# Learned Busemann Stable Leaves

- ID: 2247
- Canonical URL: https://synthcore.org/idea/2247/learned-busemann-stable-leaves
- API JSON: https://synthcore.org/api/idea/2247.json
- API Markdown: https://synthcore.org/api/idea/2247.md
- Verification status: unverified
- Source: [arXiv:2608.13534](https://arxiv.org/abs/2608.13534)
- Category: dynamics
- Solves: stability, accuracy, generalization
- ML areas: rnn, ssm, world-model, regularization
- Math tags: dynamical-systems, geometry, control-theory, optimization
- Ratings: usefulness 6/10; difficulty 6/10; novelty 8/10

## Idea description

Learn an endpoint-conditioned scalar potential whose level sets represent states with the same asymptotic behavior, analogous to the paper's stable magnetic orthospheres. Train the dynamics to contract differences within a level set while preserving differences between distinct endpoint classes, producing a latent representation organized by stable manifolds rather than Euclidean proximity.

## Mathematical statement

The paper defines a finite-time function \(F_t(p,z,q)=-T_{pzq}(0)\), with separate definitions for an interior point z and a boundary point at infinity, and defines the stable magnetic B-function by \(B_{pz}^\mu(q)=\lim_{t\to\infty}F_t(p,z,q)\). Its stable foliation through a state v with endpoint z is \(\mathcal H^s(v)=\{\dot\gamma_{qz}(0):q\in (B_{pz}^\mu)^{-1}(0)\}\). Adapt this by learning an endpoint code \(z_\phi(x)\) and scalar \(B_\phi(x,z)\), then enforce invariance and contraction for pairs \(x,y\) with the same endpoint code. The asymptotic limit is approximated by a long-horizon terminal-prediction objective.

## Key formulas

- $$F_t(p,z,q)=\begin{cases}-T_{pzq}(0),&z\in\tilde\Sigma,\\-T_{p\gamma(t)q}(0),&z\in\tilde\Sigma(\infty),\end{cases}\qquad B_{pz}^{\mu}(q)=\lim_{t\to\infty}F_t(p,z,q).$$
- $$\mathcal H^s(v)=\left\{\dot\gamma_{qz}(0)\mid q\in\left(B_{pz}^{\mu}\right)^{-1}(0)\right\}.$
- $$\mathcal L_{\mathrm{leaf}}=\mathbb E_{x,y:\,z_\phi(x)\approx z_\phi(y)}\left[\max_{0\le k\le H}\|h_\theta(x_k)-h_\theta(y_k)\|_2^2\right]+\beta\,\mathbb E\left[\left(B_\phi(x_{k+1},z)-B_\phi(x_k,z)+c\right)_+^2\right].$$

## Implementation notes

Use a recurrent world model or SSM with latent state h. Add two heads: an endpoint head \(z_\phi(h)\) predicting a coarse future-behavior label or a representation of the final observation, and a scalar B-head \(B_\phi(h,z)\). Construct positive pairs from two windows in the same trajectory that have similar terminal futures, and negative pairs from trajectories with different futures. Pseudocode: `z1=z_head(h1); z2=z_head(h2); B1=B_head(h1,stopgrad(z1)); B2=B_head(h2,stopgrad(z2)); rollout both states for H steps; leaf_loss=max_k(norm(h1_k-h2_k)^2) + alpha*contrastive(z1,z2); drift_loss=relu(B_next-B_now+c)^2; total=task+lambda*(leaf_loss+beta*drift_loss)`. Estimate the paper's asymptotic limit by an exponential moving average of finite-horizon terminal-prediction losses; do not interpret the learned B-head as an exact geometric B-function. The first experiment should use a GRU on a synthetic system with two attracting limit cycles or two stable equilibria, followed by Lorenz-63 short-horizon prediction. Compare vanilla training, contrastive latent training, and this leaf objective. The prediction is that within-leaf rollout separation decays approximately as \(C\exp(-ck)\), while cross-leaf pairs retain separated endpoint codes. Fit the slope of \(\log\|h_k-h'_k\|\); success requires a negative slope for positive pairs and a significant gap between within-leaf and cross-leaf slopes. Sweep c: training should become unstable or overconstrained when the requested decay exceeds the achievable contraction rate.

## Disclaimer

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