This page was generated from unit-7-optimization/01_Shape_Derivative_Levelset.ipynb.
7.1 Shape optimization via the shape derivative¶
In this tutorial we discuss the implementation of shape optimization algorithms to solve
\[\min_{\Omega} J(\Omega) := \int_{\Omega} f(x)\;dx, \quad f\in C^1(\mathbf{R}^d).\]
The analytic solution to this problem is
\[\Omega^*:= \{x\in \mathbf{R}^d:\; f(x)\le 0\}.\]
This problem is solved by fixing an initial guess \(\Omega_0\subset \mathbf{R}^d\) and then considering the problem
\[\min_{X} J((\mbox{id}+X)(\Omega_0)),\]
where \(X:\mathbf{R}^d \to \mathbf{R}^d\) are (at least) Lipschitz vector fields. We approximate \(X\) by a finite element function.
Initial geometry and shape function¶
We choose \(f\) as
\[\begin{split}\begin{array}{rl}
f(x,y) =& \left(\sqrt{(x - a)^2 + b y^2} - 1 \right) \left(\sqrt{(x + a)^2 + b y^2} - 1 \right) \\
& \left(\sqrt{b x^2 + (y - a)^2} - 1 \right) \left(\sqrt{b x^2 + (y + a)^2} - 1 \right) - \varepsilon.
\end{array}\end{split}\]
where \(\varepsilon = 0.001, a = 4/5, b = 2\). The corresponding zero level sets of this function are as follows. The green area indicates where \(f\) is negative.
Let us set up the geometry.
[1]:
# Generate the geometry
from ngsolve import *
# load Netgen/NGSolve and start the gui
from ngsolve.webgui import Draw
from netgen.geom2d import SplineGeometry
geo = SplineGeometry()
geo.AddCircle((0,0), r = 2.5)
ngmesh = geo.GenerateMesh(maxh = 0.08)
mesh = Mesh(ngmesh)
mesh.Curve(2)
[1]:
<ngsolve.comp.Mesh at 0x7f8c63839e00>
[2]:
#gfset.vec[:]=0
#gfset.Set((0.2*x,0))
#mesh.SetDeformation(gfset)
#scene = Draw(gfzero,mesh,"gfset", deformation = gfset)
[3]:
# define the function f and its gradient
a =4.0/5.0
b = 2
f = CoefficientFunction((sqrt((x - a)**2 + b * y**2) - 1) \
* (sqrt((x + a)**2 + b * y**2) - 1) \
* (sqrt(b * x**2 + (y - a)**2) - 1) \
* (sqrt(b * x**2 + (y + a)**2) - 1) - 0.001)
# gradient of f defined as vector valued coefficient function
grad_f = CoefficientFunction((f.Diff(x),f.Diff(y)))
[4]:
# vector space for shape gradient
VEC = H1(mesh, order=1, dim=2)
# grid function for deformation field
gfset = GridFunction(VEC)
gfX = GridFunction(VEC)
Draw(gfset)
[4]:
BaseWebGuiScene
Shape derivative¶
\[DJ(\Omega)(X) = \int_\Omega f \mbox{div}(X) + \nabla f\cdot X\;dx\]
[5]:
# Test and trial functions
PHI, PSI = VEC.TnT()
# shape derivative
dJOmega = LinearForm(VEC)
dJOmega += (div(PSI)*f + InnerProduct(grad_f, PSI) )*dx
Bilinear form¶
\[(\varphi,\psi) \mapsto b_\Omega(\varphi,\psi):= \int_\Omega (\nabla \varphi+\nabla \varphi^\top): \nabla\psi+\varphi\cdot \psi\; dx.\]
to compute the gradient \(X:= \mbox{grad}J(\Omega)\) by
\[b_\Omega(X,\psi) = DJ(\Omega)(\psi)\quad \text{ for all } \quad \psi \in H^1(\Omega)\]
[6]:
# bilinear form for H1 shape gradient; aX represents b_\Omega
aX = BilinearForm(VEC)
aX += InnerProduct(grad(PHI) + grad(PHI).trans, grad(PSI)) * dx
aX += InnerProduct(PHI, PSI) * dx
The first optimisation step¶
Fix an initial domain \(\Omega_0\) and define
\[\mathcal J_{\Omega_0}(X):= J((\mbox{id} + X)(\Omega_0)).\]
Then we have the following relation between the derivative of \(\mathcal J_{\Omega_0}\) and the shape derivative of \(J\):
\[D\mathcal J_{\Omega_0}(X_n)(X) = DJ((\mbox{id}+X_n)(\Omega_0))(X\circ(\mbox{id}+X_n)^{-1}).\]
Here
\((\mbox{id}+X_n)(\Omega_0)\) is current shape
Now we note that \(\varphi \mapsto \varphi\circ(\mbox{id}+X_n)^{-1}\) maps the finite element space on \((\mbox{id}+X_n)(\Omega_0)\) to the finite element space on \(\Omega_0\). Therefore the following are equvalent:
assembling \(\varphi \mapsto D\mathcal J_{\Omega_0}(X_n)(\varphi)\) on the fixed domain \(\Omega_0\)
assembling \(\varphi \mapsto DJ((\mbox{id}+X_n)(\Omega_0))(\varphi)\) on the deformed domain \((\mbox{id}+X_n)(\Omega_0)\).
