Algebraic interfaces for Java

Table of Contents

1. math
2. statistics
3. physics
4. algebra
- 4.1. Features
5. Compatibility
6. ConfigurationService
7. Theory testing
- 7.1. Non math related theories

$Build Status$

See this file in html for proper display of the (few) mathematical equations, and MP4-movies.

1. math

The core dependency of the modules in the project. It provides the scaffolding for:

Abstract algebras to ensure a common interface for basic operations. Java does not provide operator overloading. This provides an alternative. Implementations of these interfaces can be 'property-based' tested, to make sure that the implementation indeed adhere to the contracts.
An 'uncertain number' interface, plus an implemention of an algebraic field of uncertain numbers.
A Service Provider Interface for formatting the elements of abstract algebras. This core module only provides an implementation to format uncertain numbers, using proper rounding and scientific notation.

1.1. Abstract Algebras

The idea is that every 'abstract algebra' consists of the implementation of two interfaces

One of the extensions of org.meeuw.math.abstractalgebra.AlgebraicElement defines the properties of all elements of the algebra. It also should implement the actual operations like multiplication and addition.
One of the corresponding extensions of org.meeuw.math.abstractalgebra.AlgebraicStructure, e.g. org.meeuw.math.abstractalgebra.Field, defines properties of the structure itself, and it also serves as a container for utility method for its elements. E.g. if the structure is powerful enough to implement determinants of matrices of its elements, it does so (and more advanced structures, may do it more sophisticatedly. E.g. Ring implements determinant without using division, but in DivisionRing the implementation is optimized with use of that operation)

The terminology which is adopted is this:

Algebraic operation	operator	operator name	static operator name	result name	argument name	defined in
binary operators
operation	`*`	operate	operate	operation	operand	`MagmaElement`
addition	`+`	plus	add	sum	summand	`AdditiveSemiGroupElement`
subtraction	`-`	minus	subtract	difference	subtrahend	`AdditiveGroupElement`
multiplication	`⋅`	times	multiply	product	multiplier	`MultiplicativeSemiGroupElement`
division	`/`	dividedBy	divide	quotient	divisor	`MultiplicativeGroupElement`
exponentiation	`^`	pow	pow	power	exponent	`CompleteFieldElement`
metric or distance	`d()`	distanceTo	metric	distance		`MetricSpaceElement`
unary operators
negation	`-`	negation	negate	negation		`AdditiveGroupElement`
reciprocation	`1/`	reciprocal		reciprocal		`MultiplicativeGroupElement`
square root	`√`	sqrt	sqrt	square root	radicand	`CompleteFieldElement`
sine	`sin()`	sin	sin	sine	angle	`CompleteFieldElement`
cosine	`cos()`	cos	cos	cosine	angle	`CompleteFieldElement`
absolute value, distance to zero	\|\|	abs	abs	absolute value		`MetricSpaceElement`
identify	+	self				`AlgebraicElement`
comparison operators
equals	`=`	equals	equals	equality	object	`java.lang.Object`
loosely equals	`≈`	eq	equals	equality	other element	`AlgebraicElement`
integer operators
root	`ⁿ√x`	root	root	n-th root	base	`CompleteFieldElement`
power	`xⁿ`			n-th power	exponent	`CompleteFieldElement`
tetration	`x↑↑n` `ⁿx`				height

The methods on the elements take the name of the corresponding operator. So e.g.:

RationalNumber a, b, c;
c = a.times(b);

These methods always take the value of the element itself plus zero parameters (for the unary operators) or one parameter (for the binary operators), and create a new value from the same algebra.

Alternative terminology, like e.g. 'add' for addition would have been possible, but it was chosen to use those verbs when the operation is e.g. implemented statically (E add(e1, e2)) or are modifying the element itself.

Most implementations are strictly read-only, but at least all algebraic operations themselves should be without side effects on the algebraic element itself.

1.1.1. Algebraic structure

Every algebraic element object has a reference to (the same) instance of this structure. The structure itself defines e.g. the 'cardinality'.

Note	If the cardinality is 'countable' (< ℵ₁), the structure can also implement `Streamable` to actually produce all possible elements.

The algebraic structure also contains methods to obtain 'special elements' like the identity elements for multiplication and addition (one and zero).

