Framework for identifying causation, case (i): Bidirectional causality via functional coupling. 识别因果关系的框架,案例(i):通过功能耦合的双向因果关系
Bidirectional causality is analogous to the concept of “feedback” between two time series described by Granger (18) and is the primary case covered by Takens (19). Simply put, if variables are mutually coupled (e.g., predator and prey), they will cross map in both directions (Fig. 3A and fig. S1A). Thus, each variable can be estimated from the other (predator histories can estimate prey states). Figure 3B gives examples of the general case i. 双向因果关系类似于Granger(18)所描述的两个时间序列之间的“反馈”概念,是Takens(19)所涵盖的主要情况。简单地说,如果变量相互耦合(例如捕食者和猎物),它们将在两个方向上交叉映射(图3A和图S1A)。因此,每个变量都可以通过另一个来估计(捕食者的历史信息可以用来估计猎物现在的状态)。图3B给出了一般情况i的示例。
Notice that as the strength of coupling increases, information becomes more distinct in the affected variables. As a result, their manifolds will contain stronger historical signatures of the causes. In Fig. 1 (Eq. 1), for example, where β y , x ≫ β x , y β_{y,x} \gg β_{x,y} βy,x≫βx,y the much stronger effect of species X on Y implies faster convergence for predicting X than for Y (Fig. 3A). Thus, all things equal, the relative skill of cross mapping can indicate the relative strength of causative effect (Fig. 3B). 请注意,随着耦合强度的增加,结果变量的信息变得更加明显。因此,结果变量的流形将包含更多原因变量的历史特征。例如,在图1(等式1)中,其中 β y , x ≫ β x , y β_{y,x} \gg β_{x,y} βy,x≫βx,y 物种x对y的更强影响意味着预测x的收敛速度快于预测y(图3A)。因此,在所有条件相同的情况下,交叉映射的相对技能可以指示因果效应的相对强度(图3B)。