import unittest from SimPEG import * import simpegEM as EM from scipy.constants import mu_0 from simpegEM.Utils.Ana import hzAnalyticDipoleT import matplotlib.pyplot as plt class TDEM_bTests(unittest.TestCase): def setUp(self): cs = 5. ncx = 20 ncy = 6 npad = 20 hx = Utils.meshTensors(((0,cs), (ncx,cs), (npad,cs))) hy = Utils.meshTensors(((npad,cs), (ncy,cs), (npad,cs))) mesh = Mesh.Cyl1DMesh([hx,hy], -hy.sum()/2) model = Model.Vertical1DModel(mesh) opts = {'txLoc':0., 'txType':'VMD_MVP', 'rxLoc':np.r_[150., 0.], 'rxType':'bz', 'timeCh':np.logspace(-4,-2,20), } self.dat = EM.TDEM.DataTDEM1D(**opts) self.prb = EM.TDEM.ProblemTDEM_b(mesh, model) self.prb.setTimes([1e-5, 5e-5, 2.5e-4], [150, 150, 150]) self.sigma = np.ones(mesh.nCz)*1e-8 self.sigma[mesh.vectorCCz<0] = 0.1 self.prb.pair(self.dat) def test_analitic_b(self): bz_calc = self.dat.dpred(self.sigma) bz_ana = mu_0*hzAnalyticDipoleT(self.dat.rxLoc[0], self.prb.times, self.sigma[0]) diff = np.linalg.norm(bz_calc.flatten() - bz_ana.flatten())/np.linalg.norm(bz_ana.flatten()) self.assertTrue(diff<0.05) class TDEM_bDerivTests(unittest.TestCase): def setUp(self): cs = 5. ncx = 20 ncy = 6 npad = 20 hx = Utils.meshTensors(((0,cs), (ncx,cs), (npad,cs))) hy = Utils.meshTensors(((npad,cs), (ncy,cs), (npad,cs))) mesh = Mesh.Cyl1DMesh([hx,hy], -hy.sum()/2) model = Model.Vertical1DModel(mesh) opts = {'txLoc':0., 'txType':'VMD_MVP', 'rxLoc':np.r_[150., 0.], 'rxType':'bz', 'timeCh':np.logspace(-4,-2,20), } self.dat = EM.TDEM.DataTDEM1D(**opts) self.prb = EM.TDEM.ProblemTDEM_b(mesh, model) self.prb.setTimes([1e-5, 5e-5, 2.5e-4], [10, 10, 10]) self.sigma = np.ones(mesh.nCz)*1e-8 self.sigma[mesh.vectorCCz<0] = 0.1 self.prb.pair(self.dat) self.mesh = mesh def test_AhVec(self): """ Test that fields and AhVec produce consistent results """ sigma = np.ones(self.prb.mesh.nCz)*1e-8 sigma[self.prb.mesh.vectorCCz<0] = 0.1 u = self.prb.fields(sigma) Ahu = self.prb.AhVec(sigma, u) self.assertTrue(np.linalg.norm(Ahu.get_b(0)-1/self.prb.getDt(0)*u.get_b(-1))/np.linalg.norm(u.get_b(0)) < 1.e-2) self.assertTrue(np.linalg.norm(Ahu.get_e(0))/np.linalg.norm(u.get_e(0)) < 1.e-2) for i in range(1,u.nTimes): self.assertTrue(np.linalg.norm(Ahu.get_b(i))/np.linalg.norm(u.get_b(i)) < 1.e-2) self.assertTrue(np.linalg.norm(Ahu.get_e(i))/np.linalg.norm(u.get_e(i)) < 1.e-2) def test_AhVecVSMat_OneTS(self): self.prb.setTimes([1e-5], [1]) sigma = np.ones(self.prb.mesh.nCz)*1e-8 sigma[self.prb.mesh.vectorCCz<0] = 0.1 self.prb.makeMassMatrices(sigma) dt = self.prb.getDt(0) a11 = 1/dt*sp.eye(self.prb.mesh.nF) a12 = self.prb.mesh.edgeCurl a21 = self.prb.mesh.edgeCurl.T*self.prb.MfMui a22 = -self.prb.MeSigma A = sp.bmat([[a11,a12],[a21,a22]]) f = self.prb.fields(sigma) u1 = A*f.fieldVec() u2 = self.prb.AhVec(sigma,f).fieldVec() self.assertTrue(np.linalg.norm(u1-u2)<1e-12) def test_solveAhVSMat_OneTS(self): self.prb.setTimes([1e-5], [1]) sigma = np.ones(self.prb.mesh.nCz)*1e-8 sigma[self.prb.mesh.vectorCCz<0] = 0.1 self.prb.makeMassMatrices(sigma) dt = self.prb.getDt(0) a11 = 1/dt*sp.eye(self.prb.mesh.nF) a12 = self.prb.mesh.edgeCurl a21 = self.prb.mesh.edgeCurl.T*self.prb.MfMui a22 = -self.prb.MeSigma A = sp.bmat([[a11,a12],[a21,a22]]) f = self.prb.fields(sigma) f.set_b(np.zeros((self.prb.mesh.nF,1)),0) f.set_e(np.random.rand(self.prb.mesh.nE,1),0) u1 = self.prb.solveAh(sigma,f).fieldVec().flatten() u2 = sp.linalg.spsolve(A.tocsr(),f.fieldVec()) self.assertTrue(np.linalg.norm(u1-u2)<1e-8) def test_solveAhVsAhVec(self): prb = self.prb mesh = self.prb.mesh sigma = np.ones(self.prb.mesh.nCz)*1e-8 sigma[self.prb.mesh.vectorCCz<0] = 0.1 self.prb.makeMassMatrices(sigma) f = EM.TDEM.FieldsTDEM(prb.mesh, 1, prb.times.size, 'b') for i in range(f.nTimes): f.set_b(np.zeros((mesh.nF, 1)), i) f.set_e(np.random.rand(mesh.nE, 1), i) Ahf = prb.AhVec(sigma, f) f_test = prb.solveAh(sigma, Ahf) u1 = f.fieldVec() u2 = f_test.fieldVec() self.assertTrue(np.linalg.norm(u1-u2)<1e-8) def test_DerivG(self): """ Test the derivative of c with respect to sigma """ # Random model and perturbation sigma = np.random.rand(self.prb.mesh.nCz) f = self.prb.fields(sigma) dm = np.random.rand(self.prb.mesh.nCz) h = 1. a = np.linalg.norm(self.prb.AhVec(sigma+h*dm, f).fieldVec() - self.prb.AhVec(sigma, f).fieldVec()) b = np.linalg.norm(self.prb.AhVec(sigma+h*dm, f).fieldVec() - self.prb.AhVec(sigma, f).fieldVec() - h*self.prb.G(sigma, dm, u=f).fieldVec()) # Assuming that the gradient is exact to machine precision self.assertTrue(b<1e-16) def test_Deriv_dUdM(self): prb = self.prb prb.setTimes([1e-5, 1e-4, 1e-3], [10, 10, 10]) mesh = self.mesh sigma = self.sigma d_sig = sigma.copy() #np.random.rand(mesh.nCz) d_sig[d_sig==1e-8] = 0 num = 10 error = np.zeros(num) order = 0 hv = np.logspace(-1.2,-3, num) for i, h in enumerate(hv): f = prb.fields(sigma) fstep = prb.fields(sigma + h*d_sig) dcdm = prb.G(sigma, h*d_sig, u=f) # TODO: make negative!?!? dudm = prb.solveAh(sigma, dcdm) linear = np.linalg.norm(f.fieldVec() - fstep.fieldVec()) quad = np.linalg.norm(f.fieldVec() - fstep.fieldVec() - dudm.fieldVec()) error[i] = quad if i > 0: order = np.log(error[i]/error[i-1])/np.log(hv[i]/hv[i-1]) # print np.log(linearB/quadB)/np.log(h) print h, linear, quad, order self.assertTrue(order > 1.8) def test_Deriv_J(self): prb = self.prb prb.setTimes([1e-5, 1e-4, 1e-3], [10, 10, 10]) mesh = self.mesh sigma = self.sigma d_sig = 0.8*sigma #np.random.rand(mesh.nCz) d_sig[d_sig==1e-8] = 0 derChk = lambda m: [prb.data.dpred(m), lambda mx: -prb.J(sigma, mx)] passed = Tests.checkDerivative(derChk, sigma, plotIt=False, dx=d_sig, num=2, eps=1e-20) self.assertTrue(passed) if __name__ == '__main__': unittest.main()