[7]:
# deform current domain with gfset
mesh.SetDeformation(gfset)
# assemble on deformed domain
aX.Assemble()
dJOmega.Assemble()
mesh.UnsetDeformation()
# unset deformation
Now let’s make one optimization step with step size \(\alpha_1>0\):
\[\Omega_1 = (\mbox{id} - \alpha_1 X_0)(\Omega_0).\]
[8]:
# compute X_0
gfX.vec.data = aX.mat.Inverse(VEC.FreeDofs(), inverse="sparsecholesky") * dJOmega.vec
print("current cost ", Integrate(f*dx, mesh))
print("Gradient norm ", Norm(gfX.vec))
alpha = 20.0 / Norm(gfX.vec)
gfset.vec[:]=0
scene = Draw(gfset)
# input("A")
gfset.vec.data -= alpha * gfX.vec
mesh.SetDeformation(gfset)
#draw deformed shape
scene.Redraw()
# input("B")
print("cost after gradient step:", Integrate(f, mesh))
mesh.UnsetDeformation()
scene.Redraw()
current cost 63.58544060305382
Gradient norm 442.05184362948427
cost after gradient step: 5.043649318663175
Optimisation loop¶
Now we can set up an optimisation loop. In the second step we compute
\[\Omega_2 = (\mbox{id} - \alpha_0 X_0 - \alpha_1 X_1)(\Omega_0)\]
by the same procedure as above.
[9]:
import time
iter_max = 50
gfset.vec[:] = 0
mesh.SetDeformation(gfset)
scene = Draw(gfset,mesh,"gfset")
converged = False
alpha =0.11#100.0 / Norm(gfX.vec)
# input("A")
for k in range(iter_max):
mesh.SetDeformation(gfset)
scene.Redraw()
Jold = Integrate(f, mesh)
print("cost at iteration ", k, ': ', Jold)
# assemble bilinear form
aX.Assemble()
# assemble shape derivative
dJOmega.Assemble()
mesh.UnsetDeformation()
gfX.vec.data = aX.mat.Inverse(VEC.FreeDofs(), inverse="sparsecholesky") * dJOmega.vec
# step size control
gfset_old = gfset.vec.CreateVector()
gfset_old.data = gfset.vec
Jnew = Jold + 1
while Jnew > Jold:
gfset.vec.data = gfset_old
gfset.vec.data -= alpha * gfX.vec
mesh.SetDeformation(gfset)
Jnew = Integrate(f, mesh)
mesh.UnsetDeformation()
if Jnew > Jold:
print("reducing step size")
alpha = 0.9*alpha
else:
print("linesearch step accepted")
alpha = alpha*1.5
break
print("step size: ", alpha, '\n')
time.sleep(0.1)
Jold = Jnew
cost at iteration 0 : 63.58544060305382
linesearch step accepted
step size: 0.165
cost at iteration 1 : -0.7382553000138545
linesearch step accepted
step size: 0.2475
cost at iteration 2 : -0.8176876285984291
linesearch step accepted
step size: 0.37124999999999997
cost at iteration 3 : -0.9325785868476248
linesearch step accepted
step size: 0.556875
cost at iteration 4 : -1.0701613164084456
linesearch step accepted
step size: 0.8353125
cost at iteration 5 : -1.1521934750329594
linesearch step accepted
step size: 1.25296875
cost at iteration 6 : -1.1547483082895464
reducing step size
reducing step size
linesearch step accepted
step size: 1.52235703125
cost at iteration 7 : -1.154955094622594
reducing step size
reducing step size
reducing step size
reducing step size
linesearch step accepted
step size: 1.4982276723046877
cost at iteration 8 : -1.1550472206937452
reducing step size
reducing step size
reducing step size
reducing step size
reducing step size
linesearch step accepted
step size: 1.3270326873287925
cost at iteration 9 : -1.155453494676608
reducing step size
reducing step size
reducing step size
linesearch step accepted
step size: 1.4511102435940346
cost at iteration 10 : -1.1555977964371116
reducing step size
reducing step size
reducing step size
reducing step size
linesearch step accepted
step size: 1.4281101462330694
cost at iteration 11 : -1.1556257735480633
reducing step size
reducing step size
reducing step size
reducing step size
linesearch step accepted
step size: 1.4054746004152756
cost at iteration 12 : -1.1558550657345563
reducing step size
reducing step size
reducing step size
reducing step size
linesearch step accepted
step size: 1.3831978279986936
cost at iteration 13 : -1.1559610148052597
reducing step size
reducing step size
reducing step size
reducing step size
linesearch step accepted
step size: 1.3612741424249144
cost at iteration 14 : -1.1561635400235297
reducing step size
reducing step size
reducing step size
reducing step size
linesearch step accepted
step size: 1.3396979472674797
cost at iteration 15 : -1.156193676380524
reducing step size
reducing step size
reducing step size
reducing step size
reducing step size
linesearch step accepted
step size: 1.1866173613229614
cost at iteration 16 : -1.156357024476493
reducing step size
reducing step size
reducing step size
reducing step size
reducing step size
linesearch step accepted
step size: 1.0510285285313932
cost at iteration 17 : -1.156434437642301
reducing step size
reducing step size
reducing step size
reducing step size
linesearch step accepted
step size: 1.034369726354171
cost at iteration 18 : -1.1564637024518876
reducing step size
reducing step size
reducing step size
reducing step size
linesearch step accepted
step size: 1.0179749661914572
cost at iteration 19 : -1.1564751304725112
reducing step size
reducing step size
reducing step size
reducing step size
linesearch step accepted
step size: 1.0018400629773228
cost at iteration 20 : -1.156511537322834
reducing step size
reducing step size
reducing step size
reducing step size
linesearch step accepted
step size: 0.9859608979791321
cost at iteration 21 : -1.156564232383649
reducing step size
reducing step size
reducing step size
reducing step size
linesearch step accepted
step size: 0.9703334177461629
cost at iteration 22 : -1.1566297010931148
reducing step size
reducing step size
reducing step size
reducing step size
linesearch step accepted
step size: 0.9549536330748863
cost at iteration 23 : -1.1566995703949992
reducing step size
reducing step size
reducing step size
reducing step size
linesearch step accepted
step size: 0.9398176179906492
cost at iteration 24 : -1.156765520852471
reducing step size
reducing step size
reducing step size
linesearch step accepted
step size: 1.027690565272775
cost at iteration 25 : -1.1567676713081918
reducing step size
reducing step size
reducing step size
reducing step size
linesearch step accepted
step size: 1.0114016698132016
cost at iteration 26 : -1.1567784490425266
reducing step size
reducing step size
reducing step size
reducing step size
linesearch step accepted
step size: 0.9953709533466624
cost at iteration 27 : -1.156798428862847
reducing step size
reducing step size
reducing step size
reducing step size
linesearch step accepted
step size: 0.9795943237361178
cost at iteration 28 : -1.1568250029634704
reducing step size
reducing step size
reducing step size
reducing step size
linesearch step accepted
step size: 0.9640677537049005
cost at iteration 29 : -1.1568556930569593
reducing step size
reducing step size
reducing step size
reducing step size
linesearch step accepted
step size: 0.9487872798086778
cost at iteration 30 : -1.1568865442400496
reducing step size
reducing step size
reducing step size
reducing step size
linesearch step accepted
step size: 0.9337490014237104
cost at iteration 31 : -1.1569150632990812
reducing step size
reducing step size
reducing step size
linesearch step accepted
step size: 1.0210545330568275
cost at iteration 32 : -1.1569198549179136
reducing step size
reducing step size
reducing step size
reducing step size
linesearch step accepted
step size: 1.0048708187078768
cost at iteration 33 : -1.15692778724364
reducing step size
reducing step size
reducing step size
reducing step size
linesearch step accepted
step size: 0.988943616231357
cost at iteration 34 : -1.156938362321561
reducing step size
reducing step size
reducing step size
reducing step size
linesearch step accepted
step size: 0.97326885991409
cost at iteration 35 : -1.1569509591157165
reducing step size
reducing step size
reducing step size
reducing step size
linesearch step accepted
step size: 0.9578425484844518
cost at iteration 36 : -1.1569642145231311
reducing step size
reducing step size
reducing step size
linesearch step accepted
step size: 1.047400826767748
cost at iteration 37 : -1.1569647976305986
reducing step size
reducing step size
reducing step size
reducing step size
linesearch step accepted
step size: 1.0307995236634793
cost at iteration 38 : -1.1569668723577835
reducing step size
reducing step size
reducing step size
reducing step size
linesearch step accepted
step size: 1.0144613512134133
cost at iteration 39 : -1.156971132204393
reducing step size
reducing step size
reducing step size
reducing step size
linesearch step accepted
step size: 0.9983821387966807
cost at iteration 40 : -1.156977460360057
reducing step size
reducing step size
reducing step size
reducing step size
linesearch step accepted
step size: 0.9825577818967535
cost at iteration 41 : -1.1569856919266508
reducing step size
reducing step size
reducing step size
reducing step size
linesearch step accepted
step size: 0.96698424105369
cost at iteration 42 : -1.1569949177533807
reducing step size
reducing step size
reducing step size
reducing step size
linesearch step accepted
step size: 0.9516575408329891
cost at iteration 43 : -1.1570044550911744
reducing step size
reducing step size
reducing step size
linesearch step accepted
step size: 1.0406375209008738
cost at iteration 44 : -1.1570056477226338
reducing step size
reducing step size
reducing step size
reducing step size
linesearch step accepted
step size: 1.0241434161945948
cost at iteration 45 : -1.1570079849271222
reducing step size
reducing step size
reducing step size
reducing step size
linesearch step accepted
step size: 1.0079107430479106
cost at iteration 46 : -1.1570115458839176
reducing step size
reducing step size
reducing step size
reducing step size
linesearch step accepted
step size: 0.9919353577706012
cost at iteration 47 : -1.1570164247726291
reducing step size
reducing step size
reducing step size
reducing step size
linesearch step accepted
step size: 0.9762131823499371
cost at iteration 48 : -1.1570221572034933
reducing step size
reducing step size
reducing step size
reducing step size
linesearch step accepted
step size: 0.9607402034096908
cost at iteration 49 : -1.1570284025356303
reducing step size
reducing step size
reducing step size
linesearch step accepted
step size: 1.050569412428497
Using SciPy optimize toolbox¶
We use the package scipy.optimize and wrap the shape functions around it.
We first setup the initial geometry as a circle of radius r = 2.5 (as before)
[10]:
# The code in this cell is the same as in the example above.
from ngsolve import *
from ngsolve.webgui import Draw
from netgen.geom2d import SplineGeometry
import numpy as np
ngsglobals.msg_level = 1
geo = SplineGeometry()
geo.AddCircle((0,0), r = 2.5)
ngmesh = geo.GenerateMesh(maxh = 0.08)
mesh = Mesh(ngmesh)
mesh.Curve(2)
# define the function f
a =4.0/5.0
b = 2
f = CoefficientFunction((sqrt((x - a)**2 + b * y**2) - 1) \
* (sqrt((x + a)**2 + b * y**2) - 1) \
* (sqrt(b * x**2 + (y - a)**2) - 1) \
* (sqrt(b * x**2 + (y + a)**2) - 1) - 0.001)
# Now we define the finite element space VEC in which we compute the shape gradient
# element order
order = 1
VEC = H1(mesh, order=order, dim=2)
# define test and trial functions
PHI = VEC.TrialFunction()
PSI = VEC.TestFunction()
# define grid function for deformation of mesh
gfset = GridFunction(VEC)
gfset.Set((0,0))
# only for new gui
#scene = Draw(gfset, mesh, "gfset")
#if scene:
# scene.setDeformation(True)
# plot the mesh and visualise deformation
#Draw(gfset,mesh,"gfset")
#SetVisualization (deformation=True)
Generate Mesh from spline geometry
Curve elements, order = 2
Shape derivative¶
\[DJ(\Omega)(X) = \int_\Omega f \mbox{div}(X) + \nabla f\cdot X\;dx\]
[11]:
fX = LinearForm(VEC)
# analytic shape derivative
fX += f*div(PSI)*dx + (f.Diff(x,PSI[0])) *dx + (f.Diff(y,PSI[1])) *dx
Bilinear form for shape gradient¶
Define the bilinear form
\[(\varphi,\psi) \mapsto b_\Omega(\varphi,\psi):= \int_\Omega (\nabla \varphi+\nabla \varphi^\top): \nabla\psi + \varphi\cdot\psi\; dx\]
to compute the gradient \(X =: \mbox{grad}J\) by
\[b_\Omega(X,\psi) = DJ(\Omega)(\psi)\quad \text{ for all } \psi\in H^1(\Omega)\]
[12]:
# Cauchy-Riemann descent CR
CR = False
# bilinear form for gradient
aX = BilinearForm(VEC)
aX += InnerProduct(grad(PHI)+grad(PHI).trans, grad(PSI))*dx + InnerProduct(PHI,PSI)*dx
## Cauchy-Riemann regularisation
if CR == True:
gamma = 100
aX += gamma * (PHI.Deriv()[0,0] - PHI.Deriv()[1,1])*(PSI.Deriv()[0,0] - PSI.Deriv()[1,1]) *dx
aX += gamma * (PHI.Deriv()[1,0] + PHI.Deriv()[0,1])*(PSI.Deriv()[1,0] + PSI.Deriv()[0,1]) *dx
aX.Assemble()
invaX = aX.mat.Inverse(VEC.FreeDofs(), inverse="sparsecholesky")
gfX = GridFunction(VEC)
Wrapping the shape function and its gradient¶
Now we define the shape function \(J\) and its gradient grad(J) use the shape derivative \(DJ\). The arguments of \(J\) and grad(J) are vector in \(\mathbf{R}^d\), where \(d\) is the dimension of the finite element space.
[13]:
def J(x_):
gfset.Set((0,0))
# x_ NumPy vector
gfset.vec.FV().NumPy()[:] += x_
mesh.SetDeformation(gfset)
cost_c = Integrate (f, mesh)
mesh.UnsetDeformation()
Redraw(blocking=True)
return cost_c
def gradJ(x_, euclid = False):
gfset.Set((0,0))
# x_ NumPy vector
gfset.vec.FV().NumPy()[:] += x_
mesh.SetDeformation(gfset)
fX.Assemble()
mesh.UnsetDeformation()
if euclid == True:
gfX.vec.data = fX.vec
else:
gfX.vec.data = invaX * fX.vec
return gfX.vec.FV().NumPy().copy()
Gradient descent and Armijo rule¶
We use the functions \(J\) and grad J to define a steepest descent method.We use scipy.optimize.line_search_armijo to compute the step size in each iteration. The Arjmijo rule reads
\[J((\mbox{Id}+\alpha_k X_k)(\Omega_k)) \le J(\Omega_k) - c_1 \alpha_k \|\nabla J(\Omega_k)\|^2,\]
where \(c_1:= 1e-4\) and \(\alpha_k\) is the step size.
[14]:
# import scipy linesearch method
from scipy.optimize.linesearch import line_search_armijo
def gradient_descent(x0, J_, gradJ_):
xk_ = np.copy(x0)
# maximal iteration
it_max = 50
# count number of function evals
nfval_total = 0
print('\n')
for i in range(1, it_max):
# Compute a step size using a line_search to satisfy the Wolf
# compute shape gradient grad J
grad_xk = gradJ_(xk_,euclid = False)
# compute descent direction
pk = -gradJ_(xk_)
# eval cost function
fold = J_(xk_)
# perform armijo stepsize
if CR == True:
alpha0 = 0.15
else:
alpha0 = 0.11
step, nfval, b = line_search_armijo(J_, xk_, pk = pk, gfk = grad_xk, old_fval = fold, c1=1e-4, alpha0 = alpha0)
nfval_total += nfval
# update the shape and print cost and gradient norm
xk_ = xk_ - step * grad_xk
print('Iteration ', i, '| Cost ', fold, '| grad norm', np.linalg.norm(grad_xk))
mesh.SetDeformation(gfset)
scene.Redraw()
mesh.UnsetDeformation()
if np.linalg.norm(gradJ_(xk_)) < 1e-4:
#print('#######################################')
print('\n'+'{:<20} {:<12} '.format("##################################", ""))
print('{:<20} {:<12} '.format("### success - accuracy reached ###", ""))
print('{:<20} {:<12} '.format("##################################", ""))
print('{:<20} {:<12} '.format("gradient norm: ", np.linalg.norm(gradJ_(xk_))))
print('{:<20} {:<12} '.format("n evals f: ", nfval_total))
print('{:<20} {:<12} '.format("f val: ", fold) + '\n')
break
elif i == it_max-1:
#print('######################################')
print('\n'+'{:<20} {:<12} '.format("#######################", ""))
print('{:<20} {:<12} '.format("### maxiter reached ###", ""))
print('{:<20} {:<12} '.format("#######################", ""))
print('{:<20} {:<12} '.format("gradient norm: ", np.linalg.norm(gradJ_(xk_))))
print('{:<20} {:<12} '.format("n evals f: ", nfval_total))
print('{:<20} {:<12} '.format("f val: ", fold) + '\n')
Now we are ready to run our first optimisation algorithm:¶
[15]:
gfset.vec[:]=0
x0 = gfset.vec.FV().NumPy() # initial guess = 0
scene = Draw(gfset, mesh, "gfset")
gradient_descent(x0, J, gradJ)
Iteration 1 | Cost 63.58544060305382 | grad norm 442.05184362948427
Iteration 2 | Cost -0.7382553000138548 | grad norm 4.797321601018455
Iteration 3 | Cost -0.7857735477702836 | grad norm 4.812294575740791
Iteration 4 | Cost -0.8340109759042181 | grad norm 4.780144423275227
Iteration 5 | Cost -0.881983893823349 | grad norm 4.692239115470677
Iteration 6 | Cost -0.9285224314563473 | grad norm 4.5416445721269705
Iteration 7 | Cost -0.9723566047005454 | grad norm 4.324713174006836
Iteration 8 | Cost -1.0122540076018296 | grad norm 4.0426845384291346
Iteration 9 | Cost -1.0471892473177822 | grad norm 3.7026495061938505
Iteration 10 | Cost -1.0765050767054933 | grad norm 3.3181634341605646
Iteration 11 | Cost -1.1000167253818887 | grad norm 2.90732492212311
Iteration 12 | Cost -1.1180154448826773 | grad norm 2.490754022058955
Iteration 13 | Cost -1.1311721937086696 | grad norm 2.0883910113681714
Iteration 14 | Cost -1.140374856628482 | grad norm 1.716495522576629
Iteration 15 | Cost -1.1465562052233915 | grad norm 1.3862619992896077
Iteration 16 | Cost -1.150562263853099 | grad norm 1.1029047823734224
Iteration 17 | Cost -1.1530807854465164 | grad norm 0.8667776120942267
Iteration 18 | Cost -1.1546254652064152 | grad norm 0.6747350037169477
Iteration 19 | Cost -1.155555089885781 | grad norm 0.521601956642585
Iteration 20 | Cost -1.15610728312894 | grad norm 0.4014078557589178
Iteration 21 | Cost -1.1564329962973245 | grad norm 0.3082390241864366
Iteration 22 | Cost -1.1566250721363829 | grad norm 0.236729388003623
Iteration 23 | Cost -1.1567392179253135 | grad norm 0.18227782886365923
Iteration 24 | Cost -1.1568082424509445 | grad norm 0.14108836193150684
Iteration 25 | Cost -1.1568512059762208 | grad norm 0.11011137545386489
Iteration 26 | Cost -1.1568790758609946 | grad norm 0.08694123443763045
Iteration 27 | Cost -1.1568981259972182 | grad norm 0.06970013095634099
Iteration 28 | Cost -1.1569119356095627 | grad norm 0.05693106530543911
Iteration 29 | Cost -1.1569225483311438 | grad norm 0.04750596065756907
Iteration 30 | Cost -1.1569311374304097 | grad norm 0.040553631098362235
Iteration 31 | Cost -1.156938383007384 | grad norm 0.03540610663423864
Iteration 32 | Cost -1.1569446865801067 | grad norm 0.03155866160652364
Iteration 33 | Cost -1.1569502911220542 | grad norm 0.028637951437263578
Iteration 34 | Cost -1.156955349068707 | grad norm 0.026374404372481768
Iteration 35 | Cost -1.1569599604966179 | grad norm 0.024577686706996647
Iteration 36 | Cost -1.156964194534456 | grad norm 0.023115688255964162
Iteration 37 | Cost -1.156968101815988 | grad norm 0.021897623641615804
Iteration 38 | Cost -1.1569717214195943 | grad norm 0.020861264507975198
Iteration 39 | Cost -1.1569750846830889 | grad norm 0.019963767959634886
Iteration 40 | Cost -1.156978217692618 | grad norm 0.019175836787548774
Iteration 41 | Cost -1.1569811421514762 | grad norm 0.01847543606004201
Iteration 42 | Cost -1.1569838777622021 | grad norm 0.017849514376677995
Iteration 43 | Cost -1.1569864401815315 | grad norm 0.017283661655752045
Iteration 44 | Cost -1.1569888459059847 | grad norm 0.016770895179239378
Iteration 45 | Cost -1.156991108604262 | grad norm 0.016304608148209974
Iteration 46 | Cost -1.1569932403960685 | grad norm 0.01587952159300805
Iteration 47 | Cost -1.1569952522869715 | grad norm 0.015492096337398344
Iteration 48 | Cost -1.1569971536572254 | grad norm 0.015137178273958808
Iteration 49 | Cost -1.1569989538206762 | grad norm 0.014812385966536876
#######################
### maxiter reached ###
#######################
gradient norm: 0.014515001787420401
n evals f: 49
f val: -1.1569989538206762
L-BFGS method¶
Now we use the L-BFGS method provided by SciPy. The BFGS method requires the shape function \(J\) and its gradient grad J. We can also specify additional arguments in options, such as maximal iterations and the gradient tolerance.
In the BFGS method we replace
\[b_\Omega(X,\varphi) = \nabla J(\Omega)(\varphi)\]
by
\[H_\Omega(X,\varphi) = \nabla J(\Omega)(\varphi),\]
where \(H_\Omega\) is an approximation of the second shape derivative at \(\Omega\). On the discrete level we solve
\[\begin{split}\begin{array}{rl}
\text{ solve }\quad B_nX_k & = \nabla J(\Omega_k) \\
& \\
\text{ update } \quad s_k & := -\alpha_k p_k \\
y_k & := \nabla J_h(\Omega_{k+1}) - \nabla J_h(\Omega_k) \\
B_{k+1} &:= B_k + \frac{y_k\otimes y_k}{y_k\cdot s_k} + \frac{B_ks_k\otimes B_ks_k}{B_ky_k\cdot B_ks_k}.
\end{array}\end{split}\]
[16]:
from scipy.optimize import minimize
x0 = gfset.vec.FV().NumPy()
# options for optimiser
options = {"maxiter":1000,
"disp":True,
"gtol":1e-10}
# we use quasi-Newton method L-BFGS
minimize(J, x0, method='L-BFGS-B', jac=gradJ, options=options)
RUNNING THE L-BFGS-B CODE
* * *
Machine precision = 2.220D-16
N = 7784 M = 10
This problem is unconstrained.
At X0 0 variables are exactly at the bounds
At iterate 0 f= -1.15700D+00 |proj g|= 9.63930D-04
At iterate 1 f= -1.15700D+00 |proj g|= 1.78236D-03
At iterate 2 f= -1.15703D+00 |proj g|= 1.16772D-03
At iterate 3 f= -1.15706D+00 |proj g|= 3.35736D-03
ys=-2.003E-04 -gs= 2.740E-03 BFGS update SKIPPED
At iterate 4 f= -1.15707D+00 |proj g|= 2.24781D-03
At iterate 5 f= -1.15709D+00 |proj g|= 9.32024D-04
At iterate 6 f= -1.15709D+00 |proj g|= 1.21605D-03
At iterate 7 f= -1.15709D+00 |proj g|= 1.26745D-03
At iterate 8 f= -1.15709D+00 |proj g|= 1.29862D-03
At iterate 9 f= -1.15709D+00 |proj g|= 1.33023D-03
Bad direction in the line search;
refresh the lbfgs memory and restart the iteration.
At iterate 10 f= -1.15710D+00 |proj g|= 1.52938D-03
At iterate 11 f= -1.15710D+00 |proj g|= 4.54617D-04
At iterate 12 f= -1.15710D+00 |proj g|= 4.52943D-04
At iterate 13 f= -1.15711D+00 |proj g|= 7.05745D-04
At iterate 14 f= -1.15711D+00 |proj g|= 7.05745D-04
[16]:
fun: -1.1571110014730501
hess_inv: <7784x7784 LbfgsInvHessProduct with dtype=float64>
jac: array([ 2.14530782e-06, -6.07237181e-04, 6.13396109e-04, ...,
-5.29467877e-06, 6.19934625e-06, -5.11621928e-06])
message: b'CONVERGENCE: REL_REDUCTION_OF_F_<=_FACTR*EPSMCH'
nfev: 108
nit: 14
status: 0
success: True
x: array([-5.01427824e-04, 6.98499222e-01, -6.98449755e-01, ...,
4.66153980e-03, -1.54412643e-03, 4.61902413e-03])
* * *
Tit = total number of iterations
Tnf = total number of function evaluations
Tnint = total number of segments explored during Cauchy searches
Skip = number of BFGS updates skipped
Nact = number of active bounds at final generalized Cauchy point
Projg = norm of the final projected gradient
F = final function value
* * *
N Tit Tnf Tnint Skip Nact Projg F
7784 14 108 2 1 0 7.057D-04 -1.157D+00
F = -1.1571110014730501
CONVERGENCE: REL_REDUCTION_OF_F_<=_FACTR*EPSMCH
Warning: more than 10 function and gradient
evaluations in the last line search. Termination
may possibly be caused by a bad search direction.
Cauchy time 3.357E-04 seconds.
Subspace minimization time 5.630E-03 seconds.
Line search time 2.599E+01 seconds.
Total User time 2.625E+01 seconds.
Improving mesh quality via Cauchy-Riemann equations¶
In the previous section we computed the shape gradient grad J:= X of \(J\) at \(\Omega\) via
\[\int_{\Omega} \nabla X : \nabla \varphi + X\cdot \varphi\; dx = DJ(\Omega)(\varphi) \quad \forall \varphi \in H^1(\Omega)^d.\]
This may lead to overly streched or even degenerated triangles. One way to improve this is to modify the above equation by
\[\int_{\Omega} \nabla X : \nabla \varphi + X\cdot \varphi + \color{red}\gamma \color{red}B\color{red}X\cdot \color{red}B\color{red}\varphi\;dx = DJ(\Omega)(\varphi) \; \forall \varphi \in H^1(\Omega)^d,\]
where
\[\begin{split}B := \begin{pmatrix}
-\partial_x & \partial_y \\
\partial_y & \partial_x
\end{pmatrix}\end{split}\]
The two equations \(BX = 0\) are precisely the Cauchy-Riemann equations \(-\partial_x X_1 + \partial_y X_2=0\) and \(\partial_y X_1 + \partial_x X_2\). So the bigger \(\gamma\) the more angle preserving is the gradient. So by adding the B term we enforce conformaty with strengh \(\gamma\).
This only means we need to add the \(\alpha\) term to the above bilinear form aX:
[17]:
alpha = 100
aX += alpha * (PHI.Deriv()[0,0] - PHI.Deriv()[1,1])*(PSI.Deriv()[0,0] - PSI.Deriv()[1,1]) *dx
aX += alpha * (PHI.Deriv()[1,0] + PHI.Deriv()[0,1])*(PSI.Deriv()[1,0] + PSI.Deriv()[0,1]) *dx
[ ]: