choleskySolve
choleskySolve(
L,b):number[] |number[][]
Defined in: cholesky.js:121
Solve A x = b (or A X = B) given the Cholesky factor L of A (A = L L^T), by forward substitution (L y = b) then back substitution (L^T x = y).
Accepts either a single right-hand side vector or a matrix of them. Passing the whole set at once matters: the alternative — calling this once per column — repeats the triangular walk’s setup per column, which turns building an inverse into markedly more work than it needs. Measured on a 340x340 factor, one call with 340 right-hand sides against 340 single-vector calls: 76 ms against 371 ms.
Reads the factor’s nested rows directly rather than copying it to flat storage, and divides rather than multiplying by a reciprocal so the scaling step stays exact. The two paths accumulate in different orders — a scalar per element for one right-hand side, an axpy across the row for many — so they agree to roundoff (~1e-19 relative on the sizes measured) rather than bit for bit.
Parameters
L
number[][]
Lower triangular factor from cholesky()
b
number[] | number[][]
Right-hand side vector, or an n x k matrix of right-hand sides
Returns
number[] | number[][]
Solution, matching the shape
of b