nLab Riemannian immersion




A Riemannian immersion or isometric immersion of Riemannian manifolds is an immersion ΣX\Sigma \hookrightarrow X of their underlying smooth manifolds which is also an isometry with respect to their Riemannian metrics.

Similarly, an isometric embedding is an isometric immersion which is also an embedding of smooth manifolds, hence of the underlying topological spaces.

The isometric immersion/embedding problem is to find isometric immersions/embeddings of Riemannian manifolds into large-dimensional but flat (Euclidean) spaces (e.g. Han & Hong 2006, Han & Lewicka 2023).


Key properties of Riemannian immersion — such as whether they are harmonic — are encoded in a tensor called their “second fundamental form”, which is most immediately expressed in terms of a local Darboux coframe field adapted to the immerison.

For the following purpose:


A (local) co-frame field on a smooth manifold XX is

  1. an open cover X^X\widehat{X} \twoheadrightarrow X

  2. differential 1-formsE=(E a) a=1 dim(X)Ω dR 1(X^)E = (E^a)_{a=1}^{dim(X)} \,\in\, \Omega^1_{dR}(\widehat{X})

such that on double intersections

X^× XX^pr 2pr 1X \widehat{X} \!\times_X\! \widehat{X} \underoverset{pr_2}{pr_1}{\rightrightarrows} X

the induced metric tensors agree:

δ abpr 1 *E apr 1 *E b=δ abpr 2 *E apr 2 *E b \delta_{a b} \, pr_1^\ast E^a \otimes pr_1^\ast E^b \;\; = \;\; \delta_{a b} \, pr_2^\ast E^a \otimes pr_2^\ast E^b

where we use the Einstein summation convention, throughout.

A pair of such local co-frames is regarded as equivalent if their induced metric tensors agree on the common refinement of their respective open covers.

Local Darboux (co-)frame field

Given a Riemannian manifold (X,g)(X,g) of dimension DD, then

  • an orthonormal local frame is an open cover p:X^Xp \colon \widehat{X} \twoheadrightarrow X equipped with a DD-tuple of vector fields

    V: DTX^ V \;\colon\; \mathbb{R}^{D} \xrightarrow{\;\;} T \widehat X

    such that at each point x^X^\widehat x \,\in\, \widehat{X} we have that

    g(V,V)(x) g\big(V,V\big)(x)

    is the canonical inner product on D\mathbb{R}^D, hence in components

    g(V a,V b)(x)=δ ab; g\big(V_a, V_b\big)(x) \;=\; \delta_{a b} \,;
  • an orthonormal local co-frame is an open cover p:X^Xp \colon \widehat{X} \twoheadrightarrow X equipped with a D\mathbb{R}^D-valued differential 1-form

    E:TX^ D E \;\colon\; T\widehat{X} \xrightarrow{\;\;} \mathbb{R}^D

    such that at each point x^X^\widehat x \,\in\, \widehat{X} we have

    δ abE aE b=g. \delta_{a b} \, E^a \otimes E^b \;=\; g \,.

Now given moreover an immersion ϕ:ΣX\phi \colon \Sigma \hookrightarrow X into a Riemannian manifold


The Darboux co-frame property (2) immediately implies that the pullback of the remaining frame components

e(e aϕ *E a) adim(Σ) e \;\coloneqq\; \big( e^a \;\coloneqq\; \phi^\ast E^a \big)_{ a \leq dim(\Sigma) }

constitute a local frame on Σ\Sigma. We may summarize this by saying that a local Darboux co-frame EE gives

(3)ϕ *E a={e a foradim(Σ) 0 foradim(Σ). \phi^\ast E^a \;=\; \left\{ \begin{array}{ll} e^a & \text{for}\; a \leq dim(\Sigma) \\ 0 & \text{for}\; a \geq dim(\Sigma) \,. \end{array} \right.

(cf. Griffiths & Harris 1979 (1.13)).

