General Relativity and Gravitational Waves: Session 2. General Coordinates Thomas A. Moore — Les Houches — July 4, 2018 Overview of this session:
2.2 Definition of a Coordinate Basis 2.3 Tensors in a Coordinate Basis 2.4 The Tensor Gradient 2.5 The Geodesic Equation 2.6 Schwarzschild Geodesics 2.7 LOFs and LIFs u=–1 u=0 u=1 u=2
ew Q w=5 w=4+dw ds
eu dw ew P w=4 P due w=4 u
w=3 u=1 u=1+du
Figure 1: This drawing shows an arbitrary coordinate system in a possibly curved space, a point , the basis P vectors eu, ew at that point, and a close-up view of how we define the basis vectors so that an infinitesimal displacement ds is the vector sum of the basis vectors times simply du and dw respectively.
I am not going to emphasize rigor in what follows, but rather an intuitive approach that I think will help you quickly develop a working understanding of the mathematics. Some might complain about all of the assumptions I am leaving unstated and the hands that I am waving, but my time with you is too short. No matter how we construct our coordinate system, the di↵erential distance ds between two infinitesimally separated points in a two-dimensional space (or the spacetime separation between two events in a spacetime) is a coordinate-independent quantity, because we can measure it directly with a ruler (or a clock in spacetime) without having to define a coordinate system at all. The fundamental way that we connect arbitrary coordinates to physical reality is by specifying how the spacetime separation between two infinitesimally- separated points (or events) depends on their coordinate separations. Let’s first consider a two-dimensional flat space. In such a space, a cartesian xy coordinate system is one in which the distance ds between two infinitesimally-separated points is given by ds2 = dx2 + dy2 at all points in the space. A curvilinear coordinate system is any non-cartesian coordinate system where this simple pythagorean relationship is not true. How can we connect the coordinate-independent distance between two points with their coordinate separations in such a case? Consider arbitrary coordinatesCoordinateu, w for a 2D space. Basis When using index notation, we will interpret dxu as being equivalent to du, and dxu as being equivalent to dw, and we will assume that Greek indices have two possible values u and w. (In the last session, in the context of cartesian coordinates in non-curved spacetime, u=–1 u=0 u=1 u=2 I stated that indices could represent either t, x, y, or z, but when we use arbitrary coordinates, the indices take on values corresponding to whatever namesew the coordinate have.) I willQ also represent vectors in this two-dimensional space with the same bold-face notation as wew=5 usedw=4+dw for four-vectorsds last time. This will keep eu dw ew the notation from changing when we generalizeP to 4D spacetimes.w=4 P due Now, no matter how our u, w coordinate system is defined,w at=4 each pointu in the space, we can define P a pair of basis vectors eu, ew such that w=3 u=1 u=1+du 1. eu points tangent to the w = constant curve toward increasing values of u. Define basis vectors eu, ew such that
2. ew points tangent to the 1.u =eu constant points tangent curve toward to the increasing w = constant values curve of w. in the direction of increasing u 3. Their lengths are defined so that the displacement vector ds between a point at coordinates u, w 2. e points tangent to the u = constant curve and an infinitesimally separatedw neighboring point at coordinates u + du, w +Pdw can be written in the direction of increasingQ w µ 3. ddss= =du dueeuu ++ dwdweww = dx eµ (See figure 1.) (2.1)
(Note: the lower index on eµ tells us which basis vector, not which component, we are talking about.) Now, this only works for di↵erential separations, because the directions of the basis vectors change as we move significant distances. Moreover, vector addition as I have illustrated it in Figure 1 is really only defined in a flat space. In a curved space, the separation between and must be so small that we can treat the space as flat, just as we treat a city map as flat on the curvedP surfaceQ of the earth. (Technically, we are doing this vector addition in a flat space that is tangent to the curved space at point .) P
2 Now, ifNow, we define if we define basis vectors basis vectors this way, this then way,du thenanddudwandbecomedw become the components the components of ds in of thatds in basis that and basis and we callwe the call set the of set basis of vectors basis vectorseu, eweau,coordinateew a coordinate basis. basis.A coordinateA coordinate basis is basis generally is generally di↵erent di↵ thanerent than Now, if we define basis vectors this way, then du and dw become the components of ds in that basis and the cartesianthe cartesian coordinate coordinate basis vectors basis vectorsex, ey einx, thatey in (1) thateu (1)andeuewandmayew notmay be not perpendicular, be perpendicular, (2) eu (2)andeuewand ew we call the set of basis vectors eu, ew a coordinate basis. A coordinate basis is generally di↵erent than may notmay have not unit have length, unit length, and (3) andeu (3)and/oreu and/orew mayew changemay change in magnitude in magnitude and/or and/or direction direction as we move as we from move from the cartesianpoint coordinate topoint point. to basis point. It is vectors also It is importante alsox, ey importantin that to note (1) toe thatu noteand defining thatew may defining a not coordinate be a perpendicular, coordinate basis is basis not (2) is theeu notandonly theewwayonly toway define to define e e may not havea set unit ofa length, set basis of vectors basis and (3) vectors oru aand/or coordinate or a coordinatew may system. change system. For in magnitudeexample, For example, standard and/or standard direction polar coordinates polar as we coordinates move in from two-dimensional in two-dimensional point to point. It is also important to note that defining a coordinate basis is not the only way to define flat spaceflat and space spherical and spherical coordinates coordinates in a three-dimensional in a three-dimensional flat space flat do spacenot dousenot a coordinateuse a coordinate basis, becausebasis, because a set of basis vectors or a coordinate system. For example, standard polar coordinates in two-dimensional the basisthe vectors basis vectors in those in coordinate those coordinate systems systems always always have unit have magnitude. unit magnitude. Using Using a coordinate a coordinate basis makes basis makes flat space and spherical coordinates in a three-dimensional flat space do not use a coordinate basis, because componentscomponentsof theof displacement the displacement vector vectords simplerds simpler (at the (at expense the expense of complexity of complexity in the in basis the vectors), basis vectors), and and the basis vectors in those coordinate systems always have unit magnitude. Using a coordinate basis makes this turnsthis out turns to out be crucial to be crucial in what in follows. what follows. components of the displacement vector ds simpler (at the expense of complexity in the basis vectors), and Once weOnce have we defined have definedCoordinate a coordinate a coordinate basis for basis coordinates for Basis coordinatesu and uwand,thenwew,thenwedefine definethe componentsthe components of a of a this turns out to be crucial in what follows. four-vectorfour-vectorA at anyA at point any pointto be Atou,A bewAsuchu,Aw thatsuch that Once we have defined a coordinateP basisP for coordinates u and w,thenwedefine the components of a u w four-vector A at any point to be A ,A such that u u w w µ µ P Define a four-vector’sA = AAe=u components+AAeue+w A= Aew soe=µ thatA eµ (2.2) (2.2) A = Aue + Awe = Aµe (2.2) This ensuresThis ensures that the that components the components of dus and of dsAwandtransformA transformµ alike. alike. Now, the scalar product of ds with itself is the square of the physical distance between the endpoints of This ensures thatNow, the the components scalar product of ds ofandds Awithtransform itself is alike. the square of the physical distance between the endpoints of Now, thethe scalar displacementthe product displacement of itd represents:s with it represents:The itself metric is the in square a coordinate of the physical basis distanceis between the endpoints of the displacement it represents: 2 2 ds = dsds ⇧=dsd=(s ⇧ duds =(eu +dudweue+w)dw⇧ (eduw)e⇧u(+dudweue+w)dw ew) 2 2 2 2 2 ds = ds ⇧ ds =(=dudueue=u+du⇧dweuee+uw⇧du)e⇧u( dwdu+ dueeuu dw+⇧ edwwe+ue⇧wdwe)w dw+ dwew dw⇧ eue+w ⇧dweu +ewdw⇧ ewew ⇧ ew 2 µ ⌫ µ ⌫ µ ⌫ µ ⌫ 2 = du eu ⇧ eu=+dxdudx= dwdxeeuµdx⇧⇧ eew⌫e+µ dw⇧geµ⌫⌫ dwdx gedxwµ⌫⇧dxeu dx+ dw ew ⇧ ew (2.3) (2.3) µ ⌫ µ ⌘⌫ ⌘ = dx dx eµ ⇧ e⌫ gµ⌫ dx dx (2.3) The setThe of fourset of components four componentsgµ⌫ ⌘= geµµ⌫⇧=e⌫eµrepresents⇧ e⌫ represents the metric the metric tensor tensorof ourof two-dimensional our two-dimensional coordinate coordinate The set ofbasis. four componentsbasis. Note that Noteg becauseµ that⌫ = e becauseµ both⇧ e⌫ represents the both magnitudes the the magnitudesmetric and directions tensor and directionsof of our the two-dimensional of basis the vectors basis vectors depend coordinate depend on position, on position, the the basis. Notevalue that ofvalue becausegµ⌫ ofalsog bothµ⌫ generallyalso the generally magnitudes depends depends on and position. directions on position. But of the since But basis the since vectors dot-product the dot-product depend of on vectors position, of vectors is commutative, the is commutative, this this value of gµdefinition⌫ alsodefinition generally implies depends implies that the that on metric position. the metric is always But is sincealways symmetric: the symmetric: dot-productg↵ =gg↵ of ↵ vectors.= Thisg ↵. is means This commutative, means that the that this metric the metric of a two- of a two- definition impliesdimensionaldimensional that the space metric spacehas onlyis hasalways 3 only independent symmetric: 3 independent components,g↵ = components,g ↵. and This in means and spacetime, in that spacetime, the the metric metric the of metric has a two- 10 has independent 10 independent dimensionalcomponents spacecomponents has only(of the 3 (of independent 16 the possible 16 possible combinations components, combinations of and index in of spacetime, values, index values, 4 involve the 4 metric involve the same has the 10 value same independent repeated value repeated and the and half the half componentsof (of the theof remaining the 16 possible remaining 12 arecombinations 12 the are same the asof same index the as other values, the half, other 4 involve for half, a total for the a ofsame total 4 + value of 6 4 = + repeated 10). 