Power needed to detect an interaction In my regression model, I am testing for main effects and interaction on Y as follows:
X, X1, X*X1
(X1 is gender and there are 200 males and 250 females in the dataset.)
The interaction is not significant but one main effect (X) is significant.
My research context suggests that there should be an interaction (and the graph shows non parallel lines - see here).
I understand one of the reasons for the above outcome would be lack of power.
Question: Does detection of an intereaction require more or less power than detection of main effects? (By detection, I mean statistically significant)
I am using the general linear function in SPSS. 
 A: You need a significantly larger sample size to detect an effect for an interaction. To detect an effect of size $d$ for an interact, you need a sample size that is about 4 times larger than the sample size required to detect a main effect of size $d$.
This is because for the interaction you're essentially taking the difference in the difference between two groups so your standard error has four terms instead of two. For a clear discussion of this, see 16.4 in Regression and Other Stories available online here.
It should also be noted though that the sort of post-hoc power analysis you seem to be suggesting here is not advisable. You cannot reliably estimate the power needed for an analysis after the fact based only on your estimate for $d$.
A: If we use the question by Macro:

"To achieve the same power, does one require a greater sample size when testing an interaction than when testing a main effect?"

It does not necessarily need a bigger sample with "that" many in each dummy variable category, but you have to be aware of the problem of multicollinearity, since you create multicollinearity with the interaction term. This can lead to insignificant variables, in your example your interaction term. This problem can be cured with centering. See how this is done here with an example. Some  minor explanation of the different centering techniques]Aiken's book on interpreting interaction terms(see below for link)
With centering your R-squared values will remain the same.
Multicolliearity:
Aiken's Book:
