Subduction of lithospheric plates at convergent margins leads to transport of materials once close to or at the surface of Earth to great depths. Some of them later return to the surface by magmatism or degassing, whereas others end up being stored in the mantle for long periods of time. The fate of carbon-bearing minerals in subduction is of particular interest because they can arbitrate the long-term availability of CO2 at the surface. However, there are major gaps in the understanding of even the most fundamental processes that modulate carbon pathways at mantle depths. We use geodynamic models to understand carbonate pathways upon subduction in the form of large carbonate platforms, which were common in the Tethys realm of Europe. We conducted a series of geodynamic forward models for a 1-km-thick carbonate platform entering subduction. We show that most of the carbonate load detaches from the subducting slab and rises up diapirically through the mantle wedge and eventually mixes with the mantle lithosphere. A smaller fraction gets accreted under the forearc, whereas an even smaller fraction descends deeper into the mantle. The cold diapiric plume has a significant role in retarding silicate mantle melting above these subduction zones and promoting the formation of small-volume carbonate-rich melts and, in some cases, alkaline silica-undersaturated silicate melts. We propose that large amounts of CO2 can be stored as carbonate in the shallow uppermost lithospheric mantle.The fate of subducted carbon is of great importance for understanding deep carbon budgets on geological time scales (Dasgupta and Hirschmann, 2010; Pall et al., 2018). Carbon enters subduction zones primarily as sedimentary carbonate minerals (calcite or dolomite) in various forms, including carbonate-rich pelite, abundant calcite fracture fill in greywackes, and various forms of organic carbonate precipitation, among others (Edmond and Huh, 2003; van der Ploeg et al., 2019). One of the potentially most important sinks of carbon, subduction of large carbonate platforms (Handford and Loucks, 1993), remains poorly understood. Some carbon is incorporated in fold-and-thrust belts into the mid- to lower crust of thickened collisional orogens (Selverstone and Gutzler, 1993). Subducted carbonate is thought to be released from the forearc (Stewart and Ague, 2020) and the subarc region (Plank and Manning, 2019) by decarbonation reactions, some involving melt (Duncan and Dasgupta, 2014). The majority of that released carbon, liberated as CO2, makes its way back to the surface either via arc magmatism (Lee and Lackey, 2015) or through fault pathways as gas (Italiano et al., 2017). A small component of subducted carbon remains in the mantle lithosphere (Kelemen and Manning, 2015) or is transferred to the deep mantle, and, if deep enough, it can turn into diamonds and other high-pressure carbon-bearing phases (Shirey et al., 2013).Direct evidence for carbonate in the mantle comes from peridotite xenoliths that contain primary carbonate (Ionov et al., 1995; Ducea et al., 2005; Chen et al., 2018); these were found in xenolith suites from numerous localities and ages globally. Their primary origin is argued based on textural evidence, and the subducted, initially