ON QUATERNIONS

By

William Rowan Hamilton

(Proceedings of the Royal Irish Academy, 3 (1847), pp. 1–16.)

Edited by David R. Wilkins

1999
 On Quaternions. By Sir William R. Hamilton.
Read November 11, 1844.

[Proceedings of the Royal Irish Academy, vol. 3 (1847), pp. 1–16.]

In the theory which Sir William Hamilton submitted to the Academy in November,
1843, the name quaternion was employed to denote a certain quadrinomial expression, of
which one term was called (by analogy to the language of ordinary algebra) the real part,
while the three other terms made up together a trinomial, which (by the same analogy) was
called the imaginary part of the quaternion: the square of the former part (or term) being
always a positive, but the square of the latter part (or trinomial) being always a negative
quantity. More particularly, this imaginary trinomial was of the form ix + jy + kz, in which
x, y, z were three real and independent coefficients, or constituents, and were, in several
applications of the theory, constructed or represented by three rectangular coordinates; while
i, j, k were certain imaginary units, or symbols, subject to the following laws of combination
as regards their squares and products,

i2 = j 2 = k 2 = −1, (A)

ij = k, jk = i, ki = j, (B)
ji = −k, kj = −i, ik = −j, (C)
but were entirely free from any linear relation among themselves; in such a manner, that to
establish an equation between two such imaginary trinomials was to equate each of the three
constituents, xyz, of the one to the corresponding constituent of the other; and to equate
two quaternions was (in general) to establish four separate and distinct equations between
real quantities. Operations on such quaternions were performed, as far as possible, according
to the analogies of ordinary algebra; the distributive property of multiplication, and another,
which may be called the associative property of that operation, being, for example, retained:
with one important departure, however, from the received rules of calculation, arising from
the abandonment of the commutative property of multiplication, as not in general holding
good for the mixture of the new imaginaries; since the product ji (for example) has, by its
definition, a different sign from ij. And several constructions and conclusions, especially as
respected the geometry of the sphere, were drawn from these principles, of which some have
since been printed among the Proceedings of the Academy for the date already referred to.
The author has not seen cause, in his subsequent reflections on the subject, to abandon
any of the principles which have been thus briefly recapitulated; but he conceives that he
has been enabled to present some of them in a clearer view, as regards their bearings on
geometrical questions; and also to improve the algebraical method of applying them, or what
may be called the calculus of quaternions.

1
 Thus he has found it useful, in many applications, to dismiss the separate consideration
of the three real constituents, x, y, z, of the imaginary trinomial ix + jy + kz, and to denote
that trinomial by some single letter (taken often from the Greek alphabet). And on account
of the facility with which this so called imaginary expression, or square root of a negative
quantity, is constructed by a right line having direction in space, and having x, y, z for its
three rectangular components, or projections on three rectangular axes, he has been induced
to call the trinomial expression itself, as well as the line which it represents, a vector. A
quaternion may thus be said to consist generally of a real part and a vector. The fixing a
special attention on this part, or element, of a quaternion, by giving it a special name, and
denoting it in many calculations by a single and special sign, appears to the author to have
been an improvement in his method of dealing with the subject: although the general notion
of treating the constituents of the imaginary part as coordinates had occurred to him in his
first researches.
Regarded from a geometrical point of view, this algebraically imaginary part of a quater-
nion has thus so natural and simple a signification or representation in space, that the diffi-
culty is transferred to the algebraically real part; and we are tempted to ask what this last
can denote in geometry, or what in space might have suggested it.
By the fundamental equations of definition for the squares and products of the symbols i,
j, k, it is easy to see that any (so-called) real and positive quantity is to any vector whatever,
as that vector is to a certain real and negative quantity; this being indeed only another mode
of saying that, in this theory, every vector has a negative square. Again, the product of any
two rectangular vectors is a third vector at right angles to both the factors (but having one
or other of two opposite directions, according to the order in which those factors are taken); a
relation which may be expressed by saying, that the fourth proportional to the real unit and
to any two rectangular vectors is a third vector rectangular to both; or, conversely, that the
fourth proportional to any three rectangular vectors is a quantity distinct from every vector,
and of the kind called real in this theory, as contrasted with the kind called imaginary.
Now, in fact, what originally led the author of the present communication to conceive
(in 1843) his theory of quaternions (though he had, at a date earlier by several years, spec-
ulated on triplets and sets* of numbers, as an extension of the theory of couples, or of the
ordinary imaginaries of algebra, and also as an additional illustration of his view respecting
the Science of Pure Time), was a desire to form to himself a distinct conception, and to find
a manageable algebraical expression, of a fourth proportional to three rectangular lines, when
the directions of those lines were taken into account; as Mr. Warren† and Mr. Peacock‡ had
shewn how to conceive and express the fourth proportional to any three lines having direc-
tion, but situated in one common plane. And it has since appeared to Sir William Hamilton
that the subject of quaternions may be illustrated by considering more closely, though briefly,
this question of the determination of a fourth proportional to three rectangular directions in
space, rather in a geometrical than in an algebraical point of view.
Adopting the known results above referred to, for proportions between lines having

