Professional Documents
Culture Documents
ISSN No:-2456-2165
v1 1 1
v 1 1
2
v C , v . , v v1
w
v2 . . vw , m . , n . ,
1w
. . .
vw 1 1
w1 m1 n1
1 1
1 1
s m . , s n .
. .
1 1
( s m )1 ( s n )1
Keywords:- Spacer matrix based generalized matrix definiteness/semi-definiteness, this step is followed by a
multiplication, Spacer matrix components, embedding Completely Positive Trace Preserving
dimension, completely positive trace preserving transformation[1,8,11,12,13,15,16,18] (abbreviated as CPTP
transformations, Kraus operators transformation) uniquely determined through the embedding
dimension which is succeeded by a square root
I. INTRODUCTION transformation preserving the positive definiteness/semi-
definiteness. The last block of the transformation is the
Matrix transformations play a very important role in extraction step, it involves transformation back into the
several problems arising in data sciences, natural and social input strictly rectangular complex matrix space with a
sciences and in solving diverse types of engineering possible difference in the total variation, which is termed as
problems. A particularly interesting study in this context is “Unaccounted Variation” and is used to determine the
to transform matrices from one Matrix space to another, or efficiency of the extraction setup, using the Principal axes
to the same Matrix space, preserving one or more attributes associated with the extracted matrix as the reference axes,
of the numerical matrix array. There exist a diverse number the unaccounted variation is split among the available
of linear transformations in literature that can carry out some
of such tasks. In context of problems pertaining to data degrees of freedom (the rank of the extracted matrix Amn )
analysis, a focus is on those transformations that preserve and using this variance contribution and singular values
the total variation in the numerical data matrix (which is associated with the degrees of freedom and the reference
quantified as squared Frobenius norm of the numerical
matrix) or capturing as much variation as possible while
axes as the bases, the final matrix Aˆmn is thereby
providing dimensionality reduction and ease of data constructed, which possess the same Frobenius norm as the
visualization. It is known that the Frobenius norm is input matrix Amn but in general, possibly with a modified
Unitarily Invariant, but under unitary transformations the rank and redistributed variance per degrees of freedom. It is
variance-partitioning structure of the matrix remains also possible that rank is preserved and there is only
unaltered owing to the fact that unitary transformations variance redistribution involved.
change the Principal axes but do not provide an opportunity
for modification of the matrix rank. The paper presents the mathematical framework and
numerical case studies, involving numerical samples from
The research paper presents a mathematical framework (m=2, n=3) and (m=3, n=4) complex matrix spaces for
that allows for matrix transformation on strictly rectangular demonstration of the mathematical methodology. The
complex matrix spaces ( M mn (C ) , where m n ) with article concludes with a discussion on the results of the
preservation of the Frobenius norm and a possibility for numerical studies and insights obtained.
matrix rank readjustment. The methodology presented
operates through three distinct transformation blocks:
mapping from the Input complex matrix space to the higher
dimensional, embedded matrix space using the matrices
generated from Spacer Component matrices[3,4,5,6,7] and
associated symmetric transformation to obtain positive
The spacer component matrices Pn s and Qsm are defined as follows:
m N , n N , We consider the case where m n , the expression for the Embedding dimension, denoted by ‘s’ is given as
follows: s max(m, n) m n , under the imposed condition: m N , n N and m n , we have: s N , s m and
sn
1 0 . . 0 1n 1n . . 1
n
0 1 . . 0 1n 1n . . 1
1 n
Pns I nn ( ) n sn . . . . . . . . . .
n n s
. . . . . . . . . .
0 0 . . 1 1n 1n . . 1
n n s
1 0 . . 0
0 1 . . 0
. . . . .
. . . . .
I mm 0
0 . . 1
Qsm 1
( ) s m m 1m 1
m . . 1
m
m s m 1m 1
1 . .
m m
. . . . .
. . . . .
