Periodic Reporting for period 4 - MAGMA (Melting And Geodynamic Models of Ascent)
Reporting period: 2023-04-01 to 2023-09-30
The generation and migration of magma through the lithosphere results in spectacular geological processes, such as volcanic eruptions, giant ore deposits, large magmatic intrusions, and is responsible for the formation of continents on Earth.
Since magmatic systems develop on timescales of millions of years and are not directly accessible, we have to reconstruct them indirectly, by studying exhumed magmatic intrusions, or by using geophysical methods. Interpreting these data is complicated, as geophysical techniques only give a present-day snapshot, while the geological record yields an incomplete picture of the underlying processes. Therefore, most existing ideas on how magmatic systems work remain conceptual and are not necessarily consistent with the mechanics of the lithosphere.
The aim of this project is to develop and employ a new generation of computer models to simulate the full magmatic system in volcanic arcs in a self-consistent manner, while taking both realistic rock rheologies and evolving melt chemistry into account.
Specific aims are:
1. Develop new methods to create mechanically consistent interpretations of active magmatic plumbing systems by combining geophysical and petrological data with geodynamic inverse models.
2. Develop new tools to take the melting, crystallisation and chemical evolution of rocks into account in thermo-mechanical numerical models so we can directly compare simulations with geological observations.
3. Develop new software to simulate magmatic processes in deforming visco-elasto-plastic rocks.
4. Apply the modelling approaches to currently active magmatic systems
5. Obtain insights into the physics of magma migration over geological timescales by comparing numerical simulations with geological constraints.
This allows interpreting available data in a physically and chemically consistent manner and gives new insights in how magmatic systems evolve over geological timescales.
Since magmatic systems develop on timescales of millions of years and are not directly accessible, we have to reconstruct them indirectly, by studying exhumed magmatic intrusions, or by using geophysical methods. Interpreting these data is complicated, as geophysical techniques only give a present-day snapshot, while the geological record yields an incomplete picture of the underlying processes. Therefore, most existing ideas on how magmatic systems work remain conceptual and are not necessarily consistent with the mechanics of the lithosphere.
The aim of this project is to develop and employ a new generation of computer models to simulate the full magmatic system in volcanic arcs in a self-consistent manner, while taking both realistic rock rheologies and evolving melt chemistry into account.
Specific aims are:
1. Develop new methods to create mechanically consistent interpretations of active magmatic plumbing systems by combining geophysical and petrological data with geodynamic inverse models.
2. Develop new tools to take the melting, crystallisation and chemical evolution of rocks into account in thermo-mechanical numerical models so we can directly compare simulations with geological observations.
3. Develop new software to simulate magmatic processes in deforming visco-elasto-plastic rocks.
4. Apply the modelling approaches to currently active magmatic systems
5. Obtain insights into the physics of magma migration over geological timescales by comparing numerical simulations with geological constraints.
This allows interpreting available data in a physically and chemically consistent manner and gives new insights in how magmatic systems evolve over geological timescales.
A significant contribution of this project was to develop open-access software to simulate magmatic processes, such that the wider geoscience community will benefit from them (also beyond the duration of the MAGMA project). As many processes require 3D simulations, it implies that the software should work on laptops as well as on parallel high-performance computers. Given that the next generation computers utilize GPU’s rather than CPU’s being able to leverage that is desirable as well. In addition, the software should be reasonably easy to use and allow users to apply simulations to a particular region in a short amount of time.
We have made significant progress towards this, by extending our existing 3D thermo-mechanical code LaMEM, and by developing completely new modular software tools (using the Julia scientific language) which works on CPUs and GPUs.
The technical development evolved around the following themes:
1) Development of inverse modelling approaches. We have extended our existing open-source 3D geodynamic modelling software (LaMEM), to take adjoint-based sensitivity analysis and gradient-based inversion into account. This makes it easier to automatically test the key model parameters that, for example, control surface uplift in a volcano in complex thermos-mechanical models. We have also developed new ways to automatically change the geometry for which we invert, and an application of these approach to the Puna magmatic system allowed us to interpret available geophysical data in a mechanically consistent manner.
2) Development of a new, massively parallel, software package to perform thermodynamic Gibbs energy minimization calculations to simulate the evolving chemistry of (partially molten) rocks that takes the latest thermodynamic hydrous mafic melting models (along with several previous melting databases) into account. Our open-source code, MAGEMin, is already able to replace many of the currently employed software packages, while being faster, due to using a different algorithm. It is also particularly well-suited to integrate with other codes thanks its parallel Julia interface.
3) Exploring new approaches to couple thermodynamic and geodynamic models, using the "classical" (and slow) thermodynamic modelling approaches but with a large database of precomputed diagrams. This allowed tracking the evolving chemistry and making testable predictions that can directly be compared with observations.
4) Developing new approaches to take mode-1 plasticity into account in geodynamic thermomechanical models, which is crucial to simulate melt transport through dikes while still being able to deal with realistic visco-elasto-plastic host rock rheologies. Our methods work with both implicit finite element solvers and with pseudo-transient staggered finite difference discretizations on GPUs.
