# Mapping-Cone Boundary Consistency Loss

- ID: 511
- Canonical URL: https://synthcore.org/idea/511/mapping-cone-boundary-consistency-loss
- API JSON: https://synthcore.org/api/idea/511.json
- API Markdown: https://synthcore.org/api/idea/511.md
- Verification status: unverified
- Source: [arXiv:2607.07149](https://arxiv.org/abs/2607.07149)
- Category: regularization
- Solves: accuracy, stability, generalization
- ML areas: graph-nn, loss, regularization, world-model
- Math tags: geometry, differential-geometry, topology, linear-algebra
- Ratings: usefulness 5/10; difficulty 5/10; novelty 6/10

## Idea description

Augment a neural model with a learned target differential form and a source-side correction whose compatibility is enforced by the mapping-cone differential. For a map F from M to N, train the model so that the target quantity is closed and its pullback to M is exactly the differential of the correction, providing a structured bulk-boundary consistency constraint instead of independent feature matching.

## Mathematical statement

The paper defines the relative de Rham complex by Omega^k(F) = Omega^k(N) direct-sum Omega^(k-1)(M) and d_F(alpha,beta) = (d alpha, F* alpha - d beta). Here M is the source or boundary representation space, N is the target or bulk representation space, F:M to N is a known differentiable map, alpha is a k-form on N, beta is a (k-1)-form on M, d is the exterior derivative, and F* alpha is the pullback. A closed relative form satisfies d alpha = 0 and F* alpha = d beta. Use the concrete k=1 case: write omega(y) = a(y) transposed dy with vector field a:N to R^(dim N), and let eta:M to R. Then F* omega(x) = J_F(x) transposed a(F(x)), so the two residuals are the closure residual d omega and the boundary residual J_F(x) transposed a(F(x)) minus grad_x eta(x). The Cartan identity L_v = d_F iota_v + iota_v d_F explains why a closed relative form yields invariant pairings under compatible flows, although the MVP only needs the explicit closure and pullback-exactness equations.

## Key formulas

- $$\Omega^{k}(F):=\Omega^{k}(N)\oplus\Omega^{k-1}(M),\qquad d_{F}(\alpha,\beta):=(d\alpha,\;F^{*}\alpha-d\beta).$$
- $$d_F(\omega,\eta)=0\quad\Longleftrightarrow\quad d\omega=0\ \text{and}\ F^{*}\omega=d\eta.$$
- $$F^{*}\omega(x)=J_F(x)^{\top}a(F(x)),\qquad r_{\partial}(x)=J_F(x)^{\top}a(F(x))-\nabla_x\eta(x).$$
- $$\mathcal L_v=d_F\iota_v+\iota_vd_F.$$

## Implementation notes

Integrate this at the representation interface of a model with paired samples x in M and y = F(x) in N, such as a graph encoder with a node-to-graph map, a neural operator with interior and boundary coordinates, or a world model with a state-transition map. Add two small heads: a_theta(y), producing a vector field on target coordinates, and eta_theta(x), producing a scalar on source coordinates. For each minibatch, compute y = F(x), the Jacobian J_F(x) by autodiff or analytically, and the pullback residual r_boundary = J_F(x)^T a_theta(y) - grad_x eta_theta(x). For omega = a_i(y) d y_i, estimate the closure residual by autodiff: r_closed,ij = partial_{y_i} a_j - partial_{y_j} a_i. Optimize L = L_task + lambda_boundary times mean(||r_boundary|| squared) + lambda_closed times mean(sum_{i<j} r_closed,ij squared). Compact pseudocode is: `y = F(x); a = a_theta(y); eta = eta_theta(x); J = jacobian(F, x); rb = transpose(J) @ a - grad(eta, x); rc = antisymmetric_jacobian(a, y); loss = task_loss + lb * mean(rb**2) + lc * mean(rc**2); backprop(loss)`. The paper supplies the exact mapping-cone residual; Jacobians, gradients, and residual norms are computed numerically. Since eta has gauge freedom up to an additive constant, center it per batch or impose zero mean. First test on a synthetic 2D-to-1D boundary map and on a small Poisson or advection neural-operator benchmark, comparing a standard supervised model against the same model with this loss. Measure test error, boundary or interface violation, robustness to noisy boundary labels, and gradient variance. A positive result is lower interface error and improved out-of-domain generalization at equal parameter count, without slowing the main task loss by more than 10 percent.

## Disclaimer

AI-generated research hypothesis, automatically tested. Not peer-reviewed.