1 m 1
m . . 1
m s m
Amn M mn (C ) , Amn 0mn , rank( Amn ) = a , ( Amn ) {1 , 2 ,...., a } , such that:
1 2 .... a 0
a
Total Variation = ( Amn F )2 j 2 , we have: Bss M ss (C ) , where Bss Gsm Amn (Wsn ) H
j 1
1 1
ss ( )( BB H ) ss ( )( B H B) ss , therefore we have: trace( ss ) ( Bss F )2 ( Amn F )2 ,
2 2
( ss ) ss , ss is Positive semi-definite or Positive definite
H
1
( )t s
pt 2
1 s
, where: t 1, 2,.., s , therefore we have: pt 0 and p t 1
1 ( ) t 1
2
s
ˆ ss pt [ t ss ss ( t ss ) H ] , where: t ss ss ........ ss (‘t’ times) , i.e., 1ss ss , 2 ss ss ss ,
t 1
3
s s ss ss ss and so on
Therefore, ( ˆ ss )
H
ˆ ss , ˆ ss is Positive semi-definite or Positive definite , trace( ˆ ss ) trace( ss ) ( Amn F )2
ˆ ss M ss (C ) , rank( ˆ ss ) = d , SVD of ˆ ss : ˆ ss Rsd d d ( Rsd ) H , where we have: R H R I d d ,
d d diag (1 , 2 ,...., d ) , 1 2 .... d 0
Bˆss Rsd ( )d d ( Rsd ) H , where: ( )d d diag ( 1 , 2 ,...., d ) , therefore we have:
1 1
2 2
Amn M mn (C ) , rank( Amn ) = b , Amn is defined as follows: Amn (Gsm ) H BˆssWsn
SVD of Amn : Amn U mb bb (Vnb )
H
, where we have: U U V V I bb ,
H H
bb diag ( 1 , 2 ,...., b ) ,
b
1 2 .... b 0 , ( Amn F )2 j 2
j 1
The percentage Efficiency ( ) is defined as follows: [1 ] 100
( Amn F ) ( Amn F ) 2
2
When: ( Amn
F
) 2 ( Amn F
) 2 then Amn itself as the desired output matrix, Amn Aˆmn in this case.
ˆ ( / rank ( Amn ))
b
Construction of Aˆmn :
Case: ( Amn F
) 2 ( Amn F
)2
ˆ j [ j 2 ˆ] Where: j 1, 2,...., b
1
2
Case: ( Amn
F
) 2 ( Amn F
)2
ˆ1 [( Amn F ) 2 ] ˆ 2 ..... ˆ b 0
1
2
,
The numerical calculations are performed using the Scilab 5.4.1 Computational software
rank( X mn ) refers to the ‘ numerical rank ’ of the matrix X mn , which is given as the number of singular values of the
matrix X mn greater than the threshold parameter thd , where :
1
thd (max(m, n)) ( max ( X mn )) (( )52 )
2
m 2, n 3 , therefore s 4
1 1 0
1 0 0 4 3 2 3 2 3
3
0 1
1 2 3 4 3 2 3
0
1
P34 1 0 , Q42 , W43 ( ) ,
3 1 1 6
2 3 2 3 4 3
0 1 2 2
0 1
3 3 3
3 1 1
2 2
1 2 1 2 1 0 0 1
0
1 1 2 1 2 1 0 1 8 4 2 1
G42 ( ) , 44 0 , p1 , p2 , p3 , p4
2 2 1 1 0 1 1 15 15 15 15
0 0 0 1
1 1
Example 1:
1 1 1
A23 , therefore: rank( A23 ) = 1 , ( A23 F ) 6
2
1 1 1
( A23 ) : 6 2.44949
1
3 1
44 ( ) 1 1 1 114 , rank ( 44 ) = 1 , trace( 44 ) = 6
2 1
1 41
1 1 1
Aˆ 23
ˆ
, rank( Â23 ) = 1 , ( A 23 ) 2 6 , ( Aˆ23 ) : ˆ 2.44949
1 1 1 F
Example 2:
1 1 1
A23 , rank( A23 ) = 1 , ( A23 F ) 6
2
1 1 1
( A23 ) : 6 2.44949
Example 3:
1 1 1
A23 , rank( A23 ) = 2 , ( A23 F ) 6
2
1 1 1
( A23 ) : 1 2, 2 2 1.414214
m 3, n 4 , therefore s 5
1 1 0 0
1 0 0 0
4 0
1 0 1 0 0 0 1
0
0 1 0 0
1
1 0 0 1
4
0 0 1
1
P45 , Q53 0 1 , 55 0 1 0
0
0 1 0 1 0 0 0 1 1
4 1 1
0 1
1 3 3 3
0 0 0
0 0 0 1
4 1 1 1
3 3 3
2 3 5 2 5 2 5 2 5
2 5 23 5 2 5 2 5
1
W54 ( ) 2 5 2 5 23 5 2
4 5
5 ,
2 5 2 5 2 5 23 5
2
2 2 2
32 5 3 5 3 5
3 5 32 5 3 5
1
G53 ( ) 3 5 3 5 32 5
3 5 ,
3 3 3
3 3 3
16 8 4 2 1
p1 , p2 , p3 , p4 , p5
31 31 31 31 31
Example 1:
1 1 1 1
A34 1 1 1 1 , rank( A34 ) = 1 , ( A34 F )2 12
1 1 1 1
( A34 ) : 12 3.464102
1
1
12
55 ( ) 1 1 1 1 1 1 15 , rank( 55 ) = 1 , trace( 55 ) = 12
5
1
1
51
1 1 1 1
Aˆ34 1 1 1 1 , rank( Â34 ) = 1 , ( Aˆ34 ) 2 12 , ( Aˆ34 ) : ˆ 3.464102
F
1 1 1 1
Example 2:
1 1 1 1
A34 1 1 1 1 , rank( A34 ) = 2 , ( A34 F )2 12
1 1 1 1
( A34 ) : 1 2 2 2.828427, 2 2
Example 3:
1 1 1 1
A34 1 1 1 1 , rank( A34 ) = 3 , ( A34 F )2 12 , ( A34 ) : 1 2 3 2
1 1 1 1
The numerical case studies involving numerical [1.] Choi, M., D., Completely Positive Linear Maps on
matrices belonging to (m=2, n=3) and (m=3, n=4) complex Complex Matrices, Linear Algebra and its
matrix spaces demonstrate the framework and presents some Applications, 10, p. 285 - 290 (1975)
insights about the examples themselves. We observe that the [2.] Datta, B. N., Numerical Linear Algebra and
matrix of all ones belonging to the (m=2, n=3) matrix space Applications, SIAM
and the (m=3, n=4) matrix space retains its numerical rank [3.] Ghosh, Debopam, A Tryst with Matrices: The Matrix
under the mathematical scheme and the final matrix is equal Shell Model Formalism, 24by7 Publishing, India.
to the input matrix within the tolerances of numerical [4.] Ghosh, Debopam, A Generalized Matrix
precision available in the computational platform used in the Multiplication Scheme based on the Concept of
study. The second example of (m=2, n=3) case and that of Embedding Dimension and Associated Spacer
the (m=3, n=4) case undergo rank enhancement upon Matrices, International Journal of Innovative Science
application of the mathematical scheme and the efficiency of and Research Technology, Volume 6, Issue 1, p.