* See Transactions of the Royal Irish Academy, vol. xvii. p. 422. Dublin, 1835.
† Treatise on the Geometrical Representations of the Square Roots of Negative Quantities,
by the Rev. John Warren. Cambridge, 1828.
‡ Treatise on Algebra, by the Rev. George Peacock. Cambridge, 1830.

2
direction in a single plane (though varying a little the known manner of speaking on the
subject), it may be said that, in the horizontal plane, “West is to South as South is to East,”
and generally as any direction is to one less advanced than itself in azimuth by ninety degrees.
Let it be now assumed, as an extension of this view, that in some analogous sense there exists
a fourth proportional to the three rectangular directions, West, South, and Up; and let this
be called, provisionally, Forward, by contrast to the opposite direction, Backward, which
must be assumed to be (in the same general sense) a fourth proportional to the directions of
West, South and Down. We shall then have, inversely, Forward to Up as South to West, and
therefore, as West to North: if we admit, as it seems natural and almost necessary to do, that
(for directions, as for lengths) the inverses of equal ratios are equal; and that ratios equal to
the same ratio are equal to each other. But again, Up is to South as South to Down, and
also as North to Up: and we can scarcely avoid admitting, or defining, that (in the present
comparison of directions) ratios similarly compounded of equal ratios are to be considered
as being themselves equal ratios. Compounding, therefore, on the one hand, the ratios of
Forward to Up, and of Up to South; and on the other hand the respectively equal (or similar)
ratios of West to North, and of North to Up, we are conducted to admit that Forward is
to South as West to Up. By a reasoning exactly similar, we find that Forward is to West
as Up to South; and generally that if X, Y , Z denote any three rectangular directions such
that A : X :: Y : Z, A here denoting what we have expressed by the word Forward, then
also A : Y :: Z : X (and of course, for the same reason A : Z :: X : Y ); so that the three
directions X Y Z may be all changed together by advancing them in a ternary cycle, according
to the formula just written, without disturbing the proportionality assumed. But also, by the
principle respecting proportions of directions in one plane, we may cause any two of the three
rectangular directions XY Z to revolve together round the third, as round an axis, without
altering their ratio to each other. And by combining these two principles, it is not difficult to
see that because Forward has been supposed to be to Up as South to West, therefore the same
(as yet unknown) direction “Forward” must be supposed to be to any direction X whatever,
as any direction Y , perpendicular to X, is to that third direction Z which is perpendicular to
both X and Y , and which is obtained from Y by a right-handed (and not by a left-handed)
rotation, through a right angle, round X; in the same manner as (and because) the direction
West was so chosen as to be to the right of South, with reference to Up as an axis of rotation.
Conversely we must suppose that if any three rectangular directions, X Y Z, be arranged, as
to order of rotation, in the manner just now stated, then Z : Y :: X : A; or in other words,
we must admit, if we reason in this way at all, that the direction called already Forward, will
be the fourth proportional to Z Y X. And if we vary the order, so as to have Z to the left,
and not to the right of Y , with reference to X, then will the fourth proportional to Z Y X
become the direction which we have lately called Backward, as being the opposite to that
named Forward.
Again, since Forward is to Up as South to West, that is in a ratio compounded of the
ratios of South to East and of East to West, or in one compounded of the ratios of West to
South, and of any direction to its own opposite; or, finally, in a ratio compounded of the ratios
of Up to Forward and of Forward to Backward, that is, in the ratio of Up to Backward, we see
that the third proportional to the directions Forward and Up is in the direction Backward:
and by an exactly similar reasoning, with the help of the conclusions recently obtained, we
see that if X be any direction in tridimensional space, then A : X :: X : B; B here denoting,

