Skip to main content

Full text of "Refining the Proof of Planar Equivalence"

See other formats

CERN-PH-TH /2004-248 

FTPI-MINN-04/33, UMN-TH-2320/04 
December 17, 2004 

Refining the Proof of Planar Equivalence 

A. Armoni M. Shifman G. Veneziano "''^ 

" Department of Physics, Theory Division, CERN 
CH-1211 Geneva 23, Switzerland 

^ Department of Physics, University of Wales Swansea, 
Singleton Park, Swansea, SA2 8PP, UK 

William I. Fine Theoretical Physics Institute, University of Minnesota, 
Minneapohs, MN 55455, USA 

College de France, 11 place M. Berthelot, 75005 Paris, France 


We outline a full non-perturbative proof of planar (large- A^) equiv- 
alence between bosonic correlators in a theory with Majorana fermions 
in the adjoint representation and one with Dirac fermions in the two- 
index (anti)symmetric representation. In a particular case (one fla- 
vor), this reduces to our previous result — planar equivalence be- 
tween super- Yang-Mills theory and a non-super symmetric "orientifold 
field theory." The latter theory becomes one-flavor massless QCD at 
N = 3. 

Recently, we have argued [1] that a bosonic sector of A/" = 1 super- Yang- 
Mills (SYM) theory is equivalent, in the large- planar limit, to a corre- 
sponding sector of a non-supersymmetric Yang-Mills theory with a Dirac 
fermion in the two-index antisymmetric or symmetric representation. We 
will refer to these as the parent and daughter theories, respectively, all being 
endowed with the same gauge group, SU(A^) . The daughter theories represent 
orientifold projections of the parent one, as first discussed in Ref.[2]. As we 
shall see, all our results apply equally well to the antisymmetric (orienti-A), 
and to the symmetric (orienti-S) case. For a detailed review see Ref. [3]. 

For the orienti-A case the daughter theory reduces, at = 3, to one- 
flavor massless QCD. Thus, as an intriguing consequence of planar equiv- 
alence, one can copy, within an 0{1/N) error, non-perturbative quantities 
from SYM theory to the corresponding ones in one-flavor massless QCD [4] . 
In particular, in [5] we have obtained a very encouraging value for the quark 
condensate. Orientifold planar equivalence has further possible applications, 
both in phenomenology [6] and in string theory [7]. 

In Refs. [1, 3] we provided a perturbative proof of the planar equivalence 
and outlined a non-perturbative extension of it. In this paper we present a 
detailed analysis of non-perturbative planar equivalence (including theories 
with Nf flavors, Nf > 1), with emphasis on the assumptions made. In our 
view, this completes the non-perturbative proof, under very mild assump- 

The basic idea behind our proof is the comparison of generating func- 
tionals of appropriate gauge-invariant correlators in the parent and daughter 
theories by, flrst, integrating out their respective fermions in a flxed gauge 
background — a feature which could not be implemented for the orbifold 
projection ^ — and, then, averaging over the gauge fleld itself. In Refs. [1, 3] 
the main emphasis was on the first step. Here we mainly focus on the second. 

Let us define, for a generic Dirac fermion in the representation r, the 
generating functional. 

where .Sym is the Yang-Mills action, Jym is a source which can couple to any 

^Planar equivalence for "orbifold filed theories" was conjectured by M. Strassler in Ref. 
[8] . Orbifold theories always contain a product of gauge factors. 

DA^, D^! e-^Y^l^'-^Y^l exp (i ^+ ^'^ + J*) ^} , 



gauge-invariant operator built from gauge fields and covariant derivatives, 
and the quark (color-singlet) source can contain Lorentz 7 matrices. A 
mass term is a particular case of such quark source. We will always assume 
that a small fermion mass term is introduced for infrared regularization. It 
can be set to zero at the very end. The generating functional VVr(JYM) J^) 
is written in (1), for dcfinitcncss, in Euclidean space. This is not crucial: one 
can carry out all our derivations in Minkowski space as well. 
After the fermions are integrated out we arrive at 


r,[^, J*] = log det {i ^+ 4" r; + j*) . (3) 

For what follows it is convenient to write the effective action F^fA, J*] in the 
world-hne formalism, see [9, 10, 11, 12], as an integral over (super-)Wilson 
loops, namely^ 

1 dT 

X j VxVijj exp | - ^ dr Qi^i'' + ]^VV - ^Jl 

X TrPexp £ dr (^^"i;'^ - \vF;,r^ T;| , (4) 
where the functional integral runs over all closed paths Xh{t), 

Ai^{x) is a fixed gauge background, and 

rpa rpa rpa rpa 

— ^ adjoint > AS > S 

^Strictly speaking Eq.(4) is only valid for space-time independent currents proportional 
to 1 or 75. The extension to non-constant currents can be found in [12] for those 7-matrix 
structures and do not affect our considerations below. As discussed below and at the 
end of the paper, we also expect the same to be true for other 7-matrix structures (see 
[13]) provided suitable identifications are made for the currents in the various theories we 


are the generators for the adjoint, two-index antisymmetric and two-index 
symmetric representations, respectively. Moreover, ip^ij) are superpartners 
to ^^/^(t); they occur due to the fact that we are deahng with spin 1/2 matter. 
Eq. (4) can be written symbohcally as: 

r,[A, J^] = C^{J^)W';{A^) , (5) 


where the summation symbol also stands for the functional integrals appear- 
ing in (4). The expansion coefficients Cq,(J<i,) depend, in general, on the 
representation r through the sources J^. However, for the case at hand, 
the sources Jij, can be matched in the three theories in such a way that, to 
leading order in 1/iV, the Ca(J<i.) become representation- independent. Ex- 
amples of such a matching will be given at the end of this paper, also for the 
case of more than one flavor. With this in mind, we shall assume hereafter 
that representation dependence resides entirely in the (super) Wilson factors 
W^{Aij). Inserting the above result in (2) we arrive at 


where the angle brackets stand for the remaining functional integral (average) 
over the gauge field in the presence of a generic gluon source Jym- 

As usual, taking the logarithm of both sides of (6) picks up the connected 
contributions from the expansion of the right-hand side. 

X {Wr{A,) W^^A,) . . . W,"-iA,))^ , (7) 

where the subscript c stands for connected. In fact a subtlety, representing 
the main thrust of this paper, is related to the issue of "connectedness'' 
which, in turn, is related to the process of averaging over the gluon field the 
multi- Wilson-loop operators appearing in Eq. (7). In Ref. [1, 3] we dealt with 
a single loop (Fig. 1), now we will carefully treat multiloop averaging (see 
Fig. 2 which displays, as a particular example, five loops). 

We will compare two cases: r =adjoint and r =two-index antisymmetric. 
The dimensions of the corresponding representations are 

and iV(iV -l)/2. 


Figure 1: One fermion loop (i.e. log det {i ^+4" T" + J^)) in the gluon 
field background (shown as shaded areas). The gluon fields "inside" and 
"outside" the loop do not communicate with each other at iV oo. This is 
indicated by distinct shadings. Averaging over the gluon field inside the loop 
is independent of averaging outside. Topologically, of course, the distinction 
between inside and outside is immaterial 

respectively. Note, however, that the adjoint fermions are taken to be Majo- 
rana, while the two-index antisymmetric ones are Dirac. As a consequence, 
for r =adjoint, Eq. (4) has to be multiplied by i. Let us also note, in passing, 
that the dimension of the two-index symmetric representation is iV(A^ + l)/2. 

Our statement now is as follows: As —> oo each term in Eq. (7) for 
r —two-index antisymmetric has a corresponding term, with exactly the 
same value, forr =adjoint. The proof is based on well-known trace identities. 

Since W^{A^), for a given loop and given A^, is just the trace of one 
concrete SU(A^) group clement, written in the representation r, the following 
relations hold (see, for example, [14]): 

Wadjoint = TiUTrU^ -1 + {U ^U^)=2{TrUTrW -1) , (10) 

where U (rcsp. represents the same group element in the fundamen- 

tal (resp. antifundamental) representation of SU{N). An important point 

Ws = ^((TrC/)2 + Tr[/2) +([/ ^C/t), 
W^AS = l((Tr[/)2-Tr[/2) +([/ ^C/t), 




Figure 2: Example of fermion multiloops in the gluon field background, at 

oo. The background field outside loops 1 and 4 is the same. The 
background field inside loop 1 and outside loop 2 and 3 is the same. The 
background field inside loop 4 and outside loop 5 is the same. 

here is the occurrence of the {U — > C/^) terms in Eqs. (8), (9). The origin 
of these terms, whose presence is very natural since, for Dirac fermions, a 
representation and its complex conjugate are equivalent, can be explained as 
follows. For each given oriented contour x^{t) in (4) there exists also the 
same contour with the opposite orientation, see Fig. (3). We can therefore 
group Wilson loops pairwise (In QED this contour "pairing" is responsible 
for the Furry theorem.) and thus obtain the additional complex conjugate 
terms. For the real representations, such as the adjoint, this gives simply a 
factor 2, as on the right-hand side of Eq. (10), which cancels however in our 
case against the factor | due to the Majorana condition. 

Consider now, as the simplest example, the term with n — 1 in (7). 
At large N, after integrating over the gauge field, the terms Tr (C/^) and 
1 are subleading in 1/N with respect to the terms of 0{N^) (Trf/)^ or 
TtUTyW. Furthermore, (TrC/) = {TtW). It then follows immediately 
that, at — >• oo: 

^(W^adjoint) ^ {Ws) ^ {Was) ^ {TrUf . (11) 


Figure 3: Two opposite-orientation contours in the sum (4). 

= • O 

Figure 4: The 't Hooft double-hne representation for (M4djoint)- On the right 
we display a convenient graphic shorthand notation that we suggest to use 
in this problem. The black circle corresponds to (TrC/), the white circle 
to (TiW), while the segment connecting them indicates that both circles 
originate from one and the same fermion loop, see Fig. 1. 

Note that this common leading contribution is connected, in spite of the 
fact that, when written in terms of Wilson loops with r ^fundamental (the 
last step in Eq. (11)), it looks disconnected. It is instructive to graphically 
illustrate Eq. (11). To this end we redraw Fig. 1 using the 't Hooft double-hne 
notation, as shown in Fig. 4. 

It is easy to show that, also for higher-order terms in Eq. (7), we can 
drop the subleading contributions in Eqs. (8)-(10), namely, Ty{U^) and 1, 
so that, hereafter, we will deal with the l&rge-N limit, 

Ws = Wj,s^l{{TrUy + {TrU^f), 


^W^adjoint = TrC/TrC/t. 


Equations (12) suggest a convenient graphic notation (see again Figs. 1 
and 4). Associate with every Wadjoint a black and a white circle (related to 
Tr U and toTtU^ , respectively) connected by a short segment (just to show 
that it represents a single Wilson loop) and to either or Was a similar 
drawing with two whites or two black circles (with a factor | each). It is 
easy to see that, as ^ oo, the leading diagrams for a generic contribution 
of the form {WiW2---Wp)c in Eq. (7) are given, in the above notation, by a 
connected tree where p segments are joined through "i-vertices" , i.e. vertices 
that couple any number i {i = 1,2, . . .) of dots. By using trivial properties of 
tree diagrams, and the fact that an i- vertex gives a contribution 0{N^'^~^^), 
we arrive immediately to the conclusion that all tree diagrams are 0{N'^) 
while all loops are suppressed. It is amusing to notice that this large- 
counting resembles closely the one of closed-string amplitudes if one asso- 
ciates with each Tr [/ orTiU^a. closed string and with each shaded region a 
tree-level vertex among the closed-strings that define that region's boundary. 

