Professional Documents
Culture Documents
(Series On Knots and Everything 4) John Baez, Javier P Muniain - Gauge Fields, Knots and Gravity-World Scientific (1994)
(Series On Knots and Everything 4) John Baez, Javier P Muniain - Gauge Fields, Knots and Gravity-World Scientific (1994)
dy ‘which would amount toa left-hand rule. In shost, the ‘right-hand rule’ nonsense enters when ve unnaturally try to make the product of two clements of V to come out to an element of V, instead of M'V. This is noted in some physics books, where they say thatthe cross product of two vectors is “prewdovector or ‘axial vector! rather than a true sector. We prefer to say that the wedge product of 2 vectors lies in DV — thie is true in all dimensions Exterior algebra isan interesting subject in itself, ut we do na just vrant to generalize the cross product of vectors; we want to generalize the cross product of vector feds. Actually as aleady mentioned, it i sich better to take products of Iforms! We will do this by copying ‘our construction of AV, with the smooth functions Cm(M) on some manifold M taking the place of the real numbers, and the I-forms (Af) taking the place of the vector space V. Namely, we define the differential forms on M, denoted (A), tobe the algebra generated by 0°(i) with the relations wAp=—Bhe for all, € 9"(M). Tobe precise, we should emphasize that we form (A) as an algebra “over C™(M)’. Thi means, fest of al, that (1) consists of linear combinations of wedge products of Lforms with fune~ tons 28 corfciens, We allow all locally finite linear combinations, that is, shove for which every point pin AF has a neighborhood where only finitely many terms are soazero. Secondly, it means that 9(M) fatisies the rules of an algebra with functions taking the place of num bers. Maybe we should say again what all these rules are. We bave, for alluyy,o € MM) and fg © C=(M), wes ate, wt (uty (otatn wAluan) (wauyan,80 1M Differential Forms wAnteny (wbulAvawrutpny, fae, Rute) = fatto, (sta = fatow. We define the 0-frms, 0°(A), to be the fonctions themselves, and define the wedge product of & function with a differential form to be the ordinary product: / nw = fu. We define the product of a number ‘cand a diferential form w to be the product of the constant function € (P(A) and w. Elements that are linear combinations of products ‘of p Lform are called pforms, and we write the space of forms on Mas P(A), We have (a) = HU. Por example, sxppote M ke IRE. The Oforms on IR" ace just functions, f “The 1-frms all Look like oydat where the coefcients wy are functions, It is easy to check that the Dorms al ook lke Femdet nda where we have put in a factor of 3 because dat det = de" A doh [Also for this reason, we may aa well assume that ww = —Wy- Then on IR for example, we have w= wrads! 4 det Funda! Ade? Funds? Adet Sinilaely the 3-forms lok lke J eyeadet Ade" nds, a sand we may as well assume that djs is totally antisymmetric (that i ‘itches sign when we switch any tio indies). On IR? we get wands! nde? adel Piforms 6 ‘There are no nonzero 4-frms, forms, ete, on RY. In general, there are no nonzero pforms on aa n-dimensional manifold if» > m ‘We leave it forthe reader to show some importaxt facts about dif {erential forms in the following exercises. Exercise 48. Given a vector space V, show tat AV sa graded commu tative or aupercommutative algebra, that fv © APY and w © AYV, then ens (UPaas. ‘Show thet for any manifold I, QE) se graded commutative Exercise 47. Show that differential forms ore coneaveriant. That 5 show thet f$:M— 2 isa map from the manifold M to the manifold 1, there is unique pullback map e290) (4) ring with the usual plllack on O-forme (functions) and f-forms, and satiaing eeu) = agte Swtw = Soren wean) = Songn for allw.u € 9(8) and a ER, Exercise 48. Compare how Lforms and 2forms on BR? tanaform wnder pavity. That is, ft Ps»! be the map Playa) =(-2)-1-2) Imnown asthe ‘parity transformation’. Note tht P maps right handed bases to lefthanded bases and vice verse, Compute (a) when w i the 1form Gud, and when 4 the form heya de® de. In physic, the electri eld Bp called a vector, while the magnetic field Bie called an axial vector, because B changes sign under parity transformation, while B does not. Tn Chapter § we wil se tha itis bes to think of Ue lect td as form on space, andthe magnetic field ay a 2form, In other words, while we may be sed to thinking of«2 14 Differential Forms B = (Be, By, Bx) and B = (By, Byy Be) a8 vector fils, itis better to B= Buda + Eyly + Bale and B= Buty dz + Bye hd + Bye dy By the aboe exercise, this means that they transform diflerently under patity I the reader is frustrated because exterior algebras and difleren- tial forms wer dificult to isualise, we suggest taking © peck ahead to Figures and 4 of Chapter 5. Grassman, the inventor of the ex terice algebra, vitualied a wedge product & \--- A vy as an oriented parallelepiped with sides given by the vectors v,..,t. One must be Careful, however, because the wedge product of I-forme corresponds to 1 parallelepiped in the cotangent space. The Exterior Derivative ‘We know from the first section of thie chapter that the diflereatial is @ nice way to generalize the goed old ‘gradient’ to manifolds. As we sam, the diferential ofa function, or O-form, is « form. Now we will show how to take the diferential af pform and get a (p-+ 1)form: .90(M) + OFAN, ‘This will et us generalize the gradient, the curl and the divergence in one fell swoop, and tee that they are secretly all the same thing. The big clue is that the curl of a gradient i zero: Vx(vpa0 ‘This muggests that we make d satisfy d(df) = 0 for any function f. ‘Another clue is that the various product rales VUls) = (Wf)9+1¥e Ux(fy) = Us xot sxe Vi(fe) = Vive fiw Ve(exw) = (V-vje-vVw ‘The Exterior Derivative 8 should all be special cases of some sort of Leibuis law for differential forms, Since the diferental forms are graded commutative, it turns ‘out that we need a graded version of the Leibniz law. ‘Alerscratching our head fora while, wedefine the exterior deriva tive, or differential, to be the unique set of maps E07(M) + OF*1(0) sch that the following properties hold 1) ds9°UM) -» 0°() agrees with our previous definition. 2) dw +p) = eo + dp and (ov) = ed forall w,y © O(M) and eR. 3) d(whn) = don (Pedy for alle € (A) and w € (81). 4) dds) = 0 forall (4), ‘Toshow that these properties wniquely determine the exterior derive tive, one just needs the fact that any I-form is a locally finite linear combination of thove of the fotm df (with Functions as coeficients). This fact i easy to eee on IR", and can be shown in general axing charts. Then to ealculated of any difleratial form, say fag Nb, swe just ure cules 1) ~ 4): A fdg Nah) = af n(dg ndh) + f ndldg nh) A dg dh + faldg) dh — fag A dfdh) ido nah, "To show that d with these properties is actually well-defined, it euices (by the black magic of algebra) to show that this way of calculating d Js compatible with the eelations in the definition of differential forms. ‘The mos: important one of these is the aaticommutative law wAps naw for forms. Bor d to be well-defined, it had better be true that calew lating dlw A p) gives the same ansner as dlp hw). This i where theot 14 Differential Forms traded Leibne lave is necessary: when w and are I-forms, we have dwn) = dle) Ada nwt nde, wh dp tdo ny ae i) Let us caleulate the exterior derivative of Lforms and 2-forms on IRD, Taking any I-frm meade + ody tad, we gt hy = dag de + day Ady + deh de, hence y the rule for d of function and a ttle extra work hs = (Bj Bay) dy de (Bs Oat) rds +(asy ~Byue Ady In other words, the exterior derivative of « I-form on IR? is esentally just the curl! We need right-hand to define the evr, hoeever, while the ‘exterior derivative involves no right-hand rule. This is because d of a ‘form is «2 form; the right-hand rule only comes in when one tres $o pretend that this 24orm ia Iform, using Uhe star operator as follows: ada = (By And, as noted, this pretense is only possible in 8 