Need the maximum of the output in each time step as an Input

[AJAXJR24]

Hi I wanted to know that whether there is a way to define the maximum value of an output in a spatial domain as input? For example when my output is Temperature and I need it's max value over spatial domain at each timestep to implement in another equation which is coupled with the temperature equation.
Thanks

[lululxvi]

I don't have a good answer to this. Maybe you can update the max every certain iteration.

[AJAXJR24]

I don't have a good answer to this. Maybe you can update the max every certain iteration.

Many thanks for your time and reply. One Idea that I had which I'm not sure is right was I define the points by anchors in which for example 51 points for x and 51 points for y and 100 points for t and x,y points are for every elements of my time domain (51*51*100). Since thetrain_x_allis ordered by bc_points and train_x points then I define a new tensor in pde by the locations of my points.

bc_point = y[:4*51*100,0:1]
ic_point = y( 4*51*100 : 4*51*100 + 2*51*51, 0:1)
domain_point = y[4*51*100 + 2*51*51:,0:1]

then I use tf.reduce_max for every timestep I defined for example for the domain points it would be every 51*51 elements since my time changes every 51*51 times. then I assign them to a new tensor of the shape (batch_size,0:1). This tensor contains the T_max for every timestep in spatial domain. I've tried it but it seems it doesn't work.

the points are uniformly distributed by np.linspace(start,end,num)

Is there another way to define T_max in each iteration? should I use a for loop?

[lululxvi]

Your idea sounds good. I think it should work. Not sure why. Maybe you can double check train_x_all.

Using for loop might be very slow.