A C++ Library enabling practical engineering computation with Geometric Algebra in 3D space.
Engabra is lightweight C++ library that supports performing practical Engineering computation, modeling, analysis, and simulation in 3D space with no other software dependencies.
The main emphasis is on practical utility (not theoretical abstraction nor education, nor experimentation - as is the case for many GA software libraries). The fundamental idea behind engabra is that the library be useful for real world computation, modeling and simulation.
The focus on 3D space allows providing relationships and operations that may not be available in other dimensions. For example splitting multivectors into the sum of ComPlex and DirPlex values and using these to implement generalized inversion operations as well as other functions that may not be defined in higher dimensional spaces (e.g. circular trigonometric functions, logarithms of vectors).
Quick jump links to below:
The main utility/distinction of Engabra is that it has a very practical focus for use in coding applications that provide real-world engineering solutions - in contrast to a mathematically interesting one. Engabra is meant for use in real-world software development. It focus on ease of use, practicality, and performance - from perspective of developer as well as applications.
The types and functions are structured and intended for real-world engineering type solutions.
-
Focused on actual use (compared with experimentation or academic explorations)
-
All entities have direct physical interpretations and associations (compared with e.g. various higher-dimensional conformal algebras used in academic contexts)
-
Dumb-simple-stupid class definitions and function operators generally allow compiler to do type checking and produce easily interpretable warnings and error messages.
-
In some places, this means there are concrete implementations of overloaded functions, but this is preferable to having complex, confusing and error prone templating syntaxes.
-
Utilizes explicit constructors and mitigates implicit casting to avoid potential confusing behaviors - keeps consuming code expression very WYSIWYG.
-
Engabra is an entirely stand-alone library. No other packages need be installed to use it - other than C++ compiler (e.g. g++, clang, msvc) and general development tools of course.
The code is standard C++ and should work fine on nearly any platform that supports a modern C++ standard compiler (C++17 or later).
-
Code implementation is relatively 'vanilla' and simple (e.g. no use of fancy optimizations or 'tricks'). The library should build on nearly any (64-bit) device or platform.
- Note: (as of 2023.Q1) implementation has only been built for environments using 64-bit double representations. It is unverified how/if compatible the implementation is with 32-bit double sizes.
-
The CMake build process supports compiling and packaging on a wide variety of different operating systems, distributions and hardware platforms.
- Note the descriptions herein and associated file/directory/command name conventions are consistent with a *nix environment. However, this should not be construed as any sort of limitation on the library, only that a *nix environment was used to author many of the text files and descriptions.
For the most part, the library is very easy and natural to utilize in consuming code.
-
There are no complex templating expressions involved or required in the consuming code (other than a few relatively natural use cases - e.g. such as defining useful generic type instances like "null()" or "zero()" and a few other natural/typical template usage)
-
There are NO strange operators to memorize.
-
Operator overloading only uses the obvious and standard operators (e.g. "*" for multiplication, "+" for addition, etc).
-
For operations not standard to C++ syntax, explicit function names are used (e.g. dot(), wedge(), reverse(), dual()}
-
Division ambiguity is avoid in favor of more explicit constructs involving element inverses. I.e. there is no support for the "/" operator. Instead, multivector "quotients" are supported with explicit and specific operations via the inverse function. E.g. "inverse(vecA)vecB" or "vecAinverse(vecB)" according to whichever is actually intended.
-
Engabra is fast to use in practice.
-
Fast to compile -- although templates are used heavily within internal implementation of Engabra the instantiations are very simple so that compiler can resolve and instantiate code very quickly. Note that, in spite of the use of templates within the implementation, consuming code is generally insulated from them - other than a few cases where using templates could be expected for any library (mostly in customizing constant types)
-
Fast run time -- The run time performance is quite good. Although there are no particularly special coding techniques used (and certainly no expectation for special hardware capabilities), the implementation code makes copious use of 'const' and 'inline' specifications so that compilers can optimize the code very well.
-
Emphasizes 'const' paradigm to support better compiler optimization as well as easier and more efficient use in multi-threaded applications.
- NOTE: For runtime performance, be sure to compile with optimization enabled. This will remove many "inline" function implementations and optimize away many copy operations (that are present only for clarify of library implementation code).
With numeric libraries, it is tempting to try to override all standard C++ math operations. However, this is not necessary for useable library since many of the same results can often be expressed in alternative manner.
For example, various comparison operators can always be expressed as an appropriate combination of operator<() and logical operators (e.g. "(a > b)" can be coded easily as "(b < a)". This may require a small adaptation in style and habits.
A side benefit is that consuming code is frequently easier to read because there are fewer operators, and the resulting code patterns tend to be more standardized and uniform.
Operators (intentionally) not supported include:
-
The increment and decrement pre/post-fix operators -- these are primarily useful for integer type operations. Although these could perhaps be applied to scalar and pseudo-scalar (trivector) grades, there's little algebraic benefit in doing so.
operator++() : use explicit "foo = foo + ..." instead; operator--() : use explicit "foo = foo - ..." instead
-
Division operators -- Given the non-commutative nature of GA (at least for multiplicative operations, it's generally more clear to be explicit what is meant. E.g.
operator/() -- use combination of inverse() and multiplication - e.g.
// one of these depending on specific intent result = numerator * inverse(denominator); // division from right result = inverse(denominator) * numerator; // division from left
// for scalars, instead of "foo = bar / sca;" use something like foo = (1./sca) * bar; foo = inverse(sca) * bar;
-
Compound assignment operators. Given modern compiler optimization capabilities, the "in-place" operators often provide more of a typing shortcut than any performance difference. For GA types, there is the additional concern that multiplication and "division" do not generally commute, and there are no variants such as a hypothetical "operator=*()" or "operator=/()" to address this, which would be potentially confusing in any case. Instead utilize explicit "operate then assign" syntax such as:
operator+=() -- use "foo = foo + bar;" operator-=() -- use "foo = foo - bar;" operator*=() -- use "foo = foo * bar;", "foo = bar * foo;" operator/=() -- use "foo = foo * inverse(bar)", "foo = inverse(bar) * foo;"
As an example Engabra use, the following provides an example of accomplishing fairly general coordinate transformation operations. It illustrates instantiation of various types as well as interacting with rigid body attitude using both sequential and simultaneous angle conventions.
Consider transforming a coordinate position (vector), a force (vector) and a moment (bivector) that are expressed with respect to some 'world' reference coordinate frame in order to obtain an expressions for the force and moment in the coordinate frame of some specific structural element.
// Here the using directive makes this code snippet condensed and compact.
// Items having scope resolved by the using operator include:
// - the main data types (Vector, Bivector, Spinor, ...)
// - useful constants including
// -- basis vectors {e1, e2, e3}
// -- basis bivectors {e23, e31, e12}
// -- pi (also piHalf, piQtr, ...)
// - functions (e.g. exp(), logG2(), ...)
