Appl. Math. Inf. Sci. 8, No.

5, 2405-2416 (2014) 2405

Applied Mathematics & Information Sciences
An International Journal

http://dx.doi.org/10.12785/amis/080536

Continuous Turing Machine: Real Function

Computability and Iteration Issues
Xiaoliang Chen∗ , Wen Song, Zexia Huang and Mingwei Tang
School of Mathematics & Computer Science, Xihua University , Chen’du 610039, P. R. China

Received: 8 Sep. 2013, Revised: 6 Dec. 2013, Accepted: 7 Dec. 2013

Published online: 1 Sep. 2014

Abstract: Contemporary computer theory is governed by the discretization of continuous problems. Classical Turing machines (TMs)
are originally built to solve computation and computability problems, which main feature is discreteness. However, even some simple
numerical calculations problems, e.g., iterations in Rn , generate difficulties to be described or solved by constructing a TM. This paper
explores the computability of continuous problems by proposing a class of continuous Turing machines (CTMs) that are an extension
of TMs. CTMs can be applied to the standard for the precision of algorithms. First, computable real numbers are precisely defined by
CTMs and their computations are regarded as the running of the CTMs. CTMs introduce the coded recursive descriptions, machine
states, and operations with the characters of computer instructions in essence compared with usual computable continuous models.
Hence, they can precisely present continuous computations with the form of processes. Second, the concepts of CTM computable
and CTM handleable are proposed. Moreover, the basic concepts on approximation theory such as convergency, metric space, and
fixed-point in Rn are defined in a new space CTMRn . Finally, an iterative algorithm is shown by constructing a CTM to solve linear
equations.

Keywords: Computational mathematics, computer theory, continuous Turing machine, real number computability, iteration.

1 Introduction analyze the computability of real numbers. Mazur [5]

defines computable real functions by the proposed
The development of computability theory in information sequence computability. Kreitz and Weihrauch [6, 7] take
science begins with the generation of recursive functions into account the presentation of real number
that depend on logical theory. These recursive functions computability by introducing type-2 theory of
are considered as the precise definitions of intuitive effectiveness (TTE), which is based on the theory of
algorithms [1]. Turing describes computations by a class representations and is an approach of computable
of mathematical machines (theoretical computers), analysis. Edalat [8, 9, 10, 11] studies computable real
usually called Turing machines (TMs). The machines functions by domain theory.
precisely present the concept of computations with the
form of processes by introducing machine states and the Many constructive analysis methods are also
operations with respect to the characters of computer proposed. Moore [12] proposes µ − hierarchy to interpret
instructions. TMs are equivalent to recursive functions. recursion theory on reals and constructs flowcharts of
Hence, computability problems are equivalent to Turing continuous time to handle real number computability and
computability [2]. halting problems. Doraszelski and Satterthwaite [13]
A theory is said to be an systematic approach if its define computable real numbers by the established
deduction and reasoning depend on a standard Markov arithmetic. Blum, Cucker et al. [14, 15] analyze
mathematical model. Computations are model-based the computable problems of real numbers by constructing
processes in the solution of a given calculated problem. real-RAM models. However, the fact is that not all real
However, the existing TM computability theory cannot number computability that are described above are
properly presents computable real functions since TMs equivalent. For example, Banach-Mazur computability is
are discrete in essence [3, 4]. Thereafter, some not equivalent to Markov computability for computable
mathematical models and approaches are developed to real numbers [16]. On the other hand, the described
∗ Corresponding author e-mail: [email protected]
c 2014 NSP
Natural Sciences Publishing Cor.
2406 X. L. Chen et. al. : Continuous Turing Machine: Real Function Computability...

theories and models are difficult to compatible with the Hybrid nets et al [17, 18] are proposed. These models do
classical model TMs. Hence, this paper attempts to not means that the number of new models is increase in
construct a class of extension TM to deal with real the series of computational models. Its purpose is to
number computability. correctly and easily present, solve, and analyze a class of
Numerical analysis involves the methods for real continuous problems. A good approximation generally
number calculations. However, it does not consider proves very valuably to solve a complex problem.
computability problems. Different computational models Therefore, an approximate method is considered in TMs
can obtain inconsistent results on whether a real number in this paper.
problems is computable. For example, a serious distortion Generally, continuous models are time-related.
or an entire wrong conclusion may be obtained if the real However, CTMs are regarded as time-independent. This
numbers computability is considered by TMs. Hence, section expands TMs to CTMs by introducing an example
discrete machines do not properly demonstrate real for a non-output and two-type nondeterministic TM M1
number computability. On the other hand, although the (Fig. 1(a)). First, M1 can be constructed by the following
theory of numerical analysis makes great achievements in algorithm.
the past, its developments necessarily need a reliable
computable theory.
This paper begins with an extension from discrete Algorithm for M1 construction.
TMs to continuous TMs. Then, a class of autonomous TM M1 :=“On input ω = ω1 , ω2 : // ω1 and ω2 represent
continuous Turing machines (CTMs) is proposed in initial inputs of the two tapes, respectively.
section 2. The rationality by using CTMs to explore the DO {
step1: If there is a ‘0’ on tape1, then move it to tape2 or
computability of real numbers is considered in section 3.
retain it in tape1 in a random manner.
CTMs have mainly two strengths compared with the
step2: If there is a ‘0’ in tape2, then move it to tape1 or
usual models. Firstly, the classic methods of TMs deal retain it in tape2 in a random manner.
with the computations of natural numbers, the sets of }while .T.” //Notation .T. means that the logical condition
natural number, and the arithmetical functions. A CTM of the loop ‘while’ is always true. 
covers continuous computations and include discrete
computations. Second, it is realistic and feasible since the
concept of computable is defined by constructing CTMs. According to the view of machine computations, a
An algorithm is said to be computable if a CTM can be character ‘0’ in M1 can be considered as a certain mount
constructed for a certain input to reach an output at finite of resources. Hence, a single resource is represented by a
steps. single ‘0’ and several resources are represented by
CTMs have simple structures, basic operations, and multiple ‘0’ (0∗ ) in M1 . TMs are theoretical models of
precise descriptions of computations in the form of computers. In a real-world computer, the number of ‘0’
processes. A CTM series with respect to greater power can be represented as the amount of information. For
can be constructed by an iterative or recursive example, a single ‘0’ can be interpreted as 1G information
construction of CTMs. Iterative technology based on and ‘00000’ as 5G information.
CTMs is considered in section 4, which demonstrates an Second, a transformation is considered to divide each
approach to prove CTM-computable and to explain how ‘0’ into k equal parts. This new TM is denoted by M1k and
to construct a complex machines. Finally, a typical shown in Fig. 1(b). The world ‘block’ is assumed as an unit
example is given to illustrate real function computability, of ‘0’ in initial configurations. Each block is divided into
which can be regarded as a methodology to solve a class k. The new unit that is one k-th of block is called ‘piece’.
of computable problems. In section 5, we state the results For example shown in Fig 1, the initial configuration of
of this paper. M1 (Fig 1(a)) leads to the configuration of M1k (Fig 1(b)) in
which the resources are expressed in pieces.
Generally, the transition functions of multiple
2 Extension: discrete TM to continuous TM nondeterministic Turing machine (MNTM) have the form
The simplest way for a TM to compute four arithmetic δ : Q × Γ K → P(Q × Γ k × {L, R, S}k ),
operations is that the representation of numbers only uses
‘0’ [3], where notation ‘0’ is a character in the tape of where Q is the set of states, Γ is the tape alphabet, k is the
TMs, which is distinguished with the numerical zero. number of tapes, and P is power set. The expression
However, the representation method can lead to the δ (qi , a1 , · · · , ak ) ∈ P(q j , b1 , · · · , bk , L, . . . , R), ai , b j ∈ Γ
increase of storages. Importantly, by considering iterative
computations, a self-iterative TM can difficultly be means that if the sate of a machine is qi and read-write
constructed since computations, e.g., iterations, cannot be head 1 through k are reading symbols a1 through ak ,
easily represented by integers. Many researchers make respectively. The machine goes to one of possible states
extensions from discrete models to continuous ones, q j and writes symbols b1 through bk . Correspondingly,
where continuous automatons, continuous Petri nets, and transition functions directs each head to move left, right,

