You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Despite the fact that, as it is, the Convergent Born Series does not allow for an heterogeneous density, it is still a powerful technique, especially for low-contrast sound speed media.
In absence of density heterogeneity the Helmholtz operator is self-adjoint (up to a change of sign for the absorption term). This means that it can be possible to override the vjp operation from jax with one that uses the CBS.
Potential things to look for:
I suspect this will not allow for backpropagation through the solver operations, so there should be a clear warning for the user if one tries to find such gradients
Make sure to calculate gradients for all remaining terms, including source terms
It would be very nice to be able to mix and match the forward and adjoint solver. For example, solving for the forward map using Waveholtz and the backward map with CBS.
The text was updated successfully, but these errors were encountered:
Despite the fact that, as it is, the Convergent Born Series does not allow for an heterogeneous density, it is still a powerful technique, especially for low-contrast sound speed media.
In absence of density heterogeneity the Helmholtz operator is self-adjoint (up to a change of sign for the absorption term). This means that it can be possible to override the
vjp
operation from jax with one that uses the CBS.Potential things to look for:
The text was updated successfully, but these errors were encountered: