The geometry of sound rays in a wind

G. W. Gibbons1 and C.M. Warnick1,2

1. D.A.M.T.P., Cambridge, Wilberforce Road, Cambridge CB3 0WA, U.K.

2. Queens’ College, Cambridge, CB3 9ET, U.K.

November 2, 2018

Abstract

We survey the close relationship between sound and light rays and geometry. In the casewhere the medium is at rest, the geometry is the classical geometry of Riemann. In thecase where the medium is moving, the more general geometry known as Finsler geometry isneeded. We develop these geometries ab initio, with examples, and in particular show howsound rays in a stratified atmosphere with a wind can be mapped to a problem of circlesand straight lines.

1 Introduction: Propaga-tion of Waves in MovingMedia

Almost everyone must be familiar, by report atleast, with optical mirages. In hot regions suchas the desert, when the speed of light is greaterat low altitudes than at higher altitudes, dis-tant palm trees can appear inverted as if re-flected off a cool pool of water in a nearby oasis,see Figure 1. It is less well known that in coldregions such as the arctic, where the oppositeconditions prevail, distant ships can appear in-verted in the sky, the light from them havingbeen bent over an intervening iceberg, see Fig-ure 2. More mundanely, motorway travellerson hot dry summer days often have the discon-certing impression that there are sheets of wa-ter lying on the road some distance ahead. Theexplanation of phenomena like this is easily un-derstood using the concept of light rays subjectto Snell’s Law for a stratified medium. Alter-natively we may apply Huygens’s wave theory

according to which, in passing over the iceberg,the higher parts of the wave front move fasterthan the lower parts causing it to bend down-wards.

Mutatis Mutandis similar phenomenon canapply to sound waves whose speed increaseswith increasing absolute temperature abovezero, T , as T

12 . During a warm sunny day

therefore, when the temperature is typicallyhotter near the ground due to the sun’s rays,the sound waves should be bent upwards.On a clear night, when the temperature nearthe ground drops rapidly by radiative cooling,sound waves should be bent downwards allow-ing distant sources, such as cars on a motorway,to be heard much more clearly than during theday. One of us, living as he does a kilometre orso away from the A14, which is, judged by theamount of traffic it carries effectively a motor-way, had until recently always supposed that itwas this temperature effect that was responsi-ble for the din experienced at times when con-templating the night sky in his garden. How-ever, the temperature effect should be isotropic,

Emails: g.w.gibbons@damtp.cam.ac.uk, c.m.warnick@damtp.cam.ac.ukPre-print no. DAMTP-2011-11

1

arX

iv:1

102.

2409

v1 [

gr-q

c] 1

1 Fe

b 20

11

Figure 1: A 19th century woodcut showinga mirage in the desert. Source: ‘Elements de

Physique’, Ganot

Figure 2: Etching of a sketch by the explorerWilliam Scoresby, showing an arctic mirage.Source: ‘The life of William Scoresby’, Scoresby-

Jackson

in other words it should affect the noise of carscoming from all directions and including thoseon less busy but closer city roads. The greatervolume and speed of traffic on the A14 aloneshould not be so overwhelming.

The obvious directional influence is that ofthe wind. However, the speed of sound vs ≈1250 km per hour is so much greater than typi-cal wind speeds v ≈ 30 km per hour that simpleconvection of sound waves by the wind cannotbe responsible for any significant directional ef-fect. As pointed out to one of us by a colleague,Hugh Hunt, who made a similar point in theNew Scientist of 15th April 2009 in response toreaders’ queries, it is not the wind velocity, butits gradient, that is its variation with height,called wind shear or more technically vorticitywhich is important. As is clear from observingclouds, wind speeds, while remaining roughlyhorizontal and much slower than the speed ofsound, increase quite sharply with height, whileremaining roughly horizontal. Thus it is notjust the increased velocity |v| that matters, butits gradient or vorticity ω = curlv.

Of course, examination of the literatureshows that this is not a new observation andmany papers and textbooks contain a simple

qualitive discussion. The earliest we have dis-covered dates from 1857 [1] and is due GeorgeGabriel Stokes (1819-1903) who was appointedto the Lucasian Chair of Mathematics in Cam-bridge in 1849 and held it until his death 54years later. His explanation, elaborated on byOsborne Reynolds (1842-1912) in 1874 [2], fol-lowed very closely Huygens’s explanation for re-fraction by a gradient in the refractive index. Ifthe wind direction is towards us, and the windspeed increases with height, the higher parts ofthe wave front move faster than the lower partsand the wave front is bent over towards us. Ifon the other hand the wind direction is in theopposite direction, the wave front will be bentupwards and hence away from us. The net ef-fect is that if the vorticity ω is non zero, thesound rays are deflected in an analogous wayto the deflection of a charged particle of massm and charge e by the Lorentz Force due tomagnetic field B. In fact this is not just ananalogy but, as we shall see shortly, a precisecorrespondence for low wind speeds:

ω ≡ − e

mB . (1)

Note that since it is the velocity gradient thatmatters, it is not the direction of the wind at

2

Figure 3: The cycloid, as traced out by a point on the circumference of a rolling circle

the ground or at a height above , but their dif-ference which is important. In practice how-ever, the speed of the wind is almost alwaysmuch slower near the ground than above, andso usually it is the velocity at higher altitudeswhich matters. Thus, in principle there are twoeffects acting, and which is the more importantdepends on which induces the greater curvatureκ to the sound rays. As Stokes and Reynoldsrealised κ ≈ 1

n∂n∂z = 1

21T∂T∂z for the thermal ef-

fect and κ ≈ 1vs∂v∂z for the wind shear, where

n is the index of refraction. It is probably nocoincidence that Stokes’s work followed shortlyafter the first demonstration, by the Germanphysicist Karl Friedrich Julius Sondhauss (1815- 1886) in 1853 [3], that a balloon filled withCO2 will act as a sound lens, focussing thesound of a ticking watch so as to render it au-dible some distance away.

