Versor (vsr)

A (Fast) C++ library for Euclidean and Conformal Geometric Algebra.

Currently tested on Linux and Mac OS X

Homepage (

Versor is a C++ Library for Geometric Algebra with built-in draw routines. Version 2.0 generates optimized geometric algebra code at compile-time through template meta-programming, and supports arbitrary dimensions and metrics (limited by your compiler…).

Developer: Pablo Colapinto
gmail: wolftype


Download and Installation Instructions

Reference Guide to the Elements

Mailing List (for update notifications, to ask questions, discuss bugs, etc)

versor.js - a Javascript Port of Versor

My Master’s Thesis on the Subject

Look at the AlloSphere Research Group

As long as algebra and geometry have been separated, their progress have been slow and their uses limited; but when these two sciences have been united, they have lent each mutual forces, and have marched together towards perfection.  
-Joseph Louis Lagrange

No attention should be paid to the fact that algebra and geometry are different in appearance.
-Omar Khayyám

L’algèbre n’est qu’une géométrie écrite; la géométrie n’est qu’une algèbre figurée.
-Sophie Germain

If you want to see, learn how to act
-Heinz von Foerster 



Please see also the INSTALL guide. For version 2.0 you need C++11 support (gcc 4.7 or higher or clang 3.2 or higher)

git clone git://
cd vsr2.0
git submodule init
git submodule update

To test a graphics example

make tests/xBasic.cpp 

which both builds and runs the file.

Not Working? —

For C++11 you’ll want clang 3.2 (mac) or above or gcc 4.7 or above (linux).

For clang on a snow leopard (with thanks to Karl Yerkes for this tidbit)

brew tap home-brew/versions
brew install --HEAD llvm34 --rtti, --disable-assertions, --with-libcxx, --with-clang

If you don’t want to or can’t compile C++11 code try an older flavor of vsr. This older version runs just as fast, but is strictly 3D CGA (i.e. R4,1 metric) since I generated headers ahead of time.


  1. CLANG=path/to/clang++/ (default usr/local/bin/clang++) The makefile assumes clang is at usr/local/bin/clang++ – if you want to change that set this flag

  2. GCC=1 (default 0) If you want to build with GCC instead set GCC=1.

  3. RPI=1 (default 0) vsr also builds on the Raspberry Pi ! ( with a cross-compiler ) (GCC=1 RPI=1) (note you cannot build from the pi itself, you must use a cross-compiler)

  4. GFX=0 (default 1) if you want to build without graphics support you can set GFX=0


Versor provides operations and draw routines for Euclidean and Conformal Geometric Algebras, a relatively new spatial computing model used by physicists, engineers, and artists. Versor is designed to make graphical experimentation of geometric algebra within a C++ environment easier. You can use this library to draw geometrical things, explore spherical and hyperbolic spaces, transformations, design robots, etc. I am using it for my PhD on bio-inspired engineering.

I first developed Versor while reading “Geometric Algebra for Computer Science” by Leo Dorst, Daniel Fontijne, and Stephen Mann. It’s a fantastic book and if you’re reading this you should also consider reading that.

Built to aid in my modelling of organic forms, the initial development was funded in large part by the Olivia Long Converse Fellowship for Botanic research, courtesy of the Graduate Division at the University of California in Santa Barbara. Currently supported by the Robert W. Deutsch Foundation, this software is under a UC Regents General Public License. Feel free to use and distribute as long as copyrights and credits are maintained.

One quick word: clifford algebras and the spatial relationships they embody can often feel abstract and daunting. But it’s a twisty, boosty ride, full of weird discoveries. You’re bound to make some, so have fun!


The homogenous 5D CGA model used here was initially proposed by David Hestenes, Hongbo Li, and Alan Rockwood in 2001, and given full form and weight through the excellent and careful work of Leo Dorst, Joan and Anthony Lasenby, and Eduardo Bayro-Corrochano, and others. These researchers’ writings have helped me quite a bit. CGA is particular breed of Clifford Algebras (also known as Geometric Algebras), which operate upon combinatoric hypercomplex vector spaces that emerged from William Clifford’s attempt to fuse Hamilton’s quaternions with Grassmans’ extension algebras. Thus transformations were married with a system of abstraction. For more information, take a look at the links to the sites at the bottom of this page. For instance, for practical applications in robotics and “Geometric Cybernetics”, see Eduardo Bayro-Corrochano’s work. For some very helpful algorithms in rigid body dynamics and gravitational physics see the variety of publications by Joan and Anthony Lasenby. To get at the beginning of it all, read David Hestenes’ New Foundations for Classical Mechanics.


