How do I get working y-axis errorbars using a log scale?

Question

Adam Goldsmith 2015-9-8

编辑： John Kelly 2017-11-10

I'm trying to plot data on a semilog plot (y-axis: log, x-axis: linear), but the errorbars are seriously screwed up.

All I'm doing is loading my data and entering:

figure
errorbar(hp_v3,hp_D0,hp_D0_,'ok')
set(gca,'YScale','log')

where hp_D0_ is the vector of uncertainty values for vector hp_D0. This results in the plot attached as an image, which obviously isn't working.

I've also tried errrobarlogy, which produces literally the same plot. Any ideas?

Answer 1

Mike Garrity 2015-9-8

That looks like what happens if the bottom of the errorbar is negative.

What do you get if you do this:

h = errorbar(hp_v3,hp_D0,hp_D0_,'ok');
h.YData - h.LData

Are the resulting values negative?

If so, what's happening is that the log transform of the negative value results in a complex value which it can't transform to a point on the screen.

Here's an attempt to recreate your data:

rng default
hp_v3 = 986:1002
hp_D0 = 1e-4*rand(1,17);
hp_D0_ = 1e-4*randn(1,17);
h = errorbar(hp_v3,hp_D0,hp_D0_,'ok')
set(gca,'YScale','log');

We can tweak the LData to stay positive like this:

ylim manual
h.LData = h.YData - max(eps,h.YData-h.LData);