sedimentary, origin is demonstrated based on stable isotopes (Ducea et al., 2005). Not all carbon coming from the mantle is recycled (Stachel et al., 2009).Carbon dioxide budgets in subduction have been quantified in recent studies (e.g., Kelemen and Manning, 2015; Galvez and Pubellier, 2019; Plank and Manning, 2019). These studies acknowledge that Mediterranean subduction systems could significantly change these budgets via subduction of large amounts of carbonate. Various basins of the Tethys realm in Europe (van Hinsbergen et al., 2020) were relatively narrow and shallow basins (some were true oceanic basins; others, highly extended continental; Fig. 1) commonly covered by marine carbonate sequences of Mesozoic or Cenozoic age. The low latitude and mostly east-west orientation of these margins made them prone to carbonate formation, very different from most modern subduction systems that are found around the Pacific Ocean and that incorporate nontrivial, but much smaller, amounts of carbonate in their budget (Plank and Langmuir, 1988). Much of that carbonate mass is seen in Alpine-aged European fold-and-thrust belts as well as undeformed sequences on the continental margins that escaped subduction. One particularly important aspect of carbon pathways into the mantle that has only recently been investigated (Chen et al., 2021) is the possible mechanical transport by diapirism of large carbonate masses in the uppermost mantle above subduction zones. This process is similar to the relamination hypothesis for siliciclastic sediments (Hacker et al., 2011; Miller and Behn, 2012), which is based on earlier geodynamic forward models (Currie et al., 2007).In our study, we used geodynamic models to investigate the pathways and effects of subduction of large carbonate sequences (hundreds of kilometers wide, and a minimum of 1 km thick). These models simulate carbonate sequences that could be involved in interaction with subduction environments: (1) shallow-water carbonate sequences (including here carbonate platforms, ramps, mud-mounds, and rimmed and non-rimmed shelves) and pelagic carbonate platforms developed along continental margins; and (2) pelagic carbonate seafloor sediments developed on oceanic crusts. Figure 1 shows two snapshot paleogeographic reconstructions of the extent of carbonate platforms in the Late Triassic and the Late Jurassic, respectively, in the Tethyan domain of today's Europe (from Cosentino et al., 2010). Only a small fraction of this mass of carbonate remains preserved in the fold-and-thrust belts of Alpine Europe or on foreland basin margins and other platformal regions that escaped subduction (e.g., Moesia [in the Balkans]). Geologic constraints for the size and fate of these carbonate volumes are primarily drawn from the European Alps and Carpathian Mountains, and are described in the Supplemental Material1.We used the SOPALE numerical code (http://geodynamics.oceanography.dal.ca/sopaledoc.html) to model the coupled thermal-mechanical evolution of the lithosphere–upper mantle system (Fig. 2; Figs. S1–S3 in the Supplemental Material). The two-dimensional model domain is 2000 km wide and 600 km deep. The initial model geometry is shown in Figure S1 and consists of an oceanic plate