Examples of the interplay of these two ef-fects, which can cause sound to travel over largedistances, abound. In early June, 1666, duringthe war between the Dutch and the English,both Samuel Pepys and John Evelyn reportedin their diaries that while the sound of gun firefrom ships off the coast of Kent could be heardclearly in London, it was not audible at all atthe ports of Deal or Dover. As Pepys at thetime observed:

“This . . . makes room for a greatdispute in Philosophy: how weshould hear it and not they, thesame wind that brought it to us be-ing the same that should bring it tothem.”

Infrasound (sound of very low frequency) fromthe volcanic explosion of Krakatoa on Au-gust 27 1883 was heard to travel several timesaround the earth. During the first world war,

the noise of the very large guns on the WesternFront was often audible within a range of a 100km or so, and often beyond 200 km, but notwithin a “zone of silence” between 100 and 200km. On September 21, 1921 there occured anenormous explosion at Oppau on the Rhine andthe same phenomenon was observed, see Figure4.

One explanation for some of these phenom-ena is reflection of sound from a layer of air inthe upper atmosphere with a higher sound ve-locity. The dependence of the velocity of soundin the atmosphere follows its temperature pro-file. In the absence of wind, the temperature(and hence velocity of sound) decreases undernormal circumstances up to a height of about10 km, remains constant up to roughly 25 km,then incresases to around 50 km, where it hasa local maximum, and then has a local mini-mum at about 80 km, see figure 5. A simpleapplication of the law of refraction

n(z) sin θ(z) = constant , (2)

where the local speed of sound vs(z) ∝ 1n(z) , re-

veals that sound rays can bounce off the max-ima of vs(z) but can also be trapped at theminima. Presumably the latter effect was re-sponsible for the long distance propagation ofinfrasound from Krakatoa. A similar phenom-ena occurs for sound waves in the ocean. Thespeed of sound initially decreases with depthand then increases exhibiting a minimum valueat a depth of around 1 km at the so-called SO-FAR channel. This allows whales to communi-cate over very large distances [8, 9] and moresinisterly, submarines to snoop on other sub-marines.

For simple profiles n(z), the law of refrac-tion (2) allows simple solutions for the rays.Thus if n(z) ∝ z, one has catenaries with the

3

Figure 4: The explosion at Oppau, 1921.Filled circles represent locations where the ex-plosion was heard and empty circles locationswhere nothing was heard

Figure 5: The variation of temperature andsound speed with altitude

Source: ‘Strange sounds in the Atmosphere: Part I’, R. K. Cook, Sound 1, 12-l6 (1962).

horizontal axis z = 0 as the directrix. Thisgives a rough description of mirages in thedesert. If n(z) ∝ 1√

z, the rays are cycloids, the

curve traced out by the point on the circumfer-ence of a circle rolling without slipping alongthe horizontal axis (see Figure 3). One canimagine this as the path of a glow-worm sittingon the rim of a bicycle wheel as it rolls along inthe dark. This could decsribe the rays passingover icebergs in the arctic. More importantlyfor what follows later, this can also be achievedby assuming that n(z) ∝ 1

z , in which case therays are semi-circles centred on the horizontalaxis. The addition of the effects of wind com-plicates considerably this simple picture.

Considering for a moment fundamentalphysics, one of the clearest trends in researchover the last 100 years or so has been what onemight call the geometrisation of physics. Thisis by no means a new phenomenon. Plato’s as-sociation of the regular, or Platonic, solids withthe classical elements is perhaps the earliest at-

tempt to explain the world through geometricalintuition. Geometry, in the modern sense ofdifferentiable manifolds, really caught hold inphysics in 1915 with Einstein’s general theoryof relativity. Since then, the use of geometryhas accelerated. Modern string theory, withits 10 space-time dimensions and complicatedinternal 6-dimensional Calabi-Yau manifolds isperhaps the clearest example of this geometri-sation process.

A geometrical approach has much to recom-mend itself, even to describe the more concretephysics we have thus far been considering. Inthis article, we will discuss how the propertiesof sound and light rays can be considered in ageometrical light. We will be interested in thebehaviour of solutions of the wave equation1 ina moving medium:

[( ∂∂t−W i ∂

∂xi

)2

− c2hij ∂2

∂xi∂xj

]u(x, t) = 0 .

(3)

1Throughout this paper we will use the convention that indices i, j = 1 . . . n which are repeated should besummed over

4

Here hij is related to the speed of sound ina direction parallel to the unit vector mi byvm = c(hijmimj)

−1/2 and W i is a vector givingthe velocity of the medium at each point. Wewill be interested in particular in disturbanceswith short wavelength, which move along rays.We will start by considering the case of a staticmedium, W i = 0. The geometry of the rays inthis case is well known and we shall describesome of its properties and give some examples.We will then move on to the case of a mov-ing medium and demonstrate that this can alsobe described geometrically, as we showed in [5].We will have to loosen slightly our notion ofa geometry in order to do so but the type ofgeometry which arises, a Finsler geometry, isvery natural. In fact, this geometry first arosein our studies of light rays near rotating blackholes. The link between the effects of gravityon light rays and refraction of light and soundwaves by media can be made explicit and isthe basis for much current work on optical andacoustic black holes. These analogue modelsfor black holes allow experimentation in a lab-oratory, which would of course not be possiblewith real black holes. For this reason, a thor-ough understanding not just of the mechanismsof refraction, but the geometry of refraction isof great relevance both for terrestrial and celes-tial physics.

2 Geodesics and Fermat’sPrinciple

2.1 Optical and Acoustic Geome-tries

The elementary theory of mirrors and lensesis largely concerned with tracing the paths oflight rays on reflection (Catoptrics) or refrac-tion (Dioptrics) at a surface. Although theGreeks were uncertain whether light proceedsfrom the eye to the object (emission theory) orfrom the object to the eye intromission theory,and whether its passage is instantaneous or at afinite speed, nevertheless Heron of Alexandria(c.10-70 AD) was able to formulate the lawsof reflection in terms of a Principle of Short-

est Length from object to eye via the reflectingsurface. The occurrence of caustics shows thatthere can be more than one path, not all ofwhich are necessarily the shortest and more ac-curately we refer to the principle Principle ofStationary Length, that is the length of eachray is merely stationary among all neighbour-ing paths.

Despite important pioneering work by Ibnal-Haytham or Alhazen (965-c.1040) demolish-ing the emission theory and investigating re-fraction, it took longer to unravel the funda-mental law of Dioptrics. It it was not un-til independent work by Abu Sa‘d al-‘Ala’ ibnSahl (c.940-c.1000), Thomas Harriot (c.1560-1621) and Willebrord Snel Van Royen (1580-1626) that the familiar Law of Sines was es-tablished and Pierre Fermat (c 1601- 1665) for-mulated his unified Fermat’s Principle of Sta-tionary Time. The idea is that the slowness ofthe ray inside a medium is proportional to itsrefractive index n. Note that the finite speedof light was only demonstrated by by Ole Roe-mer (1644-1710) using the eclipse of Jupiter’smoon Io in 1676. The relation of the refractiveindex to the speed of light remained contro-versial until experiments by Armand HippolyteLouis Fizeau (1819-1896), Fresnel Augustin-Jean Fresnel (1788-1827) and George BiddellAiry (1801-1892) in the nineteenth century fi-nally established its velocity as c

n , where c is itsspeed in vacuo. According to the discreditedcorpuscular theory the opposite relation holds.Both theories give the same rays but the speedwith which light follows the rays differs. Itwould be more accurate therefore to speak ofFermat’s Principle of Stationary Optical PathLength , where the optical length of a path γ isgiven by

L =

∫γ

n(x)|dx| , (4)

where we have allowed for the possibility thatthe refractive index may depend upon position,as for example it does in a vertical stratifedmedium such as we encounter discussing mi-rages. Fermat’s principle becomes the Varia-tional Principle

δL = δ

∫γ

n(x)|dx| = 0 . (5)

5

By the time Fermat introduced this principle,Christiaan Huygens (1629-1695) had initiatedhis wave theory of light and derived Fermat’sPrinciple from it. His derivation makes it clearthat it is the optical length or optical distancewhich enters into all interference effects, andits therefore appropriate to say that all opti-cal measurements measure Optical Geometry.By the same token, measurements using soundwaves may be said to measure Acoustic Geom-etry and measurements using seismic waves, ason the earth or more recently the moon [10] tomeasure Seismic Geometry.

It is clear that these geometries will, in aninhomogeneous medium for which the depen-dence of n on x is non-trivial, differ consid-erably from Euclidean Geometry. The exis-tence of such non-Euclidean Geometries wasfirst realised by pure mathematicians in theearly part of the nineteenth century workingon the foundations of geometry. For centuries,people had been attempting to derive, start-ing from the other aximons, Euclid’s fifth ax-iom: that through any point not on a given linethere is exactly one line parallel to the first.This seemed to them so obvious that it was“neccessarily” true. Eventually they gave up,Johann Carl Friedrich Gauss (1777-1855) pri-vately and Janos Bolyai (1802 -1860) and Niko-lai Ivanovich Lobachevsky (1792-1856) pub-licly, showed the existence of two other typesof homogeneous and isotropic Congruence Ge-ometries, Spherical and Hyperbolic. The firstis easy to grasp in two dimensions since it isjust the geometry of the standard sphere, S2

with the stationary paths (or geodesics) beingthe great circles. Navigators, either by sea orair, have been using spherical geometry sinceat least the time of Columbus. It is not toodifficult to imagine a sphere in one dimensionhigher and indeed if the refractive index wereto vary as

n =n0

(1 + x2

4R2 )(6)

the optical metric would be precisely that ofthree dimensional spherical space S3 of radiusn0R. The resulting optical device is known asMaxwell’s Fish Eye Lens since all rays emanat-ing from any point xe are circles which recon-

verge onto the antipodal point xe = xe

|xe|2 . We

can verify that the optical distance along a ra-dial geodesic is given by:∫ ∞

0

n0

1 + r2

4R2

dr = πn0R, (7)

which is finite. Because, for large enough |x|,the refractive index n drops below unity, theconstruction of such a device would require themanufacture of a suitable “meta-material”. Amore practical device was invented by RudolphKarl Luneburg (1903-1949) and has

n =(2|x|2 − 1)

), |x| ≤ 1 ;

n = 1, |x| ≥ 1 . (8)

This will focus all rays incident on it a fixeddirection to the point on the circumference inthe opposite direction.

The radius of curvature of the SphericalGeometry given by (6) is noR. To obtainHyperbolic Space H3 , often called Bolyai-Lobachevsky space or just Lobachevsky Space,one needs only to let the radius of curvaturebecome pure imaginary which leads to

n =n0

(1− x2

4R2 ). (9)

The refractive index becomes infinite when|x| = 2R which should be thought of as theboundary at ‘infinity’ of hyperbolic space. Infact the reader may easily verify that the opti-cal distance along a radial path from the originx = 0 to the boundary at infinity is, by contrastwith the case of spherical space, infinite.

Jules Henri Poincare (1854-1912) gave asimple analogy which illuminates the rolesplayed by the flat Euclidean metric geometryfor which n = 1 and the curved non-Euclideangeometry for which n is given by (9) as fol-lows. Imagine a medium whose temperaturevaries with radius as 1

n occupying a ball of ra-dius |x| = 2R, as measured by a measuringrod made from a substance such as Invar whoselength is independent of temperature. If mea-sured by a different ruler made of a materialwhich expands in proportion to the tempera-ture, then the rod will shrink to zero length asit approaches the the boundary which is at zerotemperature. The boundary will thus seem to

6

be infinitely far away as measured by the sec-ond measuring rod. At the time that Poincarewrote, people were still reeling under the dis-covery that what seemed to have been well es-tablished since antiquity: that Euclid’s Geo-metrical Axioms were not logical necessities.There was thus great interest among geometersand philosophers as what was the “correct” or“real” geometry of space. For Poincare it alldepended upon how you measure it. He didhowever believe that one should always be ableto find a system of measurements in which Eu-clid’s Geometry holds.