$algebras$

Figure 1. The defined algebraic structures, with indication of the operators (and whether they are commutative), special elements, and example implementations.

1.1.2. Numbers

Some algebraic elements are like real numbers. There are several interfaces dedicated to formalising properties of that.

class/interface	description
`Scalar`	A generic interface that defines the methods to convert to java (primitive) number objects. Like `doubleValue()` and `intValue()`. It extends a few interfaces for some properties which can be applied to other structures to, like `Sizeable` and `SignedNumber`.
`ScalarFieldElement`	A `Scalar` that is also a `FieldElement`. So this is the link from number to algebra. Well-behaved field elements that also behave as a 'Number' may implement `ScalarFieldElement`
`CompleteFieldElement`	Even more similar to the everyday concept of a number are elements of an algebraic field that is 'complete'. This in some way means that is has 'no gaps', but essentially boils down to the fact that operations like taking square roots and trigonometric function are possible within the algebra.
`NumberOperations` `UncertaintyNumberOperations`	Number like structures are backed by existing classes `BigDecimal` and `Double`. These lack a common interface. Implementations of this class wrap these things with a common interface to all needed operations. . E.g. it may use `BigDecimalMath` for `BigDecimal` and `Math#log` for `Double`. The specialization `UncertaintyNumberOperations` adds the logic for propagation of uncertainties.

1.1.3. Numbers and propagation of uncertainties

Most real numbers cannot be represented exactly. It may be of interest to keep track of the uncertainty in the value, and try to propagate those uncertainties sensibly when performing operations on them.

The 'physics' module will add to this that these kinds of uncertainties may originate not only in the finite nature of representing them, but also in the limitations of actually measuring things.

The 'statistics' module introduces uncertain numbers where the uncertainty is defined as the standard deviation in a collected set of values. These numbers are examples of elements that are actually stateful, because new values can be added to the set. This should not actually change the value represented by the object though, only decrease its uncertainty. On performing operations on these kinds of objects you would receive unmodifiable stateless objects with frozen value and uncertainty.

It is not always absolutely defined how propagations must happen. Some interpretation may be needed sometimes. The choices made are currently collected in `UncertaintyNumberOperations'. This is not currently pluggable or configurable, but it may well be.

operation	formula	current uncertainty propagation algorithm
summation	$a ± Δa + b ± Δb$	$\sqrt{Δa^2 + Δb^2}$
multiplication	$a ± Δa \cdot b ± Δb$	$\mid a \cdot b \mid \cdot \sqrt{\left(\frac{Δa}{\mid a \mid + Δa }\right)^2 + \left(\frac{Δb}{\mid b \mid + Δb }\right)^2}$
exponentiation	$\left(a ± Δa\right) ^ {e ± Δe}$	$\mid a ^ e\mid \cdot \sqrt{ \left(\frac{e \cdot Δa}{a}\right)^2 + \left(\ln(a) \cdot Δe\right)^2 }$
sin/cos	$\sin(\alpha \pm \Delta\alpha)$	$\Delta\alpha$

Zero

Sometimes the value with uncertainty is exactly zero, so fractional uncertainty leads to division by zero exceptions. Therefore, for now fractional uncertainty is implemented like $ \frac{Δa}{|a| + Δa}$ (rather then $ \frac{Δa}{|a|}$), where the denominator can never become zero because the uncertainty is strictly bigger than zero.

1.1.4. Testing

In mihxil-theories for every algebraic structure interface there are 'theory' interfaces using jqwik. Tests for actual implementations implement these interfaces and provide the code to supply a bunch of example elements.

Default methods then test whether all theoretical possibilities and limitations of the algebraic structure are indeed working.

1.1.5. Implementation of `equals`/`hashCode` and `eq`

When a value has uncertainty, then equals could consider it. So objects may e.g. have different toString representation but still be equal, because the difference is considered smaller than the uncertainty, and so can be considered equal.

This is abstracted using a ConfidenceInterval concept.

In this case the hashCode must be a fixed value, because otherwise we can’t guarantee that equal values have equal hashCode.

This implies that it’s a bad idea to use uncertain values as hash keys.

Transitivity of equality

Java - and also mathematics - normally requires that the equality operator (‘=’) is transitive.

