See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/260828187

The rainbow

Article in Europhysics News · January 2006

DOI: 10.1051/epn:2006103

CITATIONS READS

2 1,743

3 authors, including:

John Ewan Inglesfield

Cardiff University
158 PUBLICATIONS 3,220 CITATIONS

SEE PROFILE

All content following this page was uploaded by John Ewan Inglesfield on 01 February 2016.

The user has requested enhancement of the downloaded file.

 features

The rainbow
Owen Davies, Jeff Wannell, and John Inglesfield,
School of Physics and Astronomy, Cardiff University, 5 The Parade, Cardiff • CF24 3YB • UK.

T he rainbow, with its arc spanning the sky, is a glorious sight

which never ceases to amaze. Rainbows have always been a
source of wonder [1], sometimes as a symbol of the gods, some-
From this geometry, the primary and secondary bows appear
as arcs making angles of 42° and 51° around the extension of a
line from the sun and passing through the observer’s head (figure
times as an evil omen, and have inspired artists (who often get it 4). As a consequence, when the sun is high in the sky the rainbow
wrong) and poets such as the Lakes poet, Wordsworth [2]: may appear against the ground. All rainbows have the same angu-
“ My heart leaps up when I behold lar size, whether they are due to a shower several miles away, or
A rainbow in the sky… ” the spray from the garden hose above the lawn.We may ask where
For physicists, part of the wonder of rainbows is the way that they we see the rainbow – do we see it at the drops, perhaps? From fig-
literally illustrate so many aspects of the nature of light: most ure 3 it seems that our eyes focus the rays contributing to the
obviously, breaking up white light into the spectrum of colours. rainbow at infinity, and the rainbow has only a direction rather
They also combine ray aspects of light, determining the angular than a position. There is another point of view (literally), which
size of the primary and secondary rainbows, and the wave nature also suggests that if the rainbow has a location, this is at infinity –
of light which produces weak “supernumerary” bows, often visible if we, as observers, move, the rainbow moves with us. This means
inside the main primary. Some of the properties which we shall that a stereoscopic view of the rainbow, with our two eyes, or the
discuss in this article are apparent in the rainbow shown figure 1, rangefinder of an old-fashioned camera, will place the rainbow at
in which we can just see the weaker secondary, some supernumer- infinity [1].
aries, and the property that the sky is light inside the rainbow and The ray theory of the rainbow can be neatly represented as polar
dark outside. plots of the scattered intensity in different directions after one, two,
or more reflections. To calculate these, we combine Fresnel’s for-
Rays of light mulae for the intensity of reflected and transmitted light [5] with
We all know of Newton’s experiment with a prism, breaking up a simple ray geometry for different angles of scattering. We obtain
beam of light into the spectrum of colours, and indeed it was New- the results shown in figure 5 for the scattered intensity for the two
ton who explained the way that the wavelength-dependence of polarizations of light. In these figures, we have light coming hori-
