Then I have to following deductions
(1) 2D graph state on a 2D lattice is universal for measurement based quantum computation, so that all quantum computations can be achieved by a 2D graph state with local measurements.
(2) Branching MERA state can be classically simulated in the meaning that local observables on such a state can be computed efficiently.
(3) If (1)(2) are true, then the local measurement on the 2D graph state can be classically simulated.
(4) From (1)(2)(3), it seems that all quantum computations can be classically simulated.
This of course not true. What's wrong with my deduction?