Suppose that one has a sufficient number of measurements to make an estimate of a measured quantity
y
y
y and report its absolute error,
±
δ
y
\pm\delta y
±δy. The absolute error
±
δ
y
\pm\delta y
±δy is represented on a Cartesian plot by extending lines of the appropriate size above and below the point
y
y
y.
If plotted on a logarithmic plot, however, absolute error bars that are symmetric on a
y
y
y vs.
x
x
x plot become asymmetric; the lower portion is longer than the upper portion.
This gives a misleading view of measurement precision, especially when measured quantities vary by several orders of magnitude. To represent error bars correctly on a log plot, one must recognize that the quantity being plotted, which we call
z
z
z, is different than the measured quantity
y
y
y.
z
=
log
(
y
)
z=\log(y)
z=log(y) The error
δ
z
\delta z
δz is
δ
z
=
δ
[
log
y
]
\delta z=\delta[\log y]
δz=δ[logy] On the assumption of small errors, a differential analysis can be used
δ
z
≈
d
z
=
d
[
log
10
e
⋅
ln
y
]
≈
0.434
δ
y
y
\delta z\approx dz=d[\log_{10}e \cdot\ln y]\approx0.434\frac{\delta y}{y}
δz≈dz=d[log10e⋅lny]≈0.434yδy The error
δ
z
\delta z
δz is thus given by the relative error in
y
y
y:
δ
z
≈
0.434
δ
y
y
\delta z\approx 0.434\frac{\delta y}{y}
δz≈0.434yδy The error bars now display correctly on a logarithmic plot.
Reference: https://faculty.washington.edu/stuve/log_error.pdf