the refractive index of the raindrop produces the colours of the zontally from the left-hand side, incident on a raindrop at the
rainbow. But what is startling about the rainbow is the sheer centre; the left-hand diagram corresponds to light polarized with
intensity of the spectrum, a result of the way that the raindrop con- its electric field perpendicular to the plane of the diagram (s-polar-
centrates the different colours in different directions. And to isation), and on the right the electric field is in the plane
understand this we must go back to Descartes [1], who in 1637 (p-polarisation). The large lobe to the right of the drop in both fig-
described the paths of rays of light through the raindrop, using the ures represents light passing through the drop, refracted but
sine law for refraction which we know today as Snell’s law. (It is without any reflection. The primary bow corresponds to the singu-
not clear whether Descartes knew of the work of Snell; Newton larity in the scattering after one reflection at ±42° with respect to the
incorrectly credited de Dominis with the explanation of the rain- incident light, showing up very strongly in s-polarisation, but
bow, and was rather casual about Descartes’ real explanation [1].) much weaker in p-polarisation: the rainbow is strongly polarised.
Rays of light are refracted as they enter the drop, and are then The figure shows very clearly how light undergoing one reflection
reflected inside the drop at the air-water interface – once in the for-
mation of the intense primary rainbow, twice for the much weaker
. Fig. 1: Rainbow above the Lake District fells: the much weaker
secondary bow – and refracted a second time as they leave the
secondary bow is just visible, with reversed colours. Several
drop (figure 2). Further reflections are possible, but as we shall see, supernumerary bows, with alternating green and violet, can be
higher order rainbows are rather theoretical. Concentrating on the seen inside the brilliant primary. The sky is distinctly darker
single reflection, figure 3 shows the paths of different rays incident outside the primary.
on the drop: ray 1, incident towards the centre of the drop, is
reflected back along its own path, but rays hitting higher up the
drop are reflected with increasing angles between the incident and
reflected rays. This continues up to a maximum angle of about 42°
reached by ray 7, determined by simple geometry from the refrac-
tive index of water. Beyond this ray, appropriately called the
Descartes ray, the angle decreases. We have a maximum – the rays
near the Descartes ray emerge in almost the same direction – and
the raindrop scatters most light at an angle of 42° to the incident
light. Taking into account the variation of the refractive index n
with wavelength – the dispersion – gives a rainbow angle of 42.2°
for red light with n = 1.332 and 40.6° for violet with n = 1.343 (the
refractive indices are taken from a very useful web site, [4]). For
two reflections (figure 2), we have a minimum angle between the
incident and scattered ray giving the secondary bow with an
angular size of about 51°. Because the rays of light bend round on
themselves, as we see from figure 2, red is on the inside of the sec-
ondary rainbow, with violet on the outside.

