Product Group S-Confinement in SUSY Gauge Theories

We propose a new set of s-confining theories with product gauge groups and no tree-level superpotential, based on a model with one antisymmetric matter field and four flavors of quarks. For each product group we find a set of gauge-invariant operators which satisfy the 't Hooft anomaly matching conditions, and we identify the dynamically generated superpotential which reproduces the classical constraints between operators. Several of these product gauge theories confine without breaking chiral symmetry, even in cases where the classical moduli space is quantum-modified. These results may be useful for composite model building, particularly in cases where small meson operators are absent, or for theories with multiple natural energy scales, and may provide new ways to break supersymmetry dynamically.


Introduction
Experimental evidence so far suggests that the Standard Model gauge group G SM = SU (3) c × SU (2) L × U (1) Y well describes the universe. Attempts to expand the gauge sector beyond G SM must therefore explain why the additional interactions have not yet presented any evidence for their existence.
There are several well-motivated ways to achieve this. The new gauge bosons and matter fields might form a "dark sector" and interact weakly (or not at all) with the particles described by the Standard Model. It is also possible for an extended gauge symmetry to be spontaneously broken to G SM at some high-energy scale which we have not yet probed. In this paper we consider an alternative in which the new dynamics are so strongly coupled that particles charged under the new interactions confine to form neutral bound states, with binding energies at the TeV scale or larger.
We focus on a particular class of N = 1 supersymmetric (susy) gauge theories with product gauge groups of the form SU (N ) 1 ×SU (N ) 2 ×. . .×SU (N ) k . Our model includes one antisymmetric tensor A αβ and four quark fields Q i α charged under SU (N ) 1 , and a series of bifundamental fields (Q i ) α β charged under adjacent gauge groups SU (N ) i × SU (N ) i+1 as shown in Table 1. This theory is an extension of a model, SU (N ) : ( + 4 + N ), which has been shown to confine [1][2][3].
Examples for Feynman diagrams 1 Moose Diagrams (1.1) Table 1: The matter content of the proposed s-confining theory is shown as a moose diagram. Each G i represents a gauged SU (N ) group, while the dashed circles represent the SU (4) L × SU (N ) R family symmetry.
We propose in the language of [4] that this SU (N ) k model is "s-confining:" that is, the theory confines smoothly in the infrared without breaking chiral symmetry, and it generates a nonvanishing superpotential that describes the interactions between the gauge invariant composite fields. Although the N = 1 s-confining theories with a simple gauge group are fully classified [5], our model is to the best of our knowledge the first with a nontrivial product gauge group and no superpotential. 1 This SU (N ) k product group model has two distinctive features which may be useful for modelbuilding. First, there are no small gauge-singlet operators: the number of fields contained in every gauge invariant operator depends on k or N . Second, the various SU (N ) i subgroups generally confine at different scales Λ i , with hierarchies based on the coupling constants g i .
Product groups of this form appear in studies of five-dimensional gauge theories [7][8][9][10]. The model shown in Table 1 can be interpreted as a k-site deconstruction of a 5d susy SU (N ) gauge theory with a Z 2 orbifold. In the 5d theory the chiral fields {A, Q} and Q k exist on opposing 4d branes, while the bifundamental Q i superfields correspond to a single bulk Q field. A natural hierarchy between the Λ i arises if the extra dimension is warped: for example, the model with Λ 1 > . . . > Λ k has A and Q on the ultraviolet brane and Q k on the infrared brane.
In Sections 1.1 and 1.2 we review the basic aspects of confining susy gauge theories. In Section 1.3 we discuss more specific properties of the A + 4Q + N Q model with an SU (N ) gauge group, including the coefficients in its dynamically generated superpotential. These coefficients do not appear in the literature, so we include our derivation in Appendix A. Section 2 contains a detailed discussion of the SU (N ) k product gauge group models and our primary results. In Section 3 we suggest other product group models which may be s-confining, as well as several counter-examples.

