Leibniz notation for vector fields

Consider a vector fieldMathworldPlanetmath V, such as

V(x,y)=(y,-x)  for (x,y)2.

Oftentimes V will be expressed in the following Leibnizian notation


which is extremely strange the first time one encounters it. After all, why not just write


where e1=(1,0), e2=(0,1)? What do partial derivativesMathworldPlanetmath have to do with anything? (That was the question I asked.)

However, if viewed in the right way, it is not so strange. For if a vector field V is given, then one natural thing that can be done with it is to differentiate a scalar-valued function f in the direction of V. In functional notation, this would be


where p is a point, V(p) is gives the particular vector in the vector field at the point p, and Df(p)v denotes the directional derivativeMathworldPlanetmathPlanetmath of f at p with respect to the direction v.

For our example, Df(p)V(p) (where p=(x,y)) equals

D1f(x,y)y+D2f(x,y)(-x) =fxy-fyx

We have written out the steps explicitly to show where the partial derivative notation for V comes from. At the second last step we considered /x and /y as operators acting on the function f.

The Leibniz notation on manifolds

But there is more. We can consider a more general situation, where V is a vector field on a manifold. In this case, because the tangent spaceMathworldPlanetmathPlanetmath to a manifold varies with each point p, we cannot fix certain basis vectors ei to describe our vector field anymore. Loosely speaking, the basis vectors now have to vary smoothly.

Suppose we have a coordinate systemMathworldPlanetmath {xi}i=1,,n on the manifold. Then the tangent vector on the manifold corresponding to an infinitesimalMathworldPlanetmathPlanetmath change in xi is often written


This makes total sense, because one of the favorite ways to define tangent vectors on abstract manifolds is to identify them with directional derivatives. The basis /xi (which varies smoothly with p) then becomes the replacement for the fixed basis ei in Euclidean spaceMathworldPlanetmath. Note that this is consistentPlanetmathPlanetmath with the calculus notation f/xi for the partial derivative of f with respect to the xi variable, because partial derivatives are merely directional derivatives with respect to the direction ei.

If {yj} is another coordinate system for the manifold, then we have the formulaMathworldPlanetmathPlanetmath (for the derivationMathworldPlanetmath, see the entry on vector fields (http://planetmath.org/VectorField))

yj=xiyjxi, (1)

We have used the Einstein summation convention above to emphasize the mnemonic cancelling of fractions.

The quantity xi/yj is the directional derivative in the direction /yj of the ith coordinateMathworldPlanetmathPlanetmath function, mapping a point p on the manifold to the coordinate xi. Notice the subtle subtle change of viewpoint here: we are not considering xi as mere “variables”, but as functions of the point p.

Formula (1), which is a linear combinationMathworldPlanetmath of vectors in a tangent space, of course looks like the chain ruleMathworldPlanetmath learned in elementary multivariate calculus, but it is much more than that: we are saying that the formula holds for general curvilinear coordinate systems on manifolds! This is definitely one of the virtues of the Leibnizian notation — making advanced concepts look simple.

As a simple example to get used to this notation, consider the function f:3 defined by f(x,y,z)=x2+y2+z2. The Euclidean space 3 can also be thought of as a manifold, and suppose we use a spherical coordinate system (r,θ,ϕ) on it.

Let us compute f/r. If f is “viewed as a function of (r,θ,ϕ)”, then f(r,θ,ϕ)=r2, so we certainly hope that f/r=2r with our definition of directional derivatives. This easily follows from the formula for differential formsMathworldPlanetmath on manifolds:


But let us see this by calculating from (1) too. We have




and substituting in (1),

fr =xrfx+yrfy+zrfz

Needless to say, this calculation can be done with the usual functional notation, but it will be somewhat clumsy. We have to say: let α be the spherical coordinateMathworldPlanetmath chart; and then D1(fα)(r,θ,ϕ)=Df(α(r,θ,ϕ))D1α(r,θ,ϕ) would be the quantity f/r. That is not to say Leibnizian notation has no disadvantages. For example, one of the typical objections to the Leibnizian formula


is that the function f means something different on the two sides of the equation. However, formula (1) partly escapes this objection: we can consider f to be a function on a manifold, ignoring the vector spaceMathworldPlanetmath structureMathworldPlanetmath of 3. Cartesian coordinatesMathworldPlanetmath (x,y,z) simply become another coordinate chart. Spherical coordinates constitute another. Then f/r (i.e. the directional derivative /r applied to f) is a natural quantity to consider, rather than “the derivative of fα’ with respect to the first variable” (which is what the functional notation says).

Physicists seem to grasp the Leibnizian formalism very readily, and it is a shame that many calculus textbooks do not fully explain the logic behind the formalism — probably because it looks to be unrigorous — but the point we are trying to drive here is that when differentialsMathworldPlanetmath are suitably interpreted, they are rigorous.