europhysicsnews number 1 • volume 37 • 17

Article available at http://www.europhysicsnews.org or http://dx.doi.org/10.1051/epn:2006103
features

is back-scattered up to this maximum angle, and beyond ±42° no

light is scattered. At ±51° there is a singularity for scattering after
two reflections – the secondary bow – again strongly polarised.
This singularity is the opposite way round from the primary bow
singularity, with two reflections scattering some light beyond ±51°,
though this does not show up on the scale of the plots.
These singularities in the scattering intensity I as a function of
angle ϑ have the form I ∝ Iϑ - ϑ 0I-1/2, where ϑ 0 is the rainbow
2 angle [7]. To derive this we must be clear what we mean by I – as
usual in treating scattering, I(ϑ )dϑ is the intensity of light scattered
into a small range of angles dϑ (this is strictly speaking in two
dimensions, and in three dimensions there is an extra factor of
sinϑ ). It follows that together with the various reflection and trans-
mission coefficients, I contains the term Idy /dϑ I, where y is the
“impact parameter” of the ray of light incident on the raindrop,
the height above ray 1 in figure 3. As ϑ (y) is parabolic at the max-
imum corresponding to the Descartes ray, Idy /dϑ I varies like
Iϑ - ϑ 0I-1/2, hence the singularity in the scattering. Except for the
primary rainbow in s-polarisation, the singularity is too narrow to
be apparent in figure 5. We shall see shortly what happens to this
singularity when the wave nature of light is considered.
The light scattered up to ±42° produces a bright sky inside the
primary, the brightness increasing as the rainbow is approached.
This can be seen quite clearly in figure 1. As no light is scattered
for one or two reflections between ±42° and ±51°, the sky appears
dark in this range of angles – this phenomenon goes under the
name Alexander’s band, perhaps reminiscent of some dance band
from the 1920’s, but in fact named for the Greek philosopher
Alexander of Aphrodisias (the names get more and more unlike-
ly) [1]. Are there higher order rainbows, corresponding to three,
four and more reflections? From figure 5 we see that three reflec-
tions produce weak singularities at ±42° in the forward direction
(for n = 1.332), and though it is not clear, four reflections produce
still weaker singularities at ±43°, almost on top of the three reflec-
3 tions peak. In principle these correspond to higher order rainbows
around the sun, but the fact that the singularities are extremely
weak and lie inside the large forward scattering lobe means that
they must be practically invisible. Bernoulli thought that the sharp-
eyed lynx or eagle might discern these higher order bows [1] – alas,
it seems very unlikely [3]. Quite frequently we have been told about
“rainbows” visible around the sun – these are invariably ice crystal
halos, and on such occasions we refer our friends to Greenler’s
“Rainbows, Halos, and Glories” [3]. Very higher order bows can in
fact be measured in laser experiments [6].
Close to the singularity, the scattering in the primary bow is 96%
s-polarised, and in the secondary 90%. This polarisation results
from the fact that the angle at which the light is reflected inside
the drops is close to the Brewster angle, at which the reflection
4 coefficient for p-polarised light is zero [5]. Taking the refractive
index for water for green light as n = 1.335, the Brewster angle is
37°, and the angle of reflection of the Descartes ray for the prima-
ry bow is 40°.At this angle the ratio of the p to s reflectivities is 0.03.
m Fig. 2: Rays of sunlight refracted and reflected inside a raindrop: For the secondary bow the angle of reflection is 45°, giving a ratio
the left-hand figure, with one reflection, shows the paths which lead of the reflectivities of 0.26 at each of the two reflections. The s-
to the primary, and the right-hand figure, with two reflections, the
polarisation of the rainbow corresponds to the electric field vector
secondary.
m Fig. 3: Parallel rays of light incident on the raindrop, with one
being tangential to the bow, and consequently it is interesting to
reflection. Ray 7, the Descartes ray, emerges at the greatest angle, view the bow through Polaroid. The segment of the rainbow which
and rays pile up in this direction. (Figure from [3].) is tangential to the plane of polarisation of the Polaroid appears rel-
m Fig. 4: Drops at an angle of 42° to the line from the sun, passing atively brighter compared with the background sky, quite a striking
through the observer’s head, scatter sunlight to form the primary effect [8].
bow. Those drops at an angle of 51° scatter sunlight to give the The scattering intensity for single reflection in p-polarisation dis-
secondary. The diagram should be rotated about this line to form plays a curious angular variation within the primary rainbow
the complete bows. (Figure from [3].) singularity (figure 5). This is a consequence of the fact for a range

