Differential Geometry in Phy Gabriel Lugo Department of Mathematical Sciences University of North Carolina at Wilmington Copyright ©1995, 1998‘This document was reproduced by the University of North Carolina at Wilmington from a camera ready copy supplied by the authors. ‘The text was generated on an desktop computer using ISTEX. Copyright ©1992, 1995 Alll rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording or otherwise, without the written permission of the authors. Printed in the United States of America,iiThese notes were developed as a supplement to a course on Differential Geometry at the advanced undergraduate, first year graduate level, which by the author has taught for several years. ‘There are many excellent good texts in Differential Geometry but very few have an early introduction to differential forms and their applications to Physics. It is the purpose of these notes to bridge some of these gaps and thus help the student get a more profound understanding of the concepts involved When appropriate, the notes also correlate classical equations to the more elegant but less intuitive modern formulation of the subject ‘These notes should be accessible to students who have completed a traditional training in Ad- vanced Calculus, Linear Algebra, and differential Equations. Students who master the entirety of this material will have gained enough background to begin a formal study of the General Theory of relativity Dr. Gabriel Lugo Mathematical Sciences Ncw Wilmington, NC 28403 Iugo@ems.unewiledu iiiivPreface 1. Vectors and Curves 11 Tangent Vectors 12 Curves in RY 1.3 Fundamental Theorem of Curves 2. Differential Forms 21 -Forms 2.2 Tensors and Forms of Higher Rank 3. Exterior Derivatives 24 The Hodge Operator 3 Connections 3.1 Frames 3.2 Curvilinear Coordinates 3.3 Covarinnt Derivative 34 Cartan Equations 4 Theory of Surfaces 4.1 Manifolds . 4.2. The First Fundamental Form 4.8 The Second Fundamental Form 44 Curvature 15, 1b a 2B 3 33, 33 35 38 40 43 43 “4 48 51Chapter 1 1.1 Tangent Vectors 1.1 Definition Buclidean n-space R” is defined as the set of ordered n-tuples p = (p!,..-p") where p* € R, for each i n. Given any two n-tuples p= (p!,....p"), a= (g1,.-.,9") and any real number ¢, we define two operations: pea = (lta 4a") ay P= (ep) With the sum and the scalar multiplication of ordered n-tuples defined this way, Euclidean space acquires the structure of a vector space of n dimensions! 1.2 Definition Let 2 be the real valued functions in R” such that 2(p) = p* for any point P= (p!,...,p"). The functions 2 are then called the natural coordinates of the the point p. When the dimension of the space n= 3, we often write: 2 = 2, 22= y and 2* 1.3 Definition A real valued function in IR" is of class C” if all the partial derivatives of the junction up to order r exist and are continuous. The space of infinitely differentiable (smooth) functions will be denoted by C™(R") In advanced calculus, vectors are usually regarded as arrows characterized by a direction and length. Vectors as thus considered as independent of their location in space. Because of physical and mathematical reasons, itis advantageous to introduce a notion of vectors which does depend on location. For example, if the vector is to represent a force acting on a rigid body, then the resulting ‘equations of motion will obviously depend on the point at which the force is applied. Ina later chapter we will also consider vectors on spaces which are curved. In these eases the position of the vectors is crucial. for instance, a unit veetor pointing north at the earth's equator, is not at all the same as a unit vector pointing north at the tropic of Capricoma, ‘This example should help motivate the following definition 1.4. Definition A tangent vector X, in R", is an ordered pair (X,p). We may regard X as an ordinary advanced calculus vector and p is the position vector of the foot the arrow. Tw these notes we will use Indices such as ffs Thon ices sul as yp, rum feom O to a indices auch as 0,8, 4, run fromm L to 2 t ving index conventions2 CHAPTER 1, VECTORS AND CURVES ‘The collection of all tangent vectors ata point p € R” is called the tangent space at p and will be denoted by T,(R”). Given two tangent vectors Xp, ¥p and a constant ¢, we can define new tangent vectors at p by (X+¥}p=Xp +¥) and (X)p = cXp. With this definition, it is easy to see that for each point p, the corresponding tangent space T,(R") at that point has the structure of a vector space, On the other hand, there is no natural way to add two tangent vectors at different, points. Let Uv be a open subset of R®, The set T(d/) consisting of the union of all tangent vectors at all points in [7 is called the tangent bundle, This object is not a vector space, but as we will see later it has much more structure than just a set, jon A vector field X in U € Ris a smooth function from U to T(U). smooth assignment of a tangent vector Xp to each point in in U. Given any two vector fields X and Y and any smooth function f, we can define new vector fields X+Y and £X by (X4Y)p = Xpt+¥ aay (EX)p = IXp Remark Since the space of smooth functions is not a field but only a ring, the operations above give the space of vector fields the structure of a ring module. The subscript notation Xp to indicate the location of a tangent veetor is some times cumbersome. At the risk of introducing some confusion, we will drop the subscript to denote a tangent vector. Hopefully, it will e clear from the context, whether we are referring to a vector or to a vector field. At the risk of introducing some confusion, we Vector fields are essential objects in physical applications. If we consider the flow of a fluid in fa region, the velocity vector field indicates the speed and direction of the flow of the fluid at that point. Other examples of vector fields in classical physics are the electric, magnetic and gravitational fields 1.6 Defi let f be aC is defined by jon Let X, be a tangent vector in an open neighborhood U of a point p € R” and function in U. ‘The directional derivative of f at the point p, in the direction of X,, Vx) = F0)-X(o), (13) where f(p) is the gradient of the function f at the point p. The notation Xplf) = Vx (Ftp) is also often used in these notes. We may think of @ tangent vector at point as an operator on the space of smooth functions in a neighborhood of the point. ‘The operator assigns to a function, the directional derivative of the function in the direction of the veetor. It is easy to generalize the notion of directional derivatives to vector fields by defining X(/)(p) = Xp(f) LT Proposition If f,g € CR", a,b € R, and X is a veetor field, then N(aftbg) = aX(s)+0NX(a) (ua) Xa) = £X(a) +9X(f) ‘The proof of this proposition follows from fundamental properties of the gradient, and it is found in any advanced calculus text Any quantity in Euclidean space which satisfies relations 1.4 is a called a Tinear derivation on the space of smooth functions. The word linear here is used in the usual sense of a linear operator in linear algebra, and the word derivation means that the operator satisfies Leibnity rule.1.2, CURVES IN RS 3 ‘The proof of the following proposition is slightly beyond the scope of this course, but the propo- sition is important because it characterizes vector fields in a coordinate independent manner, 1.8 Proposition Any li ‘This result allows us to id car derivation on C™°(R") is a vector field ntify vector fiolds with linear derivations. This step is big departure from the usual concept of a “calculus” vector. To a differential geometer, a vector is a linear operator whose inputs are functions. At each point, the output of the operator is the directional derivative ‘of the function in the direction of X. Let p €U be a point and let 2" be the coordinate functions in U. Suppose that Xp = (Xp) where the components of the Euclidean vector X are a,....a ‘Then, for any function f, the tangent vector X;, operates on J according to the formula a XU) veo (15) It is therefore natural to identify the tangent vector X, with the differential operator Sogo (16) iene Notation: We will be using Einstein's convention to suppress the si an expression contains a repeated index. ‘Thus, for example, the equat written - a » Xp = 0p aa) ‘This equation implies that the action of the vector Xp acts on the coordinate functions 2 yields the components af of the vector. In elomentary treatinents, vectors are often ident components of the vector and this may cause some confusion “The difference between a tangent vector and a vector field is that in the Latter case, the coeficients a’ are smooth functions of x', The quantities a a gale (Ggale form a basis for the tangent space 7)(R") at the point p, and any tangent vector can be written as a linear combination of these basis vectors. ‘The quantities a! are called the contravariant components of the tangent vector. Thus, for example, the Enclidean vector in R® x +4)—3k located at a point p, would correspond to the tangent. vector ty = oped tet mation symbol whenever n above could be simply ed with the a a a BUgpe +A gale — BUG 1.2 Curves in R* 1.9 Definition A curve a(t) in R is a C% map from an open subset of R into R. The curve assigns to each value of a parameter t € R, a point (2"(0),2%(t),2°(4)) in RE veR > RS b> alth='),2700),2(0)4 CHAPTER 1, VECTORS AND CURVES One may think of the parameter f as representing time, and the curve a as representing the trajectory of a moving point particle 1.10 Example Let (0) = (al + baat + ba, ast + by) ‘This equation represents a straight line passing through the point p = (bi,b2,6s), in the direction of the vector v= {0,253} 1.11 Example Let a(t) = (acoswt, asinut, bf), "This curve is called a circular helix. Geometrically, we may view the curve as the path described by the hypothenuse of a triangle with slope b, which is wrapped around a circular cylinder of radius a. ‘The projection of the helix onto the zy-plane is a circle and the curves rises at a constant rate in the z-direction Occasionally we will revert to the position vector notation x8) = (2), 2°15, 2410) (18) which is more prevalent in vector calculus and ele notation really me: itary physics textbooks. OF course, what this - a(t) = (2! oa)(t), (19) where 2* are the coordinate slot functions in an open set in R® 1.12 Definition ‘The derivative a’) of the curve is called the velocity vector and the second derivative a’()) is called the acceleration. The length v = jla‘(t)| of the velocity vector is called the speed of the eurve. ‘The components of the velocity vector are simply given by ab dx! dx? da* we (= oe). (1.10) (4 +(4) +(4) (ui) ‘The differential of dx of the classical position vector given by dx! de? de® 5 = (12) ou