On relating computation
Fran¸coise Chatelin ∗
CERFACS Technical Report TR/PA/11/37
Abstract : The report shows how the seminal ideas of Fourier, Poincar´e and Einstein blend together harmoniously to explain many features of computation in Nature and in the human mind. 1)The Fourier analysis of complex signals leads to the Fourier transform whose complex spectrum lies on the unit circle, with 4 eigenvalues +1, 1, i, i which are the 4 units of the Gaussian ring of complex integers. − − 2)The Poincar´eapproach to relativity bears on the Lorentz transformations in the field H of quaternions, using a noncommutative . Relative significant computation evolves from H to G2, where G is the alternative× algebra of octonions. 3)The Einstein perspective on relativity is based on a noncommutative + , yield- ing a geometric 2-fold information potential with 2 types of geodesics. The potential◦ lies in a R3-framed metric cloth, a non-euclidean space which is a computational construct exhibiting some features attributed to axiomatically defined geometries, either hyperbolic or elliptic. By mixing these 3 views with the quadratic logistic iteration and hypercompu- tation, we uncover an algorithmic mechanism which underlies the law of organic causality that we experience in Nature and mind.
Keywords : Einstein, Poincar´e, Fourier, noncommutative +, nonassociative , quaternions, octonions, sedenions, relativity, causal space, logistic iteration. ×
∗Ceremath, Universit´eToulouse 1 and CERFACS,42 avenue G. Coriolis 31057 Toulouse Cedex 1, France. E-mail: [email protected]
1 1 Introduction
1.1 Presentation In the report TR/PA/11/27 we studied the computational principle of relativity from the point of view of a noncommutative addition + creating a metric cloth ◦ framed into a normed linear vector space. In this report we investigate the alterna- tive point of view on relativity stemming from a noncommutative multiplication which defines recursively the quadratic Dickson algebras A of dimension 2k, k ×2 k ≥ (Chatelin 2011a, Chapter 2). The two approaches to relativity have emerged con- currently in 1905, presented respectively by Einstein and Poincar´e. In (Chatelin 2011b) we described the key role of the nonstandard Einstein addition + for weav- ◦ E ing information processing (WIP) by geometric means in the metric cloth WE. Below we present the equally important role of the multiplication when it × becomes noncommutative over the quaternions in A2 = H and nonassociative over the octonions in A3 = G, the two largest real quadratic division algebras. The rel- ativistic computational evolution in H and G is sustained by the dynamics of the logistic iteration (Chapter 6). By means of the organic logic derived from cloth ge- ometry, it wraps up the perspectives of Fourier, Poincar´eand Einstein on Relativity in computation into a holistic context whose number of dimensions can vary from 1 to 14.
1.2 A review of Dickson algebras The reader is referred to Chapter 2 in (Chatelin 2011a). The Dickson doubling process in Ak = Ak 1 Ak 1, A0 = R defines recursively two operations: − − conjugation: (x, y×)=(¯x, y), • − multiplication: •
(x, y) (x′,y′)=(x x′ y¯ y),y′ x + y x¯ ). × × − ′ × × × ′ For k 1, x A satisfies the quadratic relation x2 = xx¯ +(x +x ¯), and A ≥ ∈ k − k represents a complexified version of Ak 1 : Ak = Ak 1 Ak 1 1˜k where 1˜k = − − ⊕ − × (0, 1k 1) Ak 1 is the complex unit in Ak, k 1. − ∈ − ≥ C C For k = 1, i = (0, 1) is the complex unit of . For k 2, the 2D algebra 1˜ ˜ ˜ C ≥ C spanned by 1 = 1k and 1= 1k is isomorphic to . We write Ak = Ak 1 ⋆ 1˜ , k 2. − k ≥ For k = 2, 1˜2 = (00, 10) = e2 = j in H = C⋆Cj, and for k = 3, 1˜3 = (0000, 1000) = e4 G H C in = ⋆ e4 .
2 We recall that all Ak, k 4, have zerodivisors and an algebraic depth da(k) = k 3 ≥ 2 − 2 (Chapter 3, Chatelin 2011a). For k 3, the four algebras admit division ≥ ≤ and euclidean rings of integers (Chapter 9). Moreover, the algebraic depth of Ak is 1, 2 for k = 2, and 1 for k =0, 2, 3. { } Unless otherwise stated, all references are to the book Qualitative Computing (Chatelin 2011a).
