Merge pull request #505 from brighthe/master

Added `test ls_ solver` program

example/levelset/lsf_evolve_fem_without_reinit.py

@@ -0,0 +1 @@
+"exp"

fealpy/levelset/ls_fem_solver.py

@@ -15,25 +15,61 @@
 
 
 class LSFEMSolver(LSSolver):
+    """
+    A finite element solver for the level set evolution equation, which tracks
+    the evolution of an interface driven by a velocity field. It discretizes
+    the transport equation using Crank-Nicolson method in time and finite
+    element method in space.
+    """
     def __init__(self, space, u=None):
+        """
+        Initialize the finite element solver for the level set evolution.
+
+        Parameters:
+        - space : The finite element space over which the system is defined.
+        - u : The velocity field driving the interface evolution. It should be a vector
+            function defined on the mesh nodes.
+
+        The bilinear forms for the mass and convection are assembled here. The
+        mass matrix M represents the mass term in the equation, while the
+        convection matrix C represents the velocity field's convection effect.
+        """
         self.space = space
+
+        # Assemble the mass matrix using the ScalarMassIntegrator which represents the L2 inner product of the finite element functions.
         bform = BilinearForm(space)
         bform.add_domain_integrator(ScalarMassIntegrator())
-        self.M = bform.assembly() # TODO: 实现快速组装方法
+        self.M = bform.assembly() # TODO:Implement a fast assembly method
 
         self.u = u
+
+        # Assemble the convection matrix only if a velocity field is provided.
         if u is not None:
             bform = BilinearForm(space)
+
+            # The convection integrator accounts for the transport effect of the velocity field on the level set function.
             bform.add_domain_integrator(ScalarConvectionIntegrator(c = u))
-            self.C = bform.assembly() # TODO：实现快速组装方法
+            self.C = bform.assembly() # TODO:Implement a fast assembly method
 
     def solve(self, phi0, dt, u=None, tol=1e-8):
-        '''
-        PDE_solve
-        '''
+        """
+        Solve the level set evolution equation for one time step using the
+        provided initial condition and velocity field.
+
+        Parameters:
+        - phi0 : The initial condition for the level set function.
+        - dt : Time step size for the evolution.
+        - u : (Optional) Updated velocity field for the evolution.
+        - tol : Tolerance for the linear system solver.
+
+        The function solves for phi^{n+1} given phi^n (phi0) using the
+        discretized Crank-Nicolson scheme. It returns the updated level set
+        function after one time step.
+        """
         space = self.space
         M = self.M
 
+        # Use the provided velocity field u for this time step if given, otherwise use the previously stored velocity field.
         if u is None:
             C = self.C 
             if C is None:
@@ -42,53 +78,93 @@ def solve(self, phi0, dt, u=None, tol=1e-8):
             bform = BilinearForm(space)
             bform.add_domain_integrator(ScalarConvectionIntegrator(c = u))
             C = bform.assembly()
+
+        # The system matrix A is composed of the mass matrix and the convection matrix.
+        # It represents the Crank-Nicolson discretization of the PDE.
         A = M + (dt/2) * C 
+
+        # The right-hand side vector b for the linear system includes the effect of the previous time step's level set function and the convection.
         b = M @ phi0 - (dt/2) * C @ phi0
 
+        # Solve the linear system to find the updated level set function.
         phi0 = self.solve_system(A, b, tol = tol)
 
         return phi0
 
+
     def reinit(self, phi0, dt = 0.0001, eps = 5e-6, nt = 4, alpha = None):
         '''
-        PDE_reinit
+        Reinitialize the level set function to a signed distance function using the PDE-based reinitialization approach.
+
+        This function solves a reinitialization equation to stabilize the level set
+        function without moving the interface. The equation introduces artificial diffusion
+        to stabilize the solution process.
+
+        Parameters:
+        - phi0: The level set function to be reinitialized.
+        - dt: The timestep size for the pseudo-time evolution.
+        - eps: A small positive value to avoid division by zero and to smooth the transition across the interface.
+        - nt: The number of pseudo-time steps to perform.
+        - alpha: The artificial diffusion coefficient, which is auto-calculated based on the cell scale if not provided.
+
+        The reinitialization equation is discretized in time using a forward Euler method.
+        The weak form of the reinitialization equation is assembled and solved iteratively
+        until a stable state is reached.
+
+        Returns:
+        - phi1: The reinitialized level set function as a signed distance function.
         '''
         space = self.space
         mesh = space.mesh
 
+        # Calculate the maximum cell scale which is used to determine the artificial diffusion coefficient.
         cellscale = np.max(mesh.entity_measure('cell'))
         if alpha is None:
             alpha = 0.0625*cellscale
 
+        # Initialize the solution function and set its values to the initial level set function.
         phi1 = space.function()
         phi1[:] = phi0
         phi2 = space.function()
 
+        # Assemble the mass matrix M.
         M = self.M
 
+        # Assemble the stiffness matrix S using ScalarDiffusionIntegrator for artificial diffusion.
         bform = BilinearForm(space)
         bform.add_domain_integrator(ScalarDiffusionIntegrator())
         S = bform.assembly()
 
-        eold = 0   
+         # Initialize the old error to zero.
+        eold = 0
 
+        # Iterate for a fixed number of pseudo-time steps or until the error is below a threshold.
         for _ in range(nt):
 
+            # Define the function f for the right-hand side of the reinitialization equation.
             @barycentric
             def f(bcs, index):
                 grad = phi1.grad_value(bcs)
                 val = 1 - np.sqrt(np.sum(grad**2, -1))
                 val *= np.sign(phi0(bcs))
                 return val
 
+            # Assemble the linear form for the right-hand side of the equation.
             lform = LinearForm(space)
             lform.add_domain_integrator( ScalarSourceIntegrator(f = f) )
             b0 = lform.assembly()
+
+             # Compute the right-hand side vector for the linear system.
             b = M @ phi1 + dt * b0 - dt * alpha * (S @ phi1)
 
+            # Solve the linear system to update the level set function.
             phi2[:] = spsolve(M, b)
+
+            # Calculate the error between the new and old level set function.
             error = space.mesh.error(phi2, phi1)
-            print("重置:", error) 
+            print("Reinitialization error:", error) 
+
+            # If the error starts increasing or is below the threshold, break the loop.
             if eold < error and error< eps :
                 break
             else:

fealpy/levelset/ls_solver.py

@@ -16,18 +16,17 @@ def __call__(self, rk=None):
 
 
 class LSSolver():
-    def __init__(self, space, phi0, u, output_dir):
+    def __init__(self, space, phi0, u):
         self.space = space
         self.phi0 = phi0
         self.u = u
-        self.output_dir = output_dir
 
-    def output(self, timestep):
+    def output(self, timestep, output_dir, filename_prefix):
         mesh = self.space.mesh
-        if self.output_dir != 'None':
+        if output_dir != 'None':
             mesh.nodedata['phi'] = self.phi0
             mesh.nodedata['velocity'] = self.u
-            fname = os.path.join(self.output_dir, f'test_{timestep:010}.vtu')
+            fname = os.path.join(output_dir, f'{filename_prefix}_{timestep:010}.vtu')
             mesh.to_vtk(fname=fname)
 
     def check_gradient_norm(self, phi):
@@ -60,7 +59,7 @@ def check_gradient_norm(self, phi):
 
     def solve_system(self, A, b, tol=1e-8):
         counter = IterationCounter(disp = False)
-        result, info = lgmres(A, b, tol = tol, callback = counter)
+        result, info = lgmres(A, b, tol = tol, atol = tol, callback = counter)
         # print("Convergence info:", info)
         # print("Number of iteration of gmres:", counter.niter)
         return result

test/levelset/test_ls_fem_solver.py

@@ -0,0 +1,126 @@
+import pytest
+
+from fealpy.functionspace import LagrangeFESpace
+from fealpy.mesh.triangle_mesh import TriangleMesh
+from fealpy.decorator import cartesian
+from fealpy.levelset.ls_fem_solver import LSFEMSolver
+from fealpy.levelset.ls_fem_solver import LSSolver
+
+import numpy as np
+
+
+# A fixture for setting up the LSSolver environment with a predefined mesh and space.
+@pytest.fixture
+def ls_fem_solver_setup():
+    # Create a mesh in a unit square domain with specified number of divisions along x and y.
+    mesh = TriangleMesh.from_box([0, 1, 0, 1], nx=100, ny=100)
+    # Define a Lagrange finite element space of first order on the mesh.
+    space = LagrangeFESpace(mesh, p=1)
+
+    # Define a velocity field function for testing.
+    @cartesian
+    def velocity_field(p):
+        x = p[..., 0]
+        y = p[..., 1]
+        u = np.zeros(p.shape)
+        u[..., 0] = np.sin(np.pi * x) ** 2 * np.sin(2 * np.pi * y)
+        u[..., 1] = -np.sin(np.pi * y) ** 2 * np.sin(2 * np.pi * x)
+        return u
+
+    # Define a circle function representing the level set function for testing.
+    @cartesian
+    def circle(p):
+        x = p[..., 0]
+        y = p[..., 1]
+        return np.sqrt((x - 0.5) ** 2 + (y - 0.75) ** 2) - 0.15
+
+    # Interpolate the functions onto the finite element space.
+    phi0 = space.interpolate(circle)
+    u = space.interpolate(velocity_field, dim=2)
+
+    return space, u, phi0
+
+
+# Test the initialization of the LSFEMSolver class.
+def test_LSFEMSolver_init_without_u(ls_fem_solver_setup):
+    space, _, _ = ls_fem_solver_setup
+
+    solver_without_u = LSFEMSolver(space)
+
+    # Check that the mass matrix M is initialized.
+    assert solver_without_u.M is not None, "Mass matrix M should be initialized."
+
+    # Check that the space attribute matches the provided finite element space.
+    assert solver_without_u.space is space, "The space attribute should match the provided finite element space."
+
+    # Check that the velocity field u is None when not provided during initialization.
+    assert solver_without_u.u is None, "Velocity field u should be None when not provided during initialization."
+
+    # Check that the convection matrix C does not exist or is None when u is not provided.
+    assert not hasattr(solver_without_u, 'C') or solver_without_u.C is None, "Convection matrix C should not exist or be None when velocity u is not provided."
+
+def test_LSFEMSolver_init_with_u(ls_fem_solver_setup):
+    space, u, _ = ls_fem_solver_setup
+
+    # Instantiate the solver with the velocity field.
+    solver_with_u = LSFEMSolver(space, u=u)
+
+    # Check that the velocity field u is not None when provided during initialization.
+    assert solver_with_u.u is not None, "Velocity field u should not be None when provided during initialization."
+    # Check that the convection matrix C is initialized when u is provided.
+    assert solver_with_u.C is not None, "Convection matrix C should be initialized when velocity u is provided."
+
+
+def test_LSFEMSolver_solve(ls_fem_solver_setup):
+    space, u, phi0 = ls_fem_solver_setup
+
+    dt = 0.01  # A small time step
+
+    # Instantiate the solver with the velocity field.
+    solver_with_u = LSFEMSolver(space, u=u)
+
+    # Perform one step of the level set evolution.
+    phi1 = solver_with_u.solve(phi0, dt, u=u)
+
+    # Check if the result is a numpy array, which it should be after solving.
+    assert isinstance(phi1, np.ndarray), "The result of the solve method should be a numpy array."
+
+
+def test_LSFEMSolver_reinit(ls_fem_solver_setup):
+    space, u, _ = ls_fem_solver_setup
+
+    # 提供一个非符号距离函数 phi0
+    @cartesian
+    def non_sdf(p):
+        x = p[..., 0]
+        y = p[..., 1]
+        return (x - 0.5)**2 + (y - 0.5)**2  # 一个非符号距离函数的平方形式
+
+    phi0 = space.interpolate(non_sdf)
+    print("phi0:", phi0)
+
+    # Instantiate the solver with the velocity field.
+    solver_with_u = LSFEMSolver(space, u=u)
+
+    solver = LSSolver(space, phi0, u)
+
+    diff_avg, diff_max = solver.check_gradient_norm(phi = phi0)
+    print("diff_avg_0:", diff_avg)
+    print("diff_max_0:", diff_max)
+
+
+    # 执行重置化
+    phi1 = solver_with_u.reinit(phi0)
+    print("phi1:", phi1)
+
+    # Call the check_gradient_norm method which calculates the average and maximum difference from 1.
+    diff_avg, diff_max = solver.check_gradient_norm(phi = phi1)
+    print("diff_avg_1:", diff_avg)
+    print("diff_max_1:", diff_max)
+
+    # Assert that the average and maximum gradient norm differences should be close to 0.
+    # This means that the gradient norm is close to 1 as expected for a signed distance function.
+    assert np.isclose(diff_avg, 0, atol=1e-2), "The average difference should be close to 0."
+    assert np.isclose(diff_max, 0, atol=1e-2), "The maximum difference should be close to 0."
+
+