Sympy中nsolve无法求解非线性方程组不动点问题求助

尝试用SymPy的nsolve求解非线性方程组以确定不动点，提供初始猜测后始终报错“nsolve failed”；改用solve函数返回空列表sol=[]。尝试用解析方法求解，但不确定方法是否正确。相关代码如下：

f1_num = sp.N(f1.subs(subs))            
f2_num = sp.N(f2.subs(subs))
f3_num = sp.N(f3.subs(subs))


initial_guess = (g1_val, g2_val, 0.0)   # choose g3 guess 0

# Attempt to solve for g1,g2,g3
try:
    sol = sp.nsolve([f1_num, f2_num, f3_num], (g1, g2, g3),
                    (g1_val, g2_val, 0.02), tol=1e-12, maxsteps=200)
    results.append(sol)
except Exception as e:
    results.append(None)
    print(f"Column {j}: nsolve failed ({e})")
# fallback: try a few different initial guesses
for guess in [(g1_val, g2_val, 0.4), (1.0, 1.0, 0.5), (0.5, 0.5, 0.6)]:
    try:
        sol = sp.nsolve([f1_num, f2_num, f3_num], (g1, g2, g3), guess)
        print("sol=", sol)
        print("Found sol with guess", guess, ":", sol)
        break
    except Exception as e2:
        print("guess", guess, "failed:", e2)

g1_fp1 = [None] * num_cols
g1_fp2 = [None] * num_cols
g2_fp = [None] * num_cols

#.............................
    # Analytical solution
#............................. 
for j in range (num_cols):
   
    try:
        roots_f3 = sp.nsolve(f3_num, g1)
        if roots_f3:
            g1_fp1[j] = float(sp.re(roots_f3[0]))
            print(f"g1 roots from f3=",g1_fp1[j])
        else:
            g1_fp1[j] = np.nan
    except Exception as e:
        print(f"f3 solution failed ({e})")
        g1_fp1[j] = np.nan
#type(sol_f3)
    try:
        roots_f1 = sp.nsolve(f1_num, g1)
        if roots_f1:
            g1_fp2[j]= float(sp.re(roots_f1[0]))
            print(f"g1 roots from f1=",g1_fp2[j])
        else:
            g1_fp2[j] = np.nan
    except Exception as e:
        print(f"f1 solution failed ({e})")
        g1_fp2[j] = np.nan
    try:
        roots_f2 = sp.nsolve(f2_num, g2)
        if roots_f2:
            g2_fp[j] = float(sp.re(roots_f2[0]))
            print(f"g2 roots from f2=",g2_fp[j])
        else:
            g2_fp[j] = np.nan
    except Exception as e:
        print(f"f2 solution failed ({e})")
        g2_fp[j] = np.nan

问题排查与解决建议 #

• 不要提前数值化符号表达式：sp.N()会把符号表达式转为数值对象，nsolve更适合处理纯符号形式的方程组，数值化后可能丢失关键符号信息，导致求解失败。应保留符号表达式，求解时再代入参数，直接用符号方程组调用nsolve。
• 验证方程组是否存在解：先手动代入初始猜测计算每个方程的残差，判断初始点是否靠近解；也可以用Matplotlib绘制方程的曲线，直观查看是否存在交点。
• 调整数值求解参数：
  • 增大maxsteps值，默认迭代次数可能不足以让算法收敛；
  • 先放宽tol阈值找到粗略解，再以此为初始点进行精细求解；
  • 指定不同的求解算法，比如nsolve(..., method='lm')（Levenberg-Marquardt算法），适配不同类型的非线性方程组。
• 修正解析求解逻辑：当前单独对每个方程求解单个变量的方式错误，方程组的解需要满足所有方程同时成立，单独解单方程得到的根不是方程组的解。正确的解析求解应使用符号消元法，比如从一个方程解出某变量，代入其他方程逐步简化。
• 检查符号变量与替换参数：确保g1、g2、g3是SymPy符号变量（sp.symbols('g1 g2 g3')），且subs中的替换没有误替换掉需要求解的变量。

内容的提问来源于stack exchange，提问作者user5830353