Interface ModelData

Linear objective coefficient for each variable, in column order. The linear term is sum(colCost[j] * x[j]), or c'x. The array must contain exactly numCols finite numbers. Use 0 when a variable has no linear objective contribution; infinity is not a substitute for a bound.

Lower bound for each variable: colLower[j] <= x[j]. The array must have exactly numCols entries. Use -highs.infinity for no lower bound. A finite lower bound equal to colUpper[j] fixes the variable; normally each lower bound must not exceed its corresponding upper bound.

Optional variable names in exact column order, with exactly numCols strings. Names are copied and later used by name lookup and serialization. Empty, duplicate, or format-invalid names may be rejected by HiGHS.

Upper bound for each variable: x[j] <= colUpper[j]. The array must have exactly numCols entries. Use highs.infinity for no upper bound. A finite upper bound equal to colLower[j] fixes the variable; normally each upper bound must not be below its corresponding lower bound.

Optional symmetric matrix Q adding 0.5 * x'Qx to the objective. Its dimension must equal numCols. Omission makes the objective linear; supplying it is independent of integrality, so both may be present. See HessianInput for triangular and square storage rules.

Domain code for each variable in column order. It must have exactly numCols entries, each one of 0 continuous, 1 integer, 2 semi-continuous, 3 semi-integer, or 4 implicit integer. Prefer named highs.constants.variableType values. Omission makes every variable continuous; it does not remove a supplied quadratic Hessian.

Sparse coefficient matrix A used by every row activity Ax. Its numRows and numCols must exactly match this model. See SparseMatrixInput for the complete CSC/CSR encoding, offset, index, duplicate, and coefficient rules.

Optional name for the model as a whole. It is metadata used by supported serializers and does not affect optimization. The string is copied.

Number of decision variables, also called columns. It must be a non-negative signed 32-bit integer. colCost, colLower, colUpper, and supplied integrality/colNames vectors must have exactly this length; matrix.numCols and a supplied hessian.dimension must equal it.

Number of linear constraints, also called rows. It must be a non-negative signed 32-bit integer. rowLower, rowUpper, and supplied rowNames vectors must have exactly this length, and matrix.numRows must equal it.

Constant added to the objective independently of all variable values. It changes the reported objective and objective-based stopping criteria but not which x is feasible. Omission means 0.

Lower bound for each row activity: rowLower[i] <= sum(A[i,j] * x[j]). This is a bound on the computed left-hand side Ax, not a variable value or slack. The array must have exactly numRows entries. Use -highs.infinity when a row has no lower bound.

Optional constraint names in exact row order, with exactly numRows strings. Names are copied and later used by name lookup and serialization; empty, duplicate, or format-invalid names may be rejected.

Upper bound for each row activity: sum(A[i,j] * x[j]) <= rowUpper[i]. The array must have exactly numRows entries. Use highs.infinity when a row has no upper bound. Equal finite lower/upper values represent an equality constraint.

Whether HiGHS minimizes or maximizes the complete objective. Omission means minimize. Use highs.constants.objectiveSense.minimize or .maximize; plain numeric 1/-1 literals are deliberately not accepted by the type. The sense applies to the offset, linear term, and quadratic term together.