Prime Weighting

In Plain English#

Every constraint in a problem is different — a clause in a SAT formula is not the same as an edge-cut constraint in graph partitioning. But traditional algorithms treat constraints all the same, which causes problems: two constraints can create the same gradient direction, leading to gradient resonance and catastrophic interference.

Prime weighting solves this by giving every constraint a unique mathematical identity, like assigning each person a unique fingerprint. The assignment uses prime numbers because primes have a special property: they’re multiplicatively independent. You can’t create one prime by multiplying other primes together.

Why this matters: When each constraint has a unique prime weight, no two constraints ever produce the same gradient signature. The system always knows which constraint is causing which gradient. This eliminates an entire class of failure modes.

The Mathematics#

Each constraint $c$ receives a prime $p_c$ and weight:

$W(p_c) = \frac{1}{1 + \ln(p_c)}$

Key properties:

Monotonic: Larger primes → smaller weights (refinement, not dominance)
Logarithmic suppression: Prevents large primes from vanishing entirely
Prime Number Theorem: The $K$ -th prime $p_K \sim K \ln K$ , so weights scale consistently

Why This Formula?#

The weight formula $W(p) = 1/(1 + \ln p)$ emerges from the Prime Number Theorem:

$p_K \sim K \ln K$

This means: $\frac{1}{1 + \ln p_K} \sim \frac{1}{1 + \ln K + \ln \ln K}$

So the $K$ -th constraint receives weight $\approx 1/K$ — a natural harmonic scaling that matches the structure of modular arithmetic.

The Euler Product Connection#

The prime weights connect to the Riemann zeta function via the Euler product:

$\zeta(s) = \prod_{p \text{ prime}} \frac{1}{1 - p^{-s}}$

When constraints compose multiplicatively:

$P_{\text{mult}} = \prod_{v} \left(1 - \frac{1}{p_v^2}\right) \approx \prod_{p} \left(1 - \frac{1}{p^{2}}\right) = \frac{1}{\zeta(2)} = \frac{6}{\pi^2}$

The constraint satisfaction probability approaches $\zeta(2)^{-1}$ — a constant! This is the multiplicative bootstrap that makes the system stable.

Why Primes Are Special#

Fundamental Theorem of Arithmetic#

Every integer factors uniquely into primes. This means:

Multiplicative independence: No prime can be expressed as a product of other primes
Unique factorization: Each constraint’s gradient signature is distinct
No collisions: The mapping $c \to p_c$ is injective

Contrast with Uniform Weights#

Weight Scheme	Gradient Collision Probability	Scaling
Uniform	High (many constraints share weights)	O(1)
Random	Medium (birthday paradox)	O(√K)
Prime	Zero (guaranteed unique)	O(1)

Ablation Results#

From the NitroSAT benchmarks:

Ablation studies show 4x speedup and 75% reduction in topological complexity when using prime weights vs. uniform weights.

The prime-weighted system:

Detects phase transitions earlier and more reliably
Navigates fractures without losing progress
Scales better to large clause counts

Prime Weighting in Practice#

BAHA#

Each clause gets a prime weight; the fracture detector monitors the free energy gradient $\rho(\beta)$ .

Navokoj#

The Arithmetic Sector uses prime-weighted operators as the identity kernel — each constraint flows along its own geometric path.

Multiplicative PINN#

Each physics constraint (Navier-Stokes, Poisson, etc.) receives a distinct prime weight, enabling multiplicative enforcement without gradient conflicts.

Spectral-Multiplicative Framework#

Graph partitioning constraints receive prime weights that maintain $\rho \geq 0.99$ correlation between spectral and multiplicative functionals.

Key Insight#

Prime weighting is causal, not decorative. It’s not a heuristic — it’s a mathematical guarantee of uniqueness that emerges from the Fundamental Theorem of Arithmetic. Every other approach to constraint satisfaction has to deal with gradient collisions somehow; prime weighting eliminates them at the source.