Despite appearances both Spherical Geom-etry and Hyperbolic Geometry are, like theirsimpler Euclidean counterparts, both isotropicand homogeneous. For this reason they are can-didates for the physical geometry of space, asmeasured for example by light rays in vacuo.Indeed according to Einstein’s theory of Gen-eral Relativity just these three possibilitiescan arise in the theory of the Expanding Uni-verse proposed by Alexander AlexandrovichFriedman (1888 -1925) and Monsignor GeorgesHenri Joseph Edouard Lemaıtre (1894 - 1966).For many years cosmologists have been at-tempting to decide which best fits the observeduniverse. Based on observations of the Cos-mic Microwave Background (CMB), and otherdata, the consensus now is that it is flat Eu-clidean geometry.

Of course no realistic medium is exactly ho-mogeneous or isotropic, and in particular thespeed of light, or sound, may depend upon di-rection as well as position. A familiar exampleis provided by bi-refringence in crystals suchas calcite for which the ordinary and extra or-dinary ray have different refractive indices, noand ne. This more general situation may betaken into account using a more general geom-etry invented by Georg Friedrich Bernhard Rie-mann (1826 - 1866) called Riemannian Geome-try in which the optical or acoustic path is givenby

δL = δ

∫γ

√hij(xk)

dxi

dt

dxj

dtdt = 0 , (10)

where hij is, in three spatial dimensions, a 3×3symmetric array called the metric tensor. In-

deed one frequently introduces what is called aline element which is an expression for the in-finitesimal form of a generalised Pythagoras’stheorem: the infinitesimal distance ds betweenpoints xi and xi + dxi is given by

ds2 = hij(xk)dxidxj . (11)

Thus, for example, in the case of a uniaxial bi-refringent medium with unit field n in the ordi-nary direction the Joets-Ribotta optical metricis [4]

ds2 = n2edx

2 + (n20 − n2

e)(n.dx)2 . (12)

Riemann’s ideas are fully incorporated intoEinstein’s Theory of General Relativity. In-deed, for a static spacetime a simple form ofFermat’s Principle holds which, for example,allows one to discuss the optics of black holesin terms of an effective refractive index (in so-called isotropic coordinates)

n(x) =(

1 +GM

2c2|x|

)6(1− GM

2c2|x|

)−2

, (13)

where M is the mass and G is Newton’s con-stant. In these coordinates, the black hole eventhorizon is located at |x| = 1

2GMc2 . Note that,

because the refractive index becomes infinitethere, the event horizon is at infinite opticaldistance. A closer examination reveals that asone approaches the event horizon, the opticalgeometry approximates more and more closelythat of hyperbolic space near its boundary atinfinity as described above. We have recentlypointed out that this is a universal feature of allstatic (“non-extreme”) event horizons and usedit as a quantitative tool for discussing some oftheir puzzling physics [7].

Riemann also invented an even more gen-eral form of geometry, taken up by Paul Finsler(1894-1970) called Finsler Geometry in which amore general expression replaces (11) and thisalso arises in the optics and acoustics of mov-ing media as we shall discuss presently. In themeantime we wish to say more about hyper-bolic geometry, and in particular the hyperbolicplane H2.

7

Figure 6: A tiling of the Poincare disk. Theedges of the triangles are geodesics.

Figure 7: The same tiling, this time shownon the upper half-plane model of hyperbolicspace.

Source: Three Dimensional Geometry and Topology, Thurston

2.2 The Hyperbolic Plane

The geometry of the hyperbolic plane is in-timately connected with that of the complexnumbers. Their introduction dramatically sim-plifies many formulae. To see this we adoptunits in which the radius of curvature n0R isset to unity. We further set x = 2R(x1, x2, x3),n0 = 1 in (9). The Hyperbolic Plane , H2, isobtained by setting x3 = 0 . In order to exploitthe complex numbers we define z = x1 + ix2 toget

ds2 = 4dx2

1 + dx22

(1− x21 − x2

2)2= 4

|dz|2

(1− |z|2)2. (14)

In this representation of the hyperbolic plane,the Poincare Disc, H2 occupies the interior ofthe unit disc |z| = 1 and one may verify thatits geodesics are circular arcs which cut the unitcircle at right angles. A tiling ofH2 by triangleswhose edges are geodesics is shown as Figure 6.The triangles are all similar, so one can see howthe apparent length of a measuring rod shrinksas we approach the boundary.

We are free, in addition, to perform a co-ordinate transformation to an equivalent form.We choose a complex coordinate w = u+ iv re-

lated to z by a fractional linear transformation

w = i1 + z

1− z. (15)

This maps the unit disc into the Poincare Up-per Half Plane, v > 0. The centre of the unitdisc maps to the point w = i and the boundaryof the unit disc to the real axis v = 0. In thesecoordinates the line element becomes

ds2 =|dw|2∣∣∣w−w2

∣∣∣2 =du2 + dv2

v2. (16)

Because, as is easily verified, fractional lineartransformations take circular arcs to circulararcs, the geodesics are again circular arcs. Be-cause holomorphic maps are angle preserving,the geodeiscs are in fact semi-circles orthogonalto the real axis. Figure 7 shows the upper halfplane tiled with the same geodesic triangles asin Figure 6.

Let’s now think of u as horizontal distanceand v as vertical height in an horizontally strat-ified medium in which the refractive index ndecreases with height. If we assume that overa certain range of heights, v, to a good approx-imation

n ∝ 1

v(17)

8

Pseudo-sphere

Light-cone

Poincare Disk

X0

Figure 8: The hyperbolic plane as the pseudo-sphere in Minkoswki space. Also shown are thelight-cone X ·X = 0 and the plane X0 = 0.

the rays will be semi circles. In this way wecan easily explain the mirage mentioned in theintroduction, that in arctic or antarctic regionslight is bent over icebergs and ships behind ice-bergs are observed to float upside in the airabove. To account for the mirages seen in thedesert or on hot days driving on motorways inwhich trees or cars seem to be reflected in poolsof still water, it suffices to take the complexconjugate and work in the lower half plane.

