Skip to content

Verify that chi-squared value computed in gain matrix updater is correct #986

@stephenswat

Description

@stephenswat

We currently compute the Kálmán update step using the canonical (i.e. mentioned by Frühwirth) equation $C_k = (I - KH)C_{k-1}$, but we know that this is not correct if $K$ is non-optimal or if $C_{k-1}$ is not true, and both of these conditions fail in traccc. The correct equation, therefore, is $C_k = (I - KH)C_{k-1}(I - KH)^\intercal + KVK^\intercal$ which holds for all (even non-optimal) $K$.

Now, we compute the $\chi^2$ value similarly, using $R = (I - HK)V$. Notice that this is eerily similar to the invalid covariance update, and we are not certain that this holds for arbitrary $K$. Is there a $\chi^2$ computation that holds for all $K$?

@andiwand suggested $R = (I - HK)V(I - HK)^\intercal + HCH^\intercal$.

We should verify that this is correct and, if so, migrate to it.

Metadata

Metadata

Assignees

No one assigned

    Labels

    bugSomething isn't workingphysicsPhysics performance monitoring

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions