The SI module provides a framework for working with physical units in Python. It has no dependencies beyond Python itself, yet is fully inter-operable with Numpy’s API as well as Nutils’ own function arrays.


The SI module defines all base units and derived units of the International System of Units (SI) are predefined, as well as the full set of metric prefixes. Dimensional values are generated primarily by instantiating the Quantity type with a string value.

>>> v = SI.parse('7μN*5h/6g')

The Quantity constructor recognizes the multiplication (*) and division (/) operators to separate factors. Every factor can be prefixed with a scale and suffixed with a power. The remainder must be either a unit, or else a unit with a metric prefix.

In this example, the resulting object is of type “L/T”, i.e. length over time, which is a subtype of Quantity that stores the powers L=1 and T=-1. Many subtypes are readily defined by their physical names; others can be created through manipulation.

>>> type(v) == SI.Velocity == SI.Length / SI.Time

While Quantity can instantiate any subtype, we could have created the same object by instantiating Velocity directly, which has the advantage of verifying that the specified quantity is indeed of the desired dimension.

>>> w = SI.Velocity('8km')
Traceback (most recent call last):
TypeError: expected [L/T], got [L]

Explicit subtypes can also be used in function annotations:

>>> def f(size: SI.Length, load: SI.Force): pass

The Quantity type acts as an opaque container. As long as a quantity has a physical dimension, its value is inaccessible. The value can only be retrieved by dividing out a reference quantity, so that the result becomes dimensionless and the Quantity wrapper falls away.

>>> v / SI.parse('m/s')

To simplify this fairly common situation, any operation involving a Quantity and a string is handled by parsing the latter automatically.

>>> v / 'm/s'

A value can also be retrieved as textual output via string formatting. The syntax is similar to that of floating point values, with the desired unit taking the place of the ‘f’ suffix.

>>> f'velocity: {v:.1m/s}'
'velocity: 21.0m/s'

A Quantity container can hold an object of any type that supports arithmetical manipulation. Though an object can technically be wrapped directly, the idiomatic way is to rely on multiplication so as not to depend on the specifics of the internal reference system.

>>> import numpy
>>> F = numpy.array([1,2,3]) * SI.parse('N')

No Numpy specific methods or attributes are defined. Array manipulations must be performed via Numpy’s API, which is supported via the array protocol ([NEP 18](

>>> f'total force: {numpy.sum(F):.1N}'
'total force: 6.0N'


In case the predefined set of dimensions and units are insufficient, both can be extended. For instance, though it is not part of the official SI system, it might be desirable to add an angular dimension. This is done by creating a new Dimension instance, using a symbol that avoids the existing symbols T, L, M, I, Θ, N and J:

>>> Angle = SI.Dimension.create('Φ')

At this point, the dimension is not very useful yet as it lacks units. To rectify this we define the radian by its abbreviation ‘rad’ in terms of the provided reference quantity, and assign it to the global table of units:

>>> SI.units.rad = Angle.wrap(1.)

Additional units can be defined by relating them to pre-existing ones:

>>> import math
>>> SI.units.deg = math.pi / 180 * SI.units.rad

Alternatively, units can be defined using the same string syntax that is used by the Quantity constructor. Nevertheless, the following statement fails as we cannot define the same unit twice.

>>> SI.units.deg = '0.017453292519943295rad'
Traceback (most recent call last):
ValueError: cannot define 'deg': unit is already defined

Having defined the new units we can directly use them:

>>> angle = SI.parse('30deg')

Any function that accepts angular values will expect to receive them in a specific unit. The new Angle dimension makes this unit explicit:

>>> math.sin(angle / 'rad')