# Problem with implementing squared terms in the objective function

I'm trying to implement either one of these objective functions, but I'm having a hard time with the squared terms. I'm attaching both so you can take a look at the structure and see if you can give me any tips. Is there any way to implement either one of them?

1- Matrix notation: $$x$$: decision variable

$$1$$: column of ones

$$k$$: squared matrix

2- Summation notation: $$x$$: decision variable $$m$$: degree of the node i $$rho$$: parameter that takes into account the influence of the neighbors that surround node i $$a$$: terms of the adjacency matrix. Shows if nodes i and j are connected

• Hello and welcome to OR.SE. What are the $a_{il}$s and $m_i$s? Jul 10, 2020 at 19:18
• Actually it would be good if you could explain all the notations... Jul 10, 2020 at 19:28
• In what way are you having a hard time? Is the question how to code squared terms with either CPLEX or Gurobi (in which case you should specify what API you are using)?
– prubin
Jul 10, 2020 at 20:26
• @OguzToragay thank you for reaching out! I will edit the post with all the notations. Jul 12, 2020 at 0:58
• @prubin thanks for reaching out! yes, my problem is I don't know how to implement the second term of the objective function in the matrix notation since the term is squared. On the other hand, I tried implementing the second equation of the post, but I run into a memory error. I'm using Gurobi with Python. Jul 12, 2020 at 1:05

It's relatively easy to write $$(1^{T}Kx)^{2}$$ in standard quadratic form.
Since $$1^{T}Kx$$ is a scalar,
$$(1^{T}Kx)^{2}=(1^{T}Kx)(1^{T}Kx)^{T}=1^{T}Kxx^{T}K^{T}1$$.
$$1^{T}Kxx^{T}K^{T}1=\mbox{tr}(1^{T}Kxx^{T}K^{T}1)=\mbox{tr}(x^{T}K^{T}11^{T}Kx)=x^{T}(K^{T}11^{T}K)x$$.
Unfortunately, $$K^{T}11^{T}K$$ will be dense, so if $$x$$ is large you'll probably run out of storage.