solve_step()

template<typename NewtonStateT , typename LLFun , typename LLTupleArgs , typename CovarMat >

void stan::math::internal::CholeskyWSolverBlock::solve_step ( NewtonStateT &  state, const LLFun &  ll_fun, const LLTupleArgs &  ll_args, const CovarMat &  covariance, int  hessian_block_size, std::ostream *  msgs  )

inline

Perform one Newton step using block-diagonal Hessian solver.

Computes the block Hessian, its square root via Schur decomposition, forms B = I + W_r * Sigma * W_r, performs Cholesky factorization, and solves for the new a vector.

Template Parameters

NewtonStateT	Type of the Newton state
LLFun	Type of the log-likelihood functor
LLTupleArgs	Type of the likelihood arguments tuple
CovarMat	Type of the covariance matrix

Parameters

[in,out]	state	Shared Newton state (modified: B, b, curr().a())
[in]	ll_fun	Log-likelihood functor
[in,out]	ll_args	Additional arguments for the likelihood
[in]	covariance	Prior covariance matrix Sigma
[in]	hessian_block_size	Size of each Hessian block (must divide theta_size)
[in,out]	msgs	Output stream for diagnostic messages (may be nullptr)

Exceptions

std::domain_error

If Hessian is not positive definite or Cholesky fails

Definition at line 632 of file laplace_marginal_density_estimator.hpp.