Coordinate Space Analysis

Prior to introducing coordinate systems, our main focus had been on the construction of a geometry-centric approach to Euclidean spaces. Our efforts had yielded valuable insights, but most of our results were not well-suited for practical calculations precisely due to their geometric nature. In fact, our experience could have led us to believe that geometric insight and practical calculations are, to a certain extent, mutually exclusive.
Fortunately, that is not the case. It is the very point of Tensor Calculus to provide an effective coordinate-based analytical framework that enables practical calculations while preserving geometric insight. In this Chapter, we will begin to describe that framework and, to this end, we will begin to shift our focus from the Euclidean space to the associated coordinate space. In the coordinate space, all vectors are represented by their components and all other geometric operations (such as the dot product) are represented by systems with scalar elements (such as the metric tensor). All of the analysis is performed strictly in terms of those quantities. The advantage, of course, is that, being numerical quantities, such objects are subject to the robust techniques of Algebra, ordinary Calculus, and computational methods. At the same time, Tensor Calculus will enable us to continue to think geometrically while proceeding analytically.
By the components of a vector we, of course, mean its components with respect to the covariant basis Zi\mathbf{Z}_{i}. Recall, however, that for a general curvilinear coordinate system, the covariant basis varies from one point to another. In other words, at each point, we find a unique component space that is different from its neighbors. Thus, the term coordinate space refers to the collection of the component spaces at all points.
The covariant basis Zi\mathbf{Z}_{i} and its companion contravariant basis Zi\mathbf{Z}^{i} are equally well-suited for the decomposition of vectors at a given point. Let us first consider the components of vectors with respect to the covariant basis Zi\mathbf{Z}_{i}. These are known as the contravariant components for reasons that will become obvious shortly.

10.1.1The contravariant components

Denote by U1U^{1}, U2U^{2}, and U3U^{3}, or, collectively, UiU^{i}, the components of a vector U\mathbf{U} with respect to the covariant basis Zi\mathbf{Z}_{i}, i.e.
U=U1Z1+U2Z2+U3Z3.(10.1)\mathbf{U}=U^{1}\mathbf{Z}_{1}+U^{2}\mathbf{Z}_{2}+U^{3}\mathbf{Z}_{3}.\tag{10.1}
With the help of the summation convention, we write
U=UiZi.(10.2)\mathbf{U}=U^{i}\mathbf{Z}_{i}.\tag{10.2}
The summation convention compelled us to use a superscript to enumerate the components and therefore to use the term contravariant. As always, this choice will prove to be an accurate predictor of the manner in which UiU^{i} transforms under a coordinates transformation.
Since the covariant basis Zi\mathbf{Z}_{i} varies from one point to another, one and the same vector decomposed with respect to bases at different points in the Euclidean space will yield different contravariant components UiU^{i}. Later on, we will provide an analytical method for determining whether two sets of components at two different points in space represent equal vectors.
With the introduction of contravariant coordinates, note that the almost trivial identity
Zi=δijZj(10.3)\mathbf{Z}_{i}=\delta_{i}^{j}\mathbf{Z}_{j}\tag{10.3}
can now be interpreted in two different ways. On the one hand, it can be seen merely as an instance of the index-renaming property of the Kronecker delta in a contraction. Alternatively, the expression on the right can be seen as a linear combination for Zi\mathbf{Z}_{i} in terms of the basis vectors Z1\mathbf{Z}_{1}, Z2\mathbf{Z}_{2}, and Z3\mathbf{Z}_{3}. Thus, the elements of the Kronecker delta can also be interpreted as the collection of the contravariant components of the elements of a basis with respect to itself.

10.1.2The covariant components

When a vector U\mathbf{U} is decomposed with respect to the contravariant basis Zi\mathbf{Z}^{i}, i.e.
U=UiZi,(10.4)\mathbf{U}=U_{i}\mathbf{Z}^{i},\tag{10.4}
the resulting coefficients UiU_{i} are known as the covariant components.
Naturally, the contravariant components UiU^{i} and the covariant components UiU_{i} are closely related and one nearly can guess the relationship between them simply from the placement of the indices. Recall that the covariant and the contravariant bases are related by the identity
Zj=ZjiZi.(9.90)\mathbf{Z}_{j}=Z_{ji}\mathbf{Z}^{i}. \tag{9.90}
Substituting this relationship into the expansion U=UjZj\mathbf{U}=U^{j} \mathbf{Z}_{j}, we find
U=UjZjiZi.(10.5)\mathbf{U}=U^{j}Z_{ji}\mathbf{Z}^{i}.\tag{10.5}
Thus, since U\mathbf{U} is also given by U=UiZi\mathbf{U}=U_{i}\mathbf{Z}^{i}, we have
UjZjiZi=UiZi.(10.6)U^{j}Z_{ji}\mathbf{Z}^{i}=U_{i}\mathbf{Z}^{i}.\tag{10.6}
Equating the coefficients, we arrive at the relationship between UiU_{i} and UiU^{i}, i.e.
Ui=UjZji.(10.7)U_{i}=U^{j}Z_{ji}.\tag{10.7}
Since the metric tensor is symmetric, i.e. Zij=ZjiZ_{ij}=Z_{ji}, we can write this identity in the more pleasing way
Ui=ZijUj.(10.8)U_{i}=Z_{ij}U^{j}.\tag{10.8}
Deriving the inverse relationship
Ui=ZijUj(10.9)U^{i}=Z^{ij}U_{j}\tag{10.9}
in similar fashion is left as an exercise. Alternatively, this relationship can be derived from
Ui=ZijUj(10.8)U_{i}=Z_{ij}U^{j} \tag{10.8}
by inverting the metric tensor, as described in Section 9.5.2.
Finally, we note that the relationship between the contravariant and covariant components of a vector, captured by the identities
Ui=ZijUj(10.8)U_{i}=Z_{ij}U^{j} \tag{10.8}
and
Ui=ZijUj,(10.9)U^{i}=Z^{ij}U_{j}, \tag{10.9}
is exactly the same as the relationship between the covariant and contravariant bases. These relationships are examples of index juggling which is the subject of Chapter 11. Note that objects related by index juggling are, in fact, so closely related that we will tend to think of them as two different manifestations of the same object.

10.1.3Various methods for calculating the components

The equations
U=UiZi(10.2)\mathbf{U}=U^{i}\mathbf{Z}_{i} \tag{10.2}
and
U=UiZi(10.4)\mathbf{U}=U_{i}\mathbf{Z}^{i} \tag{10.4}
define the components but do not indicate how to calculate them. Depending on the situation, a practical calculation can be accomplished in a number of ways. In a pure geometric context illustrated in the two-dimensional figure below, one can construct a parallelogram with sides parallel to Z1\mathbf{Z}_{1} and Z2\mathbf{Z}_{2} and a vertex at the tip of U\mathbf{U}. Then the signed ratios of the lengths of the sides of the parallelogram to the lengths of the corresponding covariant basis vectors Zi\mathbf{Z}_{i} are the contravariant components UiU^{i}.
(10.10)
An alternative approach is based on decomposition by the dot product. First discussed in Section 2.4, it become our most frequently used approach thanks to its algebraic nature. We will revisit this method later in this Chapter where it will find a particularly elegant expression in the tensor notation.
Note that the two methods have a crucial feature in common: each works in any coordinate systems and each can be described exclusively in terms of quantities available in the context of the chosen coordinate system. This shared aspect of the two approaches is essential to the concept of a variant to which we now turn.
The covariant and contravariant bases Zi\mathbf{Z}_{i} and Zi\mathbf{Z}^{i}, the contravariant and covariant components UiU^{i} and UiU_{i}, and all other objects introduced in the previous Chapter are examples of variants. We will now explain the meaning of this important term.
The definition of Zi\mathbf{Z}_{i} reads
Zi=R(Z)Zi.(9.9)\mathbf{Z}_{i}=\frac{\partial\mathbf{R}\left( Z\right) }{\partial Z^{i}}. \tag{9.9}
The very fact that it contains an explicit reference to the coordinates ZiZ^{i} points to its dependence on the choice of the coordinate system. Indeed, this is so: the covariant basis, at a fixed point in space, varies from one coordinate system to another. Objects that have this property are called variants, subject to an additional requirement described in the next paragraph. Clearly, any combination of variants -- be it a sum, a product, or the result of differentiation -- is a variant in its own right. Note, however, that the term variant refers to the potential to vary from one system to another and is not a guarantee of variability. Some special combinations of variants, such as UiZiU^{i}\mathbf{Z}_{i}, have the same value in all coordinate systems. Such variants, which are the central objects of our study, are known as invariants and are described in the next Section.
Importantly, the concept of a variant carries an additional requirement: in order to be considered a variant an object must be constructed by the same algorithm in all coordinate systems. This definition is admittedly rather informal and we will therefore attempt to clarify it by illustrating why each of the objects introduced in the previous Chapter are indeed variants.