between two continental plates. Similar results would be obtained if a highly extended and thinned continental plate, such as those found in passive-margin continental shelves, were used instead of a true oceanic plate. The oceanic plate (7-km-thick oceanic crust and 64-km-thick mantle lithosphere) is overlain by a 1 km layer of carbonates, making it a 72-km-thick plate corresponding to a ca. 30 Ma oceanic plate (see the Supplemental Material for justification of the choice of thickness of the carbonate layer), and subducts below a 50-km-thick continental plate at a rate of 2.5 cm/yr. The 1 km thickness is constrained by geologic observations but is also the lower limit for computation here; thinner layers cannot be modeled yet due to computational limitations. Additional effects such as extensive serpentinization of the mantle in the downgoing plate can enhance buoyancy of the plate Other effects such as a thinner carbonate layer would decrease the effect; they are not considered here. Our focus is the carbonate layer atop the oceanic or highly attenuated continental plate, assumed to be predominantly carbonate mixed with 20% pelagic and terrigenous sediments. The density of this layer is 2700 kg/m3 at 200 °C, making it 100 kg/m3 less dense than upper and mid continental crust and ~650 kg/m3 less dense than mantle when materials are at the same temperature. Details on parametrization and constraints of the reference model and alternative models, as well as model initiation and run details, are presented in the Supplemental Material.The extent and composition of syn–carbonate diapirism partial melts predicted to be found above the most volatile-saturated solidus in Figure 3 were modeled with the pMELTS algorithm (Ghiorso et al., 2002) between 1 GPa and 3 GPa over a temperature range of 900–1300 °C at nickel–nickel oxide (NNO) oxygen fugacity. Starting composition consisted of dry (Hirose and Kushiro, 1993) and wet (0.5 wt% H2O) peridotite (average natural lherzolite with no carbonate) mixed with 2.5% to 15% CaCO3 in increments of 2.5%. Each mixture was run at constant pressure, starting with 1 GPa in steps of 0.5 GPa with 3 °C increments from initial to final temperature. Carbonate in the dry mix produces a gradual increase in melt fraction to as much as 12% at 10% carbonate in the mix, especially in the low pressure–high temperature region. However, melt fraction does not exceed 4% between 1.5 and 2.5 GPa. The presence of 0.5% H2O further increases melt fraction, although not significantly.Within the model domain, material movement is driven by both the kinematically imposed convergence and internal buoyancy forces arising from thermal and compositional variations. Figure 2, and Animation S1 in the Supplemental Material, show the evolution of the model. The model starts at 0 m.y., following initiation of subduction (Fig. 2A). The incoming carbonates accumulate at the plate margin, initially forming a wide wedge, and with continued plate convergence, the carbonates are carried into the mantle by the subducting plate (Fig. 2B). At ~14 m.y., the carbonate layer in the mantle becomes unstable and begins to buoyantly detach from the oceanic plate as a diapir that initiates at 80–100 km depth (Fig. 2C). Similar results are obtained if the thickness of the carbonate is larger (we performed model runs with carbonate as much as 2 km thick [not pictured in Figure 2]). Diapirism also takes place at lower density contrasts between carbonate and the upper mantle, to as little as 250 kg/m3 contrast (not pictured). Thicker upper-plate lithosphere would lead to delayed and deeper detachment, but models show that diapirs always form under the general scenario envisioned in our study. The upwelling diapir rapidly ascends through the mantle wedge and penetrates the continental mantle lithosphere (Fig. 2D). Ongoing subduction creates continuous diapirism that results in (1) a ponding of the carbonates either at the lithosphere-asthenosphere boundary or at the continental Moho, and (2) cooling of the mantle wedge (Figs. 2E and 2F). Additional models show that carbonate diapirism and accumulation in the mantle wedge corner readily occur for a wide range of conditions (Fig. S3).Our models predict that carbonate diapirs mix with the convective mantle and cool the mantle wedge during subduction before being added to the bottom of the continental lithosphere. A weak upper-plate lithosphere model allows for more thorough mixing of the existing mantle lithosphere (some of which is engaged in a delamination process from the backarc side) than a strong lithosphere model. For a strong lithosphere, the asthenospheric peridotite-carbonate mix does not penetrate the existing mantle lithosphere, but rather attaches to its bottom and adds to the total thickness of mantle lithosphere.Seismic velocity calculations demonstrate that over a realistic range of pressures and temperatures for the uppermost mantle, peridotite-carbonate mixes produce a seismically slow mantle, with as much as 3% lower velocities in the lowermost lithosphere relative to asthenospheric mantle. It is unlikely that such a low-density material would be prone to subsequent convective removal, so this may add carbonate to the bottom of, or into, the mantle lithosphere.Vertical thermal profiles above where the slab is located at 100 km depth at 14 and 23 m.y. after subduction initiation (Fig. 2) are shown in Figure 3 for the mantle wedge compared to the carbonated fertile peridotite solidus and the H2O + CO2–saturated peridotite solidus (Falloon and Green, 1989). This diagram shows that carbonate diapirism in the mantle wedge for relatively short-lived subduction systems retards melting because cold materials are being mixed into the mantle wedge. The H2O + CO2–saturated peridotite solidus is depressed relative to the water-only saturated peridotite, but the addition of cold materials from the downgoing slab counteracts that in terms of melt productivity. Given that the fluid-saturated fertile peridotite solidi shown in Figure 3 indicate the initiation of partial melting at the lowest temperatures possible and that most likely the actual peridotite in the wedge is less fertile (i.e., less clinopyroxene rich) and is likely not volatile saturated, and very likely not water saturated, the amount of melting in the mantle wedge of these systems is likely insignificant if carbonate material is introduced diapirically. The maximum predicted partial melt fraction under saturated conditions (H2O and CO2) is ~4%. For a near-pure limestone diapir, there is no water involved, but we assume that the presence of minor silicic sediments can add some water to the system, otherwise melting would not occur. Over time, after consumption of the carbonate platform, the wedge heats up and regains a hotter thermal regime by additional convection, and conductive heating of the carbonate diapir and its attachment to the bottom of the lithosphere take place (see below).Forward melting modeling results show that minor amounts of partial melts are either calcio-carbonatites at 2.5 GPa or highly undersaturated feldspathoid-bearing leucitite and phonolites at 1.5 GPa. It is unlikely that the mantle wedge was ever less than ~45 km thick in a Tethyan scenario, given that the upper plates of such subduction systems were continental. Our models suggest that under a scenario of subduction of a sizable amount of carbonate, melting is retarded in most situations, which is consistent with the paucity of magmatism in places like the European Alps. Rather than producing a normal-size magmatic arc, such regions (the Carpathian Mountains, Alps, etc.) have either (1) no magmatism at all for long periods in times of convergence, or (2) very limited and alkali-rich magmatism that may be explained by the processes described here.The pMELTS algorithm is not particularly accurate and at low melt fraction; a better path forward in understanding the extent and composition of melts is via experiments. Recently, Chen et al. (2021) provided some preliminary experimental data on precisely the process and physical conditions we envision here, and they obtained remarkably similar results to the pMELTS predictions (see the Supplemental Material).We quantify how much CO2 returns to the lithosphere in the upper plate by solid-state diapirism versus the amount of carbonate lost to the forearc and the fraction that is delivered to the deeper mantle along the subduction plane. A large fraction, ~35%, remains in the forearc; ~10% descends deeper into the mantle. The return via diapirism to the upper-plate lithosphere or the bottom of the upper plate is predominant; it appears to make up approximately 40%–50% of the subducted carbonate mass in the models presented here. This mechanism is important, needs to be included in the mass-balance budgets of subducted CO2 (Plank and Manning, 2019), and predicts that the shallow supra-subduction mantle can be an important long-term sink for CO2 under this scenario. For example, if one considers a 500-km-long carbonate (100% calcite) platform being subducted at a rate of 5 cm/yr and that ~40% of the subducted material returns to the upper plate through diapirism, as envisioned here, ~5 Mt of carbon per year are added to the upper plate from only one large platform.Future tests need to quantify the amount of mantle-derived materials that may have isotopic signatures indicative of carbonate additions in regions like the Alpine-Carpathian chain; recent basaltic fields exist to test for that (Wilson and Downes, 2006). Ca isotope values (DePaolo, 2004) in sedimentary carbonates are typically distinct from mantle values (Amsellem et al., 2020). Mantle lithospheric xenoliths, also common regionally (Downs 2001), can be investigated for interaction with cryptic carbonate liquids using the techniques put forward for other regions (Chen et al., 2018). Frozen calcio-carbonatites in xenoliths from the Tethyan realm, like those described in detail from the Persani volcanic field (Romania; Chalot-Prat and Arnold, 1999), can be studied to gain information on the origin of carbonate, the magnitude of carbonate recycling, and the timing of percolation of the lithospheric mantle.We acknowledge support from a Babes-Bolyai University (Romania) Fellowship (grant CNFIS-FDI-2021–0061) and Romanian UEFISCDI (Executive Unit for the Financing of Higher Education, Research, Development and Innovation) grant PN-III-P4-ID-PCCF-2016-0014. C. Currie acknowledges support and computing resources from WestGrid (Victoria, British Columbia, Canada) and Compute Canada (Toronto).

中文翻译：

---

俯冲入上地幔的碳酸盐台地底辟

岩石圈板块在会聚边缘的俯冲导致一旦接近地球表面或在地球表面处的物质被输送到很深的地方。其中一些后来通过岩浆作用或脱气返回地表，而另一些最终长期储存在地幔中。俯冲中含碳矿物的命运特别令人感兴趣，因为它们可以决定地表 CO2 的长期可用性。然而，即使是在地幔深处调节碳路径的最基本过程，在理解上也存在重大差距。我们使用地球动力学模型来了解以大型碳酸盐平台形式俯冲时的碳酸盐路径，这在欧洲的特提斯地区很常见。我们对进入俯冲的 1 公里厚的碳酸盐台地进行了一系列地球动力学正演模型。我们表明，大部分碳酸盐负荷从俯冲板片上分离出来，并通过地幔楔底面上升，最终与地幔岩石圈混合。较小的部分在前弧下吸积，而更小的部分则深入地幔。冷底辟柱在延缓这些俯冲带上方的硅酸盐地幔熔融和促进小体积富含碳酸盐熔体的形成方面具有重要作用，在某些情况下，碱性二氧化硅-不饱和硅酸盐熔体的形成。我们提出大量的 CO2 可以作为碳酸盐储存在最浅层岩石圈地幔中。俯冲碳的命运对于了解地质时间尺度上的深层碳收支非常重要（Dasgupta 和 Hirschmann，2010；Pall 等，2018 ）。碳进入俯冲带主要以沉积碳酸盐矿物（方解石或白云石）的形式以各种形式进入，包括富含碳酸盐的泥质岩、灰瓦中丰富的方解石裂缝填充物以及各种形式的有机碳酸盐沉淀等（Edmond 和 Huh，2003；van der Ploeg 等人，2019）。潜在的最重要的碳汇之一，大型碳酸盐台地的俯冲（Handford 和 Loucks，1993 年）仍然知之甚少。一些碳在褶皱和逆冲带中并入加厚碰撞造山带的中下地壳（Selverstone 和 Gutzler，1993）。俯冲碳酸盐被认为是通过脱碳反应从弧前（Stewart 和 Ague，2020 年）和弧下区域（Plank 和 Manning，2019 年）释放出来的，其中一些反应涉及熔体（Duncan 和 Dasgupta，2014 年）。大部分释放的碳以 CO2 形式释放，通过弧形岩浆作用（Lee 和 Lackey，2015 年）或通过断层通道作为气体（Italiano 等人，2017 年）返回地表。一小部分俯冲的碳保留在地幔岩石圈中（Kelemen 和 Manning，2015）或转移到地幔深处，如果足够深，它可以变成钻石和其他高压含碳相（Shirey 等人） ., 2013). 地幔中碳酸盐的直接证据来自含有初级碳酸盐的橄榄岩捕虏体 (Ionov et al., 1995; Ducea et al., 2005; Chen et al., 2018); 这些是在全球许多地区和年龄的捕食者套房中发现的。它们的主要起源是基于结构证据论证的，俯冲的最初沉积起源是基于稳定同位素证明的（Ducea et al. , 2005)。并非所有来自地幔的碳都被回收（Stachel 等人，2009 年）。最近的研究已经量化了俯冲过程中的二氧化碳预算（例如，Kelemen 和 Manning，2015 年；Galvez 和 Pubellier，2019 年；Plank 和 Manning，2019 年） . 这些研究承认，地中海俯冲系统可以通过大量碳酸盐的俯冲来显着改变这些预算。欧洲特提斯领域的各种盆地（van Hinsbergen 等，2020）是相对狭窄和浅的盆地（一些是真正的海洋盆地；另一些是高度延伸的大陆盆地；图 1）通常被中生代或新生代的海相碳酸盐岩层序所覆盖年龄。这些边缘的低纬度和大部分东西方向使它们容易形成碳酸盐，与在太平洋周围发现的大多数现代俯冲系统非常不同，这些俯冲系统在其预算中包含了非平凡但更少量的碳酸盐（Plank 和 Langmuir，1988 年）。大部分碳酸盐物质出现在高山时代的欧洲褶皱和逆冲带以及大陆边缘的未变形序列中，这些序列逃脱了俯冲。直到最近才研究的碳途径进入地幔的一个特别重要的方面（Chen 等，2021）是俯冲带上方最上层地幔中大型碳酸盐块的底辟可能通过底辟作用进行机械运输。这一过程类似于基于早期地球动力学正演模型 (Currie et al., 2007) 的硅质碎屑沉积物再分层假说 (Hacker et al., 2011; Miller and Behn, 2012)。在我们的研究中，我们使用地球动力学模型来研究大型碳酸盐序列（数百公里宽，至少 1 公里厚）俯冲的路径和影响。这些模型模拟了可能与俯冲环境相互作用的碳酸盐岩层序：（1）浅水碳酸盐岩层序（包括这里的碳酸盐台地、斜坡、泥丘以及有边和无边架）和沿大陆发育的中上层碳酸盐台地边距；(2) 在洋壳上发育的中上层碳酸盐海底沉积物。图 1 显示了今天欧洲特提斯域晚三叠世和晚侏罗世碳酸盐台地范围的两个快照古地理重建（来自 Cosentino 等，2010）。在欧洲高山的褶皱逆冲带或前陆盆地边缘和其他未发生俯冲的台地区域（例如，[巴尔干半岛的 Moesia]）中，只有一小部分碳酸盐仍然保存着。这些碳酸盐体积的大小和命运的地质限制主要来自欧洲阿尔卑斯山和喀尔巴阡山脉，并在补充材料1中进行了描述。我们使用了 SOPALE 数字代码（http://geodynamics.oceanography.dal.ca/sopaledoc .html）模拟岩石圈-上地幔系统的耦合热力学演化（图2；补充材料中的图S1-S3）。二维模型域宽 2000 公里，深 600 公里。初始模型几何如图 S1 所示，由两个大陆板块之间的海洋板块组成。如果使用高度扩展和变薄的大陆板块（例如在被动边缘大陆架中发现的那些）代替真正的海洋板块，将获得类似的结果。海洋板块（7 公里厚的洋壳和 64 公里厚的地幔岩石圈）被 1 公里的碳酸盐层覆盖，使其成为 72 公里厚的板块，相当于约 30 Ma 海洋板块（有关选择碳酸盐层厚度的理由，请参阅补充材料），并以 2.5 厘米/年的速率俯冲到 50 公里厚的大陆板块下方。1公里的厚度受地质观测的限制，但也是此处计算的下限；由于计算限制，更薄的层还不能建模。其他影响，例如下行板块中地幔的大面积蛇纹石化可以增强板块的浮力 这里不考虑它们。我们的重点是海洋或高度衰减的大陆板块顶部的碳酸盐层，假定主要是碳酸盐与 20% 的远洋和陆源沉积物混合。该层在 200 °C 时的密度为 2700 kg/m3，比上中部大陆地壳的密度低 100 kg/m3，在相同温度下的物质比地幔的密度低约 650 kg/m3。补充材料中介绍了参考模型和替代模型的参数化和约束的详细信息，以及模型启动和运行的详细信息。预测在图 3 中最易挥发饱和固相线上方发现的合成碳酸盐底辟部分熔体的范围和组成是用 pMELTS 算法（Ghiorso 等，2002）在 1 GPa 和 3 GPa 的温度范围内建模的。 900–1300 °C，氧化镍 (NNO) 氧逸度。起始成分包括干的（Hirose 和 Kushiro，1993 年）和湿的（0.5 wt% H2O）橄榄岩（不含碳酸盐的普通天然二长石），混合了 2.5% 到 15% CaCO3，增量为 2.5%。每种混合物都在恒定压力下运行，从 1 GPa 开始，以 0.5 GPa 为步长，从初始温度到最终温度以 3 °C 的增量递增。干混料中的碳酸盐会逐渐增加熔体分数，当混合物中的碳酸盐含量为 10% 时，熔体分数高达 12%，尤其是在低压-高温区域。然而，熔体分数在 1.5 和 2.5 GPa 之间不超过 4%。0.5% H2O 的存在进一步增加了熔体分数，尽管并不显着。在模型域内，材料运动是由运动学强加的收敛和由热和成分变化产生的内部浮力驱动的。图 2 和补充材料中的动画 S1 显示了模型的演变。该模型在俯冲开始后从 0 my 开始（图 2A）。进来的碳酸盐在板块边缘聚集，最初形成一个宽楔形，随着板块的持续辐合，碳酸盐被俯冲板块带入地幔（图2B）。在约 14 米时，地幔中的碳酸盐层变得不稳定并开始从海洋板块中浮力分离，形成底辟，在 80-100 公里深度处开始（图 1）。2C)。如果碳酸盐的厚度较大，则会获得类似的结果（我们使用高达 2 公里厚的碳酸盐进行了模型运行 [图 2 中未显示]）。底辟也发生在碳酸盐和上地幔之间的低密度对比下，低至 250 kg/m3 对比（未图示）。较厚的上板块岩石圈会导致分离延迟和更深，但模型表明，底辟总是在我们研究中设想的一般情况下形成。上升的底辟通过地幔楔迅速上升并穿透大陆地幔岩石圈（图2D）。持续的俯冲作用产生连续的底辟作用，导致（1）碳酸盐在岩石圈-软流圈边界或大陆莫霍面的积水，以及（2）地幔楔的冷却（图 2E 和 2F）。其他模型表明，碳酸盐底辟和地幔楔角的堆积很容易在各种条件下发生（图 S3）。我们的模型预测，碳酸盐底辟与对流地幔混合，并在俯冲过程中冷却地幔楔，然后再添加到地幔楔角。大陆岩石圈底部。与强岩石圈模型相比，较弱的上板块岩石圈模型可以更彻底地混合现有的地幔岩石圈（其中一些岩石圈从弧后侧参与分层过程）。对于强岩石圈，软流圈橄榄岩-碳酸盐混合物不会穿透现有的地幔岩石圈，而是附着在其底部并增加了地幔岩石圈的总厚度。地震速度计算表明，在最上层地幔的实际压力和温度范围内，橄榄岩-碳酸盐混合物会产生地震缓慢的地幔，相对于软流圈地幔，最下层岩石圈的速度低 3%。这种低密度材料不太可能会随后发生对流去除，因此这可能会将碳酸盐添加到地幔岩石圈的底部或地幔岩石圈中。板片位于 100 km 深度处上方的垂直热剖面 14和 23 my 俯冲开始后（图 2）显示在图 3 中，地幔楔与碳酸盐化肥沃橄榄岩固相线和 H2O + CO2 饱和橄榄岩固相线相比（Falloon 和 Green，1989）。该图显示，相对短寿命的俯冲系统，地幔楔中的碳酸盐底辟作用会延迟熔化，因为冷物质正在混合到地幔楔中。H2O + CO2 饱和橄榄岩固相线相对于纯水饱和橄榄岩有所降低，但从下行板片中添加冷材料会抵消熔体生产率方面的影响。鉴于图 3 中所示的流体饱和肥沃橄榄岩固相表明在可能的最低温度下开始部分熔融，并且楔形中的实际橄榄岩很可能肥沃程度较低（即，单斜辉石含量较低）并且可能不是饱和的，并且很可能没有水饱和，如果碳酸盐物质以底辟方式引入，这些系统的地幔楔中的熔融量可能微不足道。饱和条件（H2O 和 CO2）下的最大预测部分熔体分数约为 4%。对于近乎纯的石灰岩底辟，不涉及水，但我们假设少量硅质沉积物的存在可以为系统添加一些水，否则不会发生熔化。随着时间的推移，在消耗碳酸盐平台后，楔形加热并通过额外的对流恢复更热的热状态，并且碳酸盐底辟的传导加热及其与岩石圈底部的附着发生（见下文）。 正向熔融建模结果表明，少量的部分熔体在 2 时是钙质碳酸岩。5 GPa 或高度欠饱和的含长石的白榴石和 1.5 GPa 的响石。鉴于这种俯冲系统的上板块是大陆板块，在特提斯情景中，地幔楔的厚度不太可能小于约 45 公里。我们的模型表明，在大量碳酸盐俯冲的情况下，在大多数情况下熔化会延迟，这与欧洲阿尔卑斯山等地缺乏岩浆作用是一致的。这些地区（喀尔巴阡山脉、阿尔卑斯山等）不是产生正常大小的岩浆弧，而是（1）在辐合期很长一段时间内根本没有岩浆活动，或者（2）非常有限且富含碱的岩浆活动这可以通过这里描述的过程来解释。 pMELTS 算法不是特别准确并且在低熔体分数时；了解熔体的程度和组成的更好途径是通过实验。最近，陈等人。（2021 年）提供了一些关于我们在这里设想的过程和物理条件的初步实验数据，他们获得了与 pMELTS 预测非常相似的结果（见补充材料）。我们量化有多少二氧化碳返回到上板块的岩石圈固态底辟作用与损失到前弧的碳酸盐量以及沿俯冲平面输送到更深地幔的部分。大部分（约 35%）留在前弧；约 10% 深入地幔。通过底辟回到上板块岩石圈或上板块底部占主导地位；在此处介绍的模型中，它似乎占俯冲碳酸盐质量的约 40%–50%。这种机制很重要，需要纳入俯冲 CO2 的质量平衡预算（Plank 和 Manning，2019），并预测在这种情况下，浅层超俯冲地幔可能是 CO2 的重要长期汇。例如，如果考虑一个 500 公里长的碳酸盐（100% 方解石）平台以 5 厘米/年的速度俯冲，并且大约 40% 的俯冲物质通过底辟回到上板，正如这里所设想的，每年仅从一个大型平台向上板块添加约 5 吨碳。未来的测试需要量化可能具有同位素特征的地幔衍生材料的数量，这些同位素特征表明在阿尔卑斯-喀尔巴阡山脉等地区添加了碳酸盐；最近的玄武岩场可以对此进行测试（Wilson 和 Downes，2006 年）。沉积碳酸盐中的 Ca 同位素值 (DePaolo, 2004) 通常不同于地幔值 (Amsellem et al., 2020)。地幔岩石圈捕虏体在区域内也很常见（Downs 2001），可以使用为其他区域提出的技术（Chen 等人，2018 年）研究与神秘碳酸盐液体的相互作用。可以研究来自特提斯地区的捕虏体中的冷冻钙质碳酸岩，如 Persani 火山场（罗马尼亚；Chalot-Prat 和 Arnold，1999 年）中详细描述的那些，以获得有关碳酸盐起源、碳酸盐再循环程度的信息，以及岩石圈地幔渗流的时间。我们感谢 Babes-Bolyai 大学（罗马尼亚）奖学金（授予 CNFIS-FDI-2021-0061）和罗马尼亚 UEFISCDI（高等教育、研究、开发和创新融资执行单位）授予 PN-III-P4-ID 的支持-PCCF-2016-0014。C. Currie 感谢 WestGrid（维多利亚，不列颠哥伦比亚省，加拿大）和 Compute Canada（多伦多）提供的支持和计算资源。