seria bom mudar a escala dos dados para evitar problemas numericos ### estima para y/1000000 o <- optim(c(1.2, 1), function(p) -sum(dweibull(y/1e6, p[1], p[2], log=TRUE)), method='BFGS', hessian=TRUE) ### estima para 1000 reamostras (bootstrap) oo <- t(replicate(1000, optim(o$par, function(p) -sum(dweibull(sample(y, replace=TRUE)/1e6, p[1], p[2], log=TRUE)), method='BFGS')$par)) (tb <- rbind(sd.assintotico=sqrt(diag(solve(o$hess))), sd.bootstrap=apply(oo,2,sd))) ### visualiza estimativas bootstrap e adiciona SD assintotico par(mfrow=c(2,2), mar=c(3,3,1,1), mgp=c(2,1,0)) hist(y/1e6, main='dados') plot(function(x) dweibull(x, o$par[1], o$par[2]), 0, 2, add=T, col=2) hist(oo[,1], main='shape') abline(v=o$par[1]+c(-2,0,2)*tb[1,1], col=2, lty=c(2,1,2)) hist(oo[,2], main='scale') abline(v=o$par[2]+c(-2,0,2)*tb[1,2], col=2, lty=c(2,1,2)) Elias