ITP-UU-12/25

SPIN-12/23
arXiv:1207.3612v2 [hep-th] 15 Sep 2012

Discreteness and Determinism in Superstrings

Gerard ’t Hooft

Institute for Theoretical Physics

Utrecht University
and
Spinoza Institute
Postbox 80.195
3508 TD Utrecht, the Netherlands
e-mail: [email protected]
internet: http://www.phys.uu.nl/~thooft/

Abstract
Ideas presented in two earlier papers are applied to string theory. It had
been found that a deterministic cellular automaton in one space- and one
time dimension can be mapped onto a bosonic quantum field theory on a 1+1
dimensional lattice. We now also show that a cellular automaton in 1+1 di-
mensions that processes only ones and zeros, can be mapped onto a fermionic
quantum field theory in a similar way. The natural system to apply all of
this to is superstring theory, and we find that all classical states of a classical,
deterministic string propagating in a rectangular, D dimensional space-time
lattice, with some boolean variables on it, can be mapped onto the elements
of a specially chosen basis for a (quantized) D dimensional superstring. This
string is moderated (“regularized”) by a 1+1 dimensional lattice on its world
sheet, which may subsequently be sent to the continuum limit. The space-
time lattice in target space is not sent to the continuum, while this does not
seem to reduce its physically desirable features, including Lorentz invariance.
We claim that our observations add a new twist to discussions concerning
the interpretation of quantum mechanics, which we call the cellular automa-
ton (CA) interpretation. Detailed discussions of this interpretation, and in
particular its relation to the Bell inequalities, are now included.

July 15 and September 15, 2012

1
1. Notation

Let us briefly recapitulate our special notation: variables described by

capital Latin letters, N, P, Q, X, · · · , denote integer valued fields, (1.1)

lower case Latin letters, p, q, · · · , indicate real numbers, and (1.2)
lower case Greek letters, α, η, ξ, · · · , are numbers defined modulo 1, (1.3)

the latter being usually constrained to the interval [ − 21 , 12 ] . There will be a few ex-
ceptions, such as the lattice coordinates x, t on the world sheet, which are in lower case
because they are not used as field variables, and the Greek letter σ and the three Pauli
matrices σ1 , σ2 , σ3 , which have the eigenvalues ±1 .
Frequent use will be made of the number

ǫ ≡ e2π ≈ 535.49 · · · , so that e2πiα ≡ ǫiα , and ǫiZ = 1 if Z ∈ Z . (1.4)

If we have integers Q, P, · · · , we will often associate a Hilbert space of states to these:

| Q, P, · · ·i . Then, there will be operators ηQ , ηP , · · · , defined by

ǫiNηQ | Q, P, · · ·i = | Q + N, P, · · ·i ; ǫiNηP | Q, P, · · ·i = | Q, P + N, · · ·i . (1.5)

These operators play the role of “position” operators for the discrete “momentum” vari-
ables Q and P . In general, they will have eigenvalues restricted to be in the interval
(− 12 , 12 ] .
Hermitean operators will be called real or integer if all their eigenvalues are real or
integer. Because we wish to keep factors 2π in the exponents (to be absorbed when we
use ǫ instead of e ), our commutation rules will be

[q, p] = i/2π , (1.6)

and the quantum evolution equation for an operator O is

d
dt
O = −2πi[O, H] , (1.7)

all of which means that it is h rather than ~ that we normalize to one. Many of the
above quantities can be taken to be functions of the world sheet space-time coordinates
(x, t) or the world sheet lattice sites (x, t) .

2. Introduction

Quantum mechanics has become a phenomenally powerful and successful doctrine for the
description of all physical properties of tiny objects, whenever their sizes and mass or
energy scales are near or below those of a large molecule. It appears to be universally
accepted that quantum mechanics requires logical reasoning that at least deviates from

2
standard human experience when we think of soccer balls, automobiles, stars, planets
and live creatures such as cats. Attempting to “explain” quantum logic often drives
researchers into ways of reasoning that transcend ordinary logic even further. The idea
that quantum phenomena might be explained in terms of totally classical, down-to-earth,
underlying theories, where Hilbert spaces, tunneling, interference, and so on, play no
role whatsoever, is often categorically dismissed. “Classical thinking” belongs to the
nineteenth century when people did not know any better.
There seem to be sound reasons for this attitude. Bell’s inequalities[1] and similar
observations[2][3] are applied with mathematical rigor to prove that quantum mechanics
will be the backbone of all theories for sub-atomic particles of the future, unless, as
some string theorists have repeatedly stressed, “an even stranger form of logic might be
needed”[4].
The author of this paper takes a minority’s point of view[5][6], which is that, in order
to make further progress in our understanding of Nature, classical logic will have to be
restored, while at the same time our mathematical skills will need to be further improved.
We have reasons to believe that the mathematics of ordinary statistics can be rephrased in
a quantum mechanical language and notation; indeed this can be done in a quite natural
manner, such that one can understand where quantum phenomena come from.
Our starting point is that there are underlying theories where the physical degrees of
freedom are described as discrete bits of data. Some kind of time variable can be used
in terms of which these data evolve. The evolution resembles a computer program, and
usually we will assume that this program processes the data locally. In computer science
such a system is known as a “cellular automaton”[7].
On the one hand, we emphasize that there is no need for the reader to accept our
general philosophy or interpretation. This paper was intended just to report about some
simple mathematical transformations. On the other hand, however, we experienced such
a fierce opposition in discussions with colleagues, that, in this new version of this pa-
per, much more extensive discussions of our interpretation, “hidden variables”, and our
axiomatic formalism are added, see Sections 7—9. A reader who has already tears of
disbelief in his/her eyes, should first consult those Sections, in particular Section 8, where
also Bell’s theorem is adressed. For the others, let me briefly summarize.
Most, if not all, treatises on “hidden variables” begin with formulating what is consid-
ered to be utterly reasonable assumptions as to what such variables should do: somewhere
in their equations, they should, ‘of course’, not be quantum mechanical, but they should
‘mimic’ quantum behavior. For instance, if in genuine quantum mechanics two kinds of
measurements cannot be made at the same time, the hidden variables should be able to
do this anyway.
The authors then continue, to demonstrate with beautiful mathematical rigor, that
such variables indeed disagree with quantum mechanical predictions.
These assumptions do not apply to this work. The variables in this paper disagree
nowhere with quantum mechanics. They happen to be classical and quantum mechanical
at the same time. There is a simple mathematical transformation from one picture to

3
the other, and back. Therefore, if quantum mechanics forbids two measurements to be
done at the same time, then this theory forbids this as well. The beautiful mathematical
no go theorems do not apply here. Their assumptions, as usually formulated on page 1,
line 1, are invalid. Yet the phrase ‘hidden variables’, to some extent, does not seem to
be inappropriate here. In some extremely important ways, however, the theory adopted
here differs from the usual ‘hidden variable’ scenario. The escape from Bell’s inequalities
occurs through the route of space-like vacuum correlations, and the argument is quite
delicate. A reader who is unable to accept all this, please first continue with Section 7.
Now please allow me to explain how the ideas were conceived.
Motivated by the conviction that this is the only way to go, this work is a continuation
of a new series[8][9] where our mathematical doctrine is developed. What we wished
to demonstrate is that a theory in which states evolve as in a cellular automaton can
turn into a quantum field theory, by doing nothing more than a couple of mathematical
transformations. While a cellular automaton undergoes its (classical) evolution processes,
its observables can be cast into a Hilbert space of states, where operators can be defined
just as in quantum mechanics. These operators then obey a Schrödinger equation. This
equation itself is easy to derive and quite obvious, but the ‘quantum miracles’ it leads to
are due to extremely subtle features of the vacuum state.
So we easily arrive at the Schrödinger equation, which is linear, and it acts in a genuine
Hilbert space. This linearity invites us to consider superpositions of states, and to pass
on to a new basis of elementary states, in which a wave function evolves. What happens
in terms of the original automaton is in principle very simple. The wave function that
we use is a superposition of cellular automaton states, and this means that we have a
probabilistic distribution there. At the level of this automaton, the phases of the wave
functions mean nothing at all, and this has no consequences either, since, in terms of
the automaton states, no interference takes place. But by the time we have become
accustomed to the basis elements of the quantum field theory, these seem to evolve in a
much more contrived way, so that interference phenomena do seem to be important. One
of the reasons why we experience our world as if it is quantum mechanical, is that we have
not (yet) been able to identify the “ontological” basis of the cellular automaton states.
In the terminology used in Ref. [10], our theory is ψ -epistemic rather than ψ -ontic.
The ontological variables, or “hidden” variables λ discussed there, in our theory only
refer to the basis elements of a quantum theory, of only one very specially chosen basis.
The author’s opinion is that the findings reported here shine a new light on the ques-
tion of interpretation of quantum mechanics, but before going into any discussion about
ontological or epistemic states, EPR experiments or Bell’s inequalities, we must display
as clearly as we can the technical calculations performed here. Extensive discussions are
added at the end. The reader should then be able to judge for him- or herself whether
this affects his or her view on quantum mechanics.
The idea is simple, and now we claim that it works. Indeed, it leads to something of a
surprise. Let me disclose the surprise right away. We found a simple cellular automaton
that describes strings evolving classically on a D dimensional space-time lattice. The