c 2014 NSP
Natural Sciences Publishing Cor.
Appl. Math. Inf. Sci. 8, No. 5, 2405-2416 (2014) / www.naturalspublishing.com/Journals.asp 2407

δ1 δc and adding a piece to tape2. Hence, the evolved

0 0 tape1 0 0 tape1 q1 q2 configurations can be expressed in blocks (integer) or in
δm
Controller Controller
δc δm δ1 pieces (rational number if k is finite). Let Cik be a
M1 M1(k)
tape2 tape2
q3
configuration that is expressed in pieces in M1k and
a b c Ci = Cik /k be the corresponding configuration that is
expressed in blocks in M1 . Obviously, the computational
Evolve  q1 0 0  δ1  q2 0  δ1  q3 
configurations  q1   0 q2   0 0 q3  processes of M1 are included in the processes of M1k .
for M1   δ2   δ2  
d

{
 q2 0  δc  q2  δm  q3  M1(k)
   0 q2 0   0 0 q3 
 0 q2      Controller
δ 2 := [δ mδ c ]
 q1 0 0  δc  q1 0  δm  q2 0  tape11 1/k 1/k tape21
 q1   q1 0   
     0 q2 
tape12 1/k 1/k tape22
e 2k

{
Expressed tapes
δk
 δ1 
k
in blocks q1 k k  1 q2 k-1 k   q2 k   q3 1 q3 
q1  k q2 1 Lk Lk   
Evolve   δ2    q2
  q3 k-1
 δ 2k k k q3
 tape1k 1/k 1/k tape2k
configurations
for M 1(k) δ 1k q 1 1 δ 1k
q1 0 0  2 1−
k k   q2 0   q3 k   q3  Fig. 2: Structure of 2k-type M1K .
q1  q2 1  L0 Lk 1   
q2 q3
 δ 2k 0 0 q3
1−
Expressed   δ 2k  k     k 
in pieces
f

Fig. 1: Transformation of a TM: (a) ordinary TM M1 , (b) The fact that transition functions of a k-type TM
transformed ordinary TM M1k , represented for k=4, (c) state execute simultaneously is denoted by (δi δ j · · · δk ), where
graph of M1 , (d) evolve configurations for M1 , (e) implied evolve δi , δ j , · · · , and δk are transition functions from tape i to k,
configurations in Fig(d), and (f) evolve configurations for M1k . respectively. The structure of M1k , which can execute δ1 k
times simultaneously, is shown in Fig. 2. The execution of
δ1 in M1 equals the execution of (δ1k δ1k · · · δ1k ) in M1k . We
can change a way to describe transferred processes by
introducing some new notations. It facilitates to discuss
or to stay put. By considering M1 as an example and its the extension from discrete TMs to continuous TMs.
state graph shown in Fig 1(c), its evolutions contain three
types of transition functions: Let [δi \ δ j \ · · · \ δk ] be a class of orderly executive
δ1 : (qk , 0, ) = (qk+1 , , 0, R, R) sequences of transition functions.

δm : (qk , α , ) = (qk−1 , α , , L, L), ∀α ∈ Γ Let [δi δ j · · · δk ] be a class of synchronized executive

sequences of transition functions.
δc : (qk , , 0) = (qk , 0, , S, S)

Symbol  ∈ Γ denotes that there has not a resource at Let [δi ]α = [(δi )α ] be a class of special executive
corresponding positions on the tape, i.e., blank. Now, we sequences, which performance means that TM
consider M1k with an strategy for the segmentation of implements transition function δi total α times
resources. Its transition functions are similar to M1 . simultaneously and removes or adds α piece resources in
However, let the scale of its evolutions be the unit of its tapes to produce a new configuration, where α is a
piece. For example, δ1 , δm , and δc in M1 are represented non-negative number.
by the following functions combinations: Fig. 3(a) shows a set of possible transitions of M1 that
are concerned with two block resources. In addition to
(qk , α , k − α ) = (qk , α − 1, k − α + 1, S, S), ∀α = 1, 2, · · · , k single execution of δ1 or δ2 , multiple transitions by the

δ1k :
(qk , 1, k − 1) = (qk+1 , , k, R, R)
execution of [δ1 \ δ2 ], [δ1 ]2 , and [δ2 ]2 are also represented.
δmk : (qk , {α , }, ) = (qk−1 , {α , }, , L, L), ∀α = 1, 2, · · · , k The possible transitions of M1k for k = 4 are shown in Fig.
δck : (qk , k − α , α ) = (qk , k − α + 1, α − 1, S, S), ∀α = 1, 2, · · · , k 3(b). M1k contains many and finite multiple transitions,
e.g., [δ1 ]3 , [δ2 ]2 and [δ2 ]6 . We apostrophe read-write head
By considering M1 and its evolved configurations and state alphabet ‘qi ’ for simplification. By observing
shown in Fig. 1(d), the execution of transition function δ1
the execution of [δ1 ]3 , its transition process can be
consists of removing a block from tape1 and adding a expressed in pieces as:
block to tape2. Correspondingly, by considering M1k and
its evolved configurations shown in Fig. 1(f), the tape1 1 4  [δ1 ]
  3  
2
execution of δ1k consists of removing a piece from tape1 −−−>
tape2 3   42

c 2014 NSP
Natural Sciences Publishing Cor.
2408 X. L. Chen et. al. : Continuous Turing Machine: Real Function Computability...

Or it can be expressed in blocks as: configuration (2 − α , α ) at a moment, where α is any real

number in the range [0,2] and β is called a transferred
tape1 0.25 1  [δ1 ]
0.75 
quantity and satisfies the inequality 2 − α ≤ β ≤ 2.
  
