# uncomplicate.neanderthal.linalg

Contains type-agnostic linear algebraic functions roughly corresponding to the functionality usually defined in LAPACK (factorizations, solvers, etc.). This namespace works similarly to the uncomplicate.neanderthal.core namespace; see there for more details about the intended use.

### Cheat Sheet

- Linear equations and LU factorization: trf!, trf, ptrf!, tri!, tri, trs!, trs, sv!, sv, psv!, psv, con, det,.
- Orthogonal factorizations: qrf!, qrf, qrfp!, qrfp, qpf!, qpf, qrfp!, qrfp, rqf!, rqf, qlf!, qlf, lqf!, qlf!, qlf, org!, org,
- Linear least squares: ls!, ls, lse!, lse, gls!, gls.
- Eigen decomposition: ev!.
- Singular value decomposition (SVD): svd!, svd.

### Also see:

- LAPACK routines
- Linear Equation Computational Routines
- Linear Equations
- Orthogonal Factorizations (Q, R, L)
- Singular Value Decomposition
- Symmetric Eigenvalue Problems
- Other LAPACK documentation, as needed.

### con

`(con lu nrm nrm1?)`

`(con a nrm1-?)`

`(con a)`

Computes the reciprocal of the condition number of a triangularized matrix `lu`

.

If `nrm1?`

is true, computes the reciprocal based on 1 norm, if not, on infinity norm. `nrm`

and matching `nrm1?`

have to be supplied.

If the LU has been stale, or norm is not possible to compute, throws ExceptionInfo.

See related info about lapacke_?gecon.

### det

`(det a)`

Computes the determinant of a triangularized matrix `a`

.

If the matrix is not square, throws ExceptionInfo.

### ev!

`(ev! a w vl vr)`

`(ev! a w)`

`(ev! a vl vr)`

`(ev! a)`

Computes the eigenvalues and left and right eigenvectors of a matrix `a`

.

On exit, `a`

is overwritten with QR factors. The first 2 columns of a column-oriented GE matrix `w`

are overwritten with eigenvalues of `a`

. If `vl`

and `vr`

GE matrices are provided, they will be overwritten with left and right eigenvectors.

If the QR algorithm failed to compute all the eigenvalues, throws ExceptionInfo, with the information on the index of the first eigenvalue that converged. If `w`

is not column-oriented, or has less than 2 columns, throws ExceptionInfo. If `vl`

or `vr`

dimensions do not fit with `a`

’s dimensions, throws ExceptionInfo. If some value in the native call is illegal, throws ExceptionInfo.

See related info about lapacke_?geev.

### gls

`(gls a b d)`

Solves the generalized linear least squares problem using a generalized QR factorization.

Minimizes the 2-norm `||y||`

subject to constraints `d=ax+by`

.

See gls!.

### gls!

`(gls! a b d x y)`

Destructively solves the generalized linear least squares problem using a generalized QR factorization.

Minimizes the 2-norm `||y||`

subject to constraints `d=ax+by`

.

Overwrites `a`

with the R (rqf), `x`

with the solution, b with T (rqf), d with garbage, x ad y with solution to the GLS problem.

If the dimensions of arguments do not fit, throws ExceptionInfo. If `a`

and `b`

do not have the same layout (column or row oriented), throws ExceptionInfo. If the least squares cannot be computed the function throws ExceptionInfo. If some value in the native call is illegal, throws ExceptionInfo.

See related info at lapacke_?gglse.

### lqf

`(lqf a)`

Computes the LQ factorization of a GE `m x n`

matrix.

See qrf!.

See related info about lapacke_?gelqf.

### lqf!

`(lqf! a)`

Destructively computes the LQ factorization of a GE `m x n`

matrix.

See qrf!.

See related info about lapacke_?gelqf.

### ls

`(ls a b)`

Solves an overdetermined or underdetermined linear system `ax=b`

with full rank matrix using QR or LQ factorization.

Uses a temporary copy of`a`

for the factorization data: - QR if `m >= n`

; - LQ if `m < n`

.

Returns a new instance containing: - the least squares solution vectors if `m >= n`

- minimum norm solution vectors if `m < n`

.

If `a`

and `b`

do not have the same layout (column or row oriented), throws ExceptionInfo. If the least squares cannot be computed the function throws ExceptionInfo. If some value in the native call is illegal, throws ExceptionInfo.

See ls!. See related info about lapacke_?gels.

### ls!

`(ls! a b)`

Destructively solves an overdetermined or underdetermined linear system `AX = B`

with full rank matrix using QR or LQ factorization.

Overwrites `a`

with the factorization data: - QR if `m >= n`

; - LQ if `m < n`

.

Overwrites b with: - the least squares solution vectors if `m >= n`

- minimum norm solution vectors if `m < n`

.

