Chromitite layers require the existence of large, long-lived, and entirely molten magma chambers

An emerging and increasingly pervasive school of thought is that large, long-lived and largely molten magma chambers are transient to non-existent in Earth’s history 1–13 . 8 These ideas attempt to supplant the classical paradigm of the ‘big magma tank’ chambers 9 in which the melt differentiates, is replenished, and occasionally feeds the overlying 10 volcanoes 14–23 . The stratiform chromitites in the Bushveld Complex – the largest 11 magmatic body in the Earth’s crust 24 – however, offers strong contest to this shifting 12 concept. Several chromitites in this complex occur as layers up to 2 metres in thickness 13 and more than 400 kilometres in lateral extent, implying that chromitite-forming events 14 were chamber-wide phenomena 24–27 . Field relations and microtextural data, specifically 15 the relationship of 3D coordination number and grain size, indicate that the chromitites 16 grew as a 3D framework of touching chromite grains directly at the chamber floor from 17 a melt saturated in chromite only 28–30 . Mass-balance estimates dictate that a 1 to 4 km 18 thick column of this melt 26,31,32 is required to form each of these chromitite layers. 19 Therefore, an enormous volume of melt (>1,00,000 km 3 must involved in

the generation of all the Bushveld chromitite layers, with half of this melt being expelled 21 from the magma chamber 24,26 . We therefore argue that the very existence of thick and 22 laterally extensive chromitite layers in the Bushveld and other layered intrusions strongly 23 buttress the classical paradigm of 'big magma tank' chambers. for only a very short period of time before accumulating and erupting as lavas on the Earth's 42 surface 1-8 . Yet another group of mafic plutonists, influenced by out-of-sequence zircon 43 geochronological data 9,10,13 , have proposed that mafic plutons do not require the existence of 44 large magma chambers 12 . These are rather produced as a stack of randomly-emplaced sills, 45 with successive crystal-rich pulses often invading pre-existing cumulates 9-13 . In stark contrast 46 to these 'anti-magma-chamber' approaches, we argue here that the existence of large magma 47 chambers is indicated by laterally extensive layers of chromite-rich cumulates, which require 48 many times their own volume of magma to supply the key component, chromium (Cr). We 49 present here field and microtextural data from massive chromitites of the Bushveld Complex 50 that indicate that 'big magma tank' chambers in the Earth's crust are a reality that cannot and 51 should not be dismissed.

84
formation has been working synchronously in all parts of the superlarge chamber to produce 85 the same chromitite layer over lateral distances of up to 400 km (e.g., UG1 in Fig. 1a).

87
Field and textural evidence for in situ growth of chromite 88 The nature of this chamber-wide process can be constrained from field and textural features of 89 massive chromitite layers such as the 2 m-thick UG1 chromitite -the thickest and the best 90 exposed layer in the entire complex (Fig. 1b). This chromitite shows remarkable field 91 relationships with its respective footwall rocks. In addition to its occurrence on the planar  simplest alternative mechanism is in situ growth of chromite directly at the chamber floor from 101 a chromite-only-saturated melt 28,30 . This is the only process that allows the chromitite layer to 102 cover all the planar and irregular margins, even the places where gravity-settling of chromite 103 grains is physically impossible (i.e., "gravity-settling shadows" in which dips are 104 overturned 28,34 ) (Extended Data Fig. 1).

106
An intriguing challenge here is to decipher how in situ growth of chromite is recorded in the 107 texture of massive chromitites themselves. We have re-visited the UG1 chromitite from the 108 classical Dwars River locality 29 (Fig. 1b) where it is composed of 25-50 vol% of cumulus 109 chromite that occurs as separate idiomorphic grains or clumps of grains that are smaller than 110 0.1 mm in size (Fig. 1c, d). The chromite grains are enclosed within much larger oikocrysts of 111 plagioclase (up to 5-10 cm in size) that are clearly visible in outcrops (Fig. 1c). The traditional 112 interpretation of such layers in the frame of gravity settling models is that chromite was the  A close look at the UG1 texture ( Fig. 1d) raises, however, a simple but fundamental quandary.

120
Chromite is almost twice as dense as a basaltic melt (4,800 kg/m 3 and 2,600 kg/m 3 , 121 respectively) and is expected to settle to the chamber floor in a random closely-packed lattice     The packing density is likewise lower than the random loose packing simulation (53% vs 60%) 226 (Supplementary Data 1). The coordination number vs grain size curve for this seam shows a 227 steady linear increase with grain size, but, like the UG1 chromitite, has systematically lower 228 coordination number values than the simulation along the entire length of the trend (Fig. 3c).  earlier-formed chromite grains to produce composite 3D clusters which subsequently merged 247 into a continuous 3D framework of touching chromite grains (Fig. 4c). New crystals emerged 248 in the system mostly by self-nucleation because the activation energy for this process is much 249 lower relative to other types of nucleation 22,46 . A small portion of crystals (3 vol.%) that occur 250 as entirely discrete grains (Fig. 2d) has likely formed by homogeneous nucleation in the 251 interstitial space. We envisage that chemical differentiation of the resident melt in the chamber 252 at that time occurred by convective removal of a buoyant compositional boundary layer 50 from 253 in situ growing chromite crystals in a 3D framework (Fig. 4c). The differentiation is aided by 254 high porosity and permeability of a 3D crystal framework that permits the easy chemical  (Fig. 4a). The thickness can be 285 reduced to 2 km 32 or 1 km 26 if Cr solubility in a parental melt is to increase by its higher 286 temperature or lower fO2 32 . These estimations assume 30% of the Cr removal from a parental 287 melt 32 . One cannot remove any more Cr from the melt than that because otherwise the melt 288 will reach a cotectic with other liquidus phases (e.g., olivine or orthopyroxene) terminating 289 chromitite formation (Fig. 4b).  has been operating as a giant magma body of more than 400 km in diameter, with a column of 324 the resident melt likely attaining a few km in thickness. Thus, starting from this stage the 325 Bushveld Complex has been developed as a large, long-lived and largely molten magma 326 chamber (a true 'big tank' reservoir) in Earth's crust (Fig. 4a). The conclusion is further 327 supported by the remarkable homogeneity of Sr isotopes over an interval of more than 2.5 km 328 of the Upper Zone 56 , which indicates a melt column thickness in the chamber being that thick 329 or even thicker 57,58 . This is in contrast with the recent assertion, mostly based on out-of-330 sequence geochronology 9,13 , that depicts this giant complex as a stack of thin crystal-rich 331 sills 9,11,13 . Field relationships indicate, however, that zircon isotopic ages in these studies were 332 almost certainly misinterpreted 59 . We thus argue that the field and textural evidence from     High resolution X-ray computed tomography and 3D image analysis and quantification 501 The UG1 sample was scanned using the Zeiss Versa XRM 520 3D x-ray microscope installed