Approximate Second Fundamental Form

• Approximate second fundamental form is a collection of methods that replace direct pointwise curvature control with asymptotic, L², or variational approximations in submanifold geometry.
• These techniques impact scalar curvature, spectral bounds, and topological rigidity, providing alternatives to classical tensor evaluations in diverse geometric settings.
• Key approaches include tamed decay conditions, orthogonal decompositions, and kernel-based varifold regularizations which extend to discrete point-cloud analyses.

Approximate second fundamental form denotes a family of constructions that replace direct pointwise control of the classical second fundamental form by asymptotic bounds, $L^2$-variational principles, orthogonal decompositions, mollified varifold tensors, higher Gaussian maps, or first-order deformation formulas. For an isometric immersion $\iota:(M^m,g)\to (N^n,\bar g)$, the classical second fundamental form is the normal-valued symmetric bilinear form

$A(X,Y)=(\bar\nabla_X d\iota(Y))^\perp,$

equivalently $\bar\nabla_XY=\nabla_XY+A(X,Y)$; its trace is the mean curvature vector $H=\operatorname{tr}_gA$, and $|A|^2$ or $\|\alpha\|^2$ is the basic extrinsic curvature density. The literature surveyed here uses this tensor in several non-equivalent approximation regimes, ranging from tamed decay at infinity to explicit regularizations on point clouds (0805.0323, Simanca, 2014, Buet et al., 2019, Stepanov et al., 2024).

1. Classical tensor, trace splittings, and weak replacements

In the Riemannian immersion setting, the second fundamental form is paired with the shape operator by

$$\langle A(X,Y),\nu\rangle_{\bar g}=\langle A_\nu X,Y\rangle_g,\qquad A_\nu X=-(\bar\nabla_X\nu)^\top,$$

and the Gauss equation decomposes ambient curvature into intrinsic curvature plus quadratic terms in $A$. For tangent vector fields $X,Y,Z,W$,

$\iota:(M^m,g)\to (N^n,\bar g)$0

while

$\iota:(M^m,g)\to (N^n,\bar g)$1

These identities explain why approximation schemes for the second fundamental form immediately affect scalar curvature, spectral estimates, and variational energies (Simanca, 2014).

On a closed spacelike hypersurface $\iota:(M^m,g)\to (N^n,\bar g)$2 in a Lorentzian spacetime, the second fundamental form is denoted $\iota:(M^m,g)\to (N^n,\bar g)$3, with mean curvature $\iota:(M^m,g)\to (N^n,\bar g)$4 and trace-free part

$\iota:(M^m,g)\to (N^n,\bar g)$5

The decomposition theory in this setting singles out TT-tensors, namely symmetric, divergence-free, traceless $\iota:(M^m,g)\to (N^n,\bar g)$6-tensors, and identifies the longitudinal part through the conformal Killing operator. This already treats approximation as projection onto geometrically distinguished subspaces rather than as pointwise estimation (Stepanov et al., 2024).

In the varifold setting, the tensor is replaced by distributions built from tangent projections. For a $\iota:(M^m,g)\to (N^n,\bar g)$7-varifold $\iota:(M^m,g)\to (N^n,\bar g)$8, the $\iota:(M^m,g)\to (N^n,\bar g)$9-linear variations are

$A(X,Y)=(\bar\nabla_X d\iota(Y))^\perp,$0

If each $A(X,Y)=(\bar\nabla_X d\iota(Y))^\perp,$1 is a Radon measure, then

$A(X,Y)=(\bar\nabla_X d\iota(Y))^\perp,$2

and the weak second fundamental form $A(X,Y)=(\bar\nabla_X d\iota(Y))^\perp,$3 is defined as the unique solution of

$A(X,Y)=(\bar\nabla_X d\iota(Y))^\perp,$4

This replaces the smooth tensor by a measure-theoretic curvature object suited to nonsmooth sets and discrete data (Buet et al., 2019).

2. Tamed second fundamental form as asymptotic approximation

A complete immersion $A(X,Y)=(\bar\nabla_X d\iota(Y))^\perp,$5 into a complete ambient manifold with $A(X,Y)=(\bar\nabla_X d\iota(Y))^\perp,$6 has tamed second fundamental form when the asymptotic quantity

$A(X,Y)=(\bar\nabla_X d\iota(Y))^\perp,$7

satisfies $A(X,Y)=(\bar\nabla_X d\iota(Y))^\perp,$8, where $A(X,Y)=(\bar\nabla_X d\iota(Y))^\perp,$9 and $\bar\nabla_XY=\nabla_XY+A(X,Y)$0 are the standard comparison functions. For $\bar\nabla_XY=\nabla_XY+A(X,Y)$1, the taming condition becomes the eventual inequality

$\bar\nabla_XY=\nabla_XY+A(X,Y)$2

while for $\bar\nabla_XY=\nabla_XY+A(X,Y)$3 it becomes

$\bar\nabla_XY=\nabla_XY+A(X,Y)$4

outside a large compact set (0805.0323).

The geometric content of this definition is extrinsic convexity at infinity. For $\bar\nabla_XY=\nabla_XY+A(X,Y)$5, the choice $\bar\nabla_XY=\nabla_XY+A(X,Y)$6 yields lower Hessian bounds for $\bar\nabla_XY=\nabla_XY+A(X,Y)$7; for $\bar\nabla_XY=\nabla_XY+A(X,Y)$8, the choice $\bar\nabla_XY=\nabla_XY+A(X,Y)$9 gives the corresponding hyperbolic barrier. Integrating the resulting inequalities along minimizing geodesics forces extrinsic distance to grow at least linearly, or in the hyperbolic barrier sense. From this one obtains the main theorem: if $H=\operatorname{tr}_gA$0, then $H=\operatorname{tr}_gA$1 is compact when $H=\operatorname{tr}_gA$2 is compact, $H=\operatorname{tr}_gA$3 is proper when $H=\operatorname{tr}_gA$4 is complete noncompact, and $H=\operatorname{tr}_gA$5 has finite topology when $H=\operatorname{tr}_gA$6 is Hadamard (0805.0323).

The same asymptotic control has a spectral consequence. If $H=\operatorname{tr}_gA$7 is Hadamard with $H=\operatorname{tr}_gA$8, then for any $H=\operatorname{tr}_gA$9 with $|A|^2$0 there exist an integer $|A|^2$1 and a constant $|A|^2$2 such that

$|A|^2$3

Hence the fundamental tone is an obstruction: if $|A|^2$4 exceeds this model-space bound, $|A|^2$5 cannot be realized as a submanifold with tamed second fundamental form in such a Hadamard manifold. Examples with $|A|^2$6 include the Jorge–Meeks class of complete submanifolds of $|A|^2$7 homeomorphic to a compact manifold punctured at finitely many points and having a well-defined normal at infinity; by contrast, the positive fundamental tone of certain bounded or cylindrically bounded minimal surfaces suggests non-realizability under taming (0805.0323).

3. $|A|^2$8-variational approximation and canonical representatives

A different approximation regime is global and variational. For an isometric immersion $|A|^2$9, the squared $\|\alpha\|^2$0-norms

$\|\alpha\|^2$1

together with

$\|\alpha\|^2$2

measure extrinsic bending in integral form. The first variation of $\|\alpha\|^2$3 is governed by the Euler–Lagrange vector field

$\|\alpha\|^2$4

and critical points satisfy $\|\alpha\|^2$5. In codimension one and Einstein ambient manifolds, the equation reduces to a scalar PDE for the mean curvature $\|\alpha\|^2$6, and in a space form of curvature $\|\alpha\|^2$7 it becomes

$\|\alpha\|^2$8

for $\|\alpha\|^2$9-critical hypersurfaces (Simanca, 2014).

This variational framework defines canonical representatives of homology classes. If $$\langle A(X,Y),\nu\rangle_{\bar g}=\langle A_\nu X,Y\rangle_g,\qquad A_\nu X=-(\bar\nabla_X\nu)^\top,$$0, a canonical representative is an embedded $$\langle A(X,Y),\nu\rangle_{\bar g}=\langle A_\nu X,Y\rangle_g,\qquad A_\nu X=-(\bar\nabla_X\nu)^\top,$$1 minimizing $$\langle A(X,Y),\nu\rangle_{\bar g}=\langle A_\nu X,Y\rangle_g,\qquad A_\nu X=-(\bar\nabla_X\nu)^\top,$$2 over $$\langle A(X,Y),\nu\rangle_{\bar g}=\langle A_\nu X,Y\rangle_g,\qquad A_\nu X=-(\bar\nabla_X\nu)^\top,$$3 and, among those minimizers, having the smallest volume. Several model examples are explicit. In the principal fibration