The dual to the tangent vectors /xi

There is a subtle ambiguity in the Leibniz notation that we should also discuss here. Suppose we are on a two-dimensional manifold with coordinates u and v. The notation f/u is ambiguous, because it implicitly depends on the v coordinate as well. What we really mean when we refer to f/u is a displacement where u changes at a uniform rate of 1, and where v does not change at all.

Say, for some bizarre reason, I decided to use a coordinate system on the Euclidean plane made up of the EuclideanPlanetmathPlanetmath x coordinate and the radial r coordinate. Now when I write f/x, I mean something quite different than when I write f/x relative to the Euclidean coordinates. In the first instance the derivative is with respect to the vector field


In the second instance, the derivative is with respect to the vector field


On closer thought, we can see that /x in the elementary calculus interpretationMathworldPlanetmath has the same ambiguity, but the problem is so trivial that we often forget that it exists. For instance, if we have a function f=xyz, and we stipulate that also y=x2, then obviously f/xyz, because y is changing at the same time as x. The definition of a partial derivative with respect to x is the derivative when x changes and all the other variables are held fixed. So this rule should be applied when working with the tangent vectors /xi on a manifold too.

If we agree to use different letters for each coordinate system, and not mix them up (always a reasonable thing to do), then we will not make any mistakes arising from this ambiguity with the Leibniz notation.

Another way to understand the ambiguity is as follows. In a vector space, there is no natural isomorphism between it and its dual spaceMathworldPlanetmathPlanetmathPlanetmath (unless we involve the inner productMathworldPlanetmath or something like that). On manifolds, the role of the dual space is taken by the space of differential one-forms. (See differential forms (http://planetmath.org/DifferentialForms) for the rigorous details.) A basis for this dual space is dxj for j=1,,n. There is a basis in the tangent space that is dual to dxi: namely, these are the /xi:

dxj(xi)=δij(Kronecker delta). (2)

But if we are given a lone element dxj, we cannot produce a unique vector /xj from it (i.e. there is no isomorphismPlanetmathPlanetmathPlanetmathPlanetmath) — we need to be given the entire basis {dxj}j=1,,n. So it is not surprising why the vectors /xi should depend on each other.

Motivation for the notation of differential forms

Incidentally, the formula (2) explains the following identity involving differential forms (often seen in calculus textbooks with hardly any explanation of what it means):

df=fxjdxj. (3)

The various d’s floating around obscures the essential idea somewhat, but the derivation of this formula is basic linear algebra. Since df is a linear functionalMathworldPlanetmathPlanetmath on the tangent space, defined by df(V)=V[f], it can be written as a linear combination of the dual basis dxj. That is, for some aj, we have


And these aj are solved for by evaluating at the tangent vectors V=/xi:


giving formula (3).

For those who are not familiar with the languagePlanetmathPlanetmath of differential forms, the definition df(V)=V[f] just used might seem to be somewhat artificial, designed solely to make the classical formula (3) work out. The following comment by Spivak[2] might help clarify matters:

Classical differential geometers (and classical analysts) did not hesitate to talk about “infinitely small” changes dxi of the coordinates xi, just as Leibniz had. No one wanted to admit that this was nonsense, because true results were obtained when these infinitely small quantities were divided into each other (provided one did it in the right way). Eventually it was realized that the closest one can come to describing an infinitely small change is to describe a direction in which this change is supposed to occur, i.e. a tangent vector. Since df is supposed to be the infinitesimal change of f under an infinitesimal change of the point, df must be a function of this change, which means that df should be a function on tangent vectors. The dxi themselves then metamorphosed into functions, and it became clear that they must be distinguished from the tangent vectors /xi. Once this realization came, it was only a matter of making new definitions, which preserved the old notation, and waiting for everybody to catch up. In short, all classical notions involving infinitely small quantities became functions on tangent vectors, like df, except for quotients of infinitely small quantities, which became tangent vectors, like dc/dt.

We can also give an analogyMathworldPlanetmath as follows (also from [2]). Suppose f is a function on a manifold. Let “xi=xi(t)” be a curve on this manifold (this is the classical notation). Then by the chain rule,


where f on the left side really means f(xi(t)). The formal identity obtained by multiplying both sides by dt,


means that “true results are obtained by dividing by dt again, no matter what the functions xi(t) are.” Also, the left- and right-hand sides individually do not depend on any particular curve xi(t) at all, but really just the tangent vectors dxi/dt to that curve. Again, this leads us to realization that df and dxi should be treated as functions of a tangent vector.


Title Leibniz notation for vector fields
Canonical name LeibnizNotationForVectorFields
Date of creation 2013-03-22 15:26:54
Last modified on 2013-03-22 15:26:54
Owner stevecheng (10074)
Last modified by stevecheng (10074)
Numerical id 13
Author stevecheng (10074)
Entry type Topic
Classification msc 58A10
Classification msc 34-00
Classification msc 53-00
Related topic VectorField
Related topic DerivativeNotation
Related topic DifferentialForms
Related topic FirstOrderOperatorsInRiemannianGeometry
Related topic LieDerivative