scipy 求方程组的非负根集

js5cn81o  于 2022-11-09  发布在  其他
关注(0)|答案(2)|浏览(200)

我有一个非线性方程组,我对初始值没有很好的猜测,我需要至少一组正的根,因为在经济学中,这些变量的负值没有多大意义。


# -*- coding: utf-8 -*-

"""
Created on Sat Oct 15 21:48:56 2016

@author: Nick
"""

import scipy as sp
from scipy.optimize import root, fsolve
import numpy as np

# from scipy.optimize import *

el          = 1.1  
eg          = el   
ej          = 10  
om          = 0.3  
omg         = 0.3  
rhog        = 0.8  
xi          = 0.9  
mun         = 2
pidss       = 0.02 
muc         = 0.001
ec          = 2.00 # sims obtains 2.47
beta        = 0.998
h           = 0.8   
kappa       = 4.00 
n           = 1/3.0  
alpha       = 1/3.0  
delta       = 0.025
egs         = eg   
oms         = 0.2  
omgs        = oms  
rhom        = 0.7  
psiygap     = 1.000
psipi       = 2.500
rhoicu      = 0.800
taudss      = 0.01    # steady state tax on domestic consumption (setting it as 0 would create algebraic difficulties)
taumss      = 0.01    # steady state tax on imported consumption for domestic country
taukss      = 0.01    # steady state tax on rental income from capital for domestic country block
taunss      = 0.01    # steady state tax on labor for domestic country
tauydss     = 0.05    
gss         = 0.23    # steady state government spending as a propostion of gdp for domestic country block    
gsss        = 0.23    # steady state government spending as a propostion of gdp for foreign country block    

taudsss     = 0.01    

taumsss     = 0.01    

tauksss     = 0.01    

taunsss     = 0.01    

tauydsss    = 0.01          # steady state tax rate on output for foreign country block
    tauss       = 1.0               # Steady state terms of trade

icu = ((1+pidss)/beta) - 1 
mc = ((ej - 1)/ej)
r = (1/taukss) * ((1/beta)  - (1-delta))
rs = (1-tauksss) * ((1/beta)  - (1-delta))
KN = (mc*alpha/r)**(1/(1-alpha))
KNs = (mc*alpha/rs)**(1/(1-alpha))
psigma = (1-xi) * (1/(1-tauydss) - xi)**(-1)
psigmas = (1-xi) * (1/(1-tauydsss) - xi)**(-1)
w =     (1-alpha) * mc * (KN)**(alpha)

z = np.zeros(16)

def fun(z):
    Yd = z[0]
    N = z[1]
    X = z[2]
    I = z[3]
    Cd = z[4]
    Cm = z[5]
    Gd = z[6]
    Gm = z[7]
    Yds = z[8]
    Ns = z[9]
    Xs = z[10]
    Is = z[11]
    Cds = z[12]
    Cms = z[13]
    Gds = z[14]
    Gms = z[15]
    print (z)
    f = np.zeros(16)
    f[0]  = N - ( (X - muc)**(-ec) * ((1-alpha)/(mun)) * (mc)**(1/(1-alpha)) * (alpha/r)**(1-taunss) )
    f[1]  = Yd - ( Cd + Gd + I + ((1-n)/n) *(Cms + Gms)       )
    f[2]  = Yd - ( (KN)**(alpha)  * (psigma/(1-tauydss)**(ej)) )
    f[3]  = Cd - ( X * ((1-om)/(1+taudss)**(el)) *((1-om)*(1+taudss)**(1-el) + om * (1+taumss)**(1-el) * tauss**(1-el))**(el/(1-el))  )
    f[4]  = Gd - ( ((gss*Yd * (1-omg))/(1+taudss)**(eg) ) *((1-omg)*(1+taudss)**(1-eg) + omg* (1+taumss)**(1-eg) * tauss**(1-eg))**(eg/(1-eg) ) )
    f[5]  = I - ( delta* KN * N )
    f[6]  = Cm -( (X * (1-om)/(1+tauydss)**(el) ) *((1-om)*(1+taudss)**(1-el) + om* (1+taumss)**(1-el) * tauss**(1-el))**(el/(1-el))   )
    f[7]  = Gm - ( ((gss*Yd * (omg))/(1+taumss)**(eg) ) *((1-omg)*(1+taudss)**(1-eg) + omg* (1+taumss)**(1-eg) * tauss**(1-eg))**(eg/(1-eg) ) )
    f[8]  = Ns - ( (Xs - muc)**(-ec) * ((1-alpha)/(mun)) * (mc)**(1/(1-alpha)) * (alpha/rs)**(1-taunsss) )
    f[9]  = Yds - ( Cds + Gds + Is + (n/(1-n)) *(Cm + Gm)       )
    f[10] = Yds - ( (KNs)**(alpha)  * (psigmas/(1-tauydsss)**(ej) ) )
    f[11] = Cds - ( Xs * ((1-oms)/(1+taudsss)**(el))* ((1-oms)*(1+taudsss)**(1-el) + oms* (1+taumsss)**(1-el) * tauss**(1-el))**(el/(1-el))  )
    f[12] = Gds - ( ((gsss*Yds * (1-omgs))/(1+taudsss)**(eg) ) *((1-omgs)*(1+taudsss)**(1-eg) + omgs* (1+taumsss)**(1-eg) * tauss**(1-eg))**(eg/(1-eg) ) )
    f[13] = Is - ( delta* KNs * Ns )
    f[14] = Cms -( (Xs * (1-oms)/(1+tauydsss)**(el) ) *((1-oms)*(1+taudsss)**(1-el) + oms* (1+taumsss)**(1-el) * tauss**(1-el))**(el/(1-el))   )
    f[15] = Gms - ( ((gsss*Yds * (omgs))/(1+taumsss)**(eg) ) *((1-omgs)*(1+taudsss)**(1-eg) + omgs* (1+taumsss)**(1-eg) * tauss**(1-eg))**(eg/(1-eg) ) )
    return f