3
for shortness, the direction which has been above called Backward.
The geometrical study of the relations between directions in space, combined with a few
very simple and guiding principles respecting the composition of relations generally, might
therefore have led to the conception or assumption of a certain pair of contrasted directions,
namely, those which we have called Forward and Backward, and denoted by the letters A and
B. And these are such that if we conceive a quantitative element to be combined with each,
and give the name of positive unity to the unit of magnitude measured in the direction of
Forward, but that of Negative Unity to the same magnitude measured backward; and if we
extend to this positive unity and to lines having direction in space the received definitions
of multiplication, that “Positive Unity is to Multiplier as Multiplicand is to Product,” and
that “the product of two equal factors is the square of either;” we may then consistently and
naturally be led to assert the same results as those already enunciated from the theory of
quaternions respecting the product of two vectors, in the two principal cases, first, where
those two vectors are rectangular, and second, where they are coincident with each other.
And thus may we justify, or at least interpret and explain, the fundamental definitions (A)
(B) (C) of this theory, by regarding the symbols i j k as denoting three vector-units having
three rectangular directions in space.
But, farther, we derive from this view of the whole subject an illustration (if not a
confirmation) of the remarkable conclusion that the so-called real and positive unit +1 is not
(in this theory) to be confounded with any vector unit whatever, but is to be regarded as
of a kind essentially distinct from every vector. For this positive unit +1 is in the direction
above called Forward, and denoted by A. Now if this could coincide with a direction X in
tridimensional space, then, whatever this latter direction might be supposed to be, we could
always, by the general formula A : X :: Y : Z (where X is arbitrary), deduce the inadmissible
proportion X : X :: Y : Z, in which the two directions in one ratio are identical, but those
in the other are rectangular to each other. If then we resolve to retain the assumption of
the existence of a fourth proportional A to three rectangular directions in space, as subject
to be reasoned on at all in the way already described, and as determined in direction by
its contrast to its own opposite B (corresponding to an opposite order of rotation in the
system X Y Z), we must think of these two opposite directions A and B as merely laid down
upon a scale, but must abstain from attributing to this scale any one direction rather than
another in tridimensional space, as having such or such a zenith distance, or such or such an
azimuth, rather than such or such another. And the progression on this scale from negative
to positive infinity, obtained by combining a quantitative element with the contrast between
two opposite directions, corresponds less to the conception of space itself (though we have
seen that considerations of space might have suggested it) than to the conception of time;
the variety which it admits is not tri- by uni- dimensional; and it would, in the language of
some philosophical systems, be said to appertain rather to the notion of intensive than of
extensive magnitude. Though answering precisely to the progression of the quantities called
real in algebra, it has, when viewed from the geometrical side, somewhat the same sort of
imaginariness, and yet (it is believed) of utility, as compared with lines in space, which the
square root of an ordinary negative has, when compared with positive and negative quantities.
This analogy becomes still more complete when we observe that (in this theory) the fourth
proportional to any direction X in space, and either of the two directions A or B upon the
scale, is the direction opposite to X; so that, if a vector-unit in any determined direction X

