A novel hybrid framework for efficient higher order ODE solvers using neural networks and block methods.

Area of Science:

• Numerical Analysis
• Computational Mathematics
• Applied Mathematics

Background:

• Traditional numerical methods for higher-order ODEs often lack stability, especially for oscillatory or exponential problems.
• Converting higher-order ODEs to first-order systems can be computationally intensive and introduce inaccuracies.

Purpose of the Study:

• To introduce a novel Neural-ODE Hybrid Block Method for the direct solution of higher-order ordinary differential equations (ODEs).
• To enhance the stability, accuracy, and efficiency of solving ODEs with complex dynamic behaviors.

Main Methods:

• A hybrid approach combining neural networks for approximating solution spaces and block numerical methods for direct ODE solution.
• Development of the mathematical formulation and neural network architecture for the hybrid model.
• Convergence and stability analysis to validate the method's performance.

Main Results:

• The proposed method demonstrates superior accuracy and stability compared to existing solvers for higher-order ODEs.
• Effective handling of stiff ODEs and boundary conditions, including nonlinear problems like the Van der Pol equation.
• Numerical experiments confirm fast computation and high accuracy for linear and nonlinear problems.

Conclusions:

• The Neural-ODE Hybrid Block Method provides a robust and accurate solution for higher-order ODEs, overcoming limitations of traditional techniques.
• The method's versatility makes it applicable across various scientific and engineering disciplines requiring precise ODE solutions.
• Future research directions include extending the method to multi-dimensional systems and partial differential equations (PDEs).