 0.5 
−−−−> Similarly, if there is an execution of [δ1 ]γ , then γ satisfies
tape2 0.75   1 0.5 
0 ≤ γ ≤ 2 − α . Multiple executions taken a form of
Furthermore, we can also describe in pieces as: [(δ2 )β (δ1 )α ] are possible. TMs M1 and M1k discussed
above belong to MNTM in essence. The defined
[ δ1 ] 3 machines are recognizers [3] of languages in this paper if
(5, 3) −−−> (2, 6)
there are not special remarks.
or express in blocks as: Lemma 1.Two-type nondeterministic TM M1 has an
equivalent five-type deterministic TM Md .
[δ1 ]0.75
(1.25, 0.75) −−−−> (0.5, 1.5) Proof. The ideal is make mutual simulations between M1
and Md . The fact that M1 simulates Md is simple since
deterministic TM M1 is an special case of
nondeterministic TM Md and we only needs to construct a
tape2 nondeterministic computational branch in M1 .
On the other hand, if Md is constructed to simulate M1 ,
2 Md needs trying all possible branches of nondeterministic
δ2
[δ2 ]2 computations of M1 . The machine Md can be established
δ1 [δ1 \ δ2 ] by constructing five tapes. As shown in Fig. 4, we assume
1 that every tape has a particular function. Tape 1 and 2 are
[δ1 ]2 δ2 similar to the tapes in M1 . They contain constant strings
δ1
that copy from the initial inputs of M1 . Tape 3 and 4 are
tape1
0 1 2 simulation tapes that maintain a copy from the tapes of
M1 for a branch of its nondeterministic computations. The
(a) data in tape 3 and 4 contain evolved configurations at the
branch. The function of tape 5 is to generate the address
tape2 tape2 string ωaddress of nondeterministic computational branches
from the length one to length infinite, constantly.
δ2
Let Σ ∗ be an infinite set of all address strings, which
8 2 2
δ1 δ2 [δ2 ]6
δ1 δ2 [δ2 ]2 [δ 2 ]2−α contains all possible branches of nondeterministic
δ1 δ2 computations. Any address string consists of finite kinds
4 1 δ1 δ 2 1
δ1 δ2 of alphabets, which are connected with the number of
[δ1]3
δ1 δ2
α [δ 1 ]α states. By considering Md , its address strings consist of
δ1 δ2
δ1 tape1 tape1 three kinds of alphabets ‘1’, ‘2’, and ‘3’, which come
0 0 1 2 (block) 0 1 2 −α 2 from the subscripts of the three states of M1 (Fig. 1(c)).
0 4 8 (piece) Σ ∗ is countable according to Cantor’s theory since the
(b) (c) number of strings in some certain length is finite and the
union of denumerable countable sets is a countable set.
Fig. 3: From discrete to continuous turing machine: (a) graph of A list of Σ ∗ can be constructed by writing down all
resource transition of M1 in Figure 1(a), (b) graph of resource strings of length zero, length one, length two and so on.
transition of M1k in Figure 1(b) for k=4, and (c) graph of resource The total number of address strings can be expressed as
transition of M1∞ , for k → ∞.
∑+∞ n
n=0 3 . We can easily make a mapping from any string
to n ∈ N. Not all the address strings are valid. For
example, address string ‘212’ is valid, which represents
The number of possible multiple transitions become that the current state is q2 which configuration is
infinite if k tends to infinity. These transitions can be displayed in tape 3 and 4. The next state is q1 that is
denoted by an segment of a line between (2,0) and (0,2) obtained by executing δm , where δm is stored in controller
shown in Fig. 3(c). For example the description of the of Md . Then, the finally state is q2 that is obtained by
following transition executing stored δ1 . The process can be denoted by
[ δm \δ1 ]
[ δ2 ]β
q1 −−−−−> q2 . However, address string ‘233’ is invalid
(2 − α , α ) −−−> (2 − α + β , α − β ), since there not exists a transition function in the process
from state q3 to q3 in Fig. 1(c). The existence of invalid
implies that the configurations (2 − α , α ) and strings is reasonable since they can be considered as null
(2 − α + β , α − β ) are expressed in pieces. Transition addresses. Proved process is just a constructive process,
function δ2 can be executed β times from the we construct Md as follow:

c 2014 NSP
Natural Sciences Publishing Cor.
Appl. Math. Inf. Sci. 8, No. 5, 2405-2416 (2014) / www.naturalspublishing.com/Journals.asp 2409

Algorithm for Md construction:

TM Md :=“On input ω = ω1 , ω2 , ω3 , ω4 , ω5 : M Sd
step1: Initial tape 1 and 2. Check correctness of ω1 , ω2 .
Controller
Tape 3 to 5 stay empty.
step2: Generate a string ωaddress in tape 5 according to
the rule of the increase of string lengths.
step3: Check the validity of ωaddress by checking
transition functional grid in controller. If ωaddress is valid,
go to step4, else go to step2. B00 B B 0 B0 B1#2# #123#
step4: Copy the data in tape 1 and 2 to tape 3 and 4, single tape
respectively. Simulate the transition of states from the first
alphabet of ωaddress to the last one. Go to step2.”  Fig. 5: Structure of single-type deterministic Msd with a snapshot
in Fig. 4.

}
tape1

{
tape3
0 0 Md 2 0
input tape2 Controller tape4 simulation
tapes tapes Lemma 3.Two-type nondeterministic TM M1 has an
0 2
equivalent single-type deterministic TM Msd
tape5
Proof. By Lemma 1 and 2, we have the conclusion. 
1#2#3#11#12#13#21# #33#111# #123#
address tape Lemma 4.Two-type nondeterministic TM M1k has an
k .
equivalent single-type deterministic TM Msd
Fig. 4: Structure of five-type deterministic Md .

