I have function like

gmmparametric <- function(theta){
  
  s1 <-  etahatparametricmat[,2] - theta[1]*logk[,2] - theta[2]*logm[,2]
  s2 <-  etahatparametricmat[,1] - theta[1]*logk[,1] - theta[2]*logm[,1]
  
  olsmodel <- lm(s1 ~ s2)
  olspar <-  olsmodel$coefficients
  
  implomega <- olspar[2]
  implcte <- olspar[1]/(1-2*implomega)
  
  
  matmoment1 <- logy[,2] - implcte - theta[1]*logk[,2] - theta[2]*logm[,2] -implomega*etahatparametricmat[,1] + implomega*(implcte + theta[1]*logk[,1] + theta[2]*logm[,1])
  matmoment2 <- logy[,3] - implcte - theta[1]*logk[,3] - theta[2]*logm[,3] - implomega*etahatparametricmat[,2] + implomega*(implcte + theta[1]*logk[,2] + theta[2]*logm[,2])
  
  matmoment <- cbind(matinst1*matmoment1, matinst2*matmoment2)
  
  meanmatmoment <- colSums(matmoment)/n
  
  objectfunc <- t(meanmatmoment)%*%diag(8)%*%meanmatmoment
  
  objectfunc
}

The function is essentially computing a quadratic expression or a GMM objective function-like, but before do it, it “concentrates out” two parameters that can be obtained from an OLS problem, which depends on the variable theta, the unknown that I want to determine by minimizing gmmparametric. I want to minimize gmmparametricusing CVXR.

My intuition was to do something like

theta <- Variable(2)
object <- Minimize(gmmparametric(theta))
prob <- Problem(object)
prob.solve()

But I get the error

Error in model.frame.default(formula = s1 ~ s2, drop.unused.levels = TRUE) : 
  object is not a matrix

which I believe comes from the fact that OLS cannot be performed inside gmmparametric as theta is not an actual vector. So I was wondering if there is a way such that I can minimize gmmparametric using CVXR in this setting. I can use optim straightforwardly, but I do not how to proceed with CVXR.