using namespace engabra::g3;
// starting information expressed in (W)orld space (assumed known)
Vector const posInW{ 10., 20., 30. }; // some position in world
Vector const forceInW{ -9.81 * e3 }; // e.g. force downward (-e3)
BiVector const momentInW{ 5. * e12 }; // e.g. moment in XY-plane (e12)
// location and attitude of structure in world frame
// generally this would be from some source, design, or calculation
Vector const locSwrtW{ 100., 200. 300. }; // location in world
// specify structure attitude as ordered sequence of Azim/Elev/Roll angles
BiVector const angAzim{ .01234 * pi * e12 }; // e.g. azimuth angle of structure
BiVector const angElev{ .23456 * pi * e23 }; // e.g. elevation angle rotation
BiVector const angRoll{ .34567 * pi * e31 }; // e.g. roll angle rotation
Spinor const spinSwrtW // "spinor" rotation operator (like a quaternion)
{ exp(.5*angRoll) * exp(.5*angElev) * exp(.5*angAzim) };
BiVector const ang3D // 3D angle associated with azim/elev/roll sequence
{ 2.*logG2(spinSwrtW).theBiv }; // pick BiVector from logG2() spinor
// position, force and moment expressed in (S)tructure element frame
Vector const posInS // canonical rigid body transformation formula
{ spinSwrtW * (posInW - locSwrtW) * reverse(spinSwrtW) };
// force and moment only need to be rotated
Vector const forceInS{ spinSwrtW * forceInW * reverse(spinSwrtW) };
BiVector const momentInS{ spinSwrtW * momentInW * reverse(spinSwrtW) };
// report result
std::cout << "position in structure frame: " << posInS << std::endl;
std::cout << " force in structure frame: " << forceInS << std::endl;
std::cout << " moment in structure frame: " << momentInS << std::endl;
std::cout << "structure attitude 3D angle: " << ang3D << std::endl;// TODO - add example hereFundamental Types: For detail refer to (doxygen-generated) reference documentation for classes in file 'g3type.hpp'
-
double -- native C++ type (for convenience and performance where constructing a class 'scalar' type is unnecessary). (In algebraic speak, this is the "field" over which the 3D geometric algebra is defined).
- TODO/TBD - is this used in 'using' or 'typedef' somewhere? What about using library with other types - e.g. 'float' ?
-
Scalar -- This is mostly a class wrapper around native type in order to provide consistency with Engabra conventions. This has a single value (supports subscripting but has only single component [0]) and is of grade 0. Scalars generally quantities, or "how much" of something, and also represent magnitudes of other elements.
-
Vector -- A (classic) vector (line like) with three components (subscripts [0-2]) and of grade 1. Vectors are particularly suited for expressing position relationships, forces, velocities and accelerations.
-
BiVector -- A bivector (plane like) with three components (subscripts [0-2]) and of grade 2. BiVectors are particularly suited for encoding directed, planar concepts - including angles, moments, and rotation operations in 3D space. Bivectors square to non-positive values and often act like a pure "imaginary" number, but with a planar direction encoded within them as well.
-
TriVector -- a trivector (volume like) with one component (subscript [0]) and of grade 3. Trivectors square to non-positive values and provide a directionless pure-imaginary number.
-
MultiVector -- This is the most general element of the (3D) geometric algebra. Comprised of a Scalar (.theSca member), a Vector (.theVec member), a BiVector (.theBiv member), and a TriVector (.theTri member). Note that any and all other types and general operations of the entire algebra (and this library) can be implemented entirely in terms of MultiVectors. The components can also be accessed via subscripts [0-7] in order of Scalar([0]), Vector([1-3]), BiVector([4-6]), TriVector([7]) components.
NOTE: If/where the library currently does not support some specialized type or combination of operations, you always have the option to implement whatever you want in terms of multivector operations by first promoting the available types to MultiVectors, coding the operations with multivector operations and then casting or accessing/extracting the appropriate grade portions of the results!!
Aside from the general MultiVector composition, there are several more simple composite types that are particularly useful for certain applications.
-
Spinor -- Member of the G2 sub-algebra. The Spinor type is comprised of a Scalar (.theSca member) and Bivector (.theBiv member). This 3D spinor type is isomorphic with quaternions. (This type is often associated with the product of two vectors). The individual components can also be accessed via subscripts [0-3] with [0] accessing the Scalar and [1-3] accessing the Vector components.
-
ImSpin -- An "imaginary spinor" type that is dual to Spinor. It is comprised of a Vector (.theVec member) and TriVector (.theTri member). This type is often associated with the product of three vectors). Individual components can also be accessed via subscripts [0-3] with [0-2] accessing the Vector components and [3] accessing the TriVector component.
-
ComPlex: with Sca + Tri -- this complex type multiplicatively commutes with all other types. This type has the same behavior as does the standard library "std::complex<double>" type and can be constructed from, and cast to, std::complex.
-
DirPlex: with Vec + Biv -- this "complex vector" type comprises the spatially directed grades, vector and bivector. In general, multiplication with this type is non-commutative.
