global oceans buffered the increased atmospheric CO2, ocean waters became increasingly acidified and produced the carbonate dissolution horizons. The transfer of warm surface ocean water to intermediate depths led to thermal dissociation of seafloor methane hydrates providing the isotopically depleted carbon that produced the carbon isotopic excursion. The rapid transfer of large amounts of CO2 from the mantle to the atmosphere by kimberlite eruptions may thus provide a genetic link between mantle processes and enigmatic climatic events in the geologic record.
 This paper arose out of discussions regarding paleo-climate with a number of colleagues at McGill University, mainly Al Mucci, Eric Galbraith, Dirk Schumann and David Carozza. In addition, the ideas in this paper have benefited from discussions with William Manarik and Boswell Wing as well as statistical advice from Vincent Van Hinsberg. We appreciate the rigorous and many reviews of our manuscript, the thoughtful and helpful comments from Dante Canil were greatly appreciated.