This software is licensed under a general UC Regents General Public License. If you’re planning on using CGA inside a sellable product you should be aware that there is a vague patent on the use of 5D CGA which may limit its commercial use when encoding robotic control mechanisms, or may just limit your ability to patent the model itself. I hope and imagine it is the latter. Though powerful, elegant, and brilliant, the heart of CGA is just a quadratic equation and the arguments for the use of 5D CGA are that it is foundational and universal, the very two characteristics of a system which would make it un-patentable. The Clifford Algebras on which it is based are from the 19th century.


Typical matrix operation libraries have templated inlined functions for Vector and Matrix multiplication. Versor is similar, but on steroids, where vectors and sparse matrices of various sizes are all just called multivectors and represent geometric elements beyond just xyz directions and transformation matrices. Circles, lines, spheres, planes, points are all algebraic elements, as are operators that spin, twist, dilate, and bend those variables. Both these elements and operators are multivectors which multiply together in many many many different ways.

What’s new?

Version 2.0 compiles much faster than before, and without any silly predetermined list of allowable operations or types. Most notably, arbitrary metrics are now possible. For example, the xRoots.cpp example calculates all the Euclidean 4D reflections of a couple of point groups (F4 and D4, namely). So you can hypercube and polychoron away (8D cubes no problem!). Number of dimensions allowed are somewhat limited by your compiler infrastructure – let me know if you have a need that is not being met!

As for CGA, all the Pnt, Vec, Dll notation remains as before, but i’ve started adding utility functions since it helps people out.

auto pa = Ro::point( 1,0,0 ); auto pb = Ro::point( 0,1,0 ); auto pc = Ro::point(–1,0,0 ); auto circle = pa ^ pb ^ pc;


How does it work?

If you like mind-numbing functional template metaprogramming, take a look at the code and please let me know what you think. If you don’t, then I wouldn’t … But if you have ideas or questions please do not hesitate to contact me.


GA combines many other maths (matrix, tensor, vector, and lie algebras). It is holistic. CGA uses a particular mapping (a conformal one) of 3D Euclidean space to a 4D sphere. Operations on that hypersphere are then projected back down to 3D. That how it works in a nutshell.

A fuller treatment of this question (er, the question of why we do this) can be found in my Master’s thesis on the subject. But basically, Geometic Algebra offers a particular richness of spatial expression. Imagine needing glasses and not knowing you needed glasses. Then, when you do get glasses, the world changes unexpectedly. GA is like glasses for the inside of your brain. Conformal Geometric Algebra, especially the 5D variety enlisted here, are like x-ray glasses. One point of clarification that occurs are disambiguations of previously collapsed concepts.

For instance, the main disambiguation, is that between a Point in space and a Vector in space.
A Point has no magnitude, but a Vector does. A Point has no direction, but a Vector does. Points are null Vectors. We can make them by writing

Vec( 1,0,0 ).null();

More on that last point later … there are various binary operators defined (mainly three). We can introduce one right now, which is the dot or inner product. In mathematics, the inner product of two points pa and pb is written \(p_{a} \rfloor p_{b}\). In Versor we use the <= operator:

Point pa = Vec(1,0,0).null();
Point pb = Vec(-1,0,0).null();
Scalar squaredDist = ( pa <= pb ) * -2;

which in this case would return a Scalar value of 4. The -2 is there since the inner product really returns half the negative squared distance. We can extract the Scalar into a c++ double like so:

double squaredDist = ( pa <= pb )[0] * -2;

Points thought of as Spheres (really, Dual Spheres, more on Duality later): they are Spheres of zero radius. As such they are a type of Round element. We can also build points this way:

Round::null( 1,0,0 );

or you can pass in another element

Round::null( Vec(1,0,0) );

or use the built-in method

Point pa = Vec(1,0,0).null();

Points can also be made with the macro PT

Point pa = PT(1,0,0);

which is just “syntactic sugar” for Vec(1,0,0).null()

Speaking of Spheres, we can also make spheres with a radius this way:

DualSphere dls = Round::dls( Vec( 1,0,0 ).null(), 1 );


DualSphere dls = Round::dls( Vec( 1,0,0 ), 1 );

or, specifying the radius first and then the coordinate:

DualSphere dls = Round::dls( 1 /* <--radius */ , 1,0,0 )

all of which give a dual sphere of radius 1 at coordinate 1,0,0;


Versor is named after the one of the basic category of elements of geometric algebra.
A versor is a type of multivector which can be used to compose geometric transformations, namely reflections, translations, rotations, twists, dilations, and transversions (special conformal transformations).

More on all of those transformations later.

