1: \begin{abstract}

2:     Photonic Ising Machines constitute an emergent new paradigm of computation,

3:     geared towards tackling combinatorial optimization problems that can be

4:     reduced to the problem of finding the ground state of an Ising model.

5:     Spatial Photonic Ising Machines have proven to be advantageous for

6:     simulating fully connected large-scale spin systems. However, fine control

7:     of a general interaction matrix $J$ has so far only been accomplished

8:     through eigenvalue decomposition methods that either limit the scalability

9:     or increase the execution time of the optimization process. We introduce

10:     and experimentally validate a SPIM instance that enables direct control

11:     over the full interaction matrix, enabling the encoding of Ising

12:     Hamiltonians with arbitrary couplings and connectivity. We

13:     demonstrate the conformity of the experimentally measured Ising

14:     energy with the theoretically expected values and then proceed to solve

15:     both the unweighted and weighted graph partitioning problems, showcasing a

16:     systematic convergence to an optimal solution via simulated annealing. Our

17:     approach greatly expands the applicability of SPIMs for real-world

18:     applications without sacrificing any of the inherent advantages of the

19:     system, and paves the way to encoding the full range of NP problems that

20:     are known to be equivalent to Ising models, on SPIM devices.

21: \end{abstract}

22: