[sundials-users] Integration of matrices with 10^5 - 10^6 elements
00002e847f03e96...@listserv.llnl.gov
Hi
I am not sure if I am writing to the correct address - I apologise if that's not the case and would be grateful if you could tell me which would be an appropriate form of contact (user group, other mail etc).
I'm based in Portugal at the New University of Lisbon (NOVA), and the work below is part of a multi-year collaborative project with the University of Dalhousie and the Canadian DFO.
1. I've used CVODE to deal with solving equations for a diagenetic model for marine sediments - the model is a system of stiff equations for the biogeochemistry, and standard advection-diffusion of the chemical species vertically in the sediment layers. CVODE works brilliantly and is stunningly fast. I would like to thank the research team that developed this package;
2. I've been working on another piece of code (in C++) that simulates deposition of organic particles from fish farms to the ocean floor - it is actually this loading that drives the diagenetic model in (1). This application is parallelised and I have a used a couple of tricks to make it fast, but it uses explicit integration solvers such as Forward Euler and Runge-Kutta. The software is built as a Windows DLL and it is already fast enough for what I need. However, I wanted to try an implement a version using one of the Sundials libraries: since we also simulate the spread of dissolved substances such as ammonia, the spatial domains can be quite large, e.g. 100 X 120 XY cells and 20-30 vertical (Z) layers, i.e. an array of up to 500,000 spatial cells to be integrated at each timestep. Since this is a water simulation, there is advection-diffusion simulated among cells, but there are no stiff equations to solve, in this case we are simulating the physics rather than the biogeochemistry - the main issue would be the size of the arrays to be integrated and memory considerations. I wanted to ask what would be the best component within Sundials to use as a solver.
Apologies for the long mail and thank-you for any help and advice.
Kind regards
Joao G. Ferreira
To unsubscribe from the SUNDIALS-USERS list: write to: mailto:SUNDIALS-USERS-...@LISTSERV.LLNL.GOV