🩹 Fixes

• Unified Force Constant Nomenclature
  • Renamed angle/inversion parameters from k → c to reflect that these are stiffness coefficients in cosine-space, not traditional spring constants.
  • Rationale: Physical clarity. $K$ denotes force constants for geometric coordinates ($K(R-R_0)^2$, $K(\theta-\theta_0)^2$), while $C$ denotes coefficients for transformed spaces ($C(\cos\theta - \cos\theta_0)^2$). The effective restoring force for $C$-type potentials is geometry-dependent ($C\sin^2\theta$), fundamentally different from linear springs.

Contributors

• @TKanX Lead Developer

Full Changelog: v0.4.0...v0.4.1

🚀 Features

• New Parameter Pre-computation Methods
  • Convenience Layer: New precompute() methods convert physical constants (force fields, equilibrium geometries) into optimized kernel parameters.
  • 9 Kernels Enhanced:
    • Harmonic: $(K, R_0) \to (K/2, R_0)$
    • CosineHarmonic: $(C, \theta_0°) \to (C/2, \cos_0)$
    • ThetaHarmonic: $(K, \theta_0°) \to (K/2, \theta_{0,rad})$
    • Torsion: $(V, n, \phi_0°) \to (V/2, n, \cos_{n\phi_0}, \sin_{n\phi_0})$
    • PlanarInversion: $C \to C/2$
    • UmbrellaInversion: $(C, \psi_0°) \to (C/2, \cos_{\psi,0})$
    • LennardJones: $(D_0, R_0) \to (D_0, R_0^2)$
    • Buckingham: $(D_0, R_0, \zeta) \to (A, B, C, r_{max}^2, 2E_{max})$
    • HydrogenBond: $(D_{hb}, R_{hb}) \to (D_{hb}, R_{hb}^2)$
  • Verified: 18 new unit tests (precompute_values + precompute_round_trip) ensure correctness.

Contributors

• @TKanX Lead Developer

Full Changelog: v0.3.0...v0.4.0

🚀 Features

• Enhanced Buckingham Potential with Energy Reflection
  • Upgraded from constant penalty to energy reflection at the local maximum ($r = r_{\text{max}}$) to handle short-range divergence.
  • Parameters expanded from 4-tuple to 5-tuple: (A, B, C, r_max², 2E_max).
  • C¹ continuous: Force is continuous and equals zero at the boundary from both sides.
  • Correct physics: Energy diverges to $+\infty$ as $r \to 0$ (Pauli repulsion).
  • Branchless implementation: Uses arithmetic masking for SIMD-friendly vectorization.

🛠️ Improvements

• Removed SplinedBuckingham Potential
  • Eliminated the C² continuous rational function variant.
  • Rationale: 99% of MD simulations use Velocity Verlet integrators, which only require C¹ continuity (force continuity). C² is unnecessary for standard practice.

Contributors

• @TKanX Lead Developer

Full Changelog: v0.2.0...v0.3.0

🚀 Features

• New CosineLinear Angle Bending Kernel
  • Linear Geometry Support: Implements Eq. 10' from Mayo et al. (1990) — E = K(1 + cosθ) — for atoms with θ₀ = 180° equilibrium.
  • Correct Curvature: Replaces CosineHarmonic at the linear limit where it degenerates into a quartic well (∝ (π−θ)^4), restoring the physically correct quadratic restoring force (∝ (π−θ)^2).
  • Fastest Kernel: Constant derivative (Γ = K) yields ~1.5 Billion ops/sec — single multiply-add for energy, zero arithmetic for force.
  • Verified: Unit + integration tests included (gold-standard values at 0°, 90°, 120°, 180°).

Contributors

• @TKanX Lead Developer

Full Changelog: v0.1.0...v0.2.0

🚀 Features

1. Complete DREIDING Force Field Kernel Library

   • 5 Non-Bonded Potentials: Lennard-Jones 12-6, Buckingham Exp-6, Splined Buckingham (C² continuous), Coulomb, Hydrogen Bond 12-10
   • 2 Bond Stretch Potentials: Harmonic, Morse (anharmonic with dissociation)
   • 2 Angle Bending Potentials: Cosine Harmonic, Theta Harmonic
   • 1 Torsion Potential: Periodic Cosine with optimized multi-angle formulas (n=1,2,3 closed-form, Chebyshev for n≥4)
   • 2 Inversion Potentials: Planar (sp² centers), Umbrella (sp³ centers)

2. High-Performance Architecture

   • Zero Heap Allocation: All computations on the stack—suitable for tight MD loops and embedded systems.
   • Branchless Kernels: Mask-based conditionals for SIMD-friendly vectorization.
   • Optimized Algorithms: Precomputed coefficients, shared subexpressions, Horner's method, Chebyshev recursion.
   • ~1.4 Billion ops/sec: Cosine Harmonic throughput; ~840M/sec for Lennard-Jones on Intel i7-13620H.

3. Mathematically Rigorous

   • Analytically Verified: All force kernels are exact negative gradients of energy functions.
   • 187 Tests: Stability checks, finite-difference validation (10⁻⁴ tolerance), and physics regression tests for every potential.
   • DREIDING Compliant: Implements exact functional forms from Mayo et al. (1990).

4. Geometry-Agnostic Design

   • Stateless Kernels: Pure functions with no object lifecycle—you provide state, we return scalars.
   • Unified Force Interface: All kernels output diff factors for F += -Factor * GeometricVector pattern.
   • Separation of Concerns: Kernel layer computes derivatives; geometry layer applies coordinates.

5. Portability & Usability

   • #![no_std] Support: Fully compatible with embedded devices, OS kernels, and WASM (via libm feature).
   • Generic Precision: Unified API for both f32 and f64 via the Real trait.
   • Comprehensive Documentation: KaTeX-rendered LaTeX formulas in RustDoc for every potential.

New Contributors

• @TKanX Lead Developer

Full Changelog: https://github.com/caltechmsc/dreid-kernel/commits/v0.1.0