The Role of Di V usion-controlled Oscillatory Nucleation in the Formation of Line Rock in Pegmatite–Aplite Dikes

The George Ashley Block ( GAB), located in the Pala Pegmatite Modal or layering is a phenomenon that is seen aplite dike of 8 m thickness displaying striking mineralogical layering in plutonic igneous rocks of all compositions from high- in the aplite portion of the dike, referred to as line rock. Rhythmic temperature, low-viscosity, maﬁc complexes to low-tem- layering is characterized by garnet-rich bands alternating with perature, high-viscosity, felsic complexes. The sizes of albite–quartz–muscovite-rich bands. Cumulus textures are notably such plutonic bodies can range from the large layered absent from the layered portion of the dike. Elongated quartz intrusive complexes such as the 3000 m thick Skaergaard megacrysts are oriented perpendicular to the garnet-rich layers and complex down to dikes a few meters wide. There is, poikilitically include garnet, albite, and muscovite. Calculated however, no consensus as to the origin of the layering. crystal-free magma viscosity with 3% H 2 O is 10 6·2 Pa s and the Modal layering that results from the gravitational settling calculated settling velocity for garnet is 0·51 cm/year. Conductive of crystals from a magma is a model often invoked to cooling calculations based on emplacement of a 650 (cid:176) C dike into explain the layering in maﬁc magmatic systems. This is 150 (cid:176) C fractured gabbroic country rock at 1·5 kbar, and accounting more problematic in granitic systems because of the for latent heat of crystallization, demonstrate that the line rock higher viscosities of the melts and the smaller density portion of the dike cools to 550 (cid:176) C in about 1 year. Crystal di V erence between crystals and melt. size distribution studies also suggest very rapid nucleation and This paper presents a temperature–crystallization crystallization. Di V usion-controlled gel crystallization experiments model for the George Ashley composite pegmatite–aplite yield textures virtually identical to those observed in the layered dike, which displays well-developed rhythmic layering aplite, including rhythmic banding, colloform layering, and band (line rock) in the aplite portion of the dike. Kleck (1996) discontinuities. Thus, observed textures and calculated magmatic proposed that the layering in the George Ashley Block parameters suggest that mineralogical layering in the GAB results (GAB), a ﬂoat slide block from this dike, could be ex- from an in situ di V usion-controlled process of oscillatory nucleation plained by the gravitational settling of garnet and albite and crystallization. We propose that any event that promotes from the melt, and that many of the textural features strong undercooling has the potential to initiate rapid heterogeneous which characterize the line rock, such as o V sets in the nucleation and oscillatory crystal growth, leading to the development layers, formed as a result of soft sediment deformation of a layer of excluded components in front of the crystallization of a crystal–liquid mush. The purpose of this investigation front, and the formation of line rock. was to evaluate crystal settling as a possible mechanism for generating the layering in the GAB, as well as other possible layering mechanisms, and to constrain a crystallization model for the GAB.


Petrography
rock are generally sharp and regular in detail. Host rock Petrographic examination of the line rock samples reveals alteration adjacent to pegmatite bodies is not extensive textures that suggest in situ crystallization and rapid (Foord et al., 1986). The GAB is located on the eastern growth. The overall texture is granular with anhedral slope of Hiriart Mountain (Fig. 1) and was emplaced into to subhedral plagioclase, quartz, and muscovite, and 118-120 Ma country rock, the San Marcos Gabbronorite, euhedral garnet. Elongated quartz grains, which can be which had been extensively fractured before pegmatite as long as 1·5 mm, are commonly oriented perpendicular emplacement. Pala district pegmatites that have been to the garnet layers ( Fig. 2c and d). In some cases, quartz radiometrically dated yield 40 Ar/ 39 Ar emplacement ages poikilitically encloses plagioclase or garnet ( Fig. 2c-f ). of 99-100 Ma (Foord et al., 1991). The GAB is less Muscovite grains which can be as long as 1·5 mm are evolved than other acclaimed gem-producing dikes with randomly oriented, showing no parallelism with garnet miarolitic cavities in the Pala District. However, rare layers ( Fig. 2c-f ). Plagioclase is typically <1 mm in length tourmaline crystals up to 15 cm in length are found in and garnet is <0·5 mm in diameter (Fig. 2b). the sparse pockets associated with the core of the GAB, indicating that very late in the crystallization history, boron concentrations were sufficient to form tourmaline.
The GAB is characterized by a spectacular footwall WHOLE-ROCK CHEMISTRY quartz-albite aplite that is rhythmically layered with Fifteen samples were collected from the GAB at 60 cm garnet-rich layers (Fig. 2a). Many of the pegmatites in intervals for X-ray fluorescence (XRF) major element San Diego County have layered aplites and were first whole-rock analysis, and inductively coupled plasma mass described in detail by Jahns & Tuttle (1963). Locally, spectrometry (ICP-AES) trace-element analysis. Analayered portions of pegmatite-aplite dikes are known as lytical data along with analytical uncertainties can be line rock.
found in Table 1. The granitic samples are all peraluminous, with an average A/CNK ratio of 1·23. The average composition of the GAB falls close to the 1·5 kbar H 2 O saturated minima in the haplogranite system FIELD AND MINERALOGICAL (Fig. 4) as do the other granular pegmatite samples. The

DESCRIPTION
samples with an abundance of microcline fall closer to The GAB is a tabular dike with an exposed thickness of the Or apex, samples more enriched in quartz fall closer 8 m and a width of 15 m. A schematic vertical section to the Q apex, and the line rock sample falls along the of the dike is illustrated in Fig. 3 along with the distinctive Q-Ab join with very little of the Or component. The mineralogical and textural characteristics from bottom microcline-rich core was not plotted. to top. The lower 150 cm of the dike is composed of Overall, no fractionation trends are seen from the granular pegmatite with crystals up to 5 mm in size of bottom to the top of the GAB. Principal chemical changes quartz, plagioclase, microcline, and muscovite ( Fig. 3a are mineralogically controlled and are most apparent in and b). A few scattered aplite layers occur interspersed sample F from the line rock (300 cm above the base), with the granular pegmatite for the next 100 cm. The and sample G from the core (360 cm above the base). distinctly layered line rock begins 245 cm above the base The aplitic line rock sample F is mineralogically enriched in garnet, muscovite, and albite, and has virtually no and extends for 70 cm (Fig. 3c). It is characterized by microcline. Chemically, this sample has the highest con-muscovites analyzed were from the line rock sample F below the core (0·4 wt %) and sample H from above the centrations of Fe 2 O 3 (total iron), MnO and Na 2 O, and the lowest concentration of K 2 O. Sample G from the core (0·5 wt %). Muscovite from the remaining samples had no more than 0·1% F. Finally, metasomatic effects core has the highest concentration of K 2 O, Al 2 O 3 , B and Li, and the lowest concentration of Na 2 O, resulting from on the adjacent country rock, which would have been produced from an escaping vapor phase from the crys-the abundance of perthitic microcline, the appearance of scattered, large, tourmaline crystals, and the lack of tallizing pegmatite, are absent. Thus, we have chosen a H 2 O content of 3% for calculation purposes, which would albite crystals.
permit crystallization of at least 75% of the dike before vapor saturation occurred.