Let us start with the covariant basis. To confirm that it is a variant, let us translate its analytical definition
Zi=R(Z)Zi.(9.9)\mathbf{Z}_{i}=\frac{\partial\mathbf{R}\left( Z\right) }{\partial Z^{i}}. \tag{9.9}
into words, i.e. in order to construct the covariant basis ZiZ_{i} , differentiate the position vector R\mathbf{R} with respect to the coordinate ZiZ^{i}. This algorithm is the same in all coordinates and therefore Zi\mathbf{Z}_{i} is a variant. For the covariant metric tensor ZijZ_{ij}, note that it is constructed in two steps: 1) construct the covariant basis Zi\mathbf{Z}_{i} and 2) calculate the pairwise dot products ZiZj\mathbf{Z}_{i}\cdot \mathbf{Z}_{j}. Thus, ZijZ_{ij} is a variant. The contravariant metric tensor ZijZ^{ij} includes the additional step 3) find the matrix inverse of ZijZ_{ij}. Thus, it is also a variant. Similarly, the contravariant basis Zi\mathbf{Z}^{i} needs a further step 4) evaluate the contraction ZijZjZ^{ij}\mathbf{Z}j. Finally, the contravariant components UiU^{i} of a vector U\mathbf{U} require two steps: 1) construct the covariant basis Zi\mathbf{Z}_{i} and 2) decompose UU with respect to Zi\mathbf{Z}_{i}.
The dependence of a variant on the coordinate system may be likened to the dependence of length on the units of measurement. For example, the length of a segment may be described as 25.425.4 centimeters in the metric system or as 1010 inches in the imperial system. Meanwhile, the length of a segment is, of course, a valid concept that exists in the absence of any system of measurements. Therefore, a specific value of length in a particular measurement system can be thought of as a manifestation of length in that system. Similarly, a variant may be thought of as an entity that exists outside of any coordinate system, while the specific set of values that represent it in a particular coordinate system may be thought of as its particular coordinate manifestation. Thus, we can identify a variant with the algorithm that produces it, such as differentiation of the position vector with respect to coordinates for the covariant basis.
A variant, having different values in different coordinate systems, is said to transform from one coordinate system to another. Of great interest to us will be the transformation rule, i.e. the equation that relates the values of a variant in the different coordinate systems. We will discover that some variants, which we will call tensors, are subject to a particularly transformation rules. As you may expect, these special variants will play an instrumental role in our subject.
Finally, note that at this point, we would be hard-pressed to think of an object that is not a variant. We will encounter our first non-variant in Chapter 13 when we introduce the Jacobian which requires the presence of two coordinate systems.
The covariant basis Zi\mathbf{Z}_{i} and the contravariant components UiU^{i} of a vector U\mathbf{U} are variants. Thus, by definition, the contraction
UiZi,(10.11)U^{i}\mathbf{Z}_{i},\tag{10.11}
is also a variant. However, it is a very special kind of variant: it has the same value, namely the vector U\mathbf{U}, in all coordinate systems. Such a variant is called an invariant.
Invariants, by virtue of their independence from the choice of coordinates, are the ultimate objects of our study. Above all, the purpose of Tensor Calculus is to study the natural world (or, at least, a geometric idealization of it). Thus, every meaningful analysis must ultimately yield objects that are independent of coordinates.
Of course, the statement that the combination UiZiU^{i}\mathbf{Z}_{i} is independent of coordinates is nearly tautological since the components UiU^{i} are constructed precisely in such a way as to produce U\mathbf{U} by the contraction UiZiU^{i}\mathbf{Z}_{i}. Going forward, we will encounter invariants that arise in far more nontrivial ways. Nevertheless, the key to invariance will be exceptionally simple: the expression must be comprised of tensors and have all indices contracted away.
The combination UiZiU^{i}\mathbf{Z}_{i} also offers an insight into the eventual meaning of the terms covariant and contravariant as applied to tensors, i.e. variants subject to a special transformation rule -- or, to be more precise, one of two rules that are the inverses of each other. One of the rules is called covariant and the other -- contravariant . Covariant tensors transform in the same way as the basis Zi\mathbf{Z}_{i} -- thus co. Contravariant tensors transform in the opposite way -- thus contra. When two objects that transform according to opposite rules are combined, the two transformations cancel each other and, as a result, the value of the combination in one coordinate system equals its value in the other coordinate system. In other words, the combination is an invariant. This, in a nutshell, is how invariants are produced in Tensor Calculus.
In Section 2.4, we considered the decomposition of a vector U\mathbf{U} with respect to a basis b1,b2,b3\mathbf{b}_{1},\mathbf{b}_{2} ,\mathbf{b}_{3}. Recall that the components of U\mathbf{U}, which we at the time denoted by UiU_{i}, were given by the equations
Ui=biU(2.48)U_{i}=\mathbf{b}_{i}\cdot\mathbf{U} \tag{2.48}
for an orthonormal, a.k.a. Cartesian, basis bi\mathbf{b}_{i} ,
Ui=biUbibi(2.46)U_{i}=\frac{\mathbf{b}_{i}\cdot\mathbf{U}}{\mathbf{b}_{i}\cdot\mathbf{b}_{i}} \tag{2.46}
for an orthogonal basis bi\mathbf{b}_{i}, and
[U1U2U3]=[b1b1b1b2b1b3b2b1b2b2b2b3b3b1b3b2b3b3]1[b1Ub2Ub3U](2.54)\left[ \begin{array} {c} U_{1}\\ U_{2}\\ U_{3} \end{array} \right] =\left[ \begin{array} {ccc} \mathbf{b}_{1}\cdot\mathbf{b}_{1} & \mathbf{b}_{1}\cdot\mathbf{b}_{2} & \mathbf{b}_{1}\cdot\mathbf{b}_{3}\\ \mathbf{b}_{2}\cdot\mathbf{b}_{1} & \mathbf{b}_{2}\cdot\mathbf{b}_{2} & \mathbf{b}_{2}\cdot\mathbf{b}_{3}\\ \mathbf{b}_{3}\cdot\mathbf{b}_{1} & \mathbf{b}_{3}\cdot\mathbf{b}_{2} & \mathbf{b}_{3}\cdot\mathbf{b}_{3} \end{array} \right] ^{-1}\left[ \begin{array} {c} \mathbf{b}_{1}\cdot\mathbf{U}\\ \mathbf{b}_{2}\cdot\mathbf{U}\\ \mathbf{b}_{3}\cdot\mathbf{U} \end{array} \right] \tag{2.54}
for an arbitrary basis bi\mathbf{b}_{i}.
We will now switch to the covariant basis Zi\mathbf{Z}_{i} and the contravariant components UiU^{i}. Note that replace bi\mathbf{b}_{i} with Zi\mathbf{Z}_{i} and UiU_{i} with UiU^{i} in the above equations yields identities that are invalid from the tensor notation point of view, i.e.
          (10.12)\begin{aligned}\ \ \ \ \ \ \ \ \ \ \left(10.12\right)\end{aligned}
U^{i} & =mathbf{Z}_{i}cdotmathbf{U,}tag{-} U^{i} & =frac{mathbf{Z}_{i}cdotmathbf{U}}{mathbf{Z}_{i}cdot mathbf{Z}_{i}},text{ and }tag{-} left[ begin{array} {c} U^{1} U^{2} U^{3} end{array} right] & =left[ begin{array} {ccc} mathbf{Z}_{1}cdotmathbf{Z}_{1} & mathbf{Z}_{1}cdotmathbf{Z}_{2} & mathbf{Z}_{1}cdotmathbf{Z}_{3} mathbf{Z}_{2}cdotmathbf{Z}_{1} & mathbf{Z}_{2}cdotmathbf{Z}_{2} & mathbf{Z}_{2}cdotmathbf{Z}_{3} mathbf{Z}_{3}cdotmathbf{Z}_{1} & mathbf{Z}_{3}cdotmathbf{Z}_{2} & mathbf{Z}_{3}cdotmathbf{Z}_{3} end{array} right] ^{-1}left[ begin{array} {c} mathbf{Z}_{1}cdotmathbf{U} mathbf{Z}_{2}cdotmathbf{U} mathbf{Z}_{3}cdotmathbf{U} end{array} right] .tag{-} end{align} Recall from Section 9.5.4, that the 3×33\times3 matrix above corresponds to the covariant metric tensor ZijZ_{ij}. Therefore, its inverse corresponds to the contravariant metric tensor ZijZ^{ij}.
Despite its invalid form, the equation
(10.13)\tag{10.13}
U^{i}=mathbf{Z}_{i}cdotmathbf{U} tag{-} end{equation} is correct for an orthonormal basis Zi\mathbf{Z}_{i}, although it is, of course, limited to that special case. This is observed for most equations that violate the rules of the tensor notation: they are typically correct only in a narrow range of special cases.
Crucially, the notational flaw in the above equation is easily fixed by replacing the covariant basis with the contravariant basis, i.e.
Ui=ZiU.(10.14)U^{i}=\mathbf{Z}^{i}\cdot\mathbf{U.}\tag{10.14}
In words: the contravariant component UiU^{i} of the vector U\mathbf{U} is given by the dot product of U\mathbf{U} with the contravariant basis vector Zi\mathbf{Z}^{i}. This equation remains valid for an orthonormal basis Zi\mathbf{Z}_{i} since, in that case, Zi\mathbf{Z}^{i} coincides with Zi\mathbf{Z}_{i}. However, somewhat remarkably given its utmost simplicity, this equation is valid for an arbitrary basis Zi\mathbf{Z}_{i}. In other words, it is valid in all coordinate systems.