and the speed is, is called an infinitesimal tangent vector, and the norm |[dx|| of the infinitesimal tangent vector is called the differential of arclength ds. Clearly we have ds = ||dx|| = vdt (1.13) As we will see later in this text, the notion of infinitesimal objects needs to be treated in a more rigorous mathematical setting. At the same time, we must not discard the great intuitive value of this notion as envisioned by the masters who invented of Calculus; even at the risk of some possible confusion! ‘Thus, whereas in the more strict sense of modern differential geometry, the velocity vector is really a tangent vector and hence it should be viewed as a linear derivation on the space of functions, itis helpful to regard dx as a traditional vector which, at the infinitesimal level, gives a linear approximation to the curve.1.2. CURVES IN R® 5 If f is any smooth function on R® , we formally define a(t) in local coordinates by the formula eH) l= Soa) ay ‘The modern notation is more precise, since it takes into account that the velocity has a vector part as well a8 point of application, Given a point on the curve, the velocity of the curve acting on function, yields the ditectional derivative of that function in the diction tangential to the curve at the point in question ‘The diagram below provides a more geometrical interpretation of the the velocity vector for- mula (1.14). The map a(t) from R. to R induces a map a, from the tangent space of R. to the tangent spare of RS. ‘The image a4(:) in TR® of the tangent vector 2 is what we call «(t) qs S)= 1) Since a’(t) is a tangent vector in R® , it acts on functions in R® . ‘The action of a'(f) on a function f on R® is the same as the action of £ on the composition foa. In particular, if we apply a(t) to the coordinate functions 2', we get the components of the the tangent vector, as illustrated $e TRS TRS 5 0'(1) bog RSRGR (02) law Ale oa) | (1.5) ‘The map a. on the tangent spaces induced by the curve a is called the push-forward. Many authors use the notation da to denote the push-forward, but we prefer to avoid this notation because most students fresh out of advanced calculus have not yet been introduced to the interpretation of the differential as a linear isomorphism on tangent spaces. 1.13. Definition IC = t(s) is a smooth, real valued function and aif) is a curve in R® , we say that the curve B(s) = a{i{s) is a reparametrization of « A common reparametrization of curve is obtained by using the arclength as the parameter. Using this reparametrization is quite natural, since we know from basic physics that the rate of change of the arclength is what we call speed ds : = F = lle“ (1.16) ‘The are length is obtained by integrating the above formula s= fiona f/(S)+(4 In practice it is typically difficult to actually find an explicit arclength parametrization of a ‘curve since not only does one have calculate the integral, but also one needs to be able to find the inverse function ¢ in terms of s. Ou the other hand, from a theoretical point of view, arclength parametrizations are ideal since any curve so parametrized, has unit speed. ‘The proof of this fact is ‘a simple application of the chain rule and the inverse function theorem. Fs) foes)" (Liz)6 CHAPTER 1, VECTORS AND CURVES = o(tls))e(s) /(t()) = MON TEE a(t(s)) Teco and any vector divided by its length isa unit vector. Leibnit notation makes this even more self evident ax as all 1.14 Example Let a(f) = (acoswt, asinut, $f). Then V(t) = (-awsinwt, aw cost b), att) [ J aarsineruy + (aw conway +B du = [VR an = et, where, ‘The helix of v it speed is then given by ls) = (acos “asin =, Frenet Frames Let (s) be a curve parametrized by are length and let ‘I{s) be the vector (8) = B'(s) (1.18) The vector T(s) is tangential to the curve and it has unit length, Hereafter, we will call the unit Tangent vector. Differentiating the relation T-T=1, (19) we get 2 =0, (1.20) 80 we conclude that the vector 7” is orthogonal to T. Let N be a unit vector orthogonal to 7, and let be the scalar such that T(s) = KN (9) (21) We call N’ the unit normal to the curve, and x the curvature, ‘Taking the length of both sides of last equation, and recalling that WV has unit length, we deduce that «= |IT'(s)Il (1.22)1.2. CURVES IN RS 7 It makes sense to call « the curvature, since if Tis a unit vector, then T"(s) is not zero only direction of 7'is changing. ‘The rate of change of the direction of the tangent vector is precisely what ‘one would expect to measure how much a curve is curving. In particular, it J” = 0 at a particular point, we expect that. at that point, the curve is locally well approximated by a straight line, We now introduce a third vector B=TxN, (1.23) which we will call the binormal vector. ‘The triplet of vectors (7’,V, B) forms an orthonormal set; that is, T.T=N.N=B-B=1 T-N=T-B=N 0. (124) If we differentiate the relation B« B = 1, we find that B- BY = 0, hence Bt is orthogonal to B. Furthermore, differentiating the equation T- B = 0, we get BUT+B-T' =O. rewriting the Inst equation BUT =-T-B=-nN-B we also conclude that BY must also he orthogonal to T. This the TB-plane, so BY must be proportional to N. In other word wn only happen if BY is orthogonal to we must have Bis) -rN(s) (1.25) for some quantity 7, which we will call the torsion. The torsion is similar to the curvature in the sense that it measures the rate of change of the binormal. Since the binormal also has unit length, the only way one can have a non-zero derivative is if B is changing directions. The quantity BY then measures the rate of change in the up and down direction of an observer which is moving with the curve always faring forward in the direction of the tangent vector. ‘The binormal B is something like the flag in the back of sand dune buggy. ‘The set of basis vectors {7,.N, Bis called the Frenet Frame or the repere mobile (moving frame). ‘The advantage of this basis over the fixed (i,j,k) basis is that the Frenet. frame is naturally adapted to the curve. It propagates along with the curve with the tangent vector always pointing, in the direction of motion, whereas, the normal and binormal vectors point towards the directions in which the curve is tending to curve, In particular, a complete description of how the curve is curving can be obtained by calculating the rate of change of the frame in terms of the frame itself. 1.15 Theorem Let 3(s) be a unit speed curve with curvature x and torsion 7. ‘Then T= aN NS = 7B (126) Br —TB Proof: We only need to establish the equation for N’, Differentiating the equation N -.N = 1, we get 2N’-.N’ =, so." is orthogonal to N. Hence, N* must be a linear combination of T’ and B. ‘= aT +08. Taking the dot product of last equation with T’ and B respectively, we see that "7, and b= N'-B.8 CHAPTER 1, VECTORS AND CURVES On the other hand, differentiating the equations VT’ =0, and N - B =0, we find that -N-T) =-N-(6N)=—-8 B=-N-Bl=-N-(-rN) =r. ai We conclude that a= —n, b= 7, and thus Nia-xP +78 The Frenet frame equations (1.26) can also be written in matrix form as shown below gee am ‘The group theoretic significance of this matrix formulation is quite important and we will come back to this later when we talk about general orthonormal frames, At this time, perhaps it su to point out that the appearance of an antisymmetric matrix in the Frenet equations is not at all coincidental, ‘The following theorem provides a computational method to calculate the curvature and torsion directly from the equation of a given unit speed curve. on Let i{s) be a unit speed curve with © = 1%) BB" x a") Bra (1.28) Proof: If §{s) is a unit speed curve, we have "(s) = T. Then "(6) = aN, 38" = (nN) a ams) = WN +RNT WN +a(-8T + 7B) = WN +-0T + x78. S-[8" xa] = Te [wN x (WN +e Po [8B +427] Ba x] +arB)] ot [3 x a] aa L17 Example Consider a circle of radius r whose equation is given by alt 0) (reost,rsin1.2. CURVES IN RS 9 Then, a'(}) = (=rsint,rcost, 0) lle’ = Vrain + (roost +? /r2(sin? ¢ + cos! t) Therefore ds/dt =v and 6 = rt, which we recognize as the formula for the length of an are of circle of radius t, subtended by a central angle whose measure is ¢ radians, We conclude that Als) rsin =, rcos *,0) is a unit speed reparametrization. The curvature of the circle can now be easily computed T= Ala) =(-oom2,—sin $0) TY = (hsin 2, —e = (LSU is a very simple but important example. ‘The fact that for a circle of radius r the curvature 1/r-could not be more intuitive. A small circle has large curvature and a large circle has small curvature, As the radius of the circle approaches infinity, the circle locally looks more and more like straight line, and the curvature approaches to 0. If one were walking along a great circle on a very large sphere (like the earth) one would be perceive the space to be locally flat 1.18 Proposition Let a(t) be a curve of velocity V, acceleration A, speed v and curvature th v= om aes (1.29) = Grade Prooft Let s(t) be the arclength and let (3) be @ unit speed reparametrization. Then a(0) = is(t)) and by the chain rule v=o Hste)s"0) a A = M(t) dy 4 = Sreeriays ay . GE + alae0 CHAPTER 1, VECTORS AND CURVES Equation 1.29 is important in physics. The equation states that a particle moving along a curve in space feels a component of acceleration along the direction of motion whenever there is a change of speed, and a c n the direction of the normal whenever it changes direction, ‘The centripetal acceleration and any point is, al acceleration ‘where r is the radius of a circle which has maximal tangential contact. with the curve at the point in question. This tangential circle is ealled the oseulating, cirele. The osculating circle can be envisioned by a limiting process similar to that of the tangent to a curve in differential calculus Let p be point on the eurve, and let q1 and qs two neaeby points. ‘The three points determine a july. This circle is a “secant” approximation to the tangent circle. As the points q; and 442 approach the point p, the “secant” circle approaches the osculating circle. The osculating circle always lies in the the TN-plane, which by analogy, is called the osculating plane. circle un 1.19 Example (Helix) a) = vs) oe) = beta”(s) (ra) ara Simplifying the last express Notice that if 6 = 0 the helix collapses to a circle in the xy-plane. In this ease the formulas above reduce tok = 1/a and r = 0. The ratio x/r = a/b is particularly simple. Any curve where 1/7 = constant is called a helix, of which the circular helix is a special case, 1.20. Example (Plane curves) Let (2) = (2(2),u(é),0). Then of = (ry, 0)1.2, CURVES IN RS u ie a apg) lo’ a" yx! 1.21 Example (Cornu Spiral) Let {s) = (x(s),u(s),0), where 26) = [oka wo = [nk van (8) = (008 Si Since ||" Ijl, the curve is of unit speed, and s is indeed the are length. ‘The curvature is of the Cornu spiral is given by at aye @ el ue" The integrals (1.80) defining the coordinates of the Cornu spiral are the classical Frenel Inte These functions, as well as the spiral itself arise in the computation of the diffraction pattern of a coherent beam of light by a straight edge. In cases where the given curve a(t) is not of unit speed, the following proposition provides formulas to compute the curvature and torsion in terms of a 1.22 Proposition If a(t) is « regular curve in R® , then Tee (1.31) r= fete, (139) where (a'a’"a"") is the triple vector product [a x’ a] -a'" Proof: a a a” = VT +U«N al” = (vx N'((s())s"(0 + = RN + = BerBt.2 CHAPTER 1, VECTORS AND CURVES ‘The other terms are unimportant here because as we will see a’ x a” is proportional to B axa" = wik(T x N)=eKB le’ x a"l] = 086 ea det xattl (ol eal) al” = nr = waa _ ata ilo oF 1.3. Fundamental Theorem of Curves si ne geometrical insight into the significance of the curvature and torsion can be gained by consid- ering the Taylor series expansion of an arbitrary unit speed eurve i(s) about s = 0 BO) 2, PMO) Als) = 310) + 3'(0)s+ Sst + (1.33) Since we are assuming that » is an atclength parameter, #0) = T= 80) = (®NY(O) = BND 90) = (eT KIN + wr BY(0) = HB To + Hh No + Kom Bo Keeping only the lowest terms in the components of T, 1V, and B, we get the first order Frenet approximation to the curve Ao) = 910) + Tos-+ SroNos? + trom Bie? (1.34) The first two terms represent the linear approximation to the curve. The first three terms approximate the curve by a parabola which lies in the osculating plane (T.N-plane). If sig = 0, then locally the curve looks like a straight line. If 7» = 0, then locally the curve is a plane curve which lies on the osculating plane. In this sense, the curvature measures the deviation of the curve from being a straight line and the torsion (also called the second curvature) measures the deviation of the curve from being a plane curve 1.23. Theorem (Fundamental Theorem of Curves) Let «(s) and r(s), (s > 0) be any two analytic functions. ‘Then there exists a unique curve (unique up to its position in R® ) for which s is the arclength, «(s) its curvature and 7(s) its torsion, Proof: Pick a point in R® By an appropriate affine transformation, we may assume that this point is the origin, Pick any orthogonal frame {T,.N B}. The curve is then determined uniquely by its Taylor expansion in the Frenet frame as in equation (1.34). 1.24 Remark It is possible to prove the theorem just assuming that s(s) and r(s) are continuous. ‘The proof however, becomes much harder and we refer the reader to other standard texts for the proof, 1.25 Proposition A curve with = 0 is part of a straight line. wwe leave the proof as an exercise.1.3, FUNDAMENTAL THEOREM OF CURVES: 3 1.26 Proposition A curve a({) with r= 0 is a plane curve Proof: If = 0, then (a‘a’a"") = 0. This means that the three vectors a’, a”, and a” are linearly dependent and hence there exist functions ai(s),a2(s) and as(s) such that ‘This linear homogeneous equation will have a solution of the form = 010) + e202 +¢3, 6) = constant vectors. This curve lies in the plane (x~es)-m=0, where m=e; xeu CHAPTER 1 VECTORS AND CURVESChapter 2 2.1 1-Forms One of the most puzzling ideas in elementary calculus is the idea of the differential. In the usual definition, the differential of a dependent variable y = f(z), is given in terms of the differential of the independent variable by dy = J/(z)dz. The problem is with the quantity dz. What does de mean? What is the difference between Ar and dx’? How much “smaller” than Ax does dx have to be? There is no trivial resolution to this question. Most introductory calculus texts evade the issue by treating dz as an arbitrarily small quantity (which lacks mathematical rigor) or by simply referring to dz as an infinitesimal (a term introduced by Newton for an idea that could not otherwise be clearly defined at the time.) In this section we introduce linear algebraic tools that will allow us to interpret the differential in terms of an linear operator 2.1 Definition Let p € R, and let 7, (R") be the tangent space at p. A L-form at p is linear map 6 from T,(R") into R.. We recall that such a map must satisfy the following properties a) aX) ER, VX, ER ea) b) (aX, + OY} = a9(Xp) + 6O(¥p), Va, bER, Xp, Yp € T(R") A Leform is a smooth choice of a linear map @ as above for each point in the space, 2.2 Definition Let f:R" + R be a real-valued C™ function, We define the differential df of the function as the I-form such that aX) = XU) (2.2) for every vector field in X in R™ In other words, at any point p, the differential df of a function is an operator which assigns to ‘a tangent vector Xp, the directional derivative of the function in the direction of that vector MX)(D) = Xo(S) = S(P) Xv) (23) In particular, if we apply the differential of the coordinate functions 2 to the basis vector fields, we get Ox! Ox ‘The set of all linear functionals on a vector space is called the dual of the vector space. It is an standard theorem in linear algebra that the dual of a vector space is also a vector space of the (24)16 CHAPTER 2, DIFFERENTIAL FORMS same dimension. ‘Thus, the space TR” of all Isforms at p is a vector space which is the dual of the tangent space T)R". The space T(R") is called the cotangent space of R” at the point p. Equation (2.4) indicates that the set of differential forms {(d")y,..., (d2"}y} constitutes the basis of the cotangent space which is dual to the standard basis {(72r)p,...(s2r}p} of the tangent space. ‘The union of all the cotangent spaces as p ranges over all points in R” is called the cotangent bundle TAR"), 2.3 Proposition Let f be any smooth function in RM and let {21,...2"} be coordinate functions ina neighborhood U' of a point p. Then, the differential df is given locally by the expression Proof: ‘The differential df is by definition a I-form, so, at each point, it must be expressible as a linear combination of the basis elements {(dz")p,..-s(dz")p}. ‘Therefore, to prove the proposition, it suffices to show that the expression 2.5 applied to an arbitrary tangent vector, coincides with definition 2.2. To sce this, consider a tangent vector Xp = @(z2;}p and apply the expression above A wey, = (Lasiyai : (Fee ole) = GE te'Val V0) (26) A pny 2 = HG GS I) af det Tet Oat 2, (So de) = al Fir) -X(p) dg(X)(0) ‘The definition of differentials as linear functionals on the space of vector fields is much more satisfactory than the notion of infinitesimals, since the new definition is based on the rigorous machinery of linear algebra, Ifa is an arbitrary 1-form, then locally ay (xe)dah,-.. + an (x}de", en where the coefficients a; are C™ functions. A I-form is also called a covariant tensor of rank 1, or just simply a covector. The coefficients (a),...,a,) ate called the covariant components of the coveetor. We will adopt the convention to always write the covariant components of a covector with the indices down. Physicists often refer to the covariant components of a I-form as a covariant vector and this causes some confusion about the position of the indices. We emphasize that not all one forms ate obtained by taking the differential of a function. If there exists a function f, such that a = df, then the one form a is called exact. In vector calculus and elementary physies, exact forms are important in understanding the path independence of line integrals of conservative vector fields. As we have already noted, the cotangent space 73(R") of I-forms at a point p has a natural veclor space structure. We can easily extend the operations of addition and scalar multiplication to2.2, TENSORS AND FORMS OF HIGHER RANK a the space of all I-forms by defining (+H) (Foy(x) o(X) + 5X) 28) fo(X) for all vector fields X and all smooth functions f. 2.2 Tensors and Forms of Higher Rank As we mentioned at the beginning of this chapter, the notion of the differential di is not made precise in elementary treatments of calculus, so consequently, the differential of area drdy in R2, as well as the differential of surface area in R® also need to be revisited in a more rigorous setting. For this purpose, we introduce a new type of multiplication between forms which not only captures the essence of differentials of area and volume, but also provides a rich algebraic and geometric structure iets (which only make sense in R®) to Euclidean spaces of which is vast generalization of cross pri all dimensions, OFM FPN N) = fo(X Yi) + fo(Xe Ni) AML FLY) = FON V+ FON), VXENETIR', ff EC™R Tensor Products 2.5 Definition Let wand # be forms. The tensor product of a and i is defined as the bilinear map a.@ such that (@@AUX.Y) = a(X)8Y) 29) for all vector fields X and Y ‘Thus, for example, if'a = ajda! and ; aa a), 0 dak be) = pg) G a) bdo, then, (ao 6 a = (ante SE Mbide 5) = aidibj§f = ah A quantity of the form T = Tijdx @ da) is called a covariant tensor of rank 2, and we may think of the set {dz © de} ns a basis for all such tensors, We must caution the render again that: there is possible confusion about the location of the indices, since physicists often refer to the components T,j 8 a covariant tensor. In a similar fashion, one can also define the tensor product of vectors X and ¥ as the bilinear map X@Y such that (oYiita for any pair of arbitrary functions f and g. X(N} (2.10)18 CHAPTER 2, DIFFERENTIAL FORMS 2 then, the components of X@ Y in the basis given by ab) Any bilinear map of the form © By ave simply ya .a ; fot (2.11) is called a contravariant tensor of rank 2 in R” ‘The notion of tensor products can easily be generalized to higher rank, and in fact one can have tensors of mixed ranks. For example, a tensor of contravariant rank 2 and covariant rank in IR" is represented in local coordinates by an expression of the fori OF ak "Oa Oa © This object is alo called a tensor of type T®*. Thus, we may think of a tensor of type Tas map with threo input slots. ‘The map expeets two fnctions inthe first two slots and a veetor in the third one, The action of the map is bilinear on the two functions and linear on the vector. ‘The output is areal number, An assignment of a tensor to each point in Ris called a tensor field ii Inner Products Let x and ¥ = 6:8; be two vector fields and lot Y) = dy0'bi (2.12) a ‘The quantity g(X,Y) is an example of a bilinear map which the reader will recognize as the usual dot product 2.6 Definition bilinear map g(X, Y) on the tangent space is called a vector inner product if 1 af aX), 2 g(X,X)>0, WN 3. g(X,X) = OX Since we assume g(X,¥} to be bilinear, an inner product is completely specified by its action on ordered pairs of basis vectors. The components gy of the inner product as thus given b aa Gg gs) = 0 (2.13) where g,; is a symmetric m x n matrix, which we assume to be non-singular. By linearity, itis easy to see that if X = a',2; and Y = 6) 2, are two arbitrary vectors, then HY In this sense, an inner product can be viewed as a generalization of the dot product. The standard Buclidean inner product is obtained if we take g,; = jj. In this case the quantity g(X,X) =| X | gives the squate of the length of the vector. For this reason qi, is also called a metrie and g is called a motrie tensor Another interpretation ofthe dot product ean be seen ifinstead one considers a vector X = a! tydai ‘Phe action of the I-form on the vector gives gigail 2 oF and a L-form a aX) = (bea g2) = bales) ya's oh2.2, TENSORS AND FORMS OF HIGHER RANK 19 If we now define eu) we see that the equation above can be rewritten as gga and we recover the expression for the inner product. Equation (2.14) shows that the metric can be used as mechanism to lower indices, thus trans- forming the contravatiant components of a vector to covariant ones. If we let g'! be the i the matrix gj, that is ferse of go 945 = 85, (2.15) wwe can also raise covariant indices by the equati B= gb; (2.16) We have mentioned that the tangent and cotangent spaces of Euclidean space at a particular point are isomorphic. In view of the above discussion, we see that the metric accepts a dual interpretation; ‘one as bilinear pairing of two vectors g2T(R") x T(R") 4 R and another as a linear isomorphism oT (R") 4 7(R") that maps veetors to covectors and vice-versa. In elementary treatments of calculus authors often ignore the subtleties of differential 1-forms and tensor products and define the differential of arclength as as = gyjdeide, although, what is really meant by such an expression is ds? = gydr! @ de! er 2.7 Bxample In cylindrical coordinates, the differential of arclength is ds? (2.18) In this case the metric tensor has compon rao w-[3 2 "| (229) 001 2.8 Example In spherical coordinates ‘psin Acos.d psinO sing 5 = peosd, (220) the differential of arclength is given by ds? = dp? + pdb? + p? sin® ade”. 221) -ase the metrie tensor has components 10 a=] 0p In this (222)20 CHAPTER 2, DIFFERENTIAL FORMS Minkowski Space ‘An important object in mathematical physies is the so called Minkowski space which is can be defined as the pair Let (M;,s,g} be the pair, where Moa = (tt2'2%,2°)] taf eR) (2.23) and g is the bilinear map such that g{X.X) = 0 + (08)? +(e"? + (29) (224) ‘The matrix representing Minkowski’s metric g is given by g = diag(=1, 1, 1,1) in which case, the differential of arclength is given by ds? = gus Oda” dt dt + da! © de! + dz” © dz? + da*@ de® = nde + (de) + (de)? + (de (2.25) Note: ‘Technically speaking, Minkowski's metric is not really a metric since g{X, X) = 0 does not imply that X = 0. Non-zero vectors with zero length are called Light-like vectors ancl they are associated with with particles which travel at the speed of light (which we have set equal to 1 in our system of units.) The Minkowski metrie gua and its matrix inverse gy are also used to raise and lower indices in the space in a manner completely analogous to RV” . Thus, for example, if A is a covariant vector with components Ay = (0 Ary Aa, As), then the contravariant components of A are At = gMAy (pA, Aa, Aa) Wedge Products and n-Forms 2.9 Defi jon A map ¢: TUR") x T(R") — Ris called alternating if (X,Y) = YX) ‘The alternating property is reminiscent of determinants of square matrices which change sign if any two column vectors are switched. In fact, the determinant function is a perfect example of an alternating bilinear map on the space Mz of two by two matrices. Of course, for the definition above to apply, one has to view Myx. a8 the space of column vectors. 2.10 Definition A 2-form 6 is a map 6: TUR") x T(R") — R which is alternating and bilinear 2.11 Definition Let a and be I-forms in RM" and let X and Y be any two vector fields. The wedge product of the two I-formsis the map aA: T(R) x T(R") —+ R given by the equation (a ABYX,Y) = a(X)HY) — oY) X) (2.26)2.2, TENSORS AND FORMS OF HIGHER RANK a 2.12 Theorem If a and J are Lforms, then a A is a 2-form. Proof = We break up the proof into the following two lemmas, 2.13 Lemma The wedge product of two L-forms is alternating. Proof Let a and 9 be L-forms in R" and let X’ and ¥ be any two vector fields. then (@ABUKY) = a(X}a(Y)-a(¥)9(X) (af A(X) — a(X)207)) (0.4 3)(¥,X) 2.14 Lemma The wedge product of two I-forms is bilinear, Proof: Consider I-forms, a9, vector fields X;,X,¥ and functions f* [forms are linear functionals, we get ‘Then, since the (ABE M+ PX Y) ad FX, + FX) HY) — a(Y)IF XM + FPN) [a(Xi) + fo(X2)]4{¥) = af YL BM) + Pal X)] FlalXi)AV) + Fat X)30) + flalY)A Ni) + falV) HX) = Fla XP) + a(¥ VAX] + Fla X2} 900) + a(¥ }A(X2)] POA BUX) + Pla A AX ¥) The proof of linearity on the second slot is quite similar and it is left to the reader. 2.15, Corollary Ifa and are I-forms, then aAB=-BAa (2.27) ‘This last result tells us that, wedge products have characteristics similar to cross products of vectors in the sense that both of these products are anti-commutative. This means that we need to be careful to introduce a minus sign every time we interchange the order of the operation. Thus, for example, we have de! de} = —de! Adat ii Fj, whereas dx’ A dai = ~d' Ade! =0 since any quantity which is equal to the negative of itself must vanish, The similarity between wedge products is even more striking in the next proposition but we emphasize again that wedge products ‘are by far much more powerful than cross products, because wedge products can be computed in any dimension. 2.16 Proposition Let a = Aide’ and = Bidz be any two L-forms in R® . Then AB = (AsBy)da’ A dat (2.28) Proof: Let X and Y be arbitrary vector fields. ‘Then (aA BM(X,Y) = (Ade')(X)(Byde?)(Y) — (Ajda")(¥ (Bde? )(X) (A) By)[de"(N)dxt(¥) — de'(¥ de? (X)] (AB Nde! A dedy(X,Y)R CHAPTER 2, DIFFERENTIAL FORMS Because of the antisymmetry of the wedge product the last equation above can also be written as 0A9= SOY (AB; - A;Bi)ldz! Ade) In particular, if n = 3, then the coefficients of the wedge product are the components of the cross product of A= Ari + Aoj+ Ask and B= Bit Boj + Bok ‘de — yy and 2.17 Example Let 0 = de +dy—2eyd=. Then ONG = (a%de—yPdy) A (de + dy — 22ydz) ja Nd + ada A dy — 2xSyde A de ~ yPdy Adz ~ yPdy Ady + 2eySdy A de wd Ady ~ 28 yde Adz — dy Ade + 2eyPdy A dz (2? + uP da A dy — 22%yde Ado + Qeydy A d= sin. ‘Then 2.18 Example let x=rcos@ and y= dy Ady = (—rsin 640 +cos ddr) A (rcos Add +sin Bdr) =rsin? 0d0 A dr + reos? dr Ado = (reos? # + rsin? (de A d0) r(dv A dd) (2.29) 2.19 Remark 1. The result of the last example yields the familiar differential of area in polar coordinates 2. ‘The differential of area in polar coordinates is a special example of the change of coordinate theorem for multiple integrals. It is easy to establish that ifr = J(u, ) and y= f2(u,v), then de Ady = det} J|duAdv, where det|J is the deten of the Jacobian of the transformation 3. Quantities such as dedy and dyd= which often appear in calculus, are not well defined. In most cases what is meant by these entities are wedge products of I-forms 4, We state (without proof] that all 2-forms ¢ in R” can be expressed as linear combinations of wedge products of differentials such as 6 = Fyde! Adz (2.30) In a more elementary (ie: sloppier) treatment of this subject one could simply define forms to be gadgets which look like the quantity in equation (2.30). This is fact what. we will do in the next definition. 2.20 Definition A 3=form ¢ in R” is an object of the following type = Aijads' Adz! Ada* (231) where we assume that the wedge product of three L-forms is associative, but still alternating in the sense that if one switches any two differentials, then the entire expression changes by a minus sign. we challenge the reader to come up with a rigorous definition of three forms (or an n-form, for that matter) more in the spirit of multilinear maps. There is nothing really wrong with using2.3, EXTERIOR DERIVATIVES 23 definition refform, It is just that this definition is coordinate dependent and mathematicians in general (specially differential geometers) prefer coordinate-free definitions, theorems and proofs. And now, a little combinatorics. Let us count the number of differential forms in Euclidean spare. More specifically, we want to count the dimensions of the space of k-forms in R in the sense of vector spaces. We will thing of 0-forms as being ordinary functions. Since functions are the scalars”, the space of O-forms as a vector space has dimension 1 & Forms | Dim ‘forms f T Horms | fde!,gdz? | 2 forms | fade! Adz? | 1 R Forms Dim O-forms f T Hors Sude!, fude?, faz 3 2forms | fide” Ade’, fodx* A da', fadz' Adz? | 3 S-forms Sade! ndz? Ade 1 ‘The binomial coefficient pattern should be evident to the reader. 2.3 Exterior Derivatives In this section we introduce a differential operator which generalizes the classical gradient, cuel and divergence operators. Denote by AZ, (R*) the space of m-forns at p € R". ‘This vector space has dimension dim iy) = n)t for m n. We shall identify A(R") with the space of C™ functions at p. Also we will call A(R") the union of all A(R") as p ranges through all the points in RY In other words, we have AMR") = JAR") Ia € A(R"), then a can be written in the form Abjooiale de®! A ...de'™ (2.82) 2.21 Definition Let a be an m-formn (written in coordinates as in equation (2.32)). Th derivative of a is the (m-+I-form) da given by da dim dain In the special case where a is a 0-form, that is, a function, we write of ae a=mu CHAPTER 2, DIFFERENTIAL FORMS a) by 6) danpy=dans+(-trands YaE’ vent (234) Proof: 3} Obvious from equation (2:82) ») First we prove the proposition for a = f€ A". We have atta) = a 2h = OF aot ndat = heel nae Las ay i Daridet Fatagsite’ A dz! 0 Now, suppose that a is represented locally as in equation (2.32). 1b follows from 2.33 that d{da) = d(dAy,,. i.) Ade” Ade" ...d2= = ¢) Leta A’, se At. Th = Ag. glee" A. dee By, jg(@ deh Wd (2.35) By definition, ANB = Aig iy Bin ig dt" Ne Neda?) A (dz A... Ada) Now we take take the exterior derivative of the last equation taking into account that d(fg) = Jdg + gdf for any fanctions f and g. We get AONB) = (ACA 5g) Bin dg + Arig (By (de A. Ada) A (dat A.A dae) FAK, (da A. Ade] ALB). jg A (del A Adee) + = ia Ade A. de] (AP TAB ,..5, A (da? A. Ade!9)] = dang+(-1fands. (2.36) ‘The (=1)? factor comes in because to pass the term dB; has to perform p transpositions. through p L-forms of the type da! 2.23, Example Let a= P(z,y)de + Q(x, y)dd. Then, OP, OP. ag to = irs Binder (aes 2 nay eee ey = Payne tena = Cadena cam) ‘This example is related to Green’s theorem in R24, THE HODGE OPERATOR B 2.24 Example Let example, M(2, y}de-+.N(2,4}dy, and suppose that da = 0. ‘Then, by the previous aN al Or My which implies that N= fy and M, for some C? function f(x,y). Hence hae Ady. dav Thus, da = 0 iff Ne = fede + fydf = df. The reader should also be familiar with this example in the context of exact differential equations of first order, and conservative force fields 2.25 Definition A differential form a is called exact if da = 0. 2.26 Definition A differential form a is called closed if there exist a form 9 such that a ince dod = 0, it is clear that a closed form is also exact, ‘The converse is not. at all obvi general and we state it here without proof. 2.27 Poineare’s Lemma In a simply connected space (such as R” ), ifa differential is exact then it is closed. ‘The assumption hypothesis that the space must be simply connected is somewhat subtle. ‘The condition is reminiscent of Cauchy's integral theorem for functions of a complex variable, which states that if f(s) is holomorphic function and is a simple closed curve, then, fire ded by the curve C is not simply connected. ‘The his theorem does not hold if the region bou dard example is the integral of the complex I-form w = (1/z}dz around the unit cite bounding a punctured disk. In this ease, } 1 2.4 The Hodge-* Operator One of the important lessons that students learn in linear algebra is that all vector space of finite dimension n are isomorphic to each other. ‘Thus, for instance, the space Ps of all real polynomials in x of degree 3, and the space Maca of real 2 by 2 matrices, are basically no different than the Euclidean vector space It in terms of their vector space properties. We have alzeady encountered a number of vector spaces of finite dimension in these notes. good example of this is the tangent space TyR#, The "vector" pat a fe +0" an be mapped toa regular advanced cles vector ali + 0%) +k, by replacing 3 by and ;& by k. Of course, we must not confuse a tangent vector which is @ incar operator with a Euclidean vector which is just an ordered triplet. but as far their vector space properties, there is basically no difference. We have also observed that the tangent space 7},R” is isomorphic to the cotangent space Tp" In this case, the vector space isomorphism maps the standard basis vectors {2} to their duals {do'}. This isomorphism then transforms a contravariant vector to a covariant vector. Another interesting example is provided by the spaces (\1(R®) and A2(R3), both of which have dimension 3. It follows that these two spaces must be isomorphic. In this ease the isomorphism is given by the map de dy Adz26 CHAPTER 2, DIFFERENTIAL FORMS dy + =de Adz dz 4 de Ady (2.38) More generally, we have sen tht the dimension of the space of m-frms in Ris given by the Vinomial coefficient (fh). Since oe (i) = (2m) = AMR) = APR) (2.89) ‘To describe the isomorphism hetwvoen these two spaces, se will first need to introduce the totally antisymmetric Levi-Civita permutation symbol which is defined as follows it must be the ease that 1. ifs...) is an odd permutation of{1,...,1m) (2.40) 0 otherwise {3 iff...) is an even permutation of{1,...,m) In dimension 3, there are only 3 (3!=¢ ) nonvanishing components of & 54 in a = em = ene = 1 e192 = ena = Ga = 1 (241) ‘The permutation symbols are useful in the theory of determinants, In fact, if A= (a}) isa 3x 3 matrix, then, usi 2.41), the reader can easily verify that det = [A] = erin alee? (2.42) ‘This formula for determinants extends in an obvious manner ton x n matrices. A more thorough discussion of the Levi-Civita symbols will appear later in these notes Im RY , the Le ‘a symbol with some or all the indices up is numerically equal to the permutation symbol will all indices down ce the Euclidean metric used to raise and lower indices is 4. On the other hand, in Minkowski space, raising an index with a value of O costs a minus sign, " Thus, in cal Mca) because goo = g' since any permutation of {0,1,2,3} must contain a 0 2.28 Definition The Hodge-« operator is a linear map + : A™(R®) —» A(R) defined in standard local coordinates by the equation A ant drier y...Adgi™, (2.43) Hd? nndain) = TO i Since the forms dA. ..Adx constitute a basis of the vector space A\!"(R") and the «operator is assumed to be a linear map, equation (2.48) completely specifies the map for all m-forms,24, THE HODGE OPERATOR a 2.29 Example Consider the dimension n=3 case. then ade? = igo! Ade Fle aye? do? +e yada? A do] Jide ned = de nde lle’ nda + da 9d) ds? nds" We leave it to reader to complete the computation of the action of the operator on the other basis, forms. The results are sdz! = +de" Ade® edz? =—de! Ade® vdz® = 4da! do, (244) s(d2? Ade) (de® Ade") o(de! Ada?) (2.45) olde! nda? Ade8) = 1 (246) In particular, if J R® — R is any O-form (a function) then, af = fide" Adz* Adz) = fal where dV’ is the differential of volume, also called the volume form. Ayda! Ande? + Agdz®, and Bide! Bada? + Bada. Then, 2.30 Example Let a (Ay Bg — Ag By) + (da? A dar¥) + (A, Bg — Ag By) + (dx! Adx*) + (Ay By ~ ApB) + (da! A dz”) (ApBo — AgBo)dz? + (Ay By ~ Ag By }dx® + (A, By ~ ApBy}da* = (Ax Bde! (2.48) (ana) of the A and + operators. Ione thinks ‘The previous examples provide some insight on the he ind Ke, then it should be apparent that of the quantities de, dx® and dz* as playing the role of f ‘equations 2.44 are the differential geometry versions of the well known relations i= jxk j= -ixk k= ixj This is even more evident upon inspection of equation (2.48), which relates the A operator to the Cartesian cross product.2 CHAPTER 2, DIFFERENTIAL FORMS 2.31 Example In Minkowski + operator in this case, splits (Maa) into to S-dim subspaces AV, such that * More specifically, A is spanned by the forms {d2" Ade! da" by the forms {di® Adz®, —dx" A da®,ds* A de®). ‘The act 6. The Hodge A {de Ade!) = 2% do" Adz! = de® Adz® ade nda?) = 4% da Ada! = tar" Ade (dz Ada’) = 1% ,de* Ada! = —dr" Ade? and on AZ +(+dz" A de®) dx" Adz" Heda! nds)= delle Ada! = de" Ade? s(tdz! Adz?) = 2% de® Ade! = de" Adz® In verifying the equations above, we recall that the Levi-Civita symbols which contain an index with value 0 in the up position have an extra minus sign as a result of raising the index with g®.If FE A’(M), we will formally write F = Fy + F_, where Fx € AZ. We would like to note that the action of the dual operator on \"(.M) is such that + A?(M) —+ A(M), and «? = —1. Thus, the operator is an linear involution of the space and in fact, \3. are the eigenspaces corresponding to the two eigenvalues of this involution It is also worthwhile to calculate the duals of I-forms in Mi,s. ‘The results are edt = dr! nds? nde? ede! = tdz* Adt Ads® viz? = 4dt Ade! Ada nde! = de! dt ade? (249) Gradient, Curl and Divergence Classical differential operators which enter in Green’s and Stokes Theorems are better understood as special manifestations of the exterior differential and the Hodge-* operators in R®, Here is precisely how this works: 1, Let f: R8— R be a C* function, Then y= Pte av de (25) 2. Let a = Ajdz! be a 1-form in R®. Then (ota = 128 AH a nay a) te ean 3. Let a = Bide? Ade* + Bade’ Ade! + Bydz! Ade? be a 2-form in R®. ‘Then = PBL OB, tet n da? nat do = (OE Oe i nat (v-Byav (2.52)24, THE HODGE-* OPERATOR 29 4 Leta Bidz, then sdea=V-B (253) It is also possible to define and manipulate formulas of classical vector calculus using the per- tation symbols. For example, let a = (1, 42,4) and B = (By, Ba, Bs) be any two Buclidean vectors. Then it is easy to see that {Ax Bye = ABs and (Tx By =e, ‘To derive many classical vector identities in this formalism, it is necessary to first establish the following identity (see Ex. ()) atm = Ah ~ i, a) 2.32, Example [Ax (Bx Cle = APA g lB Che " An(e4 BjCu) = CPO AmBiCe) Ent" A™ BC) (6485, — aF83)A" BC BA" Cn = CAM by Or, rewriting in vector form Ax (Bx C)=B(A-C)—C(A-B) (2.55) Maxwell Equations ‘The classical equations of Maxwell describing electromagnetic phenomena are (2.56) We would like to formulate these equations in the language of differential forms. Let (2%, 22,28) be local coordinates in Minkowski’s space My.3. Define the Maxwell 2-form F by the ‘equation pdet Nae", uve 0482) where 0 -B, -By -Ey ry) es Ey -B, 0 Be (Seems ye Boe en Written in complete detail, Maxwell's 2-form is given by Fv = (2.58) Sgdt Ada! — Bydt A de® — Beda! \ dx? — Bydla" de® + Beds? A dz°. (259)30 CHAPTER 2, DIFFERENTIAL FORMS We also define the source eurrent I-form. jude” = pdt + Iidet + + Jad" (2.60) 2.83. Proposition Maxwell’s Equations 2.56 are equivalent to the equations 0 dred. (2.61) dF dsF Proof: ‘The proof is by direct computation using the definitions of the exterior derivative and the Hodge-« operator. ai OBs Adz” Adt Adz! — SE dz® pdt nde! + Fer has! dt Ade? — Sada? Adt Ada? + dz? Adz" Adz? — any Ot ob. Ot Collecting terms and using the antisymmetry of the wedge operator, we get, aBy Adt dz! Ada? — dx? Adz! Ada + 0B, , 0B. Oa? * Oe BE. _ OBe Oz? Ot OFe By, gin gd Fee Ghat nde! na + oy Oat ida Ade? a da + Je? Mad A de® + a 1 2 — pleat at ade “Therefore, dF = 0 + bt Oe 9 Dat + Dat + Bak which is the same as and24, THE HODGE OPERATOR aL which means that i -VxE-F=0. (2.62) ‘To verify the second set. of Maxwell equations, we first compute the dual of the current density form (2.60) using the results from example 24. We get J pila! nde? A de® + Jyde? Ndl A de® + Jndt Adel \ de® + Jyde! dt A de®, (2.63) We could now proceed to compute de F, but perhaps it is more elegant to notice that F € A\"(M) and s0, according to example (2.4), F splits into F = Fy + F. In fact, we see from (2.58) tal the components of Fy. are those of ~B and the components of F_ constitute the magnetic field vector B, Using the results of example (2.4), wo can immediately write the components of «F AP = Bydt Ada! + Bydt Ado? + Budi Ada* + Bede! pda? ~ Byda! da? + Bada? Ade, (2.68) or equivalently, (2.65) -B, -B, 0 Ey By -E, 0 Since the effect of the dual operator amounts to exchanging E+ -B B+ +E, \we can infer from equations (2.62) and (2.63) that V-B=4np and. a32 CHAPTER 2, DIFFERENTIAL FORMSChapter 3 Connections 3.1 Frames As we have already noted in chapter 1, the theory of curves in R® can be elegantly formulated by introducing orthonormal triplets of vectors which we called Frenet frames. The Frenet vectors are adapted to the curves in such a manner that the rate of change of the frame gives information about the curvature of the curve. In this chapter we will study the properties of arbitrary frames and their ‘corresponding rates of change in the direction of the various vectors in the frame. This concepts will then be applied later to special frames adapted to surfaces 3.1 Definition A coordinate frame in R” is an n-tuple of vector fields {¢),--.¢n} whieh are linearly independent at each point p in the space In local coordinates «',...2", we can alviays express the frame vectors as linear combinations of the standard basis vectors (a) where 8; = 32r We assume the matrix A lar at each point. In linear algebra, this concept is referred to as a change of basis, the difference being that in our case, the transformation matrix A depends on the position. A frame field is called orthonormal if at each point, = 8 (32) ‘Throughout this chapter, we will assume that all frame fields are orthonormal. Whereas this restriction is not necessary, it is very convenient because it simplifies considerably the formulas to compute the components of an arbitrary vector in the frame. 3.2 Proposition If {e1,...,¢,} is an orthonormal frame, then the transformation matrix is ‘orthogonal (ie: AAT = 1) Proof: ‘The proof is by direct computation. Let ¢: = 0}... Then bj = < de A AA’ > ANAL, < 4,1 > AA ou 334 CHAPTER 3. CONNECTIONS AN ARs = ANAT) ie Hence: (A Ah = dy (arya = & ATA T Given a frame vectors ¢, we can also introduce the corresponding dual coframe forms 6; by requiring, . (ej) (3.3) since the dual coframe is a set of I-forms, they can also be expressed in of local coordinates as Jinear combinations It follows from equation( 3:3), that O63) = Bide*(O.A';) BA de*(A4) Bala Bia, ‘Therefore we conclude that BA = I, so B = A~! = AT. In other words, when the frames are orthonormal we have « Ona, = Aidat, (G4) 3.3. Example Consider the transformation from Cartesian to cylindrical coordinates eos, Using the eb a oF a a0 From these equations we easily verify that the quantities eee 1S Or eas oS 5M a are3.2, CURVILINEAR COORDIN: TES 35 are a triplet of mutually orthogonal unit vectors and thus constitute an orthonormal frame, 3.4. Example For spherical coordinates( 2.20) = psindcosd = psinAsing pees, the chain rule leads to a = a a = sindeos pg, + sin sings + cos0. a a a = prosticosdg + peosdsinds: + —psinds- = -psindsing 2 + prin dose Oy Ze gle Sle In this case, the vectors (35) also constitute an orthonormal frame, ‘The fact that the chain rule in the two situations above leads to orthonormal frames is not coincidental. ‘The results are related to the orthogonality of the level surfaces 2 = constant. Since the level surfaces are orthogonal whenever they intersect, one expects the gradients of the surfaces to also be orthogonal. Transformations of this type are called triply orthogonal systems. 3.2. Curvilinear Coordinates Orthogonal transfor \drieal coordinates appear ubiquitously in mathematical physies because the geometry of a large number of problems in this area exhibit metry with respect to an axis or to the origin. In such situations, transformation to the appropriate coordinate system often result in considerable simplification of the field equations involved in the problem. It has been shown that the Laplace operator which enters into all three of main classical fields, the potential, the heat and the wave equations, is separable in 12 coordinate systems. A sim ple and efficient method to calculate the Laplacian in orthogonal coordinates can be implemented by the use of differential forms nations such as Spherical and ey 3.5 Example In spherical coordinates the differential of arc length is given by (see equation 2.21) ds? = dp? + pdb? + p? sin? ade. Let @ = dp = pao ® = psinadd. (3.6)36 CHAPTER 3, CONNECTIONS Note that these three I-forms constitute the dual coframe to the orthonormal frame which just derived in equation( 3.5). Consider a scalar field f = f(p,0,¢). We now calculate the Laplacian of f in spherical coordinates using the methods of section 24. To do this, we first compute the differential df and express the result in terms of the coframe. = Bains Bang 2 df = ae pt aa wot ag? of tins Lp ra 6 pain 06 The components df in the coframe represent the gradient in spherical coordinates. Continuing with the scheme of section 2.4, we fist apply the Hodge in terms of wedges of coordinate differentials so that we can apply the definition of the exterior derivative, ‘operator. Then we rewrite Sor ng 1 gp gry 1 For i Op poe sind 86: oo “= = sind Ldn dd — pain Edn do pind Lp dd nd 06 = 2 - oy on = psind Ede A dé —sind one a dp Add : 2 of dvdf = noe Ze do te Soi of Drawn dnndos gid Ad ip dO Finally, rewsting the diferentals back in tenn of the the eoframe, we ge a (at ae a fine mush + u noes AI One So, the Laplacian of f is given by ee oN (Ors A a Ee Bop Op. 8 “The derivation of the expression for the spherical Laplacian through the use of differential forms is elegant and leads naturally to the operator in Sturm Liouville form, ‘The process above can also be carried out for general orthogonal transformations, A change of coordinates 2! = 2'(u") leads to an orthogonal transformation if in the new coordinate system u', the line metric gui(d)? + (du)? 8) only has diagonal entries. In this ense, we choose the coframe O° = Vi! = @ = Vgndu? = hy = yijadu® = higdu® ‘The quantities (Fy, fa, ha} are classically called the weights. Please note that in the interest of connecting to classical terminology we have exchanged twa indices for one and this will cause small discrepancies with the index summation convention, We will revert to using a summation symbol3.2, CURVILINEAR COORDINATES 37 when these discrepancies occur. To satisfy the duality condition 4'(e;) = 4!, we must choose the corresponding frame veetors ¢, according to 2 = am Ou ~ hy 8 ‘(u*), Then Gradient. Let = Sle!) and 2 y= Sack = «(fe (3.9) F FO + Fol" + Fy (hi Fi}du! + (hy Fa)du® + (hg Fa)du? {hP):du', where (hi); iF LARP) OOP) a a dal 4 AME RF ay na = Par [ aur owt | TL atbPy oP ( 1_alhaFa) _ haf) ar hy dF = ch [ | = (T x Fo A(hsFs) _ (hs Fi) out aus ae - ae 1 OU Fi) _ Otho) Taha! out Out) hg Tahal ae But ) (3.10) Divergence. As before, let F = Fi" and recall that V-F = 4d F. The computation yields P “FE Fol + Fup? + Fo FAAP + Pe Ae + Fe AP (toh )atu® A du + (hy hhgFa)du A du! + (hha Faded A du » _ fAliehaFi) | Alhiha Fs ade [ ult Gut + er du! A du? Adu?38 CHAPTER 3, CONNECTIONS ‘Therefore, 1_fahetaFi) , atts) Vi Pade P= Te at On ate on oe 3.3 Covariant Derivative In this section we introduce a generalization of directional derivatives, The directional derivative ‘measures the rate of change of a function in the direction of a vector. What we want is a quantity which measures the rate of change of a veetor field in the direction of another. 3.6 Definition Let X be an arbitrary vector field in RY. A map Vx : T(R") + TCR”) is called Koszul connection if it satisfies the following properties 1. Vyx(¥) = £0x¥, 2. Vosexnh = Tay + Va} 3. Ve( $9) = Tari +a, 4, Uy SY = XUV 4 SOxY for all vector fields X,X), X2,¥,¥i, Y2 € T(R") and all smooth functions f, The definition states that the map Vy is linear on X but behaves like a linear derivation on Y. For this reason, the uantity Vir¥ is called the covariant derivative of ¥ in the direction of X. 3.7 Proposition Let ¥ = fiz be a vector field in R” , and let X another C™ veotor field. ‘Then the operator given by Tx = er piy 2 ger (3.12) defines a Koseul connection. ‘The proof just requires verification that the four properties above are satisfied, and it is left as an exercise. ‘The operator defined in this proposition is called the standard connection compatible with the standard Euclidean metric. The action of this connection on a, vector field ¥ yields a new vector field whose components are the directional derivatives of the ‘components of ¥ 3.8 Example Let 8 hted ‘then, = Pd a8 VxY = Xe Voe t Xlev gy 8 ry g oz (ety a fe text) oer egy = beg )t zeae Nae + leet wi) + 23,6 We May feo Or ~ Oy 3.9. Definition A Koszul connection Vy is compatible with the metric g(¥,Z) =< ¥,Z > if Vx S< Ux, 2 >4 (3.13)3.3, COVARIANT DERIVATIVE 39 In Euclidean space, the components of the standard frame vectors are constant, and thus their rates of change in any direction vanish. Let ¢ be arbitrary frame field with dual forms #. ‘The covariant derivatives of the ftame vectors in the directions of a vector X, will in general yield new vectors. The new vectors must be linear combinations of the the basis vectors wh (Xe $45 (Xe HA Mea whX)e Hee Hoes eYX)ey H64(XJer Heh (XJea (4) The coefficients can be more succinctly expressed using the compact index notation Vxei P(X) (3.15) 11 follows immediately that _ w(X) = OF ves) (3.16) Equivalently, one can take the inner product of both sides of equation (3.15) with ex t0 get = WIN) < epee > = WX aan Hence _ << Vaeisen >= wn X) (17) ‘The left hand side of the last equation is the inner product of two vectors, so the expression represents an array of functions. Consequently, the right hand side also represents an array of inctions, In addition, both expressions are linear on X, since by definition Vx is linear on X. We Jude that the right hand side can be interpreted as a matrix in which, each entry is a L-forms acting on the vector X to yield a function. The matrix valued quantity w, is called the conection form 3.10 Definition Let Ty be » Koszul connection and let fe} be a frame. The Christoffel symbols associated with the connection in the given frame are the functions Py given by Vees = Dijee (3.18) ‘The Christoffel symbols are the coefficients which give the representation of the rate of change of the frame vectors in the direction of the frame vectors themselves. Many physicists therefore refer to the Christoffel symbols as the connection once again giving rise to possible confusion, ‘The precise relation between the Christoffel symbols and the connection 1-forms is captured by the equations wile) = Ty, (3.19) or equivalently Phyo (3.20) Inna general frame in R there are n? entries in the connection 1-form and n® Christoffel symbols The number of independent components is reduced if one assumes that the frame is orthonormal 3.11 Proposition Let and ¢; be an orthonormal frame and Fx be a Koszu connection compatible with the metric . Then aja = wg (21)40 CHAPTER 3, CONNECTIONS Proof Since it is given that < e;,¢5 >= dij, we have 0 = Vx =< Vxenes> + <64,0x6) > = + = wh teh = ota + eho = atu thus proving that w is indeed antisymmetric. 