ut3_add#

t3toolbox.backend.ut3_linalg.ut3_add(x, y)#
def ut3_add(x: UT3Data, y: UT3Data) -> UT3Data:  # z = x + y (ranks add; NOT squashed)

Add two uniform Tucker tensor trains (direct sum): concatenate Tucker supercores along the rank axis, block-diagonalize the TT supercores, and concatenate the masks (shape via OR). Vectorized over (d,)+stack (the xnp.block acts on the last 3 axes). x,``y`` need not share padded n/r; only N, d, stack_shape must match (the frontend enforces that).

Parameters:

  • x (UT3Data)

  • y (UT3Data)

Return type:

UT3Data