Proof. Two-type TM M1k can be expressed by 2k-type TM

M1k with an unit of pieces, which is denoted by Fig. 2. We
assume that the unit of Msd is also expressed in pieces. By
Lemma 2.Five-type deterministic TM Md has an considering the 2k-type structure of M1k shown in Fig. 2,
equivalent single-type deterministic TM Msd . according to Lemma 3, the computations in the two tapes
of any level are equivalent to the computations of Msd .
Proof. The ideal is make mutual simulations between Md Hence, the computations of 2k-type M1k equal the
computations of k Msd machines. Msd k can be constructed
and Msd . The fact that Md simulates Msd is simple. We
only use any one of the tape in Md that can simulate the by combining the resources of the corresponding
single-tape of Msd . The ideal is to show how to simulate positions of k Msd machines , where resources are
Md with Msd . Fig. 5 illustrates that single tape can be used expressed in blocks. Hence, M1k is equivalent to Msdk . 
to represent five tapes. Msd simulates the functions of Md
by storing their information on its single tape, where we Lemma 5.Single-type deterministic TM Msd k has an
use symbol ‘B’ as a delimiter to separate different equivalent single-type deterministic TM Msd .
work-spaces. Tape symbol with a small ‘∧’ above it is
used to mark the position of the head on the tape. These Proof. The fact that Msdk simulates M can be achieved by
sd
new symbols have been added to the tape alphabet. In setting the parameter k = 1. Using Msd to simulate Msd k
other words, Msd contains virtual work-spaces and heads. needs little changes about the inputs of Msd . A new wider
we construct Msd as follow: range delimiter is increased to separate piece work-groups
(a piece work-group equals five original work-spaces,
which is shown in Fig. 5). The k piece work-groups are
Algorithm for Msd construction: combined by copying the initial data of Msd total k times
TM Msd :=“On input ω = Bω1 Bω2 Bω3 B ω4 Bω5 · · · :
to its single type in order of priority. Then, simulation can
step1: Msd puts its tape into the format and the formatted
ˆ ˆ ˆ be achieved by changing the unit of blocks into pieces
tape contains B0̂0BB B B · ˆ· ·
step2: Msd scans its tape from the left first ‘B’ in order to
and the extended transition functions. 
determine the symbols under the virtual heads. Then Msd
update its tape according to Md ’s transition functions. Corollary 1.Two-type nondeterministic TM M1 has an
step3: When Msd moves virtual heads to the right onto equivalent two-type nondeterministic TM M1k
another ‘B’. It means that the corresponding head of Md
has been moved onto the blank portion of the tape. Then, Proof. By lemma 3, 4 and 5, we have the conclusion. 
it continuous the simulation as before.” 
Theorem 1.Single-type deterministic TM Msd ∞ has an

equivalent single-type deterministic TM Msd .

c 2014 NSP
Natural Sciences Publishing Cor.
2410 X. L. Chen et. al. : Continuous Turing Machine: Real Function Computability...

Proof. Msd ∞ can be constructed by M k if k tends infinite. Definition 3.HALTT M = {hM K , ω i | M K is a TM that the
sd
We have its equivalent machine Msd by Lemma 4, where length of its computational history is finite or its
the number of work-spaces tends infinite. Hence, the computational history is convergent}.
mutual simulations between them are enabled. Of course, ∞
the length of the single tape of Msd is necessarily infinite Theorem 2.TM Mhalt has an equivalent single-tape
and Msd is a non-haltable TM in general sense. Note that deterministic TM Mordinary (ordinary means that it is a
machine halt in general sense equals classical TM halt [3, haltable TM).
4]. The concept corresponds to machine halt at the sense Proof. The fact that k tends infinite means that the number
of convergence of computational histories. For example, a of tapes tends infinite in unit piece. If the computational
machine is a non-haltable machine if the number of history of Mhalt ∞ is convergent, its approximate
resource divisions tends infinite. It has not acceptive and computations allows that there exists a reasonable length l
rejective states in general sense since these states cannot of computational history for transition functions at the
be achieved within a finite time. However, machine halt in sense of convergency. The evolution of Mhalt ∞ achieves
convergent meaning of computational histories is to terminal states and the machine outputs approximate
illustrates some important problems in CTMs.  computational results if the number of state transitions
Theorem 1 guarantees that the division of resources comes to l.
does not increase the power of TMs. Importantly, any TM By using Theorem 1, we can find a single-tape
with resource division has an equivalent ordinary TM no deterministic TM Msd and make the length of its tape
matter how many tapes it has and it is deterministic or tends infinite. The single-tape is divided into infinite
nondeterministic. This equivalency necessarily satisfies number of work-groups. Actually, it is just attach the
machine non-haltable in theoretical sense. Actually, the ∞ to the single-tape of M such
longitudinal k tapes in Mhalt sd
equivalency cannot be guaranteed in real-world hardware ∞ .
that Msd is equivalent to Mhalt
environments. For example, infinite division of resources In this case, if the computational history of Mhalt∞ is
is not possible by the bit restriction of computers. convergent, Msd is haltable. The haltable Msd at the sense
Hence, a more general situations should be of convergency is denoted by Mordinary . Hence, Mordinary
considered. For example, we deal with the machines that ∞ have the same computational results.
and Mhalt 
are similar to Msd ∞ , which can machine halt in general
∞ has not an equivalent TM M ∞ .
Theorem 3.Msd
sense. These machines can achieve acceptive or rejective halt
states. The judgments of these states should based on the Proof. By using Theorems 1 and 2, we only need to proof
length of computational histories since any simple that Msd is not equivalent to Mordinary . Msd cannot achieve
computation is infinite if k tends infinite. In other words, acceptive or rejective states since Msd cannot ensure
we consider the approximate calculation of Msd ∞.
machine halt. However, Mordinary has terminal states by
The classical computational history of a TM is usually considering the approximation. Hence, Msd cannot
considered as the configurations of the TM. However, if simulate the terminal states of Mordinary . Actually, Msd
real number is tacked, any configuration may has an can be regarded as the limiting state of the computations
infinite length. Hence, classical computational history of Mordinary . They are not equivalent in the sense of
cannot be applied to describe the dynamic behavior of the convergence of computational history. 
TM with real numbers. Hence, a class of computational Theorems 1 and 2 guarantee the rationality of the
history with respect to state transitions is proposed. extension from TMs to CTMs such that CTMs can be
Definition 1.A string denoted by ωh = i jk · · · is said to be formally defined. In this paper, single-tape deterministic
a computational history if the string composed of the CTMs (ordinary CTMs) and the ordinary CTMs at the
subscripts of the existent states, where every two adjacent sense of convergence in computational history are
characters i and j represents that the machine restores defined. The former are models of computational theory
the computation path from state qi to q j . for real numbers and the latter are theoretical models of
real functional approximation. Other classes of CTMs are
In this case, the computational history of an equivalent to them.
non-haltable TM is a string with infinite length, even if
the TM has finite states. For example, the states and Definition 4.An ordinary CTM is a 7-tuple, (Q, Σ , Γ , δ α ,
transitions shown in Fig. 1(c) can generate computation q1 , qLIM LIM
accept , qre ject ), where
history with infinite length. Obviously, the length of 1. Q is a finite and non-empty set of states.
computational histories is finite if the machines are 2. Σ is the set of input alphabets. It comes from R+ and
haltable since the states of them are necessarily achieve is expressed inSblocks.
acceptive or rejective states. 3. Γ = Σ {♯, ∧ , · · · } is the set of tape alphabets,
where ‘ ♯’ is a symbol of work-space delimiter, ‘ ∧’ is a
Definition 2.A computational history of M k is said to be symbol of virtual head, and ‘ ’ is blank.
convergent if there exists an absolute difference between 4. δ α : Q × Γ → Q × Γ × {L, R, S} is the transition
the output of M k and another output of M k with the length function, where α is transfer quantity that is defined as a
l of computational history, which is less than any given ε . positive rational number.