4
fact that the lattice is discrete implies that, locally, bits and bytes of information are
processed. Putting this string on a world sheet that is also a lattice (in one space- and
one time dimension), we get an ordinary, classical field theory of integers on this lattice.
The strings may also carry Boolean degrees of freedom (fields taking the values ±1 only)
on its lattice elements. The system is integrable classically, and we find streams of left-
going and right-going integers; this is the cellular automaton. It could be seen as the
simplest kind of string theory one can imagine; at this level, there is no word about
quantization, let alone bosons, fermions or supersymmetry.
But then, the procedure described in Ref[9] allows us to transform this simple au-
tomaton into a quantum field theory of bosons and fermions. There are left-movers and
right-movers, and there is a lattice cut-off. The cut-off does not affect the particle dis-
persion law: all modes with momentum below the Brillouin zone move exactly with the
(worldsheet) speed of light. There is no direct interaction yet. We did not (yet) consider
boundary conditions, so the string has infinite length. Thus, apart from the lattice cut-off
in the world sheet, this is a quantum string. After the transformation described in Ref. [9],
the space-time lattice disappears and now seems to look like a continuum. Assuming the
usual gauge fixing on the world sheet, we are left with D − 2 degrees of freedom, the
transverse coordinates.
In a similar fashion, to be described in Section 5, the Boolean degrees of freedom
can be transformed into fermionic fields, and thus we arrive at bosons and fermions.
Indeed, if we take D − 2 of these fields, they will combine with the bosons to form
N = 1 super multiplets, just as in the superstring. This way, we discover that our
simple discrete cellular string system turns into a superstring moving around in a space-
time continuum instead of a lattice. Physically, our discrete, deterministic lattice theory
is identical to the continuous, quantum mechanical (super)string theory; one is just a
mathematical transformation of the other. The identification however, is mathematically
rather contrived; we postpone further discussions to later sections.
√The original target
′
space-time turns into a lattice with fixed lattice length, aℓ =
′
2π α , where α is the string slope parameter. The target space therefore cannot be
sent to the continuum limit; yet it is mathematically transformed to become a continuous
space, in fact, a superspace.
In superstring theory, it is well-known that such a theory does not automatically
reproduce Lorentz invariance in D space-time dimensions. For this, we need to have
exactly 8 transverse dimensions, that is, a D = 10 dimensional space-time.
As for the lattice in the world sheet, the situation is more subtle. In principle, our
intention was to keep it discrete also, since the target space lattice (the lattice in space-
time) induces a natural looking discrete lattice on the world sheet. Nevertheless, in a
more advanced treatment of this theory, the world sheet lattice will probably have to be
sent to the continuum after all. This may be needed in order to get a good, Lorentz
invariant theory. To construct Lorentz boost operators, the longitudinal variables are
needed, and the algebra of the Lorentz group should harmonize with the algebra of these
constraints. This algebra requires world sheet reparametrization invariance, and this can
be treated more easily if we first pass to the continuum limit, as in the conventional

5
theory. Physically, going to the continuum limit on the world sheet does not modify the
theory; mathematically, it is a gauge transformation, using the large amount of freedom
we have in choosing the conformal world sheet coordinate transformations.
At first reading of this paper, one could decide first to read the discussions at the
end, Sections 7 —10, to see where we intend to end up. Not all problems have been
solved, because some of the math tends to become complicated. Boundary conditions,
needed to describe open and closed strings of finite length, and GSO projections, needed
to eliminate tachyon states, are hardly toughed upon.
While reading the rest of the paper, one is advised also to inspect the two previous
papers that are often referred to[8][9]. We think the implications of this work are exciting
and worth further intensive study.

3. Bosonic String Theory on the lattice

Consider a rectangular lattice in both space and time. As already emphasized, we do

not plan to send the physical value aℓ of the lattice link size to zero at any stage of the
theory. Indeed, it will be fixed by the string slope parameter α′ , whose value is not yet
known. We will normalize aℓ to one. Thus, in our space-time units, the D − 1 space-like
lattice coordinates X i and one time-like lattice variable X 0 will all take integer values.
We never Wick rotate to Euclidean space-time.
On our discrete space-time, we describe a propagating string as follows. Besides the
space-time lattice, take also a discrete, 1+1 dimensional world sheet lattice described by
integer coordinates1 (x, t) , and integer functions X µ on this lattice. We could limit
ourselves to the sites where x + t are even. The initial state is defined at the world
sheet time values t = 0 and t = 1 only: X µ (x, 0) and X µ (x, 1) . “String bits” AµL now
connect every point (x, t) with (x + 1, t + 1) while AµR connect the points (x, t) with
(x − 1, t + 1) . Thus, we have

X µ (x + 1, t + 1) − X µ (x, t) = AµL (x + 21 , t + 12 ) , (3.1)

X µ (x − 1, t + 1) − X µ (x, t) = AµR (x − 21 , t + 12 ) , (3.2)

Thus, apart from the center of mass, the entire set of initial values of the string is deter-
mined by the set of numbers AµL (x, 21 ) and AµR (x, 12 ) at all x .
These vectors will be chosen to be composed of integers, and strictly light like (later,
we may replace this constraint by a different one):

(AµL (x, 12 ))2 = 0 , (AµR (x, 21 ))2 = 0 , (3.3)

A0L (x, 21 ) ≥ 0 , A0R (x, 21 ) ≥ 0 . (3.4)

This ensures that, at t = 0 as well as at t = 1 , the initial data X µ (x, t) are not timelike
separated from both neighbors X µ (x ± 2, t) .
1
In most treatises on string theory, these are called σ and τ . But for notational consistency in this
paper, we prefer to use x and t .

6
 We now postulate the equations of motion:

X µ (x, t) = X µ (x − 1, t − 1) + X µ (x + 1, t − 1) − X µ (x, t − 2) . (3.5)

One readily derives that AµL are left-movers and AµR are right-movers:

AµL (x, t) = AµL (x + t) , AµR (x, t) = AµR (x − t) , (3.6)

and therefore, the constraints (3.3) and (3.4) hold at all times t .
The careful reader may recognize these lattice equations as a discrete version of the
bosonic relativistic string equations[11][12],

(∂x2 − ∂t2 )X µ = 0 ; (∂x XLµ (x + t))2 = 0 , (∂x XRµ (x − t))2 = 0 , (3.7)

which, in the continuum limit aℓ → 0 , would yield the conventional, classical theory.
In this paper, however, we will show that aℓ is linked to the square root of the slope
parameter α′ of the string, which we will not send to zero. Furthermore, in spite of the
appearances from Eq. (3.5), our theory will show to be a quantum theory, not a classical
one (Note, that Eqs. (3.7) hold for the quantum theory as well). The lattice artifacts will
not be observable, even at the Planck scale.
A very important property of the classical equations (3.1)—(3.6) is that they are
invariant under the discrete Lorentz group O(D − 1, 1, Z) , that is, the Lorentz group
restricted to integers in its operator matrices. At first sight, these bosonic strings seem to
be non-interacting, but one interaction mode, also invariant under O(D − 1, 1, Z) , can
be introduced: an exchange operation: if two strings hit the same space-time point X µ ,
their arms are exchanged. This may generate closed strings. The author has long been
searching for classical lattice theories, with interactions, invariant under O(D − 1, 1, Z) .
Here, finally, we have one. Note that, our theory requires also the interactions to be
classical, and this requires the exchange interaction to be unambiguous; consequently, the
bosonic string described here has to be an oriented string. And, its string interaction
constant, gs , cannot be tuned to any value; it is basically equal to one.

4. Bosonic string theory in continuous space-time

As is standard in string theories, we see that A0L,R fixes the physical time coordinates, and
Eqs. (3.3) remove one more spacial degree of freedom, so that, effectively, there are only
D − 2 free parameters left at every lattice site (x, t) . These obey the linear equations
(3.5). While the X a (x, t), a = 1, · · · , D − 2, are independent integers, we have exactly
the discrete cellular automaton described in our previous paper [9], in (D − 2) -fold. It
is described there how this system is equivalent to a quantum theory of D − 2 quantum
fields q a (x, t) , and associated field momentum variables pa (x, t) . Both q a and pa take
real values, while still living on a world sheet lattice. They obey the quantization rules:

[q a (x, t), pb (x′ , t)] = i ab

2π
δ δx,x′ . (4.1)

7
We can summarize the result of Ref. [9] as follows. The equivalence transformation starts
with the left- and right-movers, AaL,R (x) . The commutation rules for the quantum oper-
ators aaL,R (x) , associated to the classical, discrete functions AaL,R (x) , must be2

[aaL , abR ] = 0 ; [aaL (x), abL (y)] = ± 2πi δ ab if y = x ± 1 ; else 0 ; (4.2)

[aaR (x), abR (y)] = ∓ 2πi δ ab if y = x ± 1 ; else 0 . (4.3)

Let us concentrate on the left-movers, and drop the subscript L and the superscript
a . They only live on the odd sites x . Since the variables A(x) are integers, we regard
them as discrete “momentum” operators associated to “position” operators ηA (x) . Just
as in Section 1, we demand the operators ηA (x) to be periodic with period 1, so that
these variables at all points x together generate a torus with ℓ periodic dimensions, if ℓ
is the total length of a string. As explained in Ref. [9]. We could define the continuous
quantum operators a(x) as
?
a(x) = A(x) − ηA (x − 1) , (4.4)

which would obey the correct commutation rules, except where ηA (x) = ± 21 , because the
jump from + 12 to − 12 is not accounted for. A better definition is
∂ ∂
a(x) = − 2πi + ϕ({ηA (x)}) − ηA (x − 1) , (4.5)
∂ηA (x) ∂ηA (x)
where the phase function ϕ({η(x)}) is a carefully chosen functional[9], in such a way
that a(x) depends continuously on η(x) and η(x − 1) at the cross-over from 21 to − 12 ,
although this does leave an inevitable singularity when both η(x) and η(x − 1) are ± 21 .
Thus, singularities are generated at the boundary points of two successive η ’s, the
points where ηA (x) = ± 21 , ηA (x − 1) = ± 21 . At these points, wave functions must vanish
in order to keep the commutation rules (4.1) valid3 .
Let us put the indices L, R , and a back. Then, the quantum operators q a (x, t) and
pa (x, t) are defined by the equations

aaL (x + t) = pa (x, t) + q(x, t) − q a (x − 1, t) ; (4.6)

aaR (x − t) = pa (x, t) + q(x, t) − q a (x + 1, t) , (4.7)

which can easily be solved by Fourier transformation in the x coordinate.

We note that the quantum variables q a (x, t) are the transverse string variables usually
called X a (σ, τ ) in string theory, pa (x, t) = ∂τ X a (σ, τ ) , and they obey the usual commu-
tation rules. The only thing unusual to them is that the world sheet is still a lattice[13].
2
with apologies for using the letter a both for counting the D − 2 transverse degrees of freedom and
for the real valued left and right moving fields. Confusion is hardly possible.
3
In Ref. [9], the suspicion was expressed that this constraint could be removed by introducing fermionic
degrees of freedom. We refrain from continuing along that route, however, since, although we did find
that fermions may remove this constraint, they in return then deliver even more troublesome anomalies
in the fermionic (anti)commutation rules. Fermions will be introduced in the next section.

8
However, as was also emphasized in Ref. [9], the hamiltonian at world sheet momentum
values κ with |κ| < 21 , is identical to the continuum hamiltonian, so the only way in
which the world sheet lattice manifests itself here is in the presence of a Brillouin limit to
the momentum, while at values of κ within the Brillouin zone, the continuum theory is
exactly reproduced.
How much does the theory obtained here actually differ from the strictly continuous
string theory? As long as we concentrate on the transverse modes only, there is reason to
suspect that there is no difference at all. To describe highly excited strings, the high κ
values may not be needed at all; we just go to longer strings, an observation already made
by Klebanov and Susskind in Ref. [13]. To be precise, if we limit ourselves momentarily
to the description of the bulk string, we can always use local conformal transformations
on the world sheet to ensure that the values of the left- and right movers aµL,R stay within
bounds and are slowly varying.
We do note that, the constraints (3.3) and (3.4) do not quite coincide with the con-
straints usually applied in the quantum theory, since the latter are defined in terms of
the complete operators a(x) , not only their integral parts. The consequences of this for
Lorentz invariance and the usual string anomalies are yet to be investigated (see also
Subsections 10.2 and 10.3).

5. Fermions

Now consider the same world sheet lattice (x, t) , but now with Boolean degrees of freedom
on them: σ(x, t) = ±1 . Let the classical equation of motion be similar to (3.5):
σ(x, t) = σ(x − 1, t − 1) σ(x + 1, t − 1) σ(x, t − 2) . (5.1)
Again, we have left movers and right movers:
σ(x + 1, t + 1) σ(x, t) = σL (x + t + 1) , (5.2)
σ(x − 1, t + 1) σ(x, t) = σR (x − t − 1) . (5.3)
1 0
 
Identifying these operators with the Pauli σ 3 operators , we also introduce the
0 −1
operators
0 1 0 −i
   
1 2
σ = , and σ = , (5.4)
1 0 i 0
i
so that we have σL,R (x) , i = 1, 2, 3 , obeying

[σLi (x), σLj (x′ )] = 2iδx,x′ εijk σLk (x) , [σLi (x), σRj (x′ )] = 0 , (5.5)
 
and {σ i (x), σ j (x)} ≡ 21 σ i (x) σ j (x) + σ j (x) σ i (x) = δ ij . (5.6)

In 1+1 dimensions, we wish to use fermionic fields ψA (x), A = 1, 2 , obeying

ψA† (x) = ψA (x) , {ψA (x), ψB (x′ )} = δx,x′ δAB . (5.7)

9
 i
They can be obtained from the operators σL,R (x) for instance as follows:
Y Y Y
ψ1 (x) = ψL (x) = σL1 (x) σL3 (y) , ψ2 (x) = ψR (x) = σR1 (x) σR3 (y) σL3 (z) . (5.8)
y<x y<x z

This is the so-called Jordan-Wigner transformation[15]. It is non-local only if we take

products of an odd number of ψL or ψR operators. In practice, this is never needed.
Consider now the Fourier transforms of these fermionic fields:
Z 1
X 2
ψ̂A (κ) = ψA (x) ǫ−iκx
, ψA (x) = dκ ψ̂A (κ)ǫiκx , (5.9)
1
x −
2

ψ̂A† (κ) = ψ̂A (−κ) , {ψ̂A (κ), ψ̂B (κ′ )} = δ(κ + κ′ ) , (5.10)
and introduce the lattice hamiltonian operator4
Z 1
2 3
1
H=2 dκ κ ψ̂A (−κ)(−σAB )ψ̂B (κ) ; (5.11)
0

this gives
3
[H, ψ̂A (κ)] = −σAB κ ψ̂B (κ) . (5.12)
According to the Schrödinger equation (1.7), the time dependence is then
3 κt
ψ̂(κ, t) = e2πiσ ψ̂(κ, 0) , (5.13)
so that
ψ1 (x, t) = ǫiκ(x+t) ψ̂1 (κ) = ψL (x + t) , (5.14)
ψ2 (x, t) = ǫiκ(x−t) ψ̂2 (κ) = ψR (x − t) . (5.15)

Thus we see that the quantum hamiltonian (5.11) produces the correct time depen-
dence of all fermionic operators of the automaton. We now claim that it also propagates
all individual states correctly. The argument goes as follows.
Consider first the state with all “spins” up: σL3 (x, 0) = σR3 (x, 0) = +1 . One quickly
finds that, for this state, the hamiltonian vanishes. This is because, for this state, it can
only use operators of the form σA1 (x) σA1 (x + N) , but for both A = 1 and A = 2 , the
hamiltonian is antisymmetric under left-right interchange, so that the coefficient for all
these terms must cancel out.
Next, we find that, acting on this state with an arbitrary number of ψ operators,
gives all other states with left- and right movers. These now propagate the same way
both in the classical and in the quantum mechanical description.
It is a special feature of the 1+1 dimensional case that a massless fermionic lattice field
theory maps onto a simple cellular automaton. In higher dimensions, this is much more
difficult to achieve, because we cannot make use of left- and right movers. In Ref. [9], we
found the same result for the bosonic case.
4 1
The factor 2 arises from the normalization chosen in Eq, (5.6).

10
6. The superstring and its constraints

The locally independent degrees of freedom of the superstring are the bosonic coordinates
q a (x, t) and the fermionic fields ψAa (x, t) , with a = 1, · · · , D − 2 . We have seen that
these map onto the discrete integers X a (x, t) and the Boolean variables σ a (x, t) . It is
important to note, at this stage, that also the equations of motion exactly correspond to
those of the superstring: all excitations move with the local speed of light to the left or
to the right along the string world sheet.
However, the superstring also has two longitudinal coordinates q D−1 and q 0 , as well
as two longitudinal fermionic fields, ψAD−1 and ψA0 . These are not independent of the
other degrees of freedom, but determined by them via two kinds of equations: gauge fixing
equations and constraints. Even though the original string theory had a continuous world
sheet while our world sheet is a lattice, we can make a rigorous identification: the Fourier
modes of the left- and right movers, aµL,R and ψAµ , must correspond exactly with the
Fourier modes of the left- and right-movers of the continuum theory, as long as |κ| < 12 .
Higher Fourier modes of the continuum theory are set to zero. If we allow the length of
the string in its world sheet to be arbitrary, one can still obtain all string excited modes,
so that this Brillouin limitation is a minor one, in practice.
For the continuum theory, we now write the world sheet gauge fixing conditions as

a+
L = 1 , a+
R = 1 , ψ1+ = ψ2+ = 0 , (6.1)

where a± = a0 ± aD−1 , ψ ± = ψ 0 ± ψ D−1 , and the superstring constraints are:

(aµL )2 + 21 ψ1µ κ̂ ψ1µ = 0 , (aµR )2 + 21 ψ2µ κ̂ ψ2µ = 0 ; (6.2)

aµL ψ1µ = 0, aµR ψ2µ = 0, (6.3)

where κ̂ is the operator that multiplies the Fourier components ψ̂(κ) with κ . The latter
two equations now fix the Lorentz light cone minus components of these variables, since
the bosonic constraint (6.2) gives
D−2
X
a−
L = ((aaL )2 + 12 ψ1a κ̂ ψ1a ) , (6.4)
a=1

while the fermionic constraint (6.3) gives

D−2
X
ψ1− = 2 ψ1a aaL , (6.5)
a=1

and the same with aR and ψ2 .

