
[数] 二次收敛
And then, we verify that this new algorithm is well defined and it has global linear convergence and local quadratic convergence with proper assumptions.
而后证明了该算法是适定的,并且在合适的假设下保持了非内点连续化算法良好的收敛性质,即具有全局线性收敛性和局部二次收敛性。
Furthermore, the algorithm has local quadratic convergence under suitable assumptions.
在适当的假设下,证明了该算法的局部二次收敛性。
The proposed methods are proved to possess the superlinear and quadratic convergence.
该方法被证明具有超线性和二次收敛性。
The local superlinear and quadratic convergence of this two models under some mild conditions without the strict complementary condition are analysed and proved.
详细分析和论证两个模型的局部超线性收敛性及二次收敛性条件,其中并不需要严格互补条件。
The ABS algorithm for nonlinear algebraic systems is generalized. The local convergence and quadratic convergence of the generalized algorithm have also been proved.
将求解非线性方程组的ABS算法加以推广,并证明了推广了的算法具有局部收敛性和二阶收敛速率。
The local quadratic convergence rate is proved under the condition that the solution is BD-regular;
在解是BD-正则条件下,证明了算法的局部二次收敛性;
The global convergence and local quadratic convergence rate are proved under standard assumptions.
我们在标准假设下证明了该算法具有全局收敛性和局部二次收敛速度。
Global and Q-quadratic convergence are proved under certain conditions.
在适当的条件下,证明了算法的全局收敛性及Q-二次收敛性。
When A is a nonnegative definite matrix with finite maximum eigenvalue, the strong convergence of random quadratic forms is established under finite fourth moment.
当A为非负定矩阵且其最大特征根有限,随机变量存在有限的四阶矩情形下,获得了二次型的强收敛。
The rate of convergence is nearly quadratic, and is quadratic under some additional conditions.
该迭代算法具有几乎二次的收敛率,在某些条件下达到了二次收敛。
Under new control conditions, we prove convergence of the quadratic minimization problem, which improves the recent results by Xu about quadratic optimization.
在新的控制条件下,证明了二次型极小化问题的迭代算法的有效性,所得结果改进了徐洪坤关于二次型优化的最新结果。
We present a new branchandbound algorithm for solving quadratic programming problem with quadratic constraints, and analyze the convergence of the algorithm.
提出了一种解带有二次约束二次规划问题的新的分枝定界算法对该算法进行了收敛性分析。
Under one global condition on the function, instead of two, the convergence determinations are established by using quadratic and cubical majorizing functions respectively.
进一步,我们利用三次优函数技巧建立了在某种意义“更优”的收敛性准则。
For the basic differential evolution algorithm, the increasing quadratic function crossover operator was added to increase the convergence speed.
在基本差分进化算法中,融入递增二次函数交叉算子以增加算法的收敛速度。
In this paper, a lower approximating algorithm of large-scale concave quadratic programming in unbounded domain is constructed. The convergence of the algorithm is discussed.
本文给出了无界域上大规模凹二次规划的一种下逼近算法,并证明了算法的收敛性。
In view of its features, we use sequence quadratic programming planning to solve the problem effectively, in which convergence carries through fast and the stability is fine.
根据问题的特点,本文采用了收敛速度快、稳定性好的序列二次规划法对模型进行了有效求解。
Chapter 4 presents an error back propagation algorithm with quadratic momentum of the multilayer forward neural networks that will speed up the error convergence velocity.
本文提出一种带二次动量项的多层前向网络误差反传算法,提高了神经网络的误差收敛速度。
二次收敛(quadratic convergence)是数值优化算法中描述收敛速度的核心概念,指迭代算法在接近解时,误差的平方随每一步迭代呈线性下降的特性。具体而言,若存在常数$C>0$,使得迭代误差满足: $$ lim{k to infty} frac{|x{k+1} - x^|}{|x_k - x^|} = C $$ 则称该算法具有二次收敛性。其中$x^*$为精确解,$x_k$为第$k$次迭代结果。
典型应用场景包括牛顿法(Newton's method),该方法通过二阶导数信息加速收敛。例如,在求解方程$f(x)=0$时,牛顿法的迭代公式为: $$ x_{k+1} = x_k - frac{f(x_k)}{f'(xk)} $$ 当初始值靠近真实解且$f'(x^*) eq0$时,误差$|x{k+1}-x^|$与$|x_k -x^|$同阶,呈现二次收敛速度。
该性质在工程优化、机器学习参数估计等领域具有重要价值。根据《数值分析》(Burden & Faires, 2016)的定义,二次收敛算法相比线性收敛(如梯度下降法)能大幅减少达到预设精度所需的迭代次数。美国数学学会(AMS)的术语库中亦将其列为非线性方程组求解的核心评价指标之一。
Quadratic convergence(二次收敛)是数值分析和优化领域中描述迭代算法收敛速度的重要概念。以下是详细解释:
二次收敛指迭代算法在接近解时,误差的平方与上一步误差成比例减小。具体来说,若第(k)步的误差为(ek),则存在常数(C>0),使得: $$ |e{k+1}| leq C cdot |e_k| $$ 这意味着每迭代一步,误差的有效位数大约翻倍,收敛速度极快。
牛顿法(Newton-Raphson方法)
在求根问题中,若函数二阶可导且初始值足够接近真解,牛顿法会呈现二次收敛。例如,求方程(f(x)=0)的根时,迭代公式为:
$$
x_{k+1} = x_k - frac{f(x_k)}{f'(x_k)}
$$
优化中的牛顿法
用于最小化目标函数时,若目标函数在极值点附近强凸且Hessian矩阵正定,牛顿法同样具有二次收敛性。
收敛类型 | 误差关系 | 速度 |
---|---|---|
线性收敛 | ( | e_{k+1} |
超线性收敛 | ( | e_{k+1} |
二次收敛 | ( | e_{k+1} |
总结来说,二次收敛是高效算法的标志,但实际应用中需权衡其快速性与对初始条件及计算资源的要求。
【别人正在浏览】