c 2014 NSP
Natural Sciences Publishing Cor.
Appl. Math. Inf. Sci. 8, No. 5, 2405-2416 (2014) / www.naturalspublishing.com/Journals.asp 2411

configuration of MCT M } and E={(Ci j , δ jα , (Ci j )′ )|

q q
5. q1 , qLIM LIM
accept , qre ject ∈ Q are the initial state, acceptive
state and rejective state, respectively. q q δkα
Ci j , (Ci j )′ ∈ V,Ci −−> Ci′ } are sets of vertexes and edges
If we have a CTM and want to construct a CTM q q
respectively, where (Ci j )′ is a successor of Ci j .
algorithm with the form ‘0’ that is similar to ‘0’ in TM,
we only need to guarantee that Σ of the CTM comes from The CEG for CTM M1∞ (M1∞ comes from M1k ) is
the range (0, 1] (a subset of R+ ). α is defined as a positive shown in Fig. 6. The infinite configurations and state
rational number since any transfer quantity of transition transition processes can be expressed in simple form with
functions is stored in the controllers of CTMs with the finite elements (vertex, edge, and arc). The initial
form of data table. In other words, α is a fixed value and configuration is denoted by C0 = [2 ♯ 0]q1 , where q1
is not a variable. Hence, the continuity of CTMs are means that C0 is just in the state q1 , the number of both
ensured since the data of computations of CTMs are real sides of ‘♯’ represents the total amount of the resources in
numbers. the two tapes. The range of changing of them is R+ .
Definition 5.A haltable CTM at the sense of
computational history convergency is a 7-tuple,
(Q, Σ , Γ , δ α , q1 , qLIM LIM
accept , qre ject ), where
1. Q is a finite and non-empty set of states.
[δ1 ]2
2. Σ is the set of input alphabets. It comes from R+ and [δ 1 ]α [δ1 ]2−α
is expressed inSblocks. [2 # 0 ]q 1 [2 - α # α ]q 2 [0 # 2 ]q 3
3. Γ = Σ {♯, ∧ , · · · } is the set of tape alphabets,
where ‘ ♯’ is a symbol of work-space delimiter, ‘ ∧’ is a [δ 2 ]α [δ2 ]2−α[δ ]2−α
symbol of virtual head, and ‘ ’ is blank. [δ c ]α m
4. δ α : Q × Γ → Q × Γ × {L, R, S} is the transition [δ m ]α [δ c ]2 − α
function, where α is transfer quantity that is defined as a [1 #1 ]q 1 [0 # 2 ]q 2
positive rational number.
5. q1 , qm m
accept , qre ject ∈ Q is the initial state, acceptive [δ 2 ]2
state and rejective state when the length of computational
history is m, respectively. Fig. 6: CEG for CTM M1∞ (M1k , k → ∞, M1k is shown in Fig. 1(b))

The length m follows the hardware restrictions or the

computational precision demands. By considering
Definitions 4 and 5, the machines Msd ∞ is an ordinary CTM However, the moments when different classes of
∞ is a haltable CTM at the sense of convergency.
and Mhalt transition functions are executed are only recorded as the
A configuration of CTMs is similar to that in TMs, vertexes of the CEG. The weights on arcs means transfer
which contains three items: the current state, tape qualities executed according to transition functions. The
contents, and head position. It is represented by the string amount of configurations in any two adjacent vertexes
‘uqv’, where the current state is ‘q’, the current tape with the transfer quality α is infinite. The significance of
contents are ‘u’ and ‘v’ and the current head position is CEGs for CTMs are as follows:
the first symbol of ‘v’. Any symbol in ‘u’ and ‘v’ comes
–Infinite configurations are expressed by finite vertexes.
from R+ .
–The transition function grid in detail from the graph
A configuration is a description of computations of a
can be obtained.
TM at some moment. The number of configurations in a
–The state graphs by folding vertexes and arcs
haltable TM is finite. However, any change of states may
according to superscript of vertexes can be obtained.
contain infinite configurations in a CTM by the influence
of the continuous. Hence, the configurations in TMs does
not suit to represent the evolution of the computations in
CTMs. 3 CTM computable functions
Importantly, state graphs cannot express computations
of CTMs in detail. Hence, the concept of configuration A CTM computes a function by adding the inputs of the
evolution graphs (CEGs) is proposed to better describe function to its tape and halting with the outputs of the
and analyze the configuration change of CTMs. Using function on the same tape at the sense of computational
CEGs to describe some special CTMs without history convergency.
considering the complex CTM construction algorithms is
hence available. Definition 7.A function f : ΣCT∗
M → ΣCT M is a CTM
∗

computable function if some CTMs for every inputs ω can

Definition 6.The CEG of a CTM (MCT M ,C0q1 ) is a halt with just ( f (ω ))∗ on its tape at the sense of
digraph CEG (MCT M ,C0q1 ) =(V, E), where C0q1 means the computational history convergency, where ( f (ω ))∗
initial configuration of MCT M , V={Ci | Ci is a denotes precise values or theoretical values.

c 2014 NSP
Natural Sciences Publishing Cor.
2412 X. L. Chen et. al. : Continuous Turing Machine: Real Function Computability...

Definition 8.A function f : ΣCT∗

M → ΣCT M is a CTM
∗ According to transition directions (the directions of arcs)
handleable function if some CTMs for every inputs can in Fig. 7, the CTM Msubstract can achieve terminal states in
halt with the unique finite length computational history its limits. Although ( f (ω ))∗ as a CTM piecewise function
( f (ω ))m on its tape by approximation, where ( f (ω ))m has two possible values r1 − r2 and 0, it will halt with just
denotes the approximate values of length m of the value of ( f (ω ))∗ on its tape. Hence, function f is CTM
computational history. computable according to Definition 7.

The ‘unique’ in Definition 8 lies on hardware

restrictions or approximate precisions. CTM computable
[ δ 1 ] r2
functions contain classic TM computable functions since
the terminal states of TMs are special situations of the [δ 1 ]α [δ1]r2 −α
limiting forms of CTMs. Discrete characteristics of TMs [r1 # r2 ]q 0 [r1 - α # r2 − α ]qk [r1 - r2 # 0 ]qk
make them to achieve the terminal states in finite times.
Any CTM computable function is necessarily a CTM [δ 1 ] r1 − α δ3
[δ1 ]
r1
handleable function by Definition 7 and 8. In other words,
any CTM computable function can find an approximation
[0 # r2 − r1 ]qk [r2 − r1 #0]q accept

