---
title: Generalised Second Fundamental Form
url: https://www.emergentmind.com/topics/generalised-second-fundamental-form
type: topic
---

# Generalised Second Fundamental Form

Generalised second fundamental form denotes a family of extensions of the classical extrinsic tensor \(\mathrm{II}(X,Y)=(\nabla_XY)^\perp\). In the cited literature, the extension is achieved in several mathematically distinct ways: by replacing an embedded submanifold with an integrable or non-integrable distribution, by passing from Riemannian geometry to exact Courant algebroids, by interpreting second-order variation in Hodge-theoretic period maps, or by recasting the second fundamental form as a quadratic map in Lorentzian signature [1506.01697], [1808.06704], [2507.12362], [2005.03737]. Across these settings, the common role is to measure second-order failure of tangentiality or first-order stability, but symmetry, codomain, and even the meaning of “normal direction” depend on the ambient formalism.

## 1. Classical prototype and the scope of generalisation

The classical model remains the reference point. For an embedded submanifold \(S\subset (M,g)\) with Levi-Civita connection \(\nabla\) and orthogonal splitting \(T_xM=T_xS\oplus N_x\), the second fundamental form is
\[
\mathrm{II}(X,Y):=(\nabla_XY)^\perp.
\]
In the Riemannian case it is symmetric, and the Gauss formula reads
\[
\nabla_XY=\nabla^S_XY+\mathrm{II}(X,Y).
\]
This model is explicitly taken as the starting point for extensions to distributions with torsion and to non-integrable distributions [1506.01697], [1808.06704].

What changes in the generalised setting is the mechanism that replaces orthogonal normal projection. For integrable distributions one fixes a complementary distribution and projects transversely; for non-integrable distributions one compares the ambient Levi-Civita connection with the intrinsic projected connection; for exact Courant algebroids one replaces the tangent bundle by the pullback Courant algebroid and projects onto a generalised normal line; for period maps one interprets the second fundamental form as the quadratic part of the variation, taking values in a quotient of the ambient classifying-space tangent bundle [2507.12362], [2005.03737].

A persistent misconception is that a generalised second fundamental form should remain symmetric. The literature shows otherwise. In the presence of torsion, the skew part is exactly the transverse torsion; for non-integrable distributions, the skew part measures failure of involutivity; and in Lorentzian signature degeneracies introduce additional invariants beyond the usual quartet \(|H|^2\), \(K\), \(K_N\), and \(A\) [1506.01697], [1808.06704], [1503.06225].

## 2. Integrable distributions with linear connection and torsion

A direct extension of the classical submanifold tensor is developed for a smooth manifold \(M\) with a linear connection \(\nabla\), not necessarily torsion-free, together with an integrable distribution \(D\) of constant dimension \(p\) and a fixed complementary distribution \(D'\) such that \(\mathfrak X(M)=D\oplus D'\) [1506.01697]. If \((\cdot)^-\) denotes projection to \(D'\), the generalised second fundamental form is defined by
\[
\mathrm{II}_D(X,Y):=(\nabla_XY)^-, \qquad X,Y\in D.
\]

This form is genuinely bilinear on \(D\). Its central structural identity is
\[
\mathrm{II}_D(X,Y)-\mathrm{II}_D(Y,X)=T(X,Y)^-,
\]
where
\[
T(X,Y):=\nabla_XY-\nabla_YX-[X,Y]
\]
is the torsion tensor. The antisymmetric part is therefore not an error term but the transverse torsion itself. When the connection is torsion-free, \(\mathrm{II}_D\) becomes symmetric, recovering the classical behavior.

The same framework is related to the shape map of the connection. If \(A_Z\) is defined by
\[
A_Z(\xi):=\left.\frac{d}{dt}\right|_{t=0} P_t^{-1}\big((S_t)_*\xi\big),
\]
then
\[
A_Z(\xi)=\nabla_\xi Z+T(Z_x,\xi),
\]
and for the distribution one has
\[
\mathrm{II}_D(X,Y)=A_X(Y)^-.
\]
Thus the generalised second fundamental form is the transverse part of the shape map applied to tangent vectors. In codimension one, with \(D'=\operatorname{Sp}\{N\}\) determined by an integrable \(1\)-form \(\theta\) satisfying \(\theta(D)=0\) and \(\theta(N)=1\), the form becomes scalar-valued:
\[
h_D(X,Y):=\theta(\mathrm{II}_D(X,Y)), \qquad \mathrm{II}_D(X,Y)=h_D(X,Y)N.
\]
Its skew part is
\[
h_D(X,Y)-h_D(Y,X)=\theta(T(X,Y)).
\]
The codimension-one theory therefore generalises the ordinary shape-operator formalism while making explicit that torsion obstructs the self-adjoint diagonalisation familiar from hypersurface theory.

The significance of this formulation is conceptual as much as technical. It isolates precisely which part of the classical theory depends on metric orthogonality and which part depends only on a splitting and a connection. The result is a bilinear extrinsic tensor for foliated geometry that survives beyond the Levi-Civita setting.

## 3. Non-integrable distributions, curvature comparison, and constrained dynamics

For a regular distribution \(\mathcal D\subset TM\) on a Riemannian manifold \((M,g)\), a different extension treats \(\mathcal D\) itself as the analogue of a submanifold without assuming involutivity [1808.06704]. Let \(P:TM\to\mathcal D\) be the orthogonal projection associated to \(TM=\mathcal D\oplus\mathcal D^\perp\). The intrinsic connection on \(\Gamma(\mathcal D)\) is
\[
\nabla_X^{\mathcal D}Y:=P(\nabla_XY),
\]
and the second fundamental form of the distribution is
\[
B(X,Y):=\nabla_XY-\nabla_X^{\mathcal D}Y=P^\perp(\nabla_XY), \qquad X,Y\in\Gamma(\mathcal D).
\]
Equivalently,
\[
\nabla_XY=\nabla_X^{\mathcal D}Y+B(X,Y).
\]

In this setting, the symmetric and antisymmetric parts of \(B\) encode different geometric defects. The paper proves
\[
\mathcal D \text{ is involutive}\quad\Longleftrightarrow\quad B \text{ is symmetric}\quad\Longleftrightarrow\quad B^A=0,
\]
and
\[
\mathcal D \text{ is totally geodesic}\quad\Longleftrightarrow\quad B^S=0.
\]
Thus \(B^A\) measures non-integrability, while \(B^S\) measures failure of total geodesy. For a curve \(c(t)\) tangent to \(\mathcal D\),
\[
\nabla_{\dot c}\dot c=\nabla_{\dot c}^{\mathcal D}\dot c+B(\dot c,\dot c),
\]
so the normal curvature is \(\|B(\dot c,\dot c)\|\), depending only on the symmetric part.

The same formalism yields curvature comparison. The curvature of \(\nabla^{\mathcal D}\) differs from the projected ambient curvature by terms involving \(B\) and by an additional contribution from non-involutivity. In the involutive case the extra bracket term disappears and one recovers the standard Gauss equation for a foliation leaf. In the genuinely non-holonomic case, the “Gauss formula” for sectional curvature acquires new terms that record the failure of Frobenius integrability.

This extension also has a mechanical interpretation. For constrained dynamics with \(\dot c(t)\in\mathcal D_{c(t)}\) and force field \(F\),
\[
\nabla_{\dot c}\dot c=F(c)+R, \qquad R\perp\mathcal D,
\]
and projection gives
\[
P^\perp(\nabla_{\dot c}\dot c)=B(\dot c,\dot c)=P^\perp(F(c))+R.
\]
Here the generalised second fundamental form furnishes the normal acceleration generated by the constraint, so it functions as the extrinsic curvature term in nonholonomic mechanics.