The way it works in the superstring is that these constraints, as well as the longitudinal
degrees of freedom that are determined by them, could be completely ignored as long as we
are only interested in the string equations of motion. Whether this includes the superstring
interactions requires further research but in principle this should be straightforward. The

11
importance of the constraints lies entirely in the fact that they are needed to investigate
Lorentz invariance. The generators of the Lorentz transformations are determined by all
D components of the dynamic variables.
Therefore, in our mapping between discrete theories and the superstring, we may also
ignore the longitudinal degrees of freedom and the constraint equations obeyed by them,
except when we wish to study Lorentz invariance.
In our lattice theory, we originally expected invariance under discrete Lorentz trans-
formations O(D − 1, 1, Z) by adding discrete longitudinal components to the left- and
right movers AaL,R (x, t) . However, what is really needed is only the gauge fixing and the
constraints for the continuum case, because that is the system for which Lorentz invari-
ance is required. The quantum constraint equations should obey the same algebra as in
the superstring with a continuous world sheet. Details of this problem will be left for
further investigation.
Our conclusion of this section is that the theory has to be viewed as follows:

* We start with a classical (i.e. non quantum mechanical) dynamical theory that
describes D − 2 degrees of freedom Aa L,R (a = 1, · · · , D − 2) as integer-valued left-
moving and right-moving fields on a 1+1 dimensional world sheet. These degrees
of freedom act as discrete
√ coordinates on a transverse space-like lattice. The lattice
length is fixed: aℓ = 2π α′ . The theory is invariant only under discrete SO(D −
2, Z) rotations and discrete translations.

* By adding the canonical associated operators ηAL,R we can use these to fill the
spaces between the lattice sites so that a space-like D − 2 dimensional continuum
is obtained. Now, however, the fields do not commute, so we obtain a quantum
version of this system: a quantized string theory. The invariance group turns into
that of the transverse continuous rotations and translations.

* A similar procedure starting with Boolean degrees of freedom gives us fermonic

fields. Thus, we find our mapping connecting a deterministic automaton processing
integers and Boolean variables to a (quantized) superstring theory.

* Only in the quantum theory, we can now follow standard procedures to use gauge
conditions (6.1) and constraints (6.2) and (6.3) which define for us the longitudinal
space-time coordinates.

* With these longitudinal coordinates added, we use standard arguments to prove

invariance under the full Lorentz group O(D −1, 1, R) , but only if D = 10 . Clearly,
only the quantum system possesses this manifest Lorentz invariance. In the classical
automaton, this invariance is not manifest at all.

12
7. The CA interpretation of quantum mechanics

7.1. Short History

Intuitively simple and natural interpretations of quantum mechanics could be extremely

helpful for making further progress in understanding quantum gravity, string theory and
quantum cosmology. The “many world” interpretation, or the Bohm - deBroglie “pilot
waves”, and even the idea that the quantum world must be non-local, are difficult to
incorporate in models of the universe. Of course, it was attempted to adopt these view
points in numerous approaches, but the resulting picture is often opaque and not in line
with intuition. By itself, such impressions do not disqualify these interpretations, which
indeed have become fairly standard, but it seems very much worth-while to continue
searching for something better.
We should plainly ask what is really going on in a world described by quantum me-
chanics, and in order to try to get some ideas, this author constructed some models with
various degrees of sophistication[5]. These models are of course ”wrong” in the sense that
they do not describe the real world. They do not generate the Standard Model, but one
can imagine starting from such simple models and adding more and more complicated
details to make them look more realistic, in various stages.
Of course, the difficulties that arise when one tries to underpin quantum mechanics
with determinism, are well-known. Simple probabilistic theories fail in an essential way.
One or several of the usual assumptions made in deterministic theories will probably have
to be abandoned. Of course, we are fully aware of that. On the other hand, our world
seems to be extremely logical and natural.
Therefore, we decided to start an investigation at the other end. Make assumptions
that later surely will have to be amended; make some simple models, compare these with
what we know about the real world, and then modify the assumptions any way we like.
The no-go theorems tell us that a simple cellular automaton model is not likely to
work. One way to try to “amend” them, was to introduce information loss. At first sight
this would carry us even further away from quantum mechanics, but if one looks a little
more closely, one finds that one still can introduce a Hilbert space, although it becomes
much smaller and it may become holographic, but this is something we may actually
want. If one then realizes that information loss makes any mapping from a deterministic
model to quantum mechanical states fundamentally non-local - while the physics itself
stays local - then maybe the idea becomes more attractive.
Now the problem with this is that again one makes too big assumptions, and the
math is quite complicated and unattractive. So, reluctantly, we returned to a reversible,
local, deterministic automaton and asked: to what extent does this resemble quantum
mechanics, and where does it go wrong? The idea we had in mind was that we will alter
the assumptions, maybe add information loss, put in an expanding universe, but all that
comes later; first let us try to establish what goes wrong.
And here is the surprise: in a sense, nothing goes wrong. All one has to assume is

13
that quantum states can be used, even if the evolution laws themselves are deterministic.
Thus, the probability distributions are given by quantum amplitudes. The point is that,
when describing a mapping between the deterministic system and the quantum system,
one has a lot of freedom. If one looks at any one periodic mode of a deterministic system,
one can define a common contribution to the energy for all states in this mode, and this
observation introduces a large number of arbitrary conserved constants, so we are given
much freedom.
Using this freedom, we end up with quite a few models that just look interesting.
Starting with deterministic systems, we end up with quantum systems. These are real
quantum systems, not any kind of ugly concoctions. On the other hand, they are still
a far cry from the Standard Model, or even anything else that shows decent, interacting
particles.
The exception now is string theory. This is puzzling. Is the model we constructed in
this paper a counter example, showing that, what everyone tells us about fundamental
quantum mechanics being incompatible with determinism, is simply wrong? This, we do
not believe; we still believe that, somewhere, we will have to modify our assumptions, but
maybe the usual assumptions made in the no-go theorems, will have to be looked at as
well.
It does seem that people are too quick in rejecting “superdeterminism”, and maybe also
in denouncing “conspiracy”. Superdeterminism simply implies that one cannot “change
one’s mind” (about which component of a spin to measure), by “free will”, without also
having a modification of the deterministic modes of our world in the distant past. This
is obviously true in a deterministic world, but we do observe that the free will postulate
can be replaced by another one: the freedom to choose the initial state. This condition
is more precise, and we will discover that this freedom actually has to be constrained, in
order to bypass Bell’s inequality.
One has to keep in mind that what looks like ‘conspiracy’ to one observer, might
actually be a quite natural phenomenon to another. We will find that this consideration
has to be taken seriously. We took the attitude that questions of this sort can also
be settled if we can construct explicit mathematical models that can account for the
complex structure of our world. If the usual no-go theorems are insurmountable, we
should automatically be confronted with failures of our mathematical models. Curiously,
this does not seem to be happening. We are zooming in on models that become more and
more realistic.
Let us now imagine that a really interesting quantum field theory of the world can be
found that allows a mapping onto a cellular automaton (CA). Then, we would have new
support for the CA interpretation of quantum mechanics. It could still be that the real
world would require further modifications. We will be ready for that, but first allow us to
go ahead with this interpretation. As stated, it was greeted with considerable skepticism.
Well, these skeptics would not have expected the mathematical model described in this
paper, so perhaps they will react the way they should: here we have a paradox, an
interesting one, and let us see what we can conclude from it.

14
 7.2. The Cellular Automaton (CA)

The CA interpretation is embarrassingly simple. An automaton starting at time t1 in

a state Q ~ 1 , evolves and ends up in a state Q
~ 2 at t2 . The evolution is deterministic.
The simplest possibility would be that we have cells in 3-space, numbered by an index
i , containing data in the form of numbers Qi . At the beat of an external clock, the
number in each cell i is replaced, according to an algorithm that would typically look at
the number Qi itself and the numbers Qj in neighboring cells j only. In this case we
would talk of a local CA. To what extent a CA can be limited to be local remains to be
seen. Such a CA often functions as a universal computer, but we also consider simpler
test examples, which are easier to subject to simple mathematical analysis.
Our cellular automaton itself never ends up in any ‘quantum superposition’. Thus,
its evolution is classical. It is not obvious whether the evolution law is time reversible,
but in our simplest models, it is. Note, that up to this point, we are not dealing with
probabilities at all. All one has to do is compute Q~ 2 , given Q
~1 .
However, to do the calculation correctly may be very difficult, if not impossible. If all
this happens at the Planck scale, there are far too many cells, evolving far too rapidly.
Now enters quantum mechanics. According to the CA interpretation, quantum mechanics
is just a handy tool. We define a quantum Hilbert space, spanned by the states |Q(t)i ~ ,
so that these states form the basis. In this Hilbert space, the evolution law can be written
in operator form,
~ 2 |U(t1 , t2 )|Q
hQ ~ 1 i = 1 if Q
~ 1 evolves into Q
~2 ; 0 otherwise. (7.1)
Causality requires that U(t1 , t3 ) = U(t2 , t3 ) U(t1 , t2 ) . If the evolution itself is time-
independent, we can write down an operator H such that
U(t1 , t2 ) = e−iH(t2 −t1 ) . (7.2)
Such a hamiltonian is often easy to construct, and it even allows us to impose further con-
straints such as positivity and some version of locality. To calculate the evolution operator
for large time intervals, we now make as many unitary transformations to different basis
sets as we please. Irrespective of the basis, we have (d/dt)|ψi = −iH|ψi . Therefore, we
may choose any basis, to solve this equation. Then, transform back to the original basis.
If the calculation is done correctly, this gives a state that coincides with a basis element
~ 2 i at time t2 ; the outcome may never be a superposition, such as λ|Q
|Q ~ 1 (t2 ) + µ|Q
~ 2(t2 )i .
We also observe that we could have started with a totally arbitrary wave function
|ψ(t1 )i , such that the probabilities for the initial states are defined by
~ t1 ) ≡ |hQ|ψ(t
̺(Q, ~ 2
1 )i| . (7.3)
The same Schrödinger equation as we have used before then gives us the wave function
~ t2 ) = |hQ|ψ(t
at time t2 , such that the probabilities at t2 are indeed ̺(Q, ~ 2
2 )i| . This is
exactly correct just because the CA does not produce superpositions itself.
During the calculation, we may have used non-commuting operators, and we may
have noticed that the state we are interested in was an eigen state. It can never have

