convergence {lme4}R Documentation

Assessing Convergence for Fitted Models


[g]lmer fits may produce convergence warnings; these do not necessarily mean the fit is incorrect (see “Theoretical details” below). The following steps are recommended assessing and resolving convergence warnings (also see examples below):


Convergence controls

Theoretical issues

lme4 uses general-purpose nonlinear optimizers (e.g. Nelder-Mead or Powell's BOBYQA method) to estimate the variance-covariance matrices of the random effects. Assessing the convergence of such algorithms reliably is difficult. For example, evaluating the Karush-Kuhn-Tucker conditions (convergence criteria which reduce in simple cases to showing that the gradient is zero and the Hessian is positive definite) is challenging because of the difficulty of evaluating the gradient and Hessian.

We (the lme4 authors and maintainers) are still in the process of finding the best strategies for testing convergence. Some of the relevant issues are

See Also

lmerControl, isSingular


if (interactive()) {
fm1 <- lmer(Reaction ~ Days + (Days | Subject), sleepstudy)

## 1. decrease stopping tolerances
strict_tol <- lmerControl(optCtrl=list(xtol_abs=1e-8, ftol_abs=1e-8))
if (all(fm1@optinfo$optimizer=="nloptwrap")) {
    fm1.tol <- update(fm1, control=strict_tol)

## 2. center and scale predictors:
ss.CS <- transform(sleepstudy, Days=scale(Days))
fm1.CS <- update(fm1, data=ss.CS)

## 3. recompute gradient and Hessian with Richardson extrapolation
devfun <- update(fm1, devFunOnly=TRUE)
if (isLMM(fm1)) {
    pars <- getME(fm1,"theta")
} else {
    ## GLMM: requires both random and fixed parameters
    pars <- getME(fm1, c("theta","fixef"))
if (require("numDeriv")) {
    cat("hess:\n"); print(hess <- hessian(devfun, unlist(pars)))
    cat("grad:\n"); print(grad <- grad(devfun, unlist(pars)))
    cat("scaled gradient:\n")
    print(scgrad <- solve(chol(hess), grad))
## compare with internal calculations:

## compute reciprocal condition number of Hessian
H <- fm1@optinfo$derivs$Hessian

## 4. restart the fit from the original value (or
## a slightly perturbed value):
fm1.restart <- update(fm1, start=pars)
pars_x <- runif(length(pars),pars/1.01,pars*1.01)
fm1.restart2 <- update(fm1, start=pars_x,

## 5. try all available optimizers

  fm1.all <- allFit(fm1)
  ss <- summary(fm1.all)
  ss$ fixef               ## fixed effects
  ss$ llik                ## log-likelihoods
  ss$ sdcor               ## SDs and correlations
  ss$ theta               ## Cholesky factors
  ss$ which.OK            ## which fits worked


[Package lme4 version 1.1-32 Index]