The Surface Covariant Derivative

Since the emergence of surface variants with ambient indices, such as the shift tensor ZαiZ_{\alpha}^{i} and the components NiN^{i} of the unit normal, we have been faced with the need to expand the definition of the surface covariant derivative γ\nabla_{\gamma} to such objects. This is an important step in pursuing a complete analytical framework in which differential operators can be applied to all types of available objects. It is also an important step in enabling us to work with the components of vectors rather than vectors themselves.
In our quest for such a generalization of the surface covariant derivative, we can be guided by two closely related landmark relationships: the equation
αSβ=NBαβ,(2.79)\nabla_{\alpha}\mathbf{S}_{\beta}=\mathbf{N}B_{\alpha\beta}, \tag{2.79}
which serves as the definition of the curvature tensor BαβB_{\alpha\beta}, and Weingarten's equation
αN=BαβSβ.(2.91)\nabla_{\alpha}\mathbf{N}=-B_{\alpha}^{\beta}\mathbf{S}_{\beta}. \tag{2.91}
Since ZβiZ_{\beta}^{i} are the components of Sβ\mathbf{S}_{\beta} and NiN^{i} are the components of N\mathbf{N}, we should expect that the component forms of these equations read
αZβi=NiBαβ.(5.1)\nabla_{\alpha}Z_{\beta}^{i}=N^{i}B_{\alpha\beta}.\tag{5.1}
and
αNi=ZβiBαβ,(5.2)\nabla_{\alpha}N^{i}=-Z_{\beta}^{i}B_{\alpha}^{\beta},\tag{5.2}
where α\nabla_{\alpha} is the proper generalization of the surface covariant derivative. We will indeed arrive at the above equations, but a fair amount of groundwork needs to be laid down first.
Our ultimate goal is to give a definition of the surface covariant derivative γ\nabla_{\gamma} for variants with arbitrary collections of surface and ambient indices. However, rather than postulate the definition and subsequently present its properties, we will show how one can arrive at the definition on their own.
To this end, let us start with a variant TiT^{i} that features a single ambient index, and attempt to discover the proper form for
γTi(5.3)\nabla_{\gamma}T^{i}\tag{5.3}
by assuming that γ\nabla_{\gamma} has two desirable properties: one, that it satisfies the product rule and, two, that it is metrinilic with respect to the ambient basis Zi\mathbf{Z}_{i}. After all, these are the two properties that are needed in order to transfer analysis from a Euclidean space to the component space. Indeed, when the covariant derivative is applied to the combination TiZiT^{i}\mathbf{Z}_{i}, it is by the product rule that it splits into two, i.e.
γ(TiZi)=γTi Zi+TiγZi,(5.4)\nabla_{\gamma}\left( T^{i}\mathbf{Z}_{i}\right) =\nabla_{\gamma} T^{i}\ \mathbf{Z}_{i}+T^{i}\nabla_{\gamma}\mathbf{Z}_{i},\tag{5.4}
and it is by the metrinilic property that the second term vanishes, i.e.
γ(TiZi)=γTi Zi.(5.5)\nabla_{\gamma}\left( T^{i}\mathbf{Z}_{i}\right) =\nabla_{\gamma} T^{i}\ \mathbf{Z}_{i}.\tag{5.5}
Thus, it is by combination of these two properties that the covariant basis Zi\mathbf{Z}_{i} cleanly "separates" from the rest of the terms and thus facilitates conversion to component form.
For a more concrete example, consider the conversion from
αSβ=NBαβ    to    αZβi=NiBαβ.(5.6)\nabla_{\alpha}\mathbf{S}_{\beta}=\mathbf{N}B_{\alpha\beta}\text{ \ \ \ to \ \ \ }\nabla_{\alpha}Z_{\beta}^{i}=N^{i}B_{\alpha\beta}.\tag{5.6}
By the combination of the product rule and the metrinilic property, we have
αSβ=α(ZβiZi)=αZβi Zi+ZβiαZi=αZβiZi(5.7)\nabla_{\alpha}\mathbf{S}_{\beta}=\nabla_{\alpha}\left( Z_{\beta} ^{i}\mathbf{Z}_{i}\right) =\nabla_{\alpha}Z_{\beta}^{i}\ \mathbf{Z} _{i}+Z_{\beta}^{i}\nabla_{\alpha}\mathbf{Z}_{i}=\nabla_{\alpha}Z_{\beta} ^{i}\mathbf{Z}_{i}\tag{5.7}
and therefore
αSβ=NBαβ    implies    αZβi Zi=NiBαβZi.(5.8)\nabla_{\alpha}\mathbf{S}_{\beta}=\mathbf{N}B_{\alpha\beta}\text{ \ \ \ implies\ \ \ \ }\nabla_{\alpha}Z_{\beta}^{i}\ \mathbf{Z}_{i} =N^{i}B_{\alpha\beta}\mathbf{Z}_{i}.\tag{5.8}
Thus,
αZβi=NiBαβ(5.9)\nabla_{\alpha}Z_{\beta}^{i}=N^{i}B_{\alpha\beta}\tag{5.9}
follows by equating the components of matching vectors. Similarly, with the help of the product rule and the metrinilic property, we can transition from
αN=BαβSβ    to     αNi=BαβZβi.(5.10)\nabla_{\alpha}\mathbf{N}=-B_{\alpha}^{\beta}\mathbf{S}_{\beta} \text{\ \ \ \ to \ \ \ \ }\nabla_{\alpha}N^{i}=-B_{\alpha}^{\beta}Z_{\beta }^{i}.\tag{5.10}
Thus, requiring these two properties is essential for preserving our analytical framework.
