Discover millions of ebooks, audiobooks, and so much more with a free trial

Only $11.99/month after trial. Cancel anytime.

An Introduction to Differential Geometry
An Introduction to Differential Geometry
An Introduction to Differential Geometry
Ebook558 pages5 hours

An Introduction to Differential Geometry

Rating: 4 out of 5 stars

4/5

()

Read preview

About this ebook

A solid introduction to the methods of differential geometry and tensor calculus, this volume is suitable for advanced undergraduate and graduate students of mathematics, physics, and engineering. Rather than a comprehensive account, it offers an introduction to the essential ideas and methods of differential geometry.
Part 1 begins by employing vector methods to explore the classical theory of curves and surfaces. An introduction to the differential geometry of surfaces in the large provides students with ideas and techniques involved in global research. Part 2 introduces the concept of a tensor, first in algebra, then in calculus. It covers the basic theory of the absolute calculus and the fundamentals of Riemannian geometry. Worked examples and exercises appear throughout the text.
LanguageEnglish
Release dateMay 13, 2013
ISBN9780486282107
An Introduction to Differential Geometry

Related to An Introduction to Differential Geometry

Titles in the series (100)

View More

Related ebooks

Mathematics For You

View More

Related articles

Reviews for An Introduction to Differential Geometry

Rating: 4.111111111111111 out of 5 stars
4/5

9 ratings0 reviews

What did you think?

Tap to rate

Review must be at least 10 words

    Book preview

    An Introduction to Differential Geometry - T. J. Willmore

    INDEX

    Part 1

    THE THEORY OF CURVES AND SURFACES IN

    THREE-DIMENSIONAL EUCLIDEAN SPACE


    I

    THE THEORY OF SPACE CURVES

    1. Introductory remarks about space curves

    IN the theory of plane curves, a curve is usually specified either by means of a single equation or else by a parametric representation. For example, the circle centre (0, 0) and radius a is specified in Cartesian coordinates (x, y) by the single equation

    x²+y² = a²,

    or else by the parametric representation

    In the theory of space curves similar alternatives are available. However, in three-dimensional Euclidean space E3, a single equation generally represents a surface, and two equations are needed to specify a curve. Thus the curve appears as the intersection of the two surfaces represented by the two equations. Parametrically the curve may be specified in Cartesian coordinates by equations

    where X, Y, Z are real-valued functions of the real parameter u which is restricted to some interval. Alternatively, in vector notation the curve is specified by a vector-valued function

    r = R(u).

    Suppose a curve is defined by equations F(x, y, z) = 0, G(x, y, z) = 0, and it is required to find parametric equations for the curve. If F and G have continuous first derivatives and if at least one of the Jacobian determinants

    is not zero at a point (x0, y0, z0) on the curve, it is known from the theory of implicit functions that the equations F = 0, G = 0 can be solved for two of the variables in terms of the third. For example, when the first Jacobian is non-zero, the variables y and z may be expressed as functions of x, say y = Y(x), z = Z(x), which leads to the parametrization x = u, y = Y(u), z = Z(u). However, this solution is valid only for a certain range of x and it will not in general give a parametrization of the whole curve.

    Conversely, suppose a curve is given parametrically by equations (1.1) and it is required to find two equations which specify the curve. The straightforward method of solving the first equation to obtain u = f(x) and substituting in the other two equations gives y = Y(f(x)), z = Z(f(x)), but this solution may be valid only over a restricted range. Other methods of elimination may produce new difficulties as may be seen from the following example of the cubic curve given parametrically by

    Eliminate u to obtain the equations xz = y², xy = z. These equations represent two quadric surfaces which intersect not only along the given cubic but also along the x-axis.

    A parametric representation of a curve specifies not only the curve but also a particular manner in which the curve is described. This is readily seen if the parameter u is interpreted as the time and the curve is considered as the locus of a moving point. The same curve may be parametrized in other ways, and some of the properties of a particular parametric representation may be peculiar to the parametrization and therefore not an intrinsic property of the curve. In this sense, a parametric representation of the curve specifies too much.

