Data Processing Method and Apparatus Employing Parallel Processing for Solving Systems of Linear Equations

1. A data processing machine for the numerical solution of linear equations represented by Ax=b, where A=(a.sub.i,j) (1.ltoreq.i.ltoreq.n, 1.ltoreq.j.ltoreq.n, and n is an integer larger than 1) is a coefficient matrix of n rows and n columns, x=(x.sub.1, x.sub.2, . . . , x.sub.n).sup.Trans is an unknown vector and b=(b.sub.1, b.sub.2, . . . , b.sub.n).sup.Trans is a known vector, comprising: a memory; a pivot choosing section connected to said memory for choosing pivots by searching said coefficient matrix in a row direction and interchanging elements of said coefficient matrix according to a column-interchange method; a preprocessing section A.sub.1 connected to said memory for calculating Eq. 1 after said pivot choosing section chooses wherein a.sub.i,j.sup.(r) denotes (i,j) element of a coefficient matrix obtained when first to r-th columns are eliminated from A=(a.sub.i,j), b.sub.i.sup.(r) denotes i-th component of a known vector obtained when first to r-th columns are eliminated from A=(a.sub.i,j), k is an integer satisfying 1.ltoreq.k.ltoreq.n-1, wherein if n-{n/k}k=0, {n/k} denotes a maximum integer not exceeding n/k, p is an integer satisfying 0.ltoreq.p.ltoreq.{n/k}-1, and, if n-{n/k}k>0, p is an integer satisfying 0.ltoreq.p.ltoreq.{n/k}, and j is an integer satisfying pk+2.ltoreq.j.ltoreq.n; 2nd to k-th preprocessing sections A.sub.t (t is an integer satisfying 2.ltoreq.t.ltoreq.k) connected to said memory, respectively, each for calculating the following equations: ##EQU7## wherein j is integer satisfying pk+t.ltoreq.j.ltoreq.n and, after said ##EQU8## pivot choosing section chooses calculating equations /a.sub.pk+t,pk+t.sup.(pk+t-1) Eq. 10 /a.sub.pk+t,pk+t.sup.(pk+t-1) Eq. 11 for each element of a (pk+t)-th row of said coefficient matrix and for a (pk+t)-th component of said known vector wherein j is integer satisfying pk+t+1.ltoreq.j.ltoreq.n; an updating section B connected to said memory and comprised of a register set consisting of k registers for registering variables Reg and an arithmetic unit; said arithmetic unit for calculating the following equations: ##EQU9## for i and j satisfying (p+1)k+1.ltoreq.(i, j).ltoreq.n while holding variables Reg in said register set; a main controller G which, if n-{n/k}k=0, instructs said pivot choosing section, said preprocessing sections A.sub.1 to A.sub.k and said updating section B to repeat their operations for every p from zero to {n/k}-2 while incrementing p by one and, further, to execute their operations after incrementing p from p={n/k}-2 to p={n/k}-1, and if n-{n/k}k>0, instructs said pivot choosing section, said preprocessing sections A.sub.1 to A.sub.k and said updating section B to repeat their operations for every p from zero to {n/k}-1 while incrementing p by one, and instructs said pivot choosing section and said preprocessing sections A.sub.1 to A.sub.n-{n/k} to execute their operations after incrementing p by one from p={n/k}-1; a backward substitution section connected to said memory for calculating the following equations, repeatedly while decrementing i from i=n to i=1, thereby obtaining said unknown vector.

2. A data processing machine for the numerical solution of linear equations represented by Ax=b, where A=(a.sub.i,j)1.ltoreq.i.ltoreq.n, 1.ltoreq.j.ltoreq.n, and n is an integer larger than 1) is a coefficient matrix of n rows and n columns, x=(x.sub.1, x.sub.2, . . . , x.sub.n).sup.Trans is an unknown vector and b=(b.sub.1, b.sub.2, . . . , b.sub.n).sup.Trans is a known vector, comprising: a memory; a pivot choosing section connected to said memory for choosing pivots by searching said coefficient matrix in a row direction and interchanging elements of said coefficient matrix according to a column-interchange method; a preprocessing section A.sub.1 connected to said memory for calculating Eq. 1 after said pivot choosing section chooses wherein a.sub.i,j.sup.(r) denotes (i,j) element of a coefficient matrix obtained when first to r-th columns are eliminated from A=(a.sub.i,j), b.sub.i.sup.(r) denotes i-th component of a known vector obtained when first to r-th columns are eliminated from A=(a.sub.i,j), k is an integer satisfying 1.ltoreq.k.ltoreq.n-1, if n-{n/k}k=0, wherein {n/k} denotes a maximum integer not exceeding n/k, p is an integer satisfying 0.ltoreq.p.ltoreq.{n/k}-1, and, if n-{n/k}k>0, p is an integer satisfying 0.ltoreq.p.ltoreq.{n/k}, and j is an integer satisfying pk+2.ltoreq.j.ltoreq.n; 2nd to k-th preprocessing sections A.sub.t (t is an integer satisfying 2.ltoreq.t.ltoreq.k) connected to said memory, respectively, each for calculating the following equations: ##EQU10## wherein j is integer satisfying pk+t.ltoreq.j.ltoreq.n and, after said ##EQU11## pivot choosing section chooses calculating equations /a.sub.pk+t,pk+t.sup.(pk+t-1) Eq. 10 /a.sub.pk+t,pk+t.sup.(pk+t-1) Eq. 11 for each element of a (pk+t)-th row of said coefficient matrix and for a (pk+t)-th component of said known vector wherein j is integer satisfying pk+t+1.ltoreq.j.ltoreq.n; an updating section B connected to said memory and comprised of a register set consisting of k registers for registering variables Reg and an arithmetic unit; said arithmetic unit for calculating the following equations: ##EQU12## using (i, j) elements of an i-th row of said coefficient matrix for i satisfying 1.ltoreq.i.ltoreq.pk or (p+1)k+1.ltoreq.i.ltoreq.n and j satisfying (p+1)k+1.ltoreq.j.ltoreq.n while holding variables Reg in said register set; (k-1) postprocessors C.sub.1 to C.sub.k-1 each connected to said memory for calculating a.sub.pk+t+1,j.sup.(pk+t+1) Eq. 20 a.sub.pk+t+1,j.sup.(pk+t+1) Eq. 21 a.sub.pk+t+1,j.sup.(pk+t+1) Eq. 22 b.sub.pk+t+1.sup.(pk+t+1) Eq. 23 b.sub.pk+t+1.sup.(pk+t+1) Eq. 24 b.sub.pk+t+1.sup.(pk+t+1) Eq. 25 using elements of a (pk+1)-th row to a (pk+t)-th row and (pk+1)-th to (pk+t)-th components of said known vector for j satisfying pk+t+2.ltoreq.j.ltoreq.n; a main controller J which obtains said unknown vector by, if n-{n/k} k=0, instructing said pivot choosing section, said preprocessing sections A.sub.1 to A.sub.k, said updating section B and said postprocessors C.sub.1 to C.sub.k-1 to repeat their linking operations from p=0 to p=[n/k]-1, and, if n-{n/k}k>0, instructing said pivot choosing section, said preprocessing section A.sub.1 to A.sub.k, said updating section B and said postprocessors C.sub.1 to C.sub.k-1 to repeat their linking operations from p=0 to p={n/k}-1 and, thereafter, instructing after setting p={n/k}, said pivot choosing section, said preprocessing sections A.sub.1 to A.sub.n-{n/k}k, said updating section B and said postprocessors C.sub.1 to C.sub.n-{n/k}k to execute linking operations of said pivot choosing section and said preprocessing sections A.sub.1 to A.sub.n-{n/k}k, a processing wherein a number of pivots in said updating section B is set at n-{n/k}k and linking operations of said postprocessing sections C.sub.1 to C.sub.n-{n/k}k.