δ2
function in certain hardware restriction. Verse is not true. [0#0]q reject

The purpose to define CTM handleable functions is to

describe the numerical and function approximation Fig. 7: Configuration evolution graph of CT Msubstract .
problems in R+ . For example, by considering iterative
methods for matrix eigenvalues calculation, it may
possible for a computer to get an approximate value but a On the latter, A TM can get its machine coded
precise value. Computers are hardware limited. It can description < T M > and can make computations for
machine halt through the ‘overflow’. However, the coded description of other machines by self reference and
computations of the theoretical machine CTM is platform recursion theorem [3] in classic computation theory.
irrelevant , which may not lead to machine halt since the CTMs can also get coded descriptions by similar
length of computational history may be infinite. methods. A CTM computable function is a class of
Hence, we cannot to say that the computational transformation of CTM coded descriptions. For example,
processes of eigenvalues are CTM computable since the a CTM computable function f takes an input ω =< M >
machine cannot halt in precise value, and we cannot also and returns another coded description < M ′ >, which is
to say it is not CTM computable since computations δ
always access to precise value. Consequently, the kept as f :< M >−>< M ′ >, where M ′ and M recognize
definition of CTM handleable is necessary. If a function is the same language but locate on different configurations.
CTM handleable, its limiting evolved states are CTM Therefore, whether a function is CTM computable can be
computable. The concept is special occurs in CTMs and proved by constructing a CTM to compute it and
not appear in TMs, which will consider in detail in next returning unique coded strings at the sense of
section. computational history convergency.
CTM computable has two meanings: usual arithmetic Coded descriptions can greatly enhance the described
operations on reals and the transformations of machine ability of machines. For example, if we proof that a
coded descriptions. On the former, the inputs can achieve complex function ϕ = ∑ni=1 (R − ri ), R, ∀ri ∈ R+ is CTM
(not approximation) the outputs according to the computable, we only need to proof function
computations of CTMs at the sense of convergency. For δ
f1 :< r1 , r2 >−>< r1 − r2 > is CTM computable (the
example, we construct a CTM that takes an input conclusion is proved by the construction of Msubstruct ),
< r1 , r2 > r1 , r2 ∈ R+ and returns r1 − r2 . Note that the then we proof the function f2 :<
substraction is assumed as true substraction: δ
 Msubstruct1 , Msubstruct2 >−>< Msubstruct1 + Msubstruct2 > is
r1 − r2 r1 ≥ r2 CTM computable (it is easily to construct Madd since
r1 − r2 =
0 r1 < r2 adding and subtraction are similar). Finally, we construct
Let Msubstract be a CTM that can do this work. Its CEG can coded function ϕ :< f1 , f2 >. If there exists a CTM with
be shown in Fig. 7, where α ∈ (0, max(r1 , r2 )) and related the inputs of < f1 , f2 > that can outputs the unique string
transition functions are as follows: by the transition functions of ϕ , function ϕ is CTM
computable.
δ1 : (qk , m, n) = (qk+1 , m − 1, n − 1, R, R) m ∈ (0, r1 ], n ∈ Actually, substraction of real is CTM computable,
(0, r2 ] addition is the opposite of the operation of subtraction,
and multiplication is the result of continuous additions.
δ2 : (qk , 0, n) = (qre ject , 0, 0, R, R) n ∈ (0, r2 − r1 ] Hence, function ϕ is CTM computable. There are many
works that are not CTM computable. Divergent iterative
δ3 : (qk , m, 0) = (qaccept , r1 − r2 , 0, L, L) m ∈ (0, r1 − r2 ] process is an example.

c 2014 NSP
Natural Sciences Publishing Cor.
Appl. Math. Inf. Sci. 8, No. 5, 2405-2416 (2014) / www.naturalspublishing.com/Journals.asp 2413

Algorithm for CTM M f 1 construction:

Mapping reducible at the sense of CTMs is CTM M f 1 :=“On input ω = x :
considered whcih can extend the proof directions of CTM Output x, then accept.” 
computable.
Definition 9.CTM < MA > is mapping reducible to CTM
< MB >, written < MA >≤m < MB >, if there is a CTM We assume such a CTM is existent and is denoted by
computable function f : ΣCT
∗
M → ΣCT M such that for every
∗
M f 2 . The mapping from f2 to f1 is clearly exist by their
ω, geometric meanings such that for any y ∈ [0, π ], we have
< MA′ , ω ′ >input ω ⇔< MB′ , f (ω )′ >input f (ω ) < M ′f 2 , y′ >input y ⇔< M ′f 1 , f2 (y)′ >input f2 (y)
The function f is called the reduction from machine Then, function f2 (y) = cos(y), y ∈ [0, π ] is also CTM
< MA > to < MB >, where the meaning of computable according to Theorem 4.
< MA′ , ω ′ >input ω is that the coded machine < MA >
computes the string ω (ω can be a function) and
generates a new string ω ′ on its tape when < MA > halt at
the sense of computational history convergency. MA′ and 4 Iterative technology based on CTMs
MA locate in different configurations but recognize the
same language. The concept of iterations appears in computational
Actually, Definition 9 is recursive. It is possible to mathematics and the theory of programming. Almost all
construct a CTM to compute or decide a reductional high-level programming languages support iterations. The
function f . For example, < MA > and < MB > in solution of many mathematic approximation problems
Definition 9 may be the machines to computing another needs iterations. However, iterations in R are not TM
reductional functions. If a CTM computable problem is computable. In real-world, computers can handle
reducible to another problem, which is proved if the approximate and iterative computations in R. Hence, TMs
original problem is CTM computable. The following should have an ability to do and do better the works since
theorem can illustrates this ideal. TMs are a class of platform-independent theoretical
models. However, their description abilities do not match
Theorem 4.Let < MA > and < MB > be machines to usual computers. It is unreasonable that TMs are models
compute functions f1 and f2 , denoted as < MA , f1 > and of this class of computations. The proposed CTMs are the
< MA , f2 >, respectively. Function f1 is CTM < MA > expansions of TMs, which can deal with these issues.
computable if < MA >≤m < MB > and < MB > is CTM Therefore, the computable properties of iteration
computable. problems under CTMs are discussed in this section,
which can as a classical example of CTM applications.
Proof. We assume that < MB , f2 > is CTM < M >
The process of iteration in numerical calculations is
computable and function f is the reduction from < MA >
considered. It takes an initial point and an iterative
to < MB >. We construct a new CTM N to compute
formula. Then, let the obtained solution be the next initial
< MA , f1 > such that machine N can halt with just code
point and the process keeps iteration until the fixed point
<< MB′ , f2′ >∗ > on its tape.
is approximated. The computation ends when the adjacent
approximative solutions satisfy a precision requirement.
Algorithm for CTM N construction: This section introduces CTM to compute the iteration
CTM N:=“On input ω =< MA , f1 >: problems of system linear equations, which can make
step1: According to reductional function f , we have better understanding for the significance of CTM
f (ω ) =< MB , f2 >. Run machine < MB , f2 > and output computable compared with considering simple iteration
<< MB′ , f2′ >∗ > if < MB , f2 > machine halt at the sense problems. The key problems to prove iteration
of limit. computability is that the concept of convergence and a
step2: Run machine < M > with the input type of fixed-point theorem can be described by CTMs.
<< MB′ , f2′ >∗ > and output whatever < M > outputs.” On the former, we define n-dimensional real value vector
as the input of CTMs. All these CTMs constitute a space.
Then, the concepts of distance, convergence can be
Obviously, if < MA >≤m < MB > and f2 is CTM < defined. On the latter, we can use CTM recursion theorem
MB > computable, then f1 is MA computable.  to get the definitions.
An example of the proof for CTM computable by using It is necessary to ensure convergence and fixed-point
reduction is proposed. By considering a function f1 (x) = theorem. If the computations of a CTM Miteration are
x, x ∈ [−1, 1], we construct CTM M f 1 as follow: convergence and satisfy the fixed-point theorem, the
Obviously, f1 is a CTM computable function. Let iteration process is CTM Miteration handleable. Its limit
another function be f2 (y) = cos(y), y ∈ [0, π ] which case is CTM Miteration computable. Otherwise, the
computable property is not clear and the work of iteration process is neither Miteration computable nor
constructing a CTM to compute ‘cos’ is also complicated. Miteration handleable.

c 2014 NSP
Natural Sciences Publishing Cor.
2414 X. L. Chen et. al. : Continuous Turing Machine: Real Function Computability...