18 • volume 37 • number 1 europhysicsnews

 features

published in 1838 [9], following Young’s realization in 1804 that

interference causes the supernumeraries [10]. A wave-front which
we may use is the right-hand one in figure 7 and we use the axes
shown on the diagram, with the y-axis in the direction of the
Descartes ray. Then the Fresnel formula tells us that the amplitude
of the diffracted ray at a large distance from the drop, at angle ϑ to
the Descartes ray, is proportional to the — integral along the wave-
—
front, ∫ψexp[i2π(xsinϑ +y(x)cosϑ )/λ]√1+(dy/dx)2dx. Here, λ is the
wavelength of the light, and ψ is the amplitude of the electromag-
netic field over the wave-front, whose equation is y(x) [5]. The
exponential gives the phase of the contribution over the wave-
front, and the square root gives the length of the element of the
m Fig. 5: Polar diagram of scattering of light incident horizontally wave-front; positive ϑ corresponds to scattering outside the
from the left on a raindrop at the centre of the figures. Descartes ray. Why do we choose this particular wave-front for
(a) s-polarisation (electric field perpendicular to plane of figure); the Fresnel integral? We want to avoid the singularity of the later
(b) p-polarisation (electric field in plane). The numbers indicate the wave-fronts with cusps, and the wave-front where the cusp just fin-
number of reflections. The refractive index is taken as n = 1.332 ishes has a large amplitude right at the end, a point through which
(corresponding to red light). many classical rays pass. Moreover, this virtual wave-front has a
simple analytic form in the region which mainly contributes to
of angles inside the rainbow angle, there are two rays which con- the integral, y ≈ αx 3 – a result which Airy used to evaluate the
tribute to the intensity at each ϑ (15° < ϑ < 42°), each with its integral in terms of his famous function.
variation in reflectivity as the Brewster angle is approached. We In evaluating the diffraction integral, Airy assumed that the
doubt whether this scattering has been seen, as these lobes are very amplitude ψ was constant over the wave-front; neglecting the
weak compared with the s-polarisation rainbow. We should note square root for the length of wave-front, the diffraction amplitude
here that a scattering angle of 15°, the minimum angle for which for small angle ϑ is given by the Airy integral
∞
two rays contribute, corresponds to the incident ray which grazes Ψ(ϑ )∝∫ cos[2π(xϑ + αx 3)/λ]dx. This may be expressed in terms
the raindrop, the only ray for which total internal reflection occurs. 0
of the Airy function Ai, Ψ(ϑ )∝Ai(ϑ /y) , with y = (3α[— λ 2)1/3 – Airy
2π ]
Wave-fronts and waves evaluated his function by hand, but it is now immediately avail-
Rainbows produced by raindrops about 1mm in diameter or able in computer packages. The cubic coefficient α depends on
smaller often show several extra bands of colour, typically alternat- the refractive index, and for n = 1.335 it is given by α = 1.62/R2,
ing green and violet [8], inside the primary (figure 6) – these are where R is the radius of the raindrop [11]. The resulting diffraction
the supernumerary bows, produced by interference of the light intensity is shown in figure 8, for light of wavelength λ = 500 nm
waves [3]. The two rays which leave the drop for a range of direc- scattered by a drop of radius 0.5 mm, plotted as a function of
tions inside the rainbow angle have different path-lengths, and angle from the Descartes ray. We see that the singularity in ray
interfere with one another. To obtain quantitative results, we first theory is replaced by a finite peak, the primary bow in diffraction
construct the geometrical wave-front, a surface perpendicular to the theory, with a maximum at about 1/4° inside the Descartes ray. The
classical rays, on which the phase of the waves is constant. In the subsequent peaks constitute the supernumerary bows.
Huygens-Fresnel semi-classical approach, each point on the wave- The integrand in the Airy integral oscillates very rapidly as x
front is considered as a source of spherical waves, which interfere varies over the wave-front, when angle ϑ is negative, except where
with each other – this is not a full solution of the wave equations, the wave-front is perpendicular to the direction in which the ampli-
but is a good approximation when the wavelength is small com- tude is evaluated. For a range of angles inside the Descartes ray
pared with the dimensions of the object scattering the light [5]. there are two points at which the wave-front is perpendicular, cor-
Geometrical wave-fronts corresponding to light leaving the responding to two classical rays travelling in this direction. Rays
raindrop are shown in figure 7, the different curves corresponding AA’ and BB’ in figure 7 are two such rays, travelling at ϑ ≈ -5° from
to different phases of the waves, or different path-lengths the waves the Descartes ray, and at the points of intersection with the wave-
travel. The fronts which intersect the drop are, in fact, virtual, front a’ and b’ the front is perpendicular. Around these two points,
formed by extending the actual wave-fronts backwards through air
rather than through the drop. They correspond to the rays leaving
the drop extended backwards as straight lines. What we immedi-
ately notice are the cusps in the wave-fronts, which lie on the
Descartes ray (except close to the drop), and trace out a caustic.
(The bright patterns on the surface of the breakfast cup of tea
reflected from the kitchen spotlight are a familiar example of
caustics.) It is the interference between the wave-fronts on either
side of the cusps which give rise to the supernumeraries.
The next step is to use one of the wave-fronts as a source of
waves, to find the intensity as a function of scattering angle. This
calculation was first performed by Airy, described in a classic paper

c Fig. 6: Rainbow above Penyghent, North Yorkshire, with several

supernumerary bows inside the primary. The secondary bow is
barely visible.

europhysicsnews number 1 • volume 37 • 19

features

b Fig. 7: Wave-fronts of rays leaving the raindrop, n = 1.335, with

rays incident on the drop as in figure 3. The wave-fronts intersecting
the drop are “virtual”, extended backwards through air. DD’ is the
Descartes ray leaving the drop, and AA’, BB’ are the two rays leaving
the drop at a scattering angle inside the rainbow angle, which can
interfere. All these rays are extended backwards through air. Local
axes x and y are for the Fresnel integration over the wave-front.

again and again in physics, in particular as the solution of the

Schrödinger equation in a linear potential, and these approximate
expressions for the Airy function (actually the first terms in an
asymptotic expansion) are important in the mathematical analysis
of this equation [12].

From supernumeraries to fog-bows

It is remarkable that interference fringes show up as supernumer-
aries in the rainbow, when we consider that the light waves are
being scattered by raindrops 1000 times larger – we are used to
interference effects in scattering over length scales comparable with
the wave-length of light. Supernumeraries are even more remark-
the phase in the cosine integrand varies very little, and this domi- able when we consider that this effect of light waves appears in a
nates the integral. The method of stationary phase shows how the large-scale phenomenon, traversing the sky! The appearance of
integral may be determined in terms of the contributions from these these supernumeraries depends on the size of the raindrops scat-
two regions [12], and the result is that the amplitude is given tering the light, and we can explore this using the theory described
— above.
approximately by Ψ(ϑ ) ≈ cos[π/4 - –23 (Iϑ I/y)3/2]/[√ π(Iϑ I/y)1/4], ϑ
The intensity of light scattered by the raindrop at varying
negative. The corresponding intensity is shown by the dark blue angles in the primary bow is shown in figures 9a and 9b, for rain-
line in figure 8, and we see that apart from a very small range of drops of radius 0.3 mm and 0.05 mm respectively. On each figure,
angles as we approach the Descartes ray, this approximation is the three curves correspond to red, green and blue light. For rain-
amazingly accurate. The method of stationary phase gives the drops of radius 0.3 mm (figure 9a) we see that red and green give
interference pattern from the two rays travelling in direction ϑ , good strong principal peaks, with the first supernumerary of red
producing supernumeraries. Something surprising, which does not overlapping with the first peak of blue – these constitute the pri-
come out of a straightforward interference picture, is the phase mary bow, with the overlapping red and blue enhancing the violet.
shift of π/4 found in the stationary phase result given above, shift- Inside the primary we see supernumeraries, which are initially
ing the first maximum and the supernumeraries. Such phase shifts, alternating green and red + blue. It is a little dangerous to go from
and the general study of wave forms near ray caustics and singu- this figure directly to the actual appearance of the rainbow; for
larities, are a very active topic of current research, and there are this we should consider the scattering of the whole spectrum of
many papers by Berry and co-workers in this area [10]. visible light, and then use the trichromatic nature of colour vision
For positive angles, the method of steepest descents [12] may to work out the appearance of the rainbow [1]. But figure 9a, taken
be used to obtain an approximation to the Airy integral (this at face value, is consistent with the supernumeraries described by
method is the same as stationary phase if we go into the complex Minnaert in his classic book [8], as alternating “violet-pink” and
— green (the violet-pink comes from the superposition of the red and
plane), giving Ψ(ϑ ) ≈ exp[-–23 (ϑ /y)3/2]/[2√ π(Iϑ I/y)1/4], ϑ positive.
blue peaks). We can make out several supernumeraries just inside
This gives the intensity shown by the green line in figure 8, again the primary bow in figure 6, and at least on the original photograph
remarkably accurate beyond ϑ ≈ + 1/4 °. The Airy function crops up these are seen to be alternating violet and green. In nature, only a
few supernumeraries are ever visible – for one thing, raindrops are
. Fig. 8: Intensity of light, wave- unlikely to have a uniform size, and this varies the phase and wave-
length 500 nm, scattered by a length of the oscillations. Moreover, the sun has an angular
raindrop of radius 0.5 mm. Angles are diameter of about 1/2 °, again smearing out the supernumeraries.
measured from the Descartes ray. Red The supernumeraries will be less apparent with bigger drops, for
line: Airy theory; blue line, stationary which they are more closely spaced, and hence more likely to be
phase approximation for negative lost.
angles; green line, steepest descents With very small drops of radius 0.05mm, mist rather than rain,
approximation for positive angles. the principal peaks broaden and overlap completely (figure 9b).
The colours of the primary bow are completely smeared out – we
observe in this case a white rainbow, or fog-bow, the figure sug-
gesting that the supernumeraries should be quite strong. Some
photographs in the literature [1] and on the web do show super-
numeraries with fog-bows, but the only time that one of us
observed a complete white rainbow, only the principal bow was
visible, and this rather faint (figure 10). As there is no mist or fog
(the visibility of the snow-covered Helvellyn ridge is excellent), this
must be a cloud-bow, formed by droplets in the clouds.

