2
$\begingroup$

The generators for single qubit Clifford are phase $ P $ and Hadamard $ H $.

The generators for single qutrit Clifford can be found for example here

What is the set of generators for the qutrit Clifford group?

What are two $ 4 \times 4 $ matrices that generate the single qudit Clifford group for $ d=4 $?

$\endgroup$
5
  • 1
    $\begingroup$ The result in the answer of the post you link to extends to all integers (not just primes). So $H$ (Fourier) + $P$ (Phase or shear) are enough to generate the group. Also see arxiv.org/abs/1307.5087 $\endgroup$
    – unknown
    Jul 11 at 2:47
  • $\begingroup$ @unknown You claim that the linked post extends to all integers. That appears to be false. For example plugging $ p=2 $ into that definition of $ S $ gives the matrix $ Z $ which, together with $ H $ gate, does not generate the single qubit Clifford group. Similarly, plugging in $ p=4 $, the matrices $ S $ and $ H $ constructed that way do not generate a finite group which strongly suggests that they do not generate the single 4-dit Clifford group. Indeed this fact can verified by checking that the determinant 1 versions of these matrices do not generate a finite group in $ SU_4 $. $\endgroup$ Jul 11 at 21:21
  • $\begingroup$ For $p=2$, did you use equations 31 and 33?...these should work I think (there's a slight adjustment for p odd or even, eqn 32 vs 33...) $\endgroup$
    – unknown
    Jul 11 at 21:57
  • $\begingroup$ for $p=2,3,4,5$ I get group size = $192,2592,1536,30000$ $\endgroup$
    – unknown
    Jul 11 at 22:27
  • $\begingroup$ Nice! Looks like there was some confusion here. I thought you were saying that the result in the answer to the qutrit post I linked extended to all integers. That would be wrong. But if you were instead saying that the definition in equations 31 & 33 of the arxiv paper you mention extends to all integers then that sounds very promising, I'll definitely take a look! $\endgroup$ Jul 13 at 14:57

1 Answer 1

1
$\begingroup$

The generators for qudit clifford group are give here https://arxiv.org/abs/1911.08162 This is more concise than the paper in the comment and takes care of subtleties better.

Here is a short GAP program that defines these generators for any dimension:

TestA:=function(p)local Ig,Zg,Xg,Hg,Pg,grp,cen,sgrp,scen,C;
 Ig:=IdentityMat(p);
 Zg:=DiagonalMat(List([0..p-1],x->E(p)^x));
 Xg:=IdentityMat(p){\mod([0..p-1]+1,p)+1};
 Hg:=List([0..p-1],x->List([0..p-1],y->E(p)^(x*y)))/ER(p);
 Pg:=DiagonalMat(List([0..p-1],x->E(2*p)^(x*(x-\mod(p,2)))));
 grp:=Group([Hg,Pg,Zg]);cen:=Center(grp);sgrp:=Size(grp);scen:=Size(cen);
 Print(" |G| = ",String(sgrp,-5));Print(" |Cen(G)| = ",String(scen,-5));Print(" |G/Cen(G)| = ",String(sgrp/scen,-5));Print("\n");
 if(p=4)then
 Print("Z=\n");PrintArray(Zg);
 Print("X=\n");PrintArray(Xg);
 Print("H=\n");PrintArray(Hg);
 Print("P=\n");PrintArray(Pg);
 C:=KroneckerProduct([[1,0],[0,1]],[[1,1],[1,-1]]/ER(2));C:=C{[1,3,2,4]}{[1,3,2,4]};
 Zg:=C^-1*Zg*C;
 Xg:=C^-1*Xg*C;
 Hg:=C^-1*Hg*C;
 Pg:=C^-1*Pg*C;
 grp:=Group([Hg,Pg,Zg]);
 Print(" all elements are monomial? ",ForAll(Elements(grp),IsMonomialMatrix),"\n");
 fi;
return grp;end;

