![https://github.com/NaokiHori/SimpleNSSolver/blob/main/docs/source/thumbnail.gif](https://github.com/NaokiHori/SimpleNSSolver/raw/main/docs/source/thumbnail.gif)
This library numerically solves the incompressible Navier-Stokes equations (coupled with a temperature field) in two- and three-dimensional Cartesian domains using the finite-difference method.
The main objective is to develop a library where the implementation and the background knowledge are closely linked via a documentation and various examples, so that users can understand how and why things are treated.
- An energy-consistent treatment of advective, pressure-gradient, and diffusive terms, correctly replicating properties of the conservation laws.
- MPI parallelisation.
- Efficient FFT-based direct Poisson solver.
- Explicit / implicit treatments of diffusive terms in all spatial directions.
Please refer to the documentation for details.
- C compiler
- GNU Make
- MPI
- FFTW3
- Git
- Python3 with NumPy (for flow-field initialisation and for post-processing)
It should be convenient to use a proper package manager, e.g.:
sudo apt-get -y update
sudo apt-get -y install gcc libopenmpi-dev libfftw3-dev make
Also install Python3.
Installation of the Command Line Tools for Xcode
is usually required, which is followed by
brew install gcc open-mpi fftw make
Also install Python3.
Not supported. Please consider to use Windows Subsystem for Linux for instance.
Prepare workplace
mkdir -p /path/to/your/directory cd /path/to/your/directory
Get source
git clone --recurse-submodules https://github.com/NaokiHori/SimpleNSSolver cd SimpleNSSolver
Set initial condition
Here
Python3
is used to initialise the flow fields conveniently. One can giveNPY
files in different way underinitial_condition/output/
.cd initial_condition make output bash exec.sh cd ..
Build NS solver
make output make all
bash exec.sh
launches the simulator and integrate the equations in time, giving e.g.
DOMAIN
glsizes[0]: 128
glsizes[1]: 256
lengths[0]: 1.0000000e+00
lengths[1]: 2.0000000e+00
FLUID
Ra: 1.0000000e+08
Pr: 1.0000000e+01
Momentum diffusivity: 3.1622777e-04
Temperature diffusivity: 3.1622777e-05
diffusive treatment in x: implicit
diffusive treatment in y: explicit
LOGGING
next: 5.000e-01
rate: 5.000e-01
SAVE
dest: output/save/step
next: 2.000e+01
rate: 2.000e+01
STATISTICS
dest: output/stat/step
next: 1.000e+02
rate: 1.000e-01
step: 0, time: 0.0000000e+00
timemax: 2.0000000e+02, wtimemax: 6.0000000e+02
coefs: (adv) 9.500e-01, (dif) 9.500e-01
DFT-based solver is used
step 11, time 0.5, dt 4.58e-02, elapsed 2.1 [sec]
step 22, time 1.0, dt 4.58e-02, elapsed 2.2 [sec]
step 33, time 1.5, dt 4.58e-02, elapsed 2.3 [sec]
step 44, time 2.0, dt 4.58e-02, elapsed 2.4 [sec]
step 55, time 2.5, dt 4.58e-02, elapsed 2.4 [sec]
...
step 8193, time 197.5, dt 3.06e-02, elapsed 91.9 [sec]
step 8210, time 198.0, dt 2.79e-02, elapsed 92.2 [sec]
step 8228, time 198.5, dt 2.79e-02, elapsed 92.5 [sec]
step 8246, time 199.0, dt 2.90e-02, elapsed 93.0 [sec]
step 8263, time 199.5, dt 3.07e-02, elapsed 93.2 [sec]
You see that the solver (e.g. DOMAIN
and FLUID
) is initialised and parameters are loaded from the NPY
files prepared in the previous step, which is followed by the integration of the equations in time.
Several log files, snapshots of the flow fields (which are used to restart the simulation and to process the flow fields later), and collected statistics are stored in output
directory:
output
├── log
│ ├── xxxxx.dat
│ ├── yyyyy.dat
...
│ └── zzzzz.dat
├── save
│ ├── step00000xxxxx
│ ├── step00000yyyyy
...
│ └── step00000zzzzz
└── stat
└── step00000zzzzz
Log files (files under output/log
directory) are written in ASCII format, which are to monitor the progress.
For example, since I adopt the FFT-based Poisson solver in this project, local divergence of the flow field should be small enough, which is written in output/log/max_divergence.dat
:
Energy injections and dissipations are also monitored, from which the Nusselt number (computed based on several different definitions) can be extracted:
Flow fields and statistical data are stored in NPY format using SimpleNpyIO.
When Python3
with NumPy
and Matplotlib
is installed, one can easily visualise the flow fields:
or statistics (e.g., mean advective and diffusive heat transfer):
Note that all the results shown here are automatically updated to maintain / improve the code quality, and all scripts to produce the above figures are available in the examples. See the documentation for more details.
By default, this project simulates two-dimensional cases because they are easy to test and thus can be a good starting point.
When a three-dimensional version is needed, checkout 3d
branch.
Note that the main
branch contains both dimensions, which is to maintain both cases at the same time (mainly for personal use).
Please refer to the examples, where several small-scale 3D simulations are attempted as a part of the continuous integration.
Feel free to ask questions, report bugs, suggest new features, polish documentation at issues.
The development of this CFD solver is largely motivated by CaNS and AFiD.
I would like to thank Dr. Pedro Costa, Dr. Marco Rosti and Dr. Chris Howland, among others, for fruitful discussions during my time at KTH Royal Institute of Technology in Stockholm, the University of Tokyo and University of Twente.