Smooth isometric immersions of surfaces into four-space

Source and verification

This article explains family 333 of the October 6, 2026 OpenAI collection. Research conclusions are attributed to its manuscripts. The review covers statements, introductions, and selected proof mechanisms; it is not an independent verification of the complete proofs.

🏷️ Preliminaries: metrics, immersions, and regularity

A Riemannian metric $g$ assigns a positive definite inner product to each tangent plane of a smooth surface $M$. For a smooth map $F:M\to {\mathbb{R}}^4$, its induced metric is

$$\gamma (F)(v,w)=⟨dF(v),dF(w)⟩.$$

The equation $\gamma (F)=g$ says that tangent lengths and angles are preserved. Positive definiteness makes $dF$ injective: $F$ is an immersion. It need not be injective as a map, so self-intersections are permitted. An embedding would impose additional global requirements.

A surface is closed if it is compact without boundary. It may be nonorientable or disconnected. Smoothness means $C^{\infty }$, with derivatives of every order; a $C^1$ isometric immersion is a substantially different conclusion because its second fundamental form need not exist.

Manuscript result

Every closed smooth Riemannian surface admits a smooth isometric immersion into Euclidean four-space. No orientability or Gaussian-curvature hypothesis is imposed. (OpenAI, 2026)

The prerequisites are basic differential geometry and analysis of smooth functions. The proof combines oscillatory correction with smooth convergence and preserved normal geometry. An approximate metric realization, or a $C^1$ limit, would not establish the stated result.

🏷️ Normal geometry and the metric deficit

In local coordinates, the second derivative of an immersion splits into tangential and normal parts:

$$F_{ij}={\Gamma }_{ij}^k{F}_k+B_F({\partial }_i,{\partial }_j).$$

Here ${\Gamma }_{ij}^k$ are the Christoffel symbols of the induced metric, and $B_F$ is the vector-valued second fundamental form. A surface in four-space has a two-dimensional normal plane. There need not be a globally chosen orthonormal normal frame.

The construction starts with an ordinary immersion into a small round three-sphere inside ${\mathbb{R}}^4$. Scaling makes its induced metric $g_0$ strictly smaller than $g$: the difference $g-g_0$ is positive definite. Such an immersion is strictly short.

The negative radial direction gives one globally defined unit normal, even on a nonorientable surface. Its radial curvature component also ensures that $B_F$ is nonzero as a tensor at every point. This does not require a full normal frame or all normal curvatures to have one sign.

The positive metric deficit is decomposed into finitely many rank-one tensors,

$$g-g_0=\sum _{j=1}^J{a}_j^{2}d{x}_j^2.$$

Each $x_j$ is a coordinate function on a disk, while $a_j$ is smooth, positive inside that disk, and zero outside. Flat vanishing at the boundary makes the extended tensor smooth. These primitive tensors must be added successively without losing the conditions needed by later additions.

🏷️ One primitive addition

Fix coordinates $(x,y)$ and a desired addition $a^{2}d{x}^2$. For a slow map $G$ near the current immersion, set

$$\begin{array}{c}Y=G_y,C=G_{yy},\\ P=\mathrm{span}(Y,C{)}^{\perp },v=P{G}_x.\end{array}$$

The notation $P$ also denotes orthogonal projection onto this plane. A positive transverse phase Hessian ensures that $Y,C$ are independent. Thus $P$ is two-dimensional, supplying a circle of possible velocities.

Choose a periodic velocity $V$ in $P$ with

$$|V{|}^2=|v{|}^2+a^2,⟨V⟩=v,$$

where angle brackets denote the periodic average. Put $w=V-v$. Its mean is zero, so it has a periodic primitive. The proposed leading $x$-derivative is $X=G_x+w$.

Since $G_x-v$ is orthogonal to $P$,

$$\begin{array}{rl}|X{|}^2& =|G_x{|}^2+a^2,\\ ⟨X,Y⟩& =⟨G_x,G_y⟩.\end{array}$$

Thus the leading metric changes by exactly the desired rank-one tensor. A circular velocity increases one tangent length while leaving the mixed metric coefficient unchanged.

If $W$ is a periodic primitive of $w$, the actual oscillating map has leading form

$$G_{\lambda }(p)=G(p)+{\lambda }^{-1}W(p,\lambda x(p)).$$

Differentiation produces the desired velocity and lower-order derivatives of its coefficients. Merely taking $\lambda$ large gives an approximate metric.

The chosen plane contains additional second-order information. A primitive $W$ can be taken orthogonal to both $G_y$ and $G_{yy}$. Consequently,

$$\begin{array}{rl}⟨G_y,{\partial }_{y}W⟩& ={\partial }_y⟨G_y,W⟩\\ & -⟨G_{yy},W⟩=0.\end{array}$$

This cancels the first transverse metric error. Finite correctors cancel oscillatory errors to any prescribed finite order; remaining averaged errors are treated on a slower scale.

🏷️ Exact smooth correction and derivative loss

The correction theorem has quantitative hypotheses. It applies to a family $f_z$ whose derivatives through a fixed finite order have specified polynomial growth as $z\downarrow 0$, whose metric error is sufficiently small in a specified $C^R$ norm, and whose second fundamental form has a uniform pointwise nonvanishing bound.

It produces an exact smooth immersion ${\tilde{f}}_z$ satisfying

$$\begin{array}{c}\gamma ({\tilde{f}}_z)=h_{\ast },\\ \Vert {\tilde{f}}_z-f_z{\Vert }_{C^2}=o(1).\end{array}$$

The finite derivative and error-scale hypotheses are essential. This is not an arbitrary $C^2$-error correction theorem based only on nonvanishing of the second fundamental form.

The linearized equation loses derivatives, so a naive iteration can make higher derivatives diverge. The construction smooths the data used to form each increment, with estimates such as

$$\Vert S_{s}u{\Vert }_{C^{q+m}}\le C{s}^{-m}\Vert u{\Vert }_{C^q}.$$

Approximation estimates bound $u-S_{s}u$ using finitely many additional derivatives. The current immersion itself is not replaced by its smoothing. Its higher derivatives remain available until the iteration reaches an order where they can be controlled.

Two indices matter: the finite order controlled at a given stage and an arbitrary fixed derivative order in which convergence is ultimately required. The controlled order increases in blocks, with enough intervening steps to make the increments summable. This yields one map on the original surface converging in every $C^m$ norm.

The proof uses finite mode-by-mode constructions rather than assuming a global inverse for the full linearized induced-metric map. This distinction is important in the low target dimension.

🏷️ Preserving the next disk’s geometry

The primitive theorem carries boundary data for all later support disks. Along a later boundary curve, let $Y_i$ be its assigned transverse direction and set

$$A_i=B_F(Y_i,Y_i).$$

The conditions include $A_i\ne 0$ and the assertion that $A_i/|A_i|$ is not opposite to the preferred normal. At crossings, stronger positivity conditions prevent local preferred normals from becoming incompatible.

One crossing quantity has the Gauss-equation expression

$$Q_{il}=⟨A_i,A_l⟩-|Z_{il}^{\perp }{|}^2,$$

where $Z_{il}=B_F(Y_i,Y_l)$ and $Z_{il}^{\perp }$ is its component orthogonal to $A_i$. Thus $Q_{il}>0$ forces a positive inner product between the pure second-form vectors. It depends continuously on the immersion’s second derivatives where $A_i\ne 0$.

The large oscillation can rotate the second fundamental form. A small metric error alone does not preserve these conditions. The velocity loop has controlled turns, and its transverse derivatives retain the required information. The exact correction is then small in $C^2$, preserving the strict inequalities.

The map may change outside the active disk; the theorem gives arbitrarily small $C^2$ change there. Its induced metric changes by exactly the prescribed tensor, which is zero outside the disk.

🏷️ Finite global assembly

Support boundaries are chosen pairwise transverse with no triple intersections. Each phase must have positive transverse Hessian for the metric immediately before its addition. Every partial metric is prescribed by the finite tensor sum, so this requirement is arranged in advance.

At the finite set of crossings, the initial spherical immersion is modified so that its second fundamental form within the three-sphere vanishes. The radial component in four-space remains and supplies positivity. Uniform second-derivative bounds for these modifications avoid a circular choice between the number of disks and the initial geometry.

The primitive theorem is applied through the finite list. Each step realizes the next metric exactly and preserves the remaining boundary conditions. At the end,

$$g_0+\sum _{j=1}^J{a}_j^{2}d{x}_j^2=g.$$

There are finitely many global primitive additions, but an infinite smooth correction iteration inside each. This distinction explains how the construction can be finite in its geometric organization and convergent in every derivative order.

🏷️ Key steps and proof outlines

The linearized equation has a finite oscillatory inverse

For an amplitude $Z$, write its tangential covector as $u_i=⟨G_i,Z⟩$ and its normal part as $v$. Differentiating the metric gives

$$(D_{G}Z{)}_{ij}={\partial }_i{u}_j+{\partial }_j{u}_i-2{\Gamma }_{ij}^k{u}_k-2⟨B_{ij},v⟩.$$

Conjugation by the fast phase $e^{ix/\tau }$ adds $i({\delta }_{ix}{u}_j+{\delta }_{jx}{u}_i)/\tau$. Since the admissible phase has $b=B_{yy}\ne 0$, choose the component of $v$ parallel to $b$ to solve the $yy$ equation exactly. A normal component perpendicular to $b$ remains free. The other two equations become $(i/\tau )u+Lu=k$, with $L$ first order. Taking $u=(\tau /i)k$ leaves an error smaller by $\tau /s$ in norms weighted at the slow scale $s$.

Repeatedly applying this differential correction cancels the error to any prescribed finite order. Because these are differential formulas, supports remain in the original charts. The free normal amplitude creates the positive quadratic metric mean; doubled and mixed phases cancel its unwanted oscillatory terms. The source requires those finitely many mixed phases to remain admissible as well. (OpenAI, 2026)

One small increment separates amplitude from wavelength

If the displacement has size $\delta \tau$, its leading derivative has size $\delta$ and its quadratic metric contribution has size ${\delta }^2$. The small-increment theorem realizes a positive tensor $H$ with error

$$U=O_{\tau }(\delta \tau ),\gamma (G+U)-\gamma (G)-{\delta }^{2}H=O_{\tau }\left(\delta (\tau /s{)}^q+{\delta }^3/\tau \right).$$

Here $O_{\tau }$ controls each fixed norm with derivatives weighted by powers of $\tau$. The first error is the finite linear cancellation error; the second is the nonlinear remainder. Increasing $q$ handles the first, while choosing $\delta \ll \tau$ handles the second. The mean corrections arising from the large primitive loop are incorporated as finite polynomial differential perturbations, with a scale exponent fixed independently of the requested cancellation order.

Smoothing the data preserves a linear high-derivative recurrence

Let $H={\delta }^{-2}(h_{\ast }-\gamma (G))$. Construct an increment using smoothed data $G_s,H_s$, but add it to $G$. Choose the next amplitude ${\delta }^{\prime }>0$ and target the smoothed increment ${\delta }^2{H}_s-{\delta }^{\prime 2}{h}_{\ast }$. Define

$$\begin{array}{rl}{H}^{\prime }& ={\delta }^{\prime -2}(h_{\ast }-\gamma (G+U)),\\ E& =\gamma (G_s+U)-\gamma (G_s)\\ & -{\delta }^2{H}_s+{\delta }^{\prime 2}{h}_{\ast }.\end{array}$$

The exact quadratic identity is

$${\delta }^{\prime 2}(H^{\prime }-h_{\ast })={\delta }^2(H-H_s)-E-\mathcal{C}(G-G_s,U),$$

where $\mathcal{C}(V,W{)}_{ij}=⟨V_i,W_j⟩+⟨V_j,W_i⟩$. There is no square of the smoothing tail. Consequently large higher derivatives of the true current map enter the next estimate only linearly.

The source chooses $\tau =t^{6/5}$, $s=t^{11/10}$, and $\delta =t^k$. With sufficiently many vanishing moments in the smoothing kernel, the weighted norm at every fixed observation order obeys a recurrence of the form $S_m^{\prime }\le C+C_m{t}^{1/5}(1+S_m)$. At the currently controlled order this preserves a bound; at any higher fixed order repeated contraction eventually establishes one. The construction then increases the controlled order. Summability of the resulting increments in every fixed $C^m$ norm proves smooth convergence, while the total $C^2$ change preserves the geometric margins required for the next primitive addition.

See Also

The local three-space obstruction discusses failure of smooth realization even locally in target dimension three. The two results concern different dimensions and do not assert an embedding theorem.

🏷️ Manuscripts and verification scope

• Smooth isometric immersions of closed surfaces into Euclidean four-space — pinned manuscript and companion source.

The family scope page describes selected formalization coverage. The documented selection is the closed-surface immersion theorem without an orientability assumption; it is not an embedding statement. No Lean build, axiom audit, or manuscript-to-formal-statement equivalence check was performed for these notes.

OpenAI collection provides prerequisites and the reading sequence; coverage register records completed and pending families.

🐻  OpenAI 2026. Smooth isometric immersions of closed surfaces into Euclidean four-space.