You are optimizing a portfolio of highly correlated tech stocks using the Markowitz model. You feed the historical covariance matrix into the optimizer, but it outputs extremely unstable, massive weights (e.g., +1000% long AAPL, -990% short MSFT). Mathematically and from an optimization perspective, what happened, and how do you fix it by modifying the objective function?

你在用马科维茨（Markowitz）模型最优化一组高度相关的科技股。你把历史协方差矩阵扔进优化器，但结果给出了极其极端的权重（比如+1000%买入苹果，-990%做空微软）。从数学和优化的角度看，发生了什么？你如何通过修改目标函数来修复它？