    On the other hand, the specification of a curve by two equations gives too little information for the purpose of a differential geometer. Quite apart from the fact that, as has been seen, several different curves may be determined by the same pair of equations, there are other disadvantages in specifying a curve in this manner. For example, when considering the distance along a curve from a point P to a point Q, it is often necessary to specify the sense in which the curve has been described. To the differential geometer, a curve is not merely a set of points but it must have a sense of description. A parametric representation is not only a convenient way of giving a sense of description but it is also a useful tool for the further study of properties of the curve. A curve will therefore be specified by all its possible parametric representations which are equivalent in that they all describe the same curve with the same sense.

    When a curve is regarded as a set of points, it is necessary to decide to what extent a set of points must be restricted before it can be regarded as a curve. Since we shall use parametric representations, it is similarly necessary for us to decide how general a manner of description is to be considered. Since our subject is differential geometry, we restrict the manner of description accordingly. As a result, the previous question becomes trivial, for any set of points which can be parametrized in the manner we require becomes a suitable object of study.

    We are now in a position to make precise technical definitions.

    2. Definitions

    Functions of class m

    Let I be a real interval and m a positive integer. A real-valued function f defined on I is said to be of class m, or to be a Cm-function if f has an mth derivative at every point† of I and if this derivative is continuous on I.

    Briefly we can say that a Cm-function has a continuous mth derivative. When a function is infinitely differentiable we say it is of class ∞, or a Cor a C -function.

    The extension of the concept of class to real-valued functions of several real variables is evident, and we leave the formulation of the precise definition to the reader. Briefly we can say that a Cm-function of several variables admits all continuous partial derivatives of the mth order.

    A vector-valued function R = (X, Y, Z) defined on I is said to be of class m if it has an mth derivative at every point and if this derivative is continuous on I; or equivalently, if each of its components X, Y, Z is of class m. Such a function is frequently specified by the vector equation R = (X, Y, Z) or equivalently by the three equations for the Cartesian components

    never vanishes on Inever vanish simultaneously—the function is said to be regular. A regular† vector-valued function of class m is called a path of class m.

    Two paths R1, R2 of the same class m on I1, I2 are called equivalent of class m which maps I1 onto I2 and is such that R1 = R. The condition R1 = Ris equivalent to the three conditions

    admits an inverse function of the same class m.

    Any equivalence class of paths of class m determines a curve of class m. Any path R determines a unique curve and is called a parametric representation of the curve, the variable u being then called the parameter. The equations (1.1) are called parametric equations which relates two equivalent paths is called a change of parameter. It produces a change in the manner of description of the curve whilst preserving the sense.

    We now formally define a curve of class m in E3 as a set of points in E3 associated with an equivalence class of regular parametric representations of class m involving one parameter.

    As an example of two equivalent representations, consider the circular helix given by

    (i)

    (ii)

    The change in parameter in this case is

    It should be emphasized that not every property of a path is a property of the curve it represents, because some properties are peculiar to the particular parameter chosen. The properties of the curve are those which are common to all parametric representations, i.e. properties which are invariant under a change of parameter. The verification of this invariance will frequently be left to the reader.

    We note that when the function R(u) is linear, the equation r = R(u) represents a straight line.

    3. Arc length

    The distance between two points r1 = (x1, y1, z1), r2 = (x2, y2, z2) in Euclidean space is the number

    .

    If we are given a path r = R(u) and two numbers a, b (a < bis an arc of the original path joining the points corresponding to a and b. To any subdivision Δ of the interval (a, b) by points

    a = u0 < u1 < u2 < … < un = b

    there corresponds the length

    of the polygon ‘inscribed’ to the arc by joining successive points on it. Addition of further points of subdivision increases the length of the polygon (because two sides of a triangle are together greater than the third). It is therefore reasonable to define the length of the arc to be the upper bound of LΔ taken over all possible subdivisions of (a, b). This upper bound is always finite, because for any Δ

    and the right-hand member is finite and independent of Δ. (Note that we used the fact that R is at least C¹ in the first step. The inequality used in the second step follows as an easy consequence of the Schwarz inequality.)

    We now show that this upper bound is actually equal to the right-hand term of (3.1), so that this term gives a formula for the arc length. The definition of arc length implies that if a < c < b, then the arc length from a to b is the sum of the arc lengths from a to c and from c to b. We denote by s = S(u) the arc length from a to any point u. Then the arc length from u0 to u is S(u)—S(u0). We have just seen that

    and it follows from the definition of arc length that

    Hence

    This formula is equally valid if u < u0. As u tends to u. Since this is true for any u0 in the range I of the parameter, it follows that S is a function of the same class as the curve, and that

    In terms of a Cartesian parametric representation, this formula becomes

    may be written

    or, in terms of differentials,

    ds² = dx²+dy²+dz².

    never vanishes, s can be used as a new parameter. The function S is the change of parameter from s to u. In order to change from u to s, S is inverted to obtain, say, u (s); then the curve parametrized with respect to s is r = R(s)).

    The verification that s is independent of the parametrization follows immediately from the rule for changing the variable of an integral. The above argument is similar to that used by W. F. Newns (1957) in obtaining the formula for the arc length of a plane curve.

    EXAMPLE 3.1. Obtain the equations of the circular helix r = (a cos u, a sin u, bu), —∞ < u < ∞ where a > 0, referred to s c.

    , so the required equations are

    r = (a cos (s/c), a sin(s/c), bs/c).

    The range of u c.

    EXAMPLE 3.2. Find the length of the curve given as the intersection of the surfaces

    from the point (a, 0, 0) to the point (x, y, z).

    Write the equation of the curve in the parametric form

    Then from (3.6) we get

    4. Tangent, normal, and binormal

    From now on the same symbol r will be used to denote the position vector of a point on a curve and also as the function symbol of a path which represents the curve. With this convention, a curve may be represented by the equation r = r(u). The object of the convention is to free the symbol R, which will now be used to denote the position vector of a current point in space not necessarily lying on the curve.

    and let P, Q be two neighbouring points on the curve. By ‘neighbouring points’ we do not mean that the points P, Q are near in space but that the points have neighbouring values of the parameter.

    be represented by the equation r = r(u), and let P and Q. have parameters u0 and u,

    as u u0. Hence

    i.e. the unit vector along the chord PQ tends to a unit vector at P as Q P. This is called the unit tangent vector at P, and is denoted by t. Using (3.4) we have

    Observe that t, like the curve, is oriented in that it points in the direction of increasing s. The line through P parallel to t is called the tangent line at P. If R is any point on this line, the vector from the point of contact P to R is called a tangent vector at P.

    An alternative approach to the definition of tangent line is to note that there is a unique line which approximates to the curve to the first order near P—more precisely, there is a unique linear function L(u) such that

    From (4.1) it can be seen that

    The line determined by L(uat P.

    It is convenient to denote differentiation with respect to arc length by a prime; with this convention the unit tangent vector becomes

    Osculating plane

    , and let P, Q . Then the limiting position as Q P of that plane which contains the tangent line at P and the point Q is called the osculating plane at P.

    is parametrized with respect to arc length and that the parameters of P, Q are 0 and s respectively. The equation of the plane through the tangent line at P and the point Q is

    Also we have

    Using equations (4.5), (4.6) we find that the equation of the osculating plane is

    provided that the vectors r′(0), r″(0) are linearly independent. Since t² = 1, i.e. r′2 = 1, differentiation gives r′ .r″ = 0, and it follows that the vectors r′(0), r″(0) are linearly independent unless r″(0) = 0. A point P where r″ = 0 is called a point of inflexion, and the tangent line at P is called inflexional.

    When P is not a point of inflexion, it follows from (4.7) that any vector lying in the osculating plane can be expressed as

    rr

    . In particular, the vector r″ lies in the osculating plane.

    The question now arises, what happens at a point of inflexion? We now show that when a curve is analytic, we still obtain a definite osculating plane at a point of inflexion P unless the curve is a straight line. To prove this we differentiate the relation r′ .r″ = 0 to get r= 0, since r″ = 0 at P. Hence r= 0. Repetition of this argument leads to the result that r′ .r(k) = 0, where r(k) is the first non-zero derivative of r . If r(k) = 0 , then since the curve is analytic we conclude that t is constant and the curve is a straight line. If r(k) ≠ 0, then (4.6) gives

    as s → 0, and the equation of the osculating plane is

    defined by

    is a curve of class ∞ with the property that r(k)(0) = 0 .

    The osculating plane at all points with parameter u < 0 is Z = 0, while the osculating plane at all points with parameter u > 0 is Y = 0. The osculating plane at u = 0 is clearly indeterminate. Thus we see that at a point of inflexion, even a curve of class need not possess an osculating plane.

    We have dealt with the equation of the osculating plane at P at some length because it enables us to illustrate the type of difficulties which occur at a very early stage, and to express our attitude towards these difficulties in the future. A differential geometer is interested primarily in properties of curves and surfaces which hold for a ‘general’ point, and tends to ignore what happens at ‘special’ points. An analyst, however, is interested in what happens at ‘special’ points, and is particularly interested in giving examples which contradict the general statement which the geometer wishes to assert. Usually a geometer is not interested in the precise class of a curve under discussion provided that it is sufficiently high to enable him to discuss relevant properties of the curve. However, an analyst cannot be uninterested in the problem of determining the smallest class of the curve or relevant functions which enter in the hypothesis of a theorem, and the differential geometer cannot afford to ignore these questions completely. For example, we have seen that the osculating plane may not exist at a point of inflexion on a C∞-curve.

    Throughout this book we shall be concerned essentially with properties at ‘general’ points, and we shall place considerably less emphasis on what happens at ‘special’ points. Our viewpoint is that in an introductory work of this kind the reader should not be side-tracked along paths whose interest is analytical rather than geometrical. Moreover, in many cases, the reader himself will be able to supply qualifications to the general argument which may be necessary to deal with ‘special’ cases.

    EXAMPLE 4.1. Show that if a curve is given in terms of a general parameter u, then the equation of the osculating plane corresponding to (4.7) is

    . Substitution in (4.7) gives the required result.

    EXAMPLE 4.2. Find the equation of the osculating plane at a general point on the cubic curve given by r = (u, u², u³) and show that the osculating planes at any three points of the curve meet at a point lying in the plane determined by these three points.

    The equation of the osculating plane is

    which reduces to

    3u²X—3uY+Zu³ = 0.

    If u1, u2, u3 are three distinct values of the parameter, the osculating planes at these points are linearly independent and the planes meet at a point (X0, Y0, Z0). The parameters u1, u2, u3 therefore satisfy the condition

    u³—3u²X0+3uY0—Z0 = 0.

    If lX+mY+nZ+p = 0 is the equation of the plane passing through the three points, then the parameters must also satisfy the condition

    lu+mu²+nu³+p = 0.

    Since this equation has three distinct roots we have n ≠ 0; comparing coefficients in the two cubic equations gives

    The equation of the plane is thus

    3Y0X—3X0Y+ZZ0 = 0,

    and since this is satisfied by (X0, Y0, Z0) the result follows.

    The normal plane at a point P on a curve is that plane through P which is orthogonal to the tangent at P.

    The principal normal at P is the line of intersection of the normal plane and the osculating plane at P. A unit vector along the principal normal is denoted by n; its sense may be selected arbitrarily provided that it varies continuously along the curve.

    Curvature

    The arc-rate at which the tangent changes direction as P moves along the curve is the curvature | = |t′| but the sign is not determined. To find a suitable convention, we recall that t′ = r″ lies in the osculating plane and is also normal to t. It is therefore proportional to n, i.e. tn. We make the convention

    If n is chosen to vary continuously with sis determined in magnitude and sign by (4.10), but this value may change from positive to negative as P moves along the curve. It should be noted that the ambiguity of the sense of n does not apply to the vector t′ = r″ which is sometimes called the curvature vector, the curvature vector rn varies continuously along the curve, but this does not imply that n varies continuously along the curve (cf. footnote, p. 11).

