- Generated on Mon May 30 2022 16:28:11 for PetIBM by
1.9.3
|
PetIBM 0.5.4
Toolbox and applications of the immersed-boundary method for distributed-memory architectures
|
PetIBM reads input parameters from a YAML configuration file (typically called config.yaml) that is located in the simulation directory (i.e. directory where the numerical solution will be written).
If the current working directory is not the simulation directory, the user should provide the command-line argument -directory <simulation-directory>.
If the configuration file is located in another directory, the user should provide the path of the file using the command-line argument -config <file-path>. (The path can be absolute or relative to the current working directory.)
A basic YAML configuration file for a 2D run will look like:
The configuration contains three mandatory nodes: mesh, flow, parameters. The node bodies is required when running a simulation with at least one boundary immersed in the computational domain. The node probes can be used to monitor a specific scalar field variable (velocity components or pressure) in a specific subregion of the computational domain.
meshmesh controls the parameters of the structured Cartesian grid to be created. We need to provide sufficient information to generate the gridline coordinates in the direction x, y, and z. (For 2D runs, the z direction should be omitted.)
A gridline is defined by the location of a starting point (start), which corresponds to the minimum value along the gridline (for example, location of the left boundary in the x-direction).
A gridline is divided into consecutive sub-domains (subDomains), each one being defined with the coordinate of a ending point (end), a number of cells (cells), and optionally a stretching ratio (stretchRatio). The stretching ratio represents the ratio between the width of two consecutive cells. (A ratio greater than 1.0 means that the cell width increases along the gridline in the positive direction.)
Based on those information, PetIBM computes the cell widths and the user should make sure that the cells at the interface between two sub-domains are approximately of equal size.
Example:
The following node will generate a 2D structured Cartesian grid with 1704 cells in the x and y directions in the domain [-15, 15]x[-15, 15] where the mesh is uniform in the sub-domain [-0.52, 1.48]x[-2, 2] and stretched to the external boundaries with a constant ratio of 1.01.
flowflow prescribed the characteristics of the fluid as well as the initial and boundary conditions of the flow velocity.
nu is the kinematic viscosity of the fluid.
initialVelocity describes the initial velocity vector field. It is a list of length equal to the flow's dimension. The first element in the list denotes the initial value/function for u-velocity, while the second element denotes the v-velocity. The third element, if it exists, is for the w-velocity. There are two types of legal values in the list:
"cos(2*pi*x) * sin(2*pi*y)". We use SymEngine to parse symbolic expressions. So only the math operations and math functions available in SymEngine are allowed here. SymEngine is the backend of the well-known python library SymPy, so SymPy's documentation may be a good start point to know what operations and functions are available. For the independent variables, only x, y, z, t, and nu are allowed. The case at examples/navierstokes/taylorgreenvortex provides a example of using symbolic expression for initialization.initialPressure is an optional parameter and needs a single value if present in the configuration. Similar to initialVelocity, the value can be a floating point number or a symbolic expression string. If initialPressure is absent, the pressure filed will be initialized with zero.
boundaryConditions lists the type and value of a velocity component (u, v, or w) for all boundaries. The boundary locations are: xMinus and xPlus for the left and right, yMinus and yPlus for the bottom and top, and zMinus and zPlus for the front and back boundaries. (zMinus and zPLus should be omitted for 2D runs.) PetIBM implements different type of boundary conditions:
DIRICHLET); the value corresponds the value of the velocity component at the boundary;NEUMANN); the value represents the value of the derivative normal to the boundary of the velocity component;CONVECTIVE); the value is the speed at which the velocity component is convected;PERIODIC); the value has no meaning and can be set to whatever number.The following node corresponds to flow characteristics for a 3D cavity flow (initially at rest) at Reynolds number 500 (based on the kinematic viscosity, the length of the cavity, and the speed of the lid-driven wall) with a lid-driven wall at the top boundary (moving with speed 1 in the x direction) and with periodic boundary conditions in the z direction.
parametersThis node gather various parameters regarding the advancement of a simulation.
dt: the time-step size.startStep: index of the stating time step (default is zero).nt: number of time steps to compute.nsave: frequency (in number of time steps) of saving for the numerical solution.nrestart: frequency (in number of time steps) of saving for the convective and diffusive terms; those terms will required upon restart of a run at a time step different from 0.convection: time scheme for the convective terms; choices are the default explicit Euler method (EULER_EXPLICIT) or an explicit second-order Adams-Bashforth scheme (ADAMS_BASHFORTH_2).diffusion: time scheme for the diffusive terms; choices are the default implicit Euler method (EULER_IMPLICIT), an explicit Euler method (EULER_EXPLICIT), or a second-order Crank-Nicolson scheme (CRANK_NICOLSON).BN: order of the truncated Taylor series expansion of the implicit matrix A (where A is the left-hand side operator of the system for the intermediate velocity vector). The default value is 1, which leads to the identity operator scaled by the time-step size.delta: regularized delta function to use; choices are ROMA_ET_AL_1999 (3-point kernel) and PESKIN_2002 (4-point kernel).velocitySolver, poissonSolver, and forcesSolver (for the decoupled version of the immersed-boundary projection method) each references the type of hardware used to solve the linear system (either CPU or GPU) and the path (relative to the YAML configuration file) of the file containing the parameters for the linear solver.In the following example, PetIBM will run 1000 time steps (from time step 0) with a time increment of 0.01, saving the numerical solution (velocity vector field, pressure scalar field, and Lagrangian boundary forces) every 100 time steps and saving the convective and diffusive terms every 200 time steps.
In the example above, the velocity system will be solved on CPU with PETSc KSP and the parameters of the linear solver can be prescribed in the file solversPetscOptions.info. (The PETSc prefix for the arguments is -velocity_, -poisson_, and -forces_ for the velocity system, the Poisson system, and the Lagrangian forces system, respectively.)
The Poisson system will be solved on GPU devices with the NVIDIA AmgX library and the parameters of the linear solver are prescribed in the file solversAmgXOptions.info.
bodiesbodies provides information about the boundaries immersed in the computational domain.
For example, the following node stipulates that there are two immersed boundaries defined by points coordinates that can be read in the ASCII files boundary1.txt and boundary2.txt, respectively. (The path of the files can be either absolute or relative to the simulation directory.)
The file boundary1.txt contains the coordinates of the first boundary with the first line storing the number of points. The other lines contain the coordinates of points describing the boundary (2 columns for 2D simulations, 3 columns for 3D simulations).
Example of a 2D boundary with 4 points:
probesThe YAML node probes contains a sequence of probes. A probe can be used to monitor a scalar field variable (e.g., a velocity component or the pressure) either in a sub-volume of the domain or at a single point.
General configuration of a probe:
name: (optional) name of the probe; default name is unnamed.type: (required) type of the probe (VOLUME for to monitor a sub-volume, POINT to interpolate the value a single point).field: (required) name of the scalar field variable (supported variables are p for the pressure, u, v, and w for the velocity components).path: (required) path of the file to write the data (relative to the output directory) into (parent folder needs to exist).n_monitor: (optional) monitoring frequency (as a number of time steps); default value is 1.t_start: (optional) time value to start monitoring the variable; default is 0.0.t_end: (optional) time value to finish monitoring the variable; default is 1e12.Additional configuration for a volume probe:
box: (required) limits of the box in the x, y, and z directions.viewer: (optional) type of the PETSc Viewer object to use to output the data (choices are ascii and hdf5); default value is ascii.n_sum: (optional) number of time-steps over which the data at accumulated; after n_sum time-steps, the accumulated data are averaged in time and output to file; default is 0 (i.e., does not accumulate).Additional configuration for a point probe:
loc: (required) location of the point to monitor (a linear interpolation is performed at that point; bi-linear for two-dimensional runs, tri-linear for three-dimensional runs).In the following example, the computational domain extends from -1 to +1 in the x and y directions and we monitor the solution of the pressure in a sub-volume and the x-component of the velocity at a single point. We monitor the pressure every time step in the subregion [-0.5, 0.0]x[-0.5, 0.0] but accumulate (sum) the data over a period of 100 time steps; the time-averaged data is output to file every 100 time steps into HDF5 format. We monitor the x-component of the velocity every 20 time-steps at the point [0.25, 0.25]; the time-step value is output to an ASCII file every 20 time-steps. In this example, output files for the probes are saved in the solution sub-folder of the simulation directory.
1.9.3