Physics-Based Preconditioning of a Jacobian-Free Newton–Krylov Solver for the HTR-PM Secondary Circuit
Year: 2026
Type: article
Abstract: We present a mathematical framework and a numerical solution architecture for the steady-state secondary circuit of the HTR-PM power plant and use it to evaluate physics-based preconditioning for strongly coupled thermo-hydraulic networks. The governing equations are formulated as a non-linear algebraic system partitioned into four physical fields: mass flows, pressures, specific enthalpies, and the internal state of a three-zone once-through helical steam generator, whose economiser and evaporator lengths and helium interface temperatures are solved simultaneously with the cycle. Water and steam properties are evaluated from the IAPWS-IF97 Region 1 and Region 2 basic equations, with exact Newton inversions for all single-phase (P,h) and (P,s) states. The system is solved by a Jacobian-free Newton–Krylov (JFNK) method with an Eisenstat–Walker forcing term and Armijo backtracking. Two preconditioners are developed and compared: a field-split preconditioner formed by inverting the four physical field blocks, and an Additive Schwarz Preconditioned Inexact Newton (ASPIN) strategy that performs subdomain Newton sweeps over six physics-motivated subdomains. On the design case, field-split preconditioning needs a single GMRES iteration per Newton step, reducing the Krylov cost of a complete solve by a factor of more than 30 relative to the unpreconditioned iteration and about 12 relative to point Jacobi (diagonal) preconditioning, and it collapses the spectrum of the preconditioned operator onto λ=1, with max|Imλ| falling from 4.26 to below 10−9. The condition number alone does not distinguish field-split from point Jacobi; the discriminating property is eigenvalue clustering. In a randomised restart study, in which each component of the initial guess is perturbed by an independent log-uniform factor, field-split converges from 149 of 150 far-from-solution starts within 25 Newton steps, at 6-9 times lower Krylov cost than point Jacobi, and the remaining start converges when the budget is raised. ASPIN converges from 148 of 150 starts, including the only genuine stagnation observed. Our conclusions speak in favour of the physical rather than algebraic considerations for the development of preconditioners for strongly coupled thermal–hydraulics problems.
Source: Journal of Nuclear Engineering
DOI: 10.3390/jne7040067
Language: English
Subfield: Nuclear Energy and Engineering
Field: Energy
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $1,207
OpenAlex ID: W7236712197
