Mathematical Framework

Our research models basketball team dynamics as active matter using Wasserstein Gradient Flows. We treat the five-player defensive unit as a probability distribution of mass that “flows” to minimize its free energy functional.

The JKO Scheme

The Jordan-Kinderlehrer-Otto (JKO) scheme provides the variational framework for this evolution. Unlike standard gradient descent, which might allow for unrealistic “teleportation” of players, the JKO scheme restricts movement via a kinetic penalty.

Mathematical Formulation

The evolution of the defensive density is solved as a sequence of optimization problems:

\[\rho_{k+1} = \text{argmin}_{\rho} \left\{ \frac{1}{2\tau} W_2^2(\rho, \rho_k) + \mathcal{F}(\rho) \right\}\]

• Wasserstein Distance (\(W_2\)): Measures the exact physical cost of transporting defensive mass from one configuration to another.
• Inertial Penalty: The \(\frac{1}{2\tau}\) term ensures that the defense obeys physical movement constraints, preventing unnatural, high-velocity jumps.
• Gradient Flow: As the time step (\(\tau\)) approaches zero, the defense converges to a continuous path that solves a Fokker-Planck equation.

Strategic Loss Optimization

We evaluate our model using three core metrics designed to balance tactical positioning with physical realism.

1. Pressure Loss (\(L_{\text{pressure}}\))

Quantifies how effectively defenders guard offenders. It uses a Goal-Side Factor to prefer defenders staying between the offender and the basket.

2. Threat Minimization (\(L_{\text{IST}}\))

Measures the suppression of the Instantaneous Shot Threat. It evaluates the defense’s ability to “close out” on elite shooters while “sagging” off weaker perimeter threats.

3. Smoothness Loss (\(L_{\text{smoothness}}\))

Penalizes jerky or abrupt movements, promoting fluid trajectories that mimic professional athletes.

Two-Stage Bayesian Optimization

Because the IST potential introduces highly non-linear gradients, we utilized a staged approach via Optuna:

• Stage 1 (Kinematic Baseline): Optimized geometric variables (cohesion, occupancy, and ball pressure) to establish a baseline defense that obeys physical laws.
• Stage 2 (Threat Integration): Froze the kinematic parameters and isolated IST variables to tune strategic prioritization (\(W_{\text{ist}} = 1.479\)).