Theorem 5.(Recursion theorem in CTM) Let T be a CTM Theorem 6.(Fixed-point theorem in CTM) Let CT MRn be
that computes a function t : ΣCT
∗
M × ΣCT M → ΣCT M , then
∗ ∗ a complete metric space, ∃ < MT , ω >∈ CT MRn , CTM <
there exists a CTM U that computes a function u : ΣCT
∗
M→ MT , ω > possesses unique fixed-point such that
ΣCT M for every ω , such that
∗
< M, ω ∗ >=< MT , < M, ω ∗ >>
u(ω ) = t(< U, ω >) The proof is abbreviated since it is similar to that of the
fixed-point theorem in mathematic.
The proof is is abbreviated since it is similar to the This section introduces an example of solving a linear
recursion theorem in TMs [3]. Recursion theorem system equation
indicates that CTMs can output the descriptions of
n
themselves and continuous perform a computation by
these descriptions. Hence, any complex CTM can be ∑ ai j x j = bi , i = 1, 2, . . . , n
j=1
described by recursive coded methods. Theorem 5 is the
basis of the following definitions. by constructing a CTM Miteration to compute it. The
‘ω ’ is used to denote a possible string in a CTM. For equation can be denoted by the form
any x = (x1 , x2 , . . . , xn ) ∈ Rn , there exists a CTM M that n
takes ω = x1 x2 . . . xn as the input, which is written as xi = ∑ (δi j − αi j )x j + bi , i = 1, 2, . . . , n.
< M, ω >. All of these CTMs constitute a new complete j=1
space that is called CT MRn space.
The concept of distance in CT MRn space is similar to If mi j = δi j − αi j is hold, the iterative scheme can is
Euclidean distance. For any strings ω1 = x1 x2 . . . xn and represented as
ω2 = y1 y2 . . . yn , we can obtain < M, ω1 >, < M, ω2 >∈ n
∑ mi j x j
(k+1) (k)
CT MRn . The distance in CT MRn space is defined as xi = + bi , i = 1, 2, . . . , n,
follows: j=1

k = 0, 1, 2, . . .
ρ (< M, ω1 >, < M, ω2 >) =< Msqrt , <
The CTM Miteration can be constructed recursively by
Msum(i=1 to n) , < Msquare , < Msub , xi − yi >>>> . Theorem 5. Firstly, we construct CT Msub (it is similar to
Msubstract ) to compute mi j . Second, CT Msub is constructed
If x, y ∈ R, we have the distance (k)
to compute mi j x j . Finally, CT Msum is introduced to deal
ρ (< M, ω1 >, < M, ω2 >) = (k)
with ∑nj=1 mi j x j + bi . Machine Miteration can be executed
by inputting the coded descriptions of CT Msub , CT Msub ,
< Mabs , < Msub , x − y >> .
CT Msum , and their corresponding inputs on the tapes of
It is easy to prove that they satisfy distance axioms. Miteration . Fig. 8 shows the structure of Miteration .
Definition 10.Let CT MRn be a metric space and Proposition 1.Iterative computation to solve system
< M, ωn >, n = 1, 2, . . . be CTMs in CT MRn space. A linear equations is CTM handleable if it satisfies Theorem
CTM < M, ωn > is convergent to < M, ω >, written as 6 (Fixed-point theorem in CTM).
Proof. If the iterative computation does not satisfy
lim < M, ωn >=< M, ω >, fixed-point theorem, then the computation may divergent
n→∞
within two situations. First, the computational history
if ρ (< M, ωn >, < M, ω >) →< M, 0 > may be finite by embedding a controllable CTM such that
Miteration can machine halt at a given length. However, the
Definition 11.Let CT MRn be a metric space. Function fact violates the meaning of approximation.
< MT , ω >: CT MRn → CT MRn is a CTM with the ability Second, many convergent values lead to different
of contraction (it is similar to the contraction operator in approximate values, which can generate different
mathematic) if ∃ θ, 0 ≤ θ < 1, computational histories. Hence, the iterative computation
∀ < M, ω1 >, < M, ω2 >∈ CT MRn , we have is not CTM handleable by Definition 8. The proposition is
ρ (< MT , < M, ω1 >>, < MT , < M, ω1 >>) ≤ proofed. 
Actually, if the iterative computation satisfies
< Mmul , θ ♯ρ (< M, ω1 >, < M, ω2 >) > fixed-point theorem, related results are necessarily
convergent, uniqueness, and reasonable approximation.
Definition 12.Let CT MRn be a metric space. < MT , ω >: According to algorithm 6, if halting judgment condition ε
CT MRn → CT MRn . CTM < M, ω ∗ > is called the fixed- is given, Miteration is necessarily machine halt and the
point of < MT , ω >, if there exists a CTM < M, ω ∗ >∈ length of the unique computational history is finite.
CT MRn such that According to Definition 8 (CTM handleable), iterative
computation to solve system linear equation is Miteration
< M, ω ∗ >=< MT , < M, ω ∗ >> handleable.

c 2014 NSP
Natural Sciences Publishing Cor.
Appl. Math. Inf. Sci. 8, No. 5, 2405-2416 (2014) / www.naturalspublishing.com/Journals.asp 2415

Algorithm for Miteration construction:

CTM Miteration :=“On input ω = ω1 , ω2 , ω3 , B, ω5 , ω6 , ω7 , Proposition 2.If iterative computation to solve system
B, ω9 (‘B’ means no initial input on the tape): linear equations is CTM handleable and Miteration can
step1: Copy In/Output tape to d-tape3 (save x(k) ), delete machine halt with ε = 0, it is CTM Miteration computable.
checked symbol ‘•’.
step2: Format c-tape4, copy In/Output tape to c-tape4, Proposition 2 is obvious and the proof is abbreviated.
delete checked symbol ‘•’, add checked symbol ‘∧’ on
x1 ( for middle computation).
step3: In c-tape1, run machine < CT Msub > with the data 5 Conclusion and future works
of c-tape1 and d-tape1 by checked symbol ‘∧’ (compute
mi j ). Copy the result < CT Msub ′ > and current x in This paper deals with a class of time-independent
j
c-tape4 to c-tape2 and separate them with symbol ‘♯’. continuous Turing machines (CTMs) by extending from
step4: In c-tape2, run machine < CT Mmul > with the homologous discrete counterpart to continuous. Some
(k) equivalent proofs between CTMs and TMs and the
right information (compute mi j · x j ). Copy the result
< CT Mmul ′ > j to c-tape3, the subscript j comes from x j . constructive algorithms in the sense of limitations are
Initial c-tape2. given. The important concepts such as CTM, CTM
step5: If not move ‘•’ in In/Output tape to blank, go to handleable, CTM computable are defined, which provide
step6, else go to step7. an unified framework of computable theory for real
step6: If ‘∧’ in c-tape4 does not removed to blank, then numbers (real functions) computations.
consider c-tape1, c-tape4, and d-tape2, move ‘∧’ one step If the state graphs of TMs are using in CTMs, the fact
to the right, go to step3, else copy bi in d-tape3 to c-tape3 will generate infinite configurations and the incomplete
by checking ‘•’. In c-tape3, run machine < Msum > with information descriptions. Hence a graph based
the right information (compute representation methods CEGs are proposed, which can
(k+1) (k)
xi = ∑nj=1 mi j x j + bi ). Replace current xi in describe infinite configurations in CTMs by finite
In/Output tape with the result < CT Msum ′ >. Move all elements and can effectively present the behaviors of
current ‘•’ one step to the right. Initial c-tape3. The CTMs.
symbols ‘∧’ in c-tape1 and d-tape1 are moved one step to On the other hand, machine description with recursive
the right and locate on the right first symbol of ‘$’. Go to methods is proposed and a class of machine space CT MRn
step2. is defined to discuss the approximate problems in Rn .
step7: Copy In/Output tape and d-tape3 to c-tape5 and Then, some basic concepts of approximations such as
run < M jud >. If the result < M ′jud > satisfies the convergence of machine computation, distance in
precision requirement ε , output In/Output tape (display machine space CT MRn , and Fixed-point theory in CTM
approximation x(k+1) ), then machine Miteration halt, else etc, are given. Finally iteration approximate computations
initial all tapes except In/Output tape, go to step1.”  in solving system linear equations are proved to be CTM
handleable or computable.
Future works will consider the complexity theory in
Miteration CTMs. Researchers have discovered an elegant scheme to
Controller classify a problem according to there computational
difficult. Hence, we will use CTMs as measure tools to
Tape function analysis algorithmic time complexity and space
[ m ij ] < CTMsub,ω1#ω2 ># a11a12 ⋯a1n $ a21a22 ⋯a2n $⋯$ an1 ⋯ann c-tape 1 complexity.
[ m ij ⋅ x (k )
j ] < CTM mul , < CTM '
sub ># x j > c-tape 2
n
[∑ mij ⋅ x + bi ]
(k )
j < CTMsum , < CTM >1< CTM >2 ⋯< CTM >n b > c-tape 3
'
mul
'
mul
'
mul Acknowledgements
j =1
Middle
computations x1 x2 ⋯ x j ⋯ xn c-tape 4
This research was supported in part by the National
[ δ ij ] δ 11 δ 12 ⋯ δ 1 n $ δ 21 δ 22 ⋯ δ 2 n $ ⋯ $ δ n 1 ⋯ δ nn d-tape 1 Nature Science Foundation of China (No. 61271413),
[b ] i b1 b2 ⋯ bj ⋯ bn d-tape 2 Fundamental Research Funds for Science and Technology
[x ( k +1)
] x1 x2 ⋯ xj ⋯ xn
In/Output Department of Sichuan Province (No. 2013JY0089),
i tape Fundamental Research Funds for Educational Department
(k )
[x i ] x1 x2 ⋯ xj ⋯ xn d-tape 3 of Sichuan Province (No. 08ZA029), key Scientific
Halting Research Fund of Xihua University(No. z1312624), and
judgment < CTM jud , < CTM sub , x ( k +1)
#x k
># ε > c-tape 5
key Scientific Research Fund of Xihua University (No.
Fig. 8: Structure of CTM Miteration . Z1222623).

References
[1] D.P. Yang and A.S. Li, Theory of conputability. Beijing:
Science press, (1975).

c 2014 NSP
Natural Sciences Publishing Cor.
2416 X. L. Chen et. al. : Continuous Turing Machine: Real Function Computability...

[2] Z.Z. Zhao, Introduction to computational science. Beijing: Xiaoliang

Science press, (2002). Chen received B.S. and
[3] M. Sipser, Introduction to the Theory of Computation, M.S. degrees from Xihua
Boston: Course Technology Inc, (2005). University, Chengdu,
[4] J. Hopcroft, R. Motwani, and J. Ullman, Introduction to China, in 2007 and 2010,
Automata Theory, Languages, and Computation. Boston: respectively. He is currently
Addison Wesley, (2006). a Ph.D. student in School
[5] S. Mazur, A. Grzegorczyk, and H. Rasiowa, Computable of Electro-Mechanical
analysis. Pa’n stwowe Wydawn. Naukowe, (1963). Engineering, Xidian
[6] C. Kreitz and K. Weihrauch, Theoretical computer science,
University, Xi’an, China. His
38, 35-53 (1985).
research interests include supervisor control and fault
[7] K. Weihrauch, Computable analysis: An introduction.
Springer Verlag, (2000).
diagnosis of discrete event systems.
[8] A. Edalat, Information and Computation, 120, 32-48 (1995).
[9] A. Edalat, Theoretical Computer Science, 151, 163-193 Wen Song
(1995). is a professor of
[10] A. Edalat, Bulletin of Symbolic Logic, 3, 401-452 (1997). Xihua University since
[11] A. Edalat and P. Snderhauf, Theoretical Computer Science, 2005. senior member
210, 73-98 (1999). of Petri net Special
[12] C. Moore, Theoretical Computer Science, 162, 23-44 Commission. His
(1996). main research interests
[13] U. Doraszelski and M. Satterthwaite, The RAND Journal of are theory of Petri
Economics, 41, 215-243 (2010). nets and mathematical
[14] L. Blum, F. Cucker, and S. Smale, International Journal of logic.
Bifurcation and Chaos, 6, 3-26 (1996).
[15] L. Blum, Complexity and real computation. New York:
Springer Verlag, (1998). Zexia Huang received
[16] P. Hertling, Annals of Pure and Applied Logic, 132, 227-246 B.S. and M.S. degrees from
(2005). Xihua University, Chengdu,
[17] L. Recalde, E. Teruel, and M. Silva, Application and Theory China, in 2007 and 2010,
of Petri Nets, 1639, 107-126 (1999).
respectively. She received
[18] R. David and H. Alla, Discrete, Continuous, and Hybrid
Ph.D in Capital Normal
Petri Nets, Berlin: Springer-Verlag, (2005).
University, Beijing, China.
Her research interests include
function approximation
theory and mathematical
analysis.

Mingwei Tang
is an associate professor with
the School of Mathematics
and Computer Science
Technology of Xihua
University. He received
a Ph.D. degree at the School
of Computer Science and
Engineering from University
of Electronic Science and
Technology of China in 2011. His current research
interests include network security and information
hiding.

c 2014 NSP
Natural Sciences Publishing Cor.