Journal of Volcanology and Geothermal Research 153 (2006) 287 – 300

www.elsevier.com/locate/jvolgeores

Segregation processes in vesiculating crystallizing magmas

A. Costa a,b,1 , S. Blake b,⁎, S. Self b
a
INGV, Osservatorio Vesuviano, via Diocleziano 328, 80124 Napoli, Italy
b
Department of Earth Sciences, The Open University, Milton Keynes, MK7 6AA, UK
Received 21 January 2005; received in revised form 13 December 2005; accepted 16 December 2005
Available online 20 February 2006

Abstract

Vesiculation of crystallising magma can produce either a mobile vesicular magma or a rigid network of crystals containing
vesicular liquid. Where partially crystallized rigid mush underlies less-crystallized magma, such as near the base of a lava flow or
in the cumulus pile of a magma chamber, evolved interstitial melt and/or gas may escape into the main body of magma. The
consequences of this may include contamination of the overlying liquid with gas and interstitial melt, or intrusion of diapirs of
vesicular evolved liquids to form vertical vesicle cylinders and other segregation features found in many basaltic lava flows and
sills. Analog experiments were used to investigate some of the phenomena that can arise during vesiculation within a crystal mush,
which was simulated by pumping air through a porous plate that formed the floor of a container filled with a viscous liquid floored
with a layer of glass beads. Experiments used either a single liquid or two stably stratified liquids with a liquid interface either
coincident with the top of the porous layer of beads or slightly above the porous layer. For a range of liquid viscosities and air flow
rates (vesiculation rates), individual bubbles emerged from the top of the porous layer of beads and carried a thin trail of interstitial
liquid into the overlying liquid. The number of bubble trains leaving the surface of the porous bed increased with decreasing liquid
viscosity and flow rate, and with increasing bead size (and, hence, with increasing permeability). Analog vesicle cylinders,
composed of diapirs of bubbly interstitial liquid, were produced only when a layer of buoyant bubbly liquid lay above the surface
of the porous layer. The relative size of the bubbles and constrictions within the porous layer are argued to control whether
individual bubbles (leading to bubble trains) or vesicular liquid (leading to vesicle cylinders) leaves the porous layer and hence
whether vesicle cylinders can form.
© 2006 Elsevier B.V. All rights reserved.

Keywords: magma; lava; vesicles; segregations; porous medium

1. Introduction crystallization in volatile-undersaturated magma cham-

Evolved magmas and igneous rocks are often formed
during the crystallization and vesiculation of magma
but, although the dynamical processes accompanying
⁎ Corresponding author.
E-mail addresses: A.Costa@bristol.ac.uk (A. Costa),
s.blake@open.ac.uk (S. Blake), Stephen.Self@open.ac.uk (S. Self).
1
Present address: Department of Earth Sciences, University of
Bristol, Wills Memorial Building, Queen's Road, Bristol BS8 1RJ, UK.
0377-0273/$ - see front matter © 2006 Elsevier B.V. All rights reserved.
doi:10.1016/j.jvolgeores.2005.12.006
bers have been studied extensively (e.g., Sparks et al.,
1984; McBirney et al., 1985; Huppert et al., 1986; Tait
and Jaupart, 1992; Jaupart and Tait, 1995), the role of
vesiculation in igneous processes (as distinct from
volcanic eruption processes) has received comparatively
little attention. In lava flows, lava lakes, and shallow
intrusions, the magma is typically vesicular and this can
potentially introduce additional dynamical factors to the
behaviour of the magma or lava. For example, three
situations in which vesiculation and crystallization may
288 A. Costa et al. / Journal of Volcanology and Geothermal Research 153 (2006) 287–300

play important roles are in driving evolved interstitial
melt out of plutonic mushes to form highly evolved
magmas (Sisson and Bacon, 1999; Bachmann and
Bergantz, 2003); in the formation of vesicular segrega-
tions of evolved liquid within basaltic lava flows, lava
lakes and sills (Anderson et al., 1984; Rogan et al.,
1996; Goff, 1996); and in generating magmatic fluids
capable of escaping into ore-forming environments
(Candela, 1991). In these instances, crystallization
drives the compositional evolution of the melt whereas
vesiculation drives physical segregation of that evolved
melt and/or gas. The interplay between vesiculation and
crystallization, and the factors that influence that
interplay, are therefore relevant in several igneous
environments.
The general problem can be considered in terms of
the progressive crystallization of a hydrous magma
(melt plus crystals plus gas). The crystal assemblage will
be anhydrous or nearly so, so the water fraction in the
residual melt (±gas) increases. At high pressure, and
low water content, the water remains dissolved and a
point is reached when the magma becomes a porous
rigid crystal framework containing interstitial residual
melt. The transition from a mobile magma to a rigid
crystal framework containing melt occurs at solid
fractions between 0.2 and 0.5 depending on the shape
and size of the crystals (Philpotts and Carroll, 1996;
Philpotts et al., 1998; Hoover et al., 2001; Saar et al.,
2001). At low pressures or high water contents, the
residual melt becomes water-saturated and gas bubbles
exsolve. The behaviour of the crystallizing magma now
depends on the relative rates of vesiculation and
crystallization, and also on the volumetric fraction of
crystals. If vesiculation starts before crystallization has
formed a rigid network, one of two behaviours is
possible (Fig. 1). First, if the increase in gas volume
occurs faster than the increase in crystal volume, crystals
will be pushed apart by the rapidly expanding gas phase
(Fig. 1(b)) such that the magma properties become less
and less influenced by the presence of the crystals.
Secondly, in contrast, if vesiculation is slow relative to
crystallization, the magma can become a rigid crystal
network containing vesicular melt (Fig. 1(c)). The latter
situation can also arise if vesiculation does not start until
after a rigid crystal framework has formed.
Whether a vesiculating magma evolves to a rigid or a
mobile state obviously has implications for the subsequent
behaviour of the magma and its constituent phases. In this
paper we investigate the general properties of crystallizing
vesiculating magma and use the results of analog

