You can refer to woodridge's Introduction to Econometrics (third edition), specifically in Chapter 6: Multiple Regression Analysis: Other Issues.
The dependent variable y is 0- 1 dummy variable, and the independent variable x is a continuous variable. After logit regression, x is significantly positive, and the square x2 of x is significantly negative.
In empirical analysis, we usually assume that explanatory variables and explained variables are "linear". But in many cases, there may be a "nonlinear relationship" between explanatory variables and explained variables. In order to solve this problem, researchers often add square terms or even higher-order terms to the model (a common processing method in breakpoint regression analysis RDD).