# Эксперимент: Forced Variational Momentum Optimizer (#1278)

{
  "worked": false,
  "confidence": 9,
  "verdict": "Built a PyTorch ForcedVariationalMomentum optimizer and reproducible verification/benchmark script. The discrete Euler–Lagrange recurrence matched the implemented update to numerical precision, and the quadratic sweep showed a small wider stable h range than the selected heavy-ball control (0.6266 vs 0.6067) with bounded rational damping. However, on the fixed digits MLP task the idea had worse test loss and accuracy than both controls (0.2107/96.0% vs heavy-ball 0.1119/96.9%), so the claimed practical optimization win was not observed despite fewer large loss spikes and AdamW's doubled state memory.",
  "metrics": {
    "baseline": "Heavy-ball: test_loss=0.1118835, accuracy=0.968889, loss_spikes_gt25pct=132, optimizer_tensor_state_elems=3610; AdamW: test_loss=0.1004683, accuracy=0.966667, loss_spikes_gt25pct=144, optimizer_tensor_state_elems=7224",
    "idea": "Forced variational: rho=0.38504 and step=0.094806 at h=0.37,gamma=2.4,mass=1.7; DEL residual=0; quadratic stable h_max=0.62659 vs heavy-ball 0.60670; MLP test_loss=0.2107174, accuracy=0.960000, loss_spikes_gt25pct=59, optimizer_tensor_state_elems=3610"
  },
  "how_to_run": "/home/maxwelhelp/main/bin/python3 experiment.py",
  "files": [
    "experiment.py",
    "results.json"
  ],
  "limitations": "Only one small digits dataset, one primary hyperparameter setting per optimizer, and 450 minibatch updates were tested. The mass was scalar one rather than a learned or frozen diagonal preconditioner; no broad hyperparameter search, FLOP-normalized timing, weight decay, or multiple random seeds was evaluated."
}