It turns out that requiring the product rule and the metrinilic property essentially determined the surface covariant derivative. Indeed, for a first-order surface variant TiT^{i}, consider the vector field
T=TiZi.(5.11)\mathbf{T}=T^{i}\mathbf{Z}_{i}.\tag{5.11}
Apply the prospective covariant derivative γ\nabla_{\gamma} to both sides of this identity. Since T\mathbf{T} is a variant of order zero, its covariant derivative must coincide with its partial derivative, i.e.
γT=TSγ.(5.12)\nabla_{\gamma}\mathbf{T}=\frac{\partial\mathbf{T}}{\partial S^{\gamma}}.\tag{5.12}
As for the contraction TiZiT^{i}\mathbf{Z}_{i}, we have
γ(TiZi)=(TiZi)Sγ.(5.13)\nabla_{\gamma}\left( T^{i}\mathbf{Z}_{i}\right) =\frac{\partial\left( T^{i}\mathbf{Z}_{i}\right) }{\partial S^{\gamma}}.\tag{5.13}
By the presumed product rule and metrinilic property for γ\nabla_{\gamma}, and the well-established product rule for /Sγ\partial/\partial S^{\gamma}, we find
γTi Zi=TiSγZi+TiZiSγ.(5.14)\nabla_{\gamma}T^{i}\ \mathbf{Z}_{i}=\frac{\partial T^{i}}{\partial S^{\gamma }}\mathbf{Z}_{i}+T^{i}\frac{\partial\mathbf{Z}_{i}}{\partial S^{\gamma}}.\tag{5.14}
In order to evaluate the derivative Zi(S)/Sγ\partial\mathbf{Z}_{i}\left( S\right) /\partial S^{\gamma}, represent the function Zi(S)\mathbf{Z}_{i}\left( S\right) as a composition of the ambient function Zi(Z)\mathbf{Z}_{i}\left( Z\right) and the equation of the surface Zi(S)Z^{i}\left( S\right) , i.e.
Zi(S)=Zi(Z(S)).(5.15)\mathbf{Z}_{i}\left( S\right) =\mathbf{Z}_{i}\left( Z\left( S\right) \right) .\tag{5.15}
Then, by the chain rule, we have
Zi(S)Sα=Zi(Z)ZjZj(S)Sα.(5.16)\frac{\partial\mathbf{Z}_{i}\left( S\right) }{\partial S^{\alpha}} =\frac{\partial\mathbf{Z}_{i}\left( Z\right) }{\partial Z^{j}}\frac{\partial Z^{j}\left( S\right) }{\partial S^{\alpha}}.\tag{5.16}
Recall that
ZiZj=ΓijkZk.(6.45)\frac{\partial\mathbf{Z}_{i}}{\partial Z^{j}}=\Gamma_{ij}^{k}\mathbf{Z} _{k}.\tag{6.45}
Thus,
ZiSα=ZαjΓijkZk.(5.17)\frac{\partial\mathbf{Z}_{i}}{\partial S^{\alpha}}=Z_{\alpha}^{j}\Gamma _{ij}^{k}\mathbf{Z}_{k}.\tag{5.17}
Picking up where we left off,
αTi Zi=TiSαZi+TiZiSα(5.18)\nabla_{\alpha}T^{i}\ \mathbf{Z}_{i}=\frac{\partial T^{i}}{\partial S^{\alpha }}\mathbf{Z}_{i}+T^{i}\frac{\partial\mathbf{Z}_{i}}{\partial S^{\alpha}}\tag{5.18}
now becomes
αTi Zi=TiSαZi+TiZαjΓijkZk.(5.19)\nabla_{\alpha}T^{i}\ \mathbf{Z}_{i}=\frac{\partial T^{i}}{\partial S^{\alpha }}\mathbf{Z}_{i}+T^{i}Z_{\alpha}^{j}\Gamma_{ij}^{k}\mathbf{Z}_{k}.\tag{5.19}
Since TiZαjΓijkZk=ZαkΓkmiTmZiT^{i}Z_{\alpha}^{j}\Gamma_{ij}^{k}\mathbf{Z}_{k}=Z_{\alpha}^{k} \Gamma_{km}^{i}T^{m}\mathbf{Z}_{i}, we find
αTi Zi=(TiSα+ZαkΓkmiTm)Zi.(5.20)\nabla_{\alpha}T^{i}\ \mathbf{Z}_{i}=\left( \frac{\partial T^{i}}{\partial S^{\alpha}}+Z_{\alpha}^{k}\Gamma_{km}^{i}T^{m}\right) \mathbf{Z}_{i}.\tag{5.20}
Finally, matching up the components, we arrive at the final identity
αTi=TiSα+ZαkΓkmiTm.(5.21)\nabla_{\alpha}T^{i}=\frac{\partial T^{i}}{\partial S^{\alpha}}+Z_{\alpha} ^{k}\Gamma_{km}^{i}T^{m}.\tag{5.21}
It is left as an exercise to show that for a variant TjT_{j} with a subscript, the corresponding expression is
αTj=TjSαZαkΓkjmTm.(5.22)\nabla_{\alpha}T_{j}=\frac{\partial T_{j}}{\partial S^{\alpha}}-Z_{\alpha} ^{k}\Gamma_{kj}^{m}T_{m}.\tag{5.22}
Importantly, note the similarity between these relationships and the ambient covariant derivative identities
kTi=TiZk+ΓkmiTm    and          (5.23)kTj=TjZkΓkjmTm.          (5.24)\begin{aligned}\nabla_{k}T^{i} & =\frac{\partial T^{i}}{\partial Z^{k}}+\Gamma_{km} ^{i}T^{m}\text{ \ \ \ and}\ \ \ \ \ \ \ \ \ \ \left(5.23\right)\\\nabla_{k}T_{j} & =\frac{\partial T_{j}}{\partial Z^{k}}-\Gamma_{kj} ^{m}T_{m}.\ \ \ \ \ \ \ \ \ \ \left(5.24\right)\end{aligned}
We would like to reiterate that the above analysis was explorative in nature. We merely identified the properties that we wished the surface covariant derivative to have and, on the basis of those properties, derived the identity that αTi\nabla_{\alpha}T^{i} must satisfy. We will move forward by using our present experience to postulate a definition of the surface covariant derivative, for which we will then prove that the desired properties actually hold.