the extraction is observed to be around 82 % and 85% 1336 - 1343 (2021)
respectively. The third examples for both (m=2, n=3) and [5.] Ghosh, Debopam, The Analytical Expressions for the
(m=3, n=4) cases correspond to the situation of the maximal Spacer Matrices associated with Complex Matrix
possible rank; it can be observed that for both these spaces of order m by n, where m ≠ n, and other
examples there is preservation of the numerical rank under pertinent results , (Article DOI:
application of the mathematical scheme. The extraction 10.13140/RG.2.2.23283.45603) (2021)
efficiency is around 74% for the (m=2, n=3) example and it [6.] Ghosh, Debopam, Construction of an analytical
is around 81% for the (m=3, n=4) example. expression to quantify correlation between vectors
belonging to non-compatible complex coordinates
An observable drawback of the methodology is that it spaces, using the spacer matrix components and
fails to provide an output when the matrix Amn is obtained associated matrices (Article DOI:
10.13140/RG.2.2.17857.68969) (2022)
as zero, i.e. Amn 0mn . In this situation there are no [7.] Ghosh, Debopam, Quantification of Intrinsic Overlap
specific set of Principal axes available from Amn to assign in matrices belonging to strictly rectangular complex
matrix spaces, using the spacer matrix components
the unaccounted variance contribution. and associated matrices (Article DOI:
10.13140/RG.2.2.18405.06886) (2022)
In the case of Overflow, i.e. the situation:
[8.] Kraus, K., States, Effects and Operations:
( Amn F
) 2 ( Amn F
)2 the total variance is mapped Fundamental Notions of Quantum Theory, Springer
onto a single degree of freedom, which is associated with Verlag (1983)
the Principal axes pair corresponding to the largest singular [9.] Macklin, Philip A., Normal matrices for physicists ,
American Journal of Physics, 52, 513(1984)
value of Amn . Therefore in the case of Overflow, the matrix [10.] Meyer, Carl, D. , Matrix Analysis and Applied Linear
Aˆmn is a rank=1 matrix while the rank of the matrix Amn [11.]
Algebra, SIAM
Nielsen, Michael A., Chuang, Isaac L., Quantum
ˆ . In the case of
is possibly greater or equal to that of A Computation and Quantum Information, Tenth
mn
Edition, Cambridge University Press (2010)
Underflow which corresponds to the situation:
[12.] Paris, Matteo G A, The modern tools of Quantum
( Amn F ) ( Amn F ) , the unaccounted variance is
2 2
Mechanics : A tutorial on quantum states,
measurements and operations, arXiv: 1110.6815v2 [
split equally among the available degrees of freedom from
quant-ph ] (2012)
the matrix Amn and thus all the available ‘b’ degrees of [13.] Nakahara, Mikio, and Ohmi, Tetsuo, Quantum
freedom from the matrix Amn is used up in this variance Computing: From Linear Algebra to Physical
Realizations, CRC Press.
readjustment process. [14.] Sakurai, J. J., Modern Quantum Mechanics, Pearson
Education, Inc
The article presents a specific CPTP transformation
[15.] Steeb, Willi-Hans, and Hardy, Yorick, Problems and
generated from the information on the embedding
Solutions in Quantum Computing and Quantum
dimension. The mathematical elements associated with the
Information, World Scientific
Spacer component matrices can also be coupled with
[16.] Stinespring, W., F., Positive Functions on C*-
different sets of Kraus operators [8, 11, 12, 13, 15, 16, 18 ]
algebras, Proceedings of the American Mathematical
and the resulting mathematical schematic can be used to
Society, p. 211 - 216 (1955)
study and understand the functioning and limitations of the
[17.] Strang, Gilbert, Linear Algebra and its Applications,
presented mathematical framework at a much deeper level.
Fourth Edition, Cengage Learning
Follow up studies focused on this aspect of the research
[18.] Sudarshan, E., C., G., Mathews, P., M., Rau,
problem is expected to shed more light and provide a more
Jayaseetha, Stochastic Dynamics of Quantum -
complete understanding of the mathematical framework.
Mechanical Systems, Physical Review, American
Physical Society , 121 (3) , p. 920 - 924 (1961)