MAGMATIC PARAMETERS Total water and other volatiles content Temperature and pressure
There is evidence from petrological and fluid-inclusion The estimated temperature of emplacement for the GAB studies to support the idea that pegmatites do not attain was 650°C, based on mineral equilibria of the assemblage saturation in an aqueous phase until they approach their quartz, plagioclase, potassium feldspar, muscovite, and solidi (London, 1992). There is no evidence that the GAB garnet (Huang & Wyllie, 1981). As previously mentioned, magma was water saturated when it was emplaced, and the average composition of the GAB falls close to the saturation was probably only achieved in the final stages 1·5 kbar H 2 O-saturated minima in the haplogranite of core crystallization with the development of a few system (Fig. 4) corresponding to an emplacement depth small pockets. Overall boron concentrations are low in of~5 km. According to Brisbin (1986), the rheologic state the GAB and increase from both the top and the bottom of the crust at a depth of less than~7 km will be brittle of the pegmatite toward the core zone (Table 1). Fluorine and intrusion at shallow crustal levels will be accompanied abundance in muscovite determined from microprobe by brittle host-rock behavior. If the host rocks are charanalyses (E. E. Foord, analyst, unpublished data, 1993) acterized by fractures with random orientations and the also increases from both the top and the bottom of the pegmatite toward the core zone. The most F-rich lithostatic stress field is modified by horizontal tectonic VOLUME 38 NUMBER 12 DECEMBER 1997 compressive stresses, then the only fractures that will both vertical and horizontal fractures, and at greater depths would become increasingly more irregular in serve as sites for pegmatite emplacement at these depths will be oriented horizontally. However, below a depth shape. The majority of the Pala District pegmatites are subhorizontal, tabular dikes that were emplaced along of~7 km, the host rocks would behave in a more ductile fashion and the pegmatites would be emplaced along pre-existing fractures in the country rock. Thus, the Major elements in wt % were determined by X-ray fluorescence (PW-1606 instrument), using fused Li-tetraborate disks, by the method of Taggart et al. (1987), analysts D. F. Siems and J. S. Mee. Total iron is expressed as Fe 2 O 3 . Errors for each oxide are shown in parentheses for sample A. LOI, loss on ignition at 925°C; Tot H 2 O, total water, determined by Karl Fisher Titration, using the method of Jackson et al. (1987), analyst T. R. Peacock. B and Li in ppm were determined by ICP-AES, using the method of Lichte et al. (1987), analyst D. L. Fey. Errors are shown in parentheses for sample A. All analysts and equipment from US Geological Survey, Lakewood, CO.
for (X tot H2O ) >0·25 the equation is where viscosity is in pascal-seconds, and temperature, T, is in kelvins. This model is applicable to granitic melts with total H 2 O contents up to 12·3 wt % and yields melt viscosities much lower than those estimated using the method of Shaw (1972 (Baker, 1996).

Importance of conduction vs convection
To constrain a realistic cooling model, many parameters need to be evaluated, including those related to the magma and its intrusion (temperature and depth of Density emplacement, thermal diffusivity, latent heat of crys-Density, , was calculated using the method of Bottinga tallization, temperature interval of crystallization, and & Weill (1970) and the more recently determined values intrusion size), as well as knowledge about the initial of partial molar volumes from Lange & Carmichael temperature, thermal diffusivity, and geothermal gradient (1987). The density of a melt is given by of the host rock ( Jaeger, 1968). Additionally, one must evaluate whether conductive or convective processes were = X i M i X i V i (1) acting upon the magma body during crystallization. Convection and gravitational settling of crystals have long been accepted as two important processes governing where X i is the mole fraction of component i, V i its the evolution and differentiation of magmas (Bowen, partial molar volume, and M i its gram formula weight. 1928;Wager & Brown, 1967;Sparks et al., 1993). How-The magmatic density calculated for the GAB by avever, the importance of convection and crystal settling eraging all 15 whole-rock analyses is 2·31 g/cm 3 .
in thin sills and dikes has been re-evaluated by a number of workers, including Brandeis & Jaupart (1987), Marsh (1988b, 1989), Spohn et al. (1988), Gibb & Henderson (1992), and Mangan & Marsh (1992). Although many Viscosity, , was calculated using the empirical model of these models have predominantly evaluated mafic of Baker (1996). This model allows for viscosity decomposition sheet-like magma bodies, they all conclude termination of peraluminous to metaluminous granitic that either cooling in thin sills or dikes is entirely by melts at crustal pressures and temperatures. The conduction or that if convection is occurring, it will not empirical equation for a mole fraction of total water (X tot H2O ) Ζ0·25 is be vigorous.

Viscosity
Frequently, the velocity and vigor of convection has cooled to below 150°C (microcline argon-retention tembeen evaluated from the magnitude of the dimensionless perature) by 104 Ma (microcline apparent age) before Rayleigh number (Ra), the ratio of thermally induced pegmatite emplacement. This temperature of 150°C is buoyancy to viscous drag: consistent with that obtained using a geothermal gradient of 30°C/km for the GAB dike emplaced at a depth of Ra=g TL 3 / (4) 5 km. Based on phase equilibria (Huang & Wyllie, 1981), the GAB magmatic temperature of emplacement was where g is acceleration due to gravity, is the coefficient~6 50°C. The temperature interval of magmatic crysof thermal expansion, T is the temperature gradient, L tallization thus began at 650°C and, for a melt of this is the thickness of the liquid layer, is kinematic viscosity composition, should have been essentially completed by ( / ) and is thermal diffusivity. A fluid cooled at the the time the magma cooled to~550°C. The simplifying top or heated at the base should convect spontaneously assumption that thermal diffusivity is the same for both when Ra exceeds a value of~2000 (McBirney, 1993). magma and country rock was used for the cooling models. The difficulty in determining an appropriate Ra value Thermal diffusivity can be defined as for any magma body lies in choosing meaningful T and L values. Jaupart & Tait (1995) suggested that the =K/ C P (5) appropriate scales may be the thickness of the unstable where is thermal diffusivity in cm 2 s -1 , K is the coefficient boundary layer and the temperature gradient across it.
of thermal conductivity measured in cal cm -1 s -1°C-1 , Marsh (1988bMarsh ( , 1989, in evaluating the importance of is density in g cm -3 , and C P is heat capacity in cal g -1°C-1 . convection and crystal settling in sheet-like basaltic Heat capacity is defined by magma chambers, concluded that the thermal boundary layer that controls convection is a thin layer near the C P = q/ T (6) leading edge of the descending crystallization front and in which q is the amount of heat that must be added that convection will not be vigorous in thin dikes. He or removed to raise or lower the temperature of one suggested that the vigor and style of convection within a gram by one degree centigrade. Heat of crystallization, sheet-like magma body should not be measured by the q c , is the heat liberated by the crystallization of one usual measure of Ra [equation (4)], which assumes heatgram of melt that is already at the temperature where ing from below, as it suggests rates of heat transfer so the liquid and solid coexist. For most rocks, C P is~0·3 large that solidification times are unreasonably short.
cal g -1°C-1 , whereas typical values for q c are~60-100 Instead, although there may be motion within sheets of cal g -1 (McBirney, 1993). This amounts to roughly the magma, it is so slight that the thermal history is well same amount of heat in the crystallization of a melt as approximated by heat conduction alone with inclusion in raising the temperature of the melt by 200-300°C, of latent heat. Additionally, Brandeis & Jaupart (1987) which will reduce the rate of cooling of the magma as it suggested that conduction is the dominant mechanism passes through the crystallization interval. The heat of of heat transport in dikes up to widths of 100 m. Therecrystallization q c , given up during crystallization through fore, we conclude that a conductive cooling model is appropriate for the 8 m wide GAB dike. a temperature interval T 1 -T 2 , can be allowed for by adding the proportional heat per degree to the heat capacity to obtain an effective heat capacity, h ( Jaeger, 1968): Cooling parameters for the GAB h=C P + q c (T 1 −T 2 ) (7) As previously discussed, the GAB magma was intruded into the extensively fractured 118-120 Ma San Marcos This can be substituted for C P in equation (6) to obtain Gabbronorite. As a result of pegmatite emplacement in an appropriate value which accounts for the heat of the Pala District, the age of the country rock, as decrystallization. termined from both biotite and hornblende 40 Ar/ 39 Ar Using a C P of 0·3 cal g -1°C-1 , a q c of 75 cal g -1 geochronology, was reset to the time of pegmatite em- (McBirney, 1993), and a temperature interval (T 1 -T 2 ) of placement of~100 Ma (Foord et al., 1991). However, 100°C (from 650°C to 550°C) the effective heat capacity samples collected from the spatially associated Woodson (h) is 1·05 cal g -1°C-1 . Then, using a thermal conductivity Mountain Granodiorite at distances up to 15 km away (K) of 4 × 10 -3 cal cm -1 s -1°C-1 and the GAB calculated from the pegmatites have not been affected by pegmatite magma density of 2·31 g cm -3 , results in a thermal intrusion. According to Foord et al. (1991), this granodiffusivity ( ) value of 1·66 × 10 -7 m 2 s -1 . The thermal diorite was emplaced before 113 Ma (hornblende age), diffusivity obtained by not taking latent heat of crysand it cooled below 300°C (biotite argon-retention temperature) by 107 Ma (biotite apparent age), and had tallization into account, and using the same magma density, heat capacity and thermal conductivity, is growth or residence time and the degree of undercooling.
The conduction equation for a sheet of thickness 2a from the Makaopuhi lava lake, and evaluated growth and nucleation rates for crystals as well as the degree of ( Jaeger, 1968) that assumes the same thermal diffusivity, , for both magma and country rock is as follows: undercooling. The crystal population density n (number of crystals in a given size class per unit volume) is defined as where L is the characteristic linear size, and N is equal to the cumulative number of crystals per unit volume erf(y)= 2 8 y 0 exp(−u 2 du (9) less than or equal to L. The population density n of crystals can be related to crystal size L at steady state by T is the temperature at time t (in seconds) at distance x (in meters) from the mid-plane of the sheet minus the n=n 0 exp −L G (12) country rock temperature, T=T(x,t) -T cr , and T o =T m -T cr , where T m is equal to the initial magma temperature and T cr is equal to the country rock temperature.
J=n 0 G. (13) The cooling history of the GAB was modeled in two ways. The first model assumed no latent heat of crystallization was generated during the crystallization of Calculations the dike, and used a thermal diffusivity of 5·80 × 10 -7 In CSD studies, the crystal dimension measured is the m 2 s -1 . The line rock area of the dike, which is 240-300 cm longest horizontal chord. The number of crystals per above the base, cools to below 550°C in~75-125 days unit volume (N V ) can be determined from the number after emplacement (Fig. 5a). The second model, which of crystals per unit area (N A ) using the conversion accounts for latent heat of crystallization by using a modified thermal diffusivity value of 1·66 × 10 -7 m 2 s -1 , N V =(N A ) 1·5 (14) slows the rate of cooling, such that the top of the line as suggested by Wager (1961) and Kirkpatrick (1977). rock cools to below 550°C in~300 days (Fig. 5b).
Garnets in the GAB are euhedral and nearly spherical, and are thus ideal for CSD studies. Values of n were calculated numerically from measured values of N and

CRYSTAL SIZE DISTRIBUTION OF
L from BSE images of about 2500 garnets from 13 garnet

GAB GARNET LAYERS
layers, using image analysis. A cumulative CSD plot of ln n vs L for all 13 layers generates a linear trend with a Theory slope of -1/G and an intercept of n 0 , the nucleation The theory of crystal size distribution (CSD) was initially density. A linear regression through the GAB garnet data developed by chemical engineers (Randolph & Larson, points was used to calculate the slope and intercept and 1971) for evaluating the kinetics of crystallization, i.e. yields a G value of 5·52 × 10 -3 cm and an n 0 value of growth and nucleation rates, of industrial compounds.
1·9 × 10 6 cm -4 . CSD is an empirical model of crystal nucleation and Using this n 0 value and the relationship in equation growth based on a population balance of crystals in the (13), it is then possible to calculate either J or G, if one system. Crystals move into and out of a size range through of the two variables can be determined. physical movement in the system and growth into and out of the size range of interest. Marsh (1988a) developed the standard CSD theory for geologic systems, to deter-Nucleation and crystal growth rates mine the kinetics of crystallization by examining variations in grain size. On the basis of grain-size information Brandeis & Jaupart (1987) used dimensional analysis of crystallization equations for conductive cooling together it is then possible to evaluate nucleation rate, growth rate, VOLUME 38 NUMBER 12 DECEMBER 1997

Fig. 5(a)
with crystal size data from thin dikes to obtain values for Brandeis & Jaupart (1987) of between 10 5 and 10 7 s for conductive cooling of dikes less than 10 m in width. A nucleation rates J and growth rates G. The rates for both variables can be expressed as a function of temperature characteristic cooling time of >10 7 s would result in equilibrium crystallization and grain size corresponding and undercooling, following kinetic theory. Both the nucleation and growth rate functions have bell-shaped to a thicker dike. If a growth rate of 10 -8 cm s -1 is used along with the n 0 value obtained from the GAB garnet curves with respect to degree of undercooling (Brandeis CSD study, then the nucleation rate J is equal to et al., 1984). In geologic systems, the total range of possible 1·9 × 10 -2 cm -3 s -1 [from equation (13)], and the charnucleation rates is much larger than that of growth rates. acteristic cooling time ( ) is 2·69 × 10 6 s, or about 31 days. Brandeis & Jaupart (1987) suggested that local growth A growth rate of 10 -7 cm s -1 yields a characteristic cooling rates in silicate systems in general range from 10 -10 to time ( ) of 4·78 × 10 5 s, or about 5·5 days. These results 10 -8 cm s -1 , and local nucleation rates are on the order underscore the rapid crystallization time for the line rock of 10 -7 −10 -3 cm -3 s -1 . Their data from dike margins conportion of the pegmatite-aplite dike compared with the strain maximum growth rates of close to 10 -7 cm s -1 and rest of the dike based on our conductive cooling models. nucleation rates of~1 cm -3 s -1 . The characteristic cooling time, , can be related to G and J by

LAYERING MECHANISMS
For calculation purposes we have chosen growth rates Many different mechanisms have been proposed for of 10 -8 cm s -1 and 10 -7 cm s -1 , as these values yield layering in igneous rocks including gravitational settling with or without convection (Wager & Brown, 1967; characteristic cooling times ( ) consistent with those of  Huppert & Sparks, 1980), oscillatory nucleation and is only applicable for spherical grains settling in a Newtonian, non-convecting magma. Therefore, settling vecrystal settling (Hort et al., 1993); and in situ crystallization locities were only calculated for garnet and albite as from a static boundary layer ( Jackson, 1961; McBirney both muscovite and quartz are typically non-spherical, & Noyes, 1979). elongated grains. A viscosity of 10 6·2 Pa s, a magma Modal layering that results from the gravitational setdensity of 2·31 g/cm 3 , and mineral densities of 4·2 g/ tling of crystals from magma is an often invoked model cm 3 for garnet and 2·62 g/cm 3 for albite were used for to explain layering in mafic magmatic systems. A settling calculation purposes. Maximum crystal size diameters mechanism is more problematic in granitic systems be- (Fig. 2b) of 0·5 mm for garnet and 1·0 mm for albite yield cause of the higher viscosities of the magmas and the settling velocities (V s ) of 0·51 cm/yr and 0·34 cm/yr, smaller density difference between crystals and magma.
respectively. Settling velocities (V s ) were calculated for garnet and Jaupart & Tait (1995) emphasized the importance of albite for the line rock portion of the GAB using Stokes' the thermal regime of the magma on crystal settling. In Law: the absence of thermal convection, crystals nucleate and grow only in boundary layers. In this case, the V s =2gr 2 /9 (16) temperature of the magma away from the boundary where g is acceleration due to gravity, r is the radius of layer is higher than that at the boundary layer. However, the crystal, is the density difference between the if thermal convection is operating in the magma, the whole body will be cooled and crystal nucleation and crystal and the magma, and is viscosity. Stokes' Law growth can occur anywhere in the magma body. The Evidence for in situ crystallization implication of this is that the cumulus theory of grav-The lack of preferred orientation of elongated quartz itational settling relies on the existence of thermal conand muscovite grains parallel to the line rock garnet vection. According to Marsh (1988b), crystallization in layers precludes the formation of the layering as the basaltic composition sheet-like magma chambers is inresult of some type of flow process. Instead, textural ward progressing and takes place along a mushevidence such as elongated quartz and muscovite grains suspension interface. Near this interface the viscosity oriented perpendicular to the garnet layers (Fig. 2), supincreases and thus acts as a capture front that overtakes ports in situ crystallization. Elongated quartz grains poikiand traps slowly settling crystals. Crystals nucleated and litically include plagioclase, garnet and muscovite. In grown in the suspension zone can escape and settle only some instances, plagioclase is smaller in the center of the if the capture front slows to a critical rate attainable only quartz grain and is larger toward the edges. McBirney in bodies thicker than 100 m. Marsh further suggested & Noyes (1979) interpreted plagioclase laths enclosed in that sills of viscous silicic magma would have to be poikilitic pyroxenes as evidence that the phases nucleated more than~1 km thick before capture by the downward and grew together in situ. In the case of the GAB it progressing solidification front could be averted. Gibb & appears that fewer quartz nuclei formed than did the Henderson (1992) evaluated convection and crystal setother minerals, hence as the minerals grew together the tling in mafic sills, and concluded that the magmas had not quartz enclosed them, and thus stopped further growth. undergone turbulent convection and that gravitational Geochemically, the GAB shows little variation in settling did not play a significant role in modifying the whole-rock chemistry from bottom to top (with the exsills. ception of the garnet-rich layers). Because in situ crys-Hort et al. (1993) considered the role of oscillatory tallization does not change the overall magma chemistry, nucleation in the development of layering and cumulates, the geochemical implications are that fractional cryscoupled with convection. Oscillatory nucleation is initallization will not take place. In layered complexes where tiated by slight compositional variations of the liquid fractionation occurs by the removal of crystals, mineral resulting from the crystallization of one phase over anchemical trends are smooth, i.e. olivines vary from Mg other, i.e. crystallization of one phase changes the melt to Fe rich, plagioclase from a higher to lower An content. composition to favor crystallization of a second phase. However, in the GAB, garnets do not change chemically In their model, continued growth eventually allows setin a systematic way from Fe-rich to Mn-rich compositions tling, thus leading to phase layering on the floor of the toward the center of the dike (E. E. Foord, analyst, chamber. However, Hort et al. suggested that relatively unpublished data, 1993). This suggests to us that the small bodies cool too fast to nucleate and grow crystals chemistry of garnets in the GAB is controlled by the large enough to effectively settle and produce cumulates, melt chemistry of the boundary layer. and they concluded that layering is unlikely in sheet-like magmas thinner than~100 m.

Oscillatory nucleation and crystallization
Rapid nucleation and growth of albite and quartz in the line rock aided in the development of a boundary layer Model for the GAB of excluded components (initially Mn and Fe) ahead of the crystallization front, which accumulated to the point Based on the preceding arguments against settling and of local saturation. This led to the nucleation and growth convection, we have concluded that no large-scale conof garnet, which continued as long as the concentrations vection occurred in the GAB magma. Given the short of Fe and Mn were high. When Fe and Mn concentrations time necessary for conductive cooling of the GAB coupled decreased at the crystallization front, nucleation stopped with the slow Stokes' Law settling velocities for garnet and growth slowed until the concentrations built back and albite, it is unlikely that crystals nucleating and up again to saturation levels. As garnet nucleation slowed, growing at the boundary layer could escape a downward the system became dominated by the nucleation and solidifying capture front to accumulate in layers in the line crystallization of albite, quartz and muscovite. This rhythrock. Therefore, we conclude that gravitational settling mic oscillation of garnet growth repeated itself throughout cannot be the mechanism for generating the line rock the line rock interval. layering. However, as rhythmic layering does exist in the line rock, we need a mechanism for rhythmic crystallization which does not result from gravitational settling. We propose that the line rock layering resulted from GEL EXPERIMENTS a process of diffusion-controlled oscillatory nucleation and crystallization that took place in situ at the boundary To evaluate whether the textures produced in diffusioncontrolled crystallization would be similar to the observed layer. line rock textures, we designed a series of gel crys-the granular texture observed along both margins (Fig. 3).
As crystallization of the mainly anhydrous phases pro-tallization experiments. We were particularly interested in trying to reproduce the colloform structures and layer ceeded, volatile phases were concentrated as excluded components in the remaining melt in the central portion discontinuities that are observable in the line rock (Fig.  2a). Rhythmic crystallization was originally recognized of the uncrystallized dike. Had crystallization proceeded along these lines uninterrupted, a typical coarsening by Liesegang [cited by McBirney & Noyes (1979)], who suggested that banding in agates resulted from a diffusion-inward pegmatite would have resulted, with an overall granular texture up to the point of core formation, where, controlled process. He demonstrated this process of oscillatory crystallization by placing a small crystal of silver if the volatile content was sufficient and overall pressure nitrate in a dilute solution of potassium dichromate in was low enough, miarolitic cavities could have formed. gelatin. Successive bands of silver chromate would nuc-However, in the case of the GAB, after~75% crysleate and grow when supersaturation occurred. Between tallization, when the magma was close to vapor satthe bands were areas of no crystallization as a result of uration, the normal progression of crystallization was local depletion of the surrounding area in chromate ions. interrupted and rapid nucleation and crystallization en-For our experiments we dissolved potassium disued, producing the line rock portion of the dike. We chromate in gelatin to which we added a solution of believe that the trigger for the destabilization of the silver nitrate and lead nitrate in 50/50 molar quantities. system was a sudden pressure loss within the dike caused To try to simulate bands of crystals, instead of rings, we either by dike rupture or the dilatancy of the dike because added the solution as a streak on top of the gel and let of fracture propagation of the stressed country rock. This it diffuse. In some cases, the amount of solution was sudden decrease in confining pressure would increase the varied along the length of the streak. The gels were then degree of undercooling and change the crystallization dried in a desiccator, sliced into ribbons, carbon coated kinetics, initiating rapid heterogeneous nucleation and and then investigated by BSE imaging (Fig. 2g and h). The disequilibrium crystallization, as indicated by the CSD colloform-type structures seen in Fig. 2g were produced results. The ensuing rapid nucleation and growth of albite, when concentration was varied along the streak. Breaks quartz and muscovite would result in the development of in bands and variation in band width are apparent, as a boundary layer of excluded components (initially Mn well as the presence of a few scattered crystals between and Fe) ahead of the crystallization front which would the bands.
accumulate to the point of local saturation and thus set The similarity between gel crystallization textures and up the right conditions for oscillatory crystallization to those seen in line rock reinforce our belief that diffusionbegin. In this peraluminous melt, the saturation of Mn controlled crystallization is an important process in the and Fe at the crystallization front led to the nucleation formation of the GAB. The most distinctive line rock and growth of garnet which continued until the contexture in the GAB which was duplicated in the diffusioncentration of Mn and Fe decreased. The nucleation and controlled gel experiments was the colloform structures growth of garnet would then slow dramatically until the ( Fig. 2a and g).
concentration of the excluded elements built up again to saturation levels. During line rock formation, boron concentrations were not sufficiently high to permit the crystallization of significant amounts of tourmaline. Albite

DISCUSSION
and quartz nucleation and growth continued throughout. With the melt already close to vapor saturation it appears The textural relationships in the GAB dike are a reflection of the overall nucleation and growth history of the likely that the drop in pressure and subsequent rapid nucleation and crystallization of the line rock would minerals crystallizing from the dike. In most mafic dikes crystal size increases only slightly from finer-grained trigger the exsolution of a vapor phase. The volatiles would continue to be concentrated in the residual melt margins to coarser-grained interiors, suggesting that nucleation and growth rates vary only slightly and regularly along the upper crystallization boundary layer where they would inhibit nucleation and by their fluxing effect with time (Cashman & Marsh, 1988;Marsh, 1988a). It is reasonable to expect that similar kinetics of nucleation possibly cause resorption of crystals . Line rock formation ceased when sufficient volatile phases and growth should apply to granitic dikes. Such is not the case with the GAB. The abrupt change in nucleation had accumulated and the degree of undercooling decreased. Normal equilibrium granular crystallization then rates coincident with the onset of line rock formation (Fig. 3) is strong evidence that the normal crystallization resumed. Thus, the upward concentration of volatiles may be the factor that restricts line rock to the footwall process was interrupted.
Upon emplacement, nucleation and crystallization in portion of many sheet-like pegmatite-aplite dikes.
In conclusion, it appears that the GAB cooled rapidly the GAB took place along boundary layers at both the top and bottom of the magma sheet. This is reflected by by heat conduction. The rapid crystallization of the dike,