2 The hypercomputation approach to relativity
2.1 Poincar´eon relativity Let q = α + X represent a quaternion in H with real part q = α R and ℜ ∈ imaginary part q = X H = R3. One computes q2 = α2 X 2 +2αX with ℑ ∈ ℑ ∼ − q2 = α2 X 2 and q2 =2αX colinear with q. ℜ According− to (Poincar´e1905)ℑ the group of Lorentzℑ transformations leaves q2 = ℜ f R invariant when q H. This is the aspect of the mathematical content of∈ the relativity principle which∈ is related to multiplication rather than addition. Phrased into physical terms, this principle explains why it is impossible to exhibit experimentally an absolute movement of the earth by mechanical or electromagnetic means.
2.2 Significant evolutions in relativity We observe that the Lorentz invariance can be interpreted as ( q)2 = X 2 + f, which fixes q2 at f R. Among the possible evolutions of q satisfyingℜ the relativity condition weℜ choose∈ specific ones by enforcing a complementary relation between ( q)2 and q2 2 according to the ℜ ℑ Definition 2.1 The significant relative evolutions of q = α + X in H, X 0 ≥ are such that f = q2 is a real constant and at least one of the two conditions (i) q = q2 , (ii) ℜ( q)2 = q2 2 is satisfied. |ℜ | ℑ ℜ ℑ The real parameter α in q will be allowed to become octonionic (α2 < 0). We first assume that X 2 > 0 (X H). ∈ℑ
3 2.3 The linear constraint (i) q = q2 |ℜ | ℑ R 1 The condition (i) tells us that α = 2 α X for any α , hence X = 2 for α = 0. | | | | ∈ Proposition 2.1 Under the linear constraint (i), the evolution of q = α + X takes place in H with (α + X, α + X) H2 iff f 1/4. When f < 1/4, α = β1˜ with β = α , β2 = α2 >−0 and the∈ evolution≥ of − g = X + β1˜ takes− place in G with ± − (X + α 1˜,X α 1)˜ in G2. − Proof. For X = 1/2, α2 = f +1/4 0 iff f 1/4. For f < 1/4, 2 2 ˜ ≥G ≥ − α α − α = α < 0, we set the complex unit 1 in to be either α or α , that is − − α = β1˜ with β = α . The number β1+˜ X is an octonion in G = H H 1˜ which can be written g ± =(X, α ). ⊕ × ± ± 2 We observe for future reference that, for f = 0, x = α equals 1 + 1/4f, f = 0. f 1 G R Proposition 2.2 For f 4 , g = (q, q¯) satisfies g = √4f +1 . For 1 ≥ − 2− ∈ 4 2 1± ∈ f < , g′ =(X, α ) G satisfies g′ = f and g′ = f > . − 4 ∈ℑ 16 1 2 1 Proof. 1) f 4 , q = α + X with α = fx = f + 4 0, 2α = ε√4f + 1, 2 ≥ −1 1 2 2 2 ≥ ε = 1, q = f + 2 4 , g =(q + q , q¯(q +¯q)) = ±2f + 1 + ε√4f +1≥ X , ε√4f +1¯ q =− ε√4f + 1(q, q¯). 2 − − Because G is a division algebra we conclude that g = ε√4f + 1, together with g 2 =2 q 2 =2f + 1. 1 2 2 G 2) f < 4 , α = α , α + X belongs to in the form g′ = (X, α ) such 2 − 2 − 2 ℑ that g′ = X + α = f > 0. − We have learnt in Chapter 3 that computation in G can be paradoxical by lack of associativity (Section 3.9). Proposition 2.2 provides a striking example of such 1 2 G paradoxes. For f 4 the squared norm g is computed in with two values 1 and 4f + 1 ≥0 which− differ for f = 0 (i.e. x finite). The transformation ≥ T : q H g = T (q)=(q, q¯) G triggers a remarkable reduction: the octonion∈g →G is converted into− the real∈ numbers g = √4f +1 = 2 q with g = g =∈ √2f + 1 unless f = 0. The reduction in dimensions± from± 8ℜ to 1 is | | metamorphosed into a metric expansion (f > 0) or contraction ( 1 f < 0). The 2 2 4 2 g g 2f − ≤ 1 | | − rate of change for g is ρ(f) = g 2 = 2f+1 , 1 ρ < 1 for f 4 . If − ≤ ≥ − 1 1 g f = , g = (q, q¯) = and g is computed at 0 when f , | | √2. 4 √2 g − − | | →∞ →