To observe why this is so, note that the equation
Zi=ZijZj(9.89)\mathbf{Z}^{i}=Z^{ij}\mathbf{Z}_{j} \tag{9.89}
in matrix terms reads
[Z1Z2Z3]=[Z1Z1Z1Z2Z1Z3Z2Z1Z2Z2Z2Z3Z3Z1Z3Z2Z3Z3]1[Z1Z2Z3].(10.15)\left[ \begin{array} {c} \mathbf{Z}^{1}\\ \mathbf{Z}^{2}\\ \mathbf{Z}_{3} \end{array} \right] =\left[ \begin{array} {ccc} \mathbf{Z}_{1}\cdot\mathbf{Z}_{1} & \mathbf{Z}_{1}\cdot\mathbf{Z}_{2} & \mathbf{Z}_{1}\cdot\mathbf{Z}_{3}\\ \mathbf{Z}_{2}\cdot\mathbf{Z}_{1} & \mathbf{Z}_{2}\cdot\mathbf{Z}_{2} & \mathbf{Z}_{2}\cdot\mathbf{Z}_{3}\\ \mathbf{Z}_{3}\cdot\mathbf{Z}_{1} & \mathbf{Z}_{3}\cdot\mathbf{Z}_{2} & \mathbf{Z}_{3}\cdot\mathbf{Z}_{3} \end{array} \right] ^{-1}\left[ \begin{array} {c} \mathbf{Z}_{1}\\ \mathbf{Z}_{2}\\ \mathbf{Z}_{3} \end{array} \right] .\tag{10.15}
Thus, it can be seen that the equation
Ui=ZiU(10.14)U^{i}=\mathbf{Z}^{i}\cdot\mathbf{U} \tag{10.14}
is simply the tensor form of the equation
[U1U2U3]=[Z1Z1Z1Z2Z1Z3Z2Z1Z2Z2Z2Z3Z3Z1Z3Z2Z3Z3]1[Z1UZ2UZ3U](10.16)\left[ \begin{array} {c} U^{1}\\ U^{2}\\ U^{3} \end{array} \right] =\left[ \begin{array} {ccc} \mathbf{Z}_{1}\cdot\mathbf{Z}_{1} & \mathbf{Z}_{1}\cdot\mathbf{Z}_{2} & \mathbf{Z}_{1}\cdot\mathbf{Z}_{3}\\ \mathbf{Z}_{2}\cdot\mathbf{Z}_{1} & \mathbf{Z}_{2}\cdot\mathbf{Z}_{2} & \mathbf{Z}_{2}\cdot\mathbf{Z}_{3}\\ \mathbf{Z}_{3}\cdot\mathbf{Z}_{1} & \mathbf{Z}_{3}\cdot\mathbf{Z}_{2} & \mathbf{Z}_{3}\cdot\mathbf{Z}_{3} \end{array} \right] ^{-1}\left[ \begin{array} {c} \mathbf{Z}_{1}\cdot\mathbf{U}\\ \mathbf{Z}_{2}\cdot\mathbf{U}\\ \mathbf{Z}_{3}\cdot\mathbf{U} \end{array} \right]\tag{10.16}
which is valid for an arbitrary basis Zi\mathbf{Z}_{i}.
The remarkably simple equation
Ui=ZiU(10.14)U^{i}=\mathbf{Z}^{i}\cdot\mathbf{U} \tag{10.14}
can also be derived quite concisely by pure tensor means. Simply dot both sides of the equation
U=UjZj(10.2)\mathbf{U}=U^{j}\mathbf{Z}_{j} \tag{10.2}
with Zi\mathbf{Z}^{i}, i.e.
ZiU=UjZjZi.(10.17)\mathbf{Z}^{i}\cdot\mathbf{U}=U^{j}\mathbf{Z}_{j}\cdot\mathbf{Z}^{i}.\tag{10.17}
Since ZjZi=δji\mathbf{Z}_{j}\cdot\mathbf{Z}^{i}=\delta_{j}^{i} and Ujδji=UiU^{j}\delta _{j}^{i}=U^{i}, we have
ZiU=Ui(10.18)\mathbf{Z}^{i}\cdot\mathbf{U}=U^{i}\tag{10.18}
or
Ui=ZiU,(10.14)U^{i}=\mathbf{Z}^{i}\cdot\mathbf{U,} \tag{10.14}
as we set out to show.
Note that the simplicity of the equation
Ui=ZiU(10.14)U^{i}=\mathbf{Z}^{i}\cdot\mathbf{U} \tag{10.14}
does not imply that the matrix inversion inherent in linear decomposition can be circumvented. Instead, this equation hides the inversion in the contravariant basis Zi\mathbf{Z}^{i} and, in doing so, organizes the decomposition algorithm with respect to an arbitrary basis in such a way that it appears as simple as for an orthonormal basis.
It is left as an exercise to show that the covariant components UiU_{i} are given by the individual dot products with the elements Zi\mathbf{Z}_{i} of the covariant basis, i.e.
Ui=ZiU.(10.19)U_{i}=\mathbf{Z}_{i}\cdot\mathbf{U.}\tag{10.19}
The remarkable compactness of these identities is a testament to the tensor framework. While economy of notation does not imply economy of computation, the expressive power of the tensor notation is undeniable as it represents the best of two worlds: tensor expressions are as simple as those associated with Cartesian coordinates yet are valid in all coordinate systems. This pattern is found throughout Tensor Calculus, and we are about to see it again in our discussion of the coordinate space representation of the dot product.
In Section 2.6, we established the expression for the dot product of two vectors in terms of their components. Specifically -- once again switching to the symbols UiU^{i}, ViV^{i}, and Zi\mathbf{Z}_{i} -- we showed that the dot product UV\mathbf{U}\cdot\mathbf{V} is given by the matrix equation
UV=[U1U2U3][Z1Z1Z1Z2Z1Z3Z2Z1Z2Z2Z2Z3Z3Z1Z3Z2Z3Z3][V1V2V3].(2.71)\mathbf{U}\cdot\mathbf{V}= \begin{array} {c} \left[ \begin{array} {ccc} U^{1} & U^{2} & U^{3} \end{array} \right] \\ \\ \end{array} \left[ \begin{array} {ccc} \mathbf{Z}_{1}\cdot\mathbf{Z}_{1} & \mathbf{Z}_{1}\cdot\mathbf{Z}_{2} & \mathbf{Z}_{1}\cdot\mathbf{Z}_{3}\\ \mathbf{Z}_{2}\cdot\mathbf{Z}_{1} & \mathbf{Z}_{2}\cdot\mathbf{Z}_{2} & \mathbf{Z}_{2}\cdot\mathbf{Z}_{3}\\ \mathbf{Z}_{3}\cdot\mathbf{Z}_{1} & \mathbf{Z}_{3}\cdot\mathbf{Z}_{2} & \mathbf{Z}_{3}\cdot\mathbf{Z}_{3} \end{array} \right] \left[ \begin{array} {c} V^{1}\\ V^{2}\\ V^{3} \end{array} \right] . \tag{2.71}
If the 3×13\times1 matrices corresponding to UiU^{i} and ViV^{i} are denoted by UU and VV, and MM once again denotes the matrix corresponding to ZijZ_{ij}, then UV\mathbf{U}\cdot\mathbf{V} is given by
UV=UTMV.(2.72)\mathbf{U}\cdot\mathbf{V}=U^{T}MV. \tag{2.72}
If the basis is orthonormal, and therefore MM is the identity matrix, then the dot product reduces to the classical form
UV=UTV=U1V1+U2V2+U3V3.(10.20)\mathbf{U}\cdot\mathbf{V}=U^{T}V=U^{1}V^{1}+U^{2}V^{2}+U^{3}V^{3}.\tag{10.20}
You have probably already guessed that the tensor equivalent of UTMVU^{T}MV is ZijUiUjZ_{ij}U^{i}U^{j}. In fact, we have already essentially derived it in the tensor notation in Section 8.4. However, since the derivation fits on a single line, we are happy to repeat it here. Recall that
U=UiZiV=ViZi(10.21)\begin{array} {l} \mathbf{U}=U^{i}\mathbf{Z}_{i}\\ \mathbf{V}=V^{i}\mathbf{Z}_{i} \end{array}\tag{10.21}
In the second equation, rename the index ii into jj, i.e.
U=UiZiV=VjZj,(10.22)\begin{array} {l} \mathbf{U}=U^{i}\mathbf{Z}_{i}\\ \mathbf{V}=V^{j}\mathbf{Z}_{j}, \end{array}\tag{10.22}
in preparation for using the two contractions in a single expression. Now, dot both sides of the equations above, i.e.
UV=UiZiVjZj,(10.23)\mathbf{U}\cdot\mathbf{V}=U^{i}\mathbf{Z}_{i}\cdot V^{j}\mathbf{Z}_{j},\tag{10.23}
and note the chain of identities
UV=UiZiVjZj=(ZiZj)UiVj=ZijUiVj.(10.24)\mathbf{U}\cdot\mathbf{V}=U^{i}\mathbf{Z}_{i}\cdot V^{j}\mathbf{Z}_{j}=\left( \mathbf{Z}_{i}\cdot\mathbf{Z}_{j}\right) U^{i}V^{j}=Z_{ij}U^{i}V^{j}.\tag{10.24}
Thus, we have arrived at the fundamental identity
UV=ZijUiVj.(10.25)\mathbf{U}\cdot\mathbf{V}=Z_{ij}U^{i}V^{j}.\tag{10.25}
This crucial identity clarifies the meaning of the term metric in metric tensor: the metric tensor is used for evaluating dot products, and therefore lengths and angles, in the Euclidean space.