3.12 Corollary ‘The Christoffel symbols of a Kosaul connection in an orthonormal frame ate antisymmetric on the lower indice; that is Phy = Ty (3.22) We conclude that in an orthonormal frame in R" the number of independent coeficents of the connection L-form is (1/2)n(n ~ 1) since by antisymmetry, the diagonal entries are zero, and one only needs to count the number of entries in the upper triangular part of the m x n matrix wij Similaely, the number of independent Christoffel symbols gets reduced to (1/2)n®(n— 1). In the case of an orthonormal frame in R® , where gi; is diagonal, «', is also antisymmetric, so the connection equations become _ [3] 0 whixy [3] Vx|e|=| ea ‘0 8x) |] @ (3.23) “3 w(K) 3X) “0 2 Comparing the Frenet frame equation( 1.27), we notice the obvious similarity to the general frame equations above. Clearly, the Frenet frame is a special case in which the basis vectors have been adapted to a curve resulting in a simpler connection in which some of the coefficients vanish. A. further simplification occurs in the Frenet frame since here the equations represent the rate of change of the frame only along the direction of the curve rather than an arbitrary direction vector X. 3.4 Cartan Equations Perhaps, the most important contribution to the development of Differential Geometry is the work of Cartan culminating into famous equations of structure which we discuss in this chapter. First Structure Equation 3.13 Theorem Let {c;} be a frame with connection «} and dual coframe @. ‘Then aw Of Sdih+ o (3.24) Proof: Let 6 =; bbe a frame, and let 6 be the corresponding coframe. Since 6(e,), we have w= Nidzl3, CARTAN EQUATIONS a Let X be an arbitrary vector field, ‘Then = Fatal) a;X(4i) = Ad AY\X) ef AA A)(X) (ahaa), 5, " Hence, = (Yaa), oor in matrix notation w= Ada (3.29) On the other hand, taking the exterior derivative of 8, we find d= d(A“' Ade! aah), nA d= d(A')AND. we have d(A~}}A = —A“MA = —w, hence d= -2 0 (3.26) But, since A~E In other words a0 +4546) =0 Second Structure Equation Let * be a coftame in R" with connection w';. Taking the exterior derivative of the first equation of structure and recalling the properties (2.34) we get aad) + du’, 06 {AO — wt, Ad =o Substituting recursively from the first equation of structure, we get dul, A ahs A (wh, N88) dal AO Hh, Auch AO! (di, tak, Aw) AB! dai, uly Aw 0. We now introduce the following 3.14 Definition ‘The curvature 2 of a connection w is the matrix valued 2-form 2 fy duly tol, Aw (3.27) 3.15 Theorem Let # be a coframe with connection w in R” . The the curvature form vanishes (3.28) Q=dotwna CHAPTER 3, CONNECTIONS Proof Given the there is a non-singular mateix A such that @= A~Mdr and w = A~MdA, have (A) AAA On the other hand, ww = (AMA) A (ATMA) RE and y(v!, v?) : V — R® be two coordinate charts with a non empty intersection (x(U) N y(V)} # B. The two charts are said to be C™ equivalent if the map 6 = y~!x and its inverse @~ (see fig 4.1 Jaro infinitely differentiable, In more lucid terms, the def just states that two equivalent. charts x(w*) and y(v*) repre- sent different reparametrizations for the same set of points in R® Figure 4.1: Chart Equivalence 4Ba CHAPTER 4, THEORY OF SURFACES 4.3. Definition A differentiably smooth surface in R is a set of points M in R® such that 1. If p © M then p belongs to some C® chart. 2. Ifp EM belongs to two different charts x and y, then the two charts are C™ equivalent. Intuitively, we may think of a surface as consisting locally of number of patches "sewn” to each other so as to form a quilt from a global perspective. “The first condition in the definition states that each local patch looks like a piece of R?, whereas the second differ y condition indicates that the patches are joined together smoothly. An other way to state this idea is 1o say that a surface a space that is locally Euclidean and it has a differentinble structure so that the notion of differentiation makes sense. If the Euclidean space is of dimension n, the *surface” is ealled an n- 4.6. bdf The element of arclength ds? = gagdu" @ du? (49) is also called the first fundamental form. We must caution the reader that this quantity is not a form in the sense of differential geometry since ds? involves the symmetric tensor product rather than the wedge product. The first fundamental form plays such a crucial role in the theory of surfaces, that will find it convenient to introduce yet a third more modeen version. Following the same development as in the theory of curves, consider a surface .M defined locally by a function (u!, 2) a(u!,u2). We say that a quantity Xp is a tangent vector at a point p € M, if Xp is a linear derivation on the space of C® real-valued fimetions {ff :.M —+ R} on the surface. ‘The set of all tangent vectors at a point As before, a vector field X’on th choice of a tangent vector at each point on the surface and the union of all tangent spaces is called the tangent bundle TM. ‘The coordinate chart map surface is a smooth a:R?— MeR®46 CHAPTER 4. THEORY OF SURFACES induces a push-forward map ya TM defined by (VA) laguey= V(@ 2S) law Just as in the case of curves, when we revert back to classical notation to describe a surface as 2i(u'), what we really mean is (20.4)(u®), where « are the coordinate functions in R® . Particular examples of tangent vectors on M are given by the push-forward of the standard basis of TR®. ‘These tangent vectors which earlier we called xq are defined by a a Gra MA) baer Gast ef) he In this formalism, the frst fundamental form I is just the symmetric bilinear tensor defined by induced metric WAXY where X and Y are any pair of vector fields in =< XY >, (4.10) Orthogonal Parametric Curves Let V and W be vectors tangent to a surface M defined locally by a chart x(u*). Since the vectors Xaipha Span the tangent space of M at each point, the vectors Vand IW" can be written as linear combinations Vex. Wx, ‘The functions V* and 1V* are called the curvilinear coordinates of the vectors, We can caleulate the length and the inner product of the vectors using the induced Riemann netic, wie = = WOVE < xa x9 > IWIF = gaa¥°V IWIP = gop’, and < Vx, WExa >= VOW < xa, X3 > gaaVW?. ‘The angle @ subtended hy the the vectors v and Wis the given by the equation TT WW) JV, VVW im GV VE Goa WF cos = Let u® = 6(t) and wu" (8) be two curves on the surface «8 and jue oY aun = at, and bu = Sa4.2, THE FIRST FUNDAMENTAL FORM a7 represent infinitesimal tangent vectors (1.12) to the curves. ‘Thus, the angle between two infinitesimal vectors tangent to two intersecting eurves on the surface satisties dagu bul [Gap bid Gash BuP In particular, if the two curves happen to be the parametric curves, w! = constant and 1 constant then along one curve we have du! = 0, du? arbitrary, and along the second we have dur arbitrary and du? = 0. In this case, the cosine of the angle subtended by the infinitesimal tangent vectors reduces to cos = (4.12) gioduldu? Joist) Joaatrye ‘As a result, we have the following (4.13) 4.7 Proposition ‘The parametric lines are orthogonal if P = 4.8 Examples a) Sphere x = (asind cos, asind sind, acos6) xe = (acos cos, acosésing, —asin®) xs = (-asin#sind,asinBeos 6,) B= xeexpsa? Fos xe-xy G = x, ds? = ade? +a" sin? 0dé? b) Surface of Revolution x = (reas@,rsin8, f(r) x, = (cos 0,sind, (0) Xe = (-rsind,rcos?,0) Es xox = 14+ f%r) F = x -x=0 G = mar dst = [Lt fel]? + ae? «} Pseudosphere = (sin weos asin wsin v,a(cos wintans)) E = abeot?u F G = asin?u ds? = a’ cot? udu? + a) Torus48 CHAPTER 4, THEORY OF SURFACES x = ({b+acos u) cosy, (b+ acos u) sin v, asin u) eee ae a0) G = (b+acosu)® ds? = a®du? + (b+ acosu)*dv ©) Helicoid x = (weose, usin v,av) el reo wee? du + (u? + aye? 1) Catenoid an " dst = 4.3. The Second Fundamental Form Let x = x(u%) be a coordinate patch on a surface .M. Since x, and xv are tangential to the surface, ‘we can construct @ unit normal n to the surface by taking ne BRR 7 Ie xxl om Now, consider a curve on the surface given by u* = 1%(s). Without loss of generality, we assu that the curve is parametrized by arclength s so that the curve has unit speed. Using the chain rule, we se that the unit tangent vector 7’ to the curve is given by dx _ dx du® _ du* de te ds ds (4.15) Since the curve lives on the surface and the the vector 7’ is tangent to the curve, it is elear that T is also tangent to the surface. On the other hand, the vector 7” = d7'/ds does not in general have this property, so what we will do is to decompose 7 into its normal and tangential components (see fig (42) To = Kytky inom + By (4.16) where Ky = [Kol] =< Tm > The scalar quantity sy is called the normal curvature of the curve and Ky is called the geodesic curvature vector, The normal curvature measures the the curvature of x(u°(s)) resulting,4.3, THE SECOND FUNDAMEN’ AL FORM 49 Figure 4.2: Normal Curvature by the constraint of the curve to lie on a surface. ‘The geodesic curvature vector , measures the “sideward” component of the curvature in the tangent plane to the surface. Thus, if one draws a straight line on a flat piece of paper and then smoothly bend the paper into a surface, then the straight line would now acquire some curvature. Since the line was originally straight, there is no sideward component of curvature so Ky = 0 in this case. This means that the entire contribution to the curvature comes from the normal component, reflecting the fact that the only reason there is curvature here is due to the bend in the surface itself Similarly, ifone specifies a point.p € M and a direction vector Xy € T).M, one can geometrically envision the normal curvature by considering the equivalence class of all unit speed curves in M which contain the point p and whose tangent vectors line up with the direction of X. Of course, there are infinitely many such curves, but at an infinitesimal level, all these curves can be obtained by intersecting the surface with a “vertical” plane containing the vector X and the normal to M Alll curves in this equivalence class have the same normal curvature and their geodesic curvatures vanish. In this sense, the normal curvature is more of a property pertaining to a direction on the face at a point, whereas, the geodesic curvature really depends on the curve itself. It might be impossible for a hiker walking on the ondulating hills of the Ozarks to find a straight line trail, since the rolling hills of the terrain extend in all directions, However, it might be possible to walk on a path with zero geodesic curvature as long as the hiker can maintain the same compass directio ‘To find an explicit formula for the normal curvature we first differentiate equation (4.15) ‘Taking the inner product of the last equation with the normal and noticing that <>xq,m >= 0, we get ar good where bog =< Xagm > (4.18) 4.9 Definition ‘The expression TT = bagdu® du (4.19) is called the second fundamental form50 CHAPTER 4, THEORY OF SURFACES 4.10 Proposition The second fundamental form is symmeteic. Proof: In the classical formulation of the second fundamental form the proof is trivial. We have Jaa since for a C™ patch x(u2), we have Xa = Xsq because the partial derivatives co We will denote the coefficients of the second fundamental form by Daa te. 6 = bn fF = bie =< xuem> 0 that equation (4.19) can be written as IT = edu? + 2fdude + gdv (4.20) and equation (4.17) as 1 _ Edu? +2Fdudv + Gav? 80 T= eee Bilao + ge a We would also like to point out that just as the first fundamental form can be represented as T =< dx,dx>, so can we represent the second fundamental form as I] =— ‘To see this it suffices to note that differentiation of the identity < xq, >= 0 implies that < Xap M >= — < Xamy > ‘Therefore, = = udu? = =< Xqp,m> dude" = - From a computational point a view, a more useful formula for the coefficients of the second fundamental formula can be derived by first applying the classical vector identity (Ax B)-(C xD) ee Fa (4.22) ‘ocompute Ix eff? (xu Xo) = (Ru Hy) = «(EE OE] oe (4.23) Consequently, the normal vector can be written as Tel” VEG=4d, CURVATURE BL ‘Thus, we can write the coefficients bay directly as triple products involving derivatives of (x). ‘The expressions for these coefficients are p= HeXku JEG— (x Xe) ; oe ee (424) ‘The first funclamental form on a surface measures the (squate) of the dlistance hetween two infinitesimally separated points. ‘There is a similar interpretation of the second fundamental form as we show below. The second fundamental form measures the distance from a point on the surface to the tangent plane at a second infinitesimally separated point. To see this simple geometrical interpretation, consider a point xp = x(ug) €.M and a nearby point x(uj + du). Expanding on a Taylor series, we get x(ug + du") Xo + (xa)adu + Hx) optud! + We recall that the distance formula from a point x to a plane which contains xy is just the scalar projection of (x — xo) onto the normal. Since the normal to the plane at xq is the same as the unit normal to the surface and < x,,m >= 0, we find that the distance D is D = = 1 (xojaprm> dudu? = F< (oagm > dre 1 = lle [he first fundamental form (or rather, its determinant) also appears in calculus in the context of calculating the area of a parametrized surface. the reason is that if one considers an infinitesimal parallelogram subtended by the vectors x,du and x,dv, then the differential of surface area is given by the length of the cross product of these two infinitesimal tangent vectors. That is dS = Ijxy x x|| dude [ [ VEGF aa ‘The second fundamental form contains information about the shape of the surface at a point. For example, the discussion above indicates that if 6 = |bas| = eg — f2 > 0 then all the neighboring points lie on the same side of the tangent plane, and hence, the surface is concave in one direction, If at a point on a surface 6 > 0, the point is called an elliptic point, if 6 < 0, the point is called hyperbolic or a saddle point, and if b= 0, the point is called parabolic 4.4 Curvature Curvature and all related questions which surround curvature, constitute the central object of study in differential geometry. One would like to be able to answer questions such as, what quantities retain invariant as one surface is smoothly changed into another? ‘There is certainly something intrinsically different from a cone, which we can construct from a flat piece of paper and a sphere which we can not, What is it that makes these two surfaces so different? How does one caleulate the shortest path between two objects when the path is constrained to be on a surface?52 CHAPTER 4, THEORY OF SURFACES ‘These and many other questions of similar type can be quantitatively answered through the study curvature, We cannot overstate the great importance of this subject; perhaps it suffices to say that without a clear understanding of curvature, there would not be a general theory of relativity, xno concept of black holes, and even more disastrous, no Star Trek ‘The study of curvature of a hypersurface in R” (a surface of dimension n — 1) begins by trying to understand the covariant derivative of the normal to the surface. The reason is simple, If the normal to a surface is constant, then the surface is a flat hyperplane, ‘Thus, it is variations in the normal that indicate the presence of curvature, For simplicity, we constrain our discussion to surfaces in R® , but the formalism we use is applicable to any dimension, We will also introduce in this section the modern version of the second fundamental form 4.11 Definition Let X be a vector field on a surface M in R® , and let.’ be the normal vector. ‘The map L given by LX =-VxNV (4.25) is called the Gauss map In this definition we will be careful to differentiate between operators which ive on the surface and operators which live in the ambient space. We will adopt the convention of overlining objects hich live in the ambient space, the operator ¥ above being an example of one such object. ‘The Gauss map is clearly a good place to start, since it is the rate of change ofthe normal in an arbitrary direction which we wish to quantify 4.12 Definition ‘The Lie bracket [¥',¥] of two vector fields X and ¥ on a surface M is defined as the commutator [X.Y]=XY =YX, (4.26) (YA) = YX). 413 The Lie bracket of two vectors X,¥ € T(ealM) is another vector in T(M). Proof: If suffices to prove that the bracket is a linear derivation on the space of C™ fu Consider vectors X,Y, Z € T(calM) and smooth functions fg in M. ‘Then IY HZ) = XU +2) - 0 42K) XW UN + 210) = VX) = 20K) X(V(A) = YX) + XZ) = ZK) POY] +X. 2100, meaning that if f is a function on M then [X,Y] and X+¥, 2) (X + Y) ZN) — ZUX + YN) X(Z(D)) + YVAN) ~ 2X) ~ 2D) N(Z(f) ~ Z(X(N) + (ZU) — ZV) 1X, 216) + 2100) so the bracket is linear on each slot. Furthermore (Y)S +a) = XY(E +9) -YIXUS +9) X(V(N) + ¥(a)} — V(X) + X(@)) X(Y (EY) — YONA) + XY (a) — YEN) = YF Io),4d, CURVATURE 53 (WX. ¥]tf9) X{Y (Fay) — Y(X(F9)) XTFY (9) + 9¥ (4) — YUFX(9) + 9X) X (SPY (g) + FX(Y (9) + XY) + XUV) HYD (9) — V(X) — YX) = 9 (XU) = AX) — OX) + XO) — YX = FX Ila) + 9X. YI, «that the bracket acts as linear derivation of functions. 4.14 Proposition ‘The Gauss map isa linear transformation on 7(M), Proof: Linearity follows from the linearity of ©, s0 it suffices to show that Ls X —> LX maps X € T(M) to a vector LX € T(M). Since NV is the unit normal to the surface, < N,V >= 1, 30 that any derivative of < N,N > is. Assuming that the connection is compatible with the metric, Vx + 2 = 2=0. Therefore, LX is orthogonal to V, and hence it lies in TM). We recall at this point that the in the previous section we gave two equivalent definitions < dx,dx >, and < X,¥ >) of the first fundamental form. We will now do the same for the second fundamental form. 4.15 Definition ‘The second fundamental form is the bilinear map, I(X,¥) =< EXY > (427) 4.16 Remark It should be noted that the two definitions of the second fundamental form are consistent, This is easy to see if one chooses X to have components x, and Y to have components x». With these choices, LX has components —n, and 11(X,¥) becomes bag = — < Xa,ms > We also note that there is third funclamental form defined by TIX, Y) =< LX,LY >=< LEXY > (4.28) In classical notation, the thitd fundamental form would be denoted by < dn, dn >. As one would expect, the third fundamental form contains third order ‘Taylor series information about the surface. We will not treat 11/(X,¥) in much detail in this work. 4.17 Definition ‘The torsion of a connection V is the operator T such that YX.Y T(X,Y) = Fx — Pv X — [X,Y] (4.29) A connection is called torsion free if T(X,Y) = 0. In this case, UxY PrN =1N,¥) We will say much more later about the torsion and the importance of torsion free connections. For the time being, it suffices to assume that for the rest of this section, all connections are torsion fice, Using this assumption, it is possible to prove the following very important theorem,or CHAPTER 4, THEORY OF SURFACES 4.18 Theorem The Gauss map is a self adjoint operator on TM. Proof: We have already shown that L: TM —+ TM is a linear map. Recall that an operator Lona linear space is self adjoint if < LX,¥ >=< X,LY >, so that the theorem is equivalent to proving that that the second fundamental for is symmetric (/1[X,Y] = 1[¥,X]). Computing the difference of these two quantities, we get MIX,Y]- HX] = -< LY,X> <¥xNY>-< Ty 5s Since < X,N >=< ¥,N 0, and the connection is compatible with the metric, we know that = = hence MX,Y]- 1.x] = -, 0 Gif [X,¥1E TM) One of the most important topics in an introductory course linear algebra deals with the spectrum of self adjoint operators. ‘The main result in this area states that if one considers the eigenvalue equation LX 6X (4.30) then the eigenvalues are always real and eigenvectors corresponding to different eigenvalues are or ‘thogonal. In the current situation, the veetor spaces in question are the tangent spaces at each point of a surface in R® , so the dimension is 2. Hence, we expect two eigenvalues and two eigenvectors LX, Xs mX (4.31) wiX2, (4.32) 4.19 Definition ‘The eigenvalues ¢y and sy of the Gauss map L are called the prineipal eur vatures and the eigenvectors Xi and X» are called the principal directions. -veral possible situations may occur depending on the classification of the eigenvalues at each point p on the surface: 1. Ie: # xy and both eigenvalues are posi ve, then p is ealled an elliptie point 2 Weepee <0, then p is called a hyperbolic point, 3 ey 12 £0, then p is called an umbilic point. 4. iff mW» = 0, then p is called a parabolic point It is also well known from linear algebra, that the the determinant and the trace of a self adjoint operator are the only invariants under a adjoint (similarity) transformation. Clearly these invariants are important in the case of the operator L, and they deserve to be given special names. 4.20 Definition ‘The determinant & = det() is called the Gaussian curvature of M and H = (1/2)Tr(L) is called the mean curvature.4d, CURVATURE 55 Since any self-adjoint operator is diagonalizable and in a diagonal basis, the matrix repres Lis ding(c, m2) follows immediately that K = mm Ho= Heston) (4.38) 4.21 Proposition Let X and Y be any linearly independent vectors in 7M). ‘Th LX x LY (LX XY) + (X x LY) K(X «¥) 2H(X x) (4.38) Proof Since LX, LY € T(M), they can be expresses as linear combinations of the basis vectors X and Y IX = aX+hY LY = aX +be¥ computing the cross product, we get a by LX & LY oo ee (EX XY) + (Xx LY) = (a tho\(X x) det{E)(X x ¥). Similarly ‘THEY(X *Y) (@2HVIX x) 4.22 Proposition K = Hn (438) 2 Proof: Starling with equations (4.34) take ¢ vector identity (1.22). We immediately get36 CHAPTER 4, THEORY OF SURFACES ‘The result follows by taking X =x, and 4.23 Theorem (Buler) Let X; and Xz be unit eigenvectors of L and let X = (cos )X1+ (sin 8)Xa, ‘Then in? IN(X,X) Prooft Easy, Just compute 11(X,X) =< LX, >, using the fact the LX, = wiX1 , LY: and noting that the eigenvectors are orthogonal. We get, 1608? Ow: X > = < (cond) Xi + (sinO)e2N2, (c098)X1 + (sin 4).X2 > = 1008? < Xi, Ni > teas = pcos? 0+ Kysin? a,