CVXOPT can not find an optimal solution and quits with ValueError: domain error

[vr...@cornell.edu]

I have the following code:

def calculate_min_var(cov, lb, ub):

    """Calculate minimum variance portfolio subject to supplied constraints

    Inputs: covariance matrix, constraints (lb - lower bound, ub - upper bound)

    Output: asset weights

    Sum to 1 constraint is added automatically"""

    n= len(cov)

    P= cvx.matrix(cov)

    q = cvx.matrix(0.0, (n, 1))

    #G = cvx.matrix(-np.identity(n)) #x>=0

    G= cvx.matrix(np.concatenate((-np.identity(n),np.identity(n)), axis=0))

    #h = cvx.matrix(0.0, (n, 1)) #x>=0

    h= cvx.matrix(np.concatenate((-lb,ub), axis=0),tc='d')

    #A = cvx.matrix(1.0, (1, n)) #sum to 1

    A = cvx.matrix(1.0, (1, n)) #sum to 1

    b = cvx.matrix(1.0) #sum to 1

    slv.options['show_progress'] = True

    slv.options['abstol'] = 1e-7

    slv.options['reltol'] = 1e-6

               

    sol = slv.qp(P, q, G, h, A, b)

    arr= np.array(sol['x'])

    return arr

I call it using: 

t_opt_perc2= mtc.calculate_min_var(test_cov, lb_dec, ub_dec)

I have attached covariance matrix, lower bound and upper bound constraints that give me an error. (I tried scaling constraints and inputs by 100 - it did not help. Unconstrained case works fine (lb=0, ub=0))

This code is calculating constrained minimum variance portfolio with upper bound set to market cap weight * 1.2 (or any other k>1 for that matter) and lower bound set to market cap weight * 0.8 (or, once again any k less than 1). In my use case, lower and upper bounds are always symmetric (lower bound + upper bound = 2). 

This code quits with ValueError: domain error. I wasn't able to debug it myself. Covariance matrix is well conditioned.

Can please somebody explain to me what is going on and how I can fix it?

I would really appreciate your help.

Thanks in advance,

Vitaliy

[vr...@cornell.edu]

I managed to fix my own problem. Upper bound is less than 1, I hope it helps if someone experiences a similar issue

[Agnello Hupp]

I'm having the same problem.

Does the code remain the same?

Thank you for sharing