Enhancing conjugate gradient performance via a diagonal quasi-Newton strategy: applications in adaptive filtering and power allocation

Articles in Press

Document Type : Research Article

Author

Narges Bidabadi

Faculty of Mathematical Science, Yazd University, Yazd, Iran.

10.22067/ijnao.2026.97071.1808

Abstract

Unconstrained optimization is fundamental to scientific computing. While nonlinear conjugate gradient (CG) methods are valued for their low memory requirements and simplicity, their convergence may be slow. Conversely, quasi-Newton methods like BFGS achieve faster superlinear convergence but become computationally prohibitive for large-scale problems. To address this trade-off, this paper introduces a novel hybrid algorithm that combines a diagonal quasi-Newton (DQN) approach with a conjugate gradient method. This framework leverages the complementary strengths of both techniques, maintaining modest memory usage while accelerating convergence. Numerical experiments on several benchmark problems, adaptive filter design, and power allocation, demonstrate the proposed method's effectiveness. Our MATLAB-based evaluations show that the hybrid approach consistently outperforms standard CG and BFGS methods.

Keywords

Conjugate gradient

Diagonal quasi-Newton

adaptive filtering

Power allocation

Subjects

• Optimization

---