tv_precompute_probe_frame_sweep#

t3toolbox.backend.probing.tv_precompute_probe_frame_sweep(frame, ww)#

def tv_precompute_probe_frame_sweep(
        frame:   typ.Tuple[
            typ.Sequence[NDArray],          # up_tucker_cores  U. len=d
            typ.Sequence[NDArray],          # down_tt_cores    O. len=d
            typ.Sequence[NDArray],          # left_tt_cores    P. len=d
            typ.Sequence[NDArray],          # right_tt_cores   Q. len=d
        ],                                  # frame order = T3Frame.data = (up, down, left, right)
        ww:     typ.Sequence[NDArray],      # probe vectors, len=d, elm_shape=W+(Ni,)
) -> typ.Tuple[
    typ.Sequence[NDArray],  # xis.  len=d, elm_shape=W+C+(nUi,)
    typ.Sequence[NDArray],  # mus.  len=d, elm_shape=W+C+(rLi,)
    typ.Sequence[NDArray],  # nus.  len=d, elm_shape=W+C+(rR(i+1),)
    typ.Sequence[NDArray],  # etas. len=d, elm_shape=W+C+(nOi,)
]:                                          # full frame sweep -- (xis, mus, nus, etas)

The probe frame sweep (full): all four frame edge variables (xi, mu, nu, eta). The probe leaves one mode free, so its transpose’s per-mode (vector) residual must be propagated through both the left mu and right nu sweeps + the central eta combine – it cannot use the scalar-residual adjoint-state shortcut that lets apply/entries drop nu/eta (tv_precompute_apply_frame_sweep()). Reused across the probe forward / transpose of an inner solve. §6.2.2 of Alger et al. (2026); no gauge projector Π.