Hello folks,
Hope you’re doing well. Recently I have been trying to reproduce Citron202 paper “Effects of Heat‐Producing Elements on the Stability of Deep Mantle Thermochemical Piles”. The paper provided prm files. I focused on the reference case where B=0.6, HPE in dense layer is 10 times higher than ambient mantle and is constant, dense layer is 300 km thick right on CMB. The prm of this case is adapted to ASPECT 3.0.0 (Citron used v2.0.1). What surprised me is that it took more than 10 hours to initialize the particles, after that the computation is basically efficient. Citron used uniform box to generate 11 million particles. I don’t know why it’s so slow to start the run. The original Citron prm and my prm (particle_const5.prm) are attached at the end.
Instead of using uniform box, I also used reference cell to start the run immediately. I tried 2 subcases, one is reference cell + 8 particles in each direction in each cell (name it particle_const), the other one is reference cell with default settings (2 particles in each direction in each cell, name it particle_const1), total particles and min/max particle in a cell are the same as uniform box case (name it particle_const5). I found they have completely different particle configurations, esp. near 300 km, the dense layer interface (see attached plots). These different particle setups lead to different evolution statistics. particle_const case has sharp density interface at 300km, it has some 300 Myr offset in plumes formation (see Vrms_particle_const.png where Vrms is root-mean-square velocity); while particle_const1 case shows similar plume initialization as Citron. particle_const5 case (uniform box) shows the best approximation. However, all subsequent statistics are different. I don’t quite understand.
Later I realized that the thermal conductivity has been changed in energy equation for version >=2.1. So I also had a test case particle_const4 using the same thermal conductivity as in version 2.0.1. Old ASPECT used k+entropy viscosity for effective thermal conductivity to get more diffusion and be more stable. particle_const4.png shows pretty similar result from particle_const5, though tiny differences can be seen.
So I got 3 questions, one is: why are all cases different after plume initialization? The other one is: so far the case particle_const4 has identical configurations as Citron, but the results are different, why? I’m suspecting that particle module from v3.0.0 is different from that in v2.0.1. Finally, to reproduce Citron’s results, shall I compile v2.0.1 and run the code, or can we consider my results are good enough?
Any comments/ideas/suggestions are welcome! Thanks.
Best,
Mingming
particle_const5.prm (11.1 KB)
Citron_original.prm (7.4 KB)