In terms of the underlying arithmetic operations, ZijUiVjZ_{ij}U^{i}V^{j} is equivalent to the matrix expression UTMVU^{T}MV and it nearly matches its compactness. In addition, it is indifferent to the order of the multiplicative terms and lacks the need for the transpose. Furthermore, the expression ZijUiVjZ_{ij}U^{i}V^{j} is subject to further "compactization". Recall that the combination ZijVjZ_{ij}V^{j} yields the covariant component ViV_{i}. Therefore, the equation
UV=ZijUiVj(10.25)\mathbf{U}\cdot\mathbf{V}=Z_{ij}U^{i}V^{j} \tag{10.25}
can be rewritten in the form
UV=UiVi,(10.26)\mathbf{U}\cdot\mathbf{V}=U^{i}V_{i},\tag{10.26}
or, by the same token, in the form
UV=UiVi.(10.27)\mathbf{U}\cdot\mathbf{V}=U_{i}V^{i}.\tag{10.27}
Note that in the unpacked form, the last equation reads
UV=U1V1+U2V2+U3V3(10.28)\mathbf{U}\cdot\mathbf{V}=U_{1}V^{1}+U_{2}V^{2}+U_{3}V^{3}\tag{10.28}
which clearly demonstrates that we have been able to achieve the simplicity of the classical Cartesian expression found in the equation
UV=U1V1+U2V2+V3U3(2.74)\mathbf{U}\cdot\mathbf{V}=U_{1}V_{1}+U_{2}V_{2}+V_{3}U_{3} \tag{2.74}
from Chapter 2. Meanwhile, the equation
UV=U1V1+U2V2+U3V3(10.28)\mathbf{U}\cdot\mathbf{V}=U_{1}V^{1}+U_{2}V^{2}+U_{3}V^{3} \tag{10.28}
is valid in all bases. Its simplicity speaks to the expressive power of the tensor framework.
Finally, note that the length of a vector U\mathbf{U} in component form is given by
len2U=ZijUiUj,(10.29)\operatorname{len}^{2}\mathbf{U}=Z_{ij}U^{i}U^{j},\tag{10.29}
which can be expressed more compactly as
len2U=UiUi.(10.30)\operatorname{len}^{2}\mathbf{U}=U_{i}U^{i}.\tag{10.30}
Although the equation
UV=UiVi(10.27)\mathbf{U}\cdot\mathbf{V}=U_{i}V^{i} \tag{10.27}
along with its unpacked form
UV=U1V1+U2V2+U3V3(10.28)\mathbf{U}\cdot\mathbf{V}=U_{1}V^{1}+U_{2}V^{2}+U_{3}V^{3} \tag{10.28}
is valid in all coordinate systems, it will be useful to document the explicit expressions in standard coordinate systems corresponding to the equation
UV=ZijUiVj,(10.25)\mathbf{U}\cdot\mathbf{V}=Z_{ij}U^{i}V^{j}, \tag{10.25}
that expresses the dot product in terms of the contravariant coordinates only.
In Cartesian coordinates, the covariant metric tensor is represented by the identity matrix and therefore
UV=[U1U2U3][111][V1V2V3].(10.31)\mathbf{U}\cdot\mathbf{V}= \begin{array} {c} \left[ \begin{array} {ccc} U^{1} & U^{2} & U^{3} \end{array} \right] \\ \\ \end{array} \left[ \begin{array} {ccc} 1 & & \\ & 1 & \\ & & 1 \end{array} \right] \left[ \begin{array} {c} V^{1}\\ V^{2}\\ V^{3} \end{array} \right] .\tag{10.31}
In other words,
UV=U1V1+U2V2+U3V3.(10.32)\mathbf{U}\cdot\mathbf{V}=U^{1}V^{1}+U^{2}V^{2}+U^{3}V^{3}.\tag{10.32}
In polar coordinates in a two-dimensional space,
Zij corresponds to [100r2](9.43)Z_{ij}\text{ corresponds to }\left[ \begin{array} {cc} 1 & 0\\ 0 & r^{2} \end{array} \right] \tag{9.43}
and, therefore, UV\mathbf{U}\cdot\mathbf{V} is given by
UV=U1V1+r2 U2V2.(10.33)\mathbf{U}\cdot\mathbf{V}=U^{1}V^{1}+r^{2}~U^{2}V^{2}.\tag{10.33}
In cylindrical coordinates,
Zij corresponds to [1000r20001](9.44)Z_{ij}\text{ corresponds to }\left[ \begin{array} {ccc} 1 & 0 & 0\\ 0 & r^{2} & 0\\ 0 & 0 & 1 \end{array} \right] \tag{9.44}
and, therefore, UV\mathbf{U}\cdot\mathbf{V} is given by
UV=U1V1+r2 U2V2+U3V3.(10.34)\mathbf{U}\cdot\mathbf{V}=U^{1}V^{1}+r^{2}~U^{2}V^{2}+U^{3}V^{3}.\tag{10.34}
Finally, in spherical coordinates,
Zij corresponds to [1000r2000r2sin2θ](9.45)Z_{ij}\text{ corresponds to }\left[ \begin{array} {ccc} 1 & 0 & 0\\ 0 & r^{2} & 0\\ 0 & 0 & r^{2}\sin^{2}\theta \end{array} \right] \tag{9.45}
and, therefore, UV\mathbf{U}\cdot\mathbf{V} is given by
UV=U1V1+r2 U2V2+r2sin2θ U3V3.(10.35)\mathbf{U}\cdot\mathbf{V}=U^{1}V^{1}+r^{2}~U^{2}V^{2}+r^{2}\sin^{2} \theta~U^{3}V^{3}.\tag{10.35}
With every geometric vector U\mathbf{U}, we associate its contravariant components UiU^{i}. This association works in both directions: from the vector, we can determine the components, and from the components, we can reconstruct the vector. Furthermore, we can naturally associate the vector U=UiZi\mathbf{U}=U^{i}\mathbf{Z}_{i} with any first-order system UiU^{i} with scalar elements, regardless of how the system may have arisen.
Thus, there is a natural one-to-one correspondence between first-order systems and geometric vectors. In fact, this correspondence is so strong that we may go as far as referring to first-order systems as vectors. That is, we may write consider a vector UiU^{i} instead of consider a vector with components UiU^{i}. In other words, in the context of Euclidean spaces, a vector and its components are so closely linked that the terms vector and components of a vector become interchangeable.
In order to demonstrate how a practical analysis may be conducted in the component space, let us analyze the motion of a material particle in a Euclidian space. Suppose that the Euclidean space is referred to an arbitrary coordinate system ZiZ^{i} and that the trajectory of the particle is given by the equations
Zi=Zi(t),(10.36)Z^{i}=Z^{i}\left( t\right) ,\tag{10.36}
where tt represents time. The functions Zi(t)Z^{i}\left( t\right) are referred to as the equations of the motion.
Let our goal is to express the components ViV^{i} of the particle's velocity vector V\mathbf{V} and the components AiA^{i} of the acceleration vector A\mathbf{A} in terms of the equations of the motion Zi(t)Z^{i}\left( t\right) . Note that at each point along the trajectory, the vectors V\mathbf{V} and A\mathbf{A} must be decomposed with respect to the covariant basis Zi\mathbf{Z}_{i} found at that point.
In this Section, we will show that the velocity components ViV^{i} are given by the intuitive equation
Vi=dZi(t)dt.(10.37)V^{i}=\frac{dZ^{i}\left( t\right) }{dt}.\tag{10.37}
Meanwhile, the equation for the acceleration components AiA^{i} will be postponed until later since it requires the introduction of the Christoffel symbol Γjki\Gamma_{jk}^{i} which is the subject of Chapter 12.
We described the equation
Vi=dZi(t)dt(10.37)V^{i}=\frac{dZ^{i}\left( t\right) }{dt} \tag{10.37}
as intuitive since it can be easily seen to be true in an affine setting. Indeed, in affine coordinates, the position vector R\mathbf{R} along the trajectory is given by the function
R(t)=Z1(t) i+Z2(t) j+Z3(t) k,(10.38)\mathbf{R}\left( t\right) =Z^{1}\left( t\right) \mathbf{\ i}+Z^{2}\left( t\right) \ \mathbf{j}+Z^{3}\left( t\right) \ \mathbf{k,}\tag{10.38}
provided that R\mathbf{R} emanates from the origin of the coordinate system. In other words, the functions Zi(t)Z^{i}\left( t\right) represent the components of R(t)\mathbf{R}\left( t\right) with respect to the constant basis i,j,k\mathbf{i,j,k}. Differentiating both sides with respect to tt, we find that
V(t)=dZ1(t)dt i+dZ2(t)dt j+dZ3(t)dt k,(10.39)\mathbf{V}\left( t\right) =\frac{dZ^{1}\left( t\right) }{dt} \mathbf{\ i}+\frac{dZ^{2}\left( t\right) }{dt}\ \mathbf{j}+\frac {dZ^{3}\left( t\right) }{dt}\ \mathbf{k,}\tag{10.39}
which leads to the conclusion that the velocity components ViV^{i} are given by
Vi=dZi(t)dt.(10.37)V^{i}=\frac{dZ^{i}\left( t\right) }{dt}. \tag{10.37}
However, it is clear that the above derivation is limited to affine coordinates since it assumes that the coordinate basis does not vary with tt. Thus, the fact that the equation
Vi=dZi(t)dt(10.37)V^{i}=\frac{dZ^{i}\left( t\right) }{dt} \tag{10.37}
remains valid in curvilinear coordinates is perhaps somewhat unexpected.
Let us now derive the equation
Vi(t)=dZi(t)dt(10.37)V^{i}\left( t\right) =\frac{dZ^{i}\left( t\right) }{dt} \tag{10.37}
in general curvilinear coordinates. Start with the definition of the velocity V\mathbf{V}, i.e.