z = sp.optimize.root(fun, [100,100,70,30,50,20,50,20,100,100,100,100,100,100,100,100],  method='lm')

# z = fsolve(fun, [0,0,0.0,0,1,1,1,1,1,1,1,1,1,1,1,1])

print(z)

其解为以下根

success: True
       x: array([  3.64725445e-01,   1.02848541e-06,  -1.86761721e+02,
         9.52089296e-10,  -1.30733205e+02,  -1.25265418e+02,
         5.87207967e-02,   2.51660557e-02,   3.36422990e+00,
         5.18324506e-04,   8.17060628e+01,   4.87111630e-04,
         6.53648502e+01,   6.53648502e+01,   6.19018302e-01,
         1.54754576e-01])
3zwjbxry

3zwjbxry1#

给定一个根的初始估计值,数值求根算法在变量空间中沿着某个方向移动,直到找到一个根。显然,使用这种方法,要求返回的根在某个区间内是没有意义的a-这完全取决于初始估计值有多好(以及算法使用的搜索方法)。
另一种可以返回有界根的方法是将求根问题作为优化(例如最小化)问题,因为在优化问题中提供约束是有意义的。然而,你必须提供一个适当的目标函数,其最小值出现在原始函数的根上(对于这样的函数有很多选择,通常是启发式的选择)。
其中一个函数是平方和f[0]**2 + f[1]**2 + ... + f[15]**2。很明显,这个函数的最小值为零,当和的每一项都为零时,也就是在它们的根处。你可以使用Scipy的least_squares来执行这个最小化,它也允许为优化变量提供边界。
如果变量没有任何边界,并且使用相同的初始根估计值,least_squares将返回与root相同的解:
第一个
(Note z_ls.cost是在该点计算的平方和,其为(在数值精度内)零。)
现在,使用least_squares将估计值约束为非负:
第一个
返回的估计值确实有非负元素。但是,z_ls.cost(明显)大于零,表明该解 * 不是 * 根。这意味着以下两个结果之一:

  • 初始点不足以产生非负根。
  • 这个问题不存在非负根。

如果你对上述内容没有任何了解,你唯一能做的就是尝试不同的初始化值,并希望返回所需的根(直接通过root或如上所述的最小化公式)。

vh0rcniy

vh0rcniy2#

假设您要查找变量v的值,该值始终为正。
让求解器求出它的对数(log_v = log(v)),然后设置v = exp(log_v),这样当求解器在整个域中搜索log_v时,v将始终为正。

def fun(log_v):
  v = exp(log_v)
  ...[do the same math using v]

sol = fsolve(fun, (1,))
v = exp(sol[0])

相关问题