For several of the objects (the Uncertain ones) this represents a problem, because on one hand it is expected that things like (x^-1)^-1 = x, and on the other hand transitivity of equals is desired (x = y ∧ y = z → x = z).

Therefore, the elements of algebra’s have several methods for equality

eq

This is the most used equality in algebras. For uncertain valued algebras this may not be transitive, because the uncertainty is considered.

E.g. `10 ± 5 eq 14 ± 1` and `18 ± 5 eq 14 ± 1`, but `! (10 ± 5 eq  18 ± 5 )`.

For non-uncertain values, eq would behave the same as equals, the only difference being that its argument is not Object.

strictlyEquals

If the value is Uncertain then it also implements a method strictlyEquals which just compares the value without considering uncertainty. This guarantees transitivity, but e.g. reciprocity of inverse operator may not be, since e.g. because of rounding errors `(x^-1)^-1 !strictlyEquals x,

equals

Java’s equals method is implemented with strictlyEquals or with eq if the value is not uncertain (strictlyEquals is not available, and it would make no difference).

Via the CompareConfiguration configuration aspect, it can be configured though, that equals is like eq.

 withAspect(CompareConfiguration.class, compareConfiguration -> compareConfiguration.withEqualsIsStrict(false), () -> {
     /// here equals behave like eq
 }

This common case can also be accessed more concisely:

CompareConfiguration.withLooseEquals(() -> {
    // code here
});

1.2. Formatting and configuration

A service loader is provided for implementations of AlgebraicElementFormatProvider which can create instances of java.text.Format which in turn can be used to convert algebraic elements to a string. #toString can be based on it.

The formatters have access to a (thread local) configuration object (see ConfigurationService). Like this a consistent way is available to configure how e.g. uncertainties must be represented. Currently, this configuration object can only be filled by code. The base configuration object in itself is empty, but the available `AlgebraicElementFormatProvider`s communicate the 'configuration aspects' which it can use.

The service giving access to the format-providers is FormatService. This is a collection of static functions.

1.2.1. unicode

Formatting normally happens using unicode if possible. So if it is common in mathematics or physics to use super scripts, sub scripts, greek letters or other special symbols, then this will be done as good as possible using just unicode characters and modifiers.

1.3. Utilities

To implement several aspects of the groups there are provided some utility class. We describe here a few which might be of particular interest.

1.3.1. Cartesian product of streams

All countable, Streamable algebras need to implement a stream providing all elements. This is not always trivial. It may require to produce all combinations of all elements of two or more underlying streams of objects.

For finite streams this is more or less trivial. For infinite streams this is a bit more interesting.

Generic

StreamUtils provides several utilities related to streams.

The most generic implementation requires for every axis a supplier for the stream, which will be used every time the first value of the stream is needed again.

This implementation then only advances streams, and needs no state otherwise.

All combinations of 2 streams of positive integers.

All combinations of 3 streams of positive integers.

Diagonals

The 2 dimensional plane of integers traditionally can be filled by tracking diagonals. StreamUtils provides an implementation of that too. It is harder to generalize this to more dimensions, and also it requires that streams can be tracked reversely.

All combinations of 2 streams of positive integers (diagonals)

2. statistics

Implementations of UncertainDouble, which are based on calculating standard deviations on sets of incoming data, and use that as the uncertainty value.

Also, it includes some classes to keep track of 'sliding window' values of averages.

example of WindowedEventRate

WindowedEventRate rate = WindowedEventRate.builder()
            .bucketCount(50)
            .window(Duration.ofMinutes(50))
            .build();
rate.newEvent();
...
..
log.info("Measured rate: {} /s",  rate.getRate(TimeUnit.SECONDS) + " #/s");

log.info("Measured rate: {}", rate); // toString

3. physics

This module involves mostly around PhysicalNumber and its derivatives. A PhysicalNumber is a UncertainDouble, but the uncertainty is stated (it is a Measurement), and knows how to propagate those uncertainties when doing algebraic operations.

Also, a PhysicalNumber can be assigned Units. This can be used for proper displaying the value, and for dimensional analysis.