V=dR(t)dt,(10.40)\mathbf{V=}\frac{d\mathbf{R}\left( t\right) }{dt},\tag{10.40}
where R(t)\mathbf{R}\left( t\right) represents the values of the position vector along the trajectory. The function R(t)R\left( t\right) can be formed by composing the function R(Z)\mathbf{R}\left( Z\right) , i.e. the position vector a function of the coordinate ZiZ^{i}, with the equations of the motion Zi(t)Z^{i}\left( t\right) . In other words,
R(t)=R(Z(t)).(10.41)\mathbf{R}\left( t\right) =\mathbf{R}\left( Z\left( t\right) \right) .\tag{10.41}
By the chain rule, we find
V=dR(t)dt=dR(Z(t))dt=R(Z)ZidZi(t)dt.(10.42)\mathbf{V}=\frac{d\mathbf{R}\left( t\right) }{dt}=\frac{d\mathbf{R}\left( Z\left( t\right) \right) }{dt}=\frac{\partial\mathbf{R}\left( Z\right) }{\partial Z^{i}}\frac{dZ^{i}\left( t\right) }{dt}.\tag{10.42}
Since the derivative R(Z)/Zi\partial\mathbf{R}\left( Z\right) /\partial Z^{i} yields the covariant basis Zi\mathbf{Z}_{i}, we have
V=dZi(t)dtZi.(10.43)\mathbf{V}=\frac{dZ^{i}\left( t\right) }{dt}\mathbf{Z}_{i}.\tag{10.43}
Therefore, the derivatives
dZi(t)dt(10.44)\frac{dZ^{i}\left( t\right) }{dt}\tag{10.44}
are indeed the contravariant components ViV^{i} of V\mathbf{V}, i.e.
Vi=dZi(t)dt,(10.37)V^{i}=\frac{dZ^{i}\left( t\right) }{dt}, \tag{10.37}
as we set out to show.
Two surprising aspects of this identity that are worthy of note. First is the very fact that it is true despite the spatial variability of the covariant basis. As we show below, the analogous equations for the acceleration components AiA^{i}, i.e.
(10.45)\tag{10.45}
A^{i}=frac{dV^{i}left( tright) }{dt} tag{-} end{equation} or, equivalently,
(10.46)\tag{10.46}
A^{i}=frac{d^{2}Z^{i}left( tright) }{dt^{2}}, tag{-} end{equation} do not hold -- precisely due to the variability of the basis.
The second surprising aspect of the equation
Vi=dZi(t)dt,(10.37)V^{i}=\frac{dZ^{i}\left( t\right) }{dt}, \tag{10.37}
is the fact that it does not reference either the position vector or the covariant basis or any other geometric elements of the Euclidean space. Meanwhile, one and the same equations of motion
Zi=Zi(t)(10.47)Z^{i}=Z^{i}\left( t\right)\tag{10.47}
describe completely different trajectories in different coordinate systems. For example, the equations
Z1(t)=t          (10.48)Z2(t)=t2          (10.49)\begin{aligned}Z^{1}\left( t\right) & =t\ \ \ \ \ \ \ \ \ \ \left(10.48\right)\\Z^{2}\left( t\right) & =t^{2}\ \ \ \ \ \ \ \ \ \ \left(10.49\right)\end{aligned}
describe a parabola in Cartesian coordinates and a spiral in polar coordinates. Thus, the two motions are characterized by completely different velocity vectors V(t)\mathbf{V}\left( t\right) . Nevertheless, the components ViV^{i} of the velocity, i.e.
V1(t)=1          (10.50)V2(t)=2t,          (10.51)\begin{aligned}V^{1}\left( t\right) & =1\ \ \ \ \ \ \ \ \ \ \left(10.50\right)\\V^{2}\left( t\right) & =2t,\ \ \ \ \ \ \ \ \ \ \left(10.51\right)\end{aligned}
are the same for both motions.

10.8.1Example: Uniform motion along a helix

For a specific example, consider a particle moving along a helix in a three-dimensional Euclidean space. If the Euclidean space is referred to cylindrical coordinates Z1,Z2,Z3=r,θ,zZ^{1},Z^{2},Z^{3}=r,\theta,z oriented in a natural way with respect to the helix, then the equations of the motion read
r(t)=R          (10.52)θ(t)=ωt          (10.53)z(t)=at,          (10.54)\begin{aligned}r\left( t\right) & =R\ \ \ \ \ \ \ \ \ \ \left(10.52\right)\\\theta\left( t\right) & =\omega t\ \ \ \ \ \ \ \ \ \ \left(10.53\right)\\z\left( t\right) & =at,\ \ \ \ \ \ \ \ \ \ \left(10.54\right)\end{aligned}
where RR is the radius of the helix, ω\omega is the angular velocity of the circular motion, and 2πa/ω2\pi a/\omega is the span of the helix. Therefore, the components of the velocity, given by the equation
Vi=dZi(t)dt,(10.37)V^{i}=\frac{dZ^{i}\left( t\right) }{dt}, \tag{10.37}
have the following constant values
V1=0          (10.55)V2=ω          (10.56)V3=a.          (10.57)\begin{aligned}V^{1} & =0\ \ \ \ \ \ \ \ \ \ \left(10.55\right)\\V^{2} & =\omega\ \ \ \ \ \ \ \ \ \ \left(10.56\right)\\V^{3} & =a.\ \ \ \ \ \ \ \ \ \ \left(10.57\right)\end{aligned}
Of course, even though the components of the velocity vector are constant, the velocity vector itself is variable due to the variability of the accompanying basis.
The magnitude VV of the velocity can be calculated by the formula
UV=U1V1+r2 U2V2+U3V3.(10.34)\mathbf{U}\cdot\mathbf{V}=U^{1}V^{1}+r^{2}~U^{2}V^{2}+U^{3}V^{3}. \tag{10.34}
Along the helix, this formula reads
UV=U1V1+R2 U2V2+U3V3(10.58)\mathbf{U}\cdot\mathbf{V}=U^{1}V^{1}+R^{2}~U^{2}V^{2}+U^{3}V^{3}\tag{10.58}
and therefore
V2=VV=R2ω2+a2.(10.59)V^{2}=\mathbf{V}\cdot\mathbf{V}=R^{2}\omega^{2}+a^{2}.\tag{10.59}
Thus,
V=R2ω2+a2.(10.60)V=\sqrt{R^{2}\omega^{2}+a^{2}}.\tag{10.60}
It is left as an exercise to the repeat this calculation in Cartesian coordinates.

10.8.2A note on the components of acceleration

As we have already stated, the components AiA^{i} of the acceleration vector A\mathbf{A} do not equal the time derivative of the components ViV^{i} of the velocity vector, i.e.
AidVi(t)dt.(10.61)A^{i}\neq\frac{dV^{i}\left( t\right) }{dt}.\tag{10.61}
If nothing else, this is clear from the example of uniform motion along a helix that we analyzed in the previous Section. Since the components ViV^{i} of the velocity vector are constant, their derivatives vanish, i.e.
dVi(t)dt=0.(10.62)\frac{dV^{i}\left( t\right) }{dt}=0.\tag{10.62}
Meanwhile, the acceleration of the particle moving along a helix is not zero and therefore Ai0A^{i}\neq0.
The underlying reason for the failure of the formula
Ai=dVi(t)dt(-)A^{i}=\frac{dV^{i}\left( t\right) }{dt} \tag{-}
is, of course, the spatial variability of the covariant basis. In order to establish the correct relationship, let us once again start our analysis with a geometric object: the acceleration vector A\mathbf{A}, defined as the derivative of the velocity vector V(t)\mathbf{V} \left( t\right) , i.e.
A=dV(t)dt.(10.63)\mathbf{A}=\frac{d\mathbf{V}\left( t\right) }{dt}.\tag{10.63}
Since V(t)\mathbf{V}\left( t\right) is given by the equation
V(t)=Vi(t)Zi(t),(10.64)\mathbf{V}\left( t\right) =V^{i}\left( t\right) \mathbf{Z}_{i}\left( t\right) ,\tag{10.64}
where Zi(t)\mathbf{Z}_{i}\left( t\right) is a function that represents the covariant basis Zi\mathbf{Z}_{i} along the trajectory, differentiating the above equation yields
A(t)=dVi(t)dtZi+VidZi(t)dt.(10.65)\mathbf{A}\left( t\right) =\frac{dV^{i}\left( t\right) }{dt}\mathbf{Z} _{i}+V^{i}\frac{d\mathbf{Z}_{i}\left( t\right) }{dt}.\tag{10.65}
This is the crucial moment. The second term does not vanish since the derivative
dZi(t)dt(10.66)\frac{d\mathbf{Z}_{i}\left( t\right) }{dt}\tag{10.66}
represents the (nonvanishing) rate of change of the basis vector Zi\mathbf{Z}_{i} along the trajectory. As a result,
AdVi(t)dtZi(10.67)\mathbf{A}\neq\frac{dV^{i}\left( t\right) }{dt}\mathbf{Z}_{i}\tag{10.67}
and consequently
AidVi(t)dt.(10.61)A^{i}\neq\frac{dV^{i}\left( t\right) }{dt}. \tag{10.61}
Therefore, in order to advance our analysis further, we must tackle the spatial variability of the covariant basis. This line of inquiry, which leads to the concept of the Christoffel symbol, is developed in Chapter 12.

