Solver Options
Reference of the NANDRAD solver's performance options: time integration (CVODE, Euler), relative tolerance, time step limits, linear equation solvers KLU/GMRES/BiCGStab, and ILU preconditioner
Overview
The solver options are located in the simulation view on the Dynamic Simulation page in the Performance options tab. They affect only the computation speed and robustness of the numerical solution — for a converged computation not the physical results (within the tolerances). Empty input fields mean “Auto”: the solver then uses its default values.
General options
| Field | Default | Meaning |
|---|---|---|
| Relative tolerance: | 1e-5 | Accuracy of the time integration; smaller values yield more accurate results at the cost of longer computation time |
Time integration method
| Field | Default | Meaning |
|---|---|---|
| Integrator: | CVODE | Selection: CVODE-based solver [CVODE], Explicit Euler method [ExplicitEuler], Implicit Euler method [ImplicitEuler] |
| Maximum method order (1-5): | 5 | Limits the BDF order of CVODE; higher orders are more efficient but can introduce oscillations |
| Nonlinear iteration convergence coefficient: | 0.1 | Scaling factor of the Newton convergence test; smaller values enforce tighter convergence |
| First time step: | 0.1 s | First integration step; empty = automatic choice |
| Time step size limit (model consistency): | 1e-8 s | Smallest allowed time step; if the solver needs smaller steps, it aborts — useful for detecting model inconsistencies |
| Maximum time step size: | 1 h | Largest allowed time step; limits how far the solver may leap ahead |
CVODE is an implicit multi-step solver (SUNDIALS) with variable order and step size and is the default choice for the stiff systems of equations of building simulation. With the explicit Euler method, the settings for the equation solver and preconditioner do not apply.
Solver for linear systems of equations
| Field | Default | Meaning |
|---|---|---|
| Solver for linear systems of equations: | KLU | Selection: Dense solver [Dense], KLU sparse solver [KLU], Iterative GMRES solver [GMRES], Iterative BICGSTAB solver [BiCGStab] |
| Maximum dimension of the Krylov subspace: | 50 | Iterative solvers only (GMRES); larger values improve convergence but increase memory requirements |
| Linear iteration convergence coefficient: | 0.05 | Scaling factor of the linear convergence test |
KLU is a direct, sparse solver and usually the most robust choice. GMRES and BiCGStab are iterative Krylov methods that can be faster for very large systems but require a preconditioner.
Preconditioner for iterative equation-system solvers
| Field | Default | Meaning |
|---|---|---|
| Preconditioner: | ILU | Incomplete LU preconditioner [ILU] |
| Fill level (ILU): | 3 | Accuracy of the incomplete LU factorization; higher values improve the preconditioning at the cost of more memory |
These fields are only active when an iterative equation solver is selected.
Solver for hydraulic networks
The Solver for hydraulic networks section is only visible if the project contains a heat network or a supply system with a hydraulic network:
| Field | Default | Meaning |
|---|---|---|
| Newton convergence tolerance: | 0.001 | Absolute convergence tolerance of the Newton solver for the hydraulic network |
| Newton initial conditions: Use results of the previous time step as initial condition | enabled | Take the initial values of the Newton method from the last time step |
| Compute the system only in fixed time steps: with Set fixed time step: | disabled; 3 min | Do not solve the hydraulics in every integration step, but at fixed intervals — faster, but less accurate |
Good to know:
For pure building simulations, the default settings (CVODE + KLU, relative tolerance 1e-5) are almost always the best choice. If a simulation runs unusually slowly, the “Write statistics for every internal solution step” option on the start page helps to find the bottleneck before adjusting tolerances — tolerances that are too large can obscure or amplify control oscillations (thermostat, shading control).