Skip to content

QUBO / Ising Explainer

A QUBO is a quadratic unconstrained binary optimization problem. The phrase is long, but the idea is simple:

choose 0 or 1 values that minimize an objective

QUBO models are widely used because many combinatorial problems can be expressed as binary choices plus pairwise interactions. The tutorial by Glover, Kochenberger, and Du is a useful entry point, and Lucas shows how many NP-hard problems can be written in related Ising forms.

For this demo, each binary variable represents a monitoring-site decision:

x_17 = 1 means select site 17
x_17 = 0 means do not select site 17

Rewards And Penalties

QUBO solvers minimize energy. To make a good ecological site attractive, the model gives it a negative bias, which acts like a reward.

The demo uses:

  • biological value as a reward
  • environmental coverage as a reward
  • cost as a penalty
  • redundancy between similar selected sites as a pairwise penalty
  • a target-count penalty to prefer selecting exactly 12 sites

In compact form, the model is trying to minimize an energy like:

energy =
  - biological reward
  - environmental coverage reward
  + redundancy penalty
  + cost penalty
  + target-count penalty

The signs matter. Rewards lower energy because the solver minimizes. Penalties raise energy when a candidate solution violates a preference.

Why This Is Quantum-Ready

Quantum and quantum-inspired optimizers often accept binary quadratic models. That means the same problem form can be used with a local emulator today and with other compatible solvers later.

D-Wave's Ocean documentation describes this family as binary quadratic models and notes that samplers accept such models and return variable assignments that try to minimize the problem energy. See the D-Wave sampler documentation for the solver interface and the D-Wave concepts documentation for the broader model vocabulary.

This is not proof that a quantum computer would outperform a classical method. It is a way to practice the translation from environmental synthesis to a solver-ready decision model.

What The Emulator Does With The QUBO

After the QUBO is built, the local emulator searches for a low-energy set of binary decisions. In this site-selection example, low energy means the selected network balances the rewards and penalties encoded in the model:

  • reward sites with high biological value
  • reward environmental coverage
  • penalize redundant site pairs
  • penalize high cost
  • penalize selecting too many or too few sites

Different weights change the meaning of "best." That is why the workflow is useful for experimentation: researchers can rerun the same candidate table under different ecological priorities and compare how the selected sites change.

Why Baselines Are Required

A quantum-ready formulation is only useful if it is compared with something understandable. This repo uses a greedy baseline because it is transparent: pick the next site that looks best under a simple scoring rule. More serious studies should also compare against established classical optimization tools, such as mixed integer linear programming, simulated annealing, or domain tools like Marxan and prioritizr.