implementation of hard initial condition

[ryabs07]

Hi all,
I am trying to implement a Hard Initial Condition for my 2 output variables 'h' and 'u' in a 1D time dependent problem.
The initial condition is that:
if x < dam_location, h = dam_height, else h = 0
And u = 0 everywhere

I tried to implement the hard IC as follows:

def hard_constraints(a,b):
    x = a[:,0:1]
    t = a[:,1:2]
    h = b[:,0:1]
    u = b[:,1:2]
    condition = x <= dam_location
    t_zero_condition = (t == 0)  # Condition for t being 0
    # Calculate hard_IC_h for both cases
    hard_IC_h_t_zero = torch.where(condition, dam_height, downstream_depth) + h * t
    hard_IC_h_not_zero = h
    # Assign values based on conditions
    hard_IC_h = torch.where(t_zero_condition, hard_IC_h_t_zero, hard_IC_h_not_zero)
    # Calculate hard_IC_u for both cases
    hard_IC_u_t_zero = u * t
    hard_IC_u_not_zero = u
    # Assign values based on conditions for hard_IC_u
    hard_IC_u = torch.where(t_zero_condition, hard_IC_u_t_zero, hard_IC_u_not_zero)
   
    return torch.concat((hard_IC_h, hard_IC_u), dim=1)

data = dde.data.TimePDE(
    geomtime, pde, [ic_h, ic_u], num_domain=20000,  anchors=input
)
#Definition of feed forward neural network architecture
#[input layer dimension] + [hiden layer dimension]*[number of hidden layers] + [output layer dimension] , "activation function", "initializer of weights"
net = dde.nn.FNN([2] + [20] * 6 + [2], "tanh", "Glorot normal")
net.apply_output_transform(lambda x, y: hard_constraints(x,y))

The initial condition is nicely satisfied but the results for rest of the time steps are very poor and almost near to zero. I am afraid there might have been some error in applying the output transform. But I cant seem to get my head around what exactly the error is.
Any help is greatly appreciated. Thank you :)

[lululxvi]

Try soft constraints.