5) Rather than developing a single new code, we developed a range of open-source modular Julia packages, which have the advantage that they can be integrated in different codes, are easier to test, and lowers the barrier for others in the community to contribute. The software packages allow creating 3D model setups from diverse geoscientific data, integrate complex rheologies and material parameters, perform non-dimensionalisation, compute the thermal evolution of magmatic systems, or perform thermo-mechanical simulations. We focus on both pseudo-transient modelling approaches which work particularly well on GPU architectures, as well as on the more classical implicit modelling approaches that run on MPI-parallel machines. We have developed new, two-phase flow software package that takes visco-elasto-plasticity into account along with the effects of thermal expansion/contraction.
6) We made it much easier to use our various software packages in both teaching and research by providing precompiled libraries (for example of LaMEM and MAGEMin), Julia interfaces along with graphical user interfaces.
We have applied the software packages above to simulate and understand the chemical and mechanical evolution or magmatic systems on generic models of arcs, and applied inverse modelling approaches to interpret the present-day magmatic system of the Puna area in a mechanically consistent manner. We also employed models to address the magmatism in the early Earth, and highlighted the importance of magmatic processes in large scale tectonic processes (such as the creation of the Caribbean or the long term dynamics of the Galapagos plume).
We have made significant progress towards this, by extending our existing 3D thermo-mechanical code LaMEM, and by developing completely new modular software tools (using the Julia scientific language) which works on CPUs and GPUs.
The technical development evolved around the following themes:
1) Development of inverse modelling approaches. We have extended our existing open-source 3D geodynamic modelling software (LaMEM), to take adjoint-based sensitivity analysis and gradient-based inversion into account. This makes it easier to automatically test the key model parameters that, for example, control surface uplift in a volcano in complex thermos-mechanical models. We have also developed new ways to automatically change the geometry for which we invert, and an application of these approach to the Puna magmatic system allowed us to interpret available geophysical data in a mechanically consistent manner.
2) Development of a new, massively parallel, software package to perform thermodynamic Gibbs energy minimization calculations to simulate the evolving chemistry of (partially molten) rocks that takes the latest thermodynamic hydrous mafic melting models (along with several previous melting databases) into account. Our open-source code, MAGEMin, is already able to replace many of the currently employed software packages, while being faster, due to using a different algorithm. It is also particularly well-suited to integrate with other codes thanks its parallel Julia interface.
3) Exploring new approaches to couple thermodynamic and geodynamic models, using the "classical" (and slow) thermodynamic modelling approaches but with a large database of precomputed diagrams. This allowed tracking the evolving chemistry and making testable predictions that can directly be compared with observations.
4) Developing new approaches to take mode-1 plasticity into account in geodynamic thermomechanical models, which is crucial to simulate melt transport through dikes while still being able to deal with realistic visco-elasto-plastic host rock rheologies. Our methods work with both implicit finite element solvers and with pseudo-transient staggered finite difference discretizations on GPUs.
5) Rather than developing a single new code, we developed a range of open-source modular Julia packages, which have the advantage that they can be integrated in different codes, are easier to test, and lowers the barrier for others in the community to contribute. The software packages allow creating 3D model setups from diverse geoscientific data, integrate complex rheologies and material parameters, perform non-dimensionalisation, compute the thermal evolution of magmatic systems, or perform thermo-mechanical simulations. We focus on both pseudo-transient modelling approaches which work particularly well on GPU architectures, as well as on the more classical implicit modelling approaches that run on MPI-parallel machines. We have developed new, two-phase flow software package that takes visco-elasto-plasticity into account along with the effects of thermal expansion/contraction.
6) We made it much easier to use our various software packages in both teaching and research by providing precompiled libraries (for example of LaMEM and MAGEMin), Julia interfaces along with graphical user interfaces.
We have applied the software packages above to simulate and understand the chemical and mechanical evolution or magmatic systems on generic models of arcs, and applied inverse modelling approaches to interpret the present-day magmatic system of the Puna area in a mechanically consistent manner. We also employed models to address the magmatism in the early Earth, and highlighted the importance of magmatic processes in large scale tectonic processes (such as the creation of the Caribbean or the long term dynamics of the Galapagos plume).
The new thermodynamic and geodynamic software packages we developed are well beyond the current state of the art of the community. We are among the first in the geodynamics community to fully embrace the Julia scientific language and we believe that the packages we provide will be a good basis for a range of new codes over the coming years. This is particularly well-suited for our community, where the status-quo was based on several open source software packages (ASPECT, CITCOM, LaMEM) which all have 50’000-100’000 lines of code. Unsurprisingly it is very difficult for new users to contribute to these codes because it takes significant amounts of time to become familiar with it. The Julia pseudotransient solvers we have partly been using within MAGMA reduce this to several 100 lines of code, which work efficiently on one and multiple GPUs. With our packages we can further lower the barrier to go from a proof-of-concept code to a full production code (that is applied to a specific region).