Of course many laws of horizontal stratifi-cation n = n(v) will give qualitatively similarresults, but there is a considerable economy tobe made by adopting the law (17). If we dothen the line element (16) is invariant underall fractional linear transformations taking theupper half plane into itself. These are of theform

w → aw − bcw − d

, ad− bc = 1, , (18)

where a, b, c, d are real. This defines the threedimensional group SO(2, 1) ≡ SL(2,R)/Z2 iso-morphic with the Lorentz Group of three di-

mensional Minkowski spacetime. This is noaccident. Much as it is convenient to thinkof the usual 2-sphere as the set of points ata fixed distance from the origin in Euclidean3-space, there is a similar interpretation of hy-perbolic space as a pseudo-sphere in Minkowskispace. We consider the space R3, but insteadof the usual dot product, we endow it with theMinkowski product:

X ·Y = −X0Y 0 +X1Y 1 +X2Y 2 (19)

where X = (X0, X1, X2) and similarly forY. For a vector with X · X > 0, which wecall spacelike, we can define the length to be|X| =

√X ·X. The hyperbolic plane of radius

R is the set of points defined by

X ·X = −R2, X0 > 0. (20)

The pseudo-sphere is sometimes referred to asthe mass-shell. This is because when we inter-pret the Minkowski spacetime as the geometryof special relativity in two spatial dimensions2,the vectors with P·P < 0 represent momentumvectors of particles whose rest mass, m, is given

2In this case we work in units where the speed of light is 1 and a particle of energy E and momentum (p1, p2)in the two spatial dimensions would have momentum vector P = (E, p1, p2).

9

by P·P = −m2. The mass-shell thus representsall the possible velocities for a particle of a givenrest mass. It is possible to check that any vec-tor tangent to this surface is spacelike, so thatthe Minkowski inner product allows us to definethe length of such a vector. The geometry ofthis surface with this definition of length is thatof the hyperbolic plane. A Lorentz transfor-mation leaves both the Minkowski metric andthe condition (20) unchanged and so representsan isometry of the hyperbolic plane. To re-cover the Poincare disk model, we stereograph-ically project the pseudosphere from the point(−1, 0, 0) onto the plane X0 = 0 as shown inFigure 8.

3 Zermelo’s Problem andFinsler’s geometry

3.1 Zermelo’s navigation problem

In the previous section, we considered the prob-lem of finding the ray paths of a sound wavepropagating in a static medium. We showedhow Fermat’s principle of stationary time leadsus to consider the geodesics of a Rieman-nian metric. We would now like to considerhow sound waves propagate through a movingmedium. Fermat’s principle continues to apply,however we must work a little harder in orderto translate the principle into a mathematicalstatement.

We will start off by considering a problemproposed by Ernst Zermelo in 1931. Supposea boat, which can sail at a constant rate rel-ative to the water on which it sits, wishes tonavigate from point A to point B as quicklyas possible. If the water is at rest, then thecaptain should steer along a straight line join-ing A to B. More generally, the captain shouldsteer along a geodesic if the surface of the wa-ter may not be taken to be flat, for example ifthe points A and B are far enough apart thatthe Earth’s curvature should be taken into ac-count. The navigation problem for the captainin this case corresponds to finding the geodesicsof some Riemannian metric.

Now let us suppose that the body of wateron which the boat sits is not at rest, but insteadmoves with some velocity W(x), which we callthe drift. The absolute velocity of the boat isnow the sum of two components: the motion ofthe boat relative to the water, v, and W. Inorder to find the fastest route between A and Bthe captain must clearly take the drift into ac-count. In order to do this, let us first considerthe simplest situation, where the surface of thewater is a plane and the drift W is a constantvector, not changing from point to point. Thissituation is shown in Figure 9. We will assumefor convenience that the speed of the boat rel-ative to the water is c, a constant. We will alsoassume that the speed of the drift is less thanthe speed of the boat relative to the water, i.e.W2 < c2. Let’s define d = ~AB and work outhow long it will take the boat to get from Ato B assuming that v is constant. In this case,the position of the boat relative to A at time twill be

x = t(v + W). (21)

Supposing that the boat arrives at B at timeT , we have

d = T (v + W). (22)

We wish to solve for T as a function of d andW. We can do this easily by looking at theequations we get from dotting both sides of (22)with W and d:

d ·W = T (v ·W + W2), (23)

d2 = T 2(v2 + 2v ·W + W2).

Noting that v2 = c2 and eliminating v ·W be-tween the equations gives

T [d] =

√d2

c2 −W2+

(d ·W)2

(c2 −W2)2− d ·Wc2 −W2

.

(24)The condition that the speed of the drift is lessthan c guarantees that the denominators arepositive and that T [d] ≥ 0, with equality onlywhen d vanishes. We notice that if λ > 0, thenT [λd] = λT [d]. We can also check that T obeysa triangle inequality :

T [d1 + d2] ≤ T [d1] + T [d2]. (25)

This means that the time T which we havefound is in fact the least time to travel between

10

d

B

A

W

v

Figure 9: Zermelo’s navigation problem with a uniform drift

A and B, because we cannot reduce the timeby travelling along the sides of a triangle withbase AB. A simple limiting argument showsthat any curve from A to B will take longer totraverse than T .

Now we can consider a more general prob-lem, where the boat is navigating in a Rieman-nian manifold M , with metric h = hijdx

idxj .The drift is a vector field W i(x) on M , whichmay vary from point to point and whose lengthis always less than c. Suppose the captain steersthe boat so as to travel along a curve γ(s), withγ(a) = A, γ(b) = B. In order to find the timetaken to traverse this curve, we can approxi-mate it with lots of straight sections on whichthe metric and the drift are roughly constant.We have worked out how long it takes to travelalong such a line segment and we can simplyadd these times up. Passing to a limit, we findthat the time taken to travel from A to B alongthis curve is

T [γ] =1

c

∫ b

a

F [γi(s), γi(s)]ds (26)

where

F [xi, yi]/c =√α(xi, yi) + β(xi, yi)

α =hijy

iyj

c2 −W 2+

(hijyiW j)2

(c2 −W 2)2

β = −hijyiW j

c2 −W 2, (27)

with W 2 = hijWiW j This problem seems

somewhat artificial as imagining a boat navi-gating a general curved space is rather strange.

This set up is very natural, however, in the con-text of sound rays. If hij is the acoustical met-ric of a material (so that a ray moving in thedirection of the unit vector ni moves with ve-locity c/

√hijninj) and W i is a bulk motion of

the material, then T tells us the time the soundwould take to travel along γ. We are now in aposition to make a mathematical statement ofFermat’s principle for a sound wave propagat-ing through a moving medium. The sound raysare the paths which extremise the time alongthe path, or equivalently the optical length,L = Tc:

δL = cδT = 0. (28)

Before we discuss what this means for the studyof sound waves in a moving atmosphere, we willfirst discuss briefly a larger class of problemswhich have a similar form to ours.

3.2 Finsler and Randers geome-try

We have thus far been speaking somewhatloosely about geometries, without describingexactly what we mean. For our purposes, theparticular type of geometry which is of interestis a Finsler geometry. Although named afterPaul Finsler, the concept of a Finsler geometrywas introduced by Riemann in the same lec-ture that he proposed what is now known asRiemannian geometry. The defining feature ofa Finsler geometry is that for suitably well be-haved curves, one can define a curve length. Inorder to do so, one first has to define a Finslerfunction. The function F of (27) is a special

11

case. A Finsler function, in addition to an as-sumption on its smoothness, is required to havethree properties

1. Positivity: F [x, y] ≥ 0, with equality onlyif y = 0

2. Homogeneity: F [x, λy] = λF [x, y], forλ ≥ 0

3. Subadditivity: F [x, y1 + y2] ≤ F [x, y1] +F [x, y2]

Given a Finsler function F , we can then definethe length of a curve from γ(a) to γ(b) by:

L[γ] =

∫ b

a

F [γi(s), γi(s)]ds. (29)

Condition 1 ensures that the length of any non-trivial curve is positive. Condition 2 ensuresthat the length of a curve does not depend onits parameterisation. Condition 3 is necessaryso that the problem of finding curves of minimallength is well posed. This means that we cantalk about the geodesics of F as being curves ofminimal length between two points. Note thatthe ‘length’ defined in this way by the F of theprevious section is not the usual length of thecurve, so the geometry defined by this F differsfrom the Euclidean geometry of the plane.

Whilst Riemannian geometry is fairly wellunderstood, Finsler geometry in general ismuch less well studied. This is mainly becauseof the sheer variety of possible Finsler func-tions, which can be very exotic. One of thesimplest classes of Finsler function, into whichour function F of the previous section falls, arethe Randers metrics. The Finsler function of aRanders metric is given in terms of a Rieman-nian metric aij and a one-form bi as

F [x, y] =√aij(x)yiyj + bi(x)yi. (30)

This is a good Finsler function providedaijbibj < 1, where aij is the matrix inverse ofaij . One reason to study Randers metrics be-comes apparent when we consider the equationssatisfied by a curve which is a critical point of(29) when aij is the flat metric in R3. Since weare free to choose the parameterisation, we can

assume that for the curve x(s) we have x2 = 1and we then find that x(s) is a critical pointwhen

x = x× (∇× b). (31)

We see that x(s) follows the path of a particleof mass m and charge e moving in a magneticfield B = m/e(∇ × b) with unit speed. Forthis reason, the extreme curves of L are some-times referred to as magnetic geodesics. For ageneral aij , we find the generalisation of theLorentz force law in a curved space, with biacting as the vector potential for the magneticfield. By extending the notion of a metric toallow Finsler geometries, we have brought theproblem of charged particles in a magnetic fieldinto the realm of pure geometry.

3.3 A uniform magnetic field inthe Hyperbolic Plane

As an interesting example, we take x1 = u,x2 = v, with v > 0, and we will consider theRanders metric given by (30) with:

aij =ρ2

v2δij , b1 =

α

v, b2 = 0. (32)

Where ρ > 0 and α are fixed real numbers.This is a good Finsler metric, provided that|α|/ρ < 1. If α = 0, we see that this Randersmetric in fact corresponds to the Riemannianmetric considered in §2.2, that of the hyper-bolic plane of radius ρ in its ‘upper half-plane’form. If α is non-zero, b corresponds to a vectorpotential for the magnetic field which is every-where directed straight out of the plane andwhich has strength B = α/ρ, independent ofposition. Thus the geodesics of this Randersmetric will correspond to a charged particlemoving in a uniform magnetic field in the hy-perbolic plane. In order to find the geodesics,we have to find curves (u(s), v(s)) which ex-tremise the length

L = ρ

∫ds

(√u2 + v2

v+α

ρ

u

v

). (33)

Such a curve must satisfy the Euler-Lagrangeequations:

0 =d

ds

(v

v√v2 + u2

)+

√v2 + u2

v2+α

ρ

u

v2,

12

Figure 10: Some geodesics of the Randers metric given in (32), with α/ρ = 1/√

2.

0 =d

ds

(u

v√v2 + u2

+α

ρ

1

v

). (34)

In the case where α = 0, we know from abovethat the geodesics are circles which meet theline v = 0 at right angles. We also know thatwhen we have a constant magnetic field in theusual Euclidean plane that the particle trajec-tories are circles. It seems reasonable then toguess that with α non-zero, the geodesics re-main circular. We can consider a possible solu-tion of the form

u(s) = u0 + r cos s

v(s) = v0 ∓ r sin s. (35)

Notice that the circle is traversed in a clock-wise or anti-clockwise direction respectively forthe two choices of sign. For a general Finslermetric, unlike for a Riemannian metric, the di-rection of travel is important and a curve willonly be a geodesic when traversed in a partic-ular direction. Substituting into (34) we findthat we can satisfy the equations provided

v0

r=α

ρ= ±B. (36)

Thus for 0 < B < 1 we can either have clock-wise circles with centres above v = 0 or anti-clockwise circles with centres below v = 0.Since B < 1, we can interpret both cases ge-ometrically as meaning that circles which meetthe line v = 0 at an angle θ = cos−1B aregeodesics, provided they are traversed in theappropriate sense. Taking a limit where r, v0

get larger and larger with their ratio fixed, wefind that straight lines making an angle θ with

the v = 0 axis are also geodesics, providedagain that they are traversed in the correct di-rection. A little more work shows that in factany geodesic of this Randers metric is one ofthese curves. Figure 10 shows examples of thevarious cases for B = 1/

√2. For B < 0, we can

simply reverse the sense of the curves.

4 Sound in a wind

We noticed in §3.1 that Fermat’s principle tellsus that a sound ray in a medium with acousticmetric hij with a wind W i will move along ageodesic of a related Randers metric defined by

aij =hij

c2 −W 2+

WiWj

(c2 −W 2)2,

bi = − Wi

c2 −W 2, (37)

where W 2 = hijWiW j and Wi = hijW

j .Let’s firstly consider what this means in the

case where the speed of sound is constant andequal everywhere to c and the wind speed Wis small in magnitude compared to c. This isa reasonable approximation for sound waves ina realistic atmosphere. Typically W/c < 0.03and even in the strongest hurricanes, W/c doesnot exceed 0.3. For a constant, isotropic, speedof sound, the acoustic metric is simply

hij = δij . (38)

If we work to first order in w/c, then we findthat

aij =1

c2δij , b = −W

c2. (39)

13

Making use of (31) and keeping track of the fac-tors of c, we find that the path followed by asound ray is that of a particle of mass m andcharge e moving with speed c in a magneticfield given by

B = −me

(∇×W) = −meω , (40)

where ω is the vorticity of the wind. This jus-tifies our assertion in equation (1) that the vor-ticity of the wind acts like a magnetic field onthe sound rays.

A simple consequence of this correspon-dence is observed by seismologists measuringthe oscillations of the Earth after a large earth-quake. The spectral peaks are split by theEarth’s rotation. From our magnetic point ofview, the effect of the rotation is to give riseto a constant magnetic field inside the Earth.The splitting of the spectrum is in precise anal-ogy with the Zeeman effect which gives rise toa splitting of spectral lines for atoms in a mag-netic field.

4.1 A stratified example

An interesting problem to study involves com-bining a varying speed of sound with a windin a stratified atmosphere. We can gain someinsight by investigating a particular choice ofacoustic metric and wind. Let’s suppose thatthe acoustic metric for the medium at rest isthe hyperbolic metric discussed previously

ds2 = hijdxidxj = l2

dx2 + dz2

z2(41)

We’ll take this to model sound rays in an at-mosphere varying with temperature near theground, which we assume to be at z = l. Thespeed of sound at ground level is given by c.We’ll suppose there’s a horizontal wind withstrength proportional to height, so that

W = (−wzl, 0). (42)

Where w is some fixed parameter with units ofvelocity. We can easily calculate the associatedRanders metric and find the time to go along a

curve (x(s), z(s)). It is best expressed in termsof new variables:

u =cx

l√c2 − w2

, v =z

l. (43)

This just represents a rescaling of the axes bothparallel and perpendicular to the ground. Interms of u, v, the time is given by

T =l√

c2 − w2

∫ds

(√u2 + v2

v+w

c

u

v

).

(44)Thus, the sound rays follow the geodesics of aRanders metric of precisely the form we consid-ered in §3.3, with B given by w/c!

As a simple application, we can consider theproblem of tracing the sound emanating froma point source, P at ground level, which is atv = 1. We know that the sound rays in the u, vcoordinates are circles (and lines) which meetthe line v = 0 at an angle cos−1(w/c). Wealso know that these circles have a clockwisesense if their centre is above v = 0 and an anti-clockwise sense if it’s below. This is alreadyenough information to draw the rays shown inFigure 11 for the particular value w/c = 1/

√2.

We sketch outward directed geodesics whichhave a common starting point, P . The straightline geodesic through P plays an important roleas a separatrix, which separates two differentbehaviours for the trajectories. In this case,any ray emitted to the left of the separatrixwill return to the ground to the left of P , whilea ray emitted to the right returns to groundto the right. If we suppose that P is emittingsound uniformly in all directions, we see thatmore of the energy emitted is absorbed by theground on the left of P than on the right. Inthis case three times as much, which we can seeby considering the angle the straight geodesicmakes with the ground.

This example has three free parameters: c,w and l. By choosing these appropriately, wecan match (for example) the speed of sound,the speed of the wind and the the wind shearat ground level to a real, more complicated pro-file. In fact, in [5] we showed that it is possibleto construct a model based on the hyperbolicplane with four free parameters, so that one can

14

Figure 11: Some sound rays emanating from a point on the ground for w/c = 1/√

2, l = 1.The wind velocity at various heights (including the hypothetical extension below ground level)is shown at the left of the diagram. The rays are shown extending as dashed lines below groundlevel to show that they meet z = 0 at an angle of π/4.

additionally set the rate of change of speed ofsound with height at ground level as well.

5 Conclusion

In this article we have sought to show that themotion of a charged particle moving in two-dimensional Lobachevsky space, or the hyper-bolic plane, equipped with a non-uniform mag-netic field can provide a useful model for soundrays in a moving medium with a gradient inthe refractive index. We have mentioned thatin its three-dimensional version the geodesics ofLobachevsky space provide a useful model forthe motion of light rays near the event horizonof a non-rotating black hole. If the black holeis rotating then Coriolis type effects, referredto in General Relativity as the rotation of iner-tial frames, provide an effective magnetic field.These two examples by no means exhaust thepossible applications of hyperbolic geometry tophysics.

Two-dimensional surfaces abound in natureand if they have negative Gauss curvature,sometimes called anti-clastic at a point, the sur-face cannot lie on one side of its tangent planeat that point. Thus a finite smoothly embeddedsurface without edges in Euclidean space can-not have everywhere negative Gauss curvature,but a finite portion of a surface with edges may.A simple example is provided by a holly leaf.An example of great current physical interest,

following the 2010 Nobel Prize to Andre Geimand Konstantin Novoselov is a graphene surfacecontaining topological defects called disclina-tions in which some of the hexagonal latticecells have been replaced by heptagons. Theelectrical and other properties of such surfacesare of great interest, and their study entailssolving the Dirac equation in a portion of two-dimensional Lobachevsky space. The motionof charged particles on abstract finite Riemannsurfaces with no boundary or edges which haveconstant negative curvature and uniform mag-netic field are of interest in statistical mechan-ics since for week magnetic field the motion ischaotic or ergodic as it is known technically.However as the magnetic field strength is in-creased there is a sudden phase transition andthis ceases to be the case.

Three dimensional Lobachevsky space hasbeen invoked to model some aspects of quan-tum dots and the physics of four and five di-mensional Lobachevsky space and their confor-mal boundaries are currently of intense interestby String Theorists since Juan Maldacena sug-gested the famous AdS/CFT correspondencewhich has led to a number of break throughsin quantum gravity and the quantum theory ofblack holes. Without going into technical de-tails, it may be of interest to outline some fea-tures of this fascinating idea. Both in QuantumField Theory and in String Theory it is custom-ary to work in imaginary time . Thus if we start

15

in Minkowski spacetime with spacetime metric

ds2 = −c2dt2 + dx2 , (45)

we can pass to Euclidean space with positivedefinite metric

ds2 = c2dτ2 + dx2 . (46)

by setting

t = iτ , τ real . (47)

Often calculations may be performed more eas-ily in Euclidean space. We then pass back toMinkowski spacetime by setting

τ = −it , t real . (48)

This process is called a Wick Rotation and italso works for some curved spacetimes. A casein point is Anti-de-Sitter spacetime. This isa solution of Einstein’s equations with a neg-ative cosmological constant. It may be ob-tained from Lobachevsky space by a simpleWick Rotation. Maldacena’s brilliant conjec-ture is that there is a precise correspondencebetween String Theory in Anti-de-Sitter space-time on the one hand, and a special type ofquantum field theory, called a Conformal Quan-tum Field Theory, on the other hand, the lat-ter being defined on the conformal boundaryof Anti-de-Sitter spacetime. The conformalboundary of Anti-de-Sitter spactime is confor-mally related to Minkowski spacetime If we“Wick rotate” this conjecture we are led to con-jecture a correspondence between String The-ory in Lobachevsky space and Quantum Fieldtheory on its conformal boundary, the latterbeing conformally related to Euclidean space.

We hope that in this article we have made itclear that not only is a knowledge of hyperbolicgeometry and Lobachevsky space useful for un-derstanding trafffic noise, but it has a muchmuch wider range of applications in theoreti-cal physics; from cosmology to condensed mat-ter physics to String Theory and Planck scalephysics. We commend to the interested readerits study and further exploitation.

References

[1] G. C. Stokes, On the Effect of Wind onthe Intensity of Sound Report of the Br-tish Association, Dublin (1857) 22

[2] O. Reynolds, On the Refraction of Soundby the Atmosphere Proc Roy Soc A ,22(1874) 531-548

[3] C. Sondhauss, On the Refraction ofSound, Phil Mag 30 73-77

[4] A. Joets and R. Ribotta, A geometricalmodel for the propagation of light raysin an anisotropic inhomogeneous mediumOptics Communications 107 (1994) 200-204

[5] G. W. Gibbons and C. M. Warnick, TrafficNoise and the Hyperbolic Plane,’ AnnalsPhys. 325 (2010) 909 [arXiv:0911.1926 [gr-qc]].

[6] G. W. Gibbons, C. A. R. Herdeiro,C. M. Warnick and M. C. Werner, Station-ary Metrics and Optical Zermelo-Randers-Finsler Geometry, Phys. Rev. D 79 (2009)044022 [arXiv:0811.2877 [gr-qc]].

[7] G. W. Gibbons and C. M. Warnick, Uni-versal properties of the near-horizon op-tical geometry,” Phys. Rev. D 79 (2009)064031 [arXiv:0809.1571 [gr-qc]].

[8] C. S. Morawetz, Geometrical Optics andthe Singing of Whales Amer Math Monthly85 (1978) 548-554

[9] D. E. Weston and P.B. Rowlands, Guidedacoustic waves in the ocean Rep. Prog.Phys. 42 (1979) 347-387

[10] R. C. Weber et al. Seismic Detection of theLunar Core Science (2011)

16

The Authors

Gary Gibbons was born exactly 300 years af-ter Gottfried Wilhelm Leibniz. He came up toSt. Catharine’s College, Cambridge, to readNatural Sciences in 1965, specializing in The-oretical Physics. After taking Part III of theMathematical Tripos he commenced researchin DAMTP, first under D. W. Sciama and thenS. W. Hawking. After various post-doctoralappointments he was appointed to a lecture-ship in DAMTP in 1980. He is now Profes-sor of Theoretical Physics in DAMTP. He waselected to the Royal Society in 1999. He hasbeen a Professorial Fellow of Trinity Collegesince 2002.

Claude Warnick came up to Queens’ College,Cambridge, in 2001 to read Mathematics. Af-ter completing Part III of the mathematicalTripos in 2005, he studied for a PhD in theGeneral Relativity group of DAMTP, underGary Gibbons. He has been a Research Fel-low at Queens’ since 2008.

17