4 1 Proposition 2.3 For f = , the sedenion s =(g+,g ) A4 is a zerodivisor with 2 − norm 1. − ∈
1 1 2 Proof. For f = 2 , X = α = f = x = 2 , hence < g+,g >= X 2 − − 1 − − α = 0. Together with g+ = g (= ), this is a necessary and sufficient − √2 condition for s =(g+,g ) to be a zerodivisor in Zer (A4) = Aut (G) (Chapter 4). − ∼
3 The quadratic constraint (ii) ( q)2 = q2 2 ℜ ℑ We first assume that X H. ∈ℑ 3.1 Exact solutions We write (ii) α2 =4α2 X 2 =4α2(α2 f) under the form − x = 4fx(1 x) (3.1) 2 − − α 2 where x = f is the ratio of the real numbers α and f. This is the quadratic fixed-point equation known as the logistic (Section 6.8 in Chapter 6). The equation (3.1) has the two solutions a) f x = 0, b) 0 = →f x =1+ 1 , → 4f the second solution is the hyperbola solution of (i) on which X = 1 . 2 Proposition 3.1 a) The evolution of q = X remains in H for f 0 only under the condition X 2 0. ℑ ≤ ≥ b) The evolution of q = α + X for x = 1+ 1 , f = 0 is in H (resp. G2) for 4f f 1 , (resp. f < 1 ). ≥ − 4 − 4 Proof. a) x = α = 0, q = X and q2 = X 2 = f. For f 0, X 2 0; for − ≤ ≥ f > 0, X < 0 is impossible by assumption. 2 b) x = 1+ 1 , f = 0 is such that 0 < x = α < 1 for f < 1 . Necessarily 4f f − 4 f < α2 < 0 and α = β1˜ in G with β = α , X 2 > 0 hence g = X + β1=(˜ X, β). 1 1 ± 2 For f 4 , x 0 for 4 f < 0 and x> 1 for f > 0. In both cases, α > 0 implies α ≥R − and ≥X 2 =−α2 ≤f > 0. ∈ −
5 We observe on Figure 3.1 that the situations (i) and (ii) agree on the hyperbola ( X = 1 ). 2
1 1
1/4 0 1/4 0 −
X = 1 (i) X = 1 (ii) 2 2 α =0 G2 H2 H
Figure 3.1: Significant evolution in H, H2 or G2 in the plane (f, x) R2, X H ∈ ∈ℑ
3.2 Approximation by successive iteration We suppose that the resolution of (3.1) is realised by the iteration : 1 x0 = , xn = 4fxn 1(1 xn 1), n 1. (3.2) 2 − − − − ≥ 1 When n , the iterates xn do not escape to for f in [ 1, 2 ]. The inter- pretative context→∞ stemming from (f, x ) is ruled by the±∞ 5 values f − 1, 1 , 0, c, 1 n ∈ {− − 4 2 } where c is the value for the confluence of the iterates on the exact solution x = 0. 2 A According to (Nagashima and Baba 1999, p. 135), 3c = A + 2 1 where A = 1/3 − 19+3√33 , c 0.419. ≈ Lemma 3.2 c satisfies the relation 4c2(1 + c)=1. Proof. Simple check (computer algebra). The importance of the confluence parameter value c is that no odd period exists for f c. They appear in decreasing order for c < f 1 1 0.457. ≤ ≤ √2 − 4 ≃