Let us now build on our experience with the ambient covariant derivative and the explorations in the previous Section to give the following definition for the surface covariant derivative γ\nabla_{\gamma} for a variant TjiT_{j}^{i} with a representative collection of ambient indices:
γTji=TjiSγ+ZγkΓkmiTjmZγkΓkjmTmi.(5.25)\nabla_{\gamma}T_{j}^{i}=\frac{\partial T_{j}^{i}}{\partial S^{\gamma} }+Z_{\gamma}^{k}\Gamma_{km}^{i}T_{j}^{m}-Z_{\gamma}^{k}\Gamma_{kj}^{m} T_{m}^{i}.\tag{5.25}
But why stop here? Recall that we already have a definition for the surface covariant derivative for variants with surface indices, i.e.
γTβα=TβαSγ+ΓγωαTβωΓγβωTωα.(2.64)\nabla_{\gamma}T_{\beta}^{\alpha}=\frac{\partial T_{\beta}^{\alpha}}{\partial S^{\gamma}}+\Gamma_{\gamma\omega}^{\alpha}T_{\beta}^{\omega}-\Gamma _{\gamma\beta}^{\omega}T_{\omega}^{\alpha}. \tag{2.64}
Thus, we can roll the two definitions into one that applies to variants with arbitrary combinations of ambient and surface indices. For a variant TjβiαT_{j\beta}^{i\alpha} with a fully representative collection of indices, the definition reads
γTjβiα=TjβiαSγ+ZγkΓkmiTjβmαZγkΓkjmTmβiα+ΓγωαTjβiωΓγβωTjωiα.(5.26)\nabla_{\gamma}T_{j\beta}^{i\alpha}=\frac{\partial T_{j\beta}^{i\alpha} }{\partial S^{\gamma}}+Z_{\gamma}^{k}\Gamma_{km}^{i}T_{j\beta}^{m\alpha }-Z_{\gamma}^{k}\Gamma_{kj}^{m}T_{m\beta}^{i\alpha}+\Gamma_{\gamma\omega }^{\alpha}T_{j\beta}^{i\omega}-\Gamma_{\gamma\beta}^{\omega}T_{j\omega }^{i\alpha}.\tag{5.26}
As usual, this equation is interpreted as a four-part recipe for each type of index. Many of its applications rely on its properties rather than the literal definition itself. For example, as we demonstrated above, this is the case for the two objects of present interest, the shift tensor ZαiZ_{\alpha}^{i} and the components NiN^{i} of the unit normal. On the other hand, when the above definition is used in the literal sense, it is usually being applied to variants with a small number of indices.
The flagship property of the surface covariant derivative is that it produces tensor outputs for tensor inputs. More specifically, it produces tensors with one surface covariant order greater than the input. This property, known as the tensor property, is the cornerstone of the surface covariant derivative because it guarantees that its use produces geometrically meaningful objects in all combinations of coordinate systems. The demonstration of this property can be carried out according to the approach we used in Chapter TBD of Introduction to Tensor Calculus for the ambient covariant derivative i\nabla_{i}. Generalizing that approach to the surface covariant derivative is left as an important technical exercise for the reader.
We will now describe all of the remaining key properties of the surface covariant derivative. We will begin with the chain rule -- a new property that γ\nabla_{\gamma} does not share with its predecessors. The rest of the properties will be familiar from studying the ambient covariant derivative as well as the limited version of the surface covariant derivative introduced in Chapter 2.
The surface covariant derivative γ\nabla_{\gamma} applies to surface variants, such as the components NiN^{i} of the unit normal or the shift tensor ZαiZ_{\alpha}^{i}. However, the surface variant may well be the surface restriction of a variant defined in the broader ambient space. For example, γ\nabla_{\gamma} can be applied to the position vector R\mathbf{R}, the ambient covariant basis Zi\mathbf{Z}_{i}, or any other ambient variant. Of course, such variants are also subject to the ambient covariant derivative i\nabla_{i} and the role of the chain rule is to relate the two derivatives.
Naturally, ambient variants can have only ambient indices. While the chain rule holds for variants with arbitrary ambient indicial signatures, we will demonstrate it by a variant TjiT_{j}^{i} with a representative collection of indices. The chain rule reads
γTji=ZγkkTji.(5.27)\nabla_{\gamma}T_{j}^{i}=Z_{\gamma}^{k}\nabla_{k}T_{j}^{i}.\tag{5.27}
In other words, the surface covariant derivative is the "orthogonal projection" of the ambient covariant derivative.