We recall once more that the subscript c stands for connected. In order 
to ease understanding of this point we show, in Fig. 5, one of contributions in 
the parent theory (five fermion loops, see Fig. 2) in the shorthand notation 
introduced in Fig. 4. This figure represents a certain large-A^ connected 
correlator of five Wilson loops in the parent theory, namely 

{WiW2W^WiW^)c — > (TrC/2) (TrC/j) (TrC/s^TrC/gTrC/i) 

X (Tr Ul Tr U^) (Tr Ul Tr Ul) (Tr U^) . (13) 

A similar contribution in the daughter theory (see Fig. 6) is 

{W{W2W^W4W5)c — > (Tr ul) (Tr t/3) (Tr Ul Tr Tr Ui) 

X (Tr C/i Tr [/]) (Tr [/] Tr Ul) (Tr C/|) . (14) 

To complete the proof of the parent-daughter planar equivalence we now show 
that, to every such tree diagram in the adjoint theory, one can associate a 
corresponding tree shorthand diagram of the S or AS theory, having exactly 
the same value. To see that this is the case one can interchange white and 
black circles at every other vertex along the tree as shown in Fig. 6. After 


Figure 5: A particular connected contribution to an expectation value 
(1^11^21^31^41^5) c in M = 1 SYM theory. Dashed circles indicate averag- 
ing over a connected background field, for instance, the external lines of loop 
1 and loop 4 are averaged over one and the same gluon field. 


Figure 6: The same as in Fig. 5 after interchange "black circle <-> white 
circle" in vertices 5, (1,4), 2 and 3. 


doing so we arrive at a five fermion loop contribution in the daughter (A 
or S) theory. This operation obviously transforms a generic graph of the 
adjoint theory into a corresponding graph of the S or AS theory. The fact 
that one can perform the interchange "black circle white circle" at any 
vertex separately, is rather obvious. Take, for instance, the vertex (1,4) in 
Figs. 5 and 6. The above interchange is nothing but the use of an obvious 

(TV Ul Tr C/4) = (Tr C/i TV C/J) , (15) 

which is a straightforward generalization of the equality (TrU) = (Tr[/^). 
It is easy to show that the above procedure is biunivocal i.e. it associates 
to every graph of the parent theory a graph of the daughter theory, and vice 

Other one-to-one transformations of the graphs of the parent theory into 
those of the daughter theories are also possible. For instance, one can use 
the fact that Wadjoint is real even before averaging over the gluon field, see 
the second line in Eq. (12). This means, in essence, that the loops of the 
parent theory are unoriented (the consequence of the reality of the adjoint 
representation). In addition, it is not necessary to isolate and "pair" together, 
from the very beginning, contours of the opposite orientation, as shown in 
Fig. 3. One can let the sum run over all contours independently. This will 
lead to untangling the terms (TVC/)^ and (Tr U^Y in the first line in Eq. (12). 
They will appear as separate contributions. And, nevertheless, each of these 
separate contributions will have an equal counterpart in the parent theory. 

To conclude, we proved a non-perturbative equivalence between the par- 
tition function of jV = 1 SYM theory and "orientifold field theory" . The 
equivalence holds also in the presence of certain external currents. While 
we do not provide an exact detailed dictionary of the "common sector" of 
the two theories, it is clear from our proof that correlation functions that 
involve powers of Tr as well as Tr FF match at large A^. The bifermion 
operators and ^75 are also in the common sector, and so is the axial 
current ^'7^75\E'. The vector current ^7^^^ and the tensor operator ^cr^i,^ 
do not belong to the common sector, however. ^From our proof it follows 
that the bosonic hadron spectra as well as the domain wall spectra (mass and 
charge) are the same in the two theories. In general, every operator in the 
parent theory that survives the orientifold projection belongs to the common 



Finally, we would like to briefly discuss a rather obvious generalization 
which had been called [15] flavor proliferation. Our proof of planar equiva- 
lence can be readily generalized to the case of many flavors: 

SU(N) Yang-Mills theory with Nf Major ana fermions in the adjoint rep- 
resentation (non-supersymmctric if Nf > 1) is equivalent, in the common 
sector, to SU(N) Yang-Mills theory with Nf Dirac fermions in the two- 
index antisymmetric representation^ . The common sector includes all oper- 
ators built of gluon fields and covariant derivatives, and a subset of bifermion 

A few explanatory remarks are in order here regarding the determination 
of the common sector in the multiflavor case. In selecting bifermion operators 
that belong to the common sector (i.e. the set of sources Jq,) one should 
exercise caution. The pattern of flavor symmetry in these two theories are 
drastically different. In the parent theory with the Majorana fermions, the 
global flavor symmetry is SU(A^/); it is believed to be spontaneously broken 
down to SO(iV/), see e.g. Ref. [16], while in the daughter theory the pattern 
is the same as in QCD, namely 

S\J{Nf)L X S\J{Nf)R ^ S\J{Nf)v . (16) 

This is the reason why many fermion bilinears do not belong to the common 

Operationally, it can be defined as follows. Start from the parent theory 
with Nf Majorana flavors in the adjoint. Write all possible bilinears which do 
not vanish by symmetry. Perform orientifoldization and find the projection 
of the above set to the daughter. Call this " common class" C. Alternatively, 
one can also start from the daughter theory. Write all possible bilinears. 
Examine which ones of them can be elevated to the parent theory. These 
should define the same class C. 

There is a large number of fermion bilinears which do not lie in C in the 
parent theory, and the same is true for the daughter theory. These do not 
belong to the common sector. 

^In the gauge/string correspondence these operators couple to closed string modes 
which are common to type IIB and type O'B string theories. 

^In general, the m ^ and 1/N limits need not (and do not) commute. We 
consider here planar equivalence, namely we take the limit 1/N — > first. 


For fermion bilinears with no derivatives one can readily present a com- 
plete catalogue. With respect to the Lorentz symmetry they form the fol- 
lowing representations: 

(0,0), (0,1) + (1,0). 

The (0, 0) operators in the parent theory are of the type 

TrA^A"^ (17) 

where / and g are the flavor indices. To have a non-vanishing operator, 
we must symmetrize with respect to f,g. Altogether we get Nf{Nf + l)/2 
operators of the type (17) plus Nf{Nf + l)/2 complex conjugate operators. 
Let us denote the orientifold projection of Aj '-^ as 

where i,j are the color indices and rjjX are chiral (left-handed) spinors of 
the daughter theory. Each pair x , V forms one Dirac flavor. It is clear that 
(17) projects onto x^Vg + X^Vf- Dirac notation the projection is onto 

- 1 + 75 

^ / — - — , /, gf-symmetrized , 

and similarly for the complex-conjugate bilinears. The (0, 1)+(1, 0) operators 
in the parent theory are of the type 

TrA'lAjl, (19) 

with symmetrized a, (5 and antysymetrized /, There are Nf{Nf — l)/2 
operators (19) and the same amount of complex conjugate operators. They 
project onto 

^ fa^u^if^ , /, fl'-antisymmetrized, 

in the daughter theory. 

Finally, the operators (1/2, 1/2) are currents. Their classification is dis- 
cussed in sufficient detail in Rcf. [16]. The total number of currents in the 
parent theory is Nj (including one anomalous current), while the total num- 
ber of currents in the daugther theory is 2N'j (namely, Nj vector and Nj 


axial currents) . It is clear that a half of the daughter theory currents have 
no projection onto the parent one. 

What currents can be projected? To answer this question we can use, 
again, the basic projection (18). The Nj currents of the parent theory are 

Ad,,A^, f,g = 1,2,.. .,Nf. (20) 

They are projected as 

Xa,gXf3 + 'nlVl3,f- (21) 

The daughter theory has 2Nj currents, 

XccgXf}, and vlvpj- (22) 

Comparing Eqs. (21) and (22) wc conclude that the minus combination of 
the currents in (22) does not make it to the common sector. 

The analysis becomes even easier if we use the Majorana rather than 
Weyl's representation of the adjoint spinors in the parent theory. In this 
case the nonvanishing (1/2, 1/2) operators in the parent theory are 

A[^7^A^1 , A^^7/.75A^> . (23) 

where curly and square brackets denote symmetrization and antisymmctriza- 
tion, respectively. The total number of the currents (23) is Nj. Performing 
orientifoldization (i.e. replacing A ^ ^ and A ^ 'I') we get Nj currents of 
the daughter theory which belong to the common sector.^ Their charges gen- 
erate an unconventional SU(iVj) subgroup of SU(iVj) l x S\]{Nf)ji, containing 
both vector and axial transformations. 

