Abstract:This paper addresses unconstrained multiobjective optimization problems where two or more continuously differentiable functions have to be minimized. We delve into the conjugate gradient methods proposed by Lucambio Pérez and Prudente (SIAM J Optim, 28(3): 2690--2720, 2018) for such problems. Instead of the Wolfe-type line search procedure used in their work, we employ a fixed stepsize formula (or no-line-search scheme), which can mitigate the pressure of choosing stepsize caused by multiple inequalities and avoid the computational cost associated with function evaluations in specific applications. The no-line-search scheme is utilized to derive the condition of Zoutendijk's type. Global convergence encompasses the vector extensions of Fletcher--Reeves, conjugate descent, Dai--Yuan, Polak--Ribière--Polyak and Hestenes--Stiefel parameters, subject to certain mild assumptions. Additionally, numerical experiments are conducted to demonstrate the practical performance of the proposed stepsize rule, and comparative analyses are made with the multiobjective steepest descent methods using the Armijo line search and the multiobjective conjugate gradient methods using the Wolfe-type line search.
What problem does this paper attempt to address?
### What problem does this paper attempt to solve?
This paper aims to solve the **unconstrained multi - objective optimization problem**, in which two or more continuously differentiable functions need to be minimized. Specifically, the author focuses on how to effectively apply the Conjugate Gradient (CG) method to solve such problems without using Wolfe - type line search.
#### Main problems and motivations
1. **Complexity of multi - objective optimization problems**:
- In many practical applications (such as engineering, finance, environmental analysis, management science, and machine learning), it is often necessary to minimize multiple objective functions simultaneously.
- There are often conflicts between these objective functions, and it is impossible to find a unique solution that can optimize all objectives simultaneously. Therefore, a set of Pareto optimal solutions needs to be found.
2. **Limitations of existing methods**:
- **Scalarization methods**: Although widely used, they require the definition of appropriate parameters to obtain a "good" scalarization function, which requires in - depth understanding of the problem structure and is not always feasible.
- **Line search strategies**: Wolfe - type line search faces two main problems when dealing with multi - objective optimization:
- As the number of objective functions increases, multiple inequality conditions need to be satisfied simultaneously, making the selection of step size more stringent.
- The computational cost is high, especially in specific application scenarios, and frequent evaluation of objective function values will bring a significant burden.
3. **Need for new methods**:
- To overcome the above problems, the author proposes a **fixed - step - size formula** (or line - search - free scheme) to replace the traditional line search method. This method can relieve the pressure of step - size selection, avoid the computational burden associated with multiple inequalities, and improve the efficiency of the algorithm.
### Research content and contributions
- **Theoretical analysis**: The author establishes Zoutendijk - type conditions and proves the global convergence of the CG method based on Fletcher–Reeves (FR), Conjugate Descent (CD), Dai–Yuan (DY), Polak–Ribi`ere–Polyak (PRP), and Hestenes–Stiefel (HS) parameters under some mild assumptions.
- **Numerical experiments**: Through numerical experiments, the actual performance of the proposed fixed - step - size rule is demonstrated, and it is compared with the multi - objective steepest - descent method using Armijo line search and the multi - objective CG method using Wolfe - type line search.
### Formula summary
- **Form of optimization problem**:
\[
\min_{x \in \mathbb{R}^n} F(x)=(f_1(x), f_2(x), \ldots, f_m(x))^{\top}
\]
where \( F: \mathbb{R}^n \to \mathbb{R}^m \) is a continuously differentiable vector - valued function.
- **Iteration formula**:
\[
x_{k + 1}=x_k + t_k d_k, \quad k = 0,1,2,\ldots
\]
where \( d_k \in \mathbb{R}^n \) is the search direction and \( t_k>0 \) is the step size.
- **Search direction**:
\[
d_k =
\begin{cases}
v(x_k), & \text{if } k = 0, \\
v(x_k)+\beta_k d_{k - 1}, & \text{if } k\geq1,
\end{cases}
\]
where \( v(x_k) \) is the multi - objective steepest - descent direction and \( \beta_k \) is an algorithm parameter determined by different formulas.
- **Fixed - step - size formula**:
\[
t_k = -\delta\frac{\psi(x_k, d_k)}{\|d_k\|^2_{B_k}}
\]
where \( \delta>0 \) is a scalar and \( B_k