The proof of the chain rule requires a straightforward application of the definition of the surface covariant derivative. Suppose that Tji(Z)T_{j}^{i}\left( Z\right) denotes the dependence of TjiT_{j}^{i} on the ambient coordinates, while Tji(S)T_{j}^{i}\left( S\right) denotes the dependence of its surface restriction on the surface coordinates. The two functions are related by the equation
Tji(S)=Tji(Z(S)),(5.28)T_{j}^{i}\left( S\right) =T_{j}^{i}\left( Z\left( S\right) \right) ,\tag{5.28}
where Zi(S)Z^{i}\left( S\right) are the equations of the surface.
By definition, we have
γTji=Tji(S)Sγ+ZγkΓkmiTkmZγkΓkjmTmi.(5.29)\nabla_{\gamma}T_{j}^{i}=\frac{\partial T_{j}^{i}\left( S\right) }{\partial S^{\gamma}}+Z_{\gamma}^{k}\Gamma_{km}^{i}T_{k}^{m}-Z_{\gamma}^{k}\Gamma _{kj}^{m}T_{m}^{i}.\tag{5.29}
The partial derivative Tji(S)/Sγ\partial T_{j}^{i}\left( S\right) /\partial S^{\gamma} can be obtained by differentiating the identity
Tji(S)=Tji(Z(S))(5.28)T_{j}^{i}\left( S\right) =T_{j}^{i}\left( Z\left( S\right) \right) \tag{5.28}
with respect to SγS^{\gamma} which yields
Tji(S)Sγ=Tji(Z)ZjZk(S)Sγ.(5.30)\frac{\partial T_{j}^{i}\left( S\right) }{\partial S^{\gamma}} =\frac{\partial T_{j}^{i}\left( Z\right) }{\partial Z^{j}}\frac{\partial Z^{k}\left( S\right) }{\partial S^{\gamma}}.\tag{5.30}
Since Zk(S)/Sγ=Zγk\partial Z^{k}\left( S\right) /\partial S^{\gamma}=Z_{\gamma}^{k}, we have
Tji(S)Sγ=Tji(Z)ZkZγk(5.31)\frac{\partial T_{j}^{i}\left( S\right) }{\partial S^{\gamma}} =\frac{\partial T_{j}^{i}\left( Z\right) }{\partial Z^{k}}Z_{\gamma}^{k}\tag{5.31}
and, therefore,
γTji=Tji(Z)ZkZγk+ZγkΓkmiTkmZγkΓkjmTmi.(5.32)\nabla_{\gamma}T_{j}^{i}=\frac{\partial T_{j}^{i}\left( Z\right) }{\partial Z^{k}}Z_{\gamma}^{k}+Z_{\gamma}^{k}\Gamma_{km}^{i}T_{k}^{m}-Z_{\gamma} ^{k}\Gamma_{kj}^{m}T_{m}^{i}.\tag{5.32}
Upon factoring out ZγkZ_{\gamma}^{k}, we find
γTji=Zγk(Tji(Z)Zk+ΓkmiTkmΓkjmTmi).(5.33)\nabla_{\gamma}T_{j}^{i}=Z_{\gamma}^{k}\left( \frac{\partial T_{j}^{i}\left( Z\right) }{\partial Z^{k}}+\Gamma_{km}^{i}T_{k}^{m}-\Gamma_{kj}^{m}T_{m} ^{i}\right) .\tag{5.33}
Since the quantity in parentheses is precisely kTji\nabla_{k}T_{j}^{i}, we have arrived at the identity
γTji=ZγkkTji,(5.34)\nabla_{\gamma}T_{j}^{i}=Z_{\gamma}^{k}\nabla_{k}T_{j}^{i},\tag{5.34}
which is indeed the relationship we set out to show.
Note the interesting correspondence between the chain rule and the directional derivative formula
dFdl=LF(5.35)\frac{dF}{dl}=\mathbf{L}\cdot\mathbf{\nabla}F\tag{5.35}
or
dFdl=LkkF.(5.36)\frac{dF}{dl}=L^{k}\nabla_{k}F.\tag{5.36}
This formula states that dF/dldF/dl is the orthogonal projection of kF\nabla_{k}F onto the ray ll. Similarly, we can think of γTji\nabla_{\gamma}T_{j}^{i} as a "directional derivative" of TjiT_{j}^{i} along the tangent plane. Correspondingly, γTji\nabla_{\gamma}T_{j}^{i} is the orthogonal projection of kTji\nabla_{k}T_{j}^{i} onto the tangent plane.
One immediate consequence of the chain rule is the metrinilic property of the surface covariant derivative with respect to the ambient metrics. Indeed, in order to evaluate γZi\nabla_{\gamma}\mathbf{Z}_{i}, note that
γZi=ZγkkZi(5.37)\nabla_{\gamma}\mathbf{Z}_{i}=Z_{\gamma}^{k}\nabla_{k}\mathbf{Z}_{i}\tag{5.37}
and, since
kZi=0,(5.38)\nabla_{k}\mathbf{Z}_{i}=\mathbf{0,}\tag{5.38}
we conclude that
γZi=0.(5.39)\nabla_{\gamma}\mathbf{Z}_{i}=\mathbf{0.}\tag{5.39}
Naturally, the same conclusion can be reached for all other surface metrics. In summary, we have
γZi, γZi=0          (5.40)γZij, γZij, γδji, γδrsij, γδrstijk, γεijk, γεijk=0.          (5.41)\begin{aligned}\nabla_{\gamma}\mathbf{Z}_{i},\ \nabla_{\gamma}\mathbf{Z}^{i} & =\mathbf{0}\ \ \ \ \ \ \ \ \ \ \left(5.40\right)\\\nabla_{\gamma}Z_{ij},\ \nabla_{\gamma}Z^{ij},\ \nabla_{\gamma}\delta_{j} ^{i},\ \nabla_{\gamma}\delta_{rs}^{ij},\ \nabla_{\gamma}\delta_{rst} ^{ijk},\ \nabla_{\gamma}\varepsilon_{ijk},\ \nabla_{\gamma}\varepsilon^{ijk} & =0.\ \ \ \ \ \ \ \ \ \ \left(5.41\right)\end{aligned}
We will now enumerate all of the remaining properties of the surface covariant derivative starting with the complete description of the metrinilic property.