15
been in eigen states of two non-commuting operators at the same time. And there is one
restriction: there is no ontological interpretation of amplitudes

hψ1 (t1 )|ψ2 (t2 )i (7.4)

if neither |ψ1 i nor ψ2 i are basis elements. This restriction is important to remember.
An other very important point to remember is that Eq. (7.3) cannot be inverted : Any
~ corresponds to a very large class of different wave
given probability distribution ̺(Q)
~
functions ψ(Q) , since the phases may be chosen at will. So, given the probabilities, we
may need to choose different wave functions under different circumstances, a feature often
ignored in hidden variable scenarios.
Our system is quantum mechanical in the sense that we use Schrödinger equations to
solve its dynamics. It is classical in the sense that CA states evolve into CA states. We
can map all its dynamical properties from the quantum picture to the classical picture and
back. The CA states are the ontological states. The wave functions |ψi (t)i = |Q~ i (t)i are
ontological (apart from their phases), and the complete set of ontological wave functions
form an orthonormal basis.
Two more remarks: 1) The hamiltonian defined in Eq. (7.2) is far from unambiguous
(If time is limited to integer multiples of a unit ∆t , then we can add any multiple of
2π/∆t to any of the eigen values of H without modifying the evolution, and this is
not the only ambiguity). This means that one CA can be mapped onto many different
quantum models. These however, are not physically different; they are related by unitary
transformations.
2) Because of this ambiguity, and because there are very many different CAs, it may well
be that CA models are dense in the set of all quantum models. It is not true that every
quantum model can be mapped onto a CA, but it is quite possible that many interesting
quantum models can be mapped onto CAs with some margin of accuracy. In practice,
however, one would like to reflect locality properties of the quantum system into the CA,
and such a demand can make the actual construction far from easy.

7.3. Measurements

We had made an important restriction in that the amplitudes (7.4) cannot be measured if
both |ψ1 i and ψ2 i are not elements of the CA basis. Which measurements are possible?
Here, we are going to make one more assumption. It is safe to assume that all measure-
ments end up being macroscopic. Only those phenomena that can affect the position of a
pointer on a device, and eventually the position of any macroscopic body such as a person
or a planet (or a cat) are measurable.
We make the essential assumption that, statistically, a macroscopic object influences
~
the distribution of ontological values |Q(t)i of the automaton. This means that, mea-
~
suring |Q(t)i sufficiently carefully gives us all classical data we want, and with that, the
outcome of all measurements.
We emphasize now that this should also include all quantum measurements, since any

16
measurement must end up being macroscopic. Imagine now that we have employed some
fancy quantum Hilbert space to calculate the evolution transition amplitudes (7.1). If we
succeed in designing an apparatus that turns a quantum phenomenon into a macroscopic
phenomenon, then this is classically measurable, and only the transition amplitudes (7.1)
suffice to detect the outcome; the amplitudes (7.4) are not needed.
Thus, imagine that we have constructed a CA model that can be mapped onto the
Standard Model in accordance with our prescriptions, then we can perform all measure-
ments that we are accustomed to, as soon as macroscopic detection devices such as particle
detectors can be constructed out of the atoms of this model, since its pointers are classical.
A classical object is then defined by requiring that it affects the statistical distributions
of the CA to such an extent that it can be distinguished from all other objects.
Opponents of our theory now bring forward that this means that not all expectation
values of all operators can be measured. At first sight this seems to be so, but remem-
ber that we do have a system that obeys exactly the Schrödinger equations of the full,
interacting, quantum field theory. This implies that, at various occasions, systems may
seem to be in eigenstates of operators that do not commute with CA observables, while
in reality they do lead to classical effects on pointers of measuring devices. These effects
do commute with the CA observables, and that is what matters, in the end.
This viewpoint is quite natural and conventional in quantum field theories. Here, one
can derive that, given any type of field φ(x) , where x is a space-time point in Minkowski
space, then the set of vacuum expectation values
Γ(x1 , · · · , xn ) ≡ h ∅ |φ(x1) · · · φ(xn )| ∅ i , (7.5)
suffices to determine all elements of the entire scattering matrix, including expectation
values involving any of the other fields. This is because, in moment space, the connected
parts of these amplitudes display poles due to all the other field variables, which one can
single out to obtain these other amplitudes. In our theory, we simply take one of the CA
fields to play the role of φ(x) .
Thus, the bottom line is that, since all measurements are classical, all measurable
operators end up to commute with the CA operators, and hence they all commute. What
we are saying here is that quantum superpositions will never stay visible as such after a
measurement has been done.
This also explains how an ontological wave function “collapses”: since classical ob-
servables all commute with the CA observables, the ontological wave functions are always
diagonal in the classical observables and never in a superposition: Schrödinger’s cat is
alive or dead![16]
In practice, a measurement can never do more than produce an estimate of the proba-
~ . The manifold defined by these probabilities has a dimensionality
bility distribution ̺(Q)
that is only half of that of the Hilbert space of wave functions. This means that many
different wave functions produce exactly the same probability distribution. We not only
cannot determine the wave function after an experiment has been performed, but we also
cannot prepare a state with a given wave function. The best we can do to obtain informa-
tion on the wave function that we wish to detect or produce, is to mimic the construction

17
of a measuring device, such as in a Stern-Gerlach experiment, or by constructing macro-
scopic polarization filters. According to the quantum field theoretical argument given
above, such devices can be imagined if we have sufficiently non-trivial interactions in our
model.
One does not have to believe that the CA interpretation is right. Or perhaps it is
nearly right: it may well be that one ingredient, not used in the present work, should
actually be included to obtain a workable theory: classical information loss[17]. It could
also be that the expanding universe plays an essential role here, since without expansion,
it would have been difficult to imagine how the necessary cooling mechanism would have
taken place, while this would be crucial to understand why we are surrounded by a near
vacuum. The vacuum will play an important role, as we shall see.
Taking all counter arguments in consideration, it may seem difficult to defend the CA
interpretation, without any explicit model that allows us to test the various assertions.
This is why we find the construction of the model in this paper important, even if some
ingredients may still be missing, such as a detailed mathematical treatment of the super-
string constraints and the superstring interactions. But now that we seem to be homing
in to exactly the CA model we were searching for, it is of importance to describe more
precisely what the CA interpretation says. We begin with what it does not say.

8. Hidden variables

The usual no-go theorems concerning hidden variables are based upon assumptions con-
cerning the theory they plan to disprove. It is assumed that the purported hidden variables
can do the impossible: either provide for the outcome of “counterfactual” measurements,
or compare the outcome of experiments made under exactly the same initial conditions,
except for one change made in the vicinity of one point in space, but later in time.
Counterfactual measurements are measurements that cannot be made, so a good hid-
den variable theory should not be required to give the outcome of such a measurement.
Secondly, an exact repetition of an experiment with exactly the same initial conditions
(including the values of all “hidden variables”) is impossible, if one wants to change the
situation near one point later in time. A deterministic theory does not have to define
what the outcome of such measurements should be, as they are ill-defined. Instead, the
theory should tell us what does go on in a Bell experiment.
The cellular automaton (CA) interpretation of quantum mechanics, the backbone
philosophy of this paper, pretends to do exactly that.

8.1. Vacuum fluctuations

First, we have to handle a complication, and indeed an essential one. It is that, in the
world we are familiar with, all objects are surrounded by regions that we call vacuum.
The vacuum state is a unique eigen state of the hamiltonian with eigen value zero. All
other states usually considered are very near to that zero eigen state, if just a few light

18
particles are around. This means that all states of interest will have to be handled as
superpositions of a large number of — perhaps a sizeable fraction of all — CA states.
The vacuum can only be regarded as ontological states rapidly evolving into one another.
Requiring the hamiltonian to be in its lowest eigen states means that, here, the relative
phases are of importance (they are dictated by our choice of hamiltonian), and the word
‘vacuum fluctuations’ aptly describes this situation.
In spite of this chaotic picture of what the vacuum looks like in terms of ontological
states, it is in reality a very special state. There will be quite complex, ‘entangled’, cor-
relations between the many possible values of the CA’s memory cells, and it is important
to realize that these correlations will be non-trivial at spacelike distances. As soon as the
correlations differ from the vacuum ones in a statistically significant manner, we have a
state with some classical object around. This is the way in which we will characterize
physical observations. If we have significant deviations from the vacuum fluctuations, we
have a state |ψi that will be orthogonal to the vacuum, and it will be recognizible by
operators that are diagonal in the CA basis. This is a crucial observation to make the idea
that such operators suffice to do just any observation or measurement, more acceptable.
How it can be that such a vacuum state came about in our world is a question that
we will not answer completely. One may imagine that the universe started out with a
Big Bang at infinite temperature. It expanded, and with that, cooled. How the universe
came to be as it is, dominated by regions of vacuum, is then the subject of cosmology.

