API

Solver Interface

The main interface to the celerite solver is via the celerite::solver::CholeskySolver class but many of the key interfaces are abstracted into the celerite::solver::Solver.

template<typename T>
class celerite::solver::Solver

Subclassed by celerite::solver::CholeskySolver< T, SIZE >, celerite::solver::DirectSolver< T >

Public Functions

void compute(const T &jitter, const vector_t &a_real, const vector_t &c_real, const vector_t &a_comp, const vector_t &b_comp, const vector_t &c_comp, const vector_t &d_comp, const Eigen::VectorXd &A, const Eigen::MatrixXd &U, const Eigen::MatrixXd &V, const Eigen::VectorXd &x, const Eigen::VectorXd &diag) = 0

Compute the matrix and factorize

This method is overloaded by subclasses to provide the specific implementation.

Parameters

  • jitter: The jitter of the kernel.

  • a_real: The coefficients of the real terms.

  • c_real: The exponents of the real terms.

  • a_comp: The real part of the coefficients of the complex terms.

  • b_comp: The imaginary part of the of the complex terms.

  • c_comp: The real part of the exponents of the complex terms.

  • d_comp: The imaginary part of the exponents of the complex terms.

  • x: The sorted array of input coordinates.

  • diag: An array that should be added to the diagonal of the matrix. This often corresponds to measurement uncertainties and in that case, diag should be the measurement variance (i.e. sigma^2).

T dot_solve(const Eigen::VectorXd &b) const

Solve the system b^T . A^-1 . b

A previous call to solver::Solver::compute defines a matrix A and this method solves b^T . A^-1 . b for a vector b.

Parameters

  • b: The right hand side of the linear system.

T log_determinant() const

Get the log determinant of the matrix.

bool computed() const

Flag indicating if compute was successfully executed.

void compute(const T &jitter, const vector_t &a_real, const vector_t &c_real, const vector_t &a_comp, const vector_t &b_comp, const vector_t &c_comp, const vector_t &d_comp, const Eigen::VectorXd &x, const Eigen::VectorXd &diag)

Compute the matrix and factorize for purely real alphas

Parameters

  • jitter: The jitter of the kernel.

  • a_real: The coefficients of the real terms.

  • c_real: The exponents of the real terms.

  • a_comp: The real part of the coefficients of the complex terms.

  • b_comp: The imaginary part of the coefficients of the complex terms.

  • c_comp: The real part of the exponents of the complex terms.

  • d_comp: The imaginary part of the exponents of the complex terms.

  • x: The sorted array of input coordinates.

  • diag: An array that should be added to the diagonal of the matrix. This often corresponds to measurement uncertainties and in that case, diag should be the measurement variance (i.e. sigma^2).

void compute(const T &jitter, const vector_t &a_real, const vector_t &c_real, const vector_t &a_comp, const vector_t &c_comp, const vector_t &d_comp, const Eigen::VectorXd &x, const Eigen::VectorXd &diag)

Compute the matrix and factorize for purely real alphas

Parameters

  • jitter: The jitter of the kernel.

  • a_real: The coefficients of the real terms.

  • c_real: The exponents of the real terms.

  • a_comp: The real part of the coefficients of the complex terms.

  • c_comp: The real part of the exponents of the complex terms.

  • d_comp: The imaginary part of the exponents of the complex terms.

  • x: The sorted array of input coordinates.

  • diag: An array that should be added to the diagonal of the matrix. This often corresponds to measurement uncertainties and in that case, diag should be the measurement variance (i.e. sigma^2).

void compute(const T &jitter, const vector_t &a_real, const vector_t &c_real, const Eigen::VectorXd &x, const Eigen::VectorXd &diag)

Compute the matrix and factorize for a set of purely real terms

Parameters

  • jitter: The jitter of the kernel.

  • a_real: The coefficients of the real terms.

  • c_real: The exponents of the real terms.

  • x: The sorted array of input coordinates.

  • diag: An array that should be added to the diagonal of the matrix. This often corresponds to measurement uncertainties and in that case, diag should be the measurement variance (i.e. sigma^2).

void compute(const T &jitter, const vector_t &a_comp, const vector_t &c_comp, const vector_t &d_comp, const Eigen::VectorXd &x, const Eigen::VectorXd &diag)

Compute the matrix and factorize for purely complex terms with real alphas

Parameters

  • jitter: The jitter of the kernel.

  • a_comp: The real part of the coefficients of the complex terms.