5.4.1The metrinilic property

At the end of the previous Section, we demonstrated the metrinilic property of the surface covariant derivative with respect to the ambient metrics. Note, however, that the full-fledged version of the derivative introduced in this Chapter coincides with its predecessor described in Chapter 2 for objects with surface indices. Therefore, much like its predecessor, the full-fledged derivative is metrinilic with respect to the surface metrics, i.e.
 γSαβ, γSαβ, γδβα, γδρσαβ, γεαβ, γεαβ=0.(5.42)\ \nabla_{\gamma}S_{\alpha\beta},\ \nabla_{\gamma}S^{\alpha\beta} ,\ \nabla_{\gamma}\delta_{\beta}^{\alpha},\ \nabla_{\gamma}\delta_{\rho\sigma }^{\alpha\beta},\ \nabla_{\gamma}\varepsilon_{\alpha\beta},\ \nabla_{\gamma }\varepsilon^{\alpha\beta}=0.\tag{5.42}
Recall, however, that, crucially, γ\nabla_{\gamma} is not metrinilic with respect to the surface covariant basis Sα\mathbf{S}_{\alpha}, which is one important aspect in which surfaces differ from Euclidean spaces.

5.4.2The product rule

The surface covariant derivative satisfies the product rule, also known as the Leibniz rule. For example,
γ(SjαiTkβ)=γSjαi Tkβ+Sjαi γTkβ.(5.43)\nabla_{\gamma}\left( S_{j\alpha}^{i}T_{k}^{\beta}\right) =\nabla_{\gamma }S_{j\alpha}^{i}~T_{k}^{\beta}+S_{j\alpha}^{i}~\nabla_{\gamma}T_{k}^{\beta}.\tag{5.43}
The product rule can be demonstrated in the same way as the corresponding property for the ambient covariant derivative which was described in Chapter TBD of Introduction to Tensor Calculus. It is left as an exercise for the reader.

5.4.3Index juggling "across" the surface covariant derivative

The combination of the metrinilic property and the product rule implies that indices can be juggled freely "across" the surface covariant derivative. The corresponding property for the ambient derivative was discussed in Chapter TBD of Introduction to Tensor Calculus.
For ambient indices, we have
UiγVi=UiγVi,(5.44)U_{i}\nabla_{\gamma}V^{i}=U^{i}\nabla_{\gamma}V_{i},\tag{5.44}
while
Uiγ=γVi(5.45)U_{i\gamma}=\nabla_{\gamma}V_{i}\tag{5.45}
implies
Uγi=γVi.(5.46)U_{\gamma}^{i}=\nabla_{\gamma}V^{i}.\tag{5.46}
Similarly, for surface indices, we have
UαγVα=UαγVα,(5.47)U_{\alpha}\nabla_{\gamma}V^{\alpha}=U^{\alpha}\nabla_{\gamma}V_{\alpha},\tag{5.47}
while
Uαγ=γVα(5.48)U_{\alpha\gamma}=\nabla_{\gamma}V_{\alpha}\tag{5.48}
implies
Uγα=γVα.(5.49)U_{\cdot\gamma}^{\alpha}=\nabla_{\gamma}V^{\alpha}.\tag{5.49}
In summary, we are able to juggle indices freely without any regard for the presence of the surface covariant derivative. Justifying this statement is left as an exercise for the reader.

5.4.4Commutativity with contraction

The surface covariant derivative γ\nabla_{\gamma} commutes with contraction of both surface and ambient indices. The corresponding property for the ambient derivative was discussed in Section TBD of Introduction to Tensor Calculus.
Consider the expressions
γTiβiα(5.50)\nabla_{\gamma}T_{i\beta}^{i\alpha}\tag{5.50}
and
γTjαiα.(5.51)\nabla_{\gamma}T_{j\alpha}^{i\alpha}.\tag{5.51}
Each of these expressions can be interpreted in two different ways depending on the order in which the covariant derivative and contraction are applied. However, regardless of the interpretation, the expressions yield the same results. The proof of this property as left as an exercise for the reader.

5.4.5The commutative property

The surface covariant derivatives commute when applied to variants with ambient indices, i.e.
αβTi=βαTi.(5.52)\nabla_{\alpha}\nabla_{\beta}T^{i}=\nabla_{\beta}\nabla_{\alpha}T^{i}.\tag{5.52}
The assumption that the ambient space is Euclidean is essential for this property to hold. When the ambient space is not Euclidean, this commutative property would not hold. The proof of this property, which is based on deriving the identity
(αββα)Ti=ZαrZβsRmrsiTm,(5.53)\left( \nabla_{\alpha}\nabla_{\beta}-\nabla_{\beta}\nabla_{\alpha}\right) T^{i}=Z_{\alpha}^{r}Z_{\beta}^{s}R_{\cdot mrs}^{i}T^{m},\tag{5.53}
where RmrsiR_{\cdot mrs}^{i} is the (vanishing) ambient Riemann-Christoffel tensor, is left as an exercise.