8.2. What happens in a Bell experiment?

In Bell’s paper[1], the starting assumption is that a hidden variable, λ , provides for quan-
tities A(~a, λ) and B(~b, λ) , such that the expectation value for a combined measurement
is
Z
P (~a, b ) = dλ ̺(λ) A(~a, λ) B(~b, λ) ,
~ (8.1)

where ̺(λ) is positive and normalized. It is then proved, with mathematical rigor, that
this outcome disagrees with the quantum mechanical expectation values.
What this means is that the choice of the setting of a polarization filter ~a by observer
A cannot affect the action of polarization filter ~b on the photon observed by observer
B , if ~a and ~b are space-like separated. In quantum mechanics, Eq. (8.1) is violated, and
by many researchers this violation was seen as an apparent violation of either the hidden
variable concept, or of special relativity.
What does the CA theory tell us about Eq. (8.1)? Assume that the measurement
devices ~a and ~b , as well as the system that produces the quantum objects in question,
such the photons in an EPR experiment, are all spacelike separated.
“Superdeterminism” definitely applies to this system, so that there is no such thing
as “free will”, and we also do not share the revulsion often expressed against the idea
of “conspiracies”. It is however correctly pointed out in numerous investigations, that

19
this is not enough to escape from Bell’s theorem. Let us sharpen the formulation of the
problem.
The somewhat vague and misleading notion of “free will” must be replaced by an-
other demand that we think is more essential here: the freedom to choose our initial
state: a physical theory should provide meaningful predictions concerning the evolution
laws regardless our choice of the initial condition. Therefore, given the outcome of one
experiment, we should be able to predict the outcome of another experiment, if we make
just one modification, such as the setting of measuring device ~a . We keep the EPR
photons, as well as detector ~b , in the same states as before. According to Bell, as is
well-known, this leads to conflicts with Eq. (8.1). This requirement now, therefore, does
provide us with a fundamental problem. We will have to restrain the freedom to choose
the initial condition. How should this be done?
For simplicity, we consider an experiment with entangled spin 12 particles (of course,
photons with spin 1 can be handled the same way, but the notation would be a little more
clumsy). A particle polarized in the z direction is described by a wave function having
as a factor a state | ↑ i = ( 10 ) or | ↓ i = ( 01 ) . Particles polarized in the x direction have
wave functions | → i = 2−1/2 ( 11 ) or | ← i = 2−1/2 ( −1 1
) . A pair of entangled particles has
a wave function
 
1 1
√ (| ↓↑ i − | ↑↓ i) = √ (| → ←i − | ← →i = √ 1 0 1
. (8.2)
2 2 2 −1 0

If we had photons, we would be working with polarization filters, but since here we have
spin 21 particles, we should work with what we will call Stern-Gerlach detectors, or spin
detectors for short. They actually have the advantage that a particle can be detected
completely, while its spin component in one pre-assigned direction can be established.
The difficulty describing Bell experiments in the CA theory is the following. If the spin
detectors point in the z direction, each detected particle is split into the “ontological”
states | ↑ i and | ↓ i , and if the spin detector is orientated in another direction, say the x
direction, then the splitting is in terms of the states | → i and | ← i , which would then be
the ontological states. The difficulty becomes clear immediately: the entangled particles,
on their way to the spin detector, seem to know in advance which states are ontological.
But how could these entangled photons “know in advance” how to split up? How does the
CA “know” what the ontological states are? This, in a nut shell, is what many researchers
denounce as “conspiracy”. If we try to avoid it, we end up with Eq. (8.1), which cannot
be right.
The correct answer to such paradoxes can always be found by realizing how the map-
ping goes between a CA and the quantum mechanical system that we are describing, and
how a measurement takes place (see Section 9). In this case, the answer is to be found
in the structure of the vacuum state. As stated in the previous subsection, all our parti-
cles are surrounded by a vacuum, and that state is highly entangled, containing spacelike
correlations in terms of the states of the CA.
Now again consider the spin detector ~a , still spacelike separated from the place where
the entangled particles originate. Modifying ~a does not affect the wave function of the

20
entangled particles, but it does affect how this wave function is to be expressed in terms of
different ontological states. This we know for sure, because otherwise, Bell’s inequalities
would hold. How can we explain this?
The reason must be that the modified detector is surrounded by a vacuum whose
correlations are incompatible with those of the vacuum surrounding the original detector.
We know that this is possible because the vacuum has long-range, spacelike correlations.
But there is another way to see this. Let us look at the CA states. After the experiment
has been done, we have a definite ontological state, which we can extrapolate backwards
in time. Besides the entangled state (8.2), these particles can be in three other states.
These other states however do not fit with the boundary condition at times before the
particles originated, so these states do not agree with the original object surrounded by
any acceptable vacuum with appropriate correlation functions. From this it follows that
the vacuum correlations at the moment that the entangled particles were formed, were
very special, and they depend in a crucial manner on the spin detectors at some spacelike
distance away. Indeed, this could be called “conspiracy”, but it is conspiracy of an
acceptable nature: no signals can be sent faster than light. This conspiracy arises from
spacelike correlations that are quite familiar in quantum field theory; the correlations
are expressed by Green functions with points that are spacelike separated. Like the
propagators in QFT, such spacelike correlations do not vanish. The physical reason is
simple: the fluctuations arise from oscillations in a common past.
We note that, therefore, modifying the spin detector ~a without modifying the wave
function of the entangled particles is allowed, but modifying it without modifying the
ontological interpretation of these particles is not allowed. Apparently, we can’t modify
the initial state anyway we like: the definition of the vacuum state does not allow this.
This also implies that our modified notion of “free will” as formulated above is not valid.
We observe that non-locality of the CA itself is never needed for this argument. A
cellular automaton can be local in the sense that the evolution happens in small discrete
steps in the time coordinate, where every cell is updated in a way depending only on the
contents of neighboring cells.

8.3. Conspiracy

It is not so easy to dismiss the above observations as acts of “conspiracy” by Nature’s laws.
To illustrate this, let us take an example of “conspiracy” from number theory. Consider
a very large prime number P . Consider another, even larger number Q defined by

Q ≡ 2P −1 − 1 . (8.3)

What are the odds the Q can be divided by P ? A beginning student might argue that,
if P is hundreds of digits long, the odds against that are tremendous. Simple number
theory however says that Q is always dividable by P , provided P is a prime. If P is
not a prime, it nearly never is. The argument goes by simple counting, but one has to
count all the way to P . Is this a conspiracy? Perhaps, but it also is a fact.

21
 In our theory, all conspiracy goes away as soon as we realize that the most efficient way
to describe a large CA is indeed by replacing its classical evolution law by a Schrödinger
equation, which is known to yield exactly the correct results. The CA itself may seem to
be conspirational, if it is sufficiently non-trivial due to interactions. These interactions also
turn the CA into a universal computer, whose actions cannot be compressed as efficiently
as a quantum field theory.

8.4. Quantum computers

It is sometimes put forward that our findings imply that our quantum system should
be able to represent a quantum computer, whereas the CA, though very powerful, can
at best simulate a classical computer. Quantum computers are expected to solve non-
polynomial problems, which a classical computer cannot. Here, our response is that, a
quantum computer can only outperform a classical one if the latter works perfectly. The
quantum systems generated by CAs will in general be ugly, containing various types of
interactions. These interactions may well cause limitations of a fundamental type to the
performance of any quantum computer.

8.5. Classical information loss

Before phrasing the axioms of the CA interpretation, an important generalization must

be mentioned. It is conceivable that the classical automaton is not time-reversible. This
means that, given the configuration at time t = t1 , there could exist more than one
different configurations at t = t0 < t1 that evolve into the given one, or there could
be none at all (“Garden of Eden”). At first sight, this might be considered unworkable,
but in principle, such a CA can be used. What has to be done is that the CA states
must be assembled in equivalence classes: two states are equivalent if, at some time in
the foreseeable future, they both evolve into the same state. By construction then, every
equivalence class evolves into exactly one equivalence class in the future, and it has one
equivalence class in its past (that class could be the empty class, but once classes have
grown to be large enough, this case might be negligible).
These equivalence classes are very similar to gauge equivalence classes: it is impossible
to distinguish experimentally which element of the equivalence class we have, because this
information disappears after a finite amount of time.
We now associate one normalized quantum basis element to each equivalence class.
Our quantum Hilbert space then must be the Hilbert space spanned by these basis ele-
ments.
A notable feature of this Hilbert space is that it might be much smaller than if no
information loss took place. It could even be that, for a given volume in 3-space, the
equivalence classes can be associated with the structures at the surface only, in case all
information in the bulk becomes unstable. This could be an interpretation of “holography”
in quantum gravity and black hole physics.

22
 An advantage of this scenario could be that the association between quantum states
and ontological states may become non-local to a large extent, which could well explain
the apparent contradictions between the simple-minded time-reversible CA theories and
the attempted interpretation of Bell experiments in quantum mechanics as we know it.
We would have a peculiar, but desirable feature: Quantum theory is local in the sense
that commutators vanish outside the light-cone, and the classical CA theory is local as
well, but the mapping between the quantum states and the CA equivalence classes is
an apparently non-local one. This could be regarded as a possible loophole to reduce
apparent contradictions with Bell’s theorem, but this remains to be seen. As yet, the
mapping we considered does not seem to be ‘sufficiently’ non local, and it is our suspicion
that the occurence of spacelike correlations in the vacuum (subsection 8.1), perhaps in
combination with the ambiguity of the wave function, as emphasized in subsections 7.2
and 8.2, is a more elegant loophole.
It is noteworthy that the superstring theory that we worked out in this paper is time-
reversible, so information loss was not put in. On the other hand, we do know that string
theories are associated with black holes where we do have holography. This is why we put
information loss on hold in this paper. It just gives one enough room to speculate that
more complete models to be constructed in the future may handle the non-locality issues
in a more complete manner.

9. Axioms of the CA interpretation

In view of the above, we can now attempt to phrase the essentials of the CA interpretation
as succinctly as possible. We expect that these might be sharpened when the theory
becomes more mature. Let us start from the quantum system we wish to describe.

i We consider quantum theories that obey all postulates of standard quantum me-
chanics in agreement with the Copenhagen interpretation, except that we put an
important constraint on what can eventually be detected, see below, under (iv) .

ii In many interesting theories, a special basis in Hilbert space can be found that we
~
will call the ontological or CA basis. This is an orthonormal set {|Q(t)i} that
~
spans all of Hilbert space. Only these basis elements, |Q(t)i , have the property
that the evolution operator over specially chosen time steps simply permutes these
states. They describe all ontic states. We may describe them as depending on a
time variable t . The phase angles of these states are physically meaningless.

iii All other states |ψi in Hilbert space describe probabilistic distributions,
~ = |hQ|ψi|
̺(Q) ~ 2
. Again, the phases of the CA states are meaningless. However, if
we fix the phases for the CA states, we can define uniquely the evolution operators
U(t1 , t2 ) , and write these as e−iH(t2 −t1 ) .

At this stage, it is emphasized that a generic quantum state |ψi may have any set of
~
phases in its inner products hQ|ψi . Changing these phases does not have any effect on the

23
physics. This means that many different wave functions can be used to describe the same
reality. In a more conventional basis (to be referred to as the SM basis), the hamiltonian
for these different wave functions will look totally different since the phases enter into the
expressions for H . Note that it would not be a good idea to choose all phase factors of
~ states to be +1 , since we wish to use the freedom to perform superpositions.
the |Qi
The fact that (time-dependent) unitary transformations may exist as invariances will
be an important feature of the quantum theory, but such invariances may be difficult to
identify in the real world.

iv Only the projection postulate changes: one cannot project a quantum state |ψi
onto, say, an eigenstate |χi of an arbitrary operator, to measure hχ|ψi . To mea-
sure something, we must make a model of the measuring apparatus that eventually
produces a macroscopic effect. This macroscopic effect can always be represented
~
by the ontological states |Q(t)i , so that the state of the detector can be measured
by projecting |ψi on these CA states.

This presupposes that such detectors exist in the given CA model, which is far from
obvious, but we know of course that they exist in the real world. This requires the
following assumption:

v The ontological states |Qi~ , equipped with arbitrary phase angles, are assumed to be
locally dense in Hilbert space. This means that in any restricted region of space-time,
one can approach any given quantum state |ψi with a CA state; orthonormality is
restored by differences outside that region.

The accuracy of the approximation, or the question exactly how dense this subset is, may
depend on the volume, this will be worth further investigation.

vi The complete vacuum is not an ontological state, but a very delicate superposition.
Locally it can be approached by CA states, see (v) . Conversely, a vast majority
of the CA states can never be reaized in the real world because they represent
improbable amounts of energy compressed in very small regions.

vii Conversely, we can start with a cellular automaton (CA) that evolves classically.
This evolution can then be represented by an evolution operator. Standard quantum
mechanics can be recovered from that by performing unitary transformations and
constructing an appropriate hamiltonian.

This may actually lead to several quantum models representing the same CA, for instance
~ , or by adding conserved quantities to the
by redefining the phases of the states |Qi
hamiltonian.

viii Because of the non observability of the phase factors of the CA states, each proba-
~ of the CA states represents a very large class of apparently
bility distribution ̺(Q)
different wave functions.

24
This implies that replacement of a wave function by an element of the same class has no
physical effect but may be required for understanding of how Bell’s theorem is evaded.
The work presented in this paper suggests that mappings of the kind used here do
exist; in fact, they may seem to be rather easy to construct except that in most cases,
either the CA or the quantum system are too far remote from the desired theories. It
is for instance difficult to impose rotation invariance and even more difficult to impose
Lorentz invariance, let alone general coordinate invariance. It is these latter requirements
where we now have made progress.

10. Discussion and conclusions

We think that something very important has been discovered:

Both the bosonic string theory and superstring theory can be reformulated in
terms of a√ special basis of states, defined on a space-time lattice, with lattice
length 2π α′ . The evolution equations on this lattice are classical. This
allows for a CA interpretation of superstring theory.

Although the lattice is defined in Minkowski space, Lorentz invariance is not generally
guaranteed. This is where we can use established knowledge of (super)string theory to
conclude that space-time must be 26 or 10 dimensional. Discrete Lorentz invariance may
be easier to establish, although there are a few considerations that have to be taken into
account. Discrete Lorentz invariance would be evident if the superstring constraint (6.2)
would be replaced by constraint (3.3) of the discrete theory, but they do not seem to be
the same.
There are a number of issues that we briefly address in this Section.

10.1. Classical vs. quantum mechanics

The main theme of this paper is that a mapping is found between a classical (CA) de-
scription of a superstring, and the usual quantum mechanical one. Here is what we have:

- The independent degrees of freedom of strings and superstrings are the transversal
string coordinates and spinorial fields. On the world sheet of the string, they all
obey strictly linear field equations, such as:
q µ (x, t + δ) = q µ (x − δ, t) + q µ (x + δ, t) − q µ (x, t − δ) , (10.1)
where δ can be any fixed number. This simply follows from the fact that there
are left-movers and right-movers. Because of this, these degrees of freedom can
be decomposed into lattice degrees of freedom without loosing any generality. The
lattice has light like coordinates with lattice length δ , but at any stage, we can
choose δ to be something else, so it should be easy at a later stage to take the
continuum limit.

25
 - By splitting up the transverse bosonic string degrees of freedom q a (a = 1, · · · , D −
2) into their integer parts and their fractional parts (a precise description of this
splitting being a little more complex than this, see Ref. [9]), we find that also space-
time itself can be limited to being a lattice. The fractional parts actually also can
be transformed to being a lattice, simply by Fourier transforming: they form the
dual lattice.
- Only the commutators between the lattice and the dual lattice variables are non-
vanishing. The lattice degrees of freedom all commute with one another. Also the
dual lattice degrees of freedom all commute with one another. Note that the dual
lattice and the lattice itself describe the same physics.
- Because all variables on a lattice commute, a basis of states exists such that, if a
wave function starts off as having only 1s and 0s, it will always be restricted to 1s
and 0s. This means that a formalism is found in terms of which states that are
“certainly true” evolve into other states that are “certainly true”. There will never
be quantum superpositions. This basis is referred to as the ‘ontological basis’.
- If the (super)string interactions are described as exchanges of legs when strings
intersect, then it seems that these can also be described as “certainties”: there still
will be no quantum superpositions in the CA description. This however may only
be true if the string interaction constant has a fixed value.

For (super) string theory this may be an interesting observation: not all string theories
are equal; these will be special, having more predictable properties.

- Quantum superpositions will occur as soon as we try to describe the string modes in
terms of a quantum field theory. This is just the passage towards another quantum
basis. Only in terms of such a basis, indeed any of the basis choices usually employed
in quantum field theory, or quantum theory in general, interference phenomena are
perceived as real, and this is why we do encounter quantum phenomena in ordinary
physics.

Most of our symmetry transformations only exist in the quantum description, not on the
ontological basis. Rotations in the transverse dimensions are continuous only in terms of
the real variables q a (x, t) , not after we made the transition to the target space lattice
variables, where only discrete crystal transformations can be done. Lorentz transforma-
tions are even more subtle; they also involve the world sheet lattice in a complicated
way.
The philosophy used here is often attacked by using Bell’s inequalities[1]—[3] applied
to some Gedanken experiment, or some similar “quantum” arguments. The apparent
no-go arguments have to be considered carefully. They are very important to put things
in perspective, and in general they are employed to conclude that there are still essential
elements missing in the picture we have. In the present version of this paper, we addressed
the associated issues in Sections 7—9.
We emphasize that the mathematical expressions we found corroborate our views.

26
 10.2. The world sheet lattice

Our theory does away with all distance scales shorter than the lattice length aℓ in a
very effective way. The string degrees of freedom are split up in their integer parts and
fractional parts (in units where aℓ = 1 ). The fractional parts are redundant because they
merely refer to operators acting on the integer parts.
In conventional formulations of string theory, one often picks the gauge choice

∂+ q + = 21 (∂x + ∂t )q + = a+ +
L = p , (10.2)

where p+ is the total light-cone momentum, a conserved quantity. One then gets
1  tr 2 
∂+ q − = 12 (∂x + ∂t )q − = a−
L = (aL ) + C , (10.3)
p+
where C , the mass-squared of the lowest state, or intercept, is either tachyonic or it
vanishes.
For the continuum theory, the p+ dependence is of no consequence, as it is dictated
by a trivial component of the Lorentz transformations. One can choose the Lorentz frame
where it is one. In the space-time lattice, however, it is quite non-trivial, since both the
left movers AL and the right movers AR are now constrained to be integers. If we limit
ourselves to the discrete Lorentz group O(D − 1, 1, Z) , then A± L,R are all integer, but if
+
the sequence AL (x + t) takes on different integer values, then just setting these all equal
to one, as in the gauge choice (10.2), at first sight seems not to be allowed.
If at one spot, A+L takes a value N , one might consider adding N − 1 dummy lattice
sites between two consecutive string joints there, but then soon the question will arise how
to assign the other A± L,R values at these intermediate points. To answer this question,
we believe it will be best to go to the continuum limit on the world sheet, but the logic of
that must be further investigated. Our point in this subsection is, that at all stages one
has the freedom to choose the gauge, and choosing the longitudinal components of the
physical degrees of freedom serves no other purpose than to establish Lorentz invariance.
We actually have to do this in the continuum theory differently from the lattice theory,
as the next subsection shows.

10.3. The string constraints and Lorentz invariance

At first sight, the superstring constraints (6.1)—(6.5), seem to be so similar to the con-
straints (3.3) and (3.4), that one might think they can be mapped easily. This however,
does not seem to be the case. We have the problem of the superfluous world sheet lattice
coordinates, as displayed in the previous subsection. But there is also the problem that
on the lattice only a limited number of Fourier components occur, while the continuum
has infinitely many.
In the continuum, the Lorentz group that can be constructed closes algebraically,
due to the special linear properties of the commutation relations of the string excitation

27
modes, though only if D = 26 or 10 . On a finite lattice, the commutation rules among
the excited modes become more complicated and the Lorentz group does not close. This
is linked to the fact that the lattice requires a preferred gauge, which no longer holds
after a Lorentz transformation. We believe that this is a consequence of the Lorentz
group here being discrete rather than continuous. Continuous Lorentz invariance is not
self-evident, and indeed may often fail. Indeed, all we have at present is the conjecture
that continuous Lorentz invariance emerges spontaneously if we build in discrete Lorentz
invariance. Our reasons for this conjecture is that we do have emergent invariance under
continuous O(D − 2, R) rotations, which, together with invariance under the discrete
Lorentz group O(D − 1, 1, Z) , should generate continuous Lorentz invariance.
Regarding this latter argument, one may now wonder why this can go wrong in the
cases that D 6= 26 and 6= 10 . We then must note that there may be subtle features that
jeopardize discrete Lorentz invariance. There is a state counting problem.
Suppose the transverse degrees of freedom Atr L,R are all given. This means that the
longitudinal components are fixed by their products, A+ − + −
L AL and AR AR , and the fact
that all these variables are integer. This, however, does not fix their values completely, so
we must choose the world sheet gauge more precisely, or else accept that different states
are described by the same transverse data. Either we under-count or we over-count states
this way. Exactly how to do this right is still an open question. Perhaps details of the
lattice are important to resolve this problem: are all sites occupied? Do certain even or
odd positions stay empty? Do we have to consider different space-time “crystals”? One
is tempted to suspect that this is a source where the dimensionality of the lattice plays
an important role.

10.4. Why fermions? Why supersymmetry?

Fermions are added to the discrete deterministic theory by inserting D − 2 Boolean

degrees of freedom to every lattice site on the world sheet. In the time evolution, these
Boolean degrees of freedom are passed on as left- and right movers just as in a cellular
automaton. It was easy to turn these into genuine fermions by means of a Jordan Wigner
transformation[15]. In the bulk theory, each fermionic mode does not seem to be directly
linked to any of the transverse directions on the lattice, so their spin structure stays
obscure. This changes only when the constraints are considered, which means that their
properties under rotation become clear only when Lorentz transformations are considered.
Galilean transformations may suffice, so this means that the spin rotation properties are
revealed as soon as these particles move, in target space. Since both fermions and bosons
are massless and non-interacting, having exactly D − 2 fermions automatically leads to
supersymmetry, which is also necessary to remove the longitudinal modes.
In our formalism it is as yet hard to see why the presence of these fermions affect the
number D of space-time dimensions. An observation to be made is that the state counting
problem is intimately tied to the question how exactly the space-time lattice should be
occupied. Are there empty sites? In the space of integers, one finds that sometimes
it makes a difference whether world sheet lattice sites are characterized by even or odd

28
integers. This may give rise to an interesting speculation concerning the fermionic degrees
of freedom. Maybe the Boolean variable σ(x, t) indicates whether the lattice site occupied
at (x, t) is even or odd. To find out more about this, the state counting problem may
have to be addressed, and this was as yet too complicated to consider here.

10.5. The lattice length in space-time

On the world sheet lattice we saw that the even sites (where x + t is even) are detached
from the odd sites ( x + t odd). Also in the target space (space-time) lattice, it may be
important to distinguish even and odd integers. If, for instance
D−2
X
A+ −
L AL = (AaL )2 ,
a=1

(and similarly for the right-movers), we see that A+ −

L and/or AL are even or odd depending
on the number of even or odd values of AL , modulo 4. Note that if A+
a
L is odd, then so
−
is AL . So it may happen that the periodicity of the lattice is twice (or more times?) the
minimal integer number occurring there. We normalize the minimal integer on this lattice
as the number aℓ , so the lattice period may be a multiple of this. In our calculations, we
have been using units where aℓ = 1, c = 1 , and h = 2π~ = 1 . In these units,
√ √
aℓ = h α′ = 2π~ α′ . (10.4)
The string coupling constant gs is also one in our theory, since it describes the exchange
operation which is indeed unity if we also insist that the interaction respects determinism.
Since gs also fixes the strength of the gravitational force, we get that α′ must be close to
the Planck length squared, and our space-time lattice is close to the Planck length5 itself.

Acknowledgements

The author thanks in particular M. Porter, R. Maimon and P. Shor, for their constructive
criticism.

References

[1] J. S. Bell, “On the Einstein Podolsky Rosen Paradox”, Physica 1 (1964) 195.
[2] A. Einstein, B. Podolsky and N. Rosen, “Can Quantum mechanical description
of physical reality be considered complete?”, Phys. Rev. 47 (1935) 777; M. Jam-
mer, “The conceptual Development of Quantum Mechanics”, (Mc. Graw-Hill, 1966);
A. Aspect, J. Dalibard and G. Roger, “Experimental Test of Bell’s Inequalities Using
Time-Varying Analyzers”, Phys. Rev. Letters 49 (1982) 1804.
5
Remember that the actual physical value for the Planck length may depend on the compactification
scale of the extra dimensions.

29
 [3] J. Conway and S. Kochen, The Strong Free Will Theorem, arXiv:0807.3286
[quant-ph].

[4] D.J. Gross, personal communication, 25th Solvay Conference in Physics on “The
Theory of the Quantum World”, October 2011.

[5] G. ’t Hooft, “Quantization of discrete deterministic theories by Hilbert space exten-

sion”, Nucl. Phys. B342 (1990) 471; G. ’t Hooft, K. Isler and S. Kalitzin, “Quantum
field theoretic behavior of a deterministic cellular automaton”, Nucl. Phys. B 386
(1992) 495; G. ’t Hooft, “Quantummechanical behavior in a deterministic model”,
Found. Phys. Lett. 10 (1997) 105; quant-ph/9612018.

[6] M. Blasone, P. Jizba and H. Kleinert, Annals of Physics 320 (2005) 468, arXiv:
quant-ph/0504200; id., Braz. J. Phys. 35 (2005) 497, arXiv: quant-ph/0504047.

[7] S. Wolfram, “A New Kind of Science”, Wolfram Media, Inc., May 14, 2002. ISBN
1-57955-008-8.

[8] G. ’t Hooft, “Relating the quantum mechanics of discrete systems to standard canon-
ical quantum mechanics”, arXiv:1204.4926.

[9] G. ’t Hooft, “Duality between a deterministic cellular automaton and a bosonic

quantum field theory in 1+1 dimensions”, arXiv:1205.4107.

[10] L. Hardy, “Are quantum states real?”, arXiv:1205.1439.

[11] M.B. Green, J.H. Schwarz and E. Witten, Superstring Theory, ISBN-10: 0521357527
— ISBN-13: 978-0521357524, Cambridge Univ. Press, 1987.

[12] J. Polchinski, “String Theory”, Volume I, An Introduction to the Bosonic String,

ISBN 0-521-63303-6; Volume II, Superstring Theory and Beyond, ISBN 0-521-63304-
4, Cambridge University Press, 1998.

[13] I. Klebanov and L. Susskind, “Continuum strings from discrete field theories”,
Nucl. Physics B 309 (1988) 175.

[14] L. Susskind, personal communication.

[15] P. Jordan and E. Wigner, Z. Phys. 47 (1928) 631.

[16] G. ’t Hooft, “How a wave function can collapse without violating Schrödinger’s
equation, and how to understand Born’s rule”, ITP-UU-11/43, SPIN-11/34, arXiv:
1112.1811[quant-ph].

[17] G. ’t Hooft, “The mathematical basis for deterministic quantum mechanics”,

arXiv:quant-ph/0604008.

30