10.8.3The components of the tangent vector to a curve

The analysis similar to that of the motion of a material particle can be applied to abstract geometric curves. Consider a curve parameterized by a generic variable γ\gamma rather than time tt. In terms of the γ\gamma, the equations of the curve read
Zi=Zi(γ).(10.68)Z^{i}=Z^{i}\left( \gamma\right) .\tag{10.68}
Based on our foregoing discussion, we can conclude that the derivative
dZi(γ)dγ(10.69)\frac{dZ^{i}\left( \gamma\right) }{d\gamma}\tag{10.69}
represents the contravariant components of a tangent vector.
Suppose now that the curve is parameterized by arc length ss, i.e.
Zi=Zi(s).(10.70)Z^{i}=Z^{i}\left( s\right) .\tag{10.70}
In Chapter 5, we established that the derivative
T=R(s)(5.5)\mathbf{T}=\mathbf{R}^{\prime}\left( s\right) \tag{5.5}
produces a unit tangent T\mathbf{T}. Thus, the contravariant components TiT^{i} of the unit tangent T\mathbf{T} are given by the equation
Ti=dZi(s)ds.(10.71)T^{i}=\frac{dZ^{i}\left( s\right) }{ds}.\tag{10.71}
The fact that T\mathbf{T} is unit length is expressed in terms of its components by the equation
ZijTiTj=1,(10.72)Z_{ij}T^{i}T^{j}=1,\tag{10.72}
or, equivalently,
TiTi=1.(10.73)T_{i}T^{i}=1.\tag{10.73}
In Section 5.2, we considered a curve given by the vector equation
R=R(γ),(10.74)\mathbf{R}=\mathbf{R}\left( \gamma\right) ,\tag{10.74}
where γ\gamma is an arbitrary parameter along the curve. We showed that the arc length ss of the segment between the points corresponding to values γ0\gamma_{0} and γ1\gamma_{1} of the parameter γ\gamma is given by the integral
s=γ0γ1R(γ)R(γ)dγ.(5.10)s=\int_{\gamma_{0}}^{\gamma_{1}}\sqrt{\mathbf{R}^{\prime}\left( \gamma\right) \cdot\mathbf{R}^{\prime}\left( \gamma\right) }d\gamma. \tag{5.10}
On the one hand, this formula, which does not require coordinates in the ambient space, proved to be of great theoretical value. On the other hand, due to the lack of ambient coordinates, this formula cannot be used for practical calculations of the length of any concrete curve since it features geometric quantities rather than algebraic expressions. However, since we have introduced coordinates in the ambient space and are now able to work with the components of vectors, we can modify the above formula in order to make it suitable for practical calculations.
Introduce an arbitrary coordinate system ZiZ^{i} in the ambient Euclidean space and let the equations of the curve read
Zi=Zi(γ).(10.68)Z^{i}=Z^{i}\left( \gamma\right) . \tag{10.68}
As we demonstrated in the previous Section, the derivative R(γ)\mathbf{R} ^{\prime}\left( \gamma\right) is given by the equation
R(γ)=dZi(γ)dγZi(10.75)\mathbf{R}^{\prime}\left( \gamma\right) =\frac{dZ^{i}\left( \gamma\right) }{d\gamma}\mathbf{Z}_{i}\tag{10.75}
In other words,
dZidγ(10.76)\frac{dZ^{i}}{d\gamma}\tag{10.76}
are the contravariant components of R(γ)\mathbf{R}^{\prime}\left( \gamma\right) . Since the dot product of a vector U\mathbf{U} with itself is given by the formula
UU=ZijUiUj,(10.77)\mathbf{U}\cdot\mathbf{U}=Z_{ij}U^{i}U^{j},\tag{10.77}
we have
R(γ)R(γ)=ZijdZidγdZjdγ.(10.78)\sqrt{\mathbf{R}^{\prime}\left( \gamma\right) \cdot\mathbf{R}^{\prime }\left( \gamma\right) }=\sqrt{Z_{ij}\frac{dZ^{i}}{d\gamma}\frac{dZ^{j} }{d\gamma}}.\tag{10.78}
With the help of this identity, the equation
s=γ0γ1R(γ)R(γ)dγ.(5.10)s=\int_{\gamma_{0}}^{\gamma_{1}}\sqrt{\mathbf{R}^{\prime}\left( \gamma\right) \cdot\mathbf{R}^{\prime}\left( \gamma\right) }d\gamma. \tag{5.10}
becomes
s=γ0γ1Zij(γ)dZi(γ)dγdZj(γ)dγdγ.(10.79)s=\int_{\gamma_{0}}^{\gamma_{1}}\sqrt{Z_{ij}\left( \gamma\right) \frac{dZ^{i}\left( \gamma\right) }{d\gamma}\frac{dZ^{j}\left( \gamma\right) }{d\gamma}}d\gamma.\tag{10.79}
Let us once again stress one of the most crucial aspects of this formula: it is valid in arbitrary coordinates ZiZ^{i} as well as for an arbitrary parameterization γ\gamma of the curve, provided that γ0γ1\gamma_{0}\leq\gamma_{1}. The practical advantage of this formula over the coordinate-free equation
s=γ0γ1R(γ)R(γ)dγ.(5.10)s=\int_{\gamma_{0}}^{\gamma_{1}}\sqrt{\mathbf{R}^{\prime}\left( \gamma\right) \cdot\mathbf{R}^{\prime}\left( \gamma\right) }d\gamma. \tag{5.10}
is overwhelming and speaks to the immense utility of coordinate systems.
For a concrete example, let us calculate the arc length of a complete loop of the spiral in the following figure. The spiral can be described geometrically as a locus of points whose distance from a particular point OO equals the angle, in radians, to a reference ray ll. Notice that the distance from OO to the point where the spiral intersects the reference line ll for the first time is 2π2\pi.
(10.80)
Evaluating the length of this curve in a coordinate-free setting would present a formidable challenge. With the use of coordinates, however, this task becomes a routine exercise in ordinary Calculus. Let us perform the required analysis in two alternative ambient coordinate systems: Cartesian coordinates x,yx,y and polar coordinates r,θr,\theta.
First, introduce a Cartesian coordinate system x,yx,y that lines up with the reference line ll as in the following figure.
(10.81)
The equations of the curve that describe the spiral in these coordinates read
x(γ)=γcosγ          (10.82)y(γ)=γsinγ.          (10.83)\begin{aligned}x\left( \gamma\right) & =\gamma\cos\gamma\ \ \ \ \ \ \ \ \ \ \left(10.82\right)\\y\left( \gamma\right) & =\gamma\sin\gamma.\ \ \ \ \ \ \ \ \ \ \left(10.83\right)\end{aligned}
Since, in Cartesian coordinates, the metric tensor ZijZ_{ij} at all points corresponds to the identity matrix, we have
Zij(γ)dZi(γ)dγdZj(γ)dγ=x(γ)2+y(γ)2.(10.84)\sqrt{Z_{ij}\left( \gamma\right) \frac{dZ^{i}\left( \gamma\right) }{d\gamma}\frac{dZ^{j}\left( \gamma\right) }{d\gamma}}=\sqrt{x^{\prime }\left( \gamma\right) ^{2}+y^{\prime}\left( \gamma\right) ^{2}}.\tag{10.84}
The derivatives x(γ)x^{\prime}\left( \gamma\right) and y(γ)y^{\prime}\left( \gamma\right) are given by
x(γ)=cosγγsinγ       and          (10.85)y(γ)=sinγ+γcosγ,          (10.86)\begin{aligned}x^{\prime}\left( \gamma\right) & =\cos\gamma-\gamma\sin\gamma\text{ \ \ \ \ \ \ and}\ \ \ \ \ \ \ \ \ \ \left(10.85\right)\\y^{\prime}\left( \gamma\right) & =\sin\gamma+\gamma\cos\gamma,\ \ \ \ \ \ \ \ \ \ \left(10.86\right)\end{aligned}
and therefore the integrand is given by
Zij(γ)dZi(γ)dγdZj(γ)dγ=1+γ2.(10.87)\sqrt{Z_{ij}\left( \gamma\right) \frac{dZ^{i}\left( \gamma\right) }{d\gamma}\frac{dZ^{j}\left( \gamma\right) }{d\gamma}}=\sqrt{1+\gamma^{2}}.\tag{10.87}
Thus the arc length ss is given by the ordinary integral
s=02π1+γ2dγ.(10.88)s=\int_{0}^{2\pi}\sqrt{1+\gamma^{2}}d\gamma.\tag{10.88}
A routine evaluation of the integral yields the final answer
s=π4π2+1+12ln(4π2+1+2π).(10.89)s=\pi\sqrt{4\pi^{2}+1}+\frac{1}{2}\ln\left( \sqrt{4\pi^{2}+1}+2\pi\right) .\tag{10.89}
Now introduce the polar coordinate system r,θr,\theta, where the origin of the coordinate system coincides with the point OO and the polar axis coincides with the ray ll. In these coordinates the equations of the curve take on a particularly simple form, i.e.