On the other hand, the surface covariant derivatives do not commute for variants with surfaces indices. We are already familiar with this property from Chapter 2 where we discussed the fact that
(αββα)Tγ=RδαβγTδ,(7.2)\left( \nabla_{\alpha}\nabla_{\beta}-\nabla_{\beta}\nabla_{\alpha}\right) T^{\gamma}=R_{\cdot\delta\alpha\beta}^{\gamma}T^{\delta}, \tag{7.2}
where the surface Riemann-Christoffel tensor RδαβγR_{\cdot \delta\alpha\beta}^{\gamma} is given by
Rδαβγ=ΓβδγSαΓαδγSβ+ΓαωγΓβδωΓβωγΓαδω.(7.3)R_{\cdot\delta\alpha\beta}^{\gamma}=\frac{\partial\Gamma_{\beta\delta} ^{\gamma}}{\partial S^{\alpha}}-\frac{\partial\Gamma_{\alpha\delta}^{\gamma} }{\partial S^{\beta}}+\Gamma_{\alpha\omega}^{\gamma}\Gamma_{\beta\delta }^{\omega}-\Gamma_{\beta\omega}^{\gamma}\Gamma_{\alpha\delta}^{\omega}. \tag{7.3}
Note that Chapter 7 in its entirety is devoted to the study of the Riemann-Christoffel tensor.
Having established the product rule and the metrinilic properties of the surface covariant derivative γ\nabla_{\gamma}, we have retroactively validated the arguments made at the top of this Chapter by which we converted the vector identities
αSβ=NBαβ(2.79)\nabla_{\alpha}\mathbf{S}_{\beta}=\mathbf{N}B_{\alpha\beta} \tag{2.79}
and
αN=BαβSβ(2.91)\nabla_{\alpha}\mathbf{N}=-B_{\alpha}^{\beta}\mathbf{S}_{\beta} \tag{2.91}
into their component counterparts
αZβi=NiBαβ(5.1)\nabla_{\alpha}Z_{\beta}^{i}=N^{i}B_{\alpha\beta} \tag{5.1}
and
αNi=BαβZβi.(5.2)\nabla_{\alpha}N^{i}=-B_{\alpha}^{\beta}Z_{\beta}^{i}. \tag{5.2}
Like its vector counterpart, the latter formula is known as Weingarten's equation. It is left as an exercise to derive this formula by differentiating the explicit expression for the components NiN^{i}, i.e.
Ni=12εijkεβγZjβZkγ.(3.170)N^{i}=\frac{1}{2}\varepsilon^{ijk}\varepsilon_{\beta\gamma}Z_{j}^{\beta} Z_{k}^{\gamma}. \tag{3.170}
Finally, note that contracting both sides of the equation
αZβi=NiBαβ(5.1)\nabla_{\alpha}Z_{\beta}^{i}=N^{i}B_{\alpha\beta} \tag{5.1}
with NiN_{i} gives an explicit expression for the curvature tensor BβαB_{\beta }^{\alpha}, i.e.
Bαβ=NiαZβi.(5.54)B_{\alpha\beta}=N_{i}\nabla_{\alpha}Z_{\beta}^{i}.\tag{5.54}
Similarly, contracting both sides of the equation
αNi=BαβZβi(5.2)\nabla_{\alpha}N^{i}=-B_{\alpha}^{\beta}Z_{\beta}^{i} \tag{5.2}
with ZiγZ_{i\gamma} (and, subsequently, renaming γ\gamma into β\beta) yields another explicit expression for the curvature tensor, i.e.
Bαβ=ZβiαNi.(5.55)B_{\alpha\beta}=-Z_{\beta}^{i}\nabla_{\alpha}N_{i}.\tag{5.55}
When the ambient space is referred to affine coordinates, the covariant derivative reduces to the partial derivative, i.e.
Bαβ=ZβiNiSγ.(5.56)B_{\alpha\beta}=-Z_{\beta}^{i}\frac{\partial N_{i}}{\partial S^{\gamma}}.\tag{5.56}
Consider a scalar field FF defined in the ambient space. We can evaluate the directional derivative of FF along any direction at any point in the ambient space. However, at the points on the surface, there is a special direction that we might be particularly interested in -- the normal direction. The directional derivative in the normal direction is known as the normal derivative.
