What this page covers
How to form a stability-safe mean in rapidity space using nonnegative pooling weights. This is the backbone for s_sum (and other pools), keeps alignment bounded, and preserves collapse parity.
Definition (choose once per study)
w_i = |m_i|^gamma # gamma >= 0, so w_i >= 0
Weighted rapidity mean (used by s_sum and other pools)
U = sum_i [ w_i * atanh( clamp_a(a_i, eps_a) ) ]
W = sum_i w_i
mean_u = U / max( W, eps_w )
a_pool = tanh( mean_u )
Collapse and bounds
- Collapse is unaffected (pooling touches alignment only).
a_poolalways satisfies|a_pool| < 1(byclamp_aandtanh).
Domain guards
- Use
max(W, eps_w)to avoid division by zero. - All
w_i >= 0by construction (no sign flips via weights).
Notes
gamma = 0gives equal weights;gamma > 0emphasizes larger|m|.- Publish
gamma(andeps_w) in the manifest. - Keep
UandWas streaming accumulators to preserve associativity across batches/windows.
Navigation
Previous: Identities, Inverses, and Zero-Class Display (1.3)
Next: Collapse-Safety Contract (1.5)