Running in GAP gives : (takes seconds)

gap> g2:=TestA(2);;
 |G| = 192   |Cen(G)| = 8     |G/Cen(G)| = 24   
gap> g3:=TestA(3);;
 |G| = 2592  |Cen(G)| = 12    |G/Cen(G)| = 216  
gap> g4:=TestA(4);;
 |G| = 6144  |Cen(G)| = 8     |G/Cen(G)| = 768  
Z=
[ [      1,      0,      0,      0 ],
  [      0,   E(4),      0,      0 ],
  [      0,      0,     -1,      0 ],
  [      0,      0,      0,  -E(4) ] ]
X=
[ [  0,  1,  0,  0 ],
  [  0,  0,  1,  0 ],
  [  0,  0,  0,  1 ],
  [  1,  0,  0,  0 ] ]
H=
[ [        1/2,        1/2,        1/2,        1/2 ],
  [        1/2,   1/2*E(4),       -1/2,  -1/2*E(4) ],
  [        1/2,       -1/2,        1/2,       -1/2 ],
  [        1/2,  -1/2*E(4),       -1/2,   1/2*E(4) ] ]
P=
[ [     1,     0,     0,     0 ],
  [     0,  E(8),     0,     0 ],
  [     0,     0,    -1,     0 ],
  [     0,     0,     0,  E(8) ] ]
 all elements are monomial? true
gap> g5:=TestA(5);;
 |G| = 30000 |Cen(G)| = 10    |G/Cen(G)| = 3000 

As a check, the calculated $G/Cen(G)$ match the results in this paper https://arxiv.org/abs/1810.10259

The GAP code also calculates the generators in an alternate basis. This is based on this https://arxiv.org/abs/1202.3559 where the entire clifford group consists of monomial matrices (normally only the pauli's are). This can be done when $p$ is a square.

As a final note, when $p$ is prime $H$ and $P$ seem enough (so $Z$ is redundant in the generators).

$\endgroup$
5
  • $\begingroup$ Wow your code is beautiful! The single qubit and single qutrit case definitely look good, I recognize those numbers for $ G/Cen(G) $. $\endgroup$ Jul 13 at 15:01
  • 1
    $\begingroup$ Great answer +1 very interesting paper about basis where the entire Clifford group is monomial matrices if $ p $ is square. Interesting to note that Markus Heinrich here quantumcomputing.stackexchange.com/questions/13643/… and especially in chapter 4 of his dissertation uses a different definition of Clifford which differs from the standard one if the qudits are not of prime dimension. For example his single qudit Clifford group for d=4 has order $ 4^3*(4^2-1)= 960 $. I'll probably ask a new question about the reason for this alternative definition. $\endgroup$ Jul 13 at 15:50
  • $\begingroup$ Very interesting that we need to include one of the 4dit Pauli operators (X or Z ) in addition to the $ H $ and $ P $ generators to get the whole Clifford group. The paper you got this from arxiv.org/abs/1911.08162 does a great job. Looking back at the Farinholt paper you mentioned earlier arxiv.org/abs/1307.5087 it seems that the author has totally overlooked the need to include a Pauli operator in the generating set. The group constructed from his equations 31 & 33 does not contain 4dit X or 4dit Z. Indeed the group is not even irreducible. Do you think its worth emailing the guy? $\endgroup$ Jul 13 at 16:17
  • $\begingroup$ @IanGershonTeixeira It's an old paper and it looks like this was caught by others. There are many slightly different ways to define the Pauli and Clifford groups (real/complex, global phases, ...) and subtle differences are harder to resolve. That's why I like to calculate things directly when possible...I'll take a look at the other post a little later. $\endgroup$
    – unknown
    Jul 13 at 16:25
  • $\begingroup$ @IanGershonTeixeira It would be a good question to ask about $Z(p^a)$ vs $GF(p^a)$. Gottesman has a lecture on this pirsa.org/19110138 A few things are different for stabilizer codes but to me $Z(n)$ still seems more natural $\endgroup$
    – unknown
    Jul 14 at 14:25

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.