Review: Seiberg Dualities
It is generally difficult to analyze the infrared behavior of strongly coupled theories, due to the failure of perturbation theory in this limit. Seiberg, Intriligator and others have made this problem more tractable by exploiting some of the remarkable properties of supersymmetry, allowing some infrared properties of susy gauge theories to be calculated exactly [11,12]. Seiberg's infrared dualities between different phases of gauge theories were central to these developments. We summarize some of the results in this section; a more detailed review is given in [13].
Seiberg found that in SU (N ) gauge groups with F flavors of quarks and antiquarks, also known as susy QCD, the infrared behavior of the F = N and F = N +1 cases can be completely described by a set of gauge invariant operators, M = QQ, B = Q N , and B = Q N . This dual theory has no gauge interactions, so the F = N and F = N + 1 theories are said to confine: every test charge can be "screened" by creating quark-antiquark pairs from the vacuum, and a gauge-invariant Wilson loop obeys a perimeter law. Classically, the gauge invariant operators obey particular constraints, following from the Bose symmetry of the superfields and the definitions of M , B, and B. For F = N + 1, where the indices i and j refer to the family SU (F ) symmetries of the Q and Q. It has been shown [15][16][17] that Eq. (1.2) is modified quantum mechanically: where Λ b is the holomorphic scale Here θ YM is the CP -violating θ-term of the SU (N ) gauge group, g is the gauge coupling, and b = 3N − F = 2N is derived from the β function for the gauge coupling. The quantum-modified constraint Eq.

Review: S-Confinement
In the F = N + 1 case, the classical constraint equations are not modified. Instead, they are enforced by a dynamically generated superpotential [18].
which has M = B = B = 0 as a solution to the equations of motion. This vacuum corresponds to confinement without chiral symmetry breaking, which we refer to as s-confinement. More precisely, a theory is s-confining if [4]: • All infrared degrees of freedom are gauge invariant composite fields; • The infrared physics is described by a smooth effective theory, which is valid everywhere on the moduli space (including the origin); • There is a dynamically generated superpotential.
For the effective theory to be smooth, there should be no gauge invariant order parameter that can distinguish the Higgs and confined phases of the theory. The infrared degrees of freedom must also satisfy the anomaly matching conditions. Generally, the dynamically generated superpotential is determined up to an overall factor based on symmetry arguments, and by matching its equations of motion to the classical constraints. Its dependence on the holomorphic scale Λ b can be found either on dimensional grounds, or by requiring that W d is neutral under the anomalous U (1) symmetry.
The requirement that a superpotential is dynamically generated adds a powerful constraint on the matter content of any s-confining theory. An N = 1 susy theory with f massless matter superfields has a classical family symmetry of rank f +1 including the R symmetry, but the G 2 U (1) anomaly removes one linear combination of the U (1) family symmetries. This allows us to define a U (1) R symmetry such that exactly one of the matter superfields φ i has R charge, q i , with all other fields neutral. Using the normalization in which the gauginos have R charge +1, cancellation of the G 2 U (1) R anomaly requires that where µ j and µ G are the Dynkin indices of the matter fields φ j and the gluinos, respectively, with the normalization µ( ) = 1. For the dynamically generated superpotential to have R charge +2 under any of the possible anomaly-free R symmetries, it must have the form The matter content must therefore satisfy the index constraint of Csaki et al. [4]: In [5] this index constraint is used to find all N = 1 s-confining theories with one gauge group and no tree-level superpotential. Both F = N + 1 susy QCD and the A + 4Q + N Q model are included.
In theories with a product gauge group this constraint is relaxed: the number of fields exceeds the rank of the family symmetry, and it is no longer possible to identify a unique R symmetry for each field.

SU (N ) with antisymmetric tensor
Properties of the +F +(N +F −4) model have been studied by several authors [1,2,19,20]. In the F = 2 case there is a superpotential generated by a one-instanton effect; for F = 3 the theory confines, with a quantum-deformed moduli space that induces dynamical symmetry breaking; and for F = 4, the theory is s-confining. The quantum modified constraints have been derived in [19] for F = 3, but the classical constraints for the A + 4Q + N Q model do not appear in the literature. We derive the relative coefficients of the dynamically generated superpotential in Appendix A, and quote the results in this section.

Infrared operators:
In the A + 4Q + N Q model, the set of gauge invariant operators changes based on whether N is even or odd. This is due to the representation: if N = 2m is even, then the gauge invariants include the antisymmetrized products (A m ), (A m−1 Q 2 ), and (A m−2 Q 4 ), while for odd N = 2m + 1 the gauge invariants include (A m Q) and (A m−1 Q 3 ).
Below, we define the simplest gauge invariant operators for the N = 2m and N = 2m + 1 models. Both cases include the operators (QQ), (AQ 2 ), and (Q N ):  For even N ≥ 4, we also add the gauge invariants whereas for odd N ≥ 5 we include The numeric coefficients absorb the combinatoric factors from the tensors, with the convention 123...N = +1. In general, we reserve the indices a, b, α, β for gauge groups, and use the indices i, j to refer to family symmetries. Superscripts and subscripts are chosen for visual clarity, and do not signify any particular group representation. It is useful to classify the {U, V, W, X, Y, Z} fields as "baryons" and the J and K fields as "mesons," to separate the operators which scale with N from those which are independent of N . The transformation properties of these operators under the family symmetries are shown in Table 2. There is a continuous family of equivalent U (1) A × U (1) B × U (1) R charge assignments, but the choice shown in Table 2 is particularly convenient.
For N = 4, the theory contains four flavors of Q + Q. This value of N is unique in that both m A Pf A and m i j Q α i Q j α are gauge-invariant mass terms: if these masses are large compared to Λ, then every field can be integrated out above the confinement scale. This special case is discussed in Section 3.1. For N = 3 the and representations are equivalent, and the A + 4Q + 3Q model reduces to susy QCD with F = 4.
As discussed in Section 1.1, the form of the dynamically generated superpotential is determined by the representations of the matter fields. For the A + 4Q + N Q model, The sum includes all possible gauge-invariant contractions of the group indices, with some relative coefficients: (1.20) Both F odd = {J, K, X, Y, Z} and F even = {J, K, U, V, W, Z} satisfy the t' Hooft anomaly matching conditions for the mixed SU (4) 2 U (1) and SU (N ) 2 U (1) anomalies, the various U (1) 3 anomalies, and the mixed U (1) gravitational anomalies, for all U (1) symmetries listed in Table 2 except for U (1) 1 . The G 2 1 U (1) 1 anomaly breaks U (1) 1 explicitly at the scale Λ 1 , so it is not a symmetry of the infrared theory.
Dynamically generated superpotential: The number of infrared operators, dim F, is larger than the dimension of the moduli space, dim implying for both cases that the number of constraints is For odd N , the eight constraints are while for even N The index i = 1 . . . 4 refers to the SU (4) family symmetry. By taking partial derivatives of Eq. (1.19) and Eq. (1.20) and matching the equations of motion to the classical constraints, one can determine the relative coefficient of each term in the dynamically generated superpotential. The results appear below: As in susy QCD, the overall factor α cannot be determined by symmetry arguments. In principle, it is possible to add heavy quark masses and integrate out two flavors of (QQ) so as to match the F = 2 model, whose superpotential can be calculated from a one-instanton calculation analogous to F = N − 1 susy QCD. In our present study we do not perform this calculation. It is useful, however, to consider the phases of α and Λ b . As defined in Eq. (1.4), the phase of Λ b is determined by the CP -violating θ YM parameter. The phase of α is also unknown: however, because W d is charged under an unbroken U (1) R symmetry, it can be rotated by a phase without affecting the Lagrangian L ∼ dθ 2 W , so as to make α real.

Product Group Extension for an S-Confining Theory
Our interest in the product group model of Table 1 is motivated by an observation from the G 1 ×G 2 case, in which the family symmetry G 2 = SU (N ) R of the Q is weakly gauged. In the confined phase of G 1 , there are three types of operators charged under G 2 : one antisymmetric K = , four quarks J = , and N antiquarks Q 2 = . Remarkably, this is identical to the original s-confining model.
The model described in Section 1.3 can be extended indefinitely by adding more gauge groups G i and bifundamental matter Q i . As long as Λ 1 > Λ 2 . . . > Λ i > Λ i+1 , confinement under G i always produces mesons charged as + 4 under G i+1 . This is the model shown in Table 1, where the gauge group is G 1 × . . . × G k . In this section we devote our attention to the question: is this SU (N ) k theory s-confining, or is s-confinement disrupted by the product group?
There are two obvious ways in which the K + 4J + N Q 2 "k=2" model differs from the original ("k=1") s-confining theory. First, in the k = 1 model there is no tree-level superpotential, but in the k = 2 case there is a superpotential from G 1 confinement that may alter how {K, J, P } confine under G 2 . Luckily, inspection of the classical constraints shows that K, J, and Q 2 may be varied freely, as long as the baryon products {U Z, V Z, WZ} or {XZ, Y Z} vary in accordance with Eqs. (1.24) and (1.25). The second main difference is that under G 2 , the classical moduli space is modified quantum mechanically. For the k ≥ 2 theory to be s-confining, we must determine whether or not the origin remains on the moduli space.
Of the existing literature regarding susy product groups, the work of Chang and Georgi [10] on SU (N ) k extensions to F = N susy QCD is particularly useful to our present study. Our method also has some similarities to deconfinement [1,21], particularly in Section 3 when we consider Sp(2N ) groups.

Infrared Operators
To understand the infrared behavior of the theory, we develop in this section a basis of gauge invariant operators which describe the moduli space and obey anomaly matching conditions. Then in Sections 2.2 and 2.3, we find the dynamically generated superpotential and perform some consistency checks.
Let us define a basis for the anomalous U (1) charges, U (1) j=1...k , such that the anomaly coefficient A(G 2 i U (1) j ) is zero if and only if i = j, as shown in Table 3. Each U (1) i is explicitly broken at a scale associated with Λ i , so that the approximate UV symmetry is broken to ..k charges are also shown. The alternating (±) factors in the Q k charges depend on whether k is odd or even: the upper choice corresponds to odd k.
From Table 3, it is clear that combinations of the form are neutral under all of the symmetries, including the spurious U (1) i . Therefore, the dynamically generated superpotential has the form for some powers p i = 0, 1, . . . for each i = 2, 3, . . . k. Any such superpotential has an R charge of +2 under all of the possible U (1) R symmetries. Before we can find the individual terms that appear in W d , it is necessary to understand the equations of motion between the infrared operators.
To find a set of gauge invariant operators in the far infrared, let us consider the ordered case Λ 1 Λ 2 . . . Λ k . As discussed in Section 1.3, G 1 confinement produces the operators anomaly coefficients that must be calculated. This is the benefit of the strategically-defined U (1) i charges shown in Table 3: the fields {Q, A, Q 1 } are neutral under U (1) 2 . . . U (1) k , and all of these anomaly matching conditions are trivially satisfied. The fields J 1 and K 1 transform similarly to Q and A under the non-Abelian symmetries, but their U (1) B charges are different, as shown in Table 4. Table 4: Transformation properties of the composite fields in the confined phase of G 1 , in the limit where G 2 × . . . × G k is weakly gauged. The composite fields U , V , and W exist only if N is even; if N is odd, then they are replaced by X and Y .
At the scale Λ 2 < Λ 1 , the G 2 fields confine to form the following G 1 × G 2 singlets: The fields J 2 and K 2 transform under G 3 as and respectively. It is convenient to define the shorthand notation At scales below Λ 2 and above Λ 3 , the intermediate degrees of freedom are {J 2 , K 2 , B 1 , B 2 , Z 1 , Z 2 , Q 3 , . . . , Q k }. This set of fields satisfies the anomaly matching conditions for SU (4) It is straightforward to continue this procedure until all groups including G k have confined, using the following recursive operator definition: This definition can be applied to i = 1 as well if we define J 0 = Q and K 0 = A. Below the scale Λ k , all of the gauge groups have confined, and the approximate U (1) i=1...k symmetries are broken to discrete Z N groups. The charges under the remaining continuous family symmetries are shown in Table 5.
The U (1) B charges of J k and K k are positive if k is odd, and negative if k is even.
It must be shown that the basis of infrared operators is large enough to cover the moduli space. For the SU (N ) k gauge group with fields {A, Q, Q 1 , . . . , Q k }, the dimension of the moduli space is implying that there are 8k complex constraints. By rearranging Eq. (2.7) as follows, we can find 8(k − 1) of the constraint equations: for i = 2, 3 . . . k. Similarly, The eight remaining constraints are provided by 14) It is possible that these classical constraints may be quantum-modified.
Reduced operator basis: The classical constraints for B i>1 are mildly problematic, because Eqs. (2.11) and (2.12) imply that these operators are redundant: that is, they can be written as products from a smaller operator basis, {B 1 , Z 1 , Z 2 , . . . Z k }, and are therefore not independent degrees of freedom. Excitations of the B i fields above the vacuum acquire O(Λ i ) masses if they do not obey the classical constraints. These massive modes decouple at the scale Λ k , leaving only the degrees of freedom consistent with the classical (or quantum-modified) constraints. Unfortunately, anomaly cancelation depended on the fields B i=2...k : if these are not true degrees of freedom, then the anomaly matching conditions might not be satisfied. A solution to this problem can be seen by studying the X odd and Y even charges in Table 5. Their fermionic components have opposite charges under each of U (1) A , U (1) B , and U (1) R . When we calculate the anomaly coefficients for each of the mixed and pure U (1) anomalies, the contributions from each X odd cancel those from a Y even field. This is also true for the SU (4) 2 U (1) and SU (4) 3 anomalies. Therefore, we refer to X odd and Y even as an "anomaly neutral pair," indicating that they can be removed without changing any of the anomaly coefficients. Similarly, X even and Y odd also form an anomaly neutral pair.
If k is odd, then all of the operators {X 2 , Y 2 , . . . , X k , Y k } can be removed in neutral pairs. Substituting X k and Y k with their equations of motion, Eq. (2.13) becomes This is not possible if k is even. To remove all the redundant operators, we must also remove a pair {X 1 , Y even } or {X even , Y 1 }, and this is inconsistent: both X 1 and Y 1 are necessary to describe the moduli space. This can be seen if we move away from the origin along the flat direction parameterized by (A m Q), while keeping Q 1 = 0. Along this flat direction X 1 increases, but X even = 0. Therefore, X 1 describes directions on the moduli space that cannot be described by X even . Similarly, by increasing (A m−1 Q 3 ) and fixing Q 1 = 0, we can see that Y 1 is just as necessary.
Quantum modification to Eq. (2.15) could explain why the odd k and even k situations are different. If U (1) B is broken in the vacuum, then {X i , Y i } become an anomaly-neutral pair under the remaining symmetries, for any value of i = 1 . . . k. Based on F = N susy QCD, one would expect the classical relationships involving Q i and Q i+1 to be quantum-modified. Specifically, the combination (Z i−1 Z i ) has the same spurious U (1) i charge as Λ b=2N i , allowing modifications to equations such as Eq. (2.15). For example, the classical k = 4 constraint for X 4 Z 4 might become with some as-yet-unknown coefficients β i . As long as the coefficients are not zero, then the flat direction corresponding to (A m Q) = 0 with Q 1 = 0 now requires some of the Z i =1 to have nonzero expectation values. In this Z 1 = 0, ..k } is consistent at the origin of moduli space if k is odd. In this case the Z i = 0 solution remains valid far away from the origin, because every Λ b term multiplies at least one Z field. Consider Eq. (2.16) with k = 5: ...k = 0 flat direction remains on the moduli space for arbitrarily large values of (A m Q).
This does not mean that U (1) B is necessarily broken in the vacuum if k is even. Let us fix Z i = 0 for all i = 1 . . . k to ensure that U (1) B is not broken at the scale Λ i . After imposing this constraint, Eq. (2.16) becomes In this particular vacuum X 1 and Y 1 are redundant operators, and after they are removed from the calculation the U (1) B anomaly coefficients match the ultraviolet theory.