20 • volume 37 • number 1 europhysicsnews

 features

b Fig. 9: Intensity of
light scattered by
raindrops of (a) radius 0.3
mm and (b) 0.05 mm.
The different curves
correspond to red, green
and blue light, and
angles are measured
from the Descartes ray
for green light.

Beyond the rainbow References

The theory of the rainbow which we have described explains [1] R.L. Lee, Jr. and A.B. Fraser, The Rainbow Bridge: Rainbows in Art,
everything in terms of classical rays of light, even Airy theory Myth, and Science (Pennsylvania State University Press, University
boiling down to interference between two rays of light travelling in Park, PA) (2001).
the same direction. However, these theories are not the whole story [2] W. Wordsworth, “My heart leaps up” (March 1802).
of light scattering by water droplets, and they cannot begin to [3] R. Greenler, Rainbows, Halos, and Glories (Cambridge University
explain another phenomenon involving light scattering from mist Press) (1980), now available from
– the glory, and the Brocken spectre [3,8]. The full scattering theo- www.blueskyassociates.com/books.html.
ry of light by a dielectric sphere – Mie theory – is needed to
understand the glory, and computer programs are available on the [4] www.philiplaven.com
web to explore this [4]. The beauty of the physics of rainbows is [5] M. Born and E. Wolf, Principles of Optics (Cambridge University
that so much can be understood in terms of rays and simple wave Press) (1999).
theory: rainbows open our eyes to some of the fundamental prop- [6] Jearl D. Walker, Am. J. Phys. 44, 421 (1976).
erties of light.
[7] M.V. Berry and S. Klein, Eur. J. Phys. 18, 303 (1997).
Acknowledgements [8] M.G.J. Minnaert, Light and Color in the Outdoors (Springer-Verlag,
The photographs in this article were taken by Whin Inglesfield, to Berlin) (1993).
whom we are very grateful. [9] G.B. Airy, Trans. Camb. Phil. Soc. 6, 379 (1838).
[10] M.V. Berry, Phil. Trans. Roy. Soc. Lond. A 360, 1023 (2002).
About the authors
Owen Davies and Jeff Wannell both obtained the MPhys Degree [11] H.C. van de Hulst, Light Scattering by Small Particles (John Wiley,
in Cardiff in 2001, and this article is based on their final year pro- New York) (1957).
ject dissertation, supervised by John Inglesfield. Jeff has since [12] H. Jeffreys and B.S. Jeffreys, Methods of Mathematical Physics
become a banker. Owen obtained a PhD in Cardiff in 2005, on the (Cambridge University Press) (1972).
theory of DNA, and now works in the City of London.
John Inglesfield received his PhD in 1970 from Cambridge, and is
a professor of physics in Cardiff. His research interests are the the- . Fig. 10: Cloud-bow on a winter’s day above the Grasmere fells,
ory of surface electronic structure, and most recently DNA and Lake District. As an aid to the eye (the bow is faint), it appears as
photonics. an almost complete semicircle, spanning the double photograph.

europhysicsnews number 1 • volume 37 • 21

View publication stats