Repository files navigation

A finite differences simulation library.

This package lets you quickly build simple numerical simulations.

• Quickly build finite difference simulations with an arbitrary number of domains, scalars, dimensions and boundary conditions.
• Concise, clear syntax.
• Build models for CPU or GPU using Scipy or Cupy sparse libraries.
• Arbitrary derivative order and approximation accuracy.
• Partial model updates minimize overheads in iterative solutions.
• Implicit transient simulation.

pip install fastfd

# Import fastfd and select the scipy sparse library (or cupy for GPU support)importfastfdasffdffd.sparse_lib('scipy')importnumpyasnp# Define axesx=ffd.LinearAxis('x',start=0,stop=1,num=201)y=ffd.LinearAxis('y',start=0,stop=1,num=201)# Define scalarsT=ffd.Scalar('T', [x,y],accuracy=4)# Define the modelmodel=ffd.FDModel([T])# Set model governing equations (thermal diffusion)model.update_equations({'Conductivity': ((T.d('x',2)+T.d('y',2)),0),})# Set model boundary conditionsmodel.update_bocos({'Tx=1 adiabatic': (T.i[-1, :],T.d('x')[-1, :],0),'Ty=1 adiabatic': (T.i[:,-1],T.d('y')[:,-1],0),'Tx=0 sinewave': (T.i[0, :],T.i[0, :],100*np.sin(x.coords*2*np.pi)),'Ty=0 sinewave': (T.i[:,0],T.i[:,0],-100*np.sin(y.coords*2*np.pi)),})# Solve the modelresult=model.solve()

FastFD can be set to use the Cupyx sparse libraries, though support is still a little rough around the edges. The currentversion of Cupy (8.5) only supports a least squares solver ('lsqr') which is much slower than the default 'spsolve'.However, spsolve will be implemented in the upcoming release of Cupy 9.0.

See example notebooks here: ./docs/examples for a more detailed explanation of how to use.