Hg.conc<- c(2.5,22,60,90,105,144,178,210,233,256,300,400)

Y.gammaGT<-c(13.57,15.33,16.92,16.08,17.14,18.85,18.63,19.30,25.77,
             28.44,35.46,45.26)
mercury.df<-
   data.frame(Hg.conc=Hg.conc,log.conc=log(Hg.conc),resp=Y.gammaGT)

 Hg.nls.fit<- nls(resp~g0+b1*(log.conc-tau)*(log.conc>tau),
                  data=mercury.df,
                  start=list(g0=18.63,b1=38.88,tau=5.1818),
                  trace=T)
 summary(Hg.nls.fit)
 Formula: resp ~ g0 + b1 * (log.conc - tau) * (log.conc > tau)


 Formula: resp ~ g0 + b1 * (log.conc - tau) * (log.conc > tau)

 Parameters:
        Value Std. Error  t value
  g0 16.64570  0.6254910  26.6122
  b1 39.26700  3.2909700  11.9318
 tau  5.24608  0.0390842 134.2250

 Residual standard error: 1.65489 on 9 degrees of freedom

 Correlation of Parameter Estimates:
           g0       b1
  b1 8.98e-09
 tau 4.08e-01 7.75e-01


 # - now refit with the gradient provided I

 hockey<- function(g0,b1,tau,x) {
   indicate.gt.tau <- as.numeric(log.conc>tau)
   model.func<- g0 + b1*(x-tau)*indicate.gt.tau
   Temp<- cbind(1,(x-tau)*indicate.gt.tau,-b1*indicate.gt.tau)
   dimnames(Temp) <- list(NULL, c("g0","b1","tau"))
   attr(model.func,"gradient")<-Temp
   model.func
 }

 Hg.nls.fit.g<- nls(resp~hockey(g0,b1,tau,log.conc),
                  data=mercury.df,
                  start=list(g0=18.63,b1=38.88,tau=5.1818),
                  trace=T)
 summary(Hg.nls.fit.g)
 Formula: resp ~ hockey(g0, b1, tau, log.conc)


 Formula: resp ~ hockey(g0, b1, tau, log.conc)

 Parameters:
        Value Std. Error  t value
  g0 16.64570  0.6254910  26.6122
  b1 39.26700  3.2909700  11.9318
 tau  5.24608  0.0390842 134.2250

 Residual standard error: 1.65489 on 9 degrees of freedom

 Correlation of Parameter Estimates:
           g0       b1
  b1 5.55e-17
 tau 4.08e-01 7.75e-01