why incresing the regularization parameter (lambda) may increase or decrease the train error but it always increase the test error?

Question

while I was calculating the test and train error with different lambda values in the regularization equation. the train error kept decreasing with increasing lambda until a