Recall from Chapter TBD of Introduction to Tensor Calculus, that the directional derivative dF/dldF/dl of a field FF along the ray ll is given by the formula
dFdl=LF,(6.45)\frac{dF}{dl}=\mathbf{L}\cdot\mathbf{\nabla}F\mathbf{,} \tag{6.45}
where F\mathbf{\nabla}F is the gradient of FF and L\mathbf{L} is the unit vector pointing in the direction of the ray ll. Thus, at a point on a surface, the normal derivative of FF, denoted by
Fn,(5.57)\frac{\partial F}{\partial n},\tag{5.57}
is given by
Fn=NF(5.58)\frac{\partial F}{\partial n}=\mathbf{N}\cdot\mathbf{\nabla}F\tag{5.58}
which, in component form, reads
Fn=NiiF.(5.59)\frac{\partial F}{\partial n}=N^{i}\nabla_{i}F.\tag{5.59}
With the help of the normal derivative, we can decompose the ambient covariant derivative iF\nabla_{i}F along the tangent and normal directions. Recall the projection formula
δki=ZαiZkα+NiNk.(3.159)\delta_{k}^{i}=Z_{\alpha}^{i}Z_{k}^{\alpha}+N^{i}N_{k}. \tag{3.159}
Contracting both sides of this formula with iF\nabla_{i}F, we find
δkiiF=ZαiZkαiF+NiNkiF.(5.60)\delta_{k}^{i}\nabla_{i}F=Z_{\alpha}^{i}Z_{k}^{\alpha}\nabla_{i}F+N^{i} N_{k}\nabla_{i}F.\tag{5.60}
Since
δkiiF=kF,          (5.61)ZαiiF=αF,   and          NiiF=Fn,          (5.62)\begin{aligned}\delta_{k}^{i}\nabla_{i}F & =\nabla_{k}F,\ \ \ \ \ \ \ \ \ \ \left(5.61\right)\\Z_{\alpha}^{i}\nabla_{i}F & =\nabla_{\alpha}F,\text{ \ \ and}\ \ \ \ \ \ \ \ \ \ \\N^{i}\nabla_{i}F & =\frac{\partial F}{\partial n},\ \ \ \ \ \ \ \ \ \ \left(5.62\right)\end{aligned}
we discover that
kF=ZkααF+NkFn.(5.63)\nabla_{k}F=Z_{k}^{\alpha}\nabla_{\alpha}F+N_{k}\frac{\partial F}{\partial n}.\tag{5.63}
Contracting both sides with the ambient contravariant basis Zk\mathbf{Z}^{k}, we get the corresponding decomposition for the vector gradient F\mathbf{\nabla }F, i.e.
F=αF Sα+FnN,(5.64)\mathbf{\nabla}F=\nabla_{\alpha}F\ \mathbf{S}^{\alpha}+\frac{\partial F}{\partial n}\mathbf{N},\tag{5.64}
which can also be written in the invariant form
F=SF+FnN.(5.65)\mathbf{\nabla}F=\mathbf{\nabla}_{S}F+\frac{\partial F}{\partial n}\mathbf{N}.\tag{5.65}
The above equation relates various first-order derivatives of FF. We will now derive the beautiful invariant formula
ααF=iiFNiNjijF+BααNiiF.(5.66)\nabla_{\alpha}\nabla^{\alpha}F=\nabla_{i}\nabla^{i}F-N^{i}N^{j}\nabla _{i}\nabla_{j}F+B_{\alpha}^{\alpha}N^{i}\nabla_{i}F.\tag{5.66}
relating second-order derivatives of FF. Note that each term in this formula has a straightforward geometric interpretation: ααF\nabla_{\alpha} \nabla^{\alpha}F is the surface Laplacian of FF, iiF\nabla_{i}\nabla^{i}F is the ambient Laplacian of FF, NiNjijFN^{i}N^{j}\nabla_{i}\nabla_{j}F is the second-order normal derivative 2F/n2\partial^{2}F/\partial n^{2} (this requires proof which is left as an exercise), and, finally, BααNiiFB_{\alpha}^{\alpha} N^{i}\nabla_{i}F is the product of the mean curvature BααB_{\alpha}^{\alpha} and the normal derivative F/n\partial F/\partial n. In terms of the invariant symbols ΔS\Delta_{S} for the surface Laplacian and Δ\Delta for the ambient Laplacian, the above identity can be written as
ΔSF=ΔF2Fn2+BααFn.(5.67)\Delta_{S}F=\Delta F-\frac{\partial^{2}F}{\partial n^{2}}+B_{\alpha}^{\alpha }\frac{\partial F}{\partial n}.\tag{5.67}
In order to prove this identity, recall that, by the chain rule
αF=ZiαiF(5.68)\nabla^{\alpha}F=Z_{i}^{\alpha}\nabla^{i}F\tag{5.68}
and therefore
ααF=α(ZiαiF).(5.69)\nabla_{\alpha}\nabla^{\alpha}F=\nabla_{\alpha}\left( Z_{i}^{\alpha} \nabla^{i}F\right) .\tag{5.69}
Applying the product rule on the right, we find
ααF=αZiαiF+ZiααiF.(5.70)\nabla_{\alpha}\nabla^{\alpha}F=\nabla_{\alpha}Z_{i}^{\alpha}\nabla^{i} F+Z_{i}^{\alpha}\nabla_{\alpha}\nabla^{i}F.\tag{5.70}
Since
αZiα=NiBαα,(5.71)\nabla_{\alpha}Z_{i}^{\alpha}=N_{i}B_{\alpha}^{\alpha},\tag{5.71}
we have
ααF=BααNiiF+ZiααiF.(5.72)\nabla_{\alpha}\nabla^{\alpha}F=B_{\alpha}^{\alpha}N_{i}\nabla^{i} F+Z_{i}^{\alpha}\nabla_{\alpha}\nabla^{i}F.\tag{5.72}
Applying the chain rule again to the second term, i.e.
