LLCM: Leapfrog Integrator with Energy Conservation

• Leapfrog integration (LLCM) is an advanced symplectic integrator that enforces an exact discrete quasi-energy conservation law.
• The algorithm augments the discrete variational principle with a Lagrange multiplier to solve a quadratic constraint, ensuring energy invariance at each time step.
• LLCM improves long-term numerical stability and accuracy in applications like molecular dynamics, Hamiltonian Monte Carlo, plasma physics, and deep learning.

Leapfrog integration, widely employed for symplectic and energy-conserving time integration in computational physics, has found broad application in molecular simulation, Hamiltonian Monte Carlo, Bayesian inference, plasma physics, and, more recently, deep learning and stochastic modeling. The "leapfrog integrator with an energy-like conservation law" (LLCM) represents a significant advancement, embedding an explicit, exact quasi-energy constraint into the algorithmic structure. This modification retains the symplectic and momentum-preserving features of classic leapfrog while enforcing a discrete version of energy conservation, yielding improved long-time numerical stability and boundedness of the mechanical energy (Maggs, 2013).

1. Discrete Variational Principle and the Standard Leapfrog Integrator

Starting from Newton’s equations for $N$ particles of mass $m$ subject to a potential $V(q)$, the continuous action is given by

$$S[q]=\int_{t_0}^{t_1}\left[\frac{m}{2}\left\|\dot{q}(t)\right\|^2 - V(q(t))\right]\,dt.$$

Discretizing time with step size $\tau$ and positions $q_k = q(t_k)$ at times $t_k = k\tau$, a discrete Lagrangian is constructed: $$L_k(q_k, q_{k+1}) = m\,\frac{\|q_{k+1} - q_k\|^2}{2\,\tau^2} - V(q_k).$$ The discrete action principle leads to the position-only leapfrog scheme: $$m(q_{k+1} + q_{k-1} - 2q_k) + \tau^2\nabla V(q_k) = 0,$$ which is symplectic and momentum-conserving but does not conserve the continuous mechanical energy $U(q, \dot{q}) = \frac{m}{2}\|\dot{q}\|^2 + V(q)$ exactly at the discrete level (Maggs, 2013).

2. Extension to the LLCM: Exact Discrete Quasi-Energy Conservation

To enforce strict conservation of an energy-like invariant, LLCM augments the discrete Lagrangian with a Lagrange multiplier to impose

$m$0

at every step, where $m$1 is the prescribed quasi-energy. Introducing $m$2 as a discrete Lagrange multiplier and defining the quasi-energy as $m$3, the extended Lagrangian becomes

$m$4

Legendre transformation yields the momentum update

$m$5

and the extended Hamiltonian,

$m$6

Enforcing $m$7 produces the discrete quasi-energy constraint

$m$8

A further extension to a fully Hamiltonian system introduces a conjugate variable $m$9 and ensures exact preservation of $V(q)$0 by suitable design of the evolution equations (Maggs, 2013).

3. Explicit Algorithmic Realization and Quasi-Energy Invariance

The LLCM time-stepping algorithm proceeds as:

$\tau$0 The quadratic equation for $V(q)$1 is solved at each step, ensuring $V(q)$2 holds exactly. As a result, discrete quasi-energy $V(q)$3 is invariant under the LLCM map for all time steps (Maggs, 2013).

4. Symplectic Structure, Long-Time Stability, and Computational Trade-Offs

Embedding the dynamics in the extended $V(q)$4 phase space, the resulting discrete map preserves a phase-space volume form (unit Jacobian), mirroring the classic symplecticity of leapfrog. Enforced quasi-energy conservation inhibits the secular drift of mechanical energy typical of non-exact symplectic schemes, yielding improved long-term stability of the system’s energy. While the basic (forward) update is not strictly time-reversible, exact time-reversibility is recovered by alternating forward and backward steps, each requiring solution of a small implicit equation for $V(q)$5. Each step entails the solution of a $V(q)$6 quadratic for $V(q)$7 (overall $V(q)$8 cost per particle system step), in addition to force evaluation. The local integration error per step is $V(q)$9, as in standard leapfrog, but energy drift is eliminated (Maggs, 2013).

5. Implementation and Practical Considerations

• Initialization: Given $$S[q]=\int_{t_0}^{t_1}\left[\frac{m}{2}\left\|\dot{q}(t)\right\|^2 - V(q(t))\right]\,dt.$$0, solve $$S[q]=\int_{t_0}^{t_1}\left[\frac{m}{2}\left\|\dot{q}(t)\right\|^2 - V(q(t))\right]\,dt.$$1 for $$S[q]=\int_{t_0}^{t_1}\left[\frac{m}{2}\left\|\dot{q}(t)\right\|^2 - V(q(t))\right]\,dt.$$2 and set $$S[q]=\int_{t_0}^{t_1}\left[\frac{m}{2}\left\|\dot{q}(t)\right\|^2 - V(q(t))\right]\,dt.$$3.
• Step-size selection: The step $$S[q]=\int_{t_0}^{t_1}\left[\frac{m}{2}\left\|\dot{q}(t)\right\|^2 - V(q(t))\right]\,dt.$$4 is chosen to resolve system frequencies, as in any explicit symplectic integrator; in practice, $$S[q]=\int_{t_0}^{t_1}\left[\frac{m}{2}\left\|\dot{q}(t)\right\|^2 - V(q(t))\right]\,dt.$$5 remains numerically stable.
• Monitoring invariants: At each time step, $$S[q]=\int_{t_0}^{t_1}\left[\frac{m}{2}\left\|\dot{q}(t)\right\|^2 - V(q(t))\right]\,dt.$$6 is recomputed to verify quasi-energy preservation to round-off, and monitoring of the extended Hamiltonian $$S[q]=\int_{t_0}^{t_1}\left[\frac{m}{2}\left\|\dot{q}(t)\right\|^2 - V(q(t))\right]\,dt.$$7 ensures absence of hidden drifts.
• Time-reversal symmetry: Full time-reversibility is achieved by alternating forward and backward steps with implicit updates for $$S[q]=\int_{t_0}^{t_1}\left[\frac{m}{2}\left\|\dot{q}(t)\right\|^2 - V(q(t))\right]\,dt.$$8.
• Efficiency: The only additional computational overhead is the solution of the quadratic equation at each step and the maintenance of the $$S[q]=\int_{t_0}^{t_1}\left[\frac{m}{2}\left\|\dot{q}(t)\right\|^2 - V(q(t))\right]\,dt.$$9 field, with no significant per-step penalty compared to standard leapfrog (Maggs, 2013).

6. Significance, Relation to Standard Integrators, and Applications

The LLCM can be viewed as a natural extension of the classical symplectic leapfrog, which is already favored for its structure-preserving (momentum and volume) properties, but which does not strictly conserve energy. The LLCM addresses this limitation by embedding the system in an extended phase space and introducing a quasi-energy constraint, thereby enabling the realization of discrete trajectories with exact energy conservation. This innovation is particularly valuable in molecular dynamics, celestial mechanics, plasma particle simulation, and any setting where long-time boundedness of energy is essential (Maggs, 2013).

In summary, the leapfrog integrator with an energy-like conservation law enforces a discrete energy invariant alongside symplecticity and momentum conservation, yielding superior long-time stability, minimal per-step computational overhead, and a rigorous Hamiltonian underpinning (Maggs, 2013).