Theories with even N behave in essentially the same way. Under the exact family symmetries, the operator pairs {U odd , W even }, {U even , W odd }, and {V odd , V even } are anomaly-neutral. As in the odd N case, if k is even then it is not possible to remove all the redundant {U i , V i , W i } operators while preserving the anomaly matching. This leads us to expect that the classical constraint equations receive quantum modifications of the form if k is even. Either U (1) B is broken in the vacuum, or the operators {U 1 , V 1 , W 1 } are not degrees of freedom: in both cases, the IR theory satisfies t' Hooft anomaly matching. Thus, the reduced operator basis describes all infrared degrees of freedom, for both even and odd N .

Dynamically generated superpotential
In this section we find a dynamically generated superpotential in the region of parameter space with Λ 1 Λ 2 . . . Λ k . We begin by considering how the W d of Eq. (1.26) and Eq. (1.27) becomes modified at the G 2 confinement scale. Ignoring the precise relative coefficients between terms, At the scale Λ 2 , we expect J 1 and K 1 to confine to form the B 2 baryons. If we make these replacements in W (1) , it becomes It is likely that G 1 confinement changes the holomorphic scale Λ 2 to some new Λ 2 . To find the relationship between Λ 2 and Λ 2 , let us normalize the hadrons to have mass dimension +1: 2 and similarly for the baryon operators B 1 . The dynamically generated superpotential W 2 has the form 2), symmetry requirements ensure that the superpotential has the form This expression can also be derived with the same result by matching the gauge couplings at the mass threshold Λ 1 . Based on this agreement, we do not expect the superpotential W 2 to receive modifications of the form even though such terms are consistent with the family symmetries. As confinement continues, the products of intermediate mesons J 2 and K 2 can be replaced with The full superpotential is the sum Equations of motion: Let us consider equations of motion of the form ∂W/∂B 1 , where It is easy to show that these equations are for odd N , and for even N . The ∂W/∂B 2 equations yield more surprising results: for example, The classical constraint Y 2 Z 2 = Y 3 is modified, due to the appearance of X 2 in both W (1) and W (2) . For i = 2, 3 . . . (k − 1), we find The equations of motion ∂W d /∂Z i are not modified, so that for all i. Finally, the B k equations of motion are for odd N , and for even N . Recall from Section 1.3 that each gauge group SU (N ) i has a related CP parameter θ i , which determines the phase of the holomorphic scale Λ b i . Although Λ b did not appear in the k = 1 equations of motion, the phases of Λ b i do affect the equations of motion in the product group case. The overall phase of W d can still be removed by performing a U (1) R rotation; however, the relative phases between the Λ i may have physical effects.
Armed with these iterative equations of motion, we can rewrite the larger baryons B i>1 in terms of {B 1 } and the Z i fields. For example, Our guesses in Eqs. 2.16 and 2.17 as to the form of the quantum modification are correct, with β i = ±1 for each coefficient. This process is extended to arbitrary B i in the following way: each classical constraint involving products of the form (Z 1 Z 2 . . . Z j ) is modified by replacing adjacent pairs (Z i−1 Z i ) by (−Λ b i ), and each possible term is added to the product (Z 1 . . . Z j ). After making these adjustments, the k th equations of motion return the following constraints if k is odd: or if k is even: (2.47) In both cases, the origin of moduli space is a solution to the equations of motion. As we suggested in Section 2.1, if k is even then the B 1 fields are not independent degrees of freedom when Z i=1...k = 0: (2.48) Therefore, if U (1) B is a symmetry of the vacuum and k is even, then the B 1 fields are completely determined by J k and K k . After removing the B 1 fields, the t' Hooft anomaly matching conditions are satisfied. Elsewhere on the moduli space the B 1 fields may vary independently from K k and J k , U (1) B is spontaneously broken by Z i = 0, and the anomaly coefficients for the infrared symmetries match the values calculated in the ultraviolet theory.