## 4. Exact Courant algebroids and exterior generalised geometry

In exterior generalised geometry, the classical hypersurface package is transferred to an exact Courant algebroid \(E\to M\) equipped with its pairing, Courant bracket, and anchor \(\rho:E\to TM\) [2507.12362]. For an immersion \(\iota:N\hookrightarrow M\), the relevant pulled-back object is not the naive vector bundle pullback but the pullback Courant algebroid \(\iota^!E\). Generalised metrics and divergence operators are pulled back to \(\iota^!E\), and in the hypersurface case one obtains a canonical generalised normal line.

If \(N\) is a hypersurface and \(\nu\) is a chosen generalised unit normal, then
\[
E|_N\cong \iota^!E\oplus \langle \nu\rangle.
\]
With \(P_T\) and \(P_N\) the tangential and normal projections and \(\nabla\) the generalised metric connection, the generalised second fundamental form is defined for \(u,v\in\Gamma(\iota^!E)\) by
\[
\mathrm{II}(u,v):=P_N(\nabla_uv),
\]
or equivalently
\[
\mathrm{II}(u,v)=\langle \nabla_uv,\nu\rangle.
\]
This is formally identical to the classical formula, but the arguments are sections of the pullback Courant algebroid rather than vector fields, so tangent and cotangent data enter simultaneously.

The associated generalised mean curvature is the trace
\[
H:=\operatorname{tr}_{\iota^!E}\mathrm{II}=\sum_a \varepsilon_a\,\mathrm{II}(e_a,e_a),
\]
for a local generalised orthonormal frame \(\{e_a\}\). The paper then proves generalised Gauß and Codazzi equations. Schematically,
\[
R^\top(u,v,w,z)=R(u,v,w,z)+\langle \mathrm{II}(u,w),\mathrm{II}(v,z)\rangle-\langle \mathrm{II}(u,z),\mathrm{II}(v,w)\rangle,
\]
and
\[
(\nabla_u^\top \mathrm{II})(v,w)-(\nabla_v^\top \mathrm{II})(u,w)=\big(R(u,v)w\big)^\perp.
\]
These are the generalised analogues of the classical structure equations.

The split case makes the relationship to ordinary geometry explicit. When \(E\cong TM\oplus T^*M\) and the generalised metric comes from an ordinary metric \(g\) together with a \(B\)-field, the generalised second fundamental form reduces to the usual one after untwisting by the \(B\)-field. The theory therefore extends, rather than replaces, semi-Riemannian hypersurface geometry. Its stated applications include the constraint equations for the initial value formulation of the generalised Einstein equations, a generalised geometry version of the fundamental theorem for hypersurfaces, and restriction results for generalised Kähler and hyper-Kähler structures.

## 5. Hodge-theoretic and algebraic-geometric avatars

In Hodge theory, the second fundamental form is the quadratic part of a period map rather than a normal projection of \(\nabla_XY\). For the primitive cohomology of a smooth projective hypersurface \(X=\{f=0\}\subset \mathbf P^{n+1}\), tangent vectors to the universal hypersurface family are identified with degree-\(d\) elements of the Jacobian ring \(R=S/J\), and the infinitesimal variation of Hodge structure acts by multiplication in \(R\) [2005.03737]. In this description, the second fundamental form
\[
II:S^2T\to T_HD/T
\]
is given by
\[
II(u,v)=u\circ v,
\]
with \(\circ\) the ordinary composition of linear maps. Equivalently, it is determined by the product \(uv\) in the Jacobian ring. The paper further states that this implies the image lands in \(\mathfrak g^{-2,2}\), so the variation of Hodge structure satisfies new second-order partial differential equations beyond Griffiths transversality.

