Why does compute_gradient execution time increase with each epoch?

dolfin-adjoint
2

I referred to an answer in the thread Program killed when running compute_gradient on transient problem and updated my code as follows:

# Gradient computation
compute_grad_start_time = time.time()
dLdk = ReducedFunctional(L, control_k)
compute_grad_end_time = time.time()

# Retrieve vector values
vector_start_time = time.time()
grad_L_k = dLdk.derivative().vector().get_local().mean()
vector_end_time = time.time()

However, even with this modification, the execution time continues to increase as the epochs progress. Specifically, the vector_time grows as follows:

  • Epoch 1: vector_time = 0:05:52
  • Epoch 2: vector_time = 0:06:24
  • Epoch 3: vector_time = 0:06:56
  • Epoch 4: vector_time = 0:07:24