Additional tests
So far we have restricted our attention to the ordered Λ 1 > . . . > Λ k case to find the dynamically generated superpotential. Due to the holomorphy of the superpotential, changes in the Λ i hierarchy should not alter the form of the superpotential. In this section we test this supposition by considering the Λ 1 Λ i =1 case. In this limit the SU (N ) k model reduces to an SU (N ) k−1 extension to F = N susy QCD which has been studied by Chang and Georgi [10].
As Λ 1 → 0, the A and Q fields decouple from the strongly coupled Q i . Chang and Georgi find that the infrared operators involving only Q i obey the following constraints: and so on. This is exactly the same form we derived for B i≥2 in Section 2.2. At scales above O(Λ 1 ) but below Λ i>1 , the G 1 charged degrees of freedom include A, Q, and M = (Q 1 Q 2 . . . Q k ). Let us define the mass-normalized field M , (2.55) The effective scale Λ b 1 contains a product of (Q There is also a quantum modified constraint Notice that the equations of motion from Z M also determine a vacuum solution for λ: Thus, the Lagrange multiplier can be treated as a new redundant baryon operator, which should be integrated out along with the other redundant fields. Finally, let us consider regions of parameter space in which Λ 1 is neither the largest nor the smallest confinement scale. In these cases the redundant operators include a mix of B i and Z ij , all of which produce the same equations of motion in the reduced operator basis. For any arrangement, at the last confinement scale Λ f there is a dynamically generated superpotential of the form where J f , K f , and M are such that

62)
and where { J f , K f , M } are normalized to have mass dimension +1. Under the remaining gauged G f , these fields satisfy the index condition for s-confinement, j µ j − µ G = 2, and there is a dynamically generated superpotential. Lagrange multipliers λ i enforce the constraint between the operators det(Q i . . . Q j ) and {Z i . . . Z j }, and the equations of motion provide a relationship between λ i and the other hadrons. After replacing the redundant operators with their equations of motion, we find that the constraints relating {J k , K k } to {B 1 , Z i } are unchanged.
Flow: It is a necessary condition for s-confining theories that their description in terms of gaugeinvariants is valid in the Higgs phase, when some fields acquire large expectation values and spontaneously break the gauge group to a subgroup. If the low-energy theory does not s-confine, then the original theory cannot be s-confining either. This is the "flow requirement" of [4], which we use in this section to test the SU (N ) k theory.
In the J k i j Λ vacuum with A αβ = 0, the SU (N ) k group is broken to SU (N − 1) k in the classical limit. This requires a nonzero (Q i ) α β for every Q i , which break each gauged SU (N ) i to representations as follows: (2.63) The (2N − 1) broken generators of each gauge group G i =1 "eat" the combination + + 1 from Q i−1 and Q i to create (2N − 1) massive gauge superfields, leaving behind the ( , ) bifundamental fields. The G 1 group behaves somewhat differently: its broken generators "eat" the ( , 1) part of Q 1 and a linear combination of the superfields Q i=1...4 . Under SU (N − 1) 1 the field decomposes as ( ⊕ ), so that the "eaten" Q field is replaced by a component of A. After removing the massive superfields, the SU (N − 1) 1 charged matter is A + 4Q + (N − 1)Q 1 . The overall effect of J k Λ on the SU (N ) k model is to replace N with N − 1. Now let us consider the limit where A αβ Λ and J = 0. In the even N = 2m case with U 1 = Pf A Λ 1 , SU (2m) 1 is broken to Sp(2m) 1 and decomposes into Sp ⊕ 1. Here Sp is the (2m 2 − m − 1) dimensional representation of Sp(2m). There are also (2m 2 − m − 1) broken SU (2m) generators, so the superfield A = Sp is eaten.
The fields Q and Q i are not directly affected by Pf A : however, as Sp(2m) has no complex representations, Q and Q 1 are effectively (2m + 4) quarks charged in the representation of Sp(2m). This theory is known to s-confine [22]. It is likely that the Sp(2m) × SU (2m) k−1 product group theory is also s-confining: we explore this possibility in Section 3.2.
In the case where N is odd, an expectation value X 1 = A m Q Λ breaks SU (2m + 1) to Sp(2m) instead. Aside from a few extra singlets and massive gauge bosons, there is little difference between the odd N and even N cases: the infrared theory is Sp(2m) × SU (2m) k−1 .
Conclusion: Our product group extension to the A + 4Q + N Q model exhibits the behavior required for an s-confining theory. The set of gauge invariant operators {J k , K k , B 1 , Z 1...k } satisfies the t' Hooft anomaly matching conditions; the origin remains on the quantum moduli space, so the theory can confine without breaking chiral symmetry; and there is a dynamically generated superpotential. Furthermore, the operators {J k , K k , B 1 , Z 1...k } provide a smooth description of the entire moduli space: there is no gauge invariant order parameter to distinguish the confined and Higgs phases. By considering the flow along flat directions, we have also found another product group extension to an s-confining theory, Sp(2m) × SU (N ) k−1 .

Other S-Confining Theories
In the previous section we find strong evidence that the product group extension to the A+4Q+N Q model is s-confining. In this section we consider the follow-up question: how many other sconfining models can be extended into product groups? We have already suggested that Sp(2m) with (2m + 4) can be extended into an Sp(2m) × SU (N ) k−1 product group model. If this theory is not s-confining, then the SU (N ) k A + 4Q + N Q model is not s-confining either. We discuss the behavior of this theory in Section 3.2.
There are also additional possibilities for the A + 4Q + N Q model in the case where N = 4. In this special case the entire SU (4) L × SU (N ) R family symmetry can be gauged: we consider whether or not such theories are s-confining in Section 3.1. In Sections 3.3 and 3.4 we discuss the other s-confining theories in [5] with family symmetries large enough to accommodate a gauged SU (N ) subgroup. This includes susy QCD with F = N + 1 flavors, and Sp(2m) with ( + 6 ) matter for m = 2 and m = 3. We show that some of these theories are not s-confining.
Due to the lack of an index constraint on the matter content, it is difficult to conduct a systematic search for new s-confining product groups. We have seen in the A + 4Q + N Q model that G 1 confinement increases the index sum of the G 2 charged matter by +2, but other confining theories tend to change the index sum by varying amounts. Therefore, the list of theories considered in this section is presumably incomplete.
We restrict our attention to s-confining models which can be extended by gauging a subgroup of the family symmetries and adding bifundamental fields. Our goal is to determine whether product group s-confinement is possible in each model, based on the index constraint after confinement. This is sufficient to show which of the product group extensions are obviously not s-confining. A more detailed analysis is appropriate for the theories which pass this test.

Special case: SU (4)
In this section, we extend the N = 4 A + 4Q + N Q model by gauging SU (4) L × G 0 × SU (4) r R for some and r. Here G 0 is the SU (4) gauge group containing the + 4( + ) matter, and every other gauged SU (4) contains four flavors of ( + ). It is convenient to relabel the hadrons to reflect the Q ↔ Q symmetry of the matter content of the A + 4Q + 4Q model: Table 6, for = r = 2. After extending the model in this way, the model has a "left-right" symmetry which simplifies many of the calculations in this section: Above, Λ i corresponds to the group G i , while Λ i is the confinement scale of the group G i . The group G 0 × U (1) R and the field A are invariant under the discrete transformation. Table 6: Above, the original s-confining theory A + 4(Q 0 + Q 0 ) is extended on the left and right by gauging G 2 L × G 2 R and adding the Q i and Q i fields to cancel the anomalies. To extend the model beyond = r = 2, more quarks Q i and Q j can be added with alternating U (1) A and U (1) B charges.
Infrared operators: Based on our understanding of the ( = 0, r = k − 1) models developed in the previous section and the vectorlike nature of the G 0 -charged fields, we can guess the form of the gauge-invariant operators which describe the moduli space: for i = 0, 1, . . . , and j = 0, 1, . . . , r.
Only under certain conditions do we expect the basis F to obey the anