New Features in BOUT++ v5.0

BOUT++ v5.0 is a new major release, adding tons of new features, improving existing code, as well as removing some old and deprecated things. There are some breaking changes which will require modifications to your physics model, but for the vast majority we have provided tooling to automate this as much as possible.

3D Metrics

Up until now, BOUT++ has been limited to varying the metric components only in the XY plane. This release now introduces 3D metrics as a compile-time option, allow simulations of devices such as stellarators.

To enable 3D metrics, build BOUT++ like:

./configure --enable-metric-3D

or, with CMake:

cmake . -B build -DBOUT_ENABLE_METRIC_3D=ON

Changes

Types

Adding 3D metrics to BOUT++ has been a substantial effort, requiring many changes to a significant amount of the source code. The main change is that the metric components, g11, g22, and so on, as well as the grid spacing, dx, dy, dz, have changed from Field2D to Coordinates::FieldMetric: a type alias for either Field3D or Field2D depending on if BOUT++ was built with or without 3D metrics respectively.

Note

Coordinates::dz has also changed to be Field2D even without 3D metrics. This is a breaking change, in that it may be necessary to change user code in order to keep working. If you don’t use 3D metrics, wrapping the use of dz, and similarly Coordinates::zlength(), in a call to getUniform() will return a BoutReal.

The use of Coordinates::FieldMetric has been followed through the rest of the code base. If a metric component enters an expression that previously contained only Field2D and BoutReal types, the result is now a Coordinates::FieldMetric. This means that functions that previously both took and returned a Field2D now return a Coordinates::FieldMetric (we could have chosen to make the return type auto instead and rely on the compiler to deduce the correct type, but we have chosen to make the dependence on the metric dimensionality more explicit).

Because almost any operation on a vector involves the metric, the individual components of Vector2D are now also of type Coordinates::FieldMetric. Realistically, the use of Vector2D in a model making use of 3D metrics is probably ill-advised.

Indexing

3D metrics also requires changes in how fields are indexed. In BOUT_FOR loops, generally no changes are required, as they already do The Right Thing. In other cases, simply changing, for example, dx(x, y) to dx(x, y, z) is sufficient: in the 2D metric case, the third index is accepted and discarded.

Many methods and operators have been upgraded to deal with 3D metrics. For example, the LaplaceXZpetsc implementation has been modified to deal with non-zero g_{xz} terms.

FIXME WHEN COORDINATES REFACTORED

In order to simplify a lot of code, the call to Coordinates::geometry(), which calculates the connection coefficients Coordinates::G1_11 and so on, has been moved out of the Coordinates constructor. This is because computing the coefficients involves derivatives which requires Coordinates and causes all sorts of headaches and shims. As most users do not call the constructor themselves anyway, this change should not be much of an issue.

Incompatibilities

Many features of BOUT++ have been written assuming an axisymmetric coordinate system. Once 3D metrics are enabled, this is no longer (necessarily) true which breaks several features. For instance, many of the Laplacian inversion solvers use intrinsically 2D methods, and so are not available when using 3D metrics. Most of these features are runtime options, and therefore will throw an exception if you try to use them. To get a list of available Laplacian solvers, for example, you can pass the --list-laplacians flag to a compiled BOUT++ executable, which will print all the Laplacian solvers, noting which are unavailable and why.

Several boundary conditions are also incompatible with 3D metrics, unfortunately at the time of writing there is no easy way to list those that are. Several of these, such as zerolaplace have no alternative implementations, so this may mean it is not possible to run a given model with 3D metrics.

There are a few tests that don’t work with 3D metrics, mostly because they rely on one of the above incompatible methods or operators.

There is a preprocessor macro, BOUT_USE_METRIC_3D, and a constexpr bool, bout::build::use_metric_3d, which can be used to guard code that doesn’t compile or work with 3D metrics, or perhaps needs to be handled differently.

Caution should be exercised with FFT-based methods. Technically, FFTs do work with 3D metrics, but will not give the correct answer with non-constant dz.