PartialTranspose

PartialTranspose
Computes the partial transpose of a matrix
Other toolboxes required	none
Related functions	PartialMap PartialTrace
Function category	Superoperators

PartialTranspose is a function that computes the partial transpose of a matrix. The transposition may be taken on any subset of the subsystems on which the matrix acts.

Syntax

• XPT = PartialTranspose(X)
• XPT = PartialTranspose(X,SYS)
• XPT = PartialTranspose(X,SYS,DIM)

Argument descriptions

• X: A matrix to have its partial transpose returned.
• SYS (optional, default 2): A scalar or vector containing the indices of the subsystems on which the transpose is to be applied.
• DIM (optional, by default has all subsystems of equal dimension): A specification of the dimensions of the subsystems that X lives on. DIM can be provided in one of three ways:
  • If DIM is a scalar, it is assumed that X lives on the tensor product of two spaces, the first of which has dimension DIM and the second of which has dimension length(X)/DIM.
  • If $X \in M_{n_1} \otimes \cdots \otimes M_{n_p}$ then DIM should be a row vector containing the dimensions (i.e., DIM = [n_1, ..., n_p]).
  • If the subsystems aren't square (i.e., $X \in M_{m_1, n_1} \otimes \cdots \otimes M_{m_p, n_p}$) then DIM should be a matrix with two rows. The first row of DIM should contain the row dimensions of the subsystems (i.e., the m_i's) and its second row should contain the column dimensions (i.e., the n_i's). In other words, you should set DIM = [m_1, ..., m_p; n_1, ..., n_p].

Examples

A bipartite square matrix

By default, the PartialTranspose function performs the transposition on the second subsystem:

>> X = reshape(1:16,4,4)'
 
X =
 
     1     2     3     4
     5     6     7     8
     9    10    11    12
    13    14    15    16
 
>> PartialTranspose(X)
 
ans =
 
     1     5     3     7
     2     6     4     8
     9    13    11    15
    10    14    12    16

By specifying the SYS argument, you can perform the transposition on the first subsystem instead:

>> PartialTranspose(X,1)
 
ans =
 
     1     2     9    10
     5     6    13    14
     3     4    11    12
     7     8    15    16

Applying the transpose to both the first and second subsystems results in the standard transpose of X:

>> norm(PartialTranspose(X,[1,2]) - X.')
 
ans =
 
     0

Source code

Click on "expand" to the right to view the MATLAB source code for this function.

1. %%  PARTIALTRANSPOSE    Computes the partial transpose of a matrix

2. %   This function has one required argument:

3. %     X: a matrix

4. %

5. %   XPT = PartialTranspose(X) is the partial transpose of the matrix X,

6. %   where it is assumed that the number of rows and columns of X are both

7. %   perfect squares and both subsystems have equal dimension. The transpose

8. %   is applied to the second subsystem.

9. %

10. %   This function has two optional arguments:

11. %     SYS (default 2)

12. %     DIM (default has all subsystems of equal dimension)

13. %

14. %   XPT = PartialTranspose(X,SYS,DIM) gives the partial transpose of the

15. %   matrix X, where the dimensions of the (possibly more than 2)

16. %   subsystems are given by the vector DIM and the subsystems to take the

17. %   partial transpose on are given by the scalar or vector SYS. If X is

18. %   non-square, different row and column dimensions can be specified by

19. %   putting the row dimensions in the first row of DIM and the column

20. %   dimensions in the second row of DIM.

21. %

22. %   URL: http://www.qetlab.com/PartialTranspose

24. %   requires: opt_args.m, PermuteSystems.m

25. %   author: Nathaniel Johnston (nathaniel@njohnston.ca)

26. %   package: QETLAB

27. %   last updated: December 28, 2012

29. function Xpt = PartialTranspose(X,varargin)

31. dX = size(X);

32. sdX = round(sqrt(dX));

34. % set optional argument defaults: sys=2, dim=round(sqrt(length(X)))

35. [sys,dim] = opt_args({ 2, [sdX(1) sdX(1);sdX(2) sdX(2)] },varargin{:});

37. num_sys = length(dim);

39. % allow the user to enter a single number for dim

40. if(num_sys == 1)

41.     dim = [dim,dX(1)/dim];

42.     if abs(dim(2) - round(dim(2))) >= 2*dX(1)*eps

43.         error('PartialTranspose:InvalidDim','If DIM is a scalar, X must be square and DIM must evenly divide length(X); please provide the DIM array containing the dimensions of the subsystems.');

44.     end

45.     dim(2) = round(dim(2));

46.     num_sys = 2;

47. end

49. % allow the user to enter a vector for dim if X is square

50. if(min(size(dim)) == 1)

51.     dim = dim(:).'; % force dim to be a row vector

52.     dim = [dim;dim];

53. end

55. % prepare the partial transposition

56. prod_dimR = prod(dim(1,:));

57. prod_dimC = prod(dim(2,:));

58. sub_prodR = prod(dim(1,sys));

59. sub_prodC = prod(dim(2,sys));

60. sub_sys_vecR = prod_dimR*ones(1,sub_prodR)/sub_prodR;

61. sub_sys_vecC = prod_dimC*ones(1,sub_prodC)/sub_prodC;

62. perm = [sys,setdiff(1:num_sys,sys)];

64. Xpt = PermuteSystems(X,perm,dim); % permute the subsystems so that we just have to do the partial transpose on the first (potentially larger) subsystem

66. if(isnumeric(Xpt)) % if the input is a numeric matrix, perform the partial transpose operation the fastest way we know how

67.     Xpt = cell2mat(mat2cell(Xpt, sub_sys_vecR, sub_sys_vecC).'); % partial transpose on first subsystem

68. else % if the input is not numeric (such as a variable in a semidefinite program), do a slower method that avoids mat2cell (mat2cell doesn't like non-numeric arrays)

69.     Xpt = reshape(permute(reshape(Xpt,[sub_sys_vecR(1),sub_prodR,sub_sys_vecC(1),sub_prodC]),[1,4,3,2]),[prod_dimR,prod_dimC]);

70. end

72. % return the subsystems back to their original positions

73. dim(:,sys) = dim([2,1],sys);

74. dim = dim(:,perm);

75. Xpt = PermuteSystems(Xpt,perm,dim,0,1);