Interface HessianInput

Number of rows and columns in square matrix Q. When the Hessian is passed to a model, this must equal that model's numCols, because there is one Hessian coordinate per decision variable. It must be a non-negative signed 32-bit integer.

Selects which entries of symmetric matrix Q are supplied.

"triangular" stores only the lower triangle in compressed-column form, so every entry in column j must have row index i >= j. HiGHS infers the mirrored upper triangle. "square" stores every matrix entry, so each off-diagonal value must be supplied in both symmetric coordinates and the resulting matrix must be symmetric.

Zero-based row coordinate of each packed Hessian value. Position k represents matrix coordinate (indices[k], j), where column j is the unique column whose interval starts[j] <= k < starts[j + 1] contains it.

Every index must satisfy 0 <= indices[k] < dimension. In triangular mode it must additionally satisfy indices[k] >= j. A column must not contain a duplicate row index. The array length must equal values.length and the final starts entry.

Compressed-column boundaries into indices and values. Hessian column j occupies positions from starts[j] inclusive to starts[j + 1] exclusive. Equal adjacent values represent an empty column.

The array must have exactly dimension + 1 entries, begin with 0, be nondecreasing, and end with indices.length === values.length === starts[dimension]. For example, [0, 2, 3] means column 0 has two stored entries and column 1 has one.

Entries of symmetric matrix Q, parallel to indices. values[k] is the value at the row given by indices[k] and the column determined by starts. These are entries of Q itself in objective 0.5 * x'Qx; do not halve diagonal values or double off-diagonal values.

Values must be finite numbers and the length must equal indices.length and the final starts entry. For triangular encoding of Q = [[2, 1], [1, 4]], use starts: [0, 2, 3], indices: [0, 1, 1], and values: [2, 1, 4].