r(γ)=γ          (10.90)θ(γ)=γ.          (10.91)\begin{aligned}r\left( \gamma\right) & =\gamma\ \ \ \ \ \ \ \ \ \ \left(10.90\right)\\\theta\left( \gamma\right) & =\gamma.\ \ \ \ \ \ \ \ \ \ \left(10.91\right)\end{aligned}
Recall that the metric tensor ZijZ_{ij} corresponds to the matrix
[100r2].(9.43)\left[ \begin{array} {cc} 1 & 0\\ 0 & r^{2} \end{array} \right] . \tag{9.43}
Along a general curve given by the equations r=r(γ)r=r\left( \gamma\right) and θ=θ(γ)\theta=\theta\left( \gamma\right) spiral, Zij(γ)Z_{ij}\left( \gamma\right) corresponds to the matrix
[100r2(γ)].(10.92)\left[ \begin{array} {cc} 1 & 0\\ 0 & r^{2}\left( \gamma\right) \end{array} \right] .\tag{10.92}
Thus, for a general curve in polar coordinates, the integrand is given by
Zij(γ)dZi(γ)dγdZj(γ)dγ=r(γ)2+r2(γ)θ(γ)2.(10.93)\sqrt{Z_{ij}\left( \gamma\right) \frac{dZ^{i}\left( \gamma\right) }{d\gamma}\frac{dZ^{j}\left( \gamma\right) }{d\gamma}}=\sqrt{r^{\prime }\left( \gamma\right) ^{2}+r^{2}\left( \gamma\right) \theta^{\prime }\left( \gamma\right) ^{2}}.\tag{10.93}
For our specific curve, Zij(γ)Z_{ij}\left( \gamma\right) corresponds to
[100γ2].(10.94)\left[ \begin{array} {cc} 1 & 0\\ 0 & \gamma^{2} \end{array} \right] .\tag{10.94}
Since derivatives r(γ)r^{\prime}\left( \gamma\right) and θ(γ)\theta^{\prime }\left( \gamma\right) are given by
r(γ)=1 and          (10.95)θ(γ)=1,          (10.96)\begin{aligned}r^{\prime}\left( \gamma\right) & =1\text{ and}\ \ \ \ \ \ \ \ \ \ \left(10.95\right)\\\theta^{\prime}\left( \gamma\right) & =1,\ \ \ \ \ \ \ \ \ \ \left(10.96\right)\end{aligned}
the integrand is given by
ZijdZidγdZjdγ=1+γ2.(10.97)\sqrt{Z_{ij}\frac{dZ^{i}}{d\gamma}\frac{dZ^{j}}{d\gamma}}=\sqrt{1+\gamma^{2}}.\tag{10.97}
Not surprisingly, we obtained the exact same expression for the integrand since it equals the length of the vector R(γ)R^{\prime}\left( \gamma\right) and does not depend on the choice of the ambient coordinates. Thus, the rest of the analysis can proceed as before.
This simple examples illustrates the great advantage of the coordinate space approach as it enables us to convert geometric expressions into arithmetic ones which can then be evaluated by the robust techniques of ordinary Calculus.
The equation
s=γ0γ1Zij(γ)dZi(γ)dγdZj(γ)dγdγ(10.98)s=\int_{\gamma_{0}}^{\gamma_{1}}\sqrt{Z_{ij}\left( \gamma\right) \frac{dZ^{i}\left( \gamma\right) }{d\gamma}\frac{dZ^{j}\left( \gamma\right) }{d\gamma}}d\gamma\tag{10.98}
demonstrates that, in combination with the equations of the curve, the metric tensor gives us the ability to calculate the length of any curve. In other words, we do not need any additional geometric details regarding either the curve or the coordinate system.
Crucially, this connection between the metric tensor and the lengths of curves also works in the opposite direction. Namely, an ability to measure the lengths of curves can be used to calculate the metric tensor. Thus, any quantity that can be calculated solely from the metric tensor (and its derivatives) can, in fact, be calculated from the very narrow ability to measure the lengths of curves. This insight will have important implications in the study of Riemannian spaces which we will undertake in Chapter 20.
Let us begin by calculating the entry Z11Z_{11} at a fixed point AA with coordinates (A1,A2,A3)\left( A^{1},A^{2},A^{3}\right) . Consider the coordinate line corresponding to the coordinate Z1Z^{1} illustrated below in a two-dimensional figure.
(10.99)
The equations of this coordinate line read
Z1(γ)=A1+γ          (10.100)Z2(γ)=A2          (10.101)Z3(γ)=A3.          (10.102)\begin{aligned}Z^{1}\left( \gamma\right) & =A^{1}+\gamma\ \ \ \ \ \ \ \ \ \ \left(10.100\right)\\Z^{2}\left( \gamma\right) & =A^{2}\ \ \ \ \ \ \ \ \ \ \left(10.101\right)\\Z^{3}\left( \gamma\right) & =A^{3}.\ \ \ \ \ \ \ \ \ \ \left(10.102\right)\end{aligned}
For a positive γ\gamma, denote the length of this curve from the point AA to the point with coordinate γ\gamma by L(γ)L\left( \gamma\right) . Importantly, the assumption that we are able to measure the lengths of curves by some means (perhaps by taking measurements with the help of some mechanical device) implies that the function L(γ)L\left( \gamma\right) is available -- meaning that we can evaluate its values and the values of its derivatives.
Since Z1(γ)Z^{1}\left( \gamma\right) is the only function in the equations of the curve that varies with γ\gamma, the only surviving term in the sum
Zij(γ)dZidγdZjdγ(10.103)Z_{ij}\left( \gamma\right) \frac{dZ^{i}}{d\gamma}\frac{dZ^{j}}{d\gamma}\tag{10.103}
is
Z11(γ)dZ1(γ)dγdZ1(γ)dγ=Z11(γ).(10.104)Z_{11}\left( \gamma\right) \frac{dZ^{1}\left( \gamma\right) }{d\gamma }\frac{dZ^{1}\left( \gamma\right) }{d\gamma}=Z_{11}\left( \gamma\right) .\tag{10.104}
Therefore, L(γ)L\left( \gamma\right) is given by the integral
L(γ)=0γZ11(t)dt.(10.105)L\left( \gamma\right) =\int_{0}^{\gamma}\sqrt{Z_{11}\left( t\right) }dt.\tag{10.105}
By the Fundamental Theorem of Calculus, the derivative L(γ)L^{\prime}\left( \gamma\right) equals the integrand at t=γt=\gamma, i.e.
L(γ)=Z11(γ).(10.106)L^{\prime}\left( \gamma\right) =\sqrt{Z_{11}\left( \gamma\right) }.\tag{10.106}
Thus, the element Z11Z_{11} of the metric tensor at the point AA is given by the formula
Z11=L(0)2.(10.107)Z_{11}=L^{\prime}\left( 0\right) ^{2}.\tag{10.107}
Thus, we have demonstrated a way to calculate Z11Z_{11} from the lengths of curves. The same approach can be used to evaluate other "diagonal" elements Z22Z_{22} and Z33Z_{33} of ZijZ_{ij}.
To calculate the "off-diagonal" elements, say Z12Z_{12}, consider the curve given by the equations
Z1(γ)=A1+γ          (10.108)Z2(γ)=A2+γ          (10.109)Z3(γ)=A3.          (10.110)\begin{aligned}Z^{1}\left( \gamma\right) & =A^{1}+\gamma\ \ \ \ \ \ \ \ \ \ \left(10.108\right)\\Z^{2}\left( \gamma\right) & =A^{2}+\gamma\ \ \ \ \ \ \ \ \ \ \left(10.109\right)\\Z^{3}\left( \gamma\right) & =A^{3}.\ \ \ \ \ \ \ \ \ \ \left(10.110\right)\end{aligned}
Since the derivatives Zi/γ\partial Z^{i}/\partial\gamma are
Z1(γ)γ=1          (10.111)Z2(γ)γ=1          (10.112)Z3(γ)γ=0,          (10.113)\begin{aligned}\frac{\partial Z^{1}\left( \gamma\right) }{\partial\gamma} & =1\ \ \ \ \ \ \ \ \ \ \left(10.111\right)\\\frac{\partial Z^{2}\left( \gamma\right) }{\partial\gamma} & =1\ \ \ \ \ \ \ \ \ \ \left(10.112\right)\\\frac{\partial Z^{3}\left( \gamma\right) }{\partial\gamma} & =0,\ \ \ \ \ \ \ \ \ \ \left(10.113\right)\end{aligned}
we have
Zij(γ)dZidγdZjdγ=Z11(γ)+2Z12(γ)+Z22(γ)(10.114)Z_{ij}\left( \gamma\right) \frac{dZ^{i}}{d\gamma}\frac{dZ^{j}}{d\gamma }=Z_{11}\left( \gamma\right) +2Z_{12}\left( \gamma\right) +Z_{22}\left( \gamma\right)\tag{10.114}
and therefore
L(γ)=0γZ11(t)+2Z12(t)+Z22(t)dt.(10.115)L\left( \gamma\right) =\int_{0}^{\gamma}\sqrt{Z_{11}\left( t\right) +2Z_{12}\left( t\right) +Z_{22}\left( t\right) }dt.\tag{10.115}
Therefore, the derivative of L(γ)L\left( \gamma\right) at the point AA is given by
L(0)=Z11+2Z12+Z22.(10.116)L^{\prime}\left( 0\right) =\sqrt{Z_{11}+2Z_{12}+Z_{22}}.\tag{10.116}
Since the diagonal terms Z11Z_{11} and Z22Z_{22} are already available, we are able to compute Z12Z_{12} according to the formula
Z12=12(L(0)2Z11Z22).(10.117)Z_{12}=\frac{1}{2}\left( L^{\prime}\left( 0\right) ^{2}-Z_{11} -Z_{22}\right) .\tag{10.117}
The remaining off-diagonal entries can be calculated in similar fashion.