In general, an arbitrary multivector in 3D space, can be expressed (or decomposed) in terms of the sum of commuting ComPlex type and directed DirPlex type. E.g.
// isomorphic with classic complex numbers
ComPlex const complex{ someMulti.theSca, someMulti.theTri };
// the physically directed grades (vec,biv)
DirPlex const dirplex{ someMulti.theVec, someMulti.theBiv };
// together ComPlex and DirPlex types comprise a full multivector
MultiVector const multi{ complex + dirplex };
The following are less commonly utilized types - and are unsupported in the current Engabra implementation.
All operations associated with these special types can be implemented completely (albeit slightly less efficiently) by casting to g3::MultiVector type.
-
ParaVec: with Sca + Vec -- used for some physics formulations
-
ImPara: with Biv + Tri -- (dual to ParaVec)
-
NoSca: -- with Vec + Biv + Tri -- any need?
-
NoVec: -- with Sca + Biv + Tri -- any need?
-
NoBiv: -- with Sca + Vec + Tri -- any need?
-
NoTri: -- with Sca + Vec + Biv -- any need?
In general, the most noticeable restriction of not having these types is encountered when summing different types in an expression. E.g. something like:
MultiVector const foo{ aScalar + aVector + aBivector + aTriVector };
Is a seemingly natural expression. However, the compiler interprets this from left to right and first tries to produce a ParaVec result containing the scalar and vector grade quantities. This is followed by attempting to generate a NoTri type by adding that temporary ParaVec type to the bivector argument, and so on.
A workaround for this case is to either reorder the additions to produce an addition order that involves only supported types, e.g.
// temporaries are first a ComPlex and DirPlex type that are
// then added together to produce a MultiVector
MultiVector const foo
{ (aScalar + aTriVector) // a ComPlex temporary
+ (aVector + aBivector) // a DirPlex temporary
};
or to cast to MultiVector type first, e.g.
MultiVector const expMvec
{ MultiVector(aSca)
+ MultiVector(aVec)
+ MultiVector(aBiv)
+ MultiVector(aTri)
};
Note that these coding inconveniences disappear as more of the special types are implemented.
Compound entities have members of multiple grades (for example, a Spinor is composed of a Scalar and a BiVector grade member). These constituent grades may be accessed directly with the associated member variables. The member variable naming is consistent across all compound types.
.theSca : The scalar grade member of an instance
.theVec : The vector grade member of an instance
.theBiv : The bivector grade member of an instance
.theTri : The trivector grade member of an instance
From one perspective these provide the blade decomposition of a general MultiVector. I.e.
-
MultiVector = Scalar + Vector + BiVector + TriVector
-
The Scalar part is most frequently associated with the magnitude (size, strength) of quantities and other elements in the algebra, e.g. length of vector, size of an angle, etc.
-
The Vector part is probably the single type most heavily used for engineering applications.
-
The BiVector part is probably most useful to engineering applications for representing angles in 3D space and describing the (eqatorial) plane of rotation.
-
The TriVector part is probably not used much on its own for engineering applications, but when combined with the Scalar part, provides the rich capabilities of the "complex" number system used to describe waves, and oscillations.
-
The compound types can be considered as different partitioning of af general MultiVector
-
MultiVector = Spinor + ImSpin
-
The Spinor part is generally the most useful since these types are intimately connected with rotation operations and describing the attitude of components and structures. This has great utility in practice.
-
The ImSpin type is often associated with intermediate data types that occur during GA operations. Frequently it arises in the context of the product of three Vector entities.
-
-
MultiVector = ComPlex + DirPlex
-
This partition is particuarly useful for separating the multiplicatively commutative portions from a multivector and is particuarly important for defining and implementing many of the algebraic functions within this library.
-
The ComPlex type is compatible with std::complex. One of the common uses of this type is for electrical circuit and modeling vibration and oscillation of structures.
-
The DirPlex items is a bit unusual. It has application in physics for representing electro magnetic fields, but is perhaps not that useful for engineering applications.
-
The individual components of all types are accessible with the subscript operator.
