Decrease relative tolerance from 1e-5 to 1e-7

With #1819 the state goes further away from 0 with time. That meanse we rely more on the relative tolerance than the aboslute tolerance.

Reducing the default `reltol` by two orders of magnitude is enough to get only negligible leaky Terminals with occasional very brief leaks of 1e-8 m3/s. It is also enough to minimize the infiltration error to acceptable levels as can be seen in #1863 (comment).

For HWS this slows down simulation from 9.1s to 12.3s, which isn't too bad for a two orders of magnitude reduction in relative tolerance.

https://docs.sciml.ai/DiffEqDocs/stable/basics/faq/#What-does-tolerance-mean-and-how-much-error-should-I-expect

core/src/config.jl

@@ -101,7 +101,7 @@ const nodetypes = collect(keys(nodekinds))
  dtmax::Union{Float64, Nothing} = nothing
  force_dtmin::Bool = false
  abstol::Float64 = 1e-6
- reltol::Float64 = 1e-5
+ reltol::Float64 = 1e-7
  water_balance_abstol::Float64 = 1e-3
  water_balance_reltol::Float64 = 1e-2
  maxiters::Int = 1e9

core/test/docs.toml

@@ -32,7 +32,7 @@ dtmin = 0.0 # optional, default 0.0
 dtmax = 0.0 # optional, default length of simulation
 force_dtmin = false # optional, default false
 abstol = 1e-6 # optional, default 1e-6
-reltol = 1e-5 # optional, default 1e-5
+reltol = 1e-7 # optional, default 1e-7
 water_balance_abstol = 1e-3 # optional, default 1e-3
 water_balance_reltol = 1e-2 # optional, default 1e-2
 maxiters = 1e9 # optional, default 1e9

docs/concept/equations.qmd

@@ -157,8 +157,8 @@ For more a more in-depth discussion of numerical computations see [Numerical con
 
 There are many things that can influence the calculations times, for instance:
 
-- [Solver tolerance](https://diffeq.sciml.ai/stable/basics/faq/#What-does-tolerance-mean-and-how-much-error-should-I-expect):
- By default we use absolute and relative tolerances of `1e-6` and `1e-5` respectively.
+- [Solver tolerance](https://docs.sciml.ai/DiffEqDocs/stable/basics/faq/#What-does-tolerance-mean-and-how-much-error-should-I-expect):
+ By default we use absolute and relative tolerances of `1e-6` and `1e-7` respectively.
 - [ODE solvers](https://diffeq.sciml.ai/stable/solvers/ode_solve/):
  The `QNDF` method we use is robust to oscillations and massive stiffness, however other solvers should be tried as well.
 - Forcing: Every time new forcing data is injected into the model, it needs to pause.

python/ribasim/ribasim/config.py

@@ -101,7 +101,7 @@ class Solver(ChildModel):
  abstol : float
  The absolute tolerance for adaptive timestepping (Optional, defaults to 1e-6)
  reltol : float
- The relative tolerance for adaptive timestepping (Optional, defaults to 1e-5)
+ The relative tolerance for adaptive timestepping (Optional, defaults to 1e-7)
  maxiters : int
  The total number of linear iterations over the whole simulation. (Defaults to 1e9, only needs to be increased for extremely long simulations)
  sparse : bool