3. A data processing machine for the numerical solution of linear equations represented by Ax=b, where A=(a.sub.i,j)(1.ltoreq.i.ltoreq.n, 1.ltoreq.j.ltoreq.n, and n is an integer larger than 1) is a coefficient matrix of n rows and n columns, x=(x.sub.1, x.sub.2, . . . , x.sub.n).sup.Trans is an unknown vector and b=(b.sub.1, b.sub.2, . . . , b.sub.n).sup.Trans is a known vector, comprising: a network comprising P nodes .alpha..sub.0 to .alpha..sub.P-1 connected with each other, each node comprising; a memory; a pivot choosing section connected to said memory for choosing pivots by searching said coefficient matrix in a row direction and interchanging elements of said coefficient matrix according to a column-interchange method; a preprocessing section A.sub.1 connected to said memory for calculating Eq. 1 after said pivot choosing section chooses wherein a.sub.i,j.sup.(r) denotes (i,j) element of a coefficient matrix obtained when first to r-th columns are eliminated from A=(a.sub.i,j), b.sub.i.sup.(r) denotes i-th component of a known vector obtained when first to r-th columns are eliminated from A=(a.sub.i,j), k is an integer satisfying 1.ltoreq.k.ltoreq.n-1, if n-{n/k}k=0, wherein {n/k} denotes a maximum integer not exceeding n/k, p is an integer satisfying 0.ltoreq.p.ltoreq.{n/k}-1, and, if n-{n/k}k>0, p is an integer satisfying 0.ltoreq.p.ltoreq.{n/k}, and j is an integer satisfying pk+2.ltoreq.j.ltoreq.n; 2nd to k-th preprocessing sections A.sub.t (t is an integer satisfying 2.ltoreq.t.ltoreq.k) connected to said memory, respectively, each for calculating the following equations: ##EQU13## wherein j is integer satisfying pk+t.ltoreq.j.ltoreq.n and, after said pivot choosing section chooses calculating equations /a.sub.pk+t,pk+t.sup.(pk+t-1) Eq. 10 /a.sub.pk+t,pk+t.sup.(pk+t-1) Eq. 11 for each element of a (pk+t)-th row of said coefficient matrix and for a (pk+t)-th component of said known vector wherein j is integer satisfying pk+t+1.ltoreq.j.ltoreq.n; an updating section B connected to said memory and comprised of a register set consisting of k registers for registering variables Reg and an arithmetic unit; said arithmetic unit for calculating the following equations: ##EQU14## a gateway connected to said memory and provided as a junction for an external apparatus; and a transmitter connected to said memory for transmitting data between said memory and said external apparatus through said gateway; and a main controller G.sub.P for obtaining said unknown vector by executing control of (a) allocating every k rows of said coefficient matrix and every k components of each of said unknown vector and said known vector each of which has a component number equal to a row number of each of every k rows allocated to said memories of said P nodes .alpha..sub.0 to .alpha..sub.P-1 in an order of .alpha..sub.0 to .alpha..sub.P-1 cyclically until all elements of said coefficient matrix and all components of each of said unknown vector and said known vector are completely allocated to said memories of said P nodes .alpha..sub.0 to .alpha..sub.P-1, (b) if n-{n/k}k=0, instructing said P nodes .alpha..sub.0 to .alpha..sub.P-1 to repeat parallel preprocess PA.sub.1 to parallel preprocess PA.sub.k and parallel updating process PB from p=0 to p={n/k}-2 and, further, to execute parallel preprocess PA.sub.1 to parallel preprocess PA.sub.k for p={n/k}-1, and if n-{n/k}k>0, instructing said P nodes .alpha..sub.0 to .alpha..sub.P-1 to repeat parallel preprocess PA.sub.1 to parallel preprocess PA.sub.k and parallel updating process PB from p=0 to p={n/k}-1 and, further, to execute parallel preprocess PA.sub.1 to PA.sub.n-{n/k}k for p={n/k}, and (c) instructing each node to obtain values of said unknown vector using backward substitution and transmitter of each node after completion of steps of (a) and (b); said parallel preprocess PA.sub.1 including calculating Eq. 1 and Eq. 2 (pk+2.ltoreq.j.ltoreq.n) for elements of a (pk+1)-th row of said coefficient matrix and a (pk+1)-th component of said known vector at node .alpha..sub.u (0.ltoreq.u.ltoreq.P-1), after said pivot choosing section of said node chooses a pivot represented by Eq. 3, to which (pk+1)-th to (p+1)k-th rows of said coefficient matrix have been allocated, transmitting results of calculation to respective memories of said nodes other than .alpha..sub.u by said transmitter of .alpha..sub.u, calculating Eq. 15 at each updating section B of said nodes other than .alpha..sub.u for respective elements of allocated rows of said coefficient matrix in parallel to calculation of Eq. 1 and Eq. 2, and calculating Eq. 15 at said updating section B of said node .alpha..sub.u if rows other than (pk+1)-th to (p+1)k-th rows of said coefficient matrix are allocated to said node .alpha..sub.u ; said parallel preprocess PA.sub.t (2.ltoreq.t.ltoreq.k) including calculating Eq. 4, Eq. 5, . . . , Eq. 6, Eq. 7 and Eq. 8 for each element of (pk+t)-th row of said coefficient matrix and (pk+t)-th component of said known vector (pk+t.ltoreq.j.ltoreq.n) at said preprocessing section A.sub.t (2.ltoreq.t.ltoreq.k) of said node .alpha..sub.u, calculating Eq. 10 and Eq. 11, after choice of a pivot represented by Eq. 9, at said pivot choosing section for each element of (pk+t)-th row of said coefficient matrix and (pk+t)-th component of said known vector, transmitting results of calculation of respective memories of nodes other than .alpha..sub.u, and calculating ##EQU15## for allocated rows of said coefficient matrix at respective updating sections B of nodes other than .alpha..sub.u and calculating Eq. 17 at said updating section B of said node .alpha..sub.u if rows other than (pk+1)-th to (p+1)k-th rows of said coefficient matrix have been allocated to said node .alpha..sub.u ; and said parallel updating process PB including calculating Eq. 15 and Eq. 16 for ((p+1)k+1)-th row to n-th row at respective updating sections of all nodes to which ((p+1)k+1)-th row to n-th row have been allocated, respectively, while holding variables Reg in said register set.