Recall from Chapter 4 that the direction derivative dF/dldF/dl is defined as the rate of change in FF along the ray ll.
(10.118)
For a scalar field FF, we demonstrated that dF/dldF/dl is given by the equation
dFdl=FL,(4.79)\frac{dF}{dl}=\mathbf{\nabla}F\cdot\mathbf{L,} \tag{4.79}
where L\mathbf{L} is the unit vector that points in the direction of the ray ll and the vector F\mathbf{\nabla}F is the gradient of FF. In this Section, we will derive the coordinate space expression for the directional derivative which will, in turn, yield the coordinate space expression for the gradient.
Let LiL^{i} be the contravariant components of L\mathbf{L}, i.e.
L=LiZi.(10.119)\mathbf{L}=L^{i}\mathbf{Z}_{i}.\tag{10.119}
Suppose that the ambient space is referred to arbitrary coordinates ZiZ^{i} and that the equations of the ray ll when parameterized by arc length ss read
Zi=Zi(s).(10.120)Z^{i}=Z^{i}\left( s\right) .\tag{10.120}
Note that even though the ray ll is straight, the functions Zi(s)Z^{i}\left( s\right) are not necessarily linear.
Along the ray, the values of the field FF form a function of ss denoted by F(s)F\left( s\right) . By definition, the directional derivative dF/dldF/dl along the ray ll equals with F(s)F^{\prime}\left( s\right) , i.e.
dFdl=F(s).(10.121)\frac{dF}{dl}=F^{\prime}\left( s\right) .\tag{10.121}
Observe that the function F(s)F\left( s\right) can be obtained by composing the function F(Z)F\left( Z\right) , i.e. the dependence of FF on the ambient coordinates ZiZ^{i}, with the equations of the curve Zi(s)Z^{i}\left( s\right) , i.e.
F(s)=F(Z(s)).(10.122)F\left( s\right) =F\left( Z\left( s\right) \right) .\tag{10.122}
Differentiating both sides, we find
F(s)=F(Z)ZidZi(s)ds.(10.123)F^{\prime}\left( s\right) =\frac{\partial F\left( Z\right) }{\partial Z^{i}}\frac{dZ^{i}\left( s\right) }{ds}.\tag{10.123}
The collection of partial derivatives
F(Z)Zi(10.124)\frac{\partial F\left( Z\right) }{\partial Z^{i}}\tag{10.124}
will prove to be a dominant object in our narrative and therefore deserves its own symbol with a clearly indicated index placement. Thus, we will denote it by iF\nabla_{i}F, i.e.
iF=F(Z)Zi.(10.125)\nabla_{i}F=\frac{\partial F\left( Z\right) }{\partial Z^{i}}.\tag{10.125}
In Chapter 15, the symbol i\nabla_{i} will be extended to the new differential operator known as the covariant derivative.
Meanwhile, as we described in Section 10.8.3, the derivatives
dZi(s)ds(10.126)\frac{dZ^{i}\left( s\right) }{ds}\tag{10.126}
represent the components of the unit tangent to the ray ll, which is precisely the vector L\mathbf{L}. Therefore, we have
Li=dZi(s)ds.(10.127)L^{i}=\frac{dZ^{i}\left( s\right) }{ds}.\tag{10.127}
In summary, the directional derivative dF/dldF/dl is captured by the equation
dFdl=iF Li,(10.128)\frac{dF}{dl}=\nabla_{i}F\ L^{i},\tag{10.128}
which constitutes the coordinate space representation of the directional derivative. As with all other coordinate space formulas that we have encountered so far, it is valid in all coordinate systems.
The formula
dFdl=iF Li(10.128)\frac{dF}{dl}=\nabla_{i}F\ L^{i} \tag{10.128}
will now reveal to us the coordinate space representation of the gradient, which has eluded us until now. The combination iF Li\nabla_{i}F\ L^{i} represents the dot product of the two vectors
iF Zi    and    Li Zi.(10.129)\nabla_{i}F\ \mathbf{Z}^{i}\text{ \ \ \ and \ \ \ }L^{i}\ \mathbf{Z}_{i}.\tag{10.129}
The second vector is, of course, L\mathbf{L}. Since the formula
dFdl=FL(4.79)\frac{dF}{dl}=\mathbf{\nabla}F\cdot\mathbf{L} \tag{4.79}
holds for every unit vector L\mathbf{L}, we can conclude that the combination iF Zi\nabla_{i}F\ \mathbf{Z}^{i} must represent the gradient F\mathbf{\nabla}F, i.e.
F=iF Zi.(10.130)\mathbf{\nabla}F=\nabla_{i}F\ \mathbf{Z}^{i}.\tag{10.130}
This is the coordinate space representation of the gradient F\mathbf{\nabla}F that we have been seeking ever since Chapter 4.
We can now understand the underlying flaw in our original attempt at the coordinate space expression for the gradient, i.e.
F=Fxi+Fyj+Fzk,(6.66)\mathbf{\nabla}F=\frac{\partial F}{\partial x}\mathbf{i}+\frac{\partial F}{\partial y}\mathbf{j}+\frac{\partial F}{\partial z}\mathbf{k,} \tag{6.66}
found at the end of Chapter 6. In more general terms, this equation may be rewritten as
F=FZ1Z1+FZ2Z2+FZ3Z3(10.131)\mathbf{\nabla}F=\frac{\partial F}{\partial Z^{1}}\mathbf{Z}_{1} +\frac{\partial F}{\partial Z^{2}}\mathbf{Z}_{2}+\frac{\partial F}{\partial Z^{3}}\mathbf{Z}_{3}\tag{10.131}
or, equivalently,
(10.132)\tag{10.132}
mathbf{nabla}F=nabla_{1}F mathbf{Z}_{1}+nabla_{2}F mathbf{Z} _{2}+nabla_{3}F mathbf{Z}_{3}. tag*{-} end{equation} With the help of the summation sign, it can also be written as
F=iiF Zi.(-)\mathbf{\nabla}F=\sum_{i}\nabla_{i}F\ \mathbf{Z}_{i}. \tag{-}
Naturally, we cannot write this expression without the summation sign since the combination iF Zi\nabla_{i}F\ \mathbf{Z}_{i} violates the rules of the tensor notations that require the repeated indices to be of opposite flavors in a contraction. Correspondingly, we observed in Chapter 6 that when a coordinate system is "stretched" by a factor 22, both iF\nabla_{i}F and Zi\mathbf{Z}_{i} double and thus the product iF Zi\nabla_{i}F\ \mathbf{Z}_{i} quadruples. In other words, the above formula produces different vectors in different coordinate systems and is therefore -- to put it bluntly -- geometrically meaningless.
Although we have not yet studied transformation of variants under coordinate changes, we can rely on the tensor notation in order to predict the reason why the sum iiF Zi\sum_{i}\nabla_{i}F\ \mathbf{Z}_{i} quadruples. Both elements in the product have subscripts and therefore transform in the same way under a change of coordinates. Thus, if Zi\mathbf{Z}_{i} doubles then so does iF\nabla_{i}F and then their product, predictably, quadruples. The formula
F=iF Zi,(10.130)\mathbf{\nabla}F=\nabla_{i}F\ \mathbf{Z}^{i}, \tag{10.130}
by combining objects that transform by opposite rules, corrects our original mistake.
Interestingly, our original guess that the components of the gradient of FF are the partial derivatives
iF=FZi(10.125)\nabla_{i}F=\frac{\partial F}{\partial Z^{i}} \tag{10.125}
was correct. What we have originally missed, but have now corrected, was to interpret those values as the covariant components that must be combined with the contravariant basis.
We started our overall narrative in a strictly geometric setting with the geometric vector being the primary object of our study. There were several advantages to such an approach. First, it enabled us to lean heavily on our geometric intuition. Second, it allowed us to develop an analytical framework characterized by a very limited number of available operations. Finally, we were able to use the geometric space as an absolute reference, which enabled us to assure the internal consistency of our calculations.
On the other hand, what our approach lacked was the capability to analyze specific problems. We have now addressed this shortcoming by switching to the analysis of the components of vectors rather than the vectors themselves. Several examples in this Chapter have already demonstrated the great utility of this coordinate space approach. What is particularly appealing about it is that it is not a replacement but an augmentation of the geometric approach. Thus, our geometric framework will continue to provide us with an opportunity to verify our results in the geometric space. However, our emphasis will begin to shift towards the component perspective.
Exercise 10.1Derive the equation
Ui=ZiU(10.19)U_{i}=\mathbf{Z}_{i}\cdot\mathbf{U} \tag{10.19}
by the method used in Section 10.4.
Exercise 10.2Derive the equation
Ui=ZijUj(10.9)U^{i}=Z^{ij}U_{j} \tag{10.9}
from the equation
U=UiZi.(10.4)\mathbf{U}=U_{i}\mathbf{Z}^{i}. \tag{10.4}
Exercise 10.3In spherical coordinates, find the contravariant and the covariant components UiU^{i} and UiU_{i} of the unit vector pointing in the direction of the polar axis at a point with coordinates r,θ,φr,\theta,\varphi. Confirm that UiUi=1U_{i} U^{i}=1.
Exercise 10.4Analyze the helical motion described in Section 10.8.1 in Cartesian coordinates. Namely, find the contravariant components ViV^{i} of the velocity vector for a particle moving along a helix. Confirm that the resulting expression for the magnitude of the velocity vector is the same as found in Section 10.8.1.
Send feedback to Pavel