For the Torelli map \(j:\mathcal M_g\to\mathcal A_g\), the dual second fundamental form
\[
\rho=II^\vee:I_2(K_C)\to S^2H^1(C,T_C)
\]
admits a geometric realisation on \(C\times C\). The paper introduces an intrinsic section
\[
\hat\eta\in H^0\bigl(C\times C,K_{C\times C}(2\Delta)\bigr)
\]
and proves that \(\rho\) is the restriction of multiplication by \(\hat\eta\), namely
\[
\rho(Q)=Q\cdot \hat\eta.
\]
For rank \(4\) quadrics in \(I_2(K_C)\), this description gives lower bounds on \(\operatorname{rank}\rho(Q)\) and leads to improved upper bounds on the dimensions of totally geodesic subvarieties of \(\mathcal A_g\) generically contained in the Torelli locus [1907.11407].

The Torelli second fundamental form also encodes the hierarchy of higher even Gaussian maps. Using higher Schiffer variations \(\xi_p^n\), the paper proves that if \(Q\in \ker(\mu_{2n-2})\), then \(II^*(Q)\) vanishes on \(\langle \xi_p^1,\dots,\xi_p^{n-1}\rangle\), while for \(Q\in \ker(\mu_{2n-2})\setminus \ker(\mu_{2n})\),
\[
II^*(Q)(\xi_p^n)(\xi_p^n)=c_n\,\mu_{2n}(Q)(p)\neq 0
\]
for an explicit nonzero constant \(c_n\). As consequences, \(\mu_{6g-6}\) is injective for every non-hyperelliptic curve, all even Gaussian maps \(\mu_{2k}\) vanish identically for \(k>3g-3\), and rank estimates are obtained in the intermediate range [2208.14794].

An analogous lifting phenomenon occurs for Prym maps. In the unramified case, the second fundamental form of the Prym map lifts the second Gaussian map \(\mu_A\) of the Prym-canonical line bundle:
\[
m\circ II=\mu_A,
\]
and for a Schiffer variation \(\xi_P\),
\[
\mu_A(Q)(P)=-\,II(Q)(\xi_P\odot \xi_P).
\]
Degeneration to Prym-canonical binary curves is then used to prove surjectivity of \(\mu_A\) for a general Prym curve of genus \(g\ge 20\) [1105.4472]. In the ramified case, the dual Prym second fundamental form \(\rho_P\) is expressed through the Torelli second fundamental form \(\widetilde\rho\) of the covering curve:
\[
\rho_P=p\circ (\mu^\ast \widetilde\rho)\big|_G,
\]
and pointwise
\[
\rho_P(Q)(v_1\odot v_2)=\widetilde\rho(\pi Q)(v_1\odot v_2).
\]
The same paper proves that \(\rho_P\) lifts the second Gaussian map in the sense that
\[
m\circ \rho_P=-4\pi i\,\mu_2,
\]
and uses rank estimates for \(\rho_P(Q)\) to bound the dimensions of totally geodesic submanifolds and Shimura subvarieties in the Prym locus [1812.07402].

A related projective-geometric development studies the projective second fundamental form \(II_{X,x}\) of a codimension-two submanifold \(X^n\subset \mathbf P^{n+2}\), where \(II_{X,x}\) is a pencil of quadrics on \(T_x(X)\). For nonsingular intersections of two quadrics, the moduli map
\[
\mu^X:X^\circ\to MPQ,\qquad x\mapsto [II_{X,x}]
\]
is dominant, giving a negative answer to the Griffiths–Harris question in that setting, while the refined map
\[
\widetilde\mu^X:x\mapsto [II_{X,x},P_x],\qquad P_x=\ker(d_x\mu^X),
\]
does determine \(X\) up to projective automorphism when \(n>4\) [1706.00551]. Here the second fundamental form is not merely a pointwise tensor; it is organised into moduli and infinitesimal-moduli data.

## 6. Lorentzian, relativistic, and asymptotic extensions

For Lorentzian surfaces in \(\mathbb R^{2,2}\), the second fundamental form is studied as a quadratic map
\[
II_p:T_pM\to N_pM,
\]
equivalently as a quadratic map \(q:\mathbb R^{1,1}\to\mathbb R^{1,1}\) up to the left and right action of \(SO(1,1)\) [1503.06225]. The four classical invariants are the mean curvature vector \(H\), Gauss curvature \(K\), normal curvature \(K_N\), and the fourth invariant
\[
A:=\det Q_{[II]}.
\]
The associated curvature hyperbola is
\[
\mathcal H=\{q(v): |v|^2=1\},
\]
with center \(H\). In degenerate cases additional invariants \(a,b\) are required. The paper proves, in particular, that a point is quasi-umbilic if and only if there is a double lightlike asymptotic direction. This Lorentzian formulation shows that once the tangent and normal bundles both have signature \((1,1)\), the correct generalised object is a quadratic map together with its orbit data, not merely a symmetric bilinear tensor.

In general relativity, the second fundamental form \(K\) of a closed spacelike hypersurface is decomposed \(L^2\)-orthogonally into gauge and transverse-traceless parts. Writing
\[
K_0:=K-\frac{H}{n}g,
\]
the paper proves
\[
K_0=S\theta+q_{\mathrm{TT}},
\]
where \(S\theta=\delta^\ast\theta-\frac1n(\delta\theta)g\), \(q_{\mathrm{TT}}\) is transverse-traceless, and the two terms are \(L^2\)-orthogonal. Applying \(S^\ast\) yields
\[
S^\ast S\theta=S^\ast K_0.
\]
In the vacuum constraint equations, if \(H\) is constant, then \(S^\ast S\theta=0\), so \(\theta^\#\) is conformal Killing and
\[
K=\frac{H}{n}g+q_{\mathrm{TT}}.
\]
The paper explicitly interprets this as a refined splitting of the “generalized second fundamental form” into a pure gauge part and a transverse-traceless part [2406.04760].

A different extension concerns asymptotic control rather than a new tensorial codomain. For a complete immersed manifold \(M\) in an ambient space with a pole, the immersion has tamed second fundamental form if
\[
a(M):=\lim_{i\to\infty}\sup_{x\in M\setminus K_i}\bigl(P_M(x)\,|\alpha(x)|\bigr)<1,
\]
and strongly tamed second fundamental form if, for some \(\varepsilon>0\),
\[
\lim_{i\to\infty}\sup_{x\in M\setminus K_i}\bigl(P_M(x)^{1+\varepsilon}\,|\alpha(x)|\bigr)<1.
\]
In this regime, the cited paper derives properness, finite topology, absence of critical points of the extrinsic distance outside a compact set, equivalence between linear extrinsic perimeter growth and quadratic extrinsic area growth for surfaces in \(\mathbb R^n\), a generalized Chern–Osserman inequality, rigidity for nonnegative curvature surfaces, and vanishing fundamental tone under radial curvature decay assumptions [1412.4929].

Weak compactness theory delineates the boundary of the terminology. One paper on immersions with bounded second fundamental form states explicitly that it does not define a varifold-style generalized second fundamental form itself; instead it develops compactness for \(W^{2,p}\)-immersions with \(p>m\) under \(L^p\)-bounds on \(A\), local graph representations, and lower semicontinuity of \(\|A\|_{L^p}\) [1201.4562]. This suggests an important distinction: some literatures generalise the object, while others keep the classical second fundamental form and generalise only the regularity or compactness framework in which it is controlled.

Taken together, these developments show that “generalised second fundamental form” is best understood as a structural principle rather than a single formula. The principle is the extraction of second-order extrinsic data after replacing the classical tangent-normal decomposition by whatever notion of tangential versus transverse direction the ambient category provides—complementary distributions, projected intrinsic connections, Courant algebroid splittings, Hodge bundles inside classifying spaces, Lorentzian quadratic-map orbits, or gauge/TT decompositions of spacelike extrinsic curvature.

Source: https://www.emergentmind.com/topics/generalised-second-fundamental-form