$$\langle A(X,Y),\nu\rangle_{\bar g}=\langle A_\nu X,Y\rangle_g,\qquad A_\nu X=-(\bar\nabla_X\nu)^\top,$$4

with the left-invariant metrics $$\langle A(X,Y),\nu\rangle_{\bar g}=\langle A_\nu X,Y\rangle_g,\qquad A_\nu X=-(\bar\nabla_X\nu)^\top,$$5, each fiber is totally geodesic, $$\langle A(X,Y),\nu\rangle_{\bar g}=\langle A_\nu X,Y\rangle_g,\qquad A_\nu X=-(\bar\nabla_X\nu)^\top,$$6, represents the generator of $$\langle A(X,Y),\nu\rangle_{\bar g}=\langle A_\nu X,Y\rangle_g,\qquad A_\nu X=-(\bar\nabla_X\nu)^\top,$$7, has volume $$\langle A(X,Y),\nu\rangle_{\bar g}=\langle A_\nu X,Y\rangle_g,\qquad A_\nu X=-(\bar\nabla_X\nu)^\top,$$8, and under $$\langle A(X,Y),\nu\rangle_{\bar g}=\langle A_\nu X,Y\rangle_g,\qquad A_\nu X=-(\bar\nabla_X\nu)^\top,$$9 with $A$0, any minimizer is isometric to a fiber. In $A$1, any complex submanifold is $A$2-critical, $A$3-critical, and critical for $A$4; moreover $A$5 is intrinsic, and for a complex curve $A$6,

$A$7

The minimization of $A$8 within $A$9 yields a variational proof of the Kronheimer–Mrowka genus bound. In $X,Y,Z,W$0, the diagonal $X,Y,Z,W$1 is totally geodesic, Kähler, has $X,Y,Z,W$2, and is the canonical representative of the $X,Y,Z,W$3-class (Simanca, 2014).

The variational perspective treats the second fundamental form approximately in an $X,Y,Z,W$4 sense: minimizers need not vanish pointwise, but they are the canonical low-bending representatives singled out by the Euler–Lagrange system, bubble analysis, and homological constraints. This suggests an approximation principle in which the geometry of $X,Y,Z,W$5 is optimized globally rather than estimated locally.

4. Orthogonal decompositions and the Ahlfors-Laplacian approach

For closed manifolds, approximation can be phrased as an $X,Y,Z,W$6-orthogonal splitting of the second fundamental form. On a closed $X,Y,Z,W$7, $X,Y,Z,W$8,

$X,Y,Z,W$9

and the second factor is the TT space. Consequently, for the second fundamental form $\iota:(M^m,g)\to (N^n,\bar g)$00 of a closed spacelike hypersurface,

$\iota:(M^m,g)\to (N^n,\bar g)$01

or, after removing the trace,

$\iota:(M^m,g)\to (N^n,\bar g)$02

where

$\iota:(M^m,g)\to (N^n,\bar g)$03

The longitudinal/conformal part is therefore found by solving

$\iota:(M^m,g)\to (N^n,\bar g)$04

and the TT part is $\iota:(M^m,g)\to (N^n,\bar g)$05 (Stepanov et al., 2024).

The operator $\iota:(M^m,g)\to (N^n,\bar g)$06 is the Ahlfors Laplacian. The paper states

$\iota:(M^m,g)\to (N^n,\bar g)$07

equivalently

$\iota:(M^m,g)\to (N^n,\bar g)$08

and also

$\iota:(M^m,g)\to (N^n,\bar g)$09

It is formally self-adjoint, nonnegative, elliptic, and has kernel equal to the conformal Killing one-forms. Solving the elliptic equation on $\iota:(M^m,g)\to (N^n,\bar g)$10 yields the best $\iota:(M^m,g)\to (N^n,\bar g)$11 approximation of $\iota:(M^m,g)\to (N^n,\bar g)$12 by the image of the conformal Killing operator, with the TT component as the orthogonal residual (Stepanov et al., 2024).