If `a`

and `b`

do not have the same layout (column or row oriented), throws ExceptionInfo. If the least squares cannot be computed the function throws ExceptionInfo. If some value in the native call is illegal, throws ExceptionInfo.

See related info about lapacke_?gels.

### lse

`(lse a b c d)`

Solves the generalized linear least squares problem with equality constraints using RQ factorization.

See lse!

### lse!

`(lse! a b c d x)`

Destructively solves the generalized linear least squares problem with equality constraints using RQ factorization.

Minimizes the 2-norm `||ax-c||`

subject to constraints `bx=d`

.

Overwrites `a`

with the T (rqf), `x`

with the solution, b with R (rqf), d with garbage, c with the residual sum of squares.

If the dimensions of arguments do not fit, throws ExceptionInfo. If `a`

and `b`

do not have the same layout (column or row oriented), throws ExceptionInfo. If the least squares cannot be computed the function throws ExceptionInfo. If some value in the native call is illegal, throws ExceptionInfo.

See related info at lapacke_?gglse.

### org

`(org or)`

Generates the real orthogonal matrix Q of the QR, RQ, QL or LQ factorization formed by qrf! or qrfp! (or rqf! etc.).

The input is a structure containing orthogonalized matrix `:lu`

and vector `:tau`

that were previously processed by qrf! (and friends).

See related info about lapacke_?orgqr.

### org!

`(org! or)`

Destructively generates the real orthogonal matrix Q of the QR, RQ, QL or LQ factorization formed by qrf! or qrfp! (or rqf! etc.).

The input is a structure containing orthogonalized matrix `:lu`

and vector `:tau`

that were previously processed by qrf! (and friends). Overwrites the input with the appropriate portion of the resulting matrix Q.

See related info about lapacke_?orgqr.

### psv

`(psv a b)`

Solves a system of linear equations with a positive definite symmetric coefficient matrix `a`

, and multiple right sides matrix `b`

. Returns the solution matrix in a new instance.

If G is exactly singular (it can’t be used for solving a system of linear equations), throws ExceptionInfo.

See related info about psv!.

### psv!

`(psv! a b)`

Destructively solves a system of linear equations with a positive definite symmetric coefficient matrix `a`

, and multiple right sides matrix `b`

. Overwrites `b`

by the solution matrix.

Overwrites `a`

with G, and `b`

with the solution.

If G is exactly singular (it can’t be used for solving a system of linear equations), throws ExceptionInfo.

See related info about lapacke_?gesv.

### ptrf!

`(ptrf! a)`

Triangularizes a positive definite symmetric matrix `a`

. Destructively computes the Cholesky factorization of a `nxn`

matrix `a`

, and places it in a record with the key `gg`

.

Overwrites `a`

with G in the lower triangle, or G transposed in the upper triangle, depending on whether `a`

is `:lower`

or `:upper`

. Cholesky does not need pivoting.

If `a`

is exactly singular (it can’t be used for solving a system of linear equations), throws ExceptionInfo.

See related info about lapacke_?potrf.

### qlf

`(qlf a)`

Computes the QL factorization of a GE `m x n`

matrix.

See qrf!.

See related info about lapacke_?geqlf.

### qlf!

`(qlf! a)`

Destructively computes the QL factorization of a GE `m x n`

matrix.

See qrf!.

See related info about lapacke_?geqlf.

### qpf

`(qpf a)`

### qpf!

`(qpf! a)`

Destructively computes the QR factorization with pivoting of a `m x n`

matrix and places it in record that contains `:or`

and `:tau`

.

It is similar to qrf! and can replace it, with the caveat that the results have swapped columns (because of pivoting) so the solutions based on it have to be permuted back.

See related info about lapacke_?geqrf.

### qrf

`(qrf a)`

Purely computes the QR factorization of a GE `m x n`

matrix and places it in record that contains `:or`

and `:tau`

.

See qrf!

### qrf!

`(qrf! a)`

Destructively computes the QR factorization of a `m x n`

matrix and places it in record that contains `:or`

and `:tau`

.

The input is a matrix `a`

. The output overwrites the contents of `a`

. Output QR is laid out in `a`

in the following way: The elements in the upper triangle (or trapezoid) contain the `(min m n) x n`

upper triangular (or trapezoidal) matrix R. The elements in the lower triangle (or trapezoid) **below the diagonal**, with the vector `tau`

contain Q as a product of `(min m n)`

elementary reflectors. **Other routines work with Q in this representation**. If you need to compute q, call org! or org.

See related info about lapacke_?geqrf.

### qrfp

`(qrfp a)`

Purely computes the QR factorization of a GE `m x n`

matrix, with non-negative diagonal elements.

See qrf!. See related info about lapacke_?geqrfp.