...
PhysicalNumber twoLightYear = new Measurement(2, 0.1, ly);        //
PhysicalNumber oneParsec = measurement(1, 0.1, pc); // using the static import as a shortcut

assertThat(twoLightYear.plus(oneParsec).toString()).isEqualTo("5.3 ± 0.4 ly");
assertThat(oneParsec.plus(twoLightYear).toString()).isEqualTo("1.61 ± 0.13 pc");
assertThat(oneParsec.plus(twoLightYear).eq(twoLightYear.plus(oneParsec))).isTrue(); //different toString does not mean that they represent a different value
log.info("{} + {} = {}", twoLightYear, oneParsec, twoLightYear.plus(oneParsec));

Physical numbers themselves are actually only forming a multiplicative group, because they cannot be added without constraints. In this example they can only be added to each other because both values have the same dimensions (both are about distance).

Physical numbers can freely be multiplied and divided by each other.

Objects of the statistic module can be converted to 'physical numbers' like so:

event rate to measurement

WindowedEventRate rate = ...

PhysicalNumber measurement = new Measurement(rate);
PhysicalNumber rateInHours = measurement.toUnits(Units.of(SI.hour).reciprocal());

statistical number to measurement

 StatisticalDouble statisticalDouble = new StatisticalDouble();
 statisticalDouble.enter(10d, 11d, 9d);

 PhysicalNumber measurement = new Measurement(statisticalDouble, Units.of(SI.min));

 assertThat(measurement.toUnits(Units.of(SIUnit.s)).toString()).isEqualTo("600 ± 45 s");

4. algebra

This contains various implementations of the algebraic structure interfaces of mihxil-math. Like RationalNumber (modelling of rational numbers ℚ), and the rotation group SO(3).

4.1. Features

4.1.1. Real numbers

The field of real numbers. Backed by java primitive double. A RealNumber is also 'uncertain', which is used to keep track of rounding errors.

element RealNumber
structure RealField

4.1.2. Arbitrary precision real numbers

The field of reals numbers, but backed by java’s BigDecimal. This means that it supports arbitrary precision, but, since this still is not exact this still is uncertain, and rounding errors are propagated.

element BigDecimalElement
structure BigDecimalField

4.1.3. Rational numbers

The field of rational numbers. Implemented using two arbitrary sized BigIntegers.

element RationalNumber
structure RationalNumbers

Also, since division is exact in this field, this does not implement UncertainNumber.

The cardinality is countable (ℵ₀) so this does implement Streamable.

4.1.4. Permutations

The permutation group. An example of a non-abelian finite group.

element Permutation
structure PermutationGroup

This is group is finite, so streamable. This means that the group also contains an implementation of 'all permutations' (this is non-trivial, it’s using Knuth’s algorithm).

The permutation elements themselves are implemented as a java.util.function.UnaryOperator on Object[] which then performs the actual permutation.

4.1.5. Integers

The most basic algebraic structure which can be created from integers are the integers (ℤ) themselves. They form a ring:

element IntegerElement
structure Integers

4.1.6. Even integers

As an example of a 'rng' (a ring without the existence of the multiplicative identity 1), the even integers can serve

element EvenIntegerElement
structure EvenIntegers

4.1.7. Natural numbers

In the natural numbers ℕ (the non-negative integers), there can be no subtraction. So they only form a 'monoid' (both additive and multiplicative).

element NaturalNumber
structure NaturalNumbers

4.1.8. Modulo groups

Integers can be simply restricted via modulo arithmetic to form a finite ring:

element ModuloRingElement
structure ModuloRing

If the 'divisor' is a prime, then they even form a field, because the reciprocal can be defined:

element ModuleFieldElement
structure ModuloField

4.1.9. Complex numbers

Another well-known field is the field of complex numbers.

element ComplexNumber
structure ComplexNumbers

4.1.10. Quaternions

Quaternions are forming a 'non-commutative' field, a DivisionRing

element Quaternion
structure Quaternions

4.1.11. Matrix and rotation groups

SO(3)

Another non-abelian (not-commutative) multiplicative group.

element Rotation
structure RotationGroup

4.1.12. Strings

Actually one of the simplest algebraic object you can think of are the strings. They form an additive monoid, an algebraic structure with only one operation (addition).

element StringElement
structure StringMonoid

Their cardinality is only ℵ₀, so StringMonoid also contains an implementation to stream all possible strings.

4.1.13. Vector spaces

Vector spaces, which manage vectors, are basically fixed sized sets of scalars, but combine that with several vector operations like cross and inner products.

