What is the reason the gradient is zero in the discriminator using WGAN-GP?

I’m testing why the D is not converging, so once D is stopped learning I just loaded D and I’m trying to train on real and fake data only

I printed the gradient norm on all the three loss, Dreal, Dfake,and gradient penalty. they all give zero. which means the network is dead. but how comes the network is dead and how to bring it alive again.

should I re-initiate the D network?