### qrfp!

`(qrfp! a)`

Destructively computes the QR factorization of a GE `m x n`

matrix, with non-negative diagonal elements.

See qrf!. See related info about lapacke_?geqrfp.

### rqf

`(rqf a)`

### rqf!

`(rqf! a)`

Destructively computes the RQ factorization of a GE `m x n`

.

See qrf!.

See related info about lapacke_?gerqf.

### sv

`(sv a b)`

Solves a system of linear equations with a square coefficient matrix `a`

, and multiple right hand sides matrix `b`

. Returns the solution in a new matrix instance.

If `a`

is symmetric, tries to do Cholesky factorization first, and only does LDLt if it turns out not to be positive definite.

If U is exactly singular (it can’t be used for solving a system of linear equations), throws ExceptionInfo.

See related info about sv!.

### sv!

`(sv! a b)`

Destructively solves a system of linear equations with a square coefficient matrix `a`

, and multiple right hand sides matrix `b`

. Overwrites `b`

by the solution matrix.

Overwrites `a`

with L and U, and `b`

with the solution. L is stored as a lower unit triangle, and U as an upper triangle. Pivot is not retained. If you need to reuse LU, use trf, and trs, or their destructive versions.

If U is exactly singular (it can’t be used for solving a system of linear equations), throws ExceptionInfo.

See related info about lapacke_?gesv.

### svd

`(svd a u? vt?)`

`(svd a)`

Computes the singular value decomposition of a matrix `a`

, and returns a SVD record containing `:sigma`

(the singular values), `u`

(if `u?`

is true) and `vt`

(if `vt?`

is true).

If the reduction to bidiagonal form failed to converge, throws ExceptionInfo, with the information on the number of converged superdiagonals. If some value in the native call is illegal, throws ExceptionInfo.

See related info about svd!

### svd!

`(svd! a sigma u vt superb)`

`(svd! a sigma superb)`

Computes the singular value decomposition of a matrix `a`

.

On exit, `a`

’s contents is destroyed, or, if `u`

or `vt`

are `nil`

, overwritten with U or transposed V singular vectors of `a`

. `s`

is a diagonal matrix populated with sorted singular values. If the factorization does not converge, a diagonal matrix `superb`

is populated with the unconverged superdiagonal elements (see LAPACK documentation). If called without `u`

and `vt`

, U and transposed V are not computed.

If the reduction to bidiagonal form failed to converge, throws ExceptionInfo, with the information on the number of converged superdiagonals. If some value in the native call is illegal, throws ExceptionInfo.

See related info about lapacke_?gesvd.

### trf

`(trf a)`

Triangularizes a non-TR matrix `a`

. Computes the LU (or LDLt, or UDUt) factorization of a `mxn`

matrix `a`

, and places it in a record that contains `:lu`

, `:a`

and `:ipiv`

. If `a`

is positive definite symmetric, compute the Cholesky factorization, that contains only `:gg`

and no ipiv is needed.

Pivot is placed into the `:ipiv`

, a vector of **integers or longs** (if applicable).

If U is exactly singular (it can’t be used for solving a system of linear equations), throws ExceptionInfo.

### trf!

`(trf! a)`

Triangularizes a non-triangular matrix `a`

. Destructively computes the LU (or LDLt, or UDUt, or GGt) factorization of a `mxn`

matrix `a`

, and places it in a record that contains `:lu`

and `:ipiv`

.

Overwrites `a`

with L and U. L is stored as a lower unit triangle, and U as an upper triangle. Pivot is placed into the `:ipiv`

, a vector of **integers or longs**.

See related info about lapacke_?getrf.

### tri

`(tri a)`

Computes the inverse of a triangularized matrix `a`

. Returns the results in a new matrix instance.

If U is exactly singular (it can’t be used for solving a system of linear equations), throws ExceptionInfo. If the matrix is not square, throws ExceptionInfo.

See related info about tri!.

### tri!

`(tri! a)`

Destructively computes the inverse of a triangularized matrix `a`

. Overwrites `a`

(or `:lu`

) with a `nxn`

inverse.

If U is exactly singular (it can’t be used for solving a system of linear equations), throws ExceptionInfo. If the matrix is not square, throws ExceptionInfo.

See related info about lapacke_?getri.

### trs

`(trs a b)`

Solves a system of linear equations with a triangularized matrix `a`

, with multiple right hand sides matrix `b`

. Returns the results in a new matrix instance.

See related info about trs!.

### trs!

`(trs! a b)`

Destructively solves a system of linear equations with a triangularized matrix `a`

, with multiple right hand sides matrix `b`

. Overwrites `b`

by the solution matrix.

See related info about [https://software.intel.com/en-us/mkl-developer-reference-c-getrs).