Corollary 3.3 For f c (resp. f 6 Proof. Set Lf (x)= 4fx(1 x). In the intervals where exact convergence does − − 1 not take place, the iterates xn are confined by the two waveguides Lf ( 2 )= f and 2 1 2 2 − Lf ( 2 )= Lf ( f)=4f (1+f). Because 4c (1+c) = 1, it follows readily that xn > 1 is possible for− f>c. For f c, c x 1. ≤ − ≤ n ≤ Theorem 3.4 The context for the quaternion α + X, X H, stemming from the ∈ℑ computation of xn at f by (3.2) varies generically according to the table f 1 1 0 c 1 − − 4 2 @ H2 for x > 1 algebra G2 H @ n ℑ @ impossible for xn < 1 α α 1˜ 0 α real for x 0 ± n ≥ Proof. For f = 0, α2 = xf and X 2 = f(x 1) > 0 X H. 1) f [ 1, 1 [= 0 < x 1 = f α−2 < 0. The⇐⇒ context∈ℑ is G2, but for ∈ − − 4 ⇒ n ≤ ⇒ ≤ f = 1, x = 0 defines q = X H, X = 1. 2)−f [ 1 , 0] = x = α =0=∈ℑ q 2 = X 2 = f < 0. The context is H. ∈ − 4 ⇒ ⇒ − ℑ 3)f ]0, 1 ]= x = α =0= X 2 = f < 0: this is impossible. 4 2 ∈ 1 ⇒ α ⇒ 2 − 2 4)f ] 4 ,c[= xn = f < 1= X = α f < 0: impossible. ∈ 1 ⇒ ⇒ − 2 5)f [c, 2 ]= xn > 1 is possible by Corollary 3.3, hence X > 0. Therefore q H whenever∈ ⇒x > 1. In particular, if x = 1, α2 = f and q = √f is real. ∈ n n ± 3.3 X2 > 0: X is real We relax the condition X 2 > 0 which defines q as a quaternion. The case X 2 < 0 which makes no sense for X R3 can be reinterpreted as X2 > 0, hence X R : ∈ ∈ X2 = f α2 > 0 yields X , two real numbers in R, and the nature of the numbers α X −depends on α2±|(α2|< 0 is possible for x < 0). When both X and α are real,±| one| may consider that (i) they represent two different variables or (ii) they are different instances of the same unique variable. We complement Theorem 3.4 by the 1 Theorem 3.5 When X is real for 0 f 2 the evolution of α + X can take place in (i) R2, (ii) R or C. It is described≤ generically≤ by the table 7 f 0 1/4 1/2 α = 0 real if 0 x < 1 ⇒ • ≤ n algebraic 0 X 1/2 = R2 ≤|2 | ≤ ⇒ context = R (i) or R(ii) α i if xn < 0 ⇒ • ±| |= C ⇒ Proof. For 0 α i α | | | | X X | | √2f 0 √2f | | − f (X, α) R2 X = α = X + iα C ∈ | | | | q 2 ∈ x ]0, 1[ x =1/2 1/2 x< 0 ∈a) b) − ≤c) Figure 3.2: Reduction when X is real, 0 = x< 1 3.4 The alternative viewpoint of geometric algebra It is instructive to compare the interpretations of f = α2 X 2 either as q2 with − ℜ q = α + X in H (see above) or as a quadratic form in 4 real or complex variables. This latter approach, classically known as geometric algebra, is based on Clifford 8 algebras related to the signature (1, 3) of the quadratic form. Such algebras are associative and play an important role in theoretical physics (Baez 2001). They provide an alternative perspective (in mathematics and physics) which differs for 1 G2 f < 4 from the computational perspective given in Theorem 3.4, because is not associative.− We add that the interpretation shift from X H ( X 2 > 0,x> 1) to X R (X2 > 0,x< 1) which occurs for f in ]c, 1/2] and∈ℑ induces a reduction in the number∈ of dimensions is not a matter of consideration in the current presentation of Special Relativity. 4 The relative nature of the evolution of q = α+X sustained by (3.2) 4.1 The critical value x =1/2 When f < 0, the critical value 1/2 can occur for 1 f 1/2. − ≤ ≤ − f 1 Lemma 4.1 Corollary 4.2 The sedenion s = (g+,g ) in A4 is a zerodivisor for f 1/2 − ≤ − f 1 1 whenever X = α = − , X √ . q 2 2 ≤ ≤ 2 f Proof. x =1/2 X = α = −2 by Lemma 4.1. Then s has alternative ⇐⇒ q 2 2 orthogonal parts g+ and g in G with equal norm: X + α = f, hence − s 2 = 2f. − − Corollary 4.2 unravels an unsuspected link between the critical value x = 1/2 s and the construction of sedenions s in A4 such that s′ = √ 2f , s′ = 1 belong to − the zeroset Zer (A4) which is classically homeomorphic to Aut (G) (Theorem 4.7.1). This computational result is beyond the current analysis of relativistic evolution which is set in associative Clifford algebras. We believe that it plays the role of a meta-rule in the ubiquitous “self-organised” phenomena found in experimental 9 sciences. The conventional disregard for flexible (nonassociative) quadratic Dickson algebras may explain why the phenomenon has resisted mathematical analysis to the present day (Atlan 2011). √5 1 1 1 f Lemma 4.3 For − f , x = X = α = . 4 ≤ ≤ 2 2 ⇐⇒ | | | | q 2 Proof. x = 1 α2 = f = X2 X = α = f with α and X real. See 2 ⇐⇒ 2 ⇐⇒ | | | | 2 Theorem 3.5 and Figure 3.2.a) (i) and b) (ii). q 4.2 A review of the three exceptional values f 1, 1/2, 1/2 in (3.2) ∈ {− − } At the endpoints f = 1/2 and f = 1, the dynamics of (3.2) is complex, ruled by z3 = 1 and z6 = 1 respectively, and− it is real for f = 1/2. At these three values − − (3.1) can be transformed into the equation t(u)= εt(2u), with ε = 1, where t is a real or complex exponential ± f 1 1 1 − − 2 2 t sin2 exp cos ε +1 +1 1 − Moreover, the iteration (3.2) can be solved in closed form as a function of n and x [0, 1]. The formulae are listed below. 0 ∈ f iterations 2 π 1 2x = cos2ψ, x = sin ψ, ψ [0, 2 ] − n 1 ∈ 1 1 2xn = cos(2 cos− (1 2x0)), n 0 − − n − 1 ≥ 2 or ψn =2 ψ0, ψ0 = sin− √x0, xn = sin ψn. 1 π 2 x = cos θ = cos( 3 ξ), θ [0, π] 1 − π −n π ∈ 1 1/2 2 xn = cos 3 ( 2) 3 cos− (1/2 x0) , n 0 − n − − π − − ≥ or ξn =( 2) ξ0, ξ0 = θ0 − 3 − 1 2x = ey, y< 0 | − | 2n 1/2 1 2xn = (1 2x0) , x0 ]0, 1[, x0 =1/2, n 0 − or− y =2ny−, y = ln 1 ∈2x < 0 ≥ n 0 0 | − 0| 10 For f = 1, we use the definition of the Tchebychef polynomial of degree k 1: − ≥ T (d) = cos(kArcos d), d 1 to write, with d = 1 2x, d = T n (d ), n 0. We k | | ≤ − n 2 0 ≥ recall that the quantity min(max t 1 p(t) , p is a monic polynomial of degree k) is k 1 | |≤ | | achieved for p(t)= Tk(t)/2 − . With obvious changes of variable the dynamics of (3.2) takes the three remark- ably simple forms for n 0: ≥ f iteration 1 ψ =2ψ mod π − n+1 n 2 1 ξ = 2ξ mod π 2 n+1 − n 1 y =2y in R− − 2 n+1 n Some remarks are in order. For f = 1, the angle ψ refers to the real axis R; π − 1 π it is uniformly distributed on [0, 2 ]. For f = 2 , the angle ξ = 3 θ refers to the rotated axis eiπ/3R in C; ξ and θ run in opposite directions. − We consider the change of notation r = 4f, δ = 1 x = d . Then the iteration − 2 − 2 (3.2) takes for r 4, 2 one of the three recursive forms listed in Table 4.1. In ∈ { ± } 1 each case h is a continuous bijective function and h− is the inverse function. r = 4f recursion, n 0 h − ≥ d = h r h 1(d ) h(2ψ) = cos2ψ = d 4 n+1 2 − n d 1 0 x 1 2ψ [0, π] | | ≤ ⇐⇒ ≤1 ≤ ∈ y d = h (rh− ( d ) h(y)= e = d 2 | n+1| | n| | | d < 1 0