OK, here's an example, which I plotted to a 10 by 10 inch pdf. (Part of what's frustrating about using par(fig = )
et al. is that their effects are very much dependent on the size of the plotting device.)
Edited to add some explanation:
The base graphic plotting parameter par("fig")
describes/sets the location of a figure region as a proportion of the "drawing region" (which is usually the entire device, for single figure plots). It takes a length-4 vector of the form c(xmin, xmax, ymin, ymax)
consisting of numbers (proportions) between 0
and 1
.
Here I use grconvertX()
and grconvertY()
to convert x-y locations expressed in terms of the larger plot's own (a.k.a. "user"
) coordinate system into the "ndc"
(normalized device coordinates) coordinate system. The "user"
coordinate system is more human-user-friendly, while the "ndc"
is (with the caveats expressed above) the coordinate system used by par("fig")
. The grconvert*()
calls are just there to perform the translation between them.
## pdf("fig-in-fig.pdf", width=10, height=10)
curve(exp(x), from=1, to=5, lwd=5)
curve(150-exp(x), from=1, to=5, lwd=5, col="darkblue",add=T)
## Here's the bit I added.
par(fig = c(grconvertX(c(1, 3), from="user", to="ndc"),
grconvertY(c(50, 125), from="user", to="ndc")),
mar = c(4,6,1,1),
new = TRUE)
curve(exp(x), from=1, to=5, lwd=7, xlab="chi", ylab="exp(x)", cex.lab=4,axes=F)
axis(1, labels=NA,at=c(0,5))
axis(2, labels=NA,at=c(0,150))
text(1,120,"Alpha",adj=c(0,0),cex=3)
text(3.5,10,"Beta",adj=c(0,0),cex=3)
## dev.off()