DIFFUSION IN SILICATE MELTS 207 
rate of diffusion in fused rock-forming silicates. Those here 
described are to be considered of a preliminary nature. They are 
not devised with the purpose of establishing precise values for 
diffusion coefficients from which general theoretical conclusions 
might be drawn, though this is recognized as a desirable ultimate 
goal. They may serve rather to aid the geologist in deciding what 
he may and what he may not reasonably attribute to diffusion. 
THE METHOD OF EXPERIMENT 
The method followed was that of permitting the diffusion 
against gravity of a heavy liquid placed in the lower part of a 
crucible into a lighter liquid in the upper part. Ostwald has said 
that to make accurate experiments on diffusion is one of the hardest 
problems in practical physics on account of the difficulty experi- 
enced in eliminating convection currents. It is to be noted that 
this is especially true of aqueous solutions. Water is a thin liquid 
with a relatively high coefficient of thermal expansion. The 
driving force of convection, viz., difference of density, is therefore 
large and the resistance to it (viscosity) small. In silicates, how- 
ever, these conditions are reversed, the viscosity being relatively 
great and the thermal expansion relatively small. While a small 
difference of temperature may establish convection in aqueous 
solutions, it is not to be expected to have a comparable effect 
in silicates. In the case of aqueous solutions the density gradient 
resulting from the composition gradient may be very small, but 
in the case of the silicates used it is quite large. If the relative 
densities of diopside and plagioclase liquids at high temperatures 
are comparable with those at lower temperatures, it can readily be 
shown that it would be necessary to have a temperature gradient 
of 20° per mm. in order to counteract the density gradient due to 
I per cent per mm. change of composition. Gradients of com- 
position of this magnitude exist throughout most of the period of 
experiment. Such temperature gradients are, however, entirely 
lacking; not only this, but it is easy to make the moderate tem- 
perature gradient that does exist of such a sign that it acts together 
with the composition gradient instead of counteracting it. This 
is accomplished, of course, by making the temperature of the upper 