4. A data processing machine for the numerical solution of linear equations represented by Ax=b, where A=(a.sub.i,j)(1.ltoreq.i.ltoreq.n, 1.ltoreq.j.ltoreq.n, and n is an integer larger than 1) is a coefficient matrix of n rows and n columns, x=(x.sub.1, x.sub.2, . . . , x.sub.n).sup.Trans is an unknown vector and b=(b.sub.1, b.sub.2, . . . , b.sub.n).sup.Trans is a known vector, comprising: a network comprising P nodes .alpha..sub.0 to .alpha..sub.P-1 connected with each other, each node comprising; a memory; a pivot choosing section connected to said memory for choosing pivots by searching said coefficient matrix in a row direction and interchanging elements of said coefficient matrix according to a column-interchange method; a preprocessing section A.sub.1 connected to said memory for calculating Eq. 1 after said pivot choosing section chooses wherein a.sub.i,j.sup.(r) denotes (i,j) element of a coefficient matrix obtained when first to r-th columns are eliminated from A=(a.sub.i,j), b.sub.i.sup.(r) denotes i-th component of a known vector obtained when first to r-th columns are eliminated from A=(a.sub.i,j), k is an integer satisfying 1.ltoreq.k.ltoreq.n-1, if n-{n/k}k=0, wherein {n/k} denotes a maximum integer not exceeding n/k, p is an integer satisfying 0.ltoreq.p.ltoreq.{n/k}-1, and, if n-{n/k}k>0, p is an integer satisfying 0.ltoreq.p.ltoreq.{n/k}, and j is an integer satisfying pk+2.ltoreq.j.ltoreq.n; 2nd to k-th preprocessing sections A.sub.t (t is an integer satisfying 2.ltoreq.t.ltoreq.k) connected to said memory, respectively, each for calculating the following equations: ##EQU16## wherein j is integer satisfying pk+t.ltoreq.j.ltoreq.n and, after said pivot choosing section chooses calculating equations /a.sub.pk+t,pk+t.sup.(pk+t-1) Eq. 10 /a.sub.pk+t,pk+t.sup.(pk+t-1) Eq. 11 for each element of a (pk+t)-th row of said coefficient matrix and for a (pk+t)-th component of said known vector wherein j is integer satisfying pk+t+1.ltoreq.j.ltoreq.n; an updating section B connected to said memory and comprised of a register set consisting of k registers for registering variables Reg and an arithmetic unit; said arithmetic unit for calculating the following equations: ##EQU17## using (i, j) elements of i-th row of said coefficient matrix for i satisfying 1.ltoreq.i.ltoreq.pk or (p+1)k+1.ltoreq.i.ltoreq.n and j satisfying (p+1)k+1.ltoreq.j.ltoreq.n while holding variables Reg in said register set; (k-1) postprocessors C.sub.1 to C.sub.k-1 each connected to said memory for calculating a.sub.pk+t+1,j.sup.(pk+t+1) Eq. 20 a.sub.pk+t+1,j.sup.(pk+t+1) Eq. 21 a.sub.pk+t+1,j.sup.(pk+t+1) Eq. 22 b.sub.pk+t+1.sup.(pk+t+1) Eq. 23 b.sub.pk+t+1.sup.(pk+t+1) Eq. 24 b.sub.pk+t+1.sup.(pk+t+1) Eq. 25 using elements of (pk+1)-th row to (pk+t)-th row and (pk+1)-th to (pk+t)-th components of said known vector for j satisfying pk+t+2.ltoreq.j.ltoreq.n; a gateway connected to said memory and provided as a junction for an external apparatus; and a transmitter connected to said memory for transmitting data between said memory and said external apparatus through said gateway; and a main controller for obtaining said unknown vector by executing control of (a) allocating every k rows of said coefficient matrix and every k components of each of said unknown vector and said known vector each of which has a component number equal to a row number of each of every k rows allocated to said memories of said P nodes .alpha..sub.0 to .alpha..sub.P-1 in an order of .alpha..sub.0 to .alpha..sub.P-1 cyclically until all elements of said coefficient matrix and all components of each of said unknown vector and said known vector are completely allocated to said memories of said P nodes .alpha..sub.0 to .alpha..sub.P-1, (b) if n-{n/k}k=0, instructing said P nodes .alpha..sub.0 to .alpha..sub.P-1 to repeat parallel preprocess PA.sub.1 to parallel preprocessings PA.sub.2 to PA.sub.k, parallel updating process PB and a post-eliminating processing PC for every p from p=0 to p={n/k}-1 and, if n-{n/k}k>0, instructing said P nodes .alpha..sub.0 to .alpha..sub.P-1 to repeat parallel preprocessing PA.sub.1 to parallel preprocessings PA.sub.2 to PA.sub.k, parallel updating processing PB and post-eliminating processing PC for every p from p=0 to p={n/k}-1 and, further, to execute parallel preprocessings PA.sub.1 to PA.sub.n-{n/k}k, after setting p={n/k}, parallel updating processing PB, after setting a number of pivots equal to n-{n/k}k and post-eliminating processing PC; said parallel preprocess PA.sub.1 including calculating Eq. 1 and Eq. 2 (pk+2.ltoreq.j.ltoreq.n) for elements of a (pk+1)-th row of said coefficient matrix and a (pk+1)-th component of said known vector at node .alpha..sub.u (0.ltoreq.u.ltoreq.P-1), after said pivot choosing section of said node chooses a pivot represented by Eq. 3, to which (pk+1)-th to (p+1)k-th rows of said coefficient matrix have been allocated, transmitting results of calculation to respective memories of said nodes other than .alpha..sub.u by said transmitter of .alpha..sub.u, calculating Eq. 15 at each updating section of said nodes other than .alpha..sub.u for respective elements of allocated rows of said coefficient matrix in parallel to calculation of Eq. 1 and Eq. 2, and calculating Eq. 15 at said updating section of said node .alpha..sub.u if rows other than (pk+1)-th to (p+1)k-th rows of said coefficient matrix are allocated to said node .alpha..sub.u ; said parallel preprocess PA.sub.t (2.ltoreq.t.ltoreq.k) including calculating Eq. 4, Eq. 5, . . . , Eq. 6, Eq. 7 and Eq. 8 for each element of (pk+k)-th row of said coefficient matrix and (pk+t)-th component of said known vector (pk+t.ltoreq.j.ltoreq.n) at said preprocessing section A.sub.t (2.ltoreq.t.ltoreq.k) of said node .alpha..sub.u, calculating Eq. 1 and Eq. 11, after choice of a pivot represented by Eq. 9, at said pivot choosing section for each element of (pk+t)-th row of said coefficient matrix and (pk+t)-th component of said known vector, transmitting results of calculation to respective memories of nodes other than .alpha..sub.u, and calculating ##EQU18## for allocated rows of said coefficient matrix at respective updating sections B of nodes other than .alpha..sub.u and calculating Eq. 17 at said updating section B of said node .alpha..sub.u if rows other than (pk+1)-th to (p+1)k-th rows of said coefficient matrix have been allocated to said node .alpha..sub.u, and said parallel updating process PB including calculating Eq. 15 and Eq. 16 for 1.ltoreq.i.ltoreq.pk, (p+1)k+1.ltoreq.i.ltoreq.n, (p+1)k+1.ltoreq.j.ltoreq.n at respective updating sections B of all nodes to which ((p+1)k+1)-th row to n-th row have been allocated, respectively, while holding variables Reg in said register set; and said post-eliminating processing PC including calculating equations from Eq. 17 to Eq. 25 for each element of (pk+1)-th row to (pk+t)-th row of said coefficient matrix and (pk+1)-th to (pk+t)-th components of said known vector (pk+t+2.ltoreq.j.ltoreq.n, t=1, 2, . . . , k-1).

5. A data processing machine for the numerical solution of linear equations represented by Ax=b, where A=(a.sub.i,j)(1.ltoreq.i.ltoreq.n, 1.ltoreq.j.ltoreq.n, and n is an integer larger than 1) is a coefficient matrix of n rows and n columns, x=(x.sub.1, x.sub.2, . . . , x.sub.n).sup.Trans is an unknown vector and b=(b.sub.1, b.sub.2, . . . , b.sub.n).sup.Trans is a known vector, comprising: P clusters CL.sub.0 to CL.sub.P-1, connected with each other through a network, each comprising P.sub.c element processors PE.sub.1 to PE.sub.Pc connected with each other, a memory, a C gate-way for connecting each cluster with an external apparatus, and a transmitter connected to said memory for transmitting data between each cluster and said external apparatus, each element processor comprising; a memory; a pivot choosing section connected to said memory for choosing pivots by searching said coefficient matrix in a row direction and interchanging elements of said coefficient matrix according to a column-interchange method; a preprocessing section A.sub.1 connected to said memory for calculating Eq. 1 after said pivot choosing section chooses wherein a.sub.i,j.sup.(r) denotes (i,j) element of a coefficient matrix obtained when first to r-th columns are eliminated from A=(a.sub.i,j), b.sub.i.sup.(r) denotes i-th component of a known vector obtained when first to r-th columns are eliminated from A=(a.sub.i,j), k is an integer satisfying 1.ltoreq.k.ltoreq.n-1, if n-{n/k}k=0, wherein {n/k} denotes a maximum integer not exceeding n/k, p is an integer satisfying 0.ltoreq.p.ltoreq.{n/k}-1, and, if n-{n/k}k>0, p is an integer satisfying 0.ltoreq.p.ltoreq.{n/k}, and j is an integer satisfying pk+2.ltoreq.j.ltoreq.n; 2nd to k-th preprocessing sections A.sub.t (t is an integer satisfying 2.ltoreq.t.ltoreq.k) connected to said memory, respectively, each for calculating the following equations: ##EQU19## ##EQU20## wherein j is integer satisfying pk+t.ltoreq.j.ltoreq.n and, after said pivot choosing section chooses calculating equations /a.sub.pk+t,pk+t.sup.(pk+t-1) Eq. 10 /a.sub.pk+t,pk+t.sup.(pk+t-1) Eq. 11 for each element of a (pk+t)-th row of said coefficient matrix and for a (pk+t)-th component of said known vector wherein j is integer satisfying pk+t+1.ltoreq.j.ltoreq.n; an updating section B connected to said memory and comprised of a register set consisting of k registers for registering variables Reg and an arithmetic unit; said arithmetic unit for calculating the following equations: ##EQU21## for i and j satisfying (p+1)k+1.ltoreq.i, j.ltoreq.n while holding variables Reg in said register set; and a main controller G.sub.P for obtaining said unknown vector by executing control of (a) allocating every k rows of said coefficient matrix and every k components of each of said unknown vector and said known vector each of which has a component number equal to a row number of each of every k rows allocated to said memories of said P clusters CL.sub.0 to CL.sub.P-1 in an order of CL.sub.0 to CL.sub.P-1 cyclically until all elements of said coefficient matrix and all components of each of said unknown vector and said known vector are completely allocated to said memories of said P clusters CL.sub.0 to CL.sub.P-1, assuming that each element processor of each cluster is in charge of processing each one of allocated rows of said coefficient matrix and each one of allocated components of said know vector and unknown vector, (b) if n-{n/k}k=0, instructing said P clusters CL.sub.0 to CL.sub.P-1 to repeat parallel preprocessing CLA.sub.1, parallel preprocessings CLA.sub.2 to CLA.sub.k and parallel