In Versor, a Vector (or Vec) is a typical Euclidean 3D element. It can be built in the normal way:

Vec v(1,2,3);

Some built-in Vectors exist:

Vec::x x; //<-- X Direction Unit Vector Vec(1,0,0)
Vec::y y; //<-- Y Direction Unit Vector Vec(0,1,0)
Vec::z z; //<-- Z Direction Unit Vector Vec(0,0,1)

A Vector can be spun around using a Rotor, which is exactly like a quaternion. However, whereas quaternions are often built by specifying an axis and an angle, rotors are built by specifying the plane of rotation. Eventually this will make much more sense to you: in general planes are what we will be using to transform things. For instance, a reflection is a reflection in a plane. As we will see, planes can become hyperplanes which will allow for more extraordinary transformations.

The first completely new element to introduce is the Bivector, which is the plane we will use to generate our Rotor. Bivectors represent directed areas and are dual to the cross product: the cross product of two vectors in typical vector algebra returns a vector normal to the plane they define. So it is not completely new, but just sort of new.

Bivectors are also just three elements long, and are built the same way Vectors are.

Biv b(1,2,3);

Some built-in Bivectors exist:

Biv::xy xy; //<-- XY Counterclockwise Unit Area Biv(1,0,0)
Biv::xz xy; //<-- XZ Counterclockwise Unit Area Biv(0,1,0)
Biv::yz xy; //<-- YZ Counterclockwise Unit Area Biv(0,0,1)

While it is perfectly valid to write Vector, Bivector and Rotor, you’ll notice I’ve truncated them to their three letter nicknames, Vec and Rot.
That’s up to you: Both long-name and nick-name versions are valid in libvsr (they are typedef’ed to each other).

Biv b = Biv::xy;
double theta = PIOVERTWO;
Vec v1 = Vec::x.rot( b * theta )

You can also generate rotors using Gen::rot( <some bivector> ) In fact, all transformations can be generated this way, and then later applied to arbitrary elements. For instance, Motors can be generated which translate and rotate an element at the same time. This is also called a twist.

Motor m = Gen::mot(<some dual line>);   //<-- Makes A Twisting Motor around Some Dual Line
Point p = Vec(0,0,0).null().sp(m);      //<-- Applies above motor to a Point

You’ll notice there are dual versions of elements: as in a DualLine (or Dll for short). That’s because in the real world of abstract geometry, there are usually two ways of defining an element. For instance, we can build a direct Line on the Y-axis by wedging two points together, along with infinity:

Line lin = Vec(0,0,0).null() ^ Vec(0,1,0).null() ^ Inf(1);

Or we can define a line by the bivector plane that it is normal to, and a support vector that determines how far away the line is from the origin. To convert the above line into its dual representation, we just call the dual() method:

Dll dll = lin.dual();

For those who are interested, this dual representation is isomorphic to the Plücker coordinates, which are used in screw theory to twist things around. Here, too, we can use dual lines to generate transformations which twist things around them.


The examples/*.cpp files include bindings to the GLV framework for windowing and user interface controls.
A GLVApp class and GLVInterface class provide the necessary glue to get started quickly.

The interface has a built in gui, mouse info, and keyboard info stored.

static Circle circle;
Touch(inteface, circle);

Putting the above code inside your application’s onDraw() loop will enable you to click and modify geometric elements by hitting the “G”, “R” and “S” keys.
Hit any other key to deselect all elements.

Key Response
~ Toggle full screen.
SHIFT + Mouse or Arrow Keys Translate the camera in x and z directions.
CTRL+ Mouse or Arrow Keys Rotate the camera
ALT +Arrow Keys Rotate the model view around.
G Grab an Element
R Rotate an Element
S Scale an Element
Any other key Release all Elements


The elements of the algebra are geometric entities (circles, planes, spheres, etc) and operators (rotations, translations, twists, etc) which act on the elements of the algebra. All are known as multivectors since they are more than just your typical vectors.

Multivector elements are most often combined using three overloaded binary operators:

The Geometric Product of elements A and B:

A * B

multiplies two multivector elements together. This is most useful when multiplying one by the inverse of another (see ! operator, below).

The Outer Product of elements A and B:

A ^ B

“wedges” two multivectors together. Its from Grassman’s algebra of extensions, and can be thought of as a way of creating higher dimensions from smaller ones. For instance, wedging two Vectors (directed magnitudes) together returns a Bivector (a directed Area). Wedging two Points together returns a PointPair. Wedging three Points together returns a Circle.