4
had been taken for positive unity, then each of the two scalar units in the directions A and
B (in common, it is true, with every vector-unit perpendicular to X) might have been called,
by the general nomenclature of multiplication, a square root of negative one.
It is, however, a peculiarity of the calculus of quaternions, at least as lately modified by
the author, and one which seems to him important, that it select no one direction in space as
eminent above another, but treats them as all equally related to that extra-spatial, or simply
scalar direction, which has been recently called “Forward.” In this respect it differs in its
processes from the Cartesian method of coordinates, and seems often to admit of being more
simply and directly applied to the treatment of geometrical problems, because it requires
no previous selection of axes, rectangular or other. The author is, indeed, aware that the
cooperation of other and better analysts will be necessary in order to bring the method of
quaternions to anything approaching to perfection. But he hopes that an instance or two of
the facility with which some questions at least allow themselves to be treated by this method,
even in its present state, may not be without interest to the Academy. And he conceives that
two examples in particular, one relating to the composition of translations, and the other to
the composition of rotations in space, may usefully be selected for statement on the present
occasion.
As preliminary illustrations of the operations employed, it may be remarked that for any
system of lines having direction in space, it is required by many analogies (and is, for lines
in one plane included among the definitions or results of the theories of Mr. Warren and Mr.
Peacock), that the sum should be regarded as being equal to that one line which constructs
or represents the total effect of all the different rectilinear motions which are expressed by the
different summands. Vectors are therefore to be added to each other by a certain geometrical
composition, exactly analogous to the composition of motions, or of forces, and following the
same known rules. Scalars, on the other hand (that is to say, the so-called real parts of any
proposed quaternions), admitting only of a progression in quantity, and of a change of sign,
without any other changes of direction, are to be added among themselves by the known rules
of algebra, for the addition of positive and negative numbers. The addition of a scalar and a
vector to each other can be no otherwise performed, or rather indicated, than by writing their
symbols with the + sign interposed; each being, as we have seen, in some sense, imaginary
with respect to the other. These operations of addition are all of the commutative, and also
of the associative kind; that is to say, the order of all the summands may be changed, and
any group of them may be collected or associated into one partial sum.
Scalars are multiplied, as well as added, by the known rules of ordinary algebra, for the
multiplication of real numbers, positive or negative; because the positive unity of the system
has been assumed to be itself a scalar, and not a vector unit.
For the same reason, to multiply any vector by any scalar a, is in general to change its
length in a known ratio, and to preserve or reverse its direction, according as a is > or < 0;
the product is therefore a new vector, which may be denoted by aα. The same new vector
is obtained, under the form αa, when we multiply the scalar a by the vector α. If a + α and
b + β be two quaternion factors, of which a and b are the scalar parts, and α, β the vectors,
then with a view to preserving the distributive character of multiplication, it is natural to
define that the product may be distributed into the four following parts:

(a + α)(b + β) = ab + aβ + αb + αβ.

5
And if the multiplicand vector β be decomposed into two parts, or summands, one = β1 and
in the direction of the multiplier α, or in the direction exactly opposite thereto, and the other
= β2 , and in a direction perpendicular to the former (so that β1 and β2 are the projections
of β on α itself, and on the plane perpendicular to α), then it may be farther defined that
the multiplication of any one vector β by any other vector α may be accomplished by the
formula
αβ = α(β1 + β2 ) = αβ1 + αβ2 ;
in which, by what has been shewn, the partial product αβ1 is to be considered as equal to a
scalar, namely, the product of the lengths of α and β1 , taken with the sign − or +, according
as the direction of β1 coincides with, or is opposite to that of α; while the other partial
product αβ2 is a vector, of which the length is the product of the lengths of α and β2 , while
its direction is perpendicular to both of their’s, being obtained from that of β2 , by making
it revolve right-handedly through a right angle round α as an axis. These definitions, which
are compatible with the formulæ (A), (B), (C), and may serve to replace them, will be found
sufficient to prove generally, and perhaps with somewhat greater geometrical clearness than
those formulæ, the distributive and associative properties of quaternion multiplication, which
have already been stated to exist. They give easily the following corollaries, which are of very
frequent use in this calculus:

αβ + βα = 2αβ1 = 2AB cos(A, B); (a)

αβ − βα = 2αβ2 = 2γAB sin(A, B); (b)