αiF=ZαjjiF,(5.73)\nabla_{\alpha}\nabla^{i}F=Z_{\alpha}^{j}\nabla_{j}\nabla^{i}F,\tag{5.73}
yields
ααF=BααNiiF+ZiαZαjjiF.(5.74)\nabla_{\alpha}\nabla^{\alpha}F=B_{\alpha}^{\alpha}N_{i}\nabla^{i} F+Z_{i}^{\alpha}Z_{\alpha}^{j}\nabla_{j}\nabla^{i}F.\tag{5.74}
Finally, according to the projection formula
ZiαZαj=δijNiNj,(3.159)Z_{i}^{\alpha}Z_{\alpha}^{j}=\delta_{i}^{j}-N_{i}N^{j}, \tag{3.159}
we have
ααF=BααNiiF+(δijNiNj)jiF.(5.75)\nabla_{\alpha}\nabla^{\alpha}F=B_{\alpha}^{\alpha}N_{i}\nabla^{i}F+\left( \delta_{i}^{j}-N_{i}N^{j}\right) \nabla_{j}\nabla^{i}F.\tag{5.75}
Upon multiplying out the parentheses and absorbing the Kronecker delta, we arrive at the final result
ααF=iiFNiNjijF+BααNiiF.(5.76)\nabla_{\alpha}\nabla^{\alpha}F=\nabla_{i}\nabla^{i}F-N^{i}N^{j}\nabla _{i}\nabla_{j}F+B_{\alpha}^{\alpha}N^{i}\nabla_{i}F.\tag{5.76}
This completes our discussion of the surface covariant derivative and puts us in a position to continue our exploration of curvature.

5.6.1Exercises

Exercise 5.1Show that for a first-order variant TjT_{j} indexed by a subscript, the prospective surface covariant derivative γ\nabla_{\gamma}, subject to the product rule and the metrinilic property, must satisfy the identity
γTj=TjSαZγkΓkjmTm.(5.22)\nabla_{\gamma}T_{j}=\frac{\partial T_{j}}{\partial S^{\alpha}}-Z_{\gamma} ^{k}\Gamma_{kj}^{m}T_{m}. \tag{5.22}
Exercise 5.2Use the techniques employed in Chapter TBD of Introduction to Tensor Calculus to demonstrate the tensor property of the surface covariant derivative.
Exercise 5.3Show that the surface covariant derivative satisfies the product rule. For example,
γ(SαiTjβ)=γSαi Tjβ+Sαi γTjβ.(5.77)\nabla_{\gamma}\left( S_{\alpha}^{i}T_{j}^{\beta}\right) =\nabla_{\gamma }S_{\alpha}^{i}~T_{j}^{\beta}+S_{\alpha}^{i}~\nabla_{\gamma}T_{j}^{\beta}.\tag{5.77}
Exercise 5.4Justify each of the index juggling relationships in Section 5.4.3.
Exercise 5.5Show the contraction property of the surface covariant derivative.
Exercise 5.6Show that
(αββα)Tj=ZαrZβsRmrsjTm,(5.78)\left( \nabla_{\alpha}\nabla_{\beta}-\nabla_{\beta}\nabla_{\alpha}\right) T^{j}=Z_{\alpha}^{r}Z_{\beta}^{s}R_{\cdot mrs}^{j}T^{m},\tag{5.78}
where RmrsjR_{\cdot mrs}^{j} is the ambient Riemann-Christoffel tensor. Since Rmrsj=0R_{\cdot mrs}^{j}=0, we can conclude that
(αββα)Tj=0,(5.79)\left( \nabla_{\alpha}\nabla_{\beta}-\nabla_{\beta}\nabla_{\alpha}\right) T^{j}=0,\tag{5.79}
i.e. the surface covariant derivatives commute when applied to TjT^{j}.
Exercise 5.7Derive Weingarten's equation
αNi=BαβZβi(5.2)\nabla_{\alpha}N^{i}=-B_{\alpha}^{\beta}Z_{\beta}^{i} \tag{5.2}
by applying the literal definition of the surface covariant derivative to the explicit expression for the components of the normal
Ni=12εijkεβγZjβZkγ.(3.170)N^{i}=\frac{1}{2}\varepsilon^{ijk}\varepsilon_{\beta\gamma}Z_{j}^{\beta} Z_{k}^{\gamma}. \tag{3.170}
Exercise 5.8Show that when the formula
Bαβ=NiαZβi(5.54)B_{\alpha\beta}=N_{i}\nabla_{\alpha}Z_{\beta}^{i} \tag{5.54}
is used for the calculation of the curvature tensor, the term with the surface Christoffel symbol in αZβi\nabla_{\alpha}Z_{\beta}^{i} can be omitted.
Exercise 5.9Show that the collection αβF\nabla_{\alpha}\nabla_{\beta}F of second-order surface covariant derivatives is related to the ambient covariant derivatives by the identity
αβF=BαβNjjF+ZαiZβjijF.(5.80)\nabla_{\alpha}\nabla_{\beta}F=B_{\alpha\beta}N_{j}\nabla^{j}F+Z_{\alpha} ^{i}Z_{\beta}^{j}\nabla_{i}\nabla_{j}F.\tag{5.80}
Exercise 5.10Explain why the equation
ααF=ZαiiαF(5.81)\nabla_{\alpha}\nabla^{\alpha}F=Z_{\alpha}^{i}\nabla_{i}\nabla^{\alpha}F\tag{5.81}
represents an invalid application of the chain rule.
Problem 5.1Show that
2Fn2=NiNjijF.(5.82)\frac{\partial^{2}F}{\partial n^{2}}=N^{i}N^{j}\nabla_{i}\nabla_{j}F.\tag{5.82}
See Problem TBD (second directional derivative) in Introduction to Tensor Calculus.
Send feedback to Pavel