If wc allow for no-derivative bifermion operators with gluon fields in- 
cluded, the set of allowed operators expands dramatically . We will make no 
attempt at a complete classification in this case. Let us give just one example. 
With a single insertion of the gluon field, one can build combinations 

TrA^F^^A^ e'"' 

with all possible symmetry patterns for 7 , /3 and /, g. 

^For instance, in this way it is easy to check that "extra" Goldstone bosons that exist in 

the daughter theory but arc absent from the parent one [15] do not belong to the common 
sector. They can be produced only in pairs. This contribution is subleading in 


Inclusion of derivatives leads to a further enlargement of the common 
sector. The issue of a complete classification in this case is left for future 

Acknowledgments We would like to thank L. Alvarez- Gaume, R. Casal- 
buoni, L. Del-Debbio, P. Di Vecchia, M. Liischer, R. Musto and A. Ritz for 
very useful discussions. The work of M.S. was supported in part by DOE 
grant DE-FG02-94ER408. A.A. is supported by the PPARC advanced fel- 


[1] A. Armoni, M. Shifman and G. Veneziano, Nucl. Phys. B 667, 170 
(2003) [hep-th/0302163]. 

[2] A. Sagnotti, "Some properties of open string theories," hep-th/9509080; 
A. Sagnotti, Nucl. Phys. Proc. Suppl. 56B, 332 (1997) [hep-th/9702093]. 

[3] A. Armoni, M. Shifman and G. Veneziano, From super- Yang-Mills the- 
ory to QCD: Planar equivalence and its implications, in From Fields 
to Strings: Circumnavigating Theoretical Physics, Eds. M. Shifman, J. 
Wheater and A. Vainshtein, (World Scientific, Singapore, 2004), Vol. 1, 
page 353 [hep-th/0403071]. 

[4] A. Armoni, M. Shifman and G. Veneziano, Phys. Rev. Lett. 91, 191601 
(2003) [hep-th/0307097]. 

[5] A. Armoni, M. Shifman and G. Veneziano, Phys. Lett. B 579, 384 (2004) 

[6] F. Sannino and K. Tuominen, Techniorientifold, hep-ph/0405209. 

[7] P. Di Vecchia, A. Liccardo, R. Marotta and F. PezzeUa, JHEP 0409, 
050 (2004) [hep-th/0407038]. 


[8] M. J. Strasslcr, On methods for extracAing exact non-perturbative results 
in non- super symmetric gauge theories, hcp-th/0104032. 

[9] R. Casalbuoni, J. Gomis and G. Longhi, Nuovo Cim. 24A , 249 (1974); 
R. Casalbuoni, Nuovo. Cim. 33A, 389 (1976); A. Barducci, F. Bordi and 
R. Casalbuoni, Nuovo Cim. 64B, 287 (1981); A. Barducci, R. Casal- 
buoni and L. Lusanna, Nucl. Phys. B 180, 141 (1981). 

[10] L. Brink, R Di Vecchia and R S. Howe, Nucl. Rhys. B 118, 76 (1977). 

[11] M. J. Strassler, Nucl. Rhys. B 385, 145 (1992) [liep-ph/9205205]. 

[12] E. D'Hoker and D. G. Gagne, Nucl. Rhys. B 467, 272 (1996) [hep- 

[13] E. D'Hoker and D. G. Gagne, Nucl. Phys. B 467, 297 (1996) [hep- 

[14] D. J. Gross and H. Ooguri, Phys. Rev. D 58, 106002 (1998) [hep- 

[15] A. Armoni and M. Shifman, Nucl. Phys. B 670, 148 (2003) [hep- 

[16] Y. I. Kogan, M. A. Shifman and M. I. Vysotsky, Sov. J. Nucl. Phys. 42, 
318 (1985) [Yad. Fiz. 42, 504 (1985)].