numpy.linalg.eigh 适用于来自 sym.lambdify() 的 numpy 数组，但不适用于来自 np.array()

Question

这是代码。 2a 有效，但 2b 无效，即使两个矩阵具有相同的形状和相同的类型。谁能解释一下？

import sympy as sym  
import numpy as np

x1, x2 = sym.symbols('x1 x2')  
f = 2*x1**2 - 2*x1*x2 + x2**2 + 2*x1 - 2*x2  
xk = np.array([[0 , 1]])  
print("xk shape = ", xk.shape)  
fmat = sym.Matrix([f])  
H = sym.hessian(fmat, (x1, x2))  
print("Symbolic H")  
sym.pprint(H)  

# 2a: Convert the SymPy Matrix to Numpy array using lambdify and then substitute values  
Hnp = sym.lambdify((x1, x2), H, 'numpy')  
Hknp = Hnp(xk[0][0], xk[0][1])  
print("Hknp")  
sym.pprint(Hknp)  
print("Hknp type:", type(Hknp))  
print("Hknp shape = ", Hknp.shape)  
v, Q = np.linalg.eigh(Hknp)  
print("v:", v)  
print("Q:")  
sym.pprint(Q)  

# 2b: Substitute values into SymPy Matrix then convert to Numpy array  
Hks = H.subs([(x1, xk[0][0]), (x2, xk[0][1])])  
Hk = np.array(Hks)  
print("Hk")  
sym.pprint(Hk)  
print("Hk type:", type(Hk))  
print("Hk shape = ", Hk.shape)  
v, Q = np.linalg.eigh(Hk)  
print("v:", v)  
print("Q:")  
sym.pprint(Q)  

Answer 1

>>> Hk
array([[4, -2],
       [-2, 2]], dtype=object)


>>> Hknp.dtype
dtype('int64')

解决方案是Hk = np.array(Hks, dtype=np.int64)。