def test_add():
"""Test Multivariate Chebyshev polynomial addition."""
t = np.arange(27).reshape((3, 3, 3))
poly1 = MultiCheb(t)
poly2 = MultiCheb(np.ones((3, 3, 3)))
S = poly1 + poly2 # the sum of the polynomials
result = (S.coeff == (poly1.coeff + poly2.coeff))
assert result.all()
def test_evaluate_grid1():
poly = MultiCheb(np.array([[2, 0, 3], [0, -1, 0], [0, 1, 0]]))
x = np.arange(3)
xy = np.column_stack([x, x])
sol = np.polynomial.chebyshev.chebgrid2d(x, x, poly.coeff)
assert (np.all(poly.evaluate_grid(xy) == sol))
def getPoly(deg, power):
'''
A helper function for testing. Returns a random 1D polynomial of the given degree.
power is a boolean indicating whether or not the polynomial should be MultiPower.
'''
coeff = np.random.random_sample(deg + 1)
if power:
return MultiPower(coeff)
else:
return MultiCheb(coeff)
def test_evaluate():
cheb = MultiCheb(np.array([[0, 0, 0, 1], [0, 0, 0, 0], [0, 0, 1, 0]]))
value = cheb((2, 5))
assert (value == 828)
value = cheb((.25, .5))
assert (np.isclose(value, -.5625))
values = cheb([[.25, .5], [1.2, 2.2]])
print(values)
assert (np.allclose(values, [-0.5625, 52.3104]))
def chebpolyqeps(Q, eps):
dim = Q.shape[0]
polys = []
for i in range(dim):
coeff = np.zeros([3] * dim)
spot = [0] * dim
coeff[tuple(spot)] = .5
for j in range(dim):
spot[j] = 1
coeff[tuple(spot)] = Q[i, j] * eps
spot[j] = 0
spot[i] = 2
coeff[tuple(spot)] = .5
polys.append(MultiCheb(coeff))
return polys
def chebapprox(polys, a, b, deg, atol=1e-15, rtol=1e-15, ttol=1e-15):
chebcoeff = []
inf_norms = []
errors = []
for poly in polys:
coeff, _, inf_norm, error = full_cheb_approximate(
poly, a, b, deg, atol, rtol)
chebcoeff.append(coeff)
inf_norms.append(inf_norm)
errors.append(error)
chebcoeff = trim_coeffs(chebcoeff, atol, rtol, ttol, inf_norms, errors)[0]
chebpolys = []
for coeff in chebcoeff:
chebpolys.append(MultiCheb(coeff))
return chebpolys
def test_mon_mult():
"""
Tests monomial multiplication using normal polynomial multiplication.
"""
np.random.seed(4)
#Simple 2D test cases
cheb1 = MultiCheb(np.array([[0, 0, 0], [0, 0, 0], [0, 0, 1]]))
mon1 = (1, 1)
result1 = cheb1.mon_mult(mon1)
truth1 = np.array([[0, 0, 0, 0], [0, 0.25, 0, 0.25], [0, 0, 0, 0],
[0, 0.25, 0, 0.25]])
assert np.allclose(result1.coeff, truth1)
#test with random matrices
cheb2 = np.random.randint(-9, 9, (4, 4))
C1 = MultiCheb(cheb2)
C2 = cheb2poly(C1)
C3 = MultiCheb.mon_mult(C1, (1, 1))
C4 = MultiPower.mon_mult(C2, (1, 1))
C5 = poly2cheb(C4)
assert np.allclose(C3.coeff, C5.coeff)
# test results of chebyshev mult compared to power multiplication
cheb3 = np.random.randn(5, 4)
c1 = MultiCheb(cheb3)
c2 = MultiCheb(np.ones((4, 2)))
for index, i in np.ndenumerate(c2.coeff):
if sum(index) == 0:
c3 = c1.mon_mult(index)
else:
c3 = c3 + c1.mon_mult(index)
p1 = cheb2poly(c1)
p2 = cheb2poly(c2)
p3 = p1 * p2
p4 = cheb2poly(c3)
assert np.allclose(p3.coeff, p4.coeff)
# test results of chebyshev mult compared to power multiplication in 3D
cheb4 = np.random.randn(3, 3, 3)
a1 = MultiCheb(cheb4)
a2 = MultiCheb(np.ones((3, 3, 3)))
for index, i in np.ndenumerate(a2.coeff):
if sum(index) == 0:
a3 = a1.mon_mult(index)
else:
a3 = a3 + a1.mon_mult(index)
q1 = cheb2poly(a1)
q2 = cheb2poly(a2)
q3 = q1 * q2
q4 = cheb2poly(a3)
assert np.allclose(q3.coeff, q4.coeff)
def test_cheb2poly():
np.random.seed(23)
c1 = MultiCheb(np.array([[0, 0, 0], [0, 0, 0], [0, 1, 1]]))
c2 = cheb2poly(c1)
truth = np.array([[1, -1, -2], [0, 0, 0], [-2, 2, 4]])
assert np.allclose(truth, c2.coeff)
#test2
c3 = MultiCheb(np.array([[3, 1, 4, 7], [8, 3, 1, 2], [0, 5, 6, 2]]))
c4 = cheb2poly(c3)
truth2 = np.array([[5, -19, -4, 20], [7, -3, 2, 8], [-12, -2, 24, 16]])
assert np.allclose(truth2, c4.coeff)
c4_1 = poly2cheb(c4)
assert np.allclose(c3.coeff, c4_1.coeff)
#Test3
c5 = MultiCheb(
np.array([[3, 1, 3, 4, 6], [2, 0, 0, 2, 0], [3, 1, 5, 1, 8]]))
c6 = cheb2poly(c5)
truth3 = np.array([[0, -9, 12, 12, -16], [2, -6, 0, 8, 0],
[12, -4, -108, 8, 128]])
assert np.allclose(truth3, c6.coeff)
#test4 - Random 1D
matrix2 = np.random.randint(1, 100, np.random.randint(1, 10))
c7 = MultiCheb(matrix2)
c8 = cheb2poly(c7)
c9 = poly2cheb(c8)
assert np.allclose(c7.coeff, c9.coeff)
#Test5 - Random 2D
shape = list()
for i in range(2):
shape.append(np.random.randint(2, 10))
matrix1 = np.random.randint(1, 50, (shape))
c10 = MultiCheb(matrix1)
c11 = cheb2poly(c10)
c12 = poly2cheb(c11)
assert np.allclose(c10.coeff, c12.coeff)
#Test6 - Random 4D
shape = list()
for i in range(4):
shape.append(np.random.randint(2, 10))
matrix1 = np.random.randint(1, 50, (shape))
c13 = MultiCheb(matrix1)
c14 = cheb2poly(c13)
c15 = poly2cheb(c14)
assert np.allclose(c13.coeff, c15.coeff)
def chebspolyqeps(Q, eps):
dim = Q.shape[0]
polys = []
const = (1.5 * np.eye(dim) - eps * Q).sum(axis=1)
for i in range(dim):
coeff = np.zeros([3] * dim)
spot = [0] * dim
coeff[tuple(spot)] = const[i]
for j in range(dim):
spot[j] = 1
coeff[tuple(spot)] = Q[i, j] * eps
if i == j:
coeff[tuple(spot)] -= 2
spot[j] = 0
spot[i] = 2
coeff[tuple(spot)] = .5
polys.append(MultiCheb(coeff))
return polys
def _random_poly(_type, dim):
'''
Generates a random linear polynomial that has the form
c_1x_1 + c_2x_2 + ... + c_nx_n where n = dim and each c_i is a randomly
chosen integer between 0 and 1000.
Parameters
----------
_type : string
Type of Polynomial to generate. "MultiCheb" or "MultiPower".
dim : int
Degree of polynomial to generate (?).
Returns
-------
Polynomial
Randomly generated Polynomial.
'''
_vars = get_var_list(dim)
random_poly_shape = [2 for i in range(dim)]
# random_poly_coeff = np.zeros(tuple(random_poly_shape), dtype=int)
# for var in _vars:
# random_poly_coeff[var] = np.random.randint(1000)
random_poly_coeff = np.zeros(tuple(random_poly_shape), dtype=float)
#np.random.seed(42)
coeffs = np.random.rand(dim)
coeffs /= np.linalg.norm(coeffs)
for i, var in enumerate(_vars):
random_poly_coeff[var] = coeffs[i]
if _type == 'MultiCheb':
return MultiCheb(random_poly_coeff), _vars
else:
return MultiPower(random_poly_coeff), _vars
def test_paper_example():
#Power form of the polys
p1 = MultiPower(np.array([[1, -4, 0], [0, 3, 0], [1, 0,
0]])) #y^2 + 3xy - 4x +1
p2 = MultiPower(np.array([[3, 0, -2], [6, -6, 0],
[0, 0, 0]])) #-6xy -2x^2 + 6y +3
#Cheb form of the polys
c1 = MultiCheb(np.array([[2, 0, -1], [6, -6, 0], [0, 0,
0]])) #p1 in Cheb form
c2 = MultiCheb(np.array([[1.5, -4, 0], [0, 3, 0], [.5, 0,
0]])) #p2 in Cheb form
right_number_of_roots = 4
#~ ~ ~ Power Form, Mx Matrix ~ ~ ~
power_mult_roots = pr.solve([p1, p2], MSmatrix=1)
assert len(power_mult_roots) == right_number_of_roots
for root in power_mult_roots:
assert np.isclose(0, p1(root), atol=1.e-8)
assert np.isclose(0, p2(root), atol=1.e-8)
#~ ~ ~ Cheb Form, Mx Matrix ~ ~ ~
cheb_mult_roots = pr.solve([c1, c2], MSmatrix=1)
assert len(cheb_mult_roots) == right_number_of_roots
for root in cheb_mult_roots:
assert np.isclose(0, c1(root), atol=1.e-8)
assert np.isclose(0, c1(root), atol=1.e-8)
#~ ~ ~ Power Form, My Matrix ~ ~ ~
power_multR_roots = pr.solve([p1, p2], MSmatrix=2)
assert len(power_multR_roots) == right_number_of_roots
for root in power_multR_roots:
assert np.isclose(0, p1(root), atol=1.e-8)
assert np.isclose(0, p2(root), atol=1.e-8)
#~ ~ ~ Cheb Form, My Matrix ~ ~ ~
cheb_multR_roots = pr.solve([c1, c2], MSmatrix=2)
assert len(cheb_multR_roots) == right_number_of_roots
for root in cheb_multR_roots:
assert np.isclose(0, c1(root), atol=1.e-8)
assert np.isclose(0, c1(root), atol=1.e-8)
#~ ~ ~ Power Form, Pseudorandom Multiplication Matrix ~ ~ ~
power_multrand_roots = pr.solve([p1, p2], MSmatrix=0)
assert len(power_multrand_roots) == right_number_of_roots
for root in power_multrand_roots:
assert np.isclose(0, p1(root), atol=1.e-8)
assert np.isclose(0, p2(root), atol=1.e-8)
#~ ~ ~ Cheb Form, Pseudorandom Multiplication Matrix ~ ~ ~
cheb_multrand_roots = pr.solve([c1, c2], MSmatrix=0)
assert len(cheb_multrand_roots) == right_number_of_roots
for root in cheb_multrand_roots:
assert np.isclose(0, c1(root), atol=1.e-8)
assert np.isclose(0, c1(root), atol=1.e-8)
#~ ~ ~ Power Form, Division Matrix ~ ~ ~
power_div_roots = pr.solve([p1, p2], MSmatrix=-1)
assert len(power_div_roots) == right_number_of_roots
for root in power_div_roots:
assert np.isclose(0, p1(root), atol=1.e-8)
assert np.isclose(0, p2(root), atol=1.e-8)
#~ ~ ~ Cheb Form, Division Matrix ~ ~ ~
cheb_div_roots = pr.solve([c1, c2], MSmatrix=-1)
assert len(cheb_div_roots) == right_number_of_roots
for root in cheb_div_roots:
assert np.isclose(0, c1(root), atol=1.e-8)
assert np.isclose(0, c2(root), atol=1.e-8)
def subdivision_solve_nd(funcs,
a,
b,
deg,
target_deg,
interval_data,
root_tracker,
tols,
max_level,
good_degs=None,
level=0,
method='svd',
use_target_tol=False,
trust_small_evals=False):
"""Finds the common zeros of the given functions.
All the zeros will be stored in root_tracker.
Parameters
----------
funcs : list
Each element of the list is a callable function.
a : numpy array
The lower bound on the interval.
b : numpy array
The upper bound on the interval.
deg : int
The degree to approximate with in the chebyshev approximation.
target_deg : int
The degree to subdivide down to before building the Macaulay matrix.
interval_data : IntervalData
A class to run the subinterval checks and keep track of the solve
progress
root_tracker : RootTracker
A class to keep track of the roots that are found.
tols : Tolerances
The tolerances to be used.
max_level : int
The maximum level for the recursion
good_degs : numpy array
Interpoation degrees that are guaranteed to give an approximation valid
to within approx_tol.
level : int
The current level of the recursion.
method : str (optional)
The method to use when reducing the Macaulay matrix. Valid options are
svd, tvb, and qrt.
use_target_tol : bool
Whether or not to use tols.target_tol when making approximations. This
is necessary to get a sufficiently accurate approximation from which to
build the Macaulay matrix and run the solver.
"""
if level >= max_level:
# TODO Refine case where there may be a root and it goes too deep.
interval_data.track_interval("Too Deep", [a, b])
# Return potential roots if the residuals are small
root_tracker.add_potential_roots((a + b) / 2, a, b, "Too Deep.")
return
dim = len(a)
if tols.check_eval_error:
# Using the first abs_approx_tol
if not use_target_tol:
tols.abs_approx_tol = tols.abs_approx_tols[tols.currTol]
if level % tols.check_eval_freq == 0:
numSpots = (deg * 2)**len(a) - (deg)**len(a)
for func in funcs:
tols.abs_approx_tol = max(
tols.abs_approx_tol,
numSpots * get_abs_approx_tol(func, 3, a, b, dim))
# Using target_tol
else:
tols.target_tol = tols.target_tols[tols.currTol]
if level % tols.check_eval_freq == 0:
numSpots = (deg * 2)**len(a) - (deg)**len(a)
for func in funcs:
tols.target_tol = max(
tols.target_tol,
numSpots * get_abs_approx_tol(func, 3, a, b, dim))
# Buffer the interval to solve on a larger interval to account for
# corners. Right now, it's set to be 5e-10 so that on [-1, 1], the
# buffer goes out 1e-9 around the initial search interval.
# DETERMINED BY EXPERIMENTATION
interval_buffer_size = (b - a) * 5e-10
og_a = a.copy()
og_b = b.copy()
a -= interval_buffer_size
b += interval_buffer_size
cheb_approx_list = []
interval_data.print_progress()
if good_degs is None:
good_degs = [None] * len(funcs)
inf_norms = []
approx_errors = []
# Get the chebyshev approximations
num_funcs = len(funcs)
for func_num, (func, good_deg) in enumerate(zip(funcs, good_degs)):
if use_target_tol:
coeff, inf_norm, approx_error = full_cheb_approximate(
func, a, b, deg, tols.target_tol, tols.rel_approx_tol,
good_deg)
else:
coeff, inf_norm, approx_error = full_cheb_approximate(
func, a, b, deg, tols.abs_approx_tol, tols.rel_approx_tol,
good_deg)
inf_norms.append(inf_norm)
approx_errors.append(approx_error)
# Subdivides if a bad approximation
if coeff is None:
if not trust_small_evals:
approx_errors = [max(err, macheps) for err in approx_errors]
intervals = interval_data.get_subintervals(og_a, og_b,
cheb_approx_list,
approx_errors, False)
#reorder funcs. TODO: fancier things like how likely it is to pass checks
funcs2 = funcs.copy()
if func_num + 1 < num_funcs:
del funcs2[func_num]
funcs2.append(func)
for new_a, new_b in intervals:
subdivision_solve_nd(funcs2,
new_a,
new_b,
deg,
target_deg,
interval_data,
root_tracker,
tols,
max_level,
level=level + 1,
method=method,
trust_small_evals=trust_small_evals)
return
else:
# Run checks to try and throw out the interval
if not trust_small_evals:
approx_error = max(approx_error, macheps)
if interval_data.check_interval(coeff, approx_error, og_a, og_b):
return
cheb_approx_list.append(coeff)
# Reduce the degree of the approximations while not introducing too much error
coeffs, good_approx, approx_errors = trim_coeffs(cheb_approx_list,
tols.abs_approx_tol,
tols.rel_approx_tol,
inf_norms, approx_errors)
if not trust_small_evals:
approx_errors = [max(err, macheps) for err in approx_errors]
# Used if subdividing further.
# Only choose good_degs if the approximation after trim_coeffs is good.
if good_approx:
# good_degs are assumed to be 1 higher than the current approximation
# but no larger than the initial degree for more accurate performance.
good_degs = [min(coeff.shape[0], deg) for coeff in coeffs]
good_zeros_tol = max(
tols.min_good_zeros_tol,
sum(np.abs(approx_errors)) * tols.good_zeros_factor)
# Check if the degree is small enough or if trim_coeffs introduced too much error
if np.any(np.array([coeff.shape[0] for coeff in coeffs]) > target_deg +
1) or not good_approx:
intervals = interval_data.get_subintervals(og_a, og_b,
cheb_approx_list,
approx_errors, True)
for new_a, new_b in intervals:
subdivision_solve_nd(funcs,
new_a,
new_b,
deg,
target_deg,
interval_data,
root_tracker,
tols,
max_level,
good_degs,
level + 1,
method=method,
trust_small_evals=trust_small_evals,
use_target_tol=True)
# Check if any approx error is greater than target_tol for Macaulay method
elif np.any(
np.array(approx_errors) > np.array(tols.target_tol) +
tols.rel_approx_tol * np.array(inf_norms)):
intervals = interval_data.get_subintervals(og_a, og_b,
cheb_approx_list,
approx_errors, True)
for new_a, new_b in intervals:
subdivision_solve_nd(funcs,
new_a,
new_b,
deg,
target_deg,
interval_data,
root_tracker,
tols,
max_level,
good_degs,
level + 1,
method=method,
trust_small_evals=trust_small_evals,
use_target_tol=True)
# Check if everything is linear
elif np.all(np.array([coeff.shape[0] for coeff in coeffs]) == 2):
if deg != 2:
subdivision_solve_nd(funcs,
a,
b,
2,
target_deg,
interval_data,
root_tracker,
tols,
max_level,
good_degs,
level,
method=method,
trust_small_evals=trust_small_evals,
use_target_tol=True)
return
zero, cond = solve_linear(coeffs)
# Store the information and exit
zero = good_zeros_nd(zero, good_zeros_tol, good_zeros_tol)
zero = transform(zero, a, b)
zero = zeros_in_interval(zero, og_a, og_b, dim)
interval_data.track_interval("Base Case", [a, b])
root_tracker.add_roots(zero, a, b, "Base Case")
# Solve using spectral methods if stable.
else:
polys = [
MultiCheb(coeff, lead_term=[coeff.shape[0] - 1], clean_zeros=False)
for coeff in coeffs
]
res = multiplication(polys,
max_cond_num=tols.max_cond_num,
method=method)
#check for a conditioning error
if res[0] is None:
# Subdivide but run some checks on the intervals first
intervals = interval_data.get_subintervals(og_a, og_b,
cheb_approx_list,
approx_errors, True)
for new_a, new_b in intervals:
subdivision_solve_nd(funcs,
new_a,
new_b,
deg,
target_deg,
interval_data,
root_tracker,
tols,
max_level,
good_degs,
level + 1,
method=method,
trust_small_evals=trust_small_evals,
use_target_tol=True)
else:
zeros = res
zeros = good_zeros_nd(zeros, good_zeros_tol, good_zeros_tol)
zeros = transform(zeros, a, b)
zeros = zeros_in_interval(zeros, og_a, og_b, dim)
interval_data.track_interval("Macaulay", [a, b])
root_tracker.add_roots(zeros, a, b, "Macaulay")
def build_macaulay(initial_poly_list, verbose=False):
"""Constructs the unreduced Macaulay matrix. Removes linear polynomials by
substituting in for a number of variables equal to the number of linear
polynomials.
Parameters
--------
initial_poly_list: list
The polynomials in the system we are solving.
verbose : bool
Prints information about how the roots are computed.
Returns
-----------
matrix : 2d ndarray
The Macaulay matrix
matrix_terms : 2d integer ndarray
Array containing the ordered basis, where the ith row contains the
exponent/degree of the ith basis monomial
cut : int
Where to cut the Macaulay matrix for the highest-degree monomials
varsToRemove : list
The variables removed with removing linear polynomials
A : 2d ndarray
A matrix giving the linear relations between the removed variables and
the remaining variables
Pc : 1d integer ndarray
Array containing the order of the variables as the appear in the columns
of A
"""
power = is_power(initial_poly_list)
dim = initial_poly_list[0].dim
poly_coeff_list = []
degree = find_degree(initial_poly_list)
linear_polys = [poly for poly in initial_poly_list if poly.degree == 1]
nonlinear_polys = [poly for poly in initial_poly_list if poly.degree != 1]
#Choose which variables to remove if things are linear, and add linear polys to matrix
if len(linear_polys) >= 1: #Linear polys involved
#get the row rededuced linear coefficients
A, Pc = nullspace(linear_polys)
varsToRemove = Pc[:len(A)].copy()
#add to macaulay matrix
for row in A:
#reconstruct a polynomial for each row
coeff = np.zeros([2] * dim)
coeff[tuple(get_var_list(dim))] = row[:-1]
coeff[tuple([0] * dim)] = row[-1]
if not power:
poly = MultiCheb(coeff)
else:
poly = MultiPower(coeff)
poly_coeff_list = add_polys(degree, poly, poly_coeff_list)
else: #no linear
A, Pc = None, None
varsToRemove = []
#add nonlinear polys to poly_coeff_list
for poly in nonlinear_polys:
poly_coeff_list = add_polys(degree, poly, poly_coeff_list)
#Creates the matrix
# return (*create_matrix(poly_coeff_list, degree, dim, varsToRemove), A, Pc)
return create_matrix(poly_coeff_list, degree, dim, varsToRemove)
def subdivision_solve_nd(funcs,
a,
b,
deg,
target_deg,
interval_data,
root_tracker,
tols,
max_level,
good_degs=None,
level=0,
method='svd'):
"""Finds the common zeros of the given functions.
All the zeros will be stored in root_tracker.
Parameters
----------
funcs : list
Each element of the list is a callable function.
a : numpy array
The lower bound on the interval.
b : numpy array
The upper bound on the interval.
deg : int
The degree to approximate with in the chebyshev approximation.
target_deg : int
The degree to subdivide down to before building the Macaulay matrix.
interval_data : IntervalData
A class to run the subinterval checks and keep track of the solve progress
root_tracker : RootTracker
A class to keep track of the roots that are found.
tols : Tolerances
The tolerances to be used.
max_level : int
The maximum level for the recursion
good_degs : numpy array
Interpoation degrees that are guaranteed to give an approximation valid to within approx_tol.
level : int
The current level of the recursion.
method : str (optional)
The method to use when reducing the Macaulay matrix. Valid options are
svd, tvb, and qrt.
"""
if level >= max_level:
# TODO Refine case where there may be a root and it goes too deep.
interval_data.track_interval("Too Deep", [a, b])
# Return potential roots if the residuals are small
root_tracker.add_potential_roots((a + b) / 2, a, b, "Too Deep.")
return
if tols.check_eval_error:
tols.abs_approx_tol = tols.abs_approx_tols[tols.currTol]
if level % tols.check_eval_freq == 0:
numSpots = (deg * 2)**len(a) - (deg)**len(a)
for func in funcs:
tols.abs_approx_tol = max(
tols.abs_approx_tol,
numSpots * getAbsApproxTol(func, 3, a, b))
cheb_approx_list = []
interval_data.print_progress()
dim = len(a)
if good_degs is None:
good_degs = [None] * len(funcs)
inf_norms = []
approx_errors = []
#Get the chebyshev approximations
for func, good_deg in zip(funcs, good_degs):
coeff, change_sign, inf_norm, approx_error = full_cheb_approximate(
func, a, b, deg, tols.abs_approx_tol, tols.rel_approx_tol,
good_deg)
inf_norms.append(inf_norm)
approx_errors.append(approx_error)
#Subdivides if a bad approximation
if coeff is None:
intervals = get_subintervals(a, b, change_sign, None, None, None,
approx_errors)
for new_a, new_b in intervals:
subdivision_solve_nd(funcs,
new_a,
new_b,
deg,
target_deg,
interval_data,
root_tracker,
tols,
max_level,
level=level + 1,
method=method)
return
else:
#if the function changes sign on at least one subinterval, skip the checks
if np.any(change_sign):
cheb_approx_list.append(coeff)
continue
#Run checks to try and throw out the interval
if interval_data.check_interval(coeff, approx_error, a, b):
return
cheb_approx_list.append(coeff)
#Make the system stable to solve
coeffs, good_approx, approx_errors = trim_coeffs(cheb_approx_list,
tols.abs_approx_tol,
tols.rel_approx_tol,
inf_norms, approx_errors)
#Used if subdividing further.
good_degs = [coeff.shape[0] - 1 for coeff in coeffs]
good_zeros_tol = max(
tols.min_good_zeros_tol,
np.sum(np.abs(approx_errors)) * tols.good_zeros_factor)
#Check if everything is linear
if np.all(np.array([coeff.shape[0] for coeff in coeffs]) == 2):
if deg != 2:
subdivision_solve_nd(funcs,
a,
b,
2,
target_deg,
interval_data,
root_tracker,
tols,
max_level,
good_degs,
level,
method=method)
return
zero, cond = solve_linear(coeffs)
grad = [MultiCheb(c).grad(zero) for c in coeffs]
#Store the information and exit
zero = good_zeros_nd(zero, good_zeros_tol, good_zeros_tol)
zero = transform(zero, a, b)
interval_data.track_interval("Base Case", [a, b])
root_tracker.add_roots(zero, a, b, "Base Case")
#Check if anything is linear
# elif np.any(np.array([coeff.shape[0] for coeff in coeffs]) == 2):
# #Subdivide but run some checks on the intervals first
# intervals = get_subintervals(a,b,np.arange(dim),interval_data,cheb_approx_list,change_sign,approx_errors,True)
# for new_a, new_b in intervals:
# subdivision_solve_nd(method,funcs,new_a,new_b,deg, target_deg,interval_data,root_tracker,tols,max_level,good_degs,level+1)
#Runs the same things as above, but we want to get rid of that eventually so keep them seperate.
elif np.any(
np.array([coeff.shape[0]
for coeff in coeffs]) > target_deg) or not good_approx:
intervals = get_subintervals(a, b, np.arange(dim), interval_data,
cheb_approx_list, change_sign,
approx_errors, True)
for new_a, new_b in intervals:
subdivision_solve_nd(funcs,
new_a,
new_b,
deg,
target_deg,
interval_data,
root_tracker,
tols,
max_level,
good_degs,
level + 1,
method=method)
#Solve using spectral methods if stable.
else:
polys = [
MultiCheb(coeff, lead_term=[coeff.shape[0] - 1], clean_zeros=False)
for coeff in coeffs
]
try:
zeros = multiplication(polys,
max_cond_num=tols.max_cond_num,
method=method)
grad = [[poly.grad(z) for poly in polys] for z in zeros]
zeros = good_zeros_nd(zeros, good_zeros_tol, good_zeros_tol)
zeros = transform(zeros, a, b)
interval_data.track_interval("Macaulay", [a, b])
root_tracker.add_roots(zeros, a, b, "Macaulay")
except (ConditioningError, TooManyRoots) as e:
#Subdivide but run some checks on the intervals first
intervals = get_subintervals(a, b, np.arange(dim), interval_data,
cheb_approx_list, change_sign,
approx_errors, True)
for new_a, new_b in intervals:
subdivision_solve_nd(funcs,
new_a,
new_b,
deg,
target_deg,
interval_data,
root_tracker,
tols,
max_level,
good_degs,
level + 1,
method=method)
def run_n_dimension(args, radius, eigvals):
num_points = args.num_points
eps = args.eps
power = args.power
real = args.real
by_coeffs = args.coeffs
dim = args.dimension
root_pts = {}
residuals = {}
powerpolys = []
chebpolys = []
if by_coeffs:
for i in range(dim):
from yroots.polynomial import getPoly
powerpolys.append(getPoly(num_points, dim, power=True))
chebpolys.append(getPoly(num_points, dim, power=False))
else:
r = np.random.random((num_points, dim)) * radius + eps
roots = 2 * r - radius
root_pts = {'roots': np.array(list(product(*np.rot90(roots))))}
for i in range(dim):
coeffs = np.zeros((num_points + 1, ) * dim)
idx = [
slice(None),
] * dim
idx[i] = 0
coeffs[tuple(idx)] = polyfromroots(roots[:, i])
lt = [0] * dim
lt[i] = num_points
powerpolys.append(
MultiPower(coeffs)) #, lead_term=lt, clean_zeros=False))
coeffs[tuple(idx)] = chebfromroots(roots[:, i])
chebpolys.append(
MultiCheb(coeffs)) #, lead_term=lt, clean_zeros=False))
# plt.subplot(121);plt.imshow(coeffs);plt.subplot(122);plt.imshow(chebpolys[0].coeff);plt.show()
for solver in all_solvers:
if not isinstance(solver, OneDSolver) and solver.basis in [
'power', 'both'
]:
# if ((not eigvals) or solver.eigvals):
name = str(solver) + ' Power'