dimensions, while we can taked of a -form in any dimention: ~ Dy) da + (Oana ~ Ose) + (Bey ~ Bie Exe given by 49, Show that on RO the exterior derivative of any 1form 3 elute") = Baad 4 de Next, taking a 2-form on yd A dy + pay Nd2 + ads Adz ‘The Exterior Derivative 65 we get ay = dey 8 42 8 dy + dye Ny Ndz + dee A de hd = Biinyhs 32 8 dy 4 Bagels dy f d2 + Oydrdy N de dz (Basey + Bie + Ojra)dE N dy de ‘Thus the exterior derivative of a 2form on Ris just the divergence in disguise, In short, the exterior derivative has as special cases the following familiar operator: © Gradient d:9°(R) + 0°(R°) © Gurl &:a CR) -- PCR) © Divergence &:99(R!) + 9°(R) Tn fact, there isa simple formala for the exterior derivative of any Aifereatial form on IR". Let 1 stand for a multivindex, that is, a tuple (j,..-sip) of distinc integers between I and n. Let do! stand for the pform aah hen fon IRY. Then any p-form on I" can be expressed as wurde! ‘where following the Einetin summation convention we sum over all rmultiindices - We have he = day Nae by the Leibnie law, since (da!) = 0 (as can eaily be checked). More concretely, using the formula for dof a function, we have ty = (Oysr) da hd. Using this formula it it easy to derive an amazing identity alaw)66 14 Differential Forms {for aay difereatial form on IR", Just compute Alda) = dl@yurdo? ode!) = AByurde” A det A de! and note that on the one hand Ada = 3,2 lay the equality of mixed partials, but on the other hand det nde = dah nd” boy the anticommutative law. With a little thought one can see this means that d(as) i equal to the negative of itself, soit is ero. This rule is 9 important that people often write it as fu=0 cor even just e (a ID, d act like the gradient on 0-orms, the curl on Iforms and the divergence on 2-forms, a0 the identity d? = 0 contains within it the identities vxivA= and vv xe) But this identity is better, since it applies to differential forms in any dimension. In fact it applies to any manifold! Here is an easy proof that does not use coordinates. By definition, any pform on a manifold ia linear combination — with constant coeficients — of forms ike w= fads No al So it suffices to prove the identity for pforms of this sort. We have y= do Ndfi A hdl ‘The Exterior Derivative er by the Leibnis law and the fact that d{@f) = Ofor any function, Using the Leibni law and def) =O again, we obtain (du) = 0 11 turns out that the identity d® = 0 and its generalizations have profound consequences for physics slarting with Maxwell's equations tis alo the basis of a very important connection between geometry ad topology, called deRhara theory. We will explore therein Chapter 8. When we do it is important to remember that this identity is just ‘a way of saying that partial derivatives commute! As so often the case, the simplest facts in mathematics lie atthe zoot of some of the most sophisticated developments ‘We will ap up this section by showing that the exterior derivative is natural, We already discussed this for functions in Section 45 it simply meant that d commutes with pullbacks. In fact, this is tru for ‘lifereatial forma of any degree. In other words, for any map ¢:M — NV between manifolds, and any dilferental form w € O°(M), we have # (de) = dg) ‘The proof is easy. By Exercise 47, "is real-linear, 40 itsfces to treat the cate where B= hath ho Aa We then have, using Exercise 47 again together with the naturality of 1 on functions Olde) = Po Nd nerds) Fly hoo Ay BB fan AB hy MO fo NdB fio 46H) Goh $a KoA Gay) = Ag Uo nahi 6--Adh)) = doe) 1s desired.Chapter 5 Rewriting Maxwell’s Equations Hence space of itaelf, and time of tell, will sink into mere shadows, and only union of the fo shall servis. — Hermann Minkowski, ‘The First Pair of Equations ‘We now have developed enough differential gometry to generalize the fiat pair of Maxell equations, vo =o 2.08 vx 848 =o, to any manifold. We claim that they have a very beautiful form as & single equation in terms of diferential forme, fore giving away the answer, let us consider « special cae: the static ene. ‘Then we just have two equations for vector fields on space, cS “ . v-B=0, vxB=0 In the language of diferential forma, the divergence becomes the exte ror derivative on 2-forme on I. Thus, instead of treating the magnetic field as 2 vector field B = (Be, By, B) we wil treat it as the 2-form B= Baby hd + Byde he + Bude dy o0 1.5 Rewriting Maxwells Equations Similacy, the curl becomes the exterior derivative on I-forms on B®, 10 instead of treating the electric field at vector feld E = (Bs, By, Bs) wwe wil teat i as the I-form B= Buda + Eyl + Ba "The frat paie of static Maxwell's equations then become ae , aB Next consider the general time-dependent case. Now we must think ofthe electric and magnetic fields as living on spacetime. We begin by ‘working an Minkowski spacetime, R', using the standard coordinate System, which we wil number a2 (2°,24,24,2°). We will often write ‘instead of 2" for the time coordinate, and 2,9, 2 for the space cor: dinates (z,2?,2"). The electzic and magnetic feds are I-forms and 2-forms on IRS, namely B= Bade + Byly + Bade aad B= Bydy Adz + Bude de + Bud dy We can combine both fede into a unified electromagnetic field Pa 2form on BR, as follom: PoB+ENd I we want to look at all the components, raids” Ade, vee can write them out as 8 matric 0 -E, -E, 0 Be -B, Fe=l5 -B, 0 Be EL By -Be 0 ‘The First Par of Bquations n ‘The beauty of thie way of unifying the electric and magnetic fel ie that the feet pair of Masorll equations became simply aP =o, ‘To see this, fet note that dF =d{B+ Ad) = dB Hab nat ‘Then split up the exterior derivative operator into a spacelike part and a timelike part, Recall shat for any diferential form oe eve is = Ogu det nds! where J ranges overall multi-indioes and = 0,1,2,3. We can thus write du ob a sum of the spacelike part so = Burda’ nde! orhoze ranges over the epaelite’ indices 2,3, and the timelie past aE Oar = Br dod! ‘Phen we have aF = dBsdBnd UpB +d NAB (dob + dt NOE) Act 5B + (GB + dsB) Ad [Note thatthe iret term has no dt init while the second one docs, Also ‘ote that the eecoad one vanishes only ifthe expression in parentheses oes, I follows that dF = 0 is equivalent to the pair of equations 4B = 0, OB+dsB = 0. "Those are just the first pair of Maxwell equations in slighty newfangled2 15 Rewriting Maxwell's Equations Fig. 1. Splitting spacetime into space and time (One advantage of the dferential form language is its generality. We can take our spacetime to be any manifold M, of any dimension, and dlefine the electromagnetic field to be & 2form P on M. The frst pair fof Maxell equations says just that ar=o. Sometimes — but not always — we can split spacetime up into space tnd time, that i, write M ar Lx S for some manifold $ we call ‘space! If ao, we can weve t forthe utudl coordinate on I and split P into an electric and magnetic field Exercise 60. Show that any 2form P on Bx can be uniquely exprested tar BB Adin such a wy that for any local coordinates 2+ on S we have B= Bude and B= PBs de? We can also split the exterior derivative into spacelike and timelike parte a before: Exercise $1. Show tht jor any form won BL S there i «unique way tovrite da) = dt \ dua | dow auth that for an loeal coordinates 2* on S, ‘we hese dso = Byard nds, Aes = Ayan de Adel ‘Whea we split spacetime up into space and time, dF = 0 becomes ‘equivalent to the pai of equations 4B =0, OB 4+dsB=0. The Metre 3 In the static case, when OF coordinate entirely and test static equations 1B = 0, we ean forget about the £ and B as forms on space satisfying the asB = , ds =O. Note that the electric and magnet fields are only defined after we choose a way of splitting spacetime into space and time! If someone hands us @ manifold M, it may be diffeomorphic to Rx $ in many diferent ways, or in no wey at all. In special relativity one learns ‘hat diffrent inertial frames (corresponding to observers moving at constant velocity) will give diferent splittings of spacetimeinto 0, 2s called spacelike, since it points more in the space directions than the time direction. Itz is spacelike, the square foot of 2-2 represents the length ofa straight ruler that stretched from the origin to 2. Ifz has 2-2 <0 we call t timelike, since it points more in the time direction than in the space directions, The velocity Of particle moving slower than the speed of light is timelie. If «is timelike, the square root of x: measures the time a clock would tick off. fait moved from the origin to ima straight ine. If 2-2 = 0, xis called full or lightlike, since t points just as much in the time direction as in the space directions. We should add that sometimes people use she negative of our Minkowski metricand revere the definitions of ypacelike fand timelike. This i just matter of convention — but we wil always {allow the above convention! ‘The notion of metric generalizes these concepts, A semi Riemannian metric (or just"metric)) on a vector space V is & map, nVKV=R, ‘that is bilinear, or lineae in each slot: acu +e) = ealvy) + 90) g(oyewtu') = o(0,10) bealeyw') afer) = (0,2), The Metre 6 and nondegenerate: if 4(2,~) = 0 for all w € V, then v = 0. We say that » € V is spacelike timelike or null depeading on whether g(v,2) is ponitive, negative or ero, If (vw) = 0, we aay that v and w ace orthogonal. Note that null vectors are orthogonal to themselves! Givena metricon V, wecan always find an orthonormal basis for Vs that ia basis {e,) Such that g(eyy¢y) 8 0 ify # v, and it ‘The number of +1's and =1's i independent of the orthonormal basis, tad ifthe number of +1's x p and the number of —1's is 9, we say the metric has signature (p,¢). For example, Minkowski spacetime has signature (3,1), with the Minkowski metric given by n(o,w) =u? pote total + tw So far we have been talking about spacetimes that are vector spaces. [Now let AT be a manifold and consider a situation where the metric depends on where one is, A metric g on M sssigas to each point EM a metric gp on the tangent space Ty, ia @ smoothly varying ‘way. By ‘smoothly varying? we mean that if and w are smooth vector fields on M, the inner product gy(%p,up) is » smooth function on M. By the way, we usually write thie faction rimply as 9, One can show that the rmoothness condition implies that the sig- nature of gp is constant on any connected component of M. We are realy only iaterested in cases where the signature ie constant on all of 1M. Ithe signature of gis (n,0), where dim M = n, we say that is a Riemannian metric, while ifthe signature is (n — 1,1), we say that g is Lorentzian. By a semi-Riemannian manifold te mean ‘manifold equipped with a metric, and similarly for a Riemannian ‘manifold end a Lorentzian manifold. In relativity, spacetime is « Lorentzian manifld, which in the real world appease to be &-dimensional, although other cases are certainly interesting. The easiest way to get ahold of « dimensional Lorentsian ‘manifolds to take Sdimensional manifold S, space’, with a Rieman- nian metric 9p, and let IY, ‘spacetime’, be given by Ix S. Then we ‘ca define @ Lorentzian metre gad 8 fon Mas follows. Let 2 (i = 1,2,8) be local coordinates on an open subset U © S, and let t or 2° denote the coordinate on IR, that is,6 15 Rewriting Maxwell’ Equations ‘time’, Then 2 (j= 0,1,2,3) ate local coordinates on Rex U and we can define the metsic gto be that with componente 00 0 ° weal 8 ‘ o ‘This presets special sort of state spacetime, in mhich space has 1 tate that independent of ime "Phe most basi ue of a Lorentsian matric to measure distances ad times Fr example, path 70,1] > Mis spacelike, that x i ‘tangent vector is everywhere spc, me defies arclength to be [arora ys timelike, we define the proper time along 7 — that is, the time ticked of ty a'clock moving along 7 — to be [eso Orie. We will mainly be interested i some more sophisticated applicx tions ofthe metric, bowever. The most fundamental ofthese i ‘raising tnd lowering indices’, that is, converting between tangent and eotan feat vector. IfV ie a vector space equipped with a metric g, there is ‘natural way to tur an element v ¢ V into an element of V*, namely the linear funeSional g(v,") which eats another clement of V and spits cout a number. Bxercise 52. Use the nondayeneracy of the metre 10 show that the map from to V" given by verge fs an isomorphiom, this, one-to-one and onto. 1 follows that if fie semi-Riemansian manifold the metric de fines an isomorphism between each tangent space TpM and the corze sponding cotangent space Ts. We can picture this as follows: if the tangent vector v isa litle arom, the cotangent vector = g(t) is The Metsie 7 a stack of hyperplanes perpendicular tov, asin Figure 2. The reason for this is that w vanishes on vectors orthogonsl tov. ‘The key point is that one need the metric to know what ‘rthogonal” means! < Fig. 2. Tangent vector v and cotangent veetor (0,-) kOe) Similarly, we can convert between vector feds and I-forms on Mf By using the metric on space, for example, we can think ofthe electric field as 2 vector field instead of x Iform. We need to do this in order to think ofthe electri fel as “pointing in some direction Suppote M is semiRiemaanian manifold. Now that we can visu- lize Mforma on Mf as Selds of little artons, there is a nice way for us to vsualiae pforms for higher pas well. We can draw # wedge product hp of two cotangent vectors at para ttle parallelogram, asin Figure 3. So me can visualize @ 2-4orm on Mas feld of such ‘area elements. Similarly, we can draw a wedge product w fw A of three cotangent vectors ob pas a little parallelepiped, asin Figuce 4, and visualize a form as field of these ‘volume elements! — and 60 on for higher forme > <> Fig. 3. Picture of Ay € APTA* 15 Rewriting Maxwell's Equations We should not tale these pictures tae seriously. For example, if we drew (de +49) 4 (dy +4) and (dy 4 de) a (de az) thie way, we would get differen-looking paralelograms, even though they are equal as elements of APTSIR*. However, there is a lot about ‘these patllelograms that isthe same. Fest, they lie in the same plane VC MP. Second, they have the same area. And third, the pairs (dz + dy,dy-+ ds), (dy +d2,de—de) are bases for V that have the same orientation, in the usual sense of right-handed versus left-handed bases "This is true in general, which x why we think of w Ay € A273M as an “area clement’. Similar statements are true for wedge products of more: colangent vectors, Oa the other hand, itis worth noting that there fre usualy plenty of elements of A'T¢AQ that are aot wedge products Of cotangent vectors — we need to Consider near combinations, too. Sl, witha litle care one can get some good insights abot ciferential forms sing there picture, as we will see. ‘Now let us say some things about how the metric looks in the lan guage of indice. Let ue work in a chart, and let ey be basis of vector fields, Then we ean define the components ofthe metric as fellows: Sn = (0) ‘The Metre 8 If M is redimensional, gu is ann Xm matrix. The aondegeneracy condition implies thie matrix is invertible, 20 let g*” denote the inverse ‘matrix Then we have the following handy formulas, which explain why the process of eoavesting between vector fede and I-forms using the smetric is called raising and lowering indices: Exercise 58, Lets = wey bea vector fald on a chart. Show that the coreg form oy el wre the dt of forms Bxercise 54, Letw = up/ bea Lform on @ chart Show thatthe corre sponding vector field i opal tse, where of = ey [Exercise 55. Let be the Mintowsti meric on BY as defined above. Show (hat it components im the atandard basis are In general, if we have any quantity with some indices, such as 0 ° ates vee can lower or raise any index with the metic and its inverse using the Binsin summation convention. Bug, we can lower @ and get AE Mg = Sandee or mise § and get Aga ae, I we have a fot indices Hosting around it is important to keep track of their onder when we rise and lower them; otherwise things get con fusing. Note that se can even raise and lower indices on the metrie itself