5. Compatibility

This project is compiled with java 17, and provided JPMS module info, but for know is compatible with java 8.

6. ConfigurationService

ConfigurationService is responsible for managing the Configuration thread locals.

Like this it can be consulted

Accessing configuration

import org.meeuw.configuration.Configuration;
import org.meeuw.configuration.ConfigurationService;
import org.meeuw.math.text.configuration.NumberConfiguration;
import org.meeuw.math.text.configuration.UncertaintyConfiguration;

import static org.meeuw.configuration.ConfigurationService.*;
...

Configuration configuration = getConfiguration();
NumberConfiguration aspect = configuration.getAspect(NumberConfiguration.class);
int minimalExponent = aspect.getMinimalExponent();

This would however probably mainly be used in implementations.

Actual configuration can be done in two basically distinct ways.

a new configuration object can be set as a thread local
global default configuration object can be set

temporary overrides

{
    //noinspection resource
    setConfiguration(builder ->
        builder.configure(NumberConfiguration.class,
            (numberConfiguration) -> numberConfiguration.withMinimalExponent(8)
        )
    );

    //...code...
    ConfigurationService.resetToDefaults();
}

// or using Autocloseable
try (Reset ignored = setConfiguration(builder ->
    builder.configure(NumberConfiguration.class,
        (numberConfiguration) -> numberConfiguration.withMinimalExponent(8)
    )
)) {
    ;
    //...code...
}

There are some utilities in ConfigurationService that makes this process a bit easier.

temporary overrides utilities

withConfiguration((con) ->
        con.configure(UncertaintyConfiguration.class,
                (uncertaintyConfiguration) -> uncertaintyConfiguration.withNotation(UncertaintyConfiguration.Notation.PARENTHESES))
            .configure(NumberConfiguration.class,
                (numberConfiguration) -> numberConfiguration.withMinimalExponent(3))
    , () -> {
        // code

    });

Global defaults can be set similarly

setting global defaults

defaultConfiguration((configurationBuilder) ->
    configurationBuilder.configure(NumberConfiguration.class, c -> c.withMinimalExponent(4))
        .configure(UncertaintyConfiguration.class, c -> c.withNotation(UncertaintyConfiguration.Notation.PLUS_MINUS))
);

ConfigurationService itself is not actually math related, and is released in a separate artifact.

7. Theory testing

NOTE: Package structure was changed in 0.12.0.

For testing the structures and object of mihxil-math, this provides 'property-based' testing, based on jqwik.

This is provided as a set of interfaces named …Theory. Tests can implement these interfaces, and all contracts are tested. This normally requires the tests to implement a set of datapoints ore 'elements'.

7.1. Non math related theories

See org.meeuw.theories for theories not directly related to mathematical structures, but merely to java contracting like e.g.

class	goal	(example) methods
BasicObjectTheory	Tests basic properties of any java object, mainly consistency of `equals` and `hashCode`	`equalsIsReflexive` `equalsIsSymmetric` `equalsIsTransitive` `equalsIsConsistent` `equalsReturnFalseOnNull` `hashCodeIsSelfConsistent` `hashCodeIsConsistentWithEquals` `toString`
ComparableTheory	If an object is also `Comparable` then consistency of `compareTo` can be tested	`equalsConsistentWithComparable`
CharSequenceTheory	The `CharSequence` interface also has a few methods that can be tested generically	`charAtIsConsistentWithToStringCharAt` `subSequenceIsConsistent`

operation	formula	current uncertainty propagation algorithm
summation	\(a ± Δa + b ± Δb\)	\(\sqrt{Δa^2 + Δb^2}\)
multiplication	\(a ± Δa \cdot b ± Δb\)	\(\mid a \cdot b \mid \cdot \sqrt{\left(\frac{Δa}{\mid a \mid + Δa }\right)^2 + \left(\frac{Δb}{\mid b \mid + Δb }\right)^2}\)
exponentiation	\(\left(a ± Δa\right) ^ {e ± Δe}\)	\(\mid a ^ e\mid \cdot \sqrt{ \left(\frac{e \cdot Δa}{a}\right)^2 + \left(\ln(a) \cdot Δe\right)^2 }\)
sin/cos	\(\sin(\alpha \pm \Delta\alpha)\)	\(\Delta\alpha\)

Files

README.adoc

Latest commit

History