6 = 10). and the half of the remainingYou 12 can areYou easily the can same easily see as how the see we other how can we half, generalize can for generalize a total this of to 4 this four + 6 to = dimensions four 10). dimensions in spacetime. in spacetime. In spacetime, In spacetime,gµ⌫ =gµ⌫ = You caneµ easily⇧ e⌫e(whereµ see⇧ e⌫ how(where the we indices can the indices generalize now range now this overrange to four four over values) dimensions four values) represents in represents spacetime. the generalization the In generalization spacetime, of thegµ of metric⌫ the= metric tensor tensor⌘µ⌫ ⌘µ⌫ eµ ⇧ e⌫ (whereintroduced theintroduced indices in the now in last range the session. last over session. four In flat values) Inspacetime,flat representsspacetime, we can the we generalization always can always find a find cartesian of the a cartesian metric coordinate tensor coordinate⌘ basisµ⌫ where basis where the the introducedbasis in the vectorsbasis last session. vectors are everywhere are In flat everywherespacetime, orthogonal orthogonal we (e can⇧ e always (e=µ ⇧ 0e for⌫ find=µ 0 a= for cartesian⌫),µ and= ⌫), have coordinate and unit have magnitude unit basis magnitude where (e the⇧ e (e=µ ⇧ e1⌫ when= 1 when µ ⌫ 6 6 µ ⌫ ± ± basis vectorsµ = are⌫,with everywhereµ = ⌫,with1 indicating orthogonal1 indicating a ( timeeµ ⇧ ae coordinate)⌫ time= 0 for coordinate)µ = at⌫), all and events at allhave events and unit still magnitude and have still the have (e componentsµ ⇧ thee⌫ = components1 when of the of di↵ theerential di↵erential 6 6 ± µ = ⌫,withfour-displacement6 1 indicatingfour-displacement a timeds be coordinate)d simplys be simplydt, at dx, alldt, dy, events dx,and dy,dz andand. This stilldz. have is This not the is generally notcomponents generally possible of possible the in di curved↵erential in curved spacetime. spacetime. The The 6 2 2 µ ⌫ µ ⌫ four-displacementquantityquantitydsdsbe= simplydsgµ⌫ dx=dt,gdxµ dx,⌫ dxgeneralizes dy,dxandgeneralizesdz. this This idea thisis not for idea generally an forarbitrary an possiblearbitrarycoordinate incoordinate curved system spacetime. system using a using The coordinate a coordinate basis basis 2 µ ⌫ quantity dsin= angarbitraryµin⌫ dx an dxarbitraryspacetime.generalizesspacetime. The this idea quantity The for quantity andsarbitraryrepresentsds representscoordinate the spacetime the system spacetime separation using separation a coordinate if the events if the basis events at ends at of ends the of the in an arbitrarydi↵erentialspacetime.di↵erential displacement The displacement quantity haveds a have spacelikerepresents a spacelike separation, the spacetime separation, and d separation⌧ and= pd⌧ ds= if2p theis theds events2 infinitesimalis the at infinitesimal ends of spacetime the spacetime interval interval di↵erential displacement(which is have the a same spacelike as the separation, proper time and ind⌧ the= p infinitesimalds2 is the infinitesimal limit) between spacetime the events interval if they have a timelike (which is the same as the proper time in the infinitesimal limit) between the events if they have a timelike (which is theseparation. sameseparation. as the The proper point The time is point that in is the the that infinitesimal metric the metric connects limit) connects the between arbitrary the arbitrary the coordinates events coordinates if they with have the with a physical timelike the physical distances distances or or separation.intervals Theintervals point in the is that in physical the the physical metric universe connects universe behind the behind that arbitrary coordinate that coordinate coordinates system. system. with the physical distances or intervals in the physical universe behind that coordinate system. 2.2.12.2.1 Exercise: Exercise: The Metric The Metric for Spherical for Spherical Coordinates Coordinates 2.2.1 Exercise: The Metric for Spherical Coordinates ConsiderConsider✓- coordinates✓- coordinates on the on surface the surface of a sphere of a sphere of radius of radiusR, whereR, where curves curves of constant of constant✓ and ✓ andare lines are of lines of Consider ✓- coordinates on the surface of a sphere of radius R, where curves of constant ✓ and are lines of latitudelatitude and longitude, and longitude, respectively respectively (but assume (but assume that ✓ that= 0 at✓ = the 0 at north the pole, north as pole, is normal as is normal in physics, in physics, rather rather latitude and longitude, respectively (but assume that ✓ = 0 at the north pole, as is normal in physics, rather than atthan the at equator). the equator). Note that Note these that curves these curves are perpendicular are perpendicular everywhere everywhere but the but poles. the poles. By considering By considering than at the equator). Note that these curves are perpendicular everywhere but the poles. By considering what thewhat formula the formula for ds2 forbetweends2 between infinitesimally-separated infinitesimally-separated points points must be must in be terms in terms of d✓ ofandd✓d and,findd ,find what the formula for ds2 between infinitesimally-separated points must be in terms of d✓ and d ,find the metricthe metric components componentsg✓✓,g✓ g,g✓✓,g ✓✓ , and,g ✓g, andasg a functionas a function of position of position for a coordinate for a coordinate basis based basis based on the on the the metric components g✓✓,g✓ ,g ✓, and g as a function of position for a coordinate basis based on the coordinatescoordinates✓ and ✓ .and Also, . what Also, are what the are lengths the lengths of the ofe✓ theandee✓ andbasise vectorsbasis vectors as a function as a function of position? of position? coordinates ✓ and . Also, what are the lengths of the e✓ and e basis vectors as a function of position? 2.3 Tensors in a Coordinate Basis 2.3 Tensors2.3 Tensors in a Coordinate in a Coordinate Basis Basis Again, for the moment, let’s go back to considering a two-dimensional, possibly curved surface in space. Again, forAgain, the moment, for the let’s moment, go back let’s to go considering back to considering a two-dimensional, a two-dimensional, possibly curved possibly surface curved in space. surface in space. Consider a general transformation between our original coordinates u, w to new coordinates p(u, w) and Consider aConsider general transformation a general transformation between our between original our coordinates original coordinatesu, w to newu, coordinates w to newp coordinates(u, w) and p(u, w) and q(u, w). Theq(u, chain w).q(u, Therule w). for chain The partial rule chain derivatives for rule partial for partial impliesderivatives derivatives that implies infinitesimal implies that infinitesimal changes that infinitesimal in the changes new changes coordinates in the in new the are coordinates new coordinates are are
3 3 3 Now, if we define basis vectors this way, then du and dw become the components of ds in that basis and we call the set of basis vectors eu, ew a coordinate basis. A coordinate basis is generally di↵erent than the cartesian coordinate basis vectors ex, ey in that (1) eu and ew may not be perpendicular, (2) eu and ew may not have unit length, and (3) eu and/or ew may change in magnitude and/or direction as we move from point to point. It is also important to note that defining a coordinate basis is not the only way to define a set of basis vectors or a coordinate system. For example, standard polar coordinates in two-dimensional flat space and spherical coordinates in a three-dimensional flat space do not use a coordinate basis, because the basis vectors in those coordinate systems always have unit magnitude. Using a coordinate basis makes components of the displacement vector ds simpler (at the expense of complexity in the basis vectors), and this turns out to be crucial in what follows. Once we have defined a coordinate basis for coordinates u and w,thenwedefine the components of a four-vector A at any point to be Au,Aw such that P u w µ A = A eu + A ew = A eµ (2.2)
This ensures that the components of ds and A transform alike. Now, the scalar product of ds with itself is the square of the physical distance between the endpoints of the displacement it represents:
2 ds = ds ⇧ ds =(du eu + dw ew) ⇧ (du eu + dw ew) 2 2 = du eu ⇧ eu + du dw eu ⇧ ew + dw dw ew ⇧ eu + dw ew ⇧ ew = dxµdx⌫ e ⇧ e g dxµdx⌫ (2.3) µ ⌫ ⌘ µ⌫
The set of four components gµ⌫ = eµ ⇧ e⌫ represents the metric tensor of our two-dimensional coordinate basis. Note that because both the magnitudes and directions of the basis vectors depend on position, the value of gµ⌫ also generally depends on position. But since the dot-product of vectors is commutative, this definition implies that the metric is always symmetric: g↵ = g ↵. This means that the metric of a two- dimensional space has only 3 independent components, and in spacetime, the metric has 10 independent components (of the 16 possible combinations of index values, 4 involve the same value repeated and the half of the remaining 12 are the same as the other half, for a total of 4 + 6 = 10). You can easily see how we can generalize this to four dimensions in spacetime. In spacetime, gµ⌫ = eµ ⇧ e⌫ (where the indices now range over four values) represents the generalization of the metric tensor ⌘µ⌫ introduced in the last session. In flat spacetime, we can always find a cartesian coordinate basis where the basis vectors are everywhere orthogonal (eµ ⇧ e⌫ = 0 for µ = ⌫), and have unit magnitude (eµ ⇧ e⌫ = 1 when µ = ⌫,with 1 indicating a time coordinate) at all events6 and still have the components of the di↵±erential four-displacement6 ds be simply dt, dx, dy, and dz. This is not generally possible in curved spacetime. The 2 µ ⌫ quantity ds = gµ⌫ dx dx generalizes this idea for an arbitrary coordinate system using a coordinate basis in an arbitrary spacetime. The quantity ds represents the spacetime separation if the events at ends of the di↵erential displacement have a spacelike separation, and d⌧ = p ds2 is the infinitesimal spacetime interval Exercise(which is the same as the proper time in the infinitesimal limit) between the events if they have a timelike separation. The point is that the metric connects the arbitrary coordinates with the physical distances or intervals in the physical universe behind that coordinate system.
2.2.1 Exercise: The Metric for Spherical Coordinates Consider ✓- coordinates on the surface of a sphere of radius R, where curves of constant ✓ and are lines of latitude and longitude, respectively (but assume that ✓ = 0 at the north pole, as is normal in physics, rather than at the equator). Note that these curves are perpendicular everywhere but the poles. By considering what the formula for ds2 between infinitesimally-separated points must be in terms of d✓ and d ,find the metric components g✓✓,g✓ ,g ✓, and g as a function of position for a coordinate basis based on the coordinates ✓ and . Also, what are the lengths of the e✓ and e basis vectors as a function of position?
2.3 Tensors in a Coordinate Basis Again, for the moment, let’s go back to considering a two-dimensional, possibly curved surface in space. Consider a general transformation between our original coordinates u, w to new coordinates p(u, w) and q(u, w). The chain rule for partial derivatives implies that infinitesimal changes in the new coordinates are
3 related to changes in the old coordinates as follows: @p @p @q @q dp = du + dw and dq = du + dw (2.4) @u @w @u @w related to changes in the old coordinates as follows: If we consider the p, q coordinates the primed coordinate system and u, w coordinates the unprimed system, then we can write this transformation@p rule compactly@p as @q @q dp = du +Tensorsdw and dq = indu + adw coordinate(2.4) basis @u @w µ @u @w µ @x0 ⌫ dx0 = dx (2.5) If we consider the p, q coordinates the primed coordinate@x⌫ system and u, w coordinates the unprimed system, relatedthen we to can changes write this in the transformation old coordinates rule as compactly follows: as (with an implicitrelated sum over to changes the ⌫ subscript: in the oldTransformation we coordinates will consider as a follows: superscript from u, win to the p, denominator q coordinates of a partial @p @p µ @q @q derivative to equivalent to a subscript).dp = du + dwµ and@x0 dq =⌫ du + dw (2.4) dx0 =@p dx@p @q @q (2.5) Now note that in a coordinate basis,@u the values@wdp of=dp and@dux⌫ dq+ are@dw theu actualand@w componentsdq = duof+ the infinitesimaldw (2.4) displacement vector ds in the primed system and du @andu dw are@w the same in the unprimed@u system.@w Since If(with we consider an implicit the sump, q coordinates over the ⌫ subscript: the primed we coordinate will consider system a superscript and u, w coordinatesin the denominator the unprimed of a system, partial bythen definition, we can the writeIf components we this consider transformation of the anp, arbitrary q coordinates ruleor compactly vector abstractly:A thetransform as primed in coordinate the same ways system as the and componentsu, w coordinates of the the unprimed system, displacementderivative to vector, equivalent The transformation to a subscript). law for the components of A is Now note thatthen in we a coordinate can write basis, this transformation the values of dp and ruleµ dq compactlyare the actual as components of the infinitesimal µ @x0 ⌫ dx0 = µ dx (2.5) displacement vector ds in the primed system andµ du@x0and⌫ ⌫dw are the sameµ in the unprimed system. Since A = @xA @x0 (2.6) by definition, the components of an arbitrary vector0 A ⌫transform inµ the same ways⌫ as the components of the @x dx0 = ⌫ dx (2.5) (withdisplacement an implicit vector, sum The over transformation the ⌫ subscript: lawVector we for will the transformation considercomponents a superscript of A is @law:x in the denominator of a partial Again,derivative the extensionto equivalent to four-dimensional to a subscript). spacetime is straightforward. But this approach is only straight- forward if we are(with using an a coordinate implicit sum basis over, something the ⌫ subscript: that weµ will we simply will consider assume from a superscript now on. in the denominator of a partial Now note that in a coordinate basis, the valuesµ of dp@xand0 dq⌫ are the actual components of the infinitesimal derivative to equivalent to a subscript).A0 = A (2.6) displacementNote in that vector contextds thatin the equation primed 2.6 system looks and justdu like@andx⌫ thedw transformationare the same in law the for unprimed the components system. of Since a ⌫ µ µ four-vector in spacetimeNow with note@ thatx0 /@ inx areplacing coordinate⇤ basis,. Indeed, the values if you take of dp partialand dq derivativesare the actual of the components Lorentz of the infinitesimal byAgain, definition, the extension the components to four-dimensional of an arbitrary spacetime vector⌫ A istransform straightforward. in the same But this ways approach as the components is only straight- of the transformation functionsdisplacementt0(t, x, vector y, z), xd0(st,in x,Generalizes y,the z), primedy0(t, x, y, system z ),Lorentz and andz0(t,du transformation, x, y,and z),dw youare will the see same thatbecause in in fact the for unprimed a LT: system. Since displacementforward if we vector, are using The a transformationcoordinate basis law, something for the components that we will of simplyA is assume from now on. by definition, the components of anµ arbitrary vector A transform in the same ways as the components of the Note in that context that equation 2.6 looks@x0 just likeµ the transformation law for the components of a displacement vector, The transformationµ =@⇤xµ0 law⌫ for the components of A is (2.7) four-vector in spacetime with @x ⌫ /@xµ replacingA0 ⌫⇤=µ . Indeed,⌫ A if you take partial derivatives of the Lorentz(2.6) 0 @x ⌫ @x⌫ transformation functions t0(t, x, y, z), x0(t, x, y, z), y0(t, x, y, z), and z0(t, x, y,µ z), you will see that in fact µ @x0 ⌫ forAgain, that particular the extension coordinate to four-dimensional transformation. spacetime We see, is therefore, straightforward.A that0 = expressing ButA this the approach transformation is only straight- coe - (2.6) cients in terms of partial derivatives is consistent with@x µ but generalizes the@ transformationx⌫ between cartesian forward if we are using a coordinate basis, something0 = that⇤µ we will simply assume from now on. (2.7) coordinatesNote in in that inertialAgain, context frames the that extension that equation we considered to 2.6 four-dimensional looks earlier.@x just⌫ like⌫ the spacetime transformation is straightforward. law for the components But this approach of a is only straight- Now, basic partial di↵erential calculus⌫ µ implies thatµ four-vectorfor that particular inforward spacetime coordinateif with we are@ transformation.x0 using/@x areplacing coordinate We⇤ see,⌫ . basis Indeed, therefore,, something if you that take expressing that partial we will thederivatives simply transformation of assume the Lorentz coe from - now on. transformation functions t0(t, x, y, z), x0(t, x, y, zµ), y0⌫(t, x, y, z), and z0(t, x, y, z), you will see that in fact cients in terms ofNote partial in derivatives that context is consistent that@x0 equation@ withx but 2.6µ generalizes looks just the like transformation the transformation between cartesian law for the components of a ⌫ =µ µ (2.8) coordinates infour-vector inertial frames in spacetime that we considered with @⌫ x earlier./µ@↵x replacing↵ ⇤ . Indeed, if you take partial derivatives of the Lorentz @x @@0xx0 0 ⌫ Now, basic partial di↵erential calculus implies that= ⇤µ (2.7) For example, if wetransformation write this out functions for our u,t w0(t, x,p,@ y,x q z⌫),transformations,x0(t,⌫ x, y, z), y this0(t, x, says y, z that), and z0(t, x, y, z), you will see that in fact ! µ ⌫ @x0 @x µ for that particular coordinate@p @u transformation.@p @w dp We see,@ therefore,p=@ uµ @p@ thatx@0w expressingdpµ the transformation coe(2.8) - + = =1@x⌫, @x ↵ ↵+ == ⇤ ⌫=0, (2.7) cients in terms of partial@u derivatives@p @w @ isp consistentdp with0 @ butu @ generalizesq @w@x@⌫q thedq transformation between cartesian coordinates in inertial frames that we considered earlier. For example, if we write@q this@u out@q for@w our u,dq w p, q transformations,@q @u @q @w thisdq says that Now, basicfor partial that di particular↵erential+ calculus coordinate= implies=0! transformation., that + We see,= therefore,= 1 that expressing (2.9) the transformation coe - cients in@ termsu@@pp@u of@w partial@p@p@w derivativesdpdp is@ consistentu@@pq@u @w@ withp@q@w butdqdp generalizes the transformation between cartesian + = @x=1µ ,@x⌫ + = =0, Equation 2.8 basicallycoordinates says@u that in@p inertial@x@wµ/@ framesxp⌫ anddp that@0xµ/ we@x considered⌫=@representu @µq @w earlier. inverse@q transformations,dq analogous(2.8) to 0 @x⌫ @x ↵0 ↵ ⇤µ and (⇤ 1)µ inNow, flat spacetime. basic@q @u partial@q @ diw↵erentialdq calculus0 @q implies@u @q that@w dq ⌫ ⌫ + = =0, + = = 1 (2.9) ForNow example, we can if easily we write state@ this generalizedu @p out for@w our transformation@p u, wdp p, q lawstransformations,@u for@q arbitrary@µw @q tensors: this⌫ saysdq an that nth-rank tensor’s com- ! ↵ @x0 @x n ponents transform: we define an nth-rank tensor T ··· ··· to be an n-index= object µ (with 4 components) (2.8) @p @u @p µ@w ⌫ dp µ @⌫p @u @⌫ p @w↵ dp ↵ Equation 2.8 basically says that @x0 /@x and @x /@···x0 represent@x @ inversex0 transformations, analogous to thatµ transforms1 accordingµ to + = =1, + = =0, ⇤ ⌫ and (⇤ ) ⌫ in flat spacetime.@u @p @w @p dp @u @q @w @q dq For example, if we write this↵ out for⌫ our u, w p, q transformations, this says that Now we can easily state@q @ generalized↵u @ q @w transformation@x0dq @x laws@q@x@ for0u arbitrary@qµ@w tensors: dq an nth-rank tensor’s com- 0 ··· ··· ! T + = = µ =0, ↵ + T ···⌫ =··· = 1n (2.10) (2.9) ponents transform: we define@u @p an n···@th-rankw @p @@px tensor@dpu···@@xTp0 ···@··· w@u@···@xqdpto··· be@w an@nq···-index@dqp @ objectu @ (withp @w 4 components)dp that transforms according to + ··· = =1, + = =0, that is, a partial-derivative factor with theµ @ primedu⌫ @p coordinate@µw @p ⌫ indp the numerator@u for@q every@ upperw @q (superscript)dq Equation 2.8 basically says that @x0 /@x and @x /@x0 represent inverse transformations, analogous to µ 1 µ ↵ ⌫ index⇤ andand a (⇤ a partial-derivative) in flat spacetime. factor↵ with@ theq@@x0u primed@@qx coordinate@w @dqx0 in theµ denominator@ q @u for@q every@w lowerdq (sub- ⌫ ⌫ T 0 ··· ··· = T (2.10) script) index. Using methods similar to what weµ did+ in the last= session, =0, you···⌫ can··· prove+ in particular= that= 1 the (2.9) Now we can easily state generalized··· transformation@u@@xp ···@@wx0@ lawsp··· for@dpx arbitrary··· tensors:···@u @q an nth-rank@w @q tensor’sdq com- metric is a tensor with two lower indices: ↵ n ponents transform: we define an nth-rank tensor T ··· ··· to be an n-index object (with 4 components) that is, a partial-derivative factor with the primed coordinateµ ··· ⌫ in the numeratorµ ⌫ for every upper (superscript) that transformsEquation according 2.8 to basically says that@x@x↵0 @/x@⌫ x and @x /@x0 represent inverse transformations, analogous to index and a a partial-derivativeµ 1 µ factor withg the= primed0 coordinateg in the denominator for every lower(2.11) (sub- ⇤ ⌫ and (⇤ ) ⌫ in flat spacetime.µ0 ⌫ µ ↵ script) index. Using methods similar to what we↵ @ didx in@⌫x the0 last session, you can prove in particular that the ↵ @x0 @x @x0 µ Now we canT 0 easily··· ··· state= generalized transformationT ··· laws··· for arbitrary tensors:(2.10) an nth-rank tensor’s com- metric is a tensor with two lower indices: µµ⌫ µ⌫ ↵ ⌫ ⌫ that the Kronecker delta is still a tensor,··· that@gx ···defined@x0 such··· @ thatx ···g gµ↵ =··· ↵ (the matrix inverse of the n ponents transform: we define an nth-rank tensor T ··· ··· to be an n-index object (with 4 components) metric tensor), is still a tensor, that multiplying by g ↵and g⌫µ⌫ and summing··· over µ still lowers or raises a that is, a partial-derivativethat transforms factor according with the primedto @ coordinateµx⌫0 @x in the numerator for every upper (superscript) gµ0 ⌫ = µ g↵ (2.11) @x @x0 index and a a partial-derivative factor with the primed coordinate↵ in the⌫ denominator for every lower (sub- µ↵⌫ @x0 µ@⌫ x @⌫ x0 µ script)that the index. Kronecker Using delta methods is still similar a tensor, to what that weg0 ··· diddefined4 in··· the such last that session,g g you= can prove(the in matrix particular inverse that of the T = µ µ ↵ ↵ T ···⌫ ··· (2.10) ··· @µx⌫ ···@x ··· @x ··· ··· metric istensor), a tensor is still with a two tensor, lower that indices: multiplying by gµ⌫ and g and summing0 over µ still lowers or raises a ↵ ⌫ that is, a partial-derivative factor@ withx0 the@x primed coordinate in the numerator for every upper (superscript) gµ0 ⌫ = µ g↵ (2.11) index and a a partial-derivative factor@x4 @ withx0 the primed coordinate in the denominator for every lower (sub- script) index. Using methods similarµ⌫ to what we didµ⌫ in the last⌫ session, you can prove in particular that the that the Kronecker delta is still a tensor, that g defined such that g gµ↵ = ↵ (the matrix inverse of the metric is a tensor with two lower indices: µ⌫ metric tensor), is still a tensor, that multiplying by gµ⌫ and g and summing over µ still lowers or raises a ↵ ⌫ @x0 @x gµ0 ⌫ = µ g↵ (2.11) 4 @x @x0 µ⌫ µ⌫ ⌫ that the Kronecker delta is still a tensor, that g defined such that g gµ↵ = ↵ (the matrix inverse of the µ⌫ metric tensor), is still a tensor, that multiplying by gµ⌫ and g and summing over µ still lowers or raises a
