Scipy nquad报错求助：integrand()需3个参数却传入4个

Let's break down the root cause of your error first, then walk through fixing it and other potential bugs in your code:

The Immediate Error: How nquad Handles args #

Your error comes down to how you're passing the args parameter to nquad. When you write args=(t_est), Python doesn't treat this as a tuple—it just evaluates to the list t_est itself. Since t_est has two elements, nquad will split them into separate arguments and pass them to integrand alongside your integration variables a and x. That means your function gets called with a, x, t0, t1 (four arguments) instead of the a, x, t_est (three arguments) it expects.

Fix 1: Correct the args Tuple #

Change args=(t_est) to args=(t_est,)—the trailing comma tells Python this is a single-element tuple, so nquad passes the entire t_est vector as one argument to integrand:

def integration(t_est):
    return integrate.nquad(integrand, [bounds_a(), bounds_x()], args=(t_est,))[0]

Other Critical Fixes for Your Code #

Beyond the TypeError, there are a few other issues that will prevent your code from running properly:

1. Define Missing Variables #

You reference means and sigmas but never initialize them. Based on your comments (p=5 covariates, Z distribution mean=0, sigma=1), add these at the top of your code:

p = 5
means = np.zeros(p)  # Mean of 0 for all 5 covariates
sigmas = np.ones(p)  # Standard deviation of 1 for all 5 covariates
z_sigma = 1.0  # Separate sigma for the Z distribution

2. Fix the Joint Probability Calculation #

np.multiply(px) throws an error because it needs at least two arguments. To compute the joint probability of your covariates, use np.prod() to multiply all elements of px:

Px = np.prod(px)  # Replace np.multiply(px) with this

3. Fix Jacobian Integration Logic #

You can't index jacobian_integrand[0] directly—instead, pass the entire function to nquad, which will handle integrating each element of the returned list:

def integration_jac(t_est):
    def jacobian_integrand(a,x,t_est):
        t0=t_est[0]
        t1=t_est[1]
        j0 = (t0 + t1 * np.sum(x))/np.power(sigmas,2)
        j1 = np.sum(x)*(t0 + t1 * np.sum(x))/np.power(sigmas,2)
        return [integrand(a,x,t_est) * j0, integrand(a,x,t_est) * j1]
    # nquad returns the integral of each element in the list
    jac0, jac1 = integrate.nquad(jacobian_integrand, [bounds_a(), bounds_x()], args=(t_est,))
    return [jac0,jac1]

4. Fix Hessian Integration Logic #

You have two issues here: integrand.nquad is a typo (it should be integrate.nquad), and you can't index hessian_integrand directly. Adjust it to handle the 2D return value correctly:

def integration_hessian(t_est):
    t0=t_est[0]
    t1=t_est[1]
    def hessian_integrand(a,x,t_est):
        # Fix extra plus signs and sigma squaring
        h00 = (1 + np.power((t0+t1*np.sum(x)),2))/np.power(sigmas,2)
        h01 = np.sum(x)*(1 + np.power((t0+t1*np.sum(x)),2))/np.power(sigmas,2)
        h11 = np.power(np.sum(x),2)*(1 + np.power((t0+t1*np.sum(x)),2))/np.power(sigmas,2)
        return [
            [integrand(a,x,t_est)*h00, integrand(a,x,t_est)*h01],
            [integrand(a,x,t_est)*h01, integrand(a,x,t_est)*h11]
        ]
    # Extract the 2D integral results
    hess_results = integrate.nquad(hessian_integrand, [bounds_a(), bounds_x()], args=(t_est,))[0]
    entry00 = hess_results[0][0]
    entry01 = hess_results[0][1]
    entry11 = hess_results[1][1]
    return [[entry00,entry01],[entry01,entry11]]

5. Fix Z Distribution Calculation #

Your original Z calculation uses sigmas (a length-5 array) which will cause dimension mismatches. Use the z_sigma we defined earlier for the Z distribution:

# Inside integrand()
Z = np.power(a,2)/np.sqrt(2*np.pi*np.power(z_sigma,2)) * np.exp(-np.power((a-t0-t1*np.sum(x)),2)/(2*np.power(z_sigma,2)))

With all these fixes, your code should resolve the TypeError and run correctly for your integration and optimization workflow.

内容的提问来源于stack exchange，提问作者Meep