Incidentally, we may observe with GSS24, §2 that this situation (3) of Darboux co-frames, when lifted/applied to the bosonic coframe components of super spacetimes, has later come to be known as the “embedding condition” in the “super-embedding approach” to super p-branes [Sorokin 2000 (4.36-37); Bandos 2011 (2.6-2.9); Bandos & Sorokin 2023 (5.13-14)], strenghtening the original “geometrodynamical condition” of Bandos et al. 1995 (2.23), which is just the first condition in (3).


Given an immersion into a Riemannian manifold, local Darboux (co-)frames always exist.


Given an immersion ι:ΣX\iota \colon \Sigma \to X, consider any point σΣ\sigma \in \Sigma. Since the immersion is locally an embedding (see here), there exists an open neighbourhood σUΣ \sigma \in U \subset \Sigma such that ϕ |U:UX\phi_{\vert U} \colon U \to X is the embedding of a submanifold. Therefore (by this Prop.) there exists an open neighbourhood UXU' \subset X of ι(σ)X\iota(\sigma) \in X which serves as a coordinate chart XUϕ nX \supset U' \xrightarrow{\phi} \mathbb{R}^n for XX and a slice chart for ΣX\Sigma \subset X in that it exhibits ΣU\Sigma \cap U' as a rectilinear hyperplane in ϕ(U) n\phi(U') \subset \mathbb{R}^n.

Hence in this slice coordinate chart a local frame for XX is given by the nn canonical coordinate vector fields, with the first kk of them forming a local frame field on Σ\Sigma

{ 1,, k}{ 1,, k, k+1,, n}. \big\{ \partial_1, \cdots, \partial_{k} \big\} \hookrightarrow \big\{ \partial_1, \cdots, \partial_{k}, \, \partial_{k+1}, \cdots, \partial_n \big\} \,.

From this local frame the Gram-Schmidt process produces an orthonormal frame, first for Σ\Sigma and then extended to XX:

{v 1,,v k}{v 1,,v k,v k+1,,v n},g(v a,v b)=δ ab \big\{ v_1, \cdots, v_{k} \big\} \hookrightarrow \big\{ v_1, \cdots, v_{k}, \, v_{k+1}, \cdots, v_n \big\} \,, \;\;\; g(v_a, v_b) = \delta_{a b} \,

This demonstrates the existence of orthonormal local frame fields (cf. Kayban 2021, Prop. 3.1).

To obtain the desired orthonormal local co-frame field we just dualize this local frame field:

By construction, the matrix (v a μ) a,μ\big(v_a^\mu\big)_{a,\mu} of components of the above frame (given by v a μ μ=v av_a^\mu \partial_\mu = v_a) is block diagonal (the upper diagonal block being the local frame on Σ\Sigma).

This means (cf. e.g. here) that also the inverse matrix

(e μ a) μ,a((v a μ) μ,a) 1 \big(e^a_\mu\big)_{\mu,a} \,\coloneqq\, \Big(\big(v_a^\mu\big)_{\mu,a}\Big)^{-1}

is block diagonal, with its upper diagonal block being the inverse matrix of the original upper left block.

This gives the desired coframe field:

e ae μ adx μ e^{a} \,\coloneqq\, e^a_\mu \, \mathrm{d}x^\mu

which is orthonormal on XX

e ae a=g(,) e^a \otimes e_a \;=\; g(-,-)


v a μg μνv b ν=δ abg μν=e μ aδ abe ν b. v_a^\mu g_{\mu \nu} v_b^\nu \;=\; \delta_{a b} \;\;\;\; \Leftrightarrow \;\;\;\; g_{\mu \nu} \;=\; e^a_\mu \delta_{a b} e^b_\nu \,.

and which is Darboux by the block-diagonal structure of (e μ a)\big(e^a_\mu\big).

(For the further generality of sequences of Darboux frames for suitable sequences of immersions, see Giron 2020 and Chen & Giron 2021, Thm. 2.2.)


In particular, an immersion ι:ΣX\iota \colon \Sigma \hookrightarrow X of Riemannian manifolds is isometric iff around each point ι(σ)\iota(\sigma) any of its Darboux coframe fields pull back to locally induce the given metric on Σ\Sigma.

Second fundamental form

The “second fundamental form” (on the terminology cf. Rem. below) of a Riemannian immersion ΣX\Sigma \xrightarrow{\phantom{-}} X measures the transversal change of tangent vectors to Σ\Sigma, under their tangential transport along Σ\Sigma inside XX.

We first discuss the second fundamental form in terms of local Darboux co-frame fields, where its definition is most immediate, and then extract equivalent expressions in terms of covariant derivatives.

Given XX a Riemannian manifold and ϕ:ΣX\phi \colon \Sigma \to X an immersion, choose a Darboux co-frame field E(E a) a=1 dim(X)E \equiv (E^a)_{a =1}^{dim(X)} (which exists by Prop. ), hence so that

ϕ *E a=0 fora{dim(Σ)+1,,dim(X)} (e aϕ *E a) a=1 dim(Σ) is a co-frame onΣ \begin{array}{l} \phi^\ast E^a \;=\; 0 & \text{for} \; a \in \big\{dim(\Sigma)+1, \cdots, dim(X)\big\} \\ \big( e^a \;\coloneqq\; \phi^\ast E^a \big)_{a = 1}^{dim(\Sigma)} & \text{is a co-frame on}\; \Sigma \end{array}

For brevity we will say that a{1,,dim(X)}a \in \{1, \cdots, dim(X)\} is

  • tangential if a{1,,dim(Σ)}a \in \big\{1, \cdots, dim(\Sigma)\big\}

  • transversal if a{dim(Σ)+1,,dim(X)}a \in \big\{dim(\Sigma)+1, \cdots, dim(X)\big\}.

Now let Ω\Omega be the unique torsion-free connection for EE, in that

(4)dE aΩ a bE b=0. \mathrm{d} E^a - \Omega^a{}_b E^b \;=\; 0 \,.

and denote the pullback of its tangential and transversal components, respectively, by

(5)ω a b ϕ *Ω a b for tangentialand tangentialb b 1b 2 ae b 1 ϕ *Ω a b 2 for transversalaand tangentialb 2. \begin{array}{cccl} \omega^a{}_b &\coloneqq& \phi^\ast \Omega^a{}_b & \text{for tangential}\;\text{and tangential}\;b \\ {Ⅱ}^a_{b_1 b_2} e^{b_1} &\coloneqq& \phi^\ast \Omega^a{}_{b_2} & \text{for transversal}\;a\;\text{and tangential}\;b_2 \,. \end{array}

Then the pullback of the torsion constraint (4) is equivalent to the following pair of conditions on Σ\Sigma:

{de aω a be b=0 for tangentiala b 1b 2 ae b 1e b 2=0 for transversala \left\{ \begin{array}{ll} \mathrm{d} e^a - \omega^a{}_b \, e^b \;=\; 0 & \text{for tangential}\;a \\ {Ⅱ}^a_{b_1 b_2} e^{b_1} e^{b_2} \;=\; 0 & \text{for transversal}\;a \end{array} \right.

The first one just says that ω\omega is the torsion-free connection with respect to the induced coframe ee on Σ\Sigma.

The second one says that the skew symmetric part [b 1b 2] a=0{Ⅱ}^a_{[b_1 b_2]} = 0 vanishes, hence that

(6) b 1b 2 a= b 2b 1 a {Ⅱ}^a_{b_1 b_2} \;=\; {Ⅱ}^a_{b_2 b_1}

is symmetric in its tangential indices b ib_i. This symmetric tensor on Σ\Sigma is called the second fundamental form of the immersion ϕ\phi.

(e.g. Willmore 1996, p. 125-126, Berger, Bryant & Griffiths 1983, p. 819; Chavel 1993 (II.2.12); Wang 2024, Def. 2.2)


(terminology) Historically, by the “first fundamental form” authors used to refer to the pullback of the metric tensor along an immersion. While this usage is no longer practiced, the term “second fundamental form” for (6) has become standard.

(cf. Chavel 1993, Rem. II..1; Lee 2018, pp. 227)

In much of the literature the second fundamental form is alternatively expressed instead in terms of covariant derivatives of and along tangential vector fields (e.g.: Chavel 1993 §II.2; Baird & Wood 2003, §3.2; Lee 2018, pp. 225), via the following proposition

(which is well-known but seems hard to cite explicitly, cf. maybe Willmore 1996, p. 126):


The second fundamental form obtained from a local Darboux coframe field as in (5), when regarded as a tensor, namely as a morphism of vector bundles from the tensor product of the tangent bundle of Σ\Sigma to the normal bundle of XX relative to ϕ\phi

:TΣTΣN ϕX, {Ⅱ} \;\colon\; T\Sigma \otimes T \Sigma \xrightarrow{\;} N_\phi X \,,

is given by covariant derivatives of vector fields as follows:

(7)(v,w)= v Xdϕ(w)dϕ( v Σw). {Ⅱ}(v,w) \;=\; \nabla^X_{v} \, \mathrm{d}\phi(w) \;-\; \mathrm{d}\phi\big( \nabla^\Sigma_v \, w \big) \,.


Let (v a) a=1 dimX(v_a)_{a = 1}^{dim X} be a local Darboux frame (1) on XX, so that (v a) a=1 dimΣ(v_a)_{a =1}^{dim \Sigma} is a local frame on Σ\Sigma (called the tangential frame vectors, the others being the transversal ones).

It follows that pushforward of vector field in this Darboux basis is just the canonical injection

(8)dϕ σ:T σΣ T ϕ(σ)X v a(σ) v a(σ) \begin{array}{rcc} \mathrm{d}\phi_\sigma \;\colon\; T_\sigma \Sigma &\xrightarrow{\phantom{--}}& T_{\phi(\sigma)} X \\ v_a(\sigma) &\mapsto& v_a(\sigma) \end{array}

(shown for any σΣ\sigma \in \Sigma inside the given chart).

Remembering that for tangential b 1,b 2b_1, b_2 we set

(9)ϕ *Ω b 1 a b 2{ω b 1 a b 2 tangentiala b 1b 2 a transversala \phi^\ast \Omega_{b_1}{}^a{}_{b_2} \;\equiv\; \left\{ \begin{array}{ll} \omega_{b_1}{}^a{}_{b_2} & \text{tangential}\; a \\ {Ⅱ}^a_{b_1 b_2} & \text{transversal}\; a \end{array} \right.

this makes it immediate that

(10) b 1 Xdϕ(v b 2)dϕ( b 1 Σv b 2) = b 1 Xv b 2 b 1 Σv b 2 =ϕ *Ω b 1 a b 2v aω b 1 a b 2v a =II b 1b 2 av a. \begin{array}{l} \nabla^X_{b_1} \mathrm{d}\phi(v_{b_2}) - \mathrm{d}\phi\big( \nabla^\Sigma_{b_1} v_{b_2} \big) \\ \;=\; \nabla^X_{b_1} v_{b_2} - \nabla^\Sigma_{b_1} v_{b_2} \\ \;=\; \phi^\ast \Omega_{b_1}{}^{a}{}_{b_2} v_a - \omega_{b_1}{}^a{}_{b_2} v_a \\ \;=\; \text{II}^a_{b_1 b_2} v_a \mathrlap{\,.} \end{array}

In words, the computation (10) shows that the covariant derivative on XX of and along tangent vectors to Σ\Sigma is that on Σ\Sigma plus the contribution of the second fundamental form, hence their difference extracts the latter.


(further alternative expressions)
As the proof (10) of Prop. shows, the second fundamental form extracts the normal component of the ambient covariant derivative of and along tangential vectors:

II(v,w)=( v Xw) . \text{II}(v,w) \;=\; (\nabla^X_v w)^\perp \,.

In this form it appears for instance in Kobayashi & Nomizu 1963 §VII.3, Chavel 1993 (II.2.2); the relation to (7) is made explicit by Baird & Wood 2003 Def. 3.2.3, cf also Willmore 1996, p. 126.

Alternatively, the pushforward of vector fields dϕ\mathrm{d}\phi may be understood as a section of the tensor product vector bundle

dϕΓ(TΣ *ϕ *TX), \mathrm{d}\phi \;\in\; \Gamma\big( T\Sigma^\ast \otimes \phi^\ast T X \big) \,,

whence the expression (7) may be understood as the induced covariant derivative on that tensor bundle

(dϕ)(v,w) v Xdϕ(w)dϕ( v Σw). (\nabla \mathrm{d}\phi)(v,w) \;\coloneqq\; \nabla^X_{v} \mathrm{d}\phi(w) \;-\; \mathrm{d}\phi\big( \nabla^\Sigma_v w \big) \,.

In this sense the second fundamental form is simply expressed as

=dϕ. {Ⅱ} \;=\; \nabla \mathrm{d}\phi \,.

This is how it appears in Eells & Sampson 1964 p. 123, Baird & Wood 2003 (3.2.1).

Finally, one may write out the covariant derivatives in a coordinate chart in terms of the Christoffel symbols L l m nL_{l}{}^m{}_n for the Levi-Civita connection on XX, and those Γ i l j\Gamma_i{}^l{}_j of the pullback connection to Σ\Sigma:

(11) ij k = i Xdϕ( j)dϕ( i Σ j) = i jϕ l+L m l n( iϕ m)( jϕ n)Γ i k j kϕ l = i jϕ lΓ i k j kϕ l+L m l n( iϕ m)( jϕ n). \begin{array}{l} {Ⅱ}^k_{i j} \\ \;=\; \;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\; \nabla^X_{\partial_i} \mathrm{d}\phi( \partial_j ) \;\;\;\;\;\;\;\;\;\;\;\; - \mathrm{d}\phi\big( \nabla^\Sigma_{\partial_i} \partial_j \big) \\ \;=\; \overbrace{ \partial_i \partial_j \phi^l + L_{m}{}^l{}_n (\partial_i \phi^m) (\partial_j \phi^n) } - \overbrace{ \Gamma_{i}{}^k{}_j \partial_k \phi^l } \\ \;=\; \partial_i \partial_j \phi^l - \Gamma_{i}{}^k{}_j \partial_k \phi^l + L_{m}{}^l{}_n (\partial_i \phi^m) (\partial_j \phi^n) \,. \end{array}

In this guise the second fundamental form was originally given in Eells & Sampson 1964 (p. 118 & 123 with p. 111) following Eisenhart 1925, §43, reviewed by Baird & Wood 2003 (3.2.2).


(totally geodesic immersions)
If the fundamental form Ⅱ of a Riemannian immersion ϕ:,ΣX\phi \colon, \Sigma \to X vanishes, then geodesics in Σ\Sigma are taken by ϕ\phi to geodesics in XX, hence such ϕ\phi is said to be totally geodesic [Eells & Sampson 1964 p. 126, Baird & Wood 2003 Def. 3.2.1].

Tension field


The tension field of the immersion ϕ\phi is the contraction (trace) of the second fundamental form:

τ kη ab ab k=g ij ij k. \tau^k \;\coloneqq\; \eta^{a b} \, {Ⅱ}^k_{a b} \;=\; g^{i j} \, {Ⅱ}^k_{i j} \,.

From (11) we have immediately the coordinate expression

τ k=η ab ab k=g ij ij k =g ij( i jϕ lΓ i k j kϕ l)Δϕ l+g ij( iϕ m)( jϕ n)L m l n, \begin{array}{l} \tau^k \;=\; \eta^{a b} \, {Ⅱ}^k_{a b} \;=\; g^{i j} \, {Ⅱ}^k_{i j} \\ \;=\; \underset {\Delta \phi^l} { \underbrace{ g^{i j} \big( \partial_i \partial_j \phi^l - \Gamma_{i}{}^k{}_j \partial_k \phi^l \big) } } + g^{i j} (\partial_i \phi^m) (\partial_j \phi^n) L_{m}{}^l{}_n \,, \end{array}

where “Δ\Delta” denotes the Laplace operator on Σ\Sigma.

(Eells & Sampson 1964 (5), Baird & Wood 2003 Def. 3.2.4)

Harmonic immersions


(harmonic equation)
The vanishing of the tension field τ\tau (Def. ) characterizes Riemannian immersion which are harmonic maps.

(Eells & Sampson 1964 pp. 116, Baird & Wood 2003 Thm. 3.3.3)


Last revised on July 1, 2024 at 11:00:32. See the history of this page for a list of all contributions to it.