root_pts[name] = solver(powerpolys)
residuals[name] = maximal_residual(powerpolys, root_pts[name])
for solver in all_solvers:
if not isinstance(solver, OneDSolver) and solver.basis in [
'cheb', 'both'
]:
# if ((not eigvals) or solver.eigvals):
name = str(solver) + ' Cheb'
root_pts[name] = solver(chebpolys)
residuals[name] = maximal_residual(chebpolys, root_pts[name])
if args.hist:
evaluations = {}
for k, v in root_pts.items():
if k == 'roots': continue
# evaluations[k] = []
polys = powerpolys if 'Power' in k else chebpolys
# for poly in polys:
evaluations[k] = sum(np.abs(poly(root_pts[k])) for poly in polys)
ncols = len(evaluations)
# plt.figure(figsize=(12,6))
fig, ax = plt.subplots(1, ncols, sharey=True, figsize=(12, 4))
minimal = -15
maximal = 1
for i, (k, v) in enumerate(evaluations.items()):
ax[i].hist(np.clip(np.log10(v), minimal, maximal),
range=(minimal, maximal),
bins=40)
ax[i].set_xlabel(r'$log_{10}(p(r_i))$')
ax[i].set_title(k)
plt.suptitle("Eigenvalues" if eigvals else "Eigenvectors")
plt.show()
return root_pts, residuals
def base_quadratic_check(test_coeff, tol):
"""Slow nd-quadratic check to test against. """
#get the dimension and make sure the coeff tensor has all the right
# quadratic coeff spots, set to zero if necessary
dim = test_coeff.ndim
intervals = get_subintervals(-np.ones(dim), np.ones(dim), np.arange(dim),
None, None, tol)
padding = [(0, max(0, 3 - i)) for i in test_coeff.shape]
test_coeff = np.pad(test_coeff.copy(), padding, mode='constant')
#Possible extrema of qudaratic part are where D_xk = 0 for some subset of the variables xk
# with the other variables are fixed to a boundary value
#Dxk = c[0,...,0,1,0,...0] (k-spot is 1) + 4c[0,...,0,2,0,...0] xk (k-spot is 2)
# + \Sum_{j\neq k} xj c[0,...,0,1,0,...,0,1,0,...0] (k and j spot are 1)
#This gives a symmetric system of equations AX+B = 0
#We will fix different columns of X each time, resulting in slightly different
#systems, but storing A and B now will be helpful later
#pull out coefficients we care about
quad_coeff = np.zeros([3] * dim)
A = np.zeros([dim, dim])
B = np.zeros(dim)
for spot in itertools.product(np.arange(3), repeat=dim):
if np.sum(spot) < 3:
spot_array = np.array(spot)
if np.sum(spot_array != 0) == 2:
#coef of cross terms
i, j = np.where(spot_array != 0)[0]
A[i, j] = test_coeff[spot].copy()
A[j, i] = test_coeff[spot].copy()
elif np.any(spot_array == 2):
#coef of pure quadratic terms
i = np.where(spot_array != 0)[0][0]
A[i, i] = 4 * test_coeff[spot].copy()
elif np.any(spot_array == 1):
#coef of linear terms
i = np.where(spot_array != 0)[0][0]
B[i] = test_coeff[spot].copy()
quad_coeff[spot] = test_coeff[spot]
test_coeff[spot] = 0
#create a poly object for evaluations
quad_poly = MultiCheb(quad_coeff)
#The sum of the absolute values of everything else
other_sum = np.sum(np.abs(test_coeff))
def powerset(iterable):
"powerset([1,2,3]) --> () (1,) (2,) (3,) (1,2) (1,3) (2,3) (1,2,3)"
s = list(iterable)
return itertools.chain.from_iterable(itertools.combinations(s, r)\
for r in range(len(s)+1))
mask = []
for interval_num, interval in enumerate(intervals):
extreme_points = []
for fixed in powerset(np.arange(dim)):
fixed = np.array(fixed)
if len(fixed) == 0:
#fix no vars--> interior
if np.linalg.matrix_rank(A) < A.shape[0]:
#no interior critical point
continue
X = la.solve(A, -B, assume_a='sym')
#make sure it's in the domain
if np.all([
interval[0][i] <= X[i] <= interval[1][i]
for i in range(dim)
]):
extreme_points.append(quad_poly(X))
elif len(fixed) == dim:
#fix all variables--> corners
for corner in itertools.product([0, 1], repeat=dim):
#j picks if upper/lower bound. i is which var
extreme_points.append(
quad_poly(
[interval[j][i] for i, j in enumerate(corner)]))
else:
#fixed some variables --> "sides"
#we only care about the equations from the unfixed variables
unfixed = np.delete(np.arange(dim), fixed)
A_ = A[unfixed][:, unfixed]
if np.linalg.matrix_rank(A_) < A_.shape[0]:
#no solutions
continue
fixed_A = A[unfixed][:, fixed]
B_ = B[unfixed]
for side in itertools.product([0, 1], repeat=len(fixed)):
X0 = np.array([interval[j][i] for i, j in enumerate(side)])
X_ = la.solve(A_, -B_ - fixed_A @ X0, assume_a='sym')
X = np.zeros(dim)
X[fixed] = X0
X[unfixed] = X_
if np.all([
interval[0][i] <= X[i] <= interval[1][i]
for i in range(dim)
]):
extreme_points.append(quad_poly(X))
#No root if min(extreme_points) > (other_sum + tol)
# OR max(extreme_points) < -(other_sum+tol)
#Logical negation gives the boolean we want
mask.append(
np.min(extreme_points) < (other_sum + tol)
and np.max(extreme_points) > -(other_sum + tol))
return mask
def test_evaluate2():
cheb = MultiCheb(np.array([[0, 0, 0, 1], [0, 0, 0, 0], [0, 0, .5, 0]]))
value = cheb((2, 5))
assert (np.isclose(value, 656.5))
def MacaulayReduction(initial_poly_list,
max_cond_num,
macaulay_zero_tol,
verbose=False):
"""Reduces the Macaulay matrix to find a vector basis for the system of polynomials.
Parameters
--------
initial_poly_list: list
The polynomials in the system we are solving.
max_cond_num : float
The maximum condition number of the Macaulay Matrix Reduction
macaulay_zero_tol : float
What is considered 0 in the macaulay matrix reduction.
verbose : bool
Prints information about how the roots are computed.
Returns
-----------
basisDict : dict
A dictionary of terms not in the vector basis a matrixes of things in the vector basis that the term
can be reduced to.
VB : numpy array
The terms in the vector basis, each row being a term.
varsToRemove : list
The variables to remove from the basis because we have linear polysnomials
Raises
------
ConditioningError if rrqr_reduceMacaulay(...) raises a ConditioningError.
"""
power = is_power(initial_poly_list)
dim = initial_poly_list[0].dim
poly_coeff_list = []
degree = find_degree(initial_poly_list)
linear_polys = [poly for poly in initial_poly_list if poly.degree == 1]
nonlinear_polys = [poly for poly in initial_poly_list if poly.degree != 1]
#Choose which variables to remove if things are linear, and add linear polys to matrix
if len(linear_polys) == 1: #one linear
varsToRemove = [
np.argmax(np.abs(linear_polys[0].coeff[get_var_list(dim)]))
]
poly_coeff_list = add_polys(degree, linear_polys[0], poly_coeff_list)
elif len(linear_polys) > 1: #multiple linear
#get the row rededuced linear coefficients
A, Pc = nullspace(linear_polys)
varsToRemove = Pc[:len(A)].copy()
#add to macaulay matrix
for row in A:
#reconstruct a polynomial for each row
coeff = np.zeros([2] * dim)
coeff[get_var_list(dim)] = row[:-1]
coeff[tuple([0] * dim)] = row[-1]
if power:
poly = MultiPower(coeff)
else:
poly = MultiCheb(coeff)
poly_coeff_list = add_polys(degree, poly, poly_coeff_list)
else: #no linear
varsToRemove = []
#add nonlinear polys to poly_coeff_list
for poly in nonlinear_polys:
poly_coeff_list = add_polys(degree, poly, poly_coeff_list)
#Creates the matrix
matrix, matrix_terms, cuts = create_matrix(poly_coeff_list, degree, dim,
varsToRemove)
if verbose:
np.set_printoptions(suppress=False, linewidth=200)
print('\nStarting Macaulay Matrix\n', matrix)
print(
'\nColumns in Macaulay Matrix\nFirst element in tuple is degree of x, Second element is degree of y\n',
matrix_terms)
print(
'\nLocation of Cuts in the Macaulay Matrix into [ Mb | M1* | M2* ]\n',
cuts)
try:
matrix, matrix_terms = rrqr_reduceMacaulay(
matrix,
matrix_terms,
cuts,
max_cond_num=max_cond_num,
macaulay_zero_tol=macaulay_zero_tol)
except ConditioningError as e:
raise e
# TODO: rrqr_reduceMacaulay2 is not working when expected.
# if np.allclose(matrix[cuts[0]:,:cuts[0]], 0):
# matrix, matrix_terms = rrqr_reduceMacaulay2(matrix, matrix_terms, cuts, accuracy = accuracy)
# else:
# matrix, matrix_terms = rrqr_reduceMacaulay(matrix, matrix_terms, cuts, accuracy = accuracy)
if verbose:
np.set_printoptions(suppress=True, linewidth=200)
print("\nFinal Macaulay Matrix\n", matrix)
print("\nColumns in Macaulay Matrix\n", matrix_terms)
VB = matrix_terms[matrix.shape[0]:]
basisDict = makeBasisDict(matrix, matrix_terms, VB, power)
return basisDict, VB, varsToRemove
def MSMultMatrix(polys,
poly_type,
max_cond_num,
macaulay_zero_tol,
verbose=False,
MSmatrix=0):
'''
Finds the multiplication matrix using the reduced Macaulay matrix.
Parameters
----------
polys : array-like
The polynomials to find the common zeros of
poly_type : string
The type of the polynomials in polys
verbose : bool
Prints information about how the roots are computed.
MSmatrix : int
Controls which Moller-Stetter matrix is constructed. The options are:
0 (default) -- The Moller-Stetter matrix of a random polynomial
Some positive integer i < dimension -- The Moller-Stetter matrix of x_i
max_cond_num : float
The maximum condition number of the Macaulay Matrix Reduction
macaulay_zero_tol : float
What is considered 0 in the macaulay matrix reduction.
Returns
-------
multiplicationMatrix : 2D numpy array
The multiplication matrix for a random polynomial f
var_dict : dictionary
Maps each variable to its position in the vector space basis
basisDict : dict
A dictionary of terms not in the vector basis a matrixes of things in the vector basis that the term
can be reduced to.
VB : numpy array
The terms in the vector basis, each row being a term.
Raises
------
ConditioningError if MacaulayReduction(...) raises a ConditioningError.
'''
try:
basisDict, VB, varsRemoved = MacaulayReduction(
polys,
max_cond_num=max_cond_num,
macaulay_zero_tol=macaulay_zero_tol,
verbose=verbose)
except ConditioningError as e:
raise e
dim = max(f.dim for f in polys)
# Get the polynomial to make the MS matrix of
if MSmatrix == 0: #random poly
f = _random_poly(poly_type, dim)[0]
else: #multiply by x_i where i is determined by MSmatrix
xi_ind = np.zeros(dim, dtype=int)
xi_ind[MSmatrix - 1] = 1
coef = np.zeros((2, ) * dim)
coef[tuple(xi_ind)] = 1
if poly_type == "MultiPower":
f = MultiPower(np.array(coef))
elif poly_type == "MultiCheb":
f = MultiCheb(np.array(coef))
else:
raise ValueError()
if verbose:
print(
"\nCoefficients of polynomial whose Moller-Stetter matrix we construt\n",
f.coeff)
#Dictionary of terms in the vector basis their spots in the matrix.
VBdict = {}
spot = 0
for row in VB:
VBdict[tuple(row)] = spot
spot += 1
# Build multiplication matrix m_f
mMatrix = np.zeros((len(VB), len(VB)))
for i in range(VB.shape[0]):
f_coeff = f.mon_mult(VB[i], returnType='Matrix')
for term in zip(*np.where(f_coeff != 0)):
if term in VBdict:
mMatrix[VBdict[term]][i] += f_coeff[term]
else:
mMatrix[:, i] -= f_coeff[term] * basisDict[term]
# Construct var_dict
var_dict = {}
for i in range(len(VB)):
mon = VB[i]
if np.sum(mon) == 1 or np.sum(mon) == 0:
var_dict[tuple(mon)] = i
return mMatrix, var_dict, basisDict, VB
def curvature_check(coeff, tol):
poly = MultiCheb(coeff)
a = np.array([-1.] * poly.dim)
b = np.array([1.] * poly.dim)
return not can_eliminate(poly, a, b, tol)
def run_one_dimension(args, radius, eigvals):
num_points = args.num_points
eps = args.eps
power = args.power
real = args.real
by_coeffs = args.coeffs
root_pts = {}
residuals = {}
if by_coeffs:
coeffs = (np.random.random(num_points + 1) * 2 - 1) * radius
powerpoly = MultiPower(coeffs)
chebpoly = MultiCheb(coeffs)
else:
r = np.sqrt(np.random.random(num_points)) * radius + eps
angles = 2 * np.pi * np.random.random(num_points)
if power and not real:
roots = r * np.exp(angles * 1j)
else:
roots = 2 * r - radius
root_pts = {'roots': roots}
powerpoly = MultiPower(polyfromroots(roots))
chebpoly = MultiCheb(chebfromroots(roots))
n = 1000
x = np.linspace(-1, 1, n)
X, Y = np.meshgrid(x, 1j * x)
for solver in all_solvers:
if isinstance(solver, OneDSolver):
if (solver.basis == 'cheb' and args.cheb) and ((not eigvals)
or solver.eigvals):
name = str(solver)
root_pts[name] = solver(chebpoly, eigvals)
residuals[name] = maximal_residual(chebpoly, root_pts[name])
if (solver.basis == 'power'
and args.power) and ((not eigvals) or solver.eigvals):
name = str(solver)
root_pts[name] = solver(powerpoly, eigvals)
residuals[name] = maximal_residual(powerpoly, root_pts[name])
if args.hist:
evaluations = {}
for k, v in root_pts.items():
if k == 'roots': continue
poly = powerpoly if 'power' in k else chebpoly
evaluations[k] = np.abs(poly(root_pts[k]))
ncols = len(evaluations)
fig, ax = plt.subplots(1, ncols, sharey=True, figsize=(12, 4))
minimal = -20
maximal = 1
for i, (k, v) in enumerate(evaluations.items()):
ax[i].hist(np.clip(np.log10(v), minimal, maximal),
range=(minimal, maximal),
bins=40)
ax[i].set_xlabel(r'$log_{10}(p(r_i))$')
ax[i].set_title(k)
plt.suptitle("Eigenvalues" if eigvals else "Eigenvectors")
plt.show()
return root_pts, residuals
def quadratic_check_nd(test_coeff, intervals, change_sign, tol):
"""One of subinterval_checks
Finds the min of the absolute value of the quadratic part, and compares to the sum of the
rest of the terms.
Parameters
----------
test_coeff_in : numpy array
The coefficient matrix of the polynomial to check
intervals : list
A list of the intervals to check.
change_sign: list
A list of bools of whether we know the functions can change sign on the subintervals.
tol: float
The bound of the sup norm error of the chebyshev approximation.
Returns
-------
mask : list
A list of the results of each interval. False if the function is guarenteed to never be zero
in the unit box, True otherwise
"""
dim = test_coeff.ndim
padding = [(0,max(0,3-i)) for i in test_coeff.shape]
test_coeff = np.pad(test_coeff.copy(), padding, mode='constant')
A = np.zeros([dim,dim])
B = np.zeros(dim)
quad_coeff = np.zeros([3]*dim)
for spot in itertools.product(np.arange(3),repeat=dim):
if np.sum(spot) < 3:
spot_array = np.array(spot)
if np.sum(spot_array != 0) == 2:
i,j = np.where(spot_array != 0)[0]
A[i,j] = test_coeff[spot].copy()
A[j,i] = test_coeff[spot].copy()
elif np.any(spot_array == 2):
i = np.where(spot_array != 0)[0][0]
A[i,i] = 4*test_coeff[spot].copy()
elif np.any(spot_array == 1):
i = np.where(spot_array != 0)[0][0]
B[i] = test_coeff[spot].copy()
quad_coeff[spot] = test_coeff[spot]
test_coeff[spot] = 0
quad_poly = MultiCheb(quad_coeff)
#The sum of the absolute values of everything else
other_sum = np.sum(np.abs(test_coeff))
mask = []
for interval_num, interval in enumerate(intervals):
if change_sign[interval_num]:
mask.append(True)
continue
def powerset(iterable):
"powerset([1,2,3]) --> () (1,) (2,) (3,) (1,2) (1,3) (2,3) (1,2,3)"
s = list(iterable)
return itertools.chain.from_iterable(itertools.combinations(s, r)\
for r in range(len(s)+1))
extreme_points = []
for fixed in powerset(np.arange(dim)):
fixed = np.array(fixed)
if len(fixed) == 0:
others = np.arange(dim)
if np.linalg.matrix_rank(A) < A.shape[0]:
continue
X = np.linalg.solve(A, -B)
if np.all([interval[0][i] <= X[i] <= interval[1][0] for i in range(dim)]):
extreme_points.append(quad_poly(X))
elif len(fixed) == dim:
for corner in itertools.product([0,1],repeat=dim):
extreme_points.append(quad_poly([interval[j][i] for i,j in enumerate(corner)]))
else:
others = np.delete(np.arange(dim), fixed)
A_ = A[others][:,others]
if np.linalg.matrix_rank(A_) < A_.shape[0]:
continue
fixed_A = A[others][:,fixed]
B_ = B[others]
for corner in itertools.product([0,1],repeat=len(fixed)):
X0 = np.array([interval[j][i] for i,j in enumerate(corner)])
X_ = np.linalg.solve(A_, -B_-fixed_A@X0)
X = np.zeros(dim)
X[fixed] = X0
X[others] = X_
if np.all([interval[0][i] <= X[i] <= interval[1][0] for i in range(dim)]):
extreme_points.append(quad_poly(X))
extreme_points = np.array(extreme_points)
#If sign change, True
if not np.all(extreme_points > 0) and not np.all(extreme_points < 0):
mask.append(True)
continue
if np.min(np.abs(extreme_points)) - tol > other_sum:
mask.append(False)
else:
mask.append(True)
return mask