Convection and Stirring in the Earth
Convection in the Earth's mantle causes continental drift and the movement of the tectonic plates. The subduction of tectonic plates back into the mantle introduces chemical heterogeneities into the mantle, and these heterogeneities are gradually stirred and homogenised by the continuing convection of the mantle. Heterogeneities as old as 2 billion yeas are detectable through variations in trace elements and isotopes in mantle-derived rocks. In this project, the process of creation and stirring of chemical heterogeneiteis is explored in numerical models incorporating simulations of plates with large spatial variations in viscosity. Such modelling is required in order to permit quantitative interpretaions of accumulated geochemical data. A preliminary series of two-dimensional models has yielded promising results, and the models are now being extended into three dimensions and to the more extreme conditions of the early stages of the Earth's history. There are geochemical indications that a large reservoir of basaltic composition existed early in Earth history, and a major goal of this work is to explore the dynamical feasibility and implications of such a reservoir.
|
Principal Investigator Geoff DaviesGeodynamics, RSES Australian National University |
Project r01 |
|
Co-Investigator Alison LeitchGeodynamics, RSES Australian National University |
RFCD Codes 260299 |
Significant Achievements, Anticipated Outcomes and Future Work
In this project, the chemical evolution of the mantle is computed in the context of the dynamical and thermal evolution. Chemical components are included through the use of tracer points. These are stirred by mantle convection, and they can modify the convection if the tracers carry mass, simulating components of different density. The tendency of heavy components to settle has been modelled in previous work, but in that work the slow cooling of the mantle was not directly included because the high temperatures of earlier times were computationally challenging. Also, only two-dimensional models could be run with the available code. This work is now being extended to accommodate the higher mantle temperatures of early eras, and from two dimensional modelling to three dimensional modelling. The latter is possible because a three-dimensional code has been obtained from overseas colleagues.
Initial runs in two dimensions have revealed that settling of heavy components is more pronounced in the early, hotter mantle, as expected, because the viscosity of the mantle is lower. However deficiencies in the 2-D code, which had developed over nearly 30 years, were also revealed and it was decided to do a major restructuring and recasting of the code into Fortran 90. This has taken longer than expected and delayed progress on all aspects of the project. An imminent sabbatical leave for five months will also delay progress.
The 3-D code requires some adaptation to the present project, and significant progress has been made. In particular, tectonic plates are being simulated more realistically than in any previous 3-D models. Some development of plate modelling is still required, and the algorithms for tracer-point advection need some minor modification. Progress should be rapid once this part of the project is resumed.
Computational Techniques Used
The 2-D code “Conmg” is locally written. It runs on a 4-processor node with about 70% parallel efficiency. Conmg (2D) uses efficient spatial multigrid on a regular cartesian mesh. Time stepping is by alternating-direction implicit (ADI). The 3-D code “Terra” was developed at Los Alamos National Lab, US. It is a well-tested code that has been tuned on a Linux Cluster at Princeton and runs with efficiencies of 80% or more and with excellent scaling properties. Terra (3-D) is a finite-element code on an icosahedral grid mapped to the sphere. Spatial solutions use multigrid combined with conjugate gradients. Time stepping is by second-order Runge Kutta.
Publications, Awards and External Funding
External Funding and Awards
None.
Publications
None.