A and B denoting here the lengths of the lines α and β, and (A, B) the angle between them;
while γ is a vector-unit perpendicular to their plane, and such that a right-handed rotation,
equal to the angle (A, B), performed round γ, would bring the direction of α to coincide with
that of β. For example, when β = α, then B = A, (A, B) = 0, and

αβ = βα = α2 = −A2 ,

so that the length A of any vector α, in this theory, may be expressed under the form
p
A = −α2 . (c)

More generally we have the equation

αβ − βα = 0, (d)

when the lines α and β are coincident or opposite in direction; while, on the contrary, the
condition for their being at right angles to each other is expressed by the formula

αβ + βα = 0. (e)

These simple principles suffice to give, in a new way, algebraical solutions of many
geometrical problems, of various degrees of difficulty and importance. Thus, if it be required,
as an easy instance, to determine the length of the resultant of several successive rectilinear

6
motions, or the magnitude of the statical sum of several forces acting together at one point,
as a function of the amounts of those successive motions, or of those component forces, and
of their inclinations to each other, we have only to denote the components by the vectors
α1 , α2 , . . . αn , and their sum by α, the corresponding magnitudes being A1 , A2 , . . . An , and
A; and the equation
α = α1 + α2 + · · · + αn
will give, by being squared,

α2 = α12 + α22 + · · · + αn2

+ α1 α2 + α2 α1 + · · · + α1 αn + αn α1 + · · · ;

that is, by the foregoing principles (after changing all the signs),

A2 = A21 + A22 + · · · + A2n

+ 2A1 A2 cos(A1 , A2 ) + · · · + 2A1 An cos(A1 , An ) + · · · ;

a known result, it is true, but one which can scarcely be derived in any other way by so
very short a process of calculation. For it is not quite so easy, on the algebraical side of the
question, to see that

( x)2 + ( y)2 + ( z)2 = (x2 + y 2 + z 2 ) + 2 (xx0 + yy 0 + zz 0 ),

P P P P P

however easy this may be, as it is to see that

α) = (α ) + (αα0 + α0 α) :
P 2 P 2 P
( (f)

although the geometrical interpretation of the first of these two formulæ is of course more
obvious than that of the latter, to those who are familiar with the method of coordinates,
and not with the method of quaternions.
Again, let us consider the more difficult problem of the composition of any number os
successive rotations of a body, or, at first, of any one line thereof, round several successive
axes, through any angles, small or large. Let the axis of the first of these rotations have the
direction of the vector-unit α, (α2 = −1), and let the amount of the positive rotation round
this axis be denoted by a, which letter here represents still a scalar or real number. Let β be
the revolving line, considered in its original position; β 0 the same line, after it has revolved
through the angle a round the axis α. The part, or component, of β, which is in the direction
of this axis, is that which was denoted lately by β1 ; and the formula (a), when multiplied by
− 21 α, gives, as an expression for this part,

β1 = 21 (β − αβα), (g)

because it has been supposed that α2 = −1. This part of β remains unaltered by the rotation.
The other part, or component of β, is, in like manner, by (b),

β2 = 12 (β + αβα); (h)

7
and this part is to be multiplied by cos a, in order to find the part of β 0 , which is perpendicular
to α, but in the plane of α and β. Again, multiplying by α, we cause β2 to turn through a
right angle in the positive direction round α, and obtain, for the result of this rotation,

αβ2 = 21 (αβ − βα);

an expression which is the half of that marked (b), and which is to be multiplied by sin a,
in order to arrive at the remaining part of the sought line β 0 , namely, the part which is
perpendicular to the plane of α and β. Collecting, therefore, the three parts, or terms, which
have been thus separately obtained, we find,

β 0 = β1 + (cos a + α sin a)β2

= 12 (β − αβα) + 12 cos a(β + αβα) + 12 sin a(αβ − βα)
 a 2  a 2 a a
= cos .β − sin .αβα + cos sin .(αβ − βα);
2 2 2 2
that is,  a a  a a
β 0 = cos + α sin β cos − α sin ; (i)*
2 2 2 2
the operations here indicated being thus sure to make no change in the part β1 , which is in
the direction of the axis of rotation, but to cause the other part β2 to revolve round that

*[Note added during printing.]—The printing of this abstract having been delayed, the
Author desires to be permitted to append the following remarks:
If we should make, for abridgment
a
α tan = −γ,
2
the formula (i) for any single rotation might be thus written,

β 0 = (1 + γ)−1 β(1 + γ). (i0 )

And if we then made

β = ix + jy + kz, β 0 = ix0 + jy 0 + kz 0 , γ = iλ + jµ + kν,

i, j, k, being the same three rectangular vectors, or imaginary units, as in the formulæ (A)
(B) (C), but x, y, z, x0 , y 0 , z 0 , λ, µ, ν, being nine real or scalar quantities, we should obtain
the same general formula for the transformation of rectangular coordinates (with the same
geometrical meanings of the coefficients λ, µ, ν,) as that which Mr. Cayley has deduced, with
a similar view, but by a different process, and has published, with other “Results respecting
Quaternions,” in the Philosophical Magazine for February, 1845.
The present writer desires to return his sincere acknowledgments to Mr. Cayley for the
attention which he has given to the Papers on Quaternions, published in the above-mentioned
Magazine: and gladly recognizes his priority, as respects the printing of the formula just now
referred to. But while he conceives it to be very likely that Mr. Cayley, who had previously

8
axis α through an angle = a. Again, let the same line β 0 revolve round a new axis of rotation
denoted by a new vector unit α0 , through a new angle a0 , into a new position β 00 ; we shall
have, in like manner,

a0 a0 a0 a0
   
00 0 0 0
β = cos + α sin β cos − α sin ; (j)
2 2 2 2

and so on, for any number n of rotations. Let the last position of β be denoted by βn ;
and since it can easily be proved, by the theory of multiplication of quaternions, that the
continued products which present themselves admit of being thus transformed:

0 0
a(n−1) (n−1)
   
a a a
+ α(n−1) sin · · · cos + α0 sin

cos 

2 2 2 2 



  
a a an an 

cos + α sin = cos + αn sin ; 

2 2 2 2 

(k)
a0 0
  
a a a 
cos − α sin cos − α0 sin ···



2 2 2 2 



(n−1) (n−1)
  
a (n−1) a an an 

cos −α sin = cos − αn sin ; 

2 2 2 2


in which αn is a new vector unit, and an a new real angle, we find that the result of all the
n rotations is of the form
 an an   an an 
βn = cos + αn sin β cos − αn sin . (l)
2 2 2 2
It conducts, therefore to the same final position which would have been attained from the
initial position β, by a single rotation = an , round the single axis αn ; the amount and axis
of this resultant rotation being determined by either of the two equations of transformation
(k), and being independent of the direction of the line β which was operated on, so that they
are the same for all lines of the body.
If the present results be combined with the theorem marked (R), in the account, printed
in the Proceedings of the Academy, of the remarks made by the Author in November, 1843,
it will at once be seen that if the several axes of rotation be considered as terminating in the
points of a spherical polygon, and if the angles of rotation be equal respectively to the doubles
of the angles of this polygon (and be taken with proper signs or directions, determined by

published in the Cambridge Mathematical Journal some elegant researches on the rotation of
bodies, may have perceived, not only independently, but at an earlier date than he did himself,
the manner of applying quaternions to represent such a rotation; he yet hopes that he may
be allowed to mention, that a formula differing only slightly in its notation from the formula
(i) of the present abstract, with the corollaries there drawn respecting the composition of
successive finite rotations, had been exhibited to his friend and brother Professor, the Rev.
Charles Graves, of Trinity College, Dublin, in an early part of the month (October, 1844),
which preceded that communication to the Academy, of which an account is given above.

9
those angles), then the total effects of all these rotations will vanish; or, in other words, the
body will at last be brought back to the position from which it set out.
Finally, it may be mentioned that the author is in possession of a general method for
expressing by quaternions the tangent planes and normals to curved surfaces, and that in
applying this method to find the cone of tangents enveloping a given sphere, and drawn from
a given point, the geometrical impossibility of the problem, when the point is an internal one,
is expressed by the square of a vector becoming in this case positive.

10