Row-major numeric matrix backed by contiguous storage. Use this for large dense grids such as regression and volatility-surface data; small process/correlation matrices remain boxed Matrix values; object matrices retain Matrix because their elements need continuation-based FFI marshalling rather than a raw contiguous pointer.

This representation interoperates directly with hmatrix without making hasquant depend on it: Numeric.LinearAlgebra.reshape cols realMatrixData makes a row-major hmatrix matrix view with no element copy. The reverse conversion through flatten is zero-copy only for a contiguous row-major hmatrix matrix; BLAS-produced or sliced matrices can require a reorder/copy.

Contiguous numeric data for APIs whose natural size is hundreds or thousands of values. The public element type stays Double; C's double representation is used only at the FFI boundary.

A storable vector known not to be empty. The constructor is deliberately hidden; build one with singletonNonEmptyVector, consNonEmptyVector, or nonEmptyVector.

Algorithm used to salvage a matrix that is not positive semi-definite before taking its pseudo square root. Higham only works for correlation matrices.

Construct a list-backed numeric matrix for small-dimensional APIs such as process correlation and diffusion matrices. Returns a boxed Matrix Double, not a RealMatrix -- despite the shared RealMatrix spelling this is not the pair of realMatrixFromVector, which is the constructor for the contiguous RealMatrix type. Reach for this one when the dimensions are small and fixed (a correlation or diffusion matrix), and for realMatrixFromVector for large dense numerical grids such as volatility surfaces and multi-asset LSM data.

Construct a row-major numeric matrix backed by contiguous storage. This is the constructor for RealMatrix; for a small correlation/diffusion matrix use boxedRealMatrix, which yields a boxed Matrix Double instead.

Regularly spaced time-grid.

Time grid with mandatory time points. Mandatory points are guaranteed to belong to the grid. No additional points are added.

Time grid with mandatory time points. Mandatory points are guaranteed to belong to the grid. Additional points are then added with regular spacing between pairs of mandatory times in order to reach the desired number of steps.

rounds a value to the precision and rule carried by the given Rounding

Minimizes an arbitrary Haskell-defined cost function via QuantLib's general-purpose Problem/OptimizationMethod machinery -- unlike calibrate, which drives a CalibratedModel's own built-in calibration error against bound CalibrationHelpers, this takes any 'RealVector -> Double' objective. The cost function crosses back into Haskell once per outer optimizer iteration over the whole parameter vector; see withCostFunction.

Eigenvalues and eigenvectors of a real symmetric matrix, computed by the symmetric threshold Jacobi algorithm. The eigenvalues come back in decreasing order, and the eigenvectors are the columns of the returned matrix: column i belongs to the i-th eigenvalue.

Pseudo square root S of a real symmetric matrix M, i.e. the matrix with S*transpose S == M. When M is not positive semi-definite the given SalvagingAlgorithm approximates it; with SalvagingNone a non-positive-semi-definite input throws instead.

Rank-reduced pseudo square root of a real symmetric matrix: the result has rank at most maxRank. If maxRank reaches the matrix size, the given percentage of the eigenvalues' sum is retained instead.

Cholesky factor L of a symmetric positive-definite matrix M, i.e. the lower-triangular matrix with L*transpose L == M. Pass True for flexible to accept a merely positive semi-definite (rank-deficient) input, whose factor is completed with zeroes rather than producing nan.

Solves M*x == b given the Cholesky factor L of M -- the first argument is the factor returned by choleskyDecomposition, not M itself.

Mean of a caller-supplied sample. Each riskStatistics* call evaluates a fresh sample.

Standard deviation of the sample (square root of riskStatisticsVariance).

Variance of the sample, N/(N-1)-corrected.

Skewness of the sample; 0 for a gaussian distribution.

Excess kurtosis of the sample; 0 for a gaussian distribution.

Minimum sample value. Throws if the sample is empty.

Maximum sample value. Throws if the sample is empty.

Variance of the sample values falling below the sample mean (Markowitz semi-variance). Throws if fewer than two sample values fall below the mean.

Square root of riskStatisticsSemiVariance.

Variance of the sample values falling below zero. Throws if fewer than two sample values fall below zero.

Square root of riskStatisticsDownsideVariance.

Empirical y-th percentile of the sample; y must lie in (0.0, 1.0].

y-th percentile assuming the sample is gaussian (mean/standard deviation matched); y must lie in (0.0, 1.0).

Empirical value-at-risk at the given centile, which must lie in [0.9, 1.0) — a loss (non-negative), the negative of the empirical (1-centile)-th percentile floored at zero.

Value-at-risk at the given centile assuming the sample is gaussian; centile must lie in [0.9, 1.0).

Empirical expected shortfall (conditional value-at-risk) at the given centile, which must lie in [0.9, 1.0): the average of the sample values below the value-at-risk threshold. Throws if the sample is empty, or if no sample value falls below the threshold.

Expected shortfall at the given centile assuming the sample is gaussian; centile must lie in [0.9, 1.0).

Empirical potential upside (the reciprocal notion of value-at-risk, floored at zero) at the given centile, which must lie in [0.9, 1.0).

Potential upside at the given centile assuming the sample is gaussian; centile must lie in [0.9, 1.0).

Variance of the sample values falling below target (Dembo/Freeman regret). Throws if fewer than two sample values fall below target.

Probability (fraction of the sample, by count) of falling below target. Throws if the sample is empty.

Average shortfall below target, i.e. the mean of target - x over sample values x below target. Throws if no sample value falls below target.

returns the time at the given index of the grid

returns the number of times on the grid

Returns all times on the grid in contiguous storage.