-
operator: The indices start at zero and increase as follows:
Scalar : 0 Vector : 0, 1, 2 BiVector : 0, 1, 2 TriVector : 0 Spinor : (0), (1, 2, 3) ImSpinor : (0, 1, 2), (3) MultiVector: (0), (1, 2, 3), (4, 5, 6), (7) ComPlex : (0), (1) DirPlex : (0, 1, 2), (3, 4, 5)
String formatting:
-
fixed(): return a string "fixed point" representation of entity value. Optional arguments allow specifying number of digits encoded both before and after the decimal point.
-
enote(): return a string representation using "E-notation" for representing the numeric values.
Instance insert/extract:
-
operator<<(): insertion (output) operator for basic GA types.
-
operator>>(): extraction (input) operator for basic GA types.
Notes on Terminology: Some of the unitary operations of geometric algebra are associated with rather ambiguous names. For example, does "conjugate" mean the "complex conjugate" or does it mean the very different "Clifford conjugate". To avoid confusion, a couple functions are given novel names that should help remove any ambiguity. I.e. the following self-descriptive functions provide various grade sign changing operations:
-
reverse(): -- Product permutation operation: Changes algebraic sign on imaginary-like grade constituents (bivector and trivector grades). This is associated with changing the order of multiplication between vector factors in a product - i.e. what might be called a kind of "procedural inversion".
-
oddverse(): -- Space involution operation: changes algebraic sign on odd grade constituents (vector and trivector grades) - (aka "Complex Conjugate"). This is associated with reflecting all basis vectors through the origin - i.e. "space inversion" or "geometric point involution".
-
dirverse(): -- Composite involution operation: changes algebraic sign on directed grade constituents (vector and bivector grades) - (aka "Clifford Conjugate") - i.e. what might be called a kind of "direction involution".
-
dual(): -- Operator that interchanges "Real" and "Imaginary" properties: This returns the result of multiplying its argument by the unit magnitude imaginary TriVector. The operation provides the mathematical dual of its argument. In 3D space, the dual type relationships include:
- Scalar <-> TriVector
- Vector <-> BiVector
- Spinor <-> ImSpin
- ComPlex <-> ComPlex (but with 'real' and 'imaginary' roles exchanged
- DirPlex <-> DirPlex (but with Vector and BiVector roles exchanged
These are constexpr entities.
-
nan: double value representing Not-A-Number (quiet_NaN). This is used to indicate null instances by populating members with nan values.
-
null(): various constexpr functions that create "null" instances.
- TODO/TBD note template use
These involve run-time evaluation:
-
isValid(): test if an item is not a null instance. Additional specialized comparisons include:
- bothAreValid(): conjunction (&&) of isValid() called on each argument.
These are constexpr entities.
-
e0: unitary basis entity for scalar type
-
e1, e2, e3: unitary basis entities for vector type
-
e23, e31, e12: unitary basis entities for bivector type
-
e123: unitary basis entity for trivector type
These are constexpr entities/functions.
-
zero(): create zero instances of various types
- TODO/TBD note template use
-
one(): create various types instances with scalar grade value set to 1.
- TODO/TBD note template use
-
Useful double type (radian) angle values
-
pi, piHalf, piQtr: respective fractions of pi
-
turnFull, turnHalf, turnQtr: fractions of pi associated with turning motions.
-
These involve runtime evaluation.
-
nearlyEquals(): generally useful comparison if two (reasonably normalized) data values are "same enough" within practical tolerances (allowing for reasonable amount of computation noise). Additional specialized comparisons available include:
- isSmallAbs(): has a absolute value within tolerance near zero
- isSmallAbsDiff(): two values are within isSmallAbs() of each other
- nearlyEqualsAbs(): two values are same within absolute tolerance
- nearlyEqualsRel(): two values are same within relative tolerance
-
operator==(): comparisions for bit-wise identical values.
These are for types in closed subalgebras. This includes types: Scalar, Spinor, Complex, and MultiVector.
-
sq() : Square of argument (i.e. arg*arg)
-
cube(): Cube of types (i.e. arg*arg*arg)
Blade magnitude and direction (direction only valid for non-zero blades)
-
Magnitude - This is a real (Scalar) type that is always non-negative and represents the "size" or "strength" of an entity. In GA, the magnitude (aka "norm"): (associated with the square root of the scalar part of the product of an entity and its reverse):
-
magnitude(): a native double value representing the magnitude of the argument (i.e. sqrt(magSq(arg))
-
magSq(): a native double value that is the square of the magnitude of the argument (i.e. arg*reverse(arg));
-
-
Amplitude - This is a ComPlex type that has a rather abstract interpretation. Its main use is in defining an inverse of other types.
NOTE: Amplitude (as used here) is only defined for Geometric Algebras over spatial dimensions of size three (as the case here) or less. The amplitude may not exist in all cases.
-
TODO - Amplitude is one component of polar form.
-
amplitude(): a ComPlex quantity that
TODO ...
-
-
Direction - Decomposition of
-
TODO - Direction is one component of polar form.
-
direction(): for non-zero arguments, returns a unitary magnitude instance of the same type as argument (i.e. arg/magnitude(arg))
TODO... should this be in terms of amplitude?
-
-
Polar Form - Decomposition of general multivector into Amplitude and (TODO??) Direction
- pairMagDirFrom(): returns both magnitude and direction of argument (i.e. std::pair<magnitude(arg), direction(arg)>). This saves a couple computation operations in the fiarly common case when both are needed.
-
operator-(): unary negation: changes sign on all grades (sca,vec,biv,tri)
-
reverse(): -- changes algebraic sign on imaginary-like grades (biv,tri)
-
oddverse(): -- changes algebraic sign on odd grades (vec,tri)
-
dirverse(): -- changes algebraic sign on directed grades (vec,biv)
-
dual(): -- this returns the dual of it's argument:
- dual(sca) = e123*sca (is a TriVector)
- dual(vec) = e123*vec (is a BiVector)
- dual(biv) = -e123*biv (is a Vector)
- dual(tri) = -e123*tri (is a Scalar)
-
operator+(): binary addition of two items.
All operations are supported with both arguments of same time. TODO - implement support for mixed type combinations
-
operator-(): binary "subtraction" of two items.
-
operator*(): binary multiplication of two items. In large part, these operations provide the "magic" of Geometric Algebra in providing extension and contraction of grades during the multiplication process.
All operations are supported with both arguments of same time. Most commonly used mixed type operations are supported. TODO - implement support for ALL mixed type combinations
-
inverse(): multiplicative inverse of non-zero (and non-idempotent) blades (i.e. reverse(arg)/magSq(arg)).
Inverse is supported for all types that are a product of zero or more vector factors. This includes the fundamental types (S,V,B,T) as well as Spinor, ImSpinor. TODO - implement more general support
The implementation of the following functions is based in large part on the descriptions of [Chappell, et.al., PLoS]
-
exp(): exponential function (current Q1'23 for BiVectors and Spinors)
Current support includes argument types: Spinor (of which BiVector and Scalar are special cases with one grade constituent being zero). TODO - implement more general support
-
logG2(): logarithm for members of G2 sub-algebra (Spinors)
Current support includes argument types: Spinor (of which BiVector and Scalar are special cases with one grade constituent being zero). TODO - implement more general support
-
sqrtG2(): square root for members of G2 sub-algebra (Spinors)
Current support includes argument types: Spinor (of which BiVector and Scalar are special cases with one grade constituent being zero). TODO - implement more general support
-
solutionOfLinear(): solve linear multivector equation of form "AX+XB=C"
To support internal implementation and perhaps some advanced uses of the library code, there are several functions available that can aid writing template code (ref g3traits.hpp). These include the following
In the "is" sub-namespace (within engabra::g3::is::)
- is::blade(): Evaluates to true if argument is a blade (pure Scalar, Vector, BiVector, or TriVector type).
