Black-box & Simulation-based
Cost from a simulator, digital twin, or ERP: no formula, no gradients.
In plain terms
Sometimes the cost can’t be written as a formula at all: you only have a program that, given a choice, runs a simulation and tells you how good it was. Quicopt can optimize against such a “black box” directly: it tries decisions, learns from the answers, and homes in on the best one.
The technical picture
In many real problems the cost of a decision is not a formula but the output of a simulator, a digital twin, or an ERP model. There are no gradients to differentiate and no branch-and-bound proof tree to build.
Quicopt works from input–output evaluations alone: you give it a function that maps a decision to a cost, and it searches for the best decision, treating your model as a black box.
Mathematical model▾
Minimize an objective that is only available as an oracle: evaluated by simulation, with no closed form.
Example
This problem class is coming to the API soon. Today you can run LP, QP, MILP, MINLP, QUBO, PUBO and NLP models, and if this class is the one that matters to you, tell us and we’ll keep you posted.
Benchmark
How Quicopt performs on representative black-box objectives.
Measured results for this class are being prepared and will appear here.