updating process PB.sub.c from p=0 to p={n/k}-2 and, further, to execute parallel preprocessing CLA.sub.1 to parallel preprocessing CLA.sub.p-1, for p={n/k}-1, and if n-{n/k}k>0, instructing said P nodes .alpha..sub.0 to .alpha..sub.P-1 to repeat parallel preprocessing CLA.sub.k to parallel preprocessing CLA.sub.k and parallel updating process PB.sub.c from p=0 to p={n/k}-1 and, further, to execute parallel preprocesses PA.sub.1 to PA.sub.n-{n/k}k for p={n/k}, and (c) instructing each cluster to obtain values of said unknown vector using backward substitution and transmitter of each element processor of each cluster after completion of steps of (a) and (b); said parallel preprocessing CLA.sub.1 including, assuming a cluster CL.sub.u (o.ltoreq.u.ltoreq.P-1) to which (pk+1)-th to (pk+1)k-th rows have been allocated, allocating each element of (pk+1)-th row of said coefficient matrix and each element of (pk+1)-th component of said known vector to each of said element processors of said cluster CL.sub.u in turn; calculating Eq. 1 and Eq. 2 (pk+2.ltoreq.j.ltoreq.n) at respective preprocessing section A.sub.1 of said element processors of said cluster CL.sub.u simultaneously after said pivot choosing section of each element processor chooses a pivot represented by Eq. 3; transmitting results of calculation to said memories of clusters other than CL.sub.u by said transmitter of said cluster CL.sub.u ; in parallel to the above equation, calculating Eq. 12 at each updating section B of each element processor of said clusters other than CL.sub.u for each of allocated rows of said coefficient matrix; and if rows other than (pk+1)-th to (p+1)k-th rows have been allocated to said cluster CL.sub.u ; said parallel preprocessings CLA.sub.2 to CLA.sub.k including allocating each element of (pk+t)-th row (2.ltoreq.t.ltoreq.k) of said coefficient matrix and each element of (pk+t)-th component of said known vector to each of said element processors of said cluster CL.sub.u in turn; calculating Eq. 4 to Eq. 8 (pk+t.ltoreq.j.ltoreq.n) at each of said preprocessing sections A.sub.2 to A.sub.k of said element processor of said cluster CL.sub.u simultaneously; calculating Eq. 10 to Eq. 11, after choice of a pivot represented by Eq. 9 at said pivot choosing section B of each element processor, at each of said preprocessing sections A.sub.2 to A.sub.k (for pk+t+1.ltoreq.j.ltoreq.n) of each element processor simultaneously; transmitting results of calculation to each of said memories of clusters other than CL.sub.u by said transmitter of CL.sub.u ; in parallel to the above equation, calculating Eq. 17 for each row of said coefficient matrix stored in each of said memories of clusters other than CL.sub.u at each updating section B of said element processors of clusters other than CL.sub.u ; and if rows other than (pk+1)-th to (p+1)k-th rows have been allocated to said cluster CL.sub.u, ##EQU22## calculating Eq. 17 at each updating section B of said element processors in said cluster CL.sub.u ; parallel updating processing B.sub.c including calculating Eq. 15 and Eq. 16 for {(p+1)k+1}-th row to n-th row at respective updating sections of all clusters to which {(p+1)k+1}-th row to n-th row have been allocated, respectively, while holding variables Reg in said register set.

6. A data processing machine for the numerical solution of linear equations represented by Ax=b, where A=(a.sub.i,j)(1.ltoreq.i.ltoreq.n, 1.ltoreq.j.ltoreq.n, and n is an integer larger than 1) is a coefficient matrix of n rows and n columns, x=(x.sub.1, x.sub.2, . . . , x.sub.n).sup.Trans is an unknown vector and b=(b.sub.1, b.sub.2, . . . , b.sub.n).sup.Trans is a known vector, comprising: (A) P clusters CL.sub.0 to CL.sub.P-1, connected with each other through a network, each comprising P.sub.c element processors PE.sub.1 to PE.sub.Pc connected with each other, a memory, a C gate-way for connecting each cluster with an external apparatus, and a transmitter connected to said memory for transmitting data between each cluster and said external apparatus, each element processor comprising; a memory; a pivot choosing section connected to said memory for choosing pivots by searching said coefficient matrix in a row direction and interchanging elements of said coefficient matrix according to a column-interchange method; a preprocessing section A.sub.1 connected to said memory for calculating Eq. 1 after said pivot choosing section chooses wherein a.sub.i,j.sup.(r) denotes (i,j) element of a coefficient matrix obtained when first to r-th columns are eliminated from A=(a.sub.i,j), b.sub.i.sup.(r) denotes i-th component of a known vector obtained when first to r-th columns are eliminated from A=(a.sub.i,j), k is an integer satisfying 1.ltoreq.k.ltoreq.n-1, if n-{n/k}k=0, wherein {n/k} denotes a maximum integer not exceeding n/k, p is an integer satisfying 0.ltoreq.p.ltoreq.{n/k}-1, and, if n-{n/k}k>0, p is an integer satisfying 0.ltoreq.p.ltoreq.{n/k}, and j is an integer satisfying pk+2.ltoreq.j.ltoreq.n; 2nd to k-th preprocessing sections A.sub.t (t is an integer satisfying 2.ltoreq.t.ltoreq.k) connected to said memory, respectively, each for calculating the following equations: ##EQU23## wherein j is integer satisfying pk+t.ltoreq.j.ltoreq.n and, after said pivot choosing section chooses calculating equations /a.sub.pk+t,pk+t.sup.(pk+t-1) Eq. 10 /a.sub.pk+t,pk+t.sup.(pk+t-1) Eq. 11 for each element of a (pk+t)-th row of said coefficient matrix and for a (pk+t)-th component of said known vector wherein j is integer satisfying pk+t+1.ltoreq.j.ltoreq.n; an updating section B connected to said memory and comprised of a register set consisting of k :registers for registering variables Reg and an arithmetic unit; said arithmetic unit for calculating the following equations: ##EQU24## using (i, j) elements of i-th row of said coefficient matrix for i satisfying 1.ltoreq.i.ltoreq.pk or (p+1)k+1.ltoreq.i.ltoreq.n and j satisfying (p+1}k+1.ltoreq.j.ltoreq.n while holding variables Reg in said register set; (k-1) postprocessors C.sub.1 to C.sub.k-1 each connected to said pivot choosing section for calculating a.sub.pk+t+1,j.sup.(pk+t+1) Eq. 20 a.sub.pk+t+1,j.sup.(pk+t+1) Eq. 21 a.sub.pk+t+1,j.sup.(pk+t+1) Eq. 22 b.sub.pk+t+1.sup.(pk+t+1) Eq. 23 b.sub.pk+t+1.sup.(pk+t+1) Eq. 24 b.sub.pk+t+1.sup.(pk+t+1) Eq. 25 using elements of (pk+1)-th row to (pk+t)-th row and (pk+1)-th to (pk+t)-th components of said known vector for j satisfying pk+t+2.ltoreq.j.ltoreq.n; and (B) a main controller for obtaining said unknown vector by executing control of: (a) allocating every k rows of said coefficient matrix and every k components of each of said unknown vector and said known vector each of which has a component number equal to a row number of each of every k rows allocated to said P clusters CL.sub.0 to CL.sub.P-1, in an order of CL.sub.0 to CL.sub.P-1 cyclically until all elements of said coefficient matrix and all components of each of said unknown vector and said known vector are completely allocated to said memories of said P clusters CL.sub.0 to CL.sub.P-1, assuming that each element processor of each cluster is in charge of processing each one of allocated rows of said coefficient matrix and each one of allocated components of said known vector and unknown vector; and (b) if n-{n/k}k=0, instructing said P clusters CL.sub.0 to CL.sub.P-1 to repeat parallel preprocessing PA.sub.1 to parallel preprocessings PA.sub.2 to PA.sub.k and parallel updating processing PB.sub.c ' and post-eliminating processing PC.sub.c for every P from p=0 to p={n/k}-1 and, if n-{n/k}k>0, instructing said P clusters CL.sub.0 to CL.sub.P-1 to repeat parallel preprocessing PA.sub.1 to parallel preprocessing PA.sub.2 to PA.sub.k, parallel updating processing PB.sub.c ' and post-eliminating processing PC.sub.c for every P from p=0 to p={n/k}-1 and, further, to execute parallel preprocessings PA.sub.1 to PA.sub.n-{n/k}k, after setting p={n/k}, parallel updating processing PB.sub.c ', after setting a number of pivots equal to n-{n/k}k and post-eliminating processings PC.sub.1 to PC.sub.n-[n/k]k ; said parallel preprocessing PA.sub.1 including calculating Eq. 1 and Eq. 2 (pk+2.ltoreq.j.ltoreq.n) for elements of (pk+1)-th row of said coefficient matrix and (pk+1)-th component of said known vector at cluster CL.sub.u (0.ltoreq.u.ltoreq.P-1), after said pivot choosing section of said cluster CL.sub.u chooses a pivot represented by Eq. 3, to which (pk+1)-th to (p+1)k-th rows of said coefficient matrix have been allocated, transmitting results of calculation to respective memories of clusters other than CL.sub.u by said transmitter of CL.sub.u, calculating Eq. 15 at each updating section B of said clusters other than CL.sub.u for respective elements of allocated rows of said coefficient matrix in parallel to calculation of Eq. 1 and Eq. 2, and calculating Eq. 15 at said updating section B of said cluster CL.sub.u if rows other than (pk+1)-th to (p+1)k-th rows of said coefficient matrix are allocated to said cluster CL.sub.u ; said parallel preprocessings PA.sub.t (2.ltoreq.t.ltoreq.k) including calculating Eq. 4, Eq. 5, . . . , Eq. 6, Eq. 7 and Eq. 8 for each element of (pk+k)-th row of said coefficient matrix and (pk+t)-th component of said known vector (pk+t.ltoreq.j.ltoreq.n) at said preprocessing section A.sub.t (2.ltoreq.t.ltoreq.k) of said clusters CL.sub.u, calculating Eq. 1 and Eq. 11, after choice of a pivot represented by Eq. 9, at said pivot choosing section for each element of (pk+t)-th row of said coefficient matrix and (pk+t)-th component of said known vector, transmitting results of calculation to respective memories of clusters other than CL.sub.u, and calculating ##EQU25## for allocated rows of said coefficient matrix at respective updating sections B of clusters other than CL.sub.u and calculating Eq. 17 at said updating section B of said cluster CL.sub.u if rows other than (pk+1)-th to (p+1)k-th rows of said coefficient matrix have been allocated to said cluster CL.sub.u, and said parallel updating processing PB.sub.c ' including calculating Eq. 15 and Eq. 16 for 1.ltoreq.i.ltoreq.pk, (p+1)k+1.ltoreq.i .ltoreq.n, (p+1)k+1.ltoreq.j.ltoreq.n at respective updating sections B of all nodes to which ((p+1)k+1)-th row to n-th row have been allocated, respectively, while holding variables Reg in said register set; and said post-eliminating processing PC.sub.c including calculating equations from Eq. 17 to Eq. 25 for each element of (pk+1)-th to (pk+t)-th row of said coefficient matrix and (pk+1)-th to (pk+t)-th components of said known vector (pk+t+2.ltoreq.j.ltoreq.n, t=1, 2, . . . , k-1).

7. Parallel elimination method for numerical solution of linear equations represented by Ax=b wherein A=(a.sub.i,j)(1.ltoreq.i.ltoreq.n, 1.ltoreq.j.ltoreq.n, and n is an integer larger than 1) is a coefficient matrix of n columns and n rows, x=(x.sub.1, x.sub.2, . . . , x.sub.n).sup.Trans is an unknown vector and b=(b.sub.1, b.sub.2, . . . , b.sub.n).sup.Trans is a known vector with use of a parallel computer consisting of first to C-th clusters (C is an integer larger than 1) connected by a network, each cluster consisting of first to P.sub.c -th element processors (P.sub.c is an integer larger than 1) and a memory common to said first to P.sub.c -th element processors, comprising (A) data allocation step for allocating every P.sub.c rows of a coefficient matrix A.sup.(r) =(a.sub.i,j.sup.(r)) and every P.sub.c components of each of known vector b.sup.(r) and unknown vector x.sup.(r), component numbers of said P.sub.c components corresponding to row numbers of said P.sub.c rows one to one, to respective memories of said clusters in turn wherein said coefficient matrix A.sup.(r), known vector b.sup.(r) and unknown vector x.sup.(r) denote coefficient matrix, known vector and unknown vector obtained by eliminating first to r-th columns of the coefficient matrix A-(a.sub.i,j), respectively; repeating said data allocation step until all rows of the coefficient matrix A.sup.(r) and all components of each of the known vector b.sup.(r) and unknown vector x.sup.(r) have been allocated, and, further, allocating said P.sub.c rows of the coefficient matrix A.sup.(r) and P.sub.c components of each of the known and unknown vectors b.sup.(r) and x.sup.(r) to P.sub.c element processors in each cluster; (B) fundamental pre-elimination step for repeating a series of following operations from l=3 to l=P.sub.c ; choosing a pivot represented by Eq. 1 at the first element processor of the corresponding cluster wherein, if n-{n/P.sub.c }P.sub.c >0, wherein {n/P.sub.c } denotes a maximum integer not exceeding n/P.sub.c, k is an integer satisfying 0.ltoreq.k.ltoreq.{n/P.sub.c }, and, if n-{n/P.sub.c }P.sub.c =0, k is an integer satisfying 0.ltoreq.k.ltoreq.{n/P.sub.c }-1; calculating Eq. 2 and Eq. 3 /a.sub.Pck+1,Pck+1.sup.(Pck) Eq. 2 Eq. 3 and transmitting calculation results to respective memories of clusters other than those to which elements processors in charge of (P.sub.c k+2)-th to n-th rows of the coefficient matrix belong and an element processor in charge of a (P.sub.c k+1)-th row belongs, calculating Eq. 4 for the i-th row at the i-th element processor wherein P.sub.c k+2.ltoreq.i.ltoreq.n; a.sub.Pck+1,Pck+2.sup.(Pck+1) Eq. 4 calculating Eq. 5 and Eq. 6 at the second element processor of the cluster -a.sub.Pck+2,Pck+1.sup.(Pck) a.sub.Pck+1,j.sup.(Pck+1) Eq. 5 b.sub.Pck+1.sup.(Pck+1) Eq. 6 choosing a pivot represented by Eq. 7; calculating Eq. 8 and Eq. 9; /a.sub.Pck+2,Pck+2.sup.(Pck+1) Eq. 8 /a.sub.Pck+2,Pck+2.sup.(Pck+1) Eq. 9 transmitting calculation results of Eq. 8 and Eq. 9 to memories of clusters other than those to which element processors in charge of the (P.sub.c k+3)-th to n-th rows of the coefficient matrix belong and an element processor in charge of the (P.sub.c k+2)-th row belongs, calculating Eq. 10 for each of the (P.sub.c k+1)-th to n-th rows at each of element processors in charge of (P.sub.c k+1)-th to n-th rows, respectively; ##EQU26## calculating Eq. 11 and Eq. 12 at the l-th element processor of the cluster; ##EQU27## choosing a pivot represented by Eq. 13 and calculating Eq. 14 and Eq. 15; /a.sub.Pck+l,Pck+l.sup.(Pck+l-1) Eq. 14 /a.sub.Pck+l,Pck+l.sup.(Pck+l-1) Eq. 15 transmitting results of calculation of Eq. 14 and Eq. 15 to memories of clusters other than those to which element processors in charge of (P.sub.c k+l+1)-th row to n-th row belong and an element processor in charge of (P.sub.c k+l)-th row belongs; (C) multi-pivot elimination step of calculating Eq. 16 and Eq. 17 for each of ((k+1)P.sub.c +1)-th to n-th rows at each of elements processors in charge of [(k+1)P.sub.c +1]-th to n-th rows; ##EQU28## (D) repetition elimination judgment step of judging whether or not a series of operation executing said fundamental pre-elimination step in unit of cluster in turn and, thereafter, executing said multi-pivot elimination step have been repeated by {n/P.sub.c } times; (E) remainder elimination step of executing said fundamental pre-elimination step for the ([n/P.sub.c ]P.sub.c +1)-th to n-th rows of the coefficient matrix at element processors in charge of the ([n/P.sub.c ]P.sub.c +1)-th to n-th row, respectively, if n-{n/P.sub.c }P.sub.c >0 when it is judged in said repetition elimination judgement step that said series of operation have been completed; and unknown vector generation step for obtaining said unknown vector using results of steps (A) through (E).

8. The parallel elimination method as claimed in claim 7, said unknown vector generation step comprises (F) fundamental back-substitution step of setting at an element processor in charge of i-th row; (G) fundamental back transmission step of transmitting x.sub.i to the memory of the cluster to which element processors in charge of first to (i-1)-th components of the unknown vector; (H) fundamental back calculation step of calculating Eq. 19 for components in charge at element processors in charge of first to (i-1)-th components; and (I) repetition back procession step of calculating Eq. 20 by said fundamental back-substitution step in an element processor in charge of (n-l+1)-th component of each of the known and unknown vectors; repeating a series of operation executing calculation by said fundamental back calculation at respective element processors in charge of first to (n-l)-th components for l from 1 to (n-1) after transmitting x.sub.n-l+1 to a memory of a cluster to which element processors in charge of first to (n-1)-th components of each of the known and unknown vectors; and finally setting Eq. 21 by said fundamental back-substitution step

9. The parallel elimination method as claimed in claim 7 wherein, upon choosing a pivot, the following steps are executed: searching a non-zero element in an increase direction in the row number from a zero diagonal element when found at an element processor in charge of the row to which said zero diagonal element belongs; announcing the row number of the non-zero element found at the above step to other element processors; interchanging the non-zero element of the coefficient matrix having the row number having been announced with an element having a row number equal to that of said zero diagonal element; and interchanging a component of the unknown vector having a component number equal to the row number having been announced with another component of the unknown vector having a component number equal to the row number of the non-zero diagonal element.

10. The parallel elimination method as claimed in claim 7, wherein, upon choosing a pivot, the following steps are executed: searching an element having a maximum absolute value in an increase direction in the row number from a given diagonal element of the coefficient matrix at an element processor in charge of the row to which said given diagonal element belongs; announcing the row number of the element found at the above searching to element processors other than said element processor; interchanging an element having the row number announced with an element having the row number of said given diagonal number for each row at each element processor in charge of said each row; interchanging a component of the unknown vector having a component number equal to the row number announced with another component of the unknown vector having a component number equal to the row number of the given diagonal element at element processors in charge of the above two components of the unknown vector, respectively.

