why there is no gradient when the leaf tensor is updated - pytorch

Question

The following is my toy example
import torch
x = torch.tensor(3.0, requires_grad = True)
y = x**2
y.backward(retain_graph = True)
print(x.grad)
x = x + 4
y.backward(retain_graph = True)
print(x.grad)
The first print prints the gradient of x, while the second print prints nothing. Why the gradient of x disappear after x is updated by x = x + 4? Thanks.
Newly added questions:
The following code could do what I want, which update x iteratively. However, I need to add x.requires_grad = True every time it is updated. Is there any better way without using x.requires_grad = True? Thanks.
x = torch.tensor(3.0, requires_grad = True)
y = x**2
y.backward(retain_graph = True)
with torch.no_grad():
x = x + x.grad
x.requires_grad = True
y = x**2
y.backward(retain_graph = True)
print(x.grad)
Update: my solution
x = torch.tensor(3.0, requires_grad = True)
y = x**2
y.backward(retain_graph = True)
print(x.grad)
x.data = x.data + x.grad.data
x.grad.zero_()
y = x**2
y.backward(retain_graph = True)
print(x.grad)
The result of the code is
tensor(6.)
tensor(18.)
, which is exactly what I want to have. Thanks.

Answer 1

That is mainly because x=x+4 doesn't update the tensor; it creates a new tensor and assigns it to the variable x.
I changed the code to print the data pointer of the tensor in x before and after x=x+4 as follows:
import torch
x = torch.tensor(3.0, requires_grad = True)
y = x**2
y.backward(retain_graph = True)
print(x.data_ptr())
print(x.grad)
z = x
x = x + 4
y.backward(retain_graph = True)
print(x.data_ptr())
print(x.grad)
print(x is z)
print(z.grad)
The output was:
3007755542592
tensor(6.)
3007755541184
None
False
tensor(12.)
First, you will notice that the data pointer of the tensor in x has changed after x=x+4. The is because x+4 created a new tensor which is now what x holds.
Second, I kept the original tensor in another variable z. As you can see x is z returns False and the gradient of z has twice the gradient since y.backward(retain_graph = True) was called twice. z now holds the tensor which was in x before the line x=x+4.