4 related to changes in the old coordinates as follows: related to changes in the old coordinates as follows: @p @p @q @q @p @p dp = du +@q dw @qand dq = du + dw (2.4) dp = du + dw and @udq = @wdu + dw @u @w (2.4) @u @w @u @w Ifrelated we consider to changes the in thep, q oldcoordinates coordinates the as follows: primed coordinate system and u, w coordinates the unprimed system, If we consider the p, q coordinates the primed coordinate system and u, w coordinates the unprimed system, then we can write this transformation@p @p rule compactly@q as @q then we can write this transformation rule compactlydp = du as+ dw and dq = du + dw (2.4) @u @w @u @w µ µ µ @x0 ⌫ If we consider the p, q coordinatesµ the primed@x0 coordinate⌫ dx system0 = and u,dx w coordinates the unprimed system, (2.5) dx0 = dx ⌫ (2.5) then we can write this transformation rule@x⌫ compactly as @x µ (with an implicit sum over the ⌫ subscript:µ @ wex0 will⌫ consider a superscript in the denominator of a partial (with an implicit sum over the ⌫ subscript: we will considerdx0 a= superscriptdx in the denominator of a partial (2.5) derivative to equivalentderivative to a subscript).to equivalent to a subscript). @x⌫ Now note that in(with a coordinateNow an implicit note basis, that sum in the over a coordinate values the ⌫ subscript: of dp basis,and wedq the willare values consider the actual of adp superscript componentsand dq arein the theof denominator theactual infinitesimal components of a partialof the infinitesimal displacement vectordisplacementderivativeds in theto primed equivalent vector system tods ain subscript). and thedu primedand dw systemare the and samedu and in thedw unprimedare the same system. in the Since unprimed system. Since by definition, the componentsby definition,Now note of that an the inarbitrary a components coordinate vector basis, ofA an thetransform arbitrary values of indp vector theand samedq areA transform ways the actual as the components in components the sameof the ways of infinitesimal the as the components of the displacement vector ds in the primed system and du and dw are the same in the unprimed system. Since displacement vector,displacement The transformation vector, law The for transformation the components law of forA is the components of A is by definition, the components of an arbitrary vector A transform in the same ways as the components of the displacement vector, The transformation lawµ for the components of Aµ is µ @x0 ⌫ µ @x0 ⌫ A0 = A (2.6) A0 = ⌫ A µ ⌫ (2.6) @x µ @x0 ⌫ @x A0 = A (2.6) @x⌫ Again, the extensionAgain, to four-dimensional the extension spacetime to four-dimensional is straightforward. spacetime But is thisstraightforward. approach is only But straight- this approach is only straight- forward if we are usingforwardAgain, a coordinate theif extension we are basis using to four-dimensional, something a coordinate that spacetime basis we will, something is simply straightforward. assume that we from But will this now simply approach on. assume is only from straight- now on. forward if we are using a coordinate basis, something that we will simply assume from now on. Note in that contextNote that in equation that context 2.6 looks that just equation like the 2.6 transformation looks just like law the for transformation the components law of for a the components of a Note in that⌫ contextµ that equationµ 2.6 looks just like the transformation law for the components of a four-vector in spacetime with @x0 /@x replacing⌫ ⇤ µ.⌫ Indeed,µ ifµ you takeµ partial derivatives of the Lorentz four-vectorfour-vector in in spacetime spacetime with with@x0 /@@xx⌫ 0 replacing/@x replacing⇤ ⌫ . Indeed,⇤ ⌫ if. you Indeed, take partial if you derivatives take partial of the derivatives Lorentz of the Lorentz transformation functionstransformationtransformationt0(t, x, y, functions z), functionsx0(t,t x,0(t, y, x,t z0 y,(),t, zy), x,0(xt, y,0( x,t, z), x, y, y,x z0 z),(),t, andy x,0(t, y,z x,0 z( y,),t, z x,y),0( y, andt, z x,),z0 y,( yout, z x,), y, will and z), see youz0( thatt, will x, see y, in z that fact), you in fact will see that in fact µ µ @x0 µ @x0 µ µ = ⇤@⌫x0 µ (2.7) ⌫ = ⇤ ⌫ @x⌫ = ⇤ (2.7) (2.7) @x @x⌫ ⌫ for that particular coordinatefor that particular transformation. coordinate transformation. We see, therefore, We see, that therefore, expressing that expressing the transformation the transformation coe - coe - forcients that in terms particular of partial coordinateTensors derivatives transformation. is consistent in with buta We generalizescoordinate see, therefore, the transformation that expressing between basis thecartesian transformation coe - cients in terms of partialcientscoordinates derivatives in terms in inertial of is partial consistent frames that derivatives with we considered but is generalizes consistent earlier. the with transformation but generalizes between the transformation cartesian between cartesian coordinates in inertialcoordinates framesNow, basic that in partial we inertial considered di↵erential frames calculus earlier. that we implies considered that earlier. Now, basic partial di↵erential calculus implies that Now, basic partialAn di↵ importanterential calculus identity:@x µ implies@x⌫ that 0 = µ (2.8) µ ⌫ @x⌫ @x ↵ ↵ @x0 @x µ 0 µ ⌫ = @x0 @x (2.8) For example, if we write this out⌫ for our↵u, w ↵ p, q transformations, thisµ says that @x @x0 ⌫ ↵ = ↵ (2.8) ! @x @x0 For example, if we write this out for our u,@p w@u @p,p q@wtransformations,dp @p this@u says@p @ thatw dp For example, if we write this!+ out for= our=1u,, w p, q+transformations,= =0, this says that @Example:u @p @w @p dp !@u @q @w @q dq @p @u @p @w @q dp@u @q @w @pdq@u @p @@qw@u dp@q @w dq + = @p+=1@u, @=p @w=0+, dp =+ @=0p @,=u =@p 1@w dp (2.9) @u @p @w @p @u dp@p @w +@p @udp@q = @w=1@@uq@,q dq@w @q +dq = =0, @u @p @w @p dp @u @q @w @q dq @q @u @q @w dq µ ⌫@q @u µ @q @⌫ w dq Equation 2.8+ basically says= that=0@x,0 /@x and @+x /@x0 represent= = inverse 1 transformations, analogous (2.9) to µ 1 µ @q @u @q @w dq @q @u @q @w dq ⇤ ⌫ and@u @ (⇤p ) @⌫ win@ flatp spacetime.dp + @u @q = @w=0@q , dq + = = 1 (2.9) @u @p @w @p dp @u @q @w @q dq Now we can easilyµ state⌫ generalizedµ transformation⌫ laws for arbitrary tensors: an nth-rank tensor’s com- Equation 2.8 basically says that @x0 /@x and @x /@x0 represent↵ inverse transformations, analogousn to ponents transform: we define an nth-rank tensor T ··· ··· to be an n-index object (with 4 components) µ 1 µ µ ⌫ ··· µ ⌫ ⇤ ⌫ and (⇤ ) ⌫ inEquationthat flat transforms spacetime. 2.8 accordingbasically to says that @x0 /@x and @x /@x0 represent inverse transformations, analogous to µ 1 µ General tensor transformation law: Now we can easily⇤ state⌫ and generalized (⇤ ) ⌫ in transformation flat spacetime. laws↵ for arbitrary⌫ tensors: an nth-rank tensor’s com- ↵ ↵ @x 0 @x @x0 µ n 0 ··· ··· ponents transform: we defineNow we an cannth-rank easily state tensorT generalized T ···= ···µ transformationto be an n-index laws objectT ··· for⌫ (with arbitrary··· 4 components) tensors: an(2.10) nth-rank tensor’s com- ··· ···@x ···@x0 ··· @x↵ ··· ··· n that transforms accordingponents to transform: we define an nth-rank tensor T ··· ··· to be an n-index object (with 4 components) that is, a partial-derivative factor with the primed coordinate in the··· numerator for every upper (superscript) that transforms according↵ to ⌫ index and a a↵ partial-derivative @x0 factor@x with the@ primedx0 coordinateµ in the denominator for every lower (sub- 0 ··· ··· script) index.T Using methods= µ similar to what we did in↵ theT last···⌫ session,⌫··· you can prove in particular(2.10) that the ··· @x ···@↵x0 ··· @x @x···0 ···@x @x0 µ metric is a tensor with two lower indices:T 0 ··· ··· = T ···⌫ ··· (2.10) @xµ ···@x ··· @x ··· ··· that is, a partial-derivative factor with the primed coordinate··· in the↵ numerator⌫ 0 for every upper (superscript) @x0 @x index and a a partial-derivative factor with the primed coordinategµ0 ⌫ = µ in the g↵ denominator for every lower (sub-(2.11) that is, a partial-derivative factor with the@ primedx @x0 coordinate in the numerator for every upper (superscript) script) index. Using methods similar to what we did in the last session, you can prove in particular that the index and a a partial-derivative factor withµ⌫ the primed coordinateµ⌫ ⌫ in the denominator for every lower (sub- that the Kronecker delta is still a tensor, that g defined such that g gµ↵ = ↵ (the matrix inverse of the metric is a tensor with two lower indices: µ⌫ script)metric tensor), index. is Using still a tensor, methods that similar multiplying to what by gµ⌫ weand didg inand the summing last session, over µ still you lowers can prove or raises in a particular that the ↵ ⌫ metric is a tensor with two lower@x0 indices:@x gµ0 ⌫ = g↵ (2.11) @xµ @x 4 ↵ ⌫ 0 @x0 @x µ⌫ gµ0 ⌫ =µ⌫ µ ⌫ g↵ (2.11) that the Kronecker delta is still a tensor, that g defined such that g g@µ↵x =@ x↵0 (the matrix inverse of the µ⌫ metric tensor), is still a tensor, that multiplying by gµ⌫ and g andµ summing⌫ over µ still lowersµ⌫ or raises⌫ a that the Kronecker delta is still a tensor, that g defined such that g gµ↵ = ↵ (the matrix inverse of the µ⌫ metric tensor), is still a tensor, that multiplying by gµ⌫ and g and summing over µ still lowers or raises a 4 4 related to changes in the old coordinates as follows: @p @p @q @q dp = du + dw and dq = du + dw (2.4) @u @w @u @w If we consider the p, q coordinates the primed coordinate system and u, w coordinates the unprimed system, then we can write this transformation rule compactly as
µ µ @x0 ⌫ dx0 = dx (2.5) @x⌫ (with an implicit sum over the ⌫ subscript: we will consider a superscript in the denominator of a partial derivative to equivalent to a subscript). Note inNow that note context that that in aequation coordinate 2.6 looks basis, just the like values the transformation of dp and dq laware for the theactual components components of a of the infinitesimal ⌫ µ µ four-vectordisplacement in spacetime vector with @dxs0 in/@x thereplacing primed⇤ system⌫ . Indeed, and if youdu takeand partialdw are derivatives the same of the in the Lorentz unprimed system. Since transformationby definition, functions thet0 components(t, x, y, z), x0(t, of x, y,an z), arbitraryy0(t, x, y, z vector), and zA0(t,transform x, y, z), you in will the see same that in ways fact as the components of the displacement vector, The transformationµ law for the components of A is @x0 µ ⌫ = ⇤ ⌫ (2.7) Note in that context that equation 2.6 looks just@x like the transformationµ law for the components of a µ @x0 ⌫ four-vector in spacetime with @x ⌫ /@xµ replacing ⇤µ . Indeed,A if0 you= take partialA derivatives of the Lorentz (2.6) for that particular coordinate0 transformation. We⌫ see, therefore, that@ expressingx⌫ the transformation coe - transformationcients in terms functions of partialt0(t, derivatives x, y, z), x0 is(t, consistent x, y, z), y0( witht, x, y, but z), generalizes and z0(t, x, the y, z transformation), you will see that between in fact cartesian coordinatesAgain, in the inertial extension frames that to wefour-dimensional considered earlier. spacetime is straightforward. But this approach is only straight- @x µ Now,forward basic partialif we di are↵erential using calculus a coordinate implies0 basis that= ⇤µ, something that we will simply assume(2.7) from now on. @x⌫ ⌫ µ ⌫ Note in that context that equation@x0 @x 2.6 looks just like the transformation law for the components of a for that particular coordinate transformation. We⌫ see,µ therefore,= µ thatµ expressing the transformation coe (2.8)- four-vector in spacetime with @x0 /@⌫x replacing↵ ↵ ⇤ . Indeed, if you take partial derivatives of the Lorentz cients in terms of partial derivatives is consistent@ withx @ butx0 generalizes the⌫ transformation between cartesian transformation functions t (t, x, y, z), x (t, x, y, z), y (t, x, y, z), and z (t, x, y, z), you will see that in fact coordinatesFor example, in inertial if we write frames this that out we for considered0 our u, w earlier.p,0 q transformations,0 this says that 0 ! Now, basic partial di↵erential calculus implies that µ @p @u @p @w dp @p @u@x0@p @w µ dp + = µ=1,⌫ + ⌫ = ⇤=⌫ =0, (2.7) @u @p @w @p @xdp0 @x @µu @q @x@w @q dq ⌫ ↵ = ↵ (2.8) @q @u @q @w @dqx @x0 @q @u @q @w dq for that particular coordinate+ = transformation.=0, We+ see, therefore,= = 1 that expressing (2.9)the transformation coe - For example, if we write this@u out@p for@ ourw @u,p w dp p, q transformations,@u @q @w this@q saysdq that cients in terms of partial derivatives! is consistent with but generalizes the transformation between cartesian µ ⌫ µ ⌫ Equationcoordinates 2.8 basically in@p inertialsays@u that@p frames@@xw0 /@xdp thatand we@x considered/@@xp0@urepresent@ earlier.p @w inversedp transformations, analogous to µ 1 µ + = =1, + = =0, ⇤ ⌫ and (⇤Now, ) ⌫ basicin flat@u partial spacetime.@p @ diw ↵@erentialp dp calculus@u implies@q @w that@q dq Now we can easily state generalized transformation laws for arbitrary tensors: an nth-rank tensor’s com- @q @u @q @w dq ↵@q @u @q @w dq n ponents transform: we define+ an nth-rank= tensor=0, T ··· ···+toµ be an⌫=n-index= 1 object (with 2 components (2.9) @x0 @x µ n @u @p @w @p dp @u ···@q @w @q =dq (2.8) in a 2D space and 4 components in spacetime) that transforms@x⌫ according@x ↵ to ↵ µ ⌫ µ ⌫ 0 Equation 2.8 basically says that @x0 /@x and @↵x /@x0 ⌫ represent inverse transformations, analogous to µ 1 µ ↵ @x0 @x @x0 µ ⇤ andFor (⇤ example,) in flat spacetime. if we write0 ··· this··· out for our u, w p, q transformations, this says that ⌫ ⌫ T = µ T ···⌫ ··· (2.10) Now we can easily state generalized··· transformation@x ···@ lawsx0 ··· for arbitrary@!x ··· tensors:··· an nth-rank tensor’s com- @p @u @p @↵w dp @p @u @p @wn dp ponentsthat is, transform: a partial-derivative we define an factornth-rank with the tensor primedT coordinate··· ··· to in be the an numeratorn-index object for every (with upper 2 components (superscript) n + ···= =1, + = =0, in aindex 2D space and one and with 4 components the primed coordinatein spacetime)@u @p in the that@w denominator@ transformsp dp according for every lower to@u @ (subscript)q @w @ index.q dq In particular, we can prove that@q the@u metric↵ @q correctly@w⌫ dq transforms as a@q tensor@u with@q two@w lowerdq indices as ↵ @x0 @x @x0 µ follows. The coordinate-independence0 ··· ··· of the+ spacetime= separation=0, implies that + = = 1 (2.9) T = µ T ···⌫ ··· (2.10) ··· @u @@px ···@w@x@0p ··· dp@x ··· ···@u @q @w @q dq µ ⌫ ↵ Tensorsgµ0 ⌫ dx0 dx0 =ing↵ dx µadx coordinate⌫ µ ⌫ basis that is, aEquation partial-derivative 2.8 basically factor with says the that primed@x coordinate0 /@x and in the@ numeratorx /@x0 forrepresent every upper inverse (superscript) transformations, analogous to ↵ index andµ one with the primed1 µ coordinate in the denominator@x ↵ for every@x lower (subscript) index. ⇤ ⌫ and (⇤ ) ⌫ in flat spacetime.= g dx0 dx In particular, we can prove that the metric correctly↵ @x transforms@x as a tensor with two lower indices as Now we can easily state generalized✓ transformation0 ◆✓ 0 laws◆ for arbitrary tensors: an nth-rank tensor’s com- follows. The coordinate-independence of the spacetime↵ separation implies that Example: the metric tensor@x @x is a tensor ↵ n ponents transform: we define an nth-rank tensor T ··· ··· to be an n-index object (with 4 components) = g↵ dx0 dx0 µ ⌫ @↵x0 @x0 ··· that transforms accordinggµ0 ⌫ dx0 dx to0 = g↵ ✓dx dx ◆ @x↵ @x @x↵ @x µ ⌫ = µ ⌫↵g↵ ↵ dx0 dx 0⌫ = g↵↵ @x 0 @dxx00@x0 dx@x @x0 µ T 0 ···✓ @x···0 = ◆@x0 T ··· ··· (2.10) ✓ ↵ ◆✓µ ◆ ⌫ ···↵@x @x @x ···@x0 µ···⌫@x ··· ··· 0=@x @x g↵ gµ0 ⌫ dx0 dx0 (2.11) ) = @x µ @xg↵ ⌫ dx 0 dx0 @✓x @0 x 0 ◆ that is, a partial-derivative factor✓ with0 the0 primed◆ coordinate in the numerator for every upper (superscript) µ ↵⌫ Now, weindex can’t and just a divide a partial-derivative both sides by dx0 factordx@x0 because@x with the a sum primedµ can⌫ be coordinate zero even if the in the individual denominator terms in for every lower (sub- = g dx0 dx0 the sum are not. But this relation must be@ truex µ for@x arbitrary⌫ ↵ di↵erential displacements. So if I choose the script) index. Using methods similar✓ 0 to0 what◆ we did in the last session, you can prove in particular that the displacement to be entirely in the p direction (↵dw = 0), then the only nonzero term in the sum is the one metric is a tensor with two lower@ indices:x @x µ ⌫ where µ = ⌫ = p, so the quantity in parentheses0= µ must⌫ g be↵ 0 whengµ0 ⌫ µdx=0 ⌫dx=0 p. In a similar way, I can(2.11) choose ) @x0 @x0 ↵ ⌫ displacement components to show that all✓ of the other components◆@x of0 that@x quantity must be zero as well: µ ⌫ gµ0 ⌫ = µ g↵ (2.11) Now, we can’t just divide both sides by↵ dx 0 dx0 because a sum can@ bex zero↵@x even⌫ if the individual terms in @x @x @x0 @0x the sum are not. But this relation0= must be trueg forg0 arbitrary dig↵0 erential= displacements.g So if I choose the(2.12) @x µ @x ⌫ ↵ µ⌫ ) µµ⌫⌫ @xµ @x ↵ µ⌫ ⌫ displacementthat to the be Kronecker entirely in the delta✓ p direction0 is still0 a (dw tensor,= 0),◆ thenthat theg onlydefined nonzero such0 term that in theg sumgµ↵ is= the ↵ one(the matrix inverse of the µ⌫ wherewhichµ metric= is⌫ the= correctp, tensor), so the transformation quantity is still in a parentheses tensor, law for a that tensor. must multiplying Using be 0 when similarµ by= methods,g⌫µ=⌫ pand. In youg a can similarand show way, summing that I the can Kronecker choose over µ still lowers or raises a µ⌫ µ⌫ ⌫ displacementdelta is still components a tensor, that to showg defined that all such of the that otherg componentsgµ↵ = ↵ (the of that matrix quantity inverse must of the be metric zero as tensor), well: is still a tensor, that multiplying by g and gµ⌫ and summing over µ still lowers or raises a tensor index, that ↵ µ⌫ ↵ ⌫ @x @x @x0 4 @x contracting over an upper0= and lower indexg ofg a0 tensor quantityg0 = still yields ag tensor with rank n 2,(2.12) etc. @x µ @x ⌫ ↵ µ⌫ ) µ⌫ @xµ @x ↵ ✓ 0 0 ◆ 0 which is the correct transformation law for a tensor. Using4 similar methods, you can show that the Kronecker µ⌫ µ⌫ ⌫ delta is still a tensor, that g defined such that g gµ↵ = ↵ (the matrix inverse of the metric tensor), is µ⌫ still a tensor, that multiplying by gµ⌫ and g and summing over µ still lowers or raises a tensor index, that contracting over an upper and lower index of a tensor quantity still yields a tensor with rank n 2, etc.
4 related to changes in the old coordinates as follows: @p @p @q @q related to changes indp the= olddu coordinates+ dw asand follows:dq = du + dw (2.4) @u @w @u @w @p @p @q @q dp = du + dw and dq = du + dw (2.4) If we consider the p, q coordinates the primed@u coordinate@w system and u, w@coordinatesu @w the unprimed system, then we can write this transformation rule compactly as If we consider the p, q coordinates the primed coordinate system and u, w coordinates the unprimed system, µ then we can write this transformation ruleµ compactly@x0 ⌫ as dx0 = ⌫ dx (2.5) @x µ µ @x0 ⌫ dx0 = dx (2.5) (with an implicit sum over the ⌫ subscript: we will consider a@ superscriptx⌫ in the denominator of a partial derivative to equivalent to a subscript). Now note(with that an in implicit a coordinate sum over basis, the the⌫ valuessubscript: of dp weand willdq considerare the actual a superscript componentsin theof the denominator infinitesimal of a partial displacementderivative vectortods equivalentin the primed to a subscript). system and du and dw are the same in the unprimed system. Since by definition,Now the components note that in a of coordinate an arbitrary basis, vector theA valuestransform of dp and in thedq sameare the waysactual as the components componentsof the of the infinitesimal displacementdisplacement vector, The vector transformationds in the primed law for system the components and du and of dwA isare the same in the unprimed system. Since by definition, the components of an arbitrary vector A transform in the same ways as the components of the µ displacement vector, The transformationµ law@x for0 the⌫ components of A is A0 = ⌫ A (2.6) @x µ µ @x0 ⌫ A0 = A (2.6) Again, the extension to four-dimensional spacetime is straightforward.@x⌫ But this approach is only straight- forward if we are using a coordinate basis, something that we will simply assume from now on. Note inAgain, that the context extension that equation to four-dimensional 2.6 looks just spacetime like the is transformation straightforward. law But for this the components approach is only of a straight- forward if we are using⌫ a coordinateµ basis,µ something that we will simply assume from now on. four-vector in spacetime with @x0 /@x replacing ⇤ ⌫ . Indeed, if you take partial derivatives of the Lorentz Note in that context that equation 2.6 looks just like the transformation law for the components of a transformation functions t0(t, x, y, z), x0(t, x, y, z), y0(t, x, y, z), and z0(t, x, y, z), you will see that in fact ⌫ µ µ four-vector in spacetime with @x0 /@x replacing ⇤ ⌫ . Indeed, if you take partial derivatives of the Lorentz µ transformation functions t (t, x, y, z), x@(xt,0 x, y, z), y (t, x, y, z), and z (t, x, y, z), you will see that in fact 0 0 = ⇤µ 0 0 (2.7) @x⌫ ⌫ @x µ 0 = ⇤µ (2.7) for that particular coordinate transformation. We see, therefore,@x⌫ that⌫ expressing the transformation coe - cients in terms of partial derivatives is consistent with but generalizes the transformation between cartesian coordinatesfor in that inertial particular frames coordinate that we considered transformation. earlier. We see, therefore, that expressing the transformation coe - Now, basiccients partial in terms di↵ oferential partial calculus derivatives implies is consistent that with but generalizes the transformation between cartesian coordinates in inertial frames that we considered earlier. µ ⌫ Now, basic partial di↵erential calculus@x0 @ impliesx thatµ ⌫ ↵ = ↵ (2.8) @x @x0 µ ⌫ @x0 @x µ For example, if we write this out for our u, w p, q transformations,⌫ ↵ = ↵ this says that (2.8) ! @x @x0 @p @u @p @w dp @p @u @p @w dp For example, if we write+ this out= for our=1u,, w p, q +transformations,= =0 this, says that @u @p @w @p dp !@u @q @w @q dq @p @u @p @w dp @p @u @p @w dp @q @u @q @w+ dq = =1@q @, u @q @w+ dq = =0, + @u @p =@w @=0p , dp + @u @q =@w @=q 1dq (2.9) @u @p @w @p dp @u @q @w @q dq @q @u @q @w dq @q @u @q @w dq µ +⌫ =µ =0⌫ , + = = 1 (2.9) Equation 2.8 basically says that @@xu0 @/p@x @andw @p@x /dp@x0 represent@u @ inverseq @w transformations,@q dq analogous to µ 1 µ ⇤ and (⇤ ) in flat spacetime. ⌫ ⌫ µ ⌫ µ ⌫ Now weEquation can easily 2.8 state basically generalized says that transformation@x0 /@x and laws@x for/@ arbitraryx0 represent tensors: inverse an nth-rank transformations, tensor’s com- analogous to µ 1 µ ↵ ⇤ ⌫ and (⇤ ) ⌫ in flat spacetime. n ponents transform: we define an nth-rank tensor T ··· ··· to be an n-index object (with 4 components) that transformsNow according we can easily to state generalized transformation··· laws for arbitrary tensors: an nth-rank tensor’s com- ↵ n ponents transform: we define an nth-rank tensor T ··· ··· to be an n-index object (with 4 components) ↵ ⌫ ··· that transforms according↵ to @x0 @x @x0 µ 0 ··· ··· T = µ T ···⌫ ··· (2.10) ··· @x ···@x0 ↵ ··· @x ⌫ ··· ··· ↵ @x0 @x @x0 µ 0 ··· ··· that is, a partial-derivative factor withT the primed = coordinateµ in the numerator T for···⌫ every··· upper (superscript) (2.10) ··· @x ···@x0 ··· @x ··· ··· index and a a partial-derivative factor with the primed coordinate in the denominator for every lower (sub- script) index.that is,Using a partial-derivative methods similar factor to what with we the didTensors primed in the last coordinate session, in in you the cana numerator provecoordinate in for particular every upper that (superscript) the basis metric isindex a tensor and with a a two partial-derivative lower indices: factor with the primed coordinate in the denominator for every lower (sub- script) index. Using methods similar to what we did in the last session, you can prove in particular that the Other↵ tensors⌫ and tensor operations (which you can metric is a tensor with two lower indices:@x0 @x gµ0 ⌫ = proveµ in ga↵ similar way): (2.11) @x @x0 ↵ ⌫ @x0 @x µ⌫ g0 = µg⌫↵ ⌫ (2.11) that the Kronecker delta is still a tensor, that g definedµThe⌫ such Kroneckerµ that g delta:gµ↵ = (the matrix inverse of the @x @x0 ↵ metric tensor), is still a tensor, that multiplying by g and gµ⌫ and summing over µ still lowers or raises a µ⌫ µ⌫ µ⌫ ⌫ that the Kronecker delta is still a tensor, thatTheg inversedefined metric: such that g gµ↵ = ↵ (the matrix inverse of the µ⌫ metric tensor), is still a tensor, that multiplying by gµ⌫ and g and summing over µ still lowers or raises a ν 4Lowering an index: Aμ = gμνA μ μν Raising 4an index: B = g Bν Gradients tensor index, that contracting over an upper and lower index of a tensor quantity still yields a tensor with tensor index, that contracting over an upper and lower index of a tensor quantity still yields a tensor with rank n 2, and so on. The gradient of a scalar also still yields a tensor (covector): by the chain rule rank n 2, and so on. The gradientThe gradient of a scalar of a alsoscalar still is yields a covector: a tensor (covector): by the chain rule @ @x⌫ @ @x⌫ ⌫ ⌫ @ @µ0 @xµ =@ µ @⌫x= µ (@⌫ ) (2.12) ⌘ @x0 @x0 @x @x0 @µ0 µ = µ ⌫ = µ (@⌫ ) (2.12) which is the correct generalized⌘ transformation@x0 @x0 @ lawx for@ ax tensor0 with one lower index. However (unlike in which is thethe correct Lorentz generalized transformation transformation case), the gradient law for of a vector tensor (or with any one larger-rank lower index. tensor) However is not a tensor: (unlike by in the the Lorentzchain transformation and product case), rulesThe the gradient gradient of of a a vector vector (or is NOT any larger-rank a tensor: tensor) is not a tensor: by the chain and product rules ↵ ↵ ↵ ↵ @A @ @x0 ⌫ @x @ @x0 ⌫ @µ0 A µ = µ ⌫ A = µ ⌫ A @A↵ ⌘ @x@0 @@xx0 ↵ @x @x @@x0 @@xx ↵ @x @ A↵ = 0 A✓⌫ = ◆ 0✓ A⌫ ◆ µ0 µ @x µ @2x⌫ ↵ @x µ @x ↵ @A⌫ ⌘ @x0 @x0 @x 0 ⌫ @x0 @x0 @x = µ ✓ ⌫ A◆ + µ ⌫ ✓ ◆ (2.13) @2x0 ↵ @x @x @x0↵ @x ⌫ @x @x @ x0 ✓ ⌫ @x◆ @x0 @A ✓ ◆ The last term looks like= the transformationµ ⌫ A law+ for a tensorµ ⌫ quantity with one upper and one lower index,(2.13) but @x0 @x @x @x0 @x @x the first term in the last line does✓ not. This◆ term did not arise✓ in the◆ Lorentz transformation case because The last termthe looks Lorentz like transformation the transformation coe cients law arefor constant,a tensor quantity implying that with the one double upper partial and one derivatives lower index, are all but zero. the first term inThis the is last a serious line does problem, not. because This term many did physics not arise equations in the involve Lorentz calculating transformation derivatives case of because quantities the Lorentzthat transformation will generalize coe to vectorscients are or tensors constant, of higher implying rank. that We the will double address partial this problem derivatives in the are next all section. zero. This is a serious problem, because many physics equations involve calculating derivatives of quantities that will generalize2.3.1 Exercise: to vectors Parabolic or tensors coordinates. of higher rank. We will address this problem in the next section. Consider the parabolic coordinate system p, q shown in figure 2. The transformation functions from ordinary 2.3.1 Exercise: Parabolic coordinates. Consider the parabolic coordinate system p, q shown in figure 2. The transformationq = 3 m functions from ordinary q = 2 m q = 3 mq = 1 m eq q ep q = 2 mq = 0 A q = –1 m qA =y 1 m eq eq q ep Ax q = 0 q = –2 m O ep q = –3 m A A q = –1 m e y q A q = –2 m p = –4 m –2 m 0 x 2 m 4 m O ep q = –3 m Figure 2: Parabolic coordinates for a flat two-dimensional plane. Adapted from Moore, A General Relativity Workbook, University Science Books, 2013, p. 60.) p = –4 m –2 m 0 2 m 4 m cartesian x, y coordinates are p(x, y)=x and q(x, y)=y cx2 (2.14) Figure 2: Parabolic coordinates for a flat two-dimensional plane. Adapted from Moore, A General Relativity Workbook, Universitywhere c is a Science constant Books, with units 2013, of p.inverse 60.) meters. (a) Show that the inverse transformation functions are cartesian x, y coordinates are 2 p(x, y)=x x(andp, q)=qp(x,and y)=yy(p, qcx)=2 cp + q (2.14)(2.15) µ ⌫ µ ⌫ where c is a(b) constantEvaluate with all units eight of partial inverse derivatives meters. @x0 /@x and @x /@x0 .
(a) Show that(c) theThe inverse metric tensor transformation components functions for cartesian are coordinates in space are gxx = gyy =1,gxy = gyx = 0. Use the general tensor transformation rule to show that the metric tensor for p, q coordinates is x(p, q)=p and y(p, q)=cp2 + q (2.15) 1+4c2p2 2cp g0 = (2.16) µ ⌫ µ⌫ µ2cp ⌫ 1 (b) Evaluate all eight partial derivatives @x0 /@x and @x /@x0 . p q (c) The metric(d) Let tensor a vector componentsA have components for cartesianA coordinates=1,A =0inthe in spacep, q arecoordinategxx = gyy system.=1,g Findxy = thisgyx vector’s= 0. Use the generalcomponents tensor in transformation the x, y coordinate rule system to show (as that a function the metric of x tensorand y). for Butp, show q coordinates that A2 = isA ⇧ A has the same value in both coordinate systems at every position. 1+4c2p2 2cp g = (2.16) µ0 ⌫ 2cp 1 5 (d) Let a vector A have components Ap =1,Aq =0inthep, q coordinate system. Find this vector’s components in the x, y coordinate system (as a function of x and y). But show that A2 = A ⇧ A has the same value in both coordinate systems at every position.
5 tensor index, that contracting over an upper and lower index of a tensor quantity still yields a tensor with rank n 2, and so on. The gradient of a scalar also still yields a tensor (covector): by the chain rule @ @x⌫ @ @x⌫ @µ0 µ = µ ⌫ = µ (@⌫ ) (2.12) ⌘ @x0 @x0 @x @x0 which is the correct generalized transformation law for a tensor with one lower index. However (unlike in the Lorentz transformation case), the gradient of a vector (or any larger-rank tensor) is not a tensor: by the chain and product rules ↵ ↵ ↵ ↵ @A @ @x0 ⌫ @x @ @x0 ⌫ @0 A = A = A µ ⌘ @x µ @x µ @x⌫ @x µ @x @x⌫ 0 0 ✓ ◆ 0 ✓ ◆ @x @2x ↵ @x @x ↵ @A⌫ = 0 A⌫ + 0 (2.13) @x µ @x @x⌫ @x µ @x⌫ @x 0 ✓ ◆ 0 ✓ ◆ The last term looks like the transformation law for a tensor quantity with one upper and one lower index, but the first term in the last line does not. This term did not arise in the Lorentz transformation case because the Lorentz transformation coe cients are constant, implying that the double partial derivatives are all zero. This is a serious problem, because many physics equations involve calculating derivatives of quantities that will generalize to vectors or tensors of higher rank. We will address this problem in the next section.
2.3.1 Exercise: ParabolicExercise: coordinates. Parabolic coordinates Consider the parabolic coordinate system p, q shown in figure 2. The transformation functions from ordinary
q = 3 m p(x, y) = x q = 2 m 2 q = 1 m q(x, y) = y − cx eq q ep q = 0 A q = –1 m Ay eq Ax q = –2 m O ep q = –3 m
p = –4 m –2 m 0 2 m 4 m
Figure 2: Parabolic coordinates for a flat two-dimensional plane. Adapted from Moore, A General Relativity Workbook, University Science Books, 2013, p. 60.) cartesian x, y coordinates are p(x, y)=x and q(x, y)=y cx2 (2.14) where c is a constant with units of inverse meters. (a) Show that the inverse transformation functions are x(p, q)=p and y(p, q)=cp2 + q (2.15)
µ ⌫ µ ⌫ (b) Evaluate all eight partial derivatives @x0 /@x and @x /@x0 .
(c) The metric tensor components for cartesian coordinates in space are gxx = gyy =1,gxy = gyx = 0. Use the general tensor transformation rule to show that the metric tensor for p, q coordinates is 1+4c2p2 2cp g = (2.16) µ0 ⌫ 2cp 1 (d) Let a vector A have components Ap =1,Aq =0inthep, q coordinate system. Find this vector’s components in the x, y coordinate system (as a function of x and y). But show that A2 = A ⇧ A has the same value in both coordinate systems at every position.
5 tensor index, that contracting over an upper and lower index of a tensor quantity still yields a tensor with rank n 2, and so on. The gradient of a scalar also still yields a tensor (covector): by the chain rule @ @x⌫ @ @x⌫ @µ0 µ = µ ⌫ = µ (@⌫ ) (2.12) ⌘ @x0 @x0 @x @x0 which is the correct generalized transformation law for a tensor with one lower index. However (unlike in the Lorentz transformation case), the gradient of a vector (or any larger-rank tensor) is not a tensor: by the chain and product rules ↵ ↵ ↵ ↵ @A @ @x0 ⌫ @x @ @x0 ⌫ @0 A = A = A µ ⌘ @x µ @x µ @x⌫ @x µ @x @x⌫ 0 0 ✓ ◆ 0 ✓ ◆ @x @2x ↵ @x @x ↵ @A⌫ = 0 A⌫ + 0 (2.13) @x µ @x @x⌫ @x µ @x⌫ @x 0 ✓ ◆ 0 ✓ ◆ The last term looks like the transformation law for a tensor quantity with one upper and one lower index, but the first term in the last line does not. This term did not arise in the Lorentz transformation case because the Lorentz transformation coe cients are constant, implying that the double partial derivatives are all zero. This is a serious problem, because many physics equations involve calculating derivatives of quantities that will generalize to vectors or tensors of higher rank. We will address this problem in the next section.
2.3.1 Exercise: Parabolic coordinates. Consider the parabolic coordinate system p, q shown in figure 2. The transformation functions from ordinary
q = 3 m q = 2 m q = 1 m eq q ep q = 0 A q = –1 m Ay eq Ax q = –2 m O ep q = –3 m Exercise:p = –4 mParabolic –2 m 0 2 m 4 m coordinates Figure 2: Parabolic coordinates for a flat two-dimensional plane. Adapted from Moore, A General Relativity Workbook, University Science Books, 2013, p. 60.) cartesian x, y coordinates are 2 p(x, y) =p(x,x y)=x andq(xq,(x,y) y)== yy −cx2cx (2.14) where c is a constant with units of inverse meters. (a) Show that the inverse transformation functions are x(p, q)=p and y(p, q)=cp2 + q (2.15)
µ ⌫ µ ⌫ (b) Evaluate all eight partial derivatives @x0 /@x and @x /@x0 .
(c) The metric tensor components for cartesian coordinates in space are gxx = gyy =1,gxy = gyx = 0. Use the general tensor transformation rule to show that the metric tensor for p, q coordinates is 1+4c2p2 2cp g = (2.16) µ0 ⌫ 2cp 1 (d) Let a vector A have components Ap =1,Aq =0inthep, q coordinate system. Find this vector’s components in the x, y coordinate system (as a function of x and y). But show that A2 = A ⇧ A has the same value in both coordinate systems at every position.
5 2.4 The Tensor Gradient To see more clearly why the simple gradient of a vector field is not a tensor consider the simple case of a constant vector field A(xµ) in a flat two-dimensional space. In flat space at least, we can define “constant” in a coordinate-independent way by saying that at all points, the vector points in the same direction and 2.4 The Tensor Gradient has2.4 the same The magnitude. Tensor Gradient Therefore the physical gradient of such a field should be zero in any coordinate To see more clearly why the simple gradient of a vector field is not a tensor consider the simple case of a systemTo see (because more clearly the vector why the does simple not gradient change as of a we vector change field positions). is not a tensor In a consider cartesian the coordinate simple case system, of a the constant vector field Aµ(xµ) in a flat two-dimensional space. In flatµ space at least, we can define “constant” components A of suchA a fieldµ are constant, so @↵A = 0 as expected. But in a curvilinear coordinate system, in a coordinate-independentconstant vector field way(x by) saying in a flat that two-dimensional at all points, the space. vector In points flat space in the at same least, direction we can and define “constant” thein components a coordinate-independent of even a truly way constant by saying vector that field at all may points,not thebe vector constant, points because in the the same basis direction vectors and used to has the same magnitude. Therefore the physical gradient of such a field should be zero in any coordinate µ systemdefine (becausehas the the the same components vector magnitude. does change not Therefore change as one as we the goes change physical from positions). gradient point to In of point. a such cartesian a We field coordinate need should to find be system, zero a way the in any to correct coordinate@↵A in µ µ componentssuchsystem aA coordinateof (because such a field system the are vector constant, so does as to so not remove@↵ changeA = the 0 as as part expected. we change due But to positions). changes in a curvilinear in In the a cartesian coordinate basis vectors coordinate system, if we system, hope to the find the µ µ µ the componentstruecomponents change of even inA the aof truly vector such constant a fieldfunction are vector constant,A( fieldx ). may so @not↵Abe= constant, 0 as expected. because But the basis in a curvilinear vectors used coordinate to system, ⌫ µ define thetheTensor components components Gradient change of even as of one a a truly Vector. goes constant fromLet point vector us to define point. field a may We set need ofnot coe tobe find constant,cients a way because to(which correct the@ physicists↵ basisA in vectors call usedChristo to ↵el ⌫↵ µ suchsymbols a coordinatedefine the) at system components a given so as point to change remove or event as the one part goessuch due from to that changes point in to the point. basis We vectors need if to we find hope a to way find to the correct @↵A in true changesuch in a the coordinate vector function systemA so(xµ as). to removeP the part due to changes in the basis vectors if we hope to find the 2.4 The Tensor Gradient µ ⌫ Tensortrue Gradient change in of the a Vector. vector functionLet us defineA(x a). set of coe@ e↵cients ⌫⌫↵ (which physicists call Christo↵el To seesymbols more clearly) at a given why the point simple or event gradientsuch of a that vector field is not a tensor µ↵ considere⌫ the⌫ simple case of a (2.17) Tensor Gradient of a Vector. Let us define@x aµ set⌘ of coe cients ⌫↵ (which physicists call Christo↵el constant vector field A(xµ) in a flat two-dimensionalP space. In flat space at least, we can define “constant” symbols) at a given point or event @esuch that in a coordinate-independentThe numerator here way by is saying the di that↵erential at allP ↵ points, change⌫ the in vector the basis points vector in the samee as direction we move and from point to a point µ µ↵e⌫ ↵ (2.17) µ @x ⌘ P has the sameadi magnitude.↵erential displacement Therefore the physicaldx along gradient a curve of such@e where a↵ field should⌫ the other be zero coordinates in any coordinate are constant, divided by that µ↵e⌫ (2.17) systemThe (because numerator the here vector is does the di not↵erential changeµ change as we change in the positions). basis@ vectorxµ ⌘ Ine a↵ cartesianas we move coordinate from point system,to the a point e di↵erentialµ displacementµ dx (whereµ ↵ and µ have some specific values here).P Since the change @ ↵ is a componentsadi↵erentialA of displacement such a field aredx constant,along so a@ curve↵A = where 0 as expected. the other But coordinates in a curvilinear are constant, coordinate divided system, by that vectorThe (an numerator arrow here withµ is a the certain di↵erential magnitude change pointing in the basis in a vector certaine↵ direction),as we move we from can point write thatto a change point as a the componentsdi↵erential displacement of even a trulydx constant(where vector↵ and fieldµ have may not somebe specific constant, values because here). the basis Since vectors the change used to@e↵ is a µ µ P definevector thesum components (anadi over arrow↵erential the with change basis a displacement certain as vectors one magnitude goese⌫dx fromevaluatedalong point pointing to a at curvepoint. in point a certain where We need: this direction), the to is other find the a we point coordinates way can to of write correct the that are sum@↵A constant, change overin ⌫ as. aIndivided a two-dimensional by that µ P such a coordinatedi↵erential system displacement so as to removedx the(where part due↵ toand changesµ have in the some basis specific vectors values if we hope here). to find Since the the change @e↵ is a sum overspace, the there basis vectors are 8 suche⌫ evaluated coe Tensorcients; at point in four-dimensional: this gradient is the point of spacetime, the sum of over a there⌫. Invector a are two-dimensional 64 such coe cients. µ truespace, change there invectorNow the are vector consider (an 8 such arrow function coe a withcients; vectorA(x a). certain in field four-dimensionalA magnitudethatP is a pointing spacetime, function in there aof certain position. are 64 direction), such The coe productcients. we can write rule thatimplies change that as the a true ⌫ TensorNowamount Gradientsum consider overdA a of thethat vector a basis Vector.A changes field vectorsLetA that use when⌫ defineevaluated is a we functiona set move of at coe point of an position.cients arbitrary: this ⌫↵ The(which is infinitesimal the product point physicists of rule the displacement call implies sumChristo over that↵⌫el the.d Ins true(whose a two-dimensional components are symbols) at a given point or event such that P amount dspace,A that thereA changes are 8 when suchP we coe move Definecients; an arbitrary ↵Christoffel in four-dimensional infinitesimal symbols: displacement spacetime, thereds (whose are 64 components such coe arecients. some specific set of components↵ dx )is some specificNow set of consider components a vectordx )is field@Ae that is a function of position. The product rule implies that the true ↵ ⌫ e A A µ µ↵ ⌫ µ (2.17)s amount d that changes when@x we⌘ moveµµ an arbitrary@A infinitesimal displacementµ @eµ ↵ d (whose components are µdA =↵ d(@AA e )= µ dx@eµ ↵e + A dx (2.18) some specific set of componentsdA = d(A eµdx)=)is dxµ eµ + A dx µ ↵ (2.18) The numerator here is the di↵erential changeThe in physical the basisdx change vector e in↵dx asa vector we@x move↵ is then: from point@x to a point ✓ ◆ P adi↵erential displacement dxµ along a curve where✓ the other◆ coordinatesµ are constant, divided by that µ @A µ @eµ ↵ di↵erentialOne canOne displacement use can equation use equationdx 2.17µ (where and 2.17 rename↵ and and someµd renamehaveA = indices somed(A some toe specificµ)= rewrite indices values thisdx to as here). rewriteeµ Since+ A this the as changedx @e is a (2.18) dx @x↵ ↵ vector (an arrow with a certain magnitude pointingµ in a certain✓ direction),◆ we can write that change as a @A µ @⌫Aµ ↵ µ ↵ sum over the basis vectors e⌫ evaluateddA = at point +: this↵⌫ A is theeµdx pointµ of( ⌫ the↵A sum)eµdx over↵ ⌫. In aµ two-dimensional↵ (2.19) One can use equation 2.17 and@x↵d renameA = some+ indices ↵⌫⌘ Ar toe rewriteµdx this( as↵A )eµdx (2.19) space, there are 8 such coe cients; in four-dimensional P @ spacetime,x ↵ there are 64 such⌘ coer cients. µ NowThe consider quantities a ( vectorAµ) fielddx↵ Aforthat di↵erent is a function values of of@Aµ position.are by definition The product components rule implies (in whatever that the true coordinate ↵ µ ↵ dA = + µ A⌫ e dx↵ ( Aµ)e dx↵ (2.19) amount dTheA that quantitiesA changesr when ( ↵A we) movedx anfor arbitrary di↵erent infinitesimal↵ values↵⌫ of displacementµ µare byd definitions (whose↵ µ components components are (in whatever coordinate system we are using) of ther true changeSoin thisA for tensor that@x digradient↵erential is displacement. a tensor⌘ r Therefore, we define some specificsystem set of we components are using)dx↵ of)is the true change in A for that di↵erential displacement. Therefore, we define µ ↵ @Aµ The quantities ( ↵A )dx for di↵erentµ values ofµ µ⌫are by definition components (in whatever coordinate ↵A µ + A (2.20) r @A ↵ ↵⌫ @eµ µ system we are using) of theµ truer change⌘ @inx A forµ thatµ @ diA↵↵erentialµ displacement.⌫ Therefore, we define dA = d(A eµ)= dx eµ + A ↵ dx (2.18) dx ↵A @x ↵ + ↵⌫ A (2.20) to be the tensor gradient of the vector✓ field A: the◆r term involving⌘ @µx the Christo↵el symbols corrects the µ @A µ ⌫ µ Onepartial can use derivative equation 2.17 of the and field rename for the some variations indices to in rewrite the basis this vectors. as The tensor gradient ↵A must be a ↵A ↵ + ↵⌫ A (2.20) (second-rank)to be the tensor,tensor because gradient we see thatof equation the vector 2.19r field implies⌘A:@ thatx the ( termAµ)dx involving↵ yields the ther components Christo↵el of symbolsa corrects the @Aµ r↵ µ vector.partial However,to be derivative the neithertensor thed ofA gradient the= partial field derivative+of for µ theA the⌫ vector alonee variationsdx↵ nor field( theA inA: Christoµ the) thee dx term↵ basis↵el symbol involving vectors. alone the The are Christo tensors: tensor(2.19)↵el only gradient symbols the corrects↵A must the be a ↵ ↵⌫ µ ↵ µ µ ↵ r (second-rank) tensor, because@x we see that equation⌘ r 2.19 implies that ( A )dx yields theµ components of a combinationpartial given derivative by equation of the 2.20 field is a for tensor. the variations in the basis vectors. The tensor↵ gradient ↵A must be a The Tensor Gradientµ ↵ of a Covector. To determine a covector’s tensor gradient, werµ can take↵ advantager The quantitiesvector.(second-rank) ( ↵ However,A )dx for tensor, neither di↵erent because the values partial we of seeµ are that derivative by equation definition alone 2.19 components nor implies the that Christo (in whatever ( ↵Ael) symboldx coordinateyields alone the componentsare tensors: of only a the r µ ↵ systemof the wecombination fact arevector. using) that A However, ofB the givenµ istrue a scalar by neither change equation, and thein A the partialfor 2.20 tensor that is derivative di gradient a↵erential tensor. of alone displacement. a scalar nor is the the Therefore,Christo same as↵rel its we symbol ordinary define alone gradient: are tensors: only the combinationThe Tensor given Gradient by equation of a@ 2.20 Covector.Aµ isµ a tensor.@BTo determine a covector’s tensor gradient, we can take advantage µ µ @A µ µ µ⌫ µ @↵(A Bµ)=A Bµ ++A A (2.20) of theThe fact Tensor that A GradientBµ is a scalarr of↵ a@ Covector.x,⌘↵ and@x↵ the tensor↵⌫@Tox↵ determine gradient a covector’s of a scalar tensor is the gradient, same as we its can ordinary take advantage gradient: µ of the fact that A Bµ is a scalar= ,( andAµB the)=( tensorAµ gradient)B + Aµ of( a scalarB ) is the same as its (2.21) ordinary gradient: to be the tensor gradient of the vector fieldrA↵: theµ term involvingrµ↵ µ the Christor↵ ↵µel symbols corrects the µ @A µ @Bµ µ partial derivative of the field for the variations in the basis vectors.µ The tensorµ gradient A must be a Subtracting one side from the other and@↵( substitutingAµBµ)=@ whatA ↵ B weµ know+ Aµ @BµA↵ to be yields↵ @x µ@↵x ↵ r (second-rank) tensor, because we see that equation@↵(A B 2.19µ)= implies↵ B thatµ + (A↵Ar)↵dx yields the components of a µ @x µ r @x µ µ vector. However, neither the partial@A derivativeµ @ aloneBµ nor= theµ↵ Christo(Aµ B↵µµel)=( symbol↵µA alone)B areµ +µ tensors:A ( ↵ onlyBµ) the (2.21) 0= Bµ + A (=↵A )(BAµ B A)=(( ↵BµA) )B + A ( B ) (2.21) combination given by equation 2.20@x↵ is a tensor.@x↵ r r↵ µ r ↵r µ ↵ rµ r r r µ Subtracting one side fromµ ⇢ the other andµ substituting⇢ what we know A to be yields The Tensor Gradient of a Covector.@A ⇢ To@ determineBµ @A a⇢ covector’s tensor gradient, we can takeµ↵ advantage Subtractingµ one side from theµ other and substitutingµ ⌫ what weµ know ↵rA to be yields of the fact that A B is a scalar= , and⇢ B theµ + tensorA gradient⇢ ofB aµ scalar( ↵⌫ isA the)Bµ sameA as( its↵B ordinaryµ) gradient: µ @x↵ µ@x↵ @x↵ r r ⇢ @A µ ⇢ µ @Bµ µ µ @0=B @µA B + Aµ @Bµ ( µA )B µA ( B ) µµ 0=µ@A ↵⌫ Bµ +µA@Bµ ↵ ( ↵A↵ )Bµ µ A ( ↵Bµ↵) µ @ =(AAB )= @x B↵µB+⌫ A ↵B@µx↵ r r (2.22) ↵ µ @x↵ @ x@↵x µ r@x↵@x r r µµµ ⇢ µ µµµ⇢⇢ = @↵@A(AA ⇢⇢Bµ)=(µµ↵@@ABBµ)µBµ @+A@AA⇢(⇢↵Bµ)µ µ⌫ ⌫ µ µ (2.21) ==r ⇢⇢↵↵ BBµ ++AAr ↵↵ ⇢↵⇢↵BrµBµ ( ↵⌫( A↵⌫ A)Bµ)BµA (A ↵(Bµ↵) Bµ) ⇢⇢@@xx @6@xx ⇢@⇢x@x µ r r Subtracting one side from the other and substituting what we know ↵A to be yields @@Bµ r = Aµµ µ ⌫⌫ B B (2.22) @Aµ =@AB ↵ ↵↵µµB⌫⌫ ↵ ↵Bµ µ (2.22) 0= B + Aµ µ @(x↵A µ)B Aµ r( rB ) @x↵ µ @x↵ r↵ µ r↵ µ µ ⇢ µ ⇢ @A ⇢ µ @Bµ @A ⇢ µ ⌫ µ = ⇢ Bµ + A ⇢ Bµ ( ↵⌫ A )Bµ 6 A ( ↵Bµ) ⇢@x↵ @x↵ ⇢@x↵ 6 r @B = Aµ µ ⌫ B B (2.22) @x↵ ↵µ ⌫ r↵ µ
6 2.4 The Tensor Gradient To see2.4 more The clearly Tensor why the Gradient simple gradient of a vector field is not a tensor consider the simple case of a constantTo see vector more field clearlyA(xµ why) in athe flat simple two-dimensional gradient of a space. vector In field flat is space not a at tensor least, consider we can define the simple “constant” case of a in a coordinate-independentconstant vector field A(x wayµ) in by a saying flat two-dimensional that at all points, space. the In vector flat space points at least, in the we same can define direction “constant” and has thein samea coordinate-independent magnitude. Therefore way the by physical saying that gradient at all of points, such a the field vector should points be zero in the in same any coordinate direction and systemhas (because the same the magnitude. vector does Therefore not change the as physical we change gradient positions). of such In a a field cartesian should coordinate be zero in system, any coordinate the µ µ componentssystemA (becauseof such the a field vector are does constant, not change so @↵A as= we 0 change as expected. positions). But in In a a curvilinear cartesian coordinate coordinate system, system, the µ µ the componentscomponents ofA evenof such a truly a field constant are constant, vector field so @↵ mayA =not 0 asbe expected. constant, But because in a curvilinear the basis vectors coordinate used system, to µ definethe the components components of change even a astruly one constant goes from vector point field to may point.not Webe need constant, to find because a way the to basis correct vectors@↵A usedin to µ such adefine coordinate the components system so changeas to remove as one the goes part from due point to changes to point. in the We basis need vectors to find a if way we hope to correct to find@↵ theA in true changesuch a incoordinate the vector system function so asA to(xµ remove). the part due to changes in the basis vectors if we hope to find the µ ⌫ Tensortrue change Gradient in the of vector a Vector. functionLetA us(x define). a set of coe cients ⌫↵ (which physicists call Christo↵el ⌫ symbolsTensor) at a given Gradient point or of event a Vector.suchLet that us define a set of coe cients ⌫↵ (which physicists call Christo↵el symbols) at a given point or eventP such that P @e↵ ⌫ µ @e µ↵e⌫ (2.17) @x ⌘ ↵ ⌫ e (2.17) @xµ ⌘ µ↵ ⌫ The numerator here is the di↵erential change in the basis vector e↵ as we move from point to a point The numerator here is theµ di↵erential change in the basis vector e as we move from pointP to a point adi↵erential displacement dx along a curve where the other coordinates↵ are constant, dividedP by that adi↵erential displacementµ dxµ along a curve where the other coordinates are constant, divided by that di↵erential displacement dx (where ↵ and µ have some specific values here). Since the change @e↵ is a µ vectordi (an↵erential arrow displacement with a certaindx magnitude(where ↵ pointingand µ have in a some certain specific direction), values we here). can write Since that the change change@ ase↵ ais a vector (an arrow with a certain magnitude pointing in a certain direction), we can write that change as a sum over the basis vectors e⌫ evaluated at point : this is the point of the sum over ⌫. In a two-dimensional sum over the basis vectors e evaluated at pointP : this is the point of the sum over ⌫. In a two-dimensional space, there are 8 such coe cients;⌫ in four-dimensionalP spacetime, there are 64 such coe cients. Nowspace, consider there are a vector 8 such field coe Acients;that is in a four-dimensional function of position. spacetime, The there product are 64 rule such implies coe cients. that the true amount dNowA that considerA changes a vector when field we moveA that an is arbitrary a function infinitesimal of position. displacement The productds (whose rule implies components that the aretrue someamount specificd setA that of componentsA changes whendx↵)is we move an arbitrary infinitesimal displacement ds (whose components are some specific set of components dx↵)is µ @A @eµ dA = d(Aµe )= dx@A µ e + Aµ dx@e↵ (2.18) dA = d(µAµe )=dx dxµ e +@Ax↵µ µ dx↵ (2.18) µ✓ dx ◆ µ @x↵ ✓ ◆ One can use equation 2.17 and rename some indices to rewrite this as One can use equation 2.17 and rename some indices to rewrite this as @Aµ @Aµµ ⌫ ↵ µ ↵ dA = ↵ + ↵⌫ A µ eµ⌫dx (↵ ↵A )eµµdx ↵ (2.19) dA@=x + A eµ⌘dx r ( ↵A )eµdx (2.19) @x↵ ↵⌫ ⌘ r The quantities ( Aµ)dx↵ for di↵erent values of µ are by definition components (in whatever coordinate The quantities↵ ( Aµ)dx↵ for di↵erent values of µ are by definition components (in whatever coordinate system we are using)r of↵ the true change in A for that di↵erential displacement. Therefore, we define system we are using)r of the true change in A for that di↵erential displacement. Therefore, we define µ µ @A µ µ ⌫ A µ @+A A µ ⌫ (2.20) ↵ A ↵ ↵⌫+ A (2.20) r r↵⌘ @x⌘ @x↵ ↵⌫ to be the tensor gradient of the vector field A: theA term involving the Christo↵el symbols corrects the to be the tensor gradient of the vector field : the term involving the Christo↵el symbolsµ corrects the partialpartial derivative derivative of the of field the forfield the for variations the variations in the in basis the basis vectors. vectors. The The tensor tensor gradient gradient↵A Amustµ must be a be a µ ↵ r ↵ (second-rank) tensor, because we see that equation 2.19 implies that ( ↵A )dxµ yields↵ the componentsr of a (second-rank) tensor, because we see that equation 2.19 implies thatr ( ↵A )dx yields the components of a vector.vector. However, However, neither neitherTensor the partial the partial derivative derivativegradient alone alone nor the nor Christo the Christoof↵el symbol↵rael symbol covector alone alone are tensors:are tensors: only only the the combinationcombination given given by equation by equation 2.20 is 2.20 a tensor. is a tensor. The TensorThe Tensor Gradient Gradient of a Covector. of a Covector.To determineTo determine a covector’s a covector’s tensor tensor gradient, gradient, we can we can take take advantage advantage of the fact that AµB isµ a scalar, and the tensor gradient of a scalar is the same as its ordinary gradient: of the fact thatµA TensorBµ is a scalar gradient, and the of tensora scalar gradient = ordinary of a scalar gradient, is the same so as its ordinary gradient: µ µ µ @A @A µ @Bµ @Bµ @↵(A@B(µA)=µB )=Bµ +BA + Aµ ↵ µ@x↵ @x↵ µ @x↵ @x↵ µ µ µ = ↵=(A B(µA)=(µB )=(↵A )BAµµ+)BA+( A↵µB( µ)B ) (2.21) (2.21) r r↵ µ r r↵ µ r r↵ µ µ µ μ SubtractingSubtracting one side one from sideSubtracting from the other the other and and substituting and substituting substituting what what we yields, know we know for↵A arbitrary↵toA beto yields be yields r r A µ µ @A @A µ @Bµµ @Bµ µ µ µ µ 0= 0=Bµ +BA + A ( ↵A( )BAµ )BA ( A↵B( µ) B ) @x↵ @x↵ µ @x↵ @ x↵r r↵ µ r r↵ µ µ ⇢ µ ⇢ µ ⇢µ ⇢ @A ⇢@A @Bµ @B@A ⇢@A = B +⇢Aµ µ µ B ⇢( µ A⌫µ)B ⌫ Aµ( µB ) where in the last step,⇢=↵ I exchangedµ⇢↵ Bµ + A↵ the µ,↵ ⇢⌫↵indexµ⇢↵ B namesµ ↵⌫( ↵⌫ inA theµ )B thirdµ A term↵( µ↵ ofBµ the) line above so that I could ⇢@x ⇢@x @x @ x⇢@x ⇢@x r r pull out the common factor of Aµ. Since this must be true for arbitrary four-vectors A, the quantity in µ @Bµµ @Bµ⌫ ⌫ = A = A B ⌫ B ↵Bµ B (2.22)(2.22) brackets must be zero for@x all↵ @ indexx↵↵ µ values↵ rµ ⌫µ r, implying↵ µ that @B B = µ ⌫ B (2.23) So we must have: r↵6 µ 6 @x↵ ↵µ ⌫ µ⌫ The Tensor Gradient of a Tensor. The generalization to arbitrary tensors (e.g., T ) is not hard: @T µ⌫ T µ⌫ = + µ T ⌫ + µ T µ + T µ⌫ (2.24) r↵ @x↵ ↵ ↵ ↵ The general rule is that the tensor gradient of a tensor is the sum of its ordinary gradient + plus a positive Christo↵el symbol (times the tensor) for each upper index (summing over that index and the second lower Christo↵el symbol index) and a negative Christo↵el symbol term (times the tensor) for each lower index (summing over that index and the upper Christo↵el symbol index. Calculating Christo↵el Symbols. We will assume in what follows that the Christo↵el symbols are symmetric in their lower indices (which amounts to assuming that spacetime is free of “torsion”1):
↵ ↵ µ⌫ = ⌫µ (2.25)
Given this assumption, one can calculate Christo↵el symbols in terms of the metric tensor as follows (the proof is one of your homework problems):