The Inner Product of elements A and B:


There is also a Commutator product (differential)


And a few overloaded operations, including,

The Inverse:


returns \(A^{-1}\)

The Reverse:


returns \(\tilde{A}\)

And finally, since I ran out of overloadable operators, some basic methods


which returns \(\bar{A}\)


which returns \(\hat{A}\)

In summary:

Versor Math Description
A * B \(AB\) Multiplies two elements together (and, in the case of A * !B finds ratios between elements).
A ^ B \(A \wedge B\) Wedges two elements together (builds up higher dimensional elements).
A <= B \(A \rfloor B\) or \(\boldsymbol{a} \cdot B\) Contracts A out of B (returns the part of B “least like A”, sort of).
A % B \(A \times B\) Commutator, equal to \(\frac{1}{2}(AB-BA)\)
!A \(A^{-1}\) The Inverse of A.
~A \(\tilde{A}\) The Reverse of A.
A.conj() \(\bar{A}\) Conjugation.
A.inv() \(\hat{A}\) Involution.


To make the process of writing code faster, all elements of the algebra are represented by types 3 letters long. Alternatively, you can also use the long-form name.

Type Long Form Descrription
Sca Scalar A real number
Vec Vector A Directed Magnitude, or 3D Vector, typical cartesian stuff
Biv Bivector A Directed Area. Use them to make Rotors: Gen::Rot( Biv b )
Tri Trivector A Directed Volume Element
Pnt Point A Null Vector: Pnt a = Vec(1,0,0).null()
Par PointPair A 0-Sphere (Sphere on a Line): Par par = Pnt a ^ Pnt b
Cir Circle A 1-Sphere: Cir cir = Pnt a ^ Pnt b ^ Pnt c
Sph Sphere A 2-Sphere: Sph sph = Pnt a ^ Pnt b ^ Pnt c ^ Pnt d
Dls DualSphere Typedef’ed as a point: typedef Pnt Dls
Lin Line A Direct Line: e.g. Lin lin = Par par ^ Inf(1)
Dll DualLine A Dual Line: e.g. Dll dll = lin.dual()
Pln Plane A Direct Plane: e.g. Pln pln = Cir cir ^ Inf(1)
Dlp DualPlane A Dual Plane: e.g. Dlp dlp =
Flp FlatPoint
Rot Rotor Spins an Element (as a Quaternion would)
Trs Translator Translates an Element
Dil Dilator Dilates an Element
Mot Motor Twists an Element along an axis
Trv Transversor Bends an Element about the Origin
Bst Booster Bends an Element around an “Orbit”
Mnk MinkowskiPlane
Pss Pseudoscalar
Inf Infinity

There are others as well (for instance, affine planes, lines, and points) but the above are more than sufficient to start with. There are also built in macros, for instance

EP | Sphere At the Origin.
EM | Imaginary Sphere at the Origin.
PT(x,y,z) | A null Point at x,y,z

EP and EM can be invoked instead of Inf to work in non-Euclidean metrics ( Spherical and Hyperbolic, respectively)

Many Euclidean elements can be drawn by invoking Draw::Render(). Some can’t (yet) either because it wasn’t obvious how to draw them (e.g the scalar) or because I just didn’t figure out how to do it or because I forgot or was lazy. If you want something to be drawable, let me know and I’ll add it in. Or try adding it in yourself and send a pull request via github.

All elements can be dualized by invoking their dual() method

All elements can be reflected over spinors with the sp(<spinor>) method

All elements can be reflected over versors with the re(<versor>) method

The versors are constructed by the geometric entities, typically by using the Gen:: routines. Operators can also be acted on by operators – you can rotate a translation, or twist a boost.


vsr_generic_op.h and vsr_cga3D_op.h contain the bulk of the functions for generating elements from other elements. Some guidelines:


Returns Function Description
Rot Gen::rot( const Biv& b ); //<– Generate a Rotor from a Bivector
Trs Gen::trs( const Drv& v); //<– Generate a Translator from a Direction Vector
Mot Gen::mot( const Dll& d); //<– Generate a Motor from a Dual Line
Dil Gen::dil( const Pnt& p, double amt ); //<– Generate a Dilator from a Point and an amount
Trv Gen::trv( cont Tnv& v); //<– Generate a Transveror from a Tangent Vector
Bst Gen::bst( const Par& p); //<– Generate a Booster from a Point Pair


In addition to the above “even” spinors, we can also reflect. Reflections (in a sphere, circle, or point pair, or over a line or plane ) can be calculated by writing

Pnt p = PT(1,0,0);
Pnt r = CXY(1) ); //Reflection of a point in a circle
r = r / r[3];           //Renormalization of a point

The re() method calculates as C*v.inv()*~C. With a versor C and an element v you might also try C * v * !C. Inversion in a circle or a sphere may change the weight of the element (for at Point at x, it will change it by x^2)