  • c_comp: The real part of the exponents of the complex terms.

  • d_comp: The imaginary part of the exponents of the complex terms.

  • x: The sorted array of input coordinates.

  • diag: An array that should be added to the diagonal of the matrix. This often corresponds to measurement uncertainties and in that case, diag should be the measurement variance (i.e. sigma^2).

void compute(const T &jitter, const vector_t &a_comp, const vector_t &b_comp, const vector_t &c_comp, const vector_t &d_comp, const Eigen::VectorXd &x, const Eigen::VectorXd &diag)

Compute the matrix and factorize for purely complex terms

Parameters

  • jitter: The jitter of the kernel.

  • a_comp: The real part of the coefficients of the complex terms.

  • b_comp: The imaginary part of the coefficients of the complex terms.

  • c_comp: The real part of the exponents of the complex terms.

  • d_comp: The imaginary part of the exponents of the complex terms.

  • x: The sorted array of input coordinates.

  • diag: An array that should be added to the diagonal of the matrix. This often corresponds to measurement uncertainties and in that case, diag should be the measurement variance (i.e. sigma^2).

Fast Solver

The celerite::solver::CholeskySolver is an \(\mathcal{O}(N)\) solver that exploits the semi-separable structure of the matrix to solve and compute determinants.

template<typename T, int SIZE = Eigen::Dynamic>
class celerite::solver::CholeskySolver : public celerite::solver::Solver<T>

Public Functions

void compute(const T &jitter, const vector_t &a_real, const vector_t &c_real, const vector_t &a_comp, const vector_t &b_comp, const vector_t &c_comp, const vector_t &d_comp, const Eigen::VectorXd &A, const Eigen::MatrixXd &U, const Eigen::MatrixXd &V, const Eigen::VectorXd &x, const Eigen::VectorXd &diag)

Compute the Cholesky factorization of the matrix

Parameters

  • jitter: The jitter of the kernel.

  • a_real: The coefficients of the real terms.

  • c_real: The exponents of the real terms.

  • a_comp: The real part of the coefficients of the complex terms.

  • b_comp: The imaginary part of the of the complex terms.

  • c_comp: The real part of the exponents of the complex terms.

  • d_comp: The imaginary part of the exponents of the complex terms.

  • x: The sorted array of input coordinates.

  • diag: An array that should be added to the diagonal of the matrix. This often corresponds to measurement uncertainties and in that case, diag should be the measurement variance (i.e. sigma^2).

matrix_t solve(const Eigen::MatrixXd &b) const

Solve the system b^T . A^-1 . b

A previous call to solver::CholeskySolver::compute defines a matrix A and this method solves b^T . A^-1 . b for a vector b.

Parameters

  • b: The right hand side of the linear system.

T dot_solve(const Eigen::VectorXd &b) const

Solve the system b^T . A^-1 . b

A previous call to solver::CholeskySolver::compute defines a matrix A and this method solves b^T . A^-1 . b for a vector b.

Parameters

  • b: The right hand side of the linear system.

matrix_t dot_L(const Eigen::MatrixXd &z) const

Compute the dot product of the square root of a celerite matrix

This method computes L.z where A = L.L^T is the matrix defined in compute.

Parameters

  • z: The matrix z from above.

matrix_t dot(const T &jitter, const vector_t &a_real, const vector_t &c_real, const vector_t &a_comp, const vector_t &b_comp, const vector_t &c_comp, const vector_t &d_comp, const vector_t &A, const matrix_t &U, const matrix_t &V, const Eigen::VectorXd &x, const Eigen::MatrixXd &z)

Compute the dot product of a celerite matrix with another matrix

Parameters

  • jitter: The jitter of the kernel.

  • a_real: The coefficients of the real terms.

  • c_real: The exponents of the real terms.

  • a_comp: The real part of the coefficients of the complex terms.

  • b_comp: The imaginary part of the of the complex terms.

  • c_comp: The real part of the exponents of the complex terms.

  • d_comp: The imaginary part of the exponents of the complex terms.

  • x: The sorted array of input coordinates.

  • z: The matrix that will be dotted with the celerite matrix.

vector_t predict(const Eigen::VectorXd &y, const Eigen::VectorXd &x) const

Compute the conditional mean for a GP in O(N)

This method computes K_*.(K+sigma)^{-1}.y where K is the matrix defined in compute.

Parameters

  • y: The data vector to condition on

  • x: The input coordinates to compute the predictions at