Constraints: Controlling the Search Space
By default, the optimizer can suggest any combination of variable values within the bounds you defined. Constraints let you restrict this search space — ruling out unsafe conditions, enforcing physical limits, or encoding domain knowledge.
Four Constraint Types
Type | What it does | Example |
Exclusion | Block specific values or ranges for one variable | Exclude temperature 100–150 °C |
Conditional Exclusion | Block values for one variable depending on another | If pH is 5–7, exclude temperature 50–100 °C |
Linear | Enforce a mathematical relationship between variables | Component A + B + C = 100% |
Subset | Limit how many variables can be active at once | At most 3 out of 6 catalysts can be used |
Read the detail article for each type:
Exclusion Constraint — block ranges or categories for a single variable
Conditional Exclusion — block values based on another variable's value
Linear Constraint — enforce weighted sums (mixtures, budgets)
Subset Constraint — control how many variables are active simultaneously
Which Constraint Do I Need?
Question 1: Is this about a single variable or a relationship between variables?
Single variable — Use Exclusion. Block specific values unconditionally.
Relationship — Go to Question 2.
Question 2: What kind of relationship?
"If X then exclude Y" — Use Conditional Exclusion. One variable's value restricts another.
Mathematical sum or budget — Use Linear Constraint. Variables must satisfy an equation.
Limit how many variables are used — Use Subset Constraint. Cap the number of active parameters.
Tips
You can add multiple constraints of any type to a single experiment.
Constraints work with both numerical and categorical variables (except Linear and Subset, which are numerical only).
Be careful not to over-constrain: if constraints eliminate too much of the search space, the optimizer may struggle to find valid suggestions.
Constraint names are optional but recommended — a descriptive name helps your team understand why the constraint exists.