This decomposition aligns directly with the vacuum constraint equations. Writing $\iota:(M^m,g)\to (N^n,\bar g)$13, the momentum constraint is

$\iota:(M^m,g)\to (N^n,\bar g)$14

Combining it with $\iota:(M^m,g)\to (N^n,\bar g)$15 yields an elliptic equation for $\iota:(M^m,g)\to (N^n,\bar g)$16; in the constant mean curvature case, $\iota:(M^m,g)\to (N^n,\bar g)$17, so $\iota:(M^m,g)\to (N^n,\bar g)$18, $\iota:(M^m,g)\to (N^n,\bar g)$19 is conformal Killing, and one may gauge away the longitudinal part, leaving

$\iota:(M^m,g)\to (N^n,\bar g)$20

For maximal hypersurfaces $\iota:(M^m,g)\to (N^n,\bar g)$21, the Hamiltonian constraint becomes

$\iota:(M^m,g)\to (N^n,\bar g)$22

with equality implying $\iota:(M^m,g)\to (N^n,\bar g)$23. In this setting, approximation means projection of $\iota:(M^m,g)\to (N^n,\bar g)$24 onto trace, TT, and longitudinal sectors that are orthogonal in $\iota:(M^m,g)\to (N^n,\bar g)$25 and geometrically adapted to the Einstein constraints (Stepanov et al., 2024).

5. Varifold regularization and point-cloud approximate second fundamental forms

The most literal notion of approximate second fundamental form in the surveyed literature arises from regularization of varifolds. Starting from the weak curvature data $\iota:(M^m,g)\to (N^n,\bar g)$26 and the averaged projector

$\iota:(M^m,g)\to (N^n,\bar g)$27

the weak second fundamental form is given explicitly by

$\iota:(M^m,g)\to (N^n,\bar g)$28

The associated Weingarten-type tensor is

$\iota:(M^m,g)\to (N^n,\bar g)$29

Regularization is then introduced through radial kernels $\iota:(M^m,g)\to (N^n,\bar g)$30, producing

$\iota:(M^m,g)\to (N^n,\bar g)$31

and the approximate second fundamental form

$\iota:(M^m,g)\to (N^n,\bar g)$32

For rectifiable varifolds with bounded variations, $\iota:(M^m,g)\to (N^n,\bar g)$33, $\iota:(M^m,g)\to (N^n,\bar g)$34, hence $\iota:(M^m,g)\to (N^n,\bar g)$35 and $\iota:(M^m,g)\to (N^n,\bar g)$36 almost everywhere; quantitative convergence estimates are also given for discrete approximations $\iota:(M^m,g)\to (N^n,\bar g)$37 (Buet et al., 2019).

The same framework extends to point clouds. For a point-cloud varifold $\iota:(M^m,g)\to (N^n,\bar g)$38, the orthogonal approximate tensor simplifies to explicit weighted sums. In codimension one, after projection onto a unit normal and restriction to tangent coordinates, the eigenvalues of the resulting matrix are the approximate principal curvatures, while the trace and determinant provide approximate mean and Gaussian curvature. The numerical tests on the dragon point cloud, a genus-$\iota:(M^m,g)\to (N^n,\bar g)$39 surface, and a cube with and without Gaussian noise exhibit stable curvature concentration on smooth and sharp features, with robustness improved by increasing neighborhood size (Buet et al., 2019).

A later generalization introduces operator-dependent approximations. For a linear operator field $\iota:(M^m,g)\to (N^n,\bar g)$40, the approximate mean curvature is defined by

$\iota:(M^m,g)\to (N^n,\bar g)$41

and the approximate second fundamental form becomes

$\iota:(M^m,g)\to (N^n,\bar g)$42

Under the natural kernel pair condition

$\iota:(M^m,g)\to (N^n,\bar g)$43

the theory recovers the classical mean curvature for several operator choices $\iota:(M^m,g)\to (N^n,\bar g)$44, and for unit-density integral varifolds with $\iota:(M^m,g)\to (N^n,\bar g)$45, $\iota:(M^m,g)\to (N^n,\bar g)$46, and $\iota:(M^m,g)\to (N^n,\bar g)$47, the corresponding $\iota:(M^m,g)\to (N^n,\bar g)$48 converges to the classical second fundamental form. In particular,

$\iota:(M^m,g)\to (N^n,\bar g)$49

The same regularized operators drive continuous and discrete point-cloud motions by approximate mean curvature, together with internal and external sphere-barrier principles when $\iota:(M^m,g)\to (N^n,\bar g)$50 (Sagueni, 8 Sep 2025).

6. Derived invariants, higher Gaussian maps, and mean-value rigidity

Approximation also appears through first-order deformation formulas. For a smooth hypersurface $\iota:(M^m,g)\to (N^n,\bar g)$51 with unit normal $\iota:(M^m,g)\to (N^n,\bar g)$52, the tensor $\iota:(M^m,g)\to (N^n,\bar g)$53 is the Weingarten map. Under a velocity field with normal speed $\iota:(M^m,g)\to (N^n,\bar g)$54, the shape derivative is

$\iota:(M^m,g)\to (N^n,\bar g)$55

while the mean curvature and, in $\iota:(M^m,g)\to (N^n,\bar g)$56, the Gauss curvature satisfy

$\iota:(M^m,g)\to (N^n,\bar g)$57

$\iota:(M^m,g)\to (N^n,\bar g)$58

These identities provide first-order updates of the second fundamental form and of its eigenvalue invariants under small normal deformations, and they underpin the Newton-type scheme for minimizing curvature-dependent shape functionals (Chicco-Ruiz et al., 2017).

In Hodge-theoretic geometry, the second fundamental form of the Torelli map is approximated by higher even Gaussian maps. For a non-hyperelliptic curve $\iota:(M^m,g)\to (N^n,\bar g)$59, the second fundamental form

$\iota:(M^m,g)\to (N^n,\bar g)$60

is identified, up to constant, with the Hodge–Gaussian map $\iota:(M^m,g)\to (N^n,\bar g)$61, and its composition with multiplication recovers the second Gaussian map: $\iota:(M^m,g)\to (N^n,\bar g)$62 If $\iota:(M^m,g)\to (N^n,\bar g)$63 with $\iota:(M^m,g)\to (N^n,\bar g)$64 even, then

$\iota:(M^m,g)\to (N^n,\bar g)$65

and

$\iota:(M^m,g)\to (N^n,\bar g)$66

Moreover, for any non-hyperelliptic curve of genus $\iota:(M^m,g)\to (N^n,\bar g)$67, $\iota:(M^m,g)\to (N^n,\bar g)$68 is injective and $\iota:(M^m,g)\to (N^n,\bar g)$69 for all $\iota:(M^m,g)\to (N^n,\bar g)$70. This suggests that higher even Gaussian maps provide a computable proxy for the second fundamental form of the Torelli map along higher Schiffer directions (Frediani, 2022).

A final global approximation principle is by mean value. For a complex projective manifold $\iota:(M^m,g)\to (N^n,\bar g)$71 with second fundamental form $\iota:(M^m,g)\to (N^n,\bar g)$72, the averaged squared length

$\iota:(M^m,g)\to (N^n,\bar g)$73

satisfies

$\iota:(M^m,g)\to (N^n,\bar g)$74

where $\iota:(M^m,g)\to (N^n,\bar g)$75 is the hyperplane section bundle and $\iota:(M^m,g)\to (N^n,\bar g)$76 is the sectional genus. Under the threshold $\iota:(M^m,g)\to (N^n,\bar g)$77, the possibilities are completely classified: strict inequality forces the minimal-degree cases $\iota:(M^m,g)\to (N^n,\bar g)$78, $\iota:(M^m,g)\to (N^n,\bar g)$79, the Veronese surface, or rational normal scrolls, while equality corresponds to del Pezzo manifolds or scrolls over elliptic curves. Stronger pinching gives rigidity: $\iota:(M^m,g)\to (N^n,\bar g)$80 and for $\iota:(M^m,g)\to (N^n,\bar g)$81,

$\iota:(M^m,g)\to (N^n,\bar g)$82

Here the approximation is by the mean $\iota:(M^m,g)\to (N^n,\bar g)$83-size of the second fundamental form rather than by a local tensor, but the conclusion is again rigidity of the underlying geometry (Li, 2019).

Taken together, these frameworks show that “approximate second fundamental form” is not a single construction but a cluster of technically distinct ideas. In complete submanifold theory it is an asymptotic decay condition; in geometric analysis it is an $\iota:(M^m,g)\to (N^n,\bar g)$84-energy or an orthogonal projection; in varifold theory it is a mollified curvature tensor with convergence theorems and point-cloud formulas; and in algebraic and Hodge geometry it is encoded by averaged extrinsic energy or higher Gaussian maps. The common feature is that the full pointwise tensor is replaced by a controlled surrogate that preserves enough curvature information to recover topology, spectra, variational optimality, or rigidity.