    A necessary and sufficient condition that a curve be a straight line is that = 0 at all points. Any straight line has equation of the form r = as+b, where a and b are constant vectors. Hence t = a and t′ = 0= 0 identically, then r″ = 0 which gives on integration r = as+b, which is the equation of a straight line. Thus the condition is also sufficient.

    The binormal line at P is the normal in a direction orthogonal to the osculating plane. The sense of the unit vector b along the binormal is chosen so that the triad t, n, b form a right-handed system of axes, i.e.

    Torsion

    As P moves along a curve the arc-rate at which the osculating plane turns about the tangent is called the torsion . We now obtain the relation

    Since b² = 1 it follows that b .b′ = 0 and b′ lies in the osculating plane. Also b .t = 0 implies that b′ .t+b .t′ = 0; but as b .t′ = bn) = 0 it follows that b′ is orthogonal to t. But as b′ lies in the osculating plane it must be parallel to n. Thus the equation |b| is the absolute magnitude of the torsion. The negative sign in (4.12) is introduced since, as a convention, torsion is regarded as positive when the rotation of the osculating plane as s increases is in the direction of a right-handed screw travelling in the direction of tis determined only in magnitude.

    Let be a curve for which b varies differentiably with arc length. Then a necessary and sufficient condition that be a plane curve is that = 0 at all points= 0, then b must be a constant vector. In this case the identity r′ .b = 0 implies (r .b)′ = 0, from which r .b = constant, showing that the curve is plane. Note that the C∞-curve considered on p. 9 is not a plane curve but the torsion is zero where it exists.

    EXAMPLE 4.3. Prove that [r′, r. Equations (4.10) and (4.11) give r′ × rb. Differentiate and use (4.12) to get rbn. Scalar multiplication by r″ then gives the required result.

    is a necessary and sufficient condition that the curve be plane. Evidently

    = 0 always . The condition is thus necessary and sufficient.

    EXAMPLE 4.5. Calculate the curvature and torsion of the cubic curve given by r = (u, u², u³).

    Evidently

    Differentiate to get

    Take the vector product of (ii) and (i) to get

    Differentiate to get

    Take the scalar product of (ii) and (iv) to get

    ,

    for any given curve can be calculated by mere substitution, but it is advisable to use the method given and to treat each curve on its merits.

    Serret-Frenet formulae

    The relations

    are known as the Serret-Frenet formulae, and these underlie many investigations in the theory of curves and surfaces. The reader is well advised to commit them to memory.

    The first and third relations have already been obtained. The second relation follows from differentiating the identity n = b × t to get nn × t+b n bt.

    The behaviour of a curve in the neighbourhood of one of its points may be investigated by means of relations (4.13). At a point P on the curve let axes Ox, Oy, Oz be taken along t, n, and b, and let X, Y, Z be the coordinates of a neighbouring point Q and if s denotes the small arc length PQ, then, using Taylor’s theorem,

    When the derivatives of r are substituted from (4.13) this becomes

    the coefficients being evaluated at P. It follows that as a first-order approximation the chord PQ is along the tangent: its projection on the principal normal is a magnitude of the second order, and its projection on the binormal is of the third order.

    From equation (4.14) two relations can be deduced which are analogous to Newton’s formula for the curvature of plane curves. These are

    Moreover, it is easy to show that

    ≠ 0, the arc length PQ differs from the chord PQ by terms of the third order in s.

    It is convenient to refer to the plane determined by the tangent and binormal at P as the rectifying plane.

    EXAMPLE 4.6. Show that the projection of the curve near P .

    These results follow immediately from equations (4.14) by retaining only the lowest powers of s in each equation and then eliminating s from the equations in pairs.

    EXAMPLE 4.7. Show that the length of the common perpendicular d of the tangents at two near points distance s apart is approximately given by

    Let P, Q have parameters 0 and s respectively. The unit tangent vectors at P and Q are r′(0), r′(s),

    Enjoying the preview?
    Page 1 of 1