11. Parallel elimination method for numerical solution of linear equations represented by Ax=b wherein A=(a.sub.i,j)(1.ltoreq.i.ltoreq.n, 1.ltoreq.j.ltoreq.n, and n is an integer larger than 1) is a coefficient matrix of n columns and n rows, x=(x.sub.1, x.sub.2, . . . , x.sub.n).sup.Trans is an unknown vector and b=(b.sub.1, b.sub.2, . . . , b.sub.n).sup.Trans is a known vector with use of a parallel computer consisting of first to C-th clusters (C is an integer larger than 1) connected by a network, each cluster consisting of first to P.sub.c -th element processors (P.sub.c is an integer larger than 1) and a memory common to said first to P.sub.c -th element processors, comprising (A) data allocation step for allocating every P.sub.c rows of a coefficient matrix A.sup.(r) =(a.sub.ij.sup.(r)) and every P.sub.c components of each of known vector b.sup.(r) and unknown vector x.sup.(r), component numbers of said P.sub.c components corresponding to row numbers of said P.sub.c rows one to one, to respective memories of said clusters in turn wherein said coefficient matrix A.sup.(r), known vector b.sup.(r) and unknown vector x.sup.(r) denote coefficient matrix, known vector and unknown vector obtained by eliminating first to r-th columns of the coefficient matrix A-(a.sub.ij), respectively; repeating said data allocation step until all rows of the coefficient matrix A.sup.(r) and all components of each of the known vector b.sup.(r) and unknown vector x.sup.(r) have been allocated, and, further, allocating said P.sub.c rows of the coefficient matrix A.sup.(r) and P.sub.c components of each of the known and unknown vectors b.sup.(r) and x.sup.(r) to P.sub.c element processors in each cluster; (B) fundamental pre-elimination step for repeating a series of following operations from l=3 to l=P.sub.c ; choosing a pivot represented by Eq. 1 at the first element processor of the corresponding cluster wherein, if n-{n/P.sub.c }P.sub.c >0, wherein {n/P.sub.c } denotes a maximum integer not exceeding n/P.sub.c, k is an integer satisfying 0.ltoreq.k.ltoreq.{n/P.sub.c }, and, if n-{n/P.sub.c }P.sub.c =0, k is an integer satisfying 0.ltoreq.k.ltoreq.{n/P.sub.c }-1; calculating Eq. 2 and Eq. 3 /a.sub.Pck+1,Pck+1.sup.(Pck) Eq. 2 Eq. 3 and transmitting calculation results to respective memories of clusters other than those to which element processors in charge of (P.sub.c k+2)-th to n-th rows of the coefficient matrix belong and an element processor in charge of a (P.sub.c k+1)-th row belongs, calculating Eq. 4 for the i-th row at the i-th element processor wherein P.sub.c k+2.ltoreq.i.ltoreq.n; a.sub.Pck+1,Pck+2.sup.(Pck+1) Eq. 4 calculating Eq. 5 and Eq. 6 at the second element processor of the cluster -a.sub.Pck+2,Pck+1.sup.(Pck) a.sub.Pck+1,j.sup.(Pck+1) Eq. 5 b.sub.Pck+1.sup.(Pck+1) Eq. 6 choosing a pivot represented by Eq. 7; calculating Eq. 8 and Eq. 9; /a.sub.Pck+2,Pck+2.sup.(Pck+1) Eq. 8 /a.sub.Pck+2,Pck+2.sup.(Pck+1) Eq. 9 transmitting calculation results of Eq. 8 and Eq. 9 to memories of clusters other than those to which element processors in charge of the (P.sub.c k+3)-th to n-th rows of the coefficient matrix belong and an element processor in charge of the (P.sub.c k+2)-th row belongs, calculating Eq. 10 for each of the (P.sub.c k+1)-th to n-th rows at each of element processors in charge of (P.sub.c k+1)-th to n-th rows, respectively; ##EQU29## calculating Eq. 11 and Eq. 12 at the l-th element processor of the cluster; ##EQU30## choosing a pivot represented by Eq. 13 and calculating Eq. 14 and Eq. 15; /a.sub.Pck+l,Pck+l.sup.(Pck+l-1) Eq. 14 /a.sub.Pck+l,Pck+l.sup.(Pck+l-1) Eq. 15 transmitting results of calculation of Eq. 14 and Eq. 15 to memories of clusters other than those to which element processors in charge of (P.sub.c k+l+1)-th row to n-th row belong and an element processor in charge of (P.sub.c k+l)-th row belongs; (C) multi-pivot elimination step of calculating Eq. 16 and Eq. 17 for each of ((k+1)P.sub.c +1)-th to n-th rows at each of elements processors in charge of [(k+1)P.sub.c +1]-th to n-th rows; ##EQU31## (D) fundamental post-elimination step of calculating Eq. 18 and Eq. 19 at each element processor; a.sub.i+1,j.sup.(r+1) Eq. 18 Eq. 19 (E) post-elimination procession step of repeating the following operation at respective element processors in charge of (P.sub.c k+1)-th to (P.sub.c k+q)-th rows of the coefficient matrix from q=1 to q=P.sub.c -1, said operation executing said fundamental post-elimination step for (P.sub.c k+1)-th to (P.sub.c k+q)-th rows of the coefficient matrix simultaneously after setting l=-w+q+1 in each of Eq. 18 and Eq. 19 for (P.sub.c k+w)-th row (1.ltoreq.w.ltoreq.q); (F) repetition elimination judgment step of judging whether or not a series of operations have been repeated by {n/P.sub.c } times, said series of operations executing said fundamental pre-elimination step for every P.sub.c rows and, then, executing said multi-pivot elimination procession step and post-elimination procession step at each cluster; (G) remainder elimination step of executing, if n-{n/P.sub.c }P.sub.ac >0 at the time when it is judged that said series of operations have been repeated by {n/P.sub.ac } times at repetition elimination step, said fundamental pre-elimination step, said multi-pivot elimination step and post-elimination procession step for remaining ([n/P.sub.c ]P.sub.c +1)-th to n-th rows of the coefficient matrix at respective element processors in charge of them (H) unknown vector generation step for obtaining said unknown vector using results of steps (A) through (G).

12. The parallel elimination method as claimed in claim 11 wherein, upon choosing a pivot, the following steps are executed; searching a non-zero element in an increase direction in the row number from a zero diagonal element when found at an element processor in charge of the row to which said zero diagonal element belongs; announcing the row number of the non-zero element found at the above step to other element processors; interchanging the non-zero element of the coefficient matrix having the row number having been announced with an element having a row number equal to that of said zero diagonal element; and interchanging a component of the unknown vector having a component number equal to the row number having been announced with another component of the unknown vector having a component number equal to the row number of the non-zero diagonal element.

13. The parallel elimination method as claimed in claim 12 wherein, upon choosing a pivot, the following steps are executed; searching an element having a maximum absolute value in an increase direction in the row number from a given diagonal element of the coefficient matrix at an element processor in charge of the row to which said zero diagonal element belongs; announcing the row number of the element found at the above searching to element processors other than said element processor; interchanging an element having the row number announced with an element having a row number of said given diagonal number for each row at each element processor in charge of said each row; interchanging a component of the unknown vector having a component number equal to the row number announced with another component of the unknown vector having a component number equal to the row number of the given diagonal element at element processors in charge of the above two components of the unknown vector, respectively.