Clone this repository to local system then use the CMake to package or install - e.g. using the example commands below
If you have a code project that is also using CMake, you can request cmake to automatically include Engabra by doing the following.
In the SomeProject/CMakeLists.txt file include:
# Use CMake command to locate and attach Engabra package
find_package(Engabra REQUIRED NO_MODULE)
# Informational messages to confirm status and report version ID
message(Engabra Found: ${Engabra_FOUND})
message(Engabra Version: ${Engabra_VERSION})
# Include Engabra in target build - e.g.
target_link_libraries(
${SomeTarget}
PRIVATE
Engabra::Engabra
)
TBD
Code is implemented in standard C++ (C++11? 17?). The library and distribution packages can be built with the CMake utility. The reference documentation is created with Doxygen as part of the cmake build process.
Tools for building the library:
-
C++ compiler (standard C++17 or later) - e.g. gcc, clang, msvc
-
CMake utilties (cmake, ctest, cpack) (v3.13 or later)
-
Doxygen utility
Relative to top engabra directory, the directory structure includes several subdirectories:
-
./include -- library main source code
- ./include/Engabra -- main include file with which to access library.
-
./src -- currently (2023.03) not used (is header only project)
-
./test -- project test programs (integrated with 'CTest')
-
./cmake -- contains source files used in the CMake generator system
-
./doc -- files and information used for Doxygen document production
-
./demo -- a few simple demonstration programs for reference which can be Compiling Demo Programs compiled as described below).
- ./helloEngabra -- example of include, namespaces and basic types
The following summarizes the organization of header files for the Engabra library. The majority of library code is implemented as inline definitions within these headers.
Detail on the contents of each can be found in the (doxygen-generated) reference documentation. This reference documentation is created as part of the library build and installation process. Ref section Reference Documentation
-
Engabra -- for matter of style (has "#include "engabra.hpp")
-
engabra.hpp -- includes the headers listed below
Fundamentals: (included by top level include, but useful names to know for accessing the reference documentation.
-
g3type.hpp -- defines fundamental types
-
g3const.hpp -- useful constants (e.g. basis entities, zero<>, null<>, pi, etc)
-
g3compare.hpp -- functions for comparing numeric values
-
g3validity.hpp -- functions for checking validity (supporting the null object paradigm)
Fundamental algebraic operations.
-
g3ops.hpp -- includes all other g3ops* files in one place
-
g3opsUni.hpp -- unitary operations (e.g. reverse, negation, duals, etc)
-
g3opsAdd.hpp -- addition operators e.g. op+() and op-() for various types
-
g3opsMul.hpp -- multiplication operators (e.g. op*() for various types)
Input output support functions
- g3io.hpp -- string formatting, op<<(), op>>() overloads, etc
General algebraic functions
- g3func.hpp -- math functions (e.g. exponentials, logarithms, roots, etc)
Advanced utilities
- g3traits.hpp -- constructs useful for advanced template constructions
Internal implementation:
- g3publish.hpp -- declarations to promote various function declarations for use in global namespace.
-
engabra - overall containing namespace
-
engabra::g3 - sub space for 3D geometric operations. An often convenient coding technique is to employ the 'using' directive to shorten the namespace resolution (unless you have another 'g3' namespace).
E.g. as in
engabra::g3::Vector const aVec{ 1., 2., 3. };
// or
using namespace engabra;
g3::Vector const aVec{ 1., 2., 3. };
// or
using namespace engabra::g3;
Vector const aVec{ 1., 2., 3. };The library uses and supports a null object pattern paradigm. E.g.
using namespace engabra;
// construct a null instance
g3::Vector const aNullVec{ g3::null<Vector>() };
// after which guarantees (false == g3::isValid(aNullVec))
// check validity of return value (assuming someFunction supports null paradigm)
g3::Vector const someVec{ someFunction() };
if (g3::isValid(someVec))
{
// okay to use someVec (it's not null)
}Default initializations follow C++ convention for initialization of aggregate types - e.g. no initialization can be assumed.
engabra::g3::Vector const vecUnInitialized;Note that this is of limited use in light of the emphasis on use of const as described in the next section. I.e. in general, all entities should be initialized explicitly (to values, by function return, using a 'null' or 'zero' type template) - otherwise, it's generally good to delay (or eliminate) instantiating an instance until there's something with which to initialize it.
The library implementation makes copious use of 'const' to support high performance (and what many people consider good coding practice). The use of 'const' data instances is encouraged to the extent that this is assumed in place of supporting self-modifying types.
Therefore, the library does not have support for self modifying operators like "*=", "++", etc. Instead, in place of updating an entity in place, simply create a new instance at will and discard the old one.
using namespace engabra;
g3::Vector const beforeVec{ someInitFunction() };
g3::Vector const afterVec{ beforeVec + g3::Vector{1., 2., 3.} };In general, this is not a problem for higher level applications. Compiler optimization for modern compilers is typically sufficiently advanced that there are virtually no (speed or space) performance penalties in expressing source code with the extra construction/destruction.
Engabra uses the CMake build system generator utility.
Clone Example:
$ cd /repos # or wherever, '/repos' is used as example here
$ git clone https://github.com/Stellacore/engabra.git
Build Example:
$ # -- Compile everything (including doxygen documentation)
$ mkdir <someBuildDir> && cd <someBuildDir>
$ cmake \
-DCMAKE_BUILD_TYPE=Release \
-DCMAKE_INSTALL_PREFIX=/tmpLocal/ \
-DCMAKE_PREFIX_PATH=/tmpLocal/ \
/repos/engabra
$ cmake --build . --target all -j `nproc`
$ # -- Run library unit tests
$ ctest -j `nproc`
$ # -- Create distribution packages
$ cpack
$ # -- Installation
$ dpkg --contents ./Engabra*
$ sudo apt-get install ./Engabra-0.1.0-Linux.deb # Install
$ sudo apt-get remove engabra # Remove
```
NOTE:
The source version identity check in test_helloEngabra may fail
if the build environment is associated with a different user ID
than the one owning the git repo. To address this, from the build
environment, issue a git command similar to:
$ git config --global --add safe.directory /repos/Engabra
This allows the build process to run the "git describe" command
which extracts source code identifier to be included as version
info in the build result.
```
* cmake --build . --target all -j `nproc`
Assuming that Engabra is cloned to /repos/engabra, then
$ mkdir <someBuildDir> && cd <someBuildDir>
$ cmake \
-DCMAKE_BUILD_TYPE=Release \
/repos/engabra/demos
$ make -j `nproc`
$ # -- Run individual programs as desired (relative to build dir) e.g.
$ ./demos/helloEngabra
Provided "doxygen" tool is available, library reference documentation (on classes, files, functions, and more) is generated automatically during the cmake build process described in section Code Build Commands.
After building system (with cmake), point browser to:
<cmakeBuildDir>/doc/html/index.html
After installation of the package documents are found in appropriate system install locations. E.g. on linux systems, often in:
<installPath>/share/doc/engabra/html/index.html
-
Chappell JM, Iqbal A, Gunn LJ, Abbott D (2015) Functions of Multivector Variables. PLoS ONE 10(3): e0116943. doi:10.1371/journal.pone.0116943, (at journals.plos.org)
-
Chappell JM, Iqbal A, Abbott D (2015), "Geometric Algebra: A natural representation of three-space", arXiv:1101.3619v3 [physics.hist-ph] 16 Apr 2015, (at arXiv)
-
Chappell JM, Hartnett JG, Iannella N, Iqbal A and Abbott D (2016) Time As a Geometric Property of Space. Front. Phys. 4:44. doi: 10.3389/fphy.2016.00044 (at Frontiers in Physics)
-
Chappell,... "Deriving time from the geometry of space" https://arxiv.org/pdf/1501.04857v2.pdf
-
Chappell,... "A new description of space and time using Clifford multivectors" https://arxiv.org/pdf/1205.5195.pdf
-
https://www.researchgate.net/publication/281948670_Some_remarks_on_Cl3_and_Lorentz_transformations