Fig. 1. Cartoon showing how a crystallizing hydrous magma may evolve from an initial stage shown in (a) into (b) an inflated mobile vesicular
magma, or (c) a rigid porous crystalline matrix containing vesicular melt. In (b) crystal volume has increased at the expense of liquid volume, but the
increase in gas volume has been much larger. In (c), gas volume has increased less than the crystal volume.
 A. Costa et al. / Journal of Volcanology and Geothermal Research 153 (2006) 287–300 289

experiments to discuss some of the conditions under which where ρ0 is the density of anhydrous liquid with the
vesiculation plays an important role in magma behaviour composition of the initial liquid and β takes account of
and how differentiated interstitial liquid may segregate changes in major element composition and water
from an immobile crystal mush. In Section 2 we model the content along the liquid line of descent. Using the
relative proportions of crystals and vesicles that can model of Ochs and Lange (1999) and the results of
develop in magmas of different water contents, solubilities, calculations plotted by Spera (1999), we use β = 3.1 per
and pressures. Section 3 reports laboratory experiments mass fraction dissolved water.
that give some insights on how the properties of a porous The density of the crystals is ρx, and is assumed to be
medium and the liquid it contains can affect the way in constant (2800 kg m− 3). The density of the gas is given
which bubbles are transported through the porous medium, by the gas law
and how bubbles and liquid can be transferred from a mush
mP
into an overlying liquid. Section 4 considers the geological qg ¼ ð5Þ
implications of the modelling results, with particular RT
reference to basaltic lavas and their segregation features,
such as vesicle cylinders (Goff, 1996). with the molecular mass of water m = 0.018 kg and the
gas constant R = 8.314 J K− 1 mol− 1.
2. Modelling crystallization and vesiculation The volumes of liquid, gas and crystals in a unit mass
of magma are then
The proportions of liquid, gas and crystals in a
1−Wx −Wg
cooling magma can be modelled as follows. We Vl ¼ ð6Þ
consider a magma that contains a total mass fraction ql
of water Wwi and examine how the volume fractions of Wg
Vg ¼ ð7Þ
gas, liquid and crystals change with isobaric cooling qg
from the liquidus temperature. For modelling purposes, Wx
the mass fraction of crystals (Wx) is taken to increase Vx ¼ ð8Þ
qx
linearly with cooling:
From which the total porosity of the magma
TL −T
Wx ¼ ð1Þ Vl þ Vg
TL −TS U¼ ð9Þ
Vl þ Vg þ Vx

where T L and T S are the liquidus and solidus and the vesicularity of the melt
temperatures, respectively.
The crystals are assumed to be anhydrous, so the Vg
V ¼ ð10Þ
water content of the remaining melt is Vl þ Vg

Wwi
WW ¼ ð2Þ can be calculated.
1−Wx We model the evolving porosity and vesicularity of a
mafic magma with TL = 1150 °C and TS = 850 °C in
unless the saturation limit Ws = aP is less than Ww (a is a which the water solubility constant changes as a linear
composition-dependent solubility constant and P is function of temperature from a = 2.5 × 10− 6 Pa− 1 / 2 to
pressure), in which case the water content of the liquid is 4.1 × 10− 6 Pa− 1 / 2, these being typical values for basalt
Ws and the excess water is present as a gas comprising a and rhyolite (Sparks, 1978), so as to model the change in
mass fraction solubility with evolving melt composition.
Two situations in which crystallization leads to
Wwi −Ws ð1−Wx Þ
Wg ¼ ð3Þ water saturation and vesiculation are considered. The
1−Ws first (Fig. 2(a)) is a plutonic magma with 2 wt.% water
emplaced at 150 MPa. Here, vesiculation does not start
The liquid density (ρl) is a function of the dissolved until after ca. 55% crystallization and, although the
water content volume of gas increases with further crystallization, the
total porosity continues to decrease while gas bubbles
ql ¼ q0 ðl−bWW Þ ð4Þ are exsolving. So, although the total volume of the
290 A. Costa et al. / Journal of Volcanology and Geothermal Research 153 (2006) 287–300
magma increases, the mush becomes more solid with
time, as in the transition from Fig. 1(a) to (c).
In the second situation, basalt has degassed to
atmospheric pressure (Wwi = 0.0008) and then crystal-
lises at the base of a lava flow. Considering a 1 m thick
flow, and hence a pressure of 1.25 × 105 Pa, vesiculation
starts after ca. 18% crystallization and, because the
pressure is low and the molar volume of gas is large,
total porosity increases even though the mass of crystals
is increasing (Fig. 2(b)). Here the magma becomes less
solid with time, as in the transition from Fig. 1(a) to (b),
and could therefore be entrained into lava moving in the
interior of the flow. At a pressure of 5 × 105 Pa,
equivalent to the base of a 4 m thick lava, vesiculation
does not start until ca. 65% crystallization (Fig. 2(c)),
which is likely to be after a rigid crystalline framework
Fig. 2. Calculated vesicularity of interstitial melt within a magma as a
function of total porosity (i.e., 1 − volume fraction of crystals). Models
assume an initial basaltic liquid, with properties described in the text.
(a) Example of plutonic conditions: initial water content 2 wt.%,
pressure 150 MPa. (b) and (c) Examples of conditions in a lava flow:
initial water content 0.08 wt.%, pressure of (b) 0.125, and (c) 0.5 MPa.
has developed. In this case, the total volume of the
magma increases, but because the solid fraction forms a
rigid framework, the volume expansion is accommo-
dated either by some of the vesicular liquid leaving the
porous framework, or by bubbles preferentially escap-
ing from the porous framework.
Considering the case where a crystal-poor magma
overlies a region of highly crystalline mush, a situation
that can be found in cumulus piles at the bottom of
magma chambers and partly crystallized zones at the
bottom of lava flows, sills, and lava lakes, one possible
mode of behaviour is that the gas bubbles escape from
the mush, leaving degassed liquid in the pore spaces
(Fig. 3(a)). Alternatively, bubble-bearing interstitial
melt may convect out of the mush, being replaced by
denser melt from the overlying layer (Fig. 3(b)). A third
possibility is that the overall expansion caused by
vesiculation is accommodated by bubbly liquid being
exuded from the crystal framework, as in Fig. 3(c). If the
exuded layer becomes sufficiently thick, then a
Rayleigh–Taylor instability may occur and produce
plumes of vesicular differentiated melt as in Fig. 3(d).
There are therefore a number of possible outcomes, with
different geological consequences.
3. Laboratory experiments
A series of laboratory experiments were carried out
to study systems in which vesiculation occurs in a liquid
layer within a porous medium that underlies a second,
denser, liquid layer. This resembles some previously
studied situations, but with important differences.
Convective overturn between non-vesicular liquid in a
porous layer and an overlying liquid layer has been
studied by Kerr and Tait (1985) and Tait and Jaupart
(1992) in the context of solidification at the base of
magma chambers. Convective overturn between vesic-
ular liquid and non-vesicular liquid in a chamber but
without any porous layers has been studied by Turner et
al. (1983), Thomas et al. (1993), and Phillips and Woods
(2002). The present problem is therefore a hybrid of
these two cases, with the added constraint that the liquid
in the lower part of the system is more evolved and
therefore more viscous than the overlying liquid.
The effect of the porous medium was studied by doing
experiments in which air was pumped through a sintered
plate into a container holding a viscous liquid and, in
some cases, a porous layer of glass beads (Fig. 4). The
experiments used a cylindrical glass Buchner funnel with
a base made from a sintered disc of porosity grade 3
(nominal pore index 16 to 40 μm). The cylindrical part of
the funnel had a 12 cm diameter and 8.8 cm height and
 A. Costa et al. / Journal of Volcanology and Geothermal Research 153 (2006) 287–300 291

Fig. 3. Cartoons showing three possible consequences of vesiculation within a partially crystallized basal layer of a magma chamber or lava. Patterned
polygons represent crystals that form a three dimensional rigid network. Dark grey represents evolved interstitial liquid produced by in situ
crystallization and is overlain by uncrystallized liquid (light grey). (a) Gas bubbles and entrained liquid escape into the upper layer; (b) convective
overturn; (c) expansion of vesicular magma expels a layer of bubbly differentiated interstitial liquid which may become unstable, leading to (d)
vesicular plumes of the differentiated interstitial liquid.

was filled to a height of 6 to 8 cm with liquid. The glass
beads had nominal diameters of either 1.5, 3 or 6 mm. A
peristaltic pump was used to pump air through the porous
plate in the bottom of the funnel at a constant volumetric
rate (Q), expressed as a flow rate per unit area, q.
3.1. Physical properties of liquids used in the
experiments
We used liquids with different physical properties:
aqueous solutions of golden syrup (denoted GS) at
different water (W) fractions, and glycerine. The
compositions of the aqueous solutions are denoted
GScW100−c where c is the percentage by mass of GS.
The physical properties of golden syrup depend
strongly on temperature and water content. In the
temperature range from 18 to 30 °C we found that the
viscosity of GS is described well by the exponential
equation:
with μ in Pa s and T in Celsius (see Fig. 5(a)).
Moreover, GS viscosity changes dramatically with
water content, especially for low water fractions.
Using the falling sphere technique and a tube visco-
meter, we found that the dependence of viscosity on
the mass fraction of water (w) at the reference tem-
perature of 21 °C is approximately described by:
0:424
lðwÞc − 0:797 ð12Þ
w þ 0:007
in the range 0 ≤ w ≤ 0.5 (see Fig. 5(b)). The few
available measurements did not permit a more
accurate fitting but this relation shows clearly the
strong influence of w on viscosity. In contrast, GS
surface tension and GS density change only linearly
with water fraction. The surface tension (σ in N m− 1)
of GS solutions was estimated according to

lðT Þcexp½6:9947−0:13714T  ð11Þ rðwÞc0:08 ð1−wÞ ð13Þ

292 A. Costa et al. / Journal of Volcanology and Geothermal Research 153 (2006) 287–300

which depends on the rise velocity of an isolated bubble

Ub as given by Stokes law:

Dqgrb2
Ub ¼ ð16Þ
3l
where rb is the bubble radius and Δρg the buoyancy of
the bubble, and the bubble Bond number

Dqgrb2
Bob ¼ ð17Þ
r
The size of individual bubbles, the amount of
deformation they experienced, and the number of bubble
trains varied with the viscosity of the liquid, for a fixed air
flux of q = 0.013 cm s− 1. In the case of 100% GS, we
estimated a bubble diameter of the order of 1 cm and hence
a Bond number Bob of the order of 10 and Reb = O(10− 3).
Strong bubble deformation and bubble interaction were
evident, in accord with the analysis of Manga and Stone
(1994). For glycerine, the biggest bubble diameters were
about 2 mm (Reb∼0.01 and Bob∼1) and the bubbles rose in

Fig. 4. Sketches of the four types of experiment in which air was

pumped through the sintered porous plate in the bottom of a Buchner
funnel containing liquid layers.

based on Llewellin et al. (2002). The dependence of

density (kg m− 3) on water fraction at 21 °C was
determined by weighing known volumes and is given
by:

qðwÞc1412−352 w ð14Þ

The properties of glycerine at 21 °C were μ = 1.5 Pa s,

ρ = 1260 kg m− 3 and σ = 0.063 N m− 1.

3.2. Vesiculation in a single liquid layer

In the first set of experiments (Type 1 in Fig. 4),

about 0.5 l of liquid was used with no glass beads. Air
bubbles escaped from points across the surface of the
sintered disc into the liquid, forming a number of
discrete bubble trains. The characteristic dimensionless
numbers (Manga and Stone, 1994) are the bubble
Reynolds number
Fig. 5. (a) Temperature °C dependence of the viscosity Pa s of 100%
qUb rb
Reb ¼ ð15Þ Golden Syrup. (b) Viscosity of aqueous solutions of Golden Syrup at
l 21 °C, as a function of the water fraction.
 A. Costa et al. / Journal of Volcanology and Geothermal Research 153 (2006) 287–300
trains that interacted. In all the other cases (GS75W25 and
GS50W50) bubble diameters were less than about 1 mm,
Bob was always b1 and, in accord with Manga and Stone
(1994), bubble deformation was not clearly evident.
These experiments were then repeated, but with a
porous layer, about 3.5 cm deep, of equal-sized glass
beads at the bottom of the funnel (Type 2 experiments in
Fig. 4). Different permeability conditions were obtained
by using different bead diameters. The porosity, ε, was
estimated from the depth of the layer and the mass of
glass beads in the layer; the porosity was about 0.3 to
0.4. Two types of behaviour were observed.
First, in all experiments except those with relatively
low viscosity (GS75W25 and GS50W50) and large beads,
bubbles coalesced within the porous layer and became
so large as to form meandering channels of air in the
space between the glass beads. These channels were
visible around the walls of the funnel. As a result of
coalescence and channelling, the bubbles that rose into
the liquid above the porous layer were larger than in the
equivalent experiment without the layer of beads. The
number of bubble trains was less than in the equivalent
experiment with no porous layer. In the second type of
behaviour, encountered in GS75W25 and GS50W50
experiments with large beads, some coalescence
occurred but channelling was not seen. The escaping
bubbles were small, although cyclic formation of a few
bigger bubbles by coalescing was observed.
A criterion that seems to distinguish between the two
different behaviours is:
2
dbead =dbubble
2
z20 ð18Þ
where dbead is the bead diameter and dbubble is the
diameter of the bubble that formed in the equivalent
experiment without a porous layer. When this ratio is
smaller than the critical value, bubble rise was signifi-
cantly influenced by porous effects, in that the number
of bubble trains emerging from the upper surface of the
porous layer was clearly reduced, and bubble dimension
increased, with respect to the case where the porous
layer was absent.
Increasing the flux of air from 0.013 to 0.083 cm s− 1
had no discernable effect on the number of bubble
trains, only on the vigour of bubble production, except
in experiments with low viscosity and 6 mm beads. In
these latter experiments, increasing the flow rate led to a
decrease in the number of bubble trains. This is attri-
buted to the increased likelihood of bubbles intercepting
each other within the tortuous space between beads as
2
the flow rate of air increased. Considering that dbead is
proportional to the permeability of the porous layer and
293
2
that dbubble is a function of other physical parameters
such as viscosity, the criterion given by (18) is likely to
be related to a more general expression involving the
properties of the porous medium, the gas flux, and the
properties of the liquid.
The experiments (Table 1) covered a range of vis-
cosity (0.04 to 60 Pa s), air flow rate (1.5 to 9.4 cm3 s− 1;
q = 0.013 to 0.083 cm s− 1), bead diameter (1.5 to 6 mm),
surface tension (0.04 to 0.08 N m− 1), liquid density
(1.21 to 1.41 g cm− 3) and porosity (0.3 to 0.4).
Dimensionless parameters that involve these variables
include the capillary number
lu
Ca ¼ ð19Þ
r
where u is a representative speed in the pores of the
porous layer, u∼q / ε, and a normalised hydraulic
conductivity
kgq
M¼ ð20Þ
lq
where k is the permeability, which can be calculated
from the Blake–Dullien–Carman relation (Dullien,
1979)
e3 r2
k¼ ð21Þ
150ð1−eÞ2
in which r is the bead radius and ε is the porosity of the
porous medium. The experiments ranged from
M ≈ 10− 3 and Ca ≈ 0.3 to M ≈ 20 and Ca ≈ 3 × 10− 4
(Fig. 6(a)) in an inversely correlated band that arises
because M and Ca are both influenced by the three
orders of magnitude variation in the viscosity of the
liquids used. The highest numbers of bubble trains
(∼100) were produced in experiments that had M N 10
and Ca b 10− 3.
Given that M and Ca are correlated, the ratio
Ca l2 q2
R¼ ¼ ð22Þ
M kgqer
is a useful compounding of the important variables and
it covers more than seven orders of magnitude. Plotting
the number of bubble trains against R (Fig. 6(b)) shows
that 5 to 10 trains were typical for R ≥ 0.01, but the
number increases as R decreases. In practise, the number
of trains cannot increase indefinitely and must be limited
by the number of pore spaces exposed at the top of the
porous layer. For a cross-sectional area A, this limit will
be of order A / d2 bead, which is 102 to 103 in our
apparatus.
 294
Table 1

A. Costa et al. / Journal of Volcanology and Geothermal Research 153 (2006) 287–300
Summary of experimental conditions and results
Experiment Substance Bead diam./mm Porosity k/m2 μ/Pa s ρ/kg m− 3 σ/N m− 1 Q/m3 s− 1 M Ca R = Ca / M Bubble diam. Approx. number
of bubble trains
A Glycerol 0 1.5 1260 0.063 1.5E − 06 b1–2 mm 35
B Glycerol 1.5 0.34 3.38E − 10 1.5 1260 0.063 1.5E − 06 2.10E − 02 9.29E − 03 4.42E − 01 about b5mm 7
C Glycerol 3 0.29 7.26E − 10 1.5 1260 0.063 1.5E − 06 4.51E − 02 1.09E − 02 2.42E − 01 2–5 mm 7
D Glycerol 6 0.33 4.80E − 09 1.5 1260 0.063 1.5E − 06 2.98E − 01 9.57E − 03 3.21E − 02 1–5 mm 7
E GS 0 59 1412 0.08 1.5E − 06 about 1 cm 25
G GS 3 0.29 7.26E − 10 59 1412 0.08 1.5E − 06 1.28E − 03 3.37E − 01 2.63E + 02 1–1.5 cm 10
H GS 6 0.33 4.80E − 09 59 1412 0.08 1.5E − 06 8.50E − 03 2.97E − 01 3.49E + 01 about b2 cm 7
I GS75W25 0 0.85 1324 0.06 1.5E − 06 about b1 mm 40
L GS75W25 1.5 0.34 3.38E − 10 0.85 1324 0.06 1.5E − 06 3.90E − 02 5.53E − 03 1.42E − 01 b2–3 mm 10
N GS75W25 6 0.33 4.80E − 09 0.85 1324 0.06 1.5E − 06 5.53E − 01 5.70E − 03 1.03E − 02 1 to 5 mm 15
O GS50W50 0 0.039 1236 0.04 1.5E − 06 b1 mm 150
P GS50W50 1.5 0.34 3.38E − 10 0.039 1236 0.04 1.5E − 06 7.93E − 01 3.81E − 04 4.80E − 04 about b3 mm 30
R GS50W50 6 0.33 4.80E − 09 0.039 1236 0.04 1.5E − 06 1.13E + 01 3.92E − 04 3.48E − 05 b1 mm;5 mm 80
300304A GS60SW40 6 0.41 1.19E − 08 0.0533 1288 0.045 1.5E − 06 2.12E + 01 3.83E − 04 1.81E − 05 b0.5 – 2 mm 100
300304C GS60SW40 6 0.41 1.19E − 08 0.0533 1288 0.045 1.7E − 05 1.87E + 00 4.34E − 03 2.32E − 03 20
300304D GS60SW40 6 0.41 1.19E − 08 0.0533 1288 0.045 9.4E − 06 3.39E + 00 2.40E − 03 7.09E − 04 2–3 mm 30
300304E ⁎ GS60SW40 6 0.40 1.07E − 08 0.0533 1288 0.045 1.5E − 06 1.91E + 01 3.93E − 04 2.06E − 05 80
300304F Mix 6 0.40 1.07E − 08 0.0212 1213 0.045 1.5E − 06 4.51E + 01 1.56E − 04 3.46E − 06 40
070404 ⁎⁎ GS 3 0.39 2.39E − 09 47 1405 0.08 1.5E − 06 5.28E − 03 2.00E − 01 3.78E + 01 7
011004A GS60W40 3 0.38 2.14E − 09 0.069 1281 0.048 9.4E − 06 4.69E − 01 3.15E − 03 6.71E − 03 2–3 mm 15
011004B GS60W40 3 0.38 2.14E − 09 0.069 1281 0.048 1.7E − 05 2.59E − 01 5.69E − 03 2.19E − 02 2–3 mm 15
011004C GS60W40 3 0.38 2.14E − 09 0.069 1281 0.048 3.5E − 06 1.26E + 00 1.17E − 03 9.30E − 04 2–3 mm 15
011004D GS60W40 3 0.38 2.14E − 09 0.069 1281 0.048 1.5E − 06 2.94E + 00 5.02E − 04 1.71E − 04 2–3 mm 10
* Upper layer of GS50W50; μ = 0.0133 Pa s, ρ = 1187 kg m− 3.
** Upper layer of GS60SW40; μ = 0.0533 Pa s; ρ = 1288 kg m− 3.
 A. Costa et al. / Journal of Volcanology and Geothermal Research 153 (2006) 287–300 295

Fig. 6. (a) Plot of Ca versus M for the experiments in Table 1. (b) Plot of R versus number of bubble trains estimated in experiments tabulated in Table 1.
Error bars of ±30% indicate the uncertainty in estimating the number of active trains at any one time. Diamonds refer to experiments with a single
liquid. Square symbols are for experiment 070404 (Δρ = 124 kg m− 3, μ = 0.053 Pa s) and 300304E (Δρ = 101 kg m− 3, μ = 0.013 Pa s) in which the liquid
above the porous medium had a lower density than the liquid in the porous medium (i.e., the configuration of Fig. 5c).

3.3. Vesiculation in a two-layer liquid
A further set of experiments investigated bubble
motion through a two layer liquid system (Types 3 and 4
in Fig. 4). Specifically, we studied bubbles passing from
a high viscosity liquid to a less viscous liquid with and
without a layer of beads in the lower liquid. We used
100% GS, 75% GS, 50% GS, glycerine and aqueous
solutions. Salt was used to increase the density of the
upper liquid and so reduce the density difference
between the two layers. The bubble Reynolds number
of the liquids ranged from O(0.001) to O(10), and the
Bond number from O(0.1) to O(10),
3.3.1. Experiments with no porous layer
Two different regimes were observed in experiments
using the set up shown as Type 3 in Fig. 4: one where
bubbles passing through the interface dragged a long
narrow tail of the more dense and more viscous liquid
across the interface (cf. Fig. 3(a) and another where
bubbly plumes of lower, more viscous liquid rose into
the upper layer (Fig. 7).
The first of these regimes was observed in experi-
ments with Reb ≥ O(10− 2) and Bob ≤ O(1) and is in
keeping with the study of Manga and Stone (1995) on
the passage of drops through a density interface. The tail
of lower-layer liquid transported by the bubbles was less
well-developed when the lower liquid viscosity and the
viscosity contrast of two liquids were small. When a
bubble passed across the interface, each bubble tended
to follow the bubble immediately above it through the
entrained liquid, forming an aligned column of lower
liquid in the upper layer sustained by bubble train rise.
Once bubbles had reached the surface of the liquid, the
dense liquid that they had entrained sank back into the
liquid. Typical bubble Reynolds numbers in the lower
liquid in these cases were roughly close to unity. As the
viscosity of the lower layer was increased (Reb ≪ 1),
296 A. Costa et al. / Journal of Volcanology and Geothermal Research 153 (2006) 287–300
Fig. 7. Plume rise induced by bubbles: the lower pale liquid is bubbly 100% GS and the upper liquid is a 50% GS + 50% aqueous salt solution. At the
bottom of the pictures the relative time is reported.
interactions between bubbles became more and more
important. Similar to the experiments of Manga and
Stone (1995) with Teflon spheres passing from a higher
to lower viscosity liquid, bubbles passing through the
interface tended to cluster and entrain more lower layer
liquid than the equivalent number of single bubbles.
In the second regime, bubbly plumes of lower liquid
were carried into the upper liquid (see Fig. 7) and
occurred for large lower layer viscosity values and a
small difference in the density of the upper and lower
liquids. Experiments in this regime had Reb ≤ O(10− 3)
and Bob ≥ O(10). The most suitable conditions for a
bubbly plume to form was when the lower liquid was
100% GS containing many small bubbles such that its
density was slightly less than that of the 50% or 75% GS
diluted with salt water that comprised the upper liquid
layer. Thus, the mechanism for producing diapirs and
plumes of lower layer liquid was the Rayleigh–Taylor
instability caused by the unstable density stratification
possibly enhanced by the passage of large bubbles
through the fluids.
3.3.2. Experiments with a porous layer and stable
stratification
In these experiments the lower layer liquid was
contained within a porous layer of glass beads (Type 4 in
Fig. 4) and was denser than the overlying liquid. We
found that bubble trains emerged from the porous layer
and carried a tail of lower liquid behind them, much as
described in Section 3.3.1, and illustrated in Fig. 2(a).
Two pairs of experiments (G and 070404; 300304A and
300304E) compared the situation where the fluid above
the porous layer was the same as that within the porous
layer with the situation where the upper fluid was less
dense and less viscous than the lower layer. The results
are reported in Table 1 and Fig. 7 and showed that the
number of bubble trains seen in an experiment was
independent of the density difference between the liquid
layers and, furthermore, was no different from experi-
ments in which only the lower layer liquid was used.
This is not surprising as Section 3.2 showed that the
properties of the porous layer and the liquid within it
control the bubble behaviour in that layer.
3.3.3. Experiments with a porous layer and an unstable
stratification
Experiments in which bubble-free liquid was placed in
the porous layer below a denser liquid overturned rapidly,
with plumes of buoyant liquid flowing from many pores
in the porous layer's surface, as reported in similar expe-
riments with bubble-free liquids by Kerr and Tait (1985).
The contrasting case in which the lower layer was vesi-
cular and slightly less dense than the upper layer by virtue
of its vesicularity was investigated in an experiment
(070904) that used a layer of bubble-rich Golden Syrup as
the lower liquid, contained in a porous layer of 6 mm
diameter beads, and a denser GS95W5 solution for the
upper liquid. In one part of the apparatus there was a
shallow depression in the top of the porous layer, such that
a pool of bubbly GS lay directly below denser GS95W5.
This arrangement is essentially the same as in the expe-
riment shown in Fig. 7 and simulates the situation where
vesiculation in the lower layer has forced bubbly liquid to
emerge from the porous layer and form a thin layer above
it, as in Fig. 3(c). In this area, plumes of bubbly GS rose
into the upper liquid, as sketched in Fig. 3(d). Plumes were
not observed where the top of the lower fluid and the top
of the porous medium coincided. Rather than developing
 A. Costa et al. / Journal of Volcanology and Geothermal Research 153 (2006) 287–300
the type of plumes shown in Fig. 3(b), this system did not
overturn. The role of the porous layer was to suppress the
liquids overturning, possibly because the beads stopped
the bubbles in any incipiently rising region of bubbly GS
from being carried through the porous medium, restoring
stability. When the gas flux was turned on, bubbles passed
through the lower liquid and emerged to form bubble
trains with trails of lower layer liquid (similar to Fig. 3(a)).
At flow rates (q) of 1.3 × 10− 4 and 1.5 × 10− 3 m s− 1 about
10 bubble trains were active over the surface of the porous
layer, which is consistent with results in Fig. 6 for very
viscous (high R) lower layer liquids.
3.4. Summary of experimental observations
The presence of a porous medium altered the pattern
of bubble transfer from that produced in the absence of a
porous medium. The effect was to channel bubbles such
that bubbles emerged from the top of the vesiculating
layer at intermittently active sites. The number of sites
was independent of the dimensionless parameter R for
R N 10 − 2 but increased as R decreased and was
independent of the properties of the overlying liquid.
The pattern of bubble release is thus a function of the
properties of the porous medium (permeability and
porosity), vesiculation rate, and fluid properties (viscos-
ity, density and surface tension). In experiments with two
fluid layers, bubbles leaving the porous layer carried a
trail of lower layer fluid into the upper fluid. These
experiments produced the type of behaviour sketched in
Fig. 3(a) but did not lead to wholesale overturning (Fig. 3
(b)), possibly because the timescale for overturning was
longer than the timescale for bubbles to escape. In the
case where a layer of buoyant vesicular viscous fluid lay
above the porous medium, plumes of this layer rose into
the overlying liquid (Fig. 3(d)).
4. Discussion: implications of the experimental
results for understanding igneous segregation
processes
Many basaltic lava flows, lava lakes, and minor
intrusions contain vesicular bodies that are composi-
tionally more evolved than the host flow (e.g., Wright
and Okamura, 1977; Anderson et al., 1984; Goff, 1996).
Geochemical studies show that the evolved vesicular
rock can be explained by in situ fractional crystallization
of the host flow (Rogan et al., 1996; Stephenson et al.,
2000) followed by some sort of physical segregation of
evolved vesicular melt. This section briefly describes
segregations in basaltic rocks and then compares them
to the processes seen in the experiments.
297
4.1. Geological constraints from basaltic rocks
Three forms of segregated body have been docu-
mented in the literature, and two or more can occur in
the same lava:
(1) Sub-horizontal sheets that may form by draining
of interstitial melt into tears and sags in the upper
solidification zone of lavas and intrusions (Marsh,
2002), by compaction (Philpotts et al., 1996), or
by horizontal injection (Thordarson, 2000). These
are referred to as vesicle/vesicular sheets or
segregation veins (Greenough et al., 1999).
(2) Rounded patches and vesicles that are partly filled
or lined with differentiated liquid, sometimes
pooled at the bottom of the patch and found
throughout all but the upper and lower margins of
some lava flows and minor intrusions.
(3) Sub-vertical cylinders of vesicular differentiated
rock that may have an apparently regular spacing
and may therefore have formed by some sort of
diapiric process in or near the partly crystallized
region at the base of the flow (Helz et al., 1989;
Goff, 1996). These structures are generally known
as vesicle cylinders. They are for the most part
vertical and start a few centimetres to tens of
centimetres from the base of the lava flow, in a
zone that also contains pipe vesicles. Cylinder
diameter is generally between about 2 and 20 cm
and, for a given flow, the distance between
cylinders is roughly regular (Rogan et al., 1996;
Goff, 1996). In the case of the Museum flow,
Columbia River Basalt, a fairly horizontal expo-
sure providing a plan view of the interior of a 5 m
thick pahoehoe sheet flow lobe at Sentinel Bluffs,
Sentinel Gap (Washington State), shows that the
distance between 4 and 8 cm diameter vesicle
cylinders is 15 to 20 cm (T. Thordarson and S.
Self, unpublished data). The vesicle cylinders
make up about 14% of the exposed area. Rogan et
al. (1996) reported a crude positive correlation
between radius and spacing of vesicle cylinders in
examples from Auckland, New Zealand. Near the
top of a flow, vesicle cylinders may intersect
vesicle sheets (Rogan et al., 1996; Goff, 1996; Self
et al., 1997). Characteristically, the estimated
vesicularity of these features is greater than that
of their hosts; Rogan et al. (1996) found a factor
from two to five times.
Notable characteristics of segregation vesicles and ves-
icle cylinders in basaltic lavas are their high vesicularity
298 A. Costa et al. / Journal of Volcanology and Geothermal Research 153 (2006) 287–300

(in comparison with their host), an association with
pipe vesicles, sub-vertical orientation, apparently reg-
ular spacing, an ‘origin’ near the base of the lava, a
chemical composition formed by between 35% and
55% (mean 43%) crystallization of the host magma,
liquid viscosities that are several times larger than
those of the host lavas, and densities that are less than
those of the host liquid (Fig. 8). The density difference
will be enhanced by the greater vesicularity of the
segregated material.
4.2. Insights from experiments
The laboratory experiments reported here provide some
insights into how the vesicular segregations of evolved
liquid may form near the base of a cooling basaltic lava.
Both the experiments and basalts are characterized by
similar Re and Bo numbers (Reb 10− 8 to 10− 2; Bo∼10, e.g.,
Manga and Stone, 1994). The typical pattern in the
experiments was that vesiculation caused individual
bubbles to leave the porous zone, carrying a coating and a
tail of liquid from the porous layer. This mechanism may
account for individual vesicles of evolved segregated
material (type (2) in Section 4.1), and is an alternative to
the gas-driven filter pressing mechanism discussed by
Smith (1967), Anderson et al. (1984), Sanders (1986), and
Rogan et al. (1996).
In contrast, well-developed vesicle cylinders with
sub-equal spacing imply a buoyancy-driven instability
of Rayleigh–Taylor type (Goff, 1996) in which diapirs
of bubbly liquid rise from a layer of vesicular evolved
magma near the base of the lava (Fig. 3(c),(d)). Our

Fig. 8. (a) A plot of the relative density difference (ρhost − ρsegr) / ρhost versus viscosity ratio μsegr / μhost of segregation and host liquids calculated from
compositional data (Rogan et al., 1996; Goff, 1996; Stephenson et al., 2000; Caroff et al., 1999). For each segregation–host pair, the viscosity and
density was calculated at the liquidus temperature of the segregation composition using the computer software PELE (PC version of MELTS software
by Ghiorso and Sack, 1995) and CONFLOW (L. Mastin (US Geological Survey) and M. Ghiorso (University of Washington)). (b) A histogram
showing the amount of crystallization required to derive the segregated liquid from the host composition, calculated from the balance of K2O
assuming complete incompatibility of K2O during crystallization. The average value is 43%.
 A. Costa et al. / Journal of Volcanology and Geothermal Research 153 (2006) 287–300 299

Fig. 9. Cartoons illustrating the influence of the relative sizes of constrictions within a porous medium and the bubbles within interstitial liquid on the
way in which bubbles and dense liquid may escape from the porous medium. (a) Small constrictions and large bubbles lead to preferential escape of
bubbles. (b) Large constrictions and small bubbles allow bubbly liquid to flow out of the porous medium.

porous medium experiments only produced this type
of behaviour when a layer of buoyant vesicular lower
liquid had pooled above the porous medium (Section
3.3.3). This case was also mimicked by experiments
with the two layer liquid system illustrated by Fig. 7.
We hypothesize that the ratio of pore size to bubble
size will be important in determining whether a buoyant
layer can be established above the porous layer. Fig. 9(a)
shows a case where the bubbles are larger than the
constrictions between crystals so bubble movement
requires a lot of deformation and can lead to coalescence
and channelling of gas, as was seen in experiments with
glass beads. In this regime, the increase in the volume of
the interstitial material is accommodated by escape of
gas and a little liquid because the constrictions between
crystals act to filter the bubbles from the liquid. In
contrast, Fig. 9(b) shows a more open texture develop-
ing and if the bubbles remain smaller than the con-
strictions, then the increase in volume due to vesiculation
can be accommodated by vesicular liquid flowing
unhindered from the porous medium. The material that
remains in the porous medium has the same vesicularity
as that which escaped. If the layer of escaped vesicular
liquid is thick enough, a Rayleigh–Taylor instability
will generate diapirs of the segregated material, with a
spacing of about 10 times the layer depth (Lister and
Kerr, 1989). The layer of expelled vesicular melt
needed to generate the ca. 20 cm spaced vesicle cy-
linders of the Museum flow (Section 4.1) would there-
fore have only been about 2 cm thick. This hypothesis
predicts a thin source layer for the segregated material
and, because the material that escapes from the porous
layer is compositionally the same as that which remains
in the porous layer, the part of the lava where the
segregated liquid is generated would still be compo-
sitionally similar to the original magma. Paradoxically,
the observation by Goff (1996, page 182) that ‘there is
little evidence that such layers [that is, layers of the low
density vesicular evolved liquid that give rise to vesicle
cylinders by a Rayleigh–Taylor instability] once
existed in the flows’ is entirely consistent with this
mechanism for the formation of vesicle cylinders.
Using the results from our study (summarised in
Section 3.4), we conclude by presenting the following
general model of the formation of vesicle cylinders. In
accordance with field evidence (Rogan et al., 1996;
Goff, 1996; Thordarson, 2000) we consider that the
lava is stagnant at the time when the preserved vesicle
cylinders form. Heat is transferred to the ground by
conduction and crystals start to form in a layer adjacent
to the bottom. Once a rigid permeable framework of
crystals has formed (about 35% to 55% crystals), gas
exsolution and expansion expels vesicular melt up-
wards, forming a thin layer of vesicular evolved liquid.
On becoming a few centimetres thick, this layer be-
comes gravitationally unstable and produces diapirs or
plumes of vesicular melt that rise into the liquid interior
of the stagnant lava (cf. Fig. 3(c),(d)). As crystallization
of the basal region gradually progresses into the flow, a
continuous sequence of diapirs can be generated until
the flow solidifies.
Acknowledgements
We thank Janet Sumner for help with some of the
measurements and experiments, Andy Lloyd for draft-
ing a number of diagrams, and two reviewers for their
helpful comments. We gratefully acknowledge funding
300 A. Costa et al. / Journal of Volcanology and Geothermal Research 153 (2006) 287–300
from the Volcano Dynamics Research Training Network
under EC contract HPRN-CT-1999-00060.
References
Anderson, A.T., Swihart, G.H., Artioli, G., Geiger, C.A., 1984.
Segregation vesicles, gas filter-pressing, and igneous differentia-
tion. J. Geol. 92, 55–72.
Bachmann, O., Bergantz, G.W., 2003. Rejuvenation of the Fish
Canyon magma body: a window into the evolution of large-
volume silicic magma systems. Geology 31, 789–792.
Candela, P.A., 1991. Physics of aqueous phase evolution in plutonic
environments. Am. Mineral. 76, 1081–1091.
Caroff, M., Maury, R.C., Cotten, J., Clement, J.P., 1999. Segregation
structures in vapor-differentiated basaltic flows. Bull. Volcanol. 62,
171–187.
Dullien, F.A.L., 1979. Porous Media Fluid Transport and Pore
Structure. Academic Press, New York, NY. 396 pp.
Ghiorso, M.S., Sack, R.O., 1995. Chemical mass-transfer in magmatic
processes: 4. A revised and internally-consistent thermodynamic
model for the interpolation and extrapolation of liquid-solid
equilibria in magmatic systems at elevated-temperatures and
pressures. Contrib. Mineral. Petrol. 119, 197–212.
Goff, F., 1996. Vesicle cylinders in vapor-differentiated basalt flows. J.
Volcanol. Geotherm. Res. 71, 167–185.
Greenough, J.D., Lee, C.-Y., Fryer, B.J., 1999. Evidence for volatile-
influenced differentiation in a layered alkali basalt flow, Penghu
Islands, Taiwan. Bull. Volcanol. 60, 412–424.
Helz, R.T., Kirschenbaum, H., Marinenko, J.W., 1989. Diapiric
transfer of melt in Kilauea Iki lava lake, Hawaii: a quick, efficient
process of igneous differentiation. Geol. Soc. Amer. Bull. 101,
578–594.
Hoover, S.R., Cashman, K.V., Manga, M., 2001. The yield strength of
subliquidus basalts—experimental results. J. Volcanol. Geotherm.
Res. 107, 1–18.
Huppert, H.E., Sparks, R.S.J., Wilson, J.R., Hallworth, M.A., 1986.
Cooling and crystallization at an inclined plane. Earth Planet. Sci.
Lett. 79, 319–328.
Jaupart, C., Tait, S., 1995. Dynamics of differentiation in magma
reservoirs. J. Geophys. Res. 100, 17615–17636.
Kerr, R.C., Tait, S.R., 1985. Convective exchange between pore fluid
and an overlying reservoir of denser fluid: a post-cumulus process
in layered intrusions. Earth Planet. Sci. Lett. 75, 147–156.
Lister, J.R., Kerr, R.C., 1989. The effect of geometry on the
gravitational instability of a buoyant region of viscous fluid. J.
Fluid Mech. 202, 577–594.
Llewellin, E.W., Mader, H.M., Wilson, S.D.R., 2002. The rheology of
bubbly liquid. Proc. R. Soc. Lond., A 458, 987–1016.
Manga, M., Stone, H.A., 1994. Interactions between bubbles in
magmas and lavas: effects of bubble deformation. J. Volcanol.
Geotherm. Res. 63, 267–279.
Manga, M., Stone, H.A., 1995. Low Reynolds number motion of
bubbles, drops and rigid spheres through fluid–fluid interfaces. J.
Fluid Mech. 287, 279–298.
Marsh, B.D., 2002. On bimodal differentiation by solidification front
instability in basaltic magmas: Part 1. Basic mechanics. Geochim.
Cosmochim. Acta 66, 2211–2229.
McBirney, A.R., Baker, B.H., Nilson, R.H., 1985. Liquid fraction-
ation: Part I. Basic principles and experimental simulations. J.
Volcanol. Geotherm. Res. 24, 1–24.
Ochs, F.A., Lange, R.A., 1999. The density of hydrous magmatic
liquids. Science 283, 1314–1317.
Phillips, J.C., Woods, A.W., 2002. Suppression of large-scale magma
mixing by melt-volatile separation. Earth Planet. Sci. Lett. 204,
47–60.
Philpotts, A.R., Carroll, M., 1996. Physical properties of partly melted
tholeiitic basalt. Geology 24, 1029–1032.
Philpotts, A.R., Carroll, M., Hill, J.M., 1996. Crystal-mush compac-
tion and the origin of pegmatitic segregation sheets in a thick
flood-basalt flow in the Mesozoic Hartford Basin, Connecticut. J.
Petrol. 37, 811–836.
Philpotts, A.R., Shi, J., Brustman, C., 1998. Role of plagioclase crystal
chains in the differentiation of partly crystallized basaltic magma.
Nature 395, 343–346.
Rogan, W., Blake, S., Smith, I., 1996. In situ chemical fractionation in
thin basaltic lava flows: examples from the Auckland volcanic
field, New Zealand, and a general physical model. J. Volcanol.
Geotherm. Res. 74, 89–99.
Saar, M.O., Manga, M., Cashman, K.V., Fremouw, S., 2001.
Numerical models of the onset of yield strength in crystal-melt
suspensions. Earth Planet. Sci. Lett. 187, 367–379.
Sanders, I.S., 1986. Gas filter-pressing origin for segregation vesicles
in dykes. Geol. Mag. 123, 67–72.
Self, S., Keszthelyi, L., Thordarson, T., 1997. The importance of
pahoehoe. Annu. Rev. Earth Planet. Sci. 26, 81–110.
Sisson, T.W., Bacon, C.R., 1999. Gas-driven filter pressing in magmas.
Geology 27, 613–616.
Smith, R.E., 1967. Segregation vesicles in basaltic lava. Am. J. Sci.
265, 696–713.
Sparks, R.S.J., 1978. The dynamics of bubble formation in magmas: a
review and analysis. J. Volcanol. Geotherm. Res. 3, 1–37.
Sparks, R.S.J., Huppert, H.E., Turner, J.S., 1984. The fluid dynamics
of evolving magma chambers. Phil. Trans. Roy. Soc. Lond. Ser. A
310, 511–534.
Spera, F., 1999. A slippery problem with explosive consequences.
Science 283, 1275–1278.
Stephenson, P.J., Zhang, M., Spry, M., 2000. Fractionation modelling
of segregations in the Toomba Basalt, north Queensland. Aust. J.
Earth Sci. 47, 291–300.
Tait, S., Jaupart, C., 1992. Compositional convection in a reactive
crystalline mush and melt differentiation. J. Geophys. Res. 97,
6735–6756.
Thomas, N., Tait, S., Koyaguchi, T., 1993. Mixing of stratified liquids
by the motion of gas bubbles: application to magma mixing. Earth
Planet. Sci. Lett. 115, 161–175.
Thordarson, T., 2000. Physical volcanology of lava flows on Surtsey,
Iceland: a preliminary report. Surtsey Res. 11, 109–126.
Turner, J.S., Huppert, H.E., Sparks, R.S.J., 1983. An experimental
investigation of volatile exsolution in evolving magma chambers.
J. Volcanol. Geotherm. Res. 16, 263–277.
Wright, T.L., Okamura, R.T., 1977. Cooling and crystallization of
tholeiitic basalt, 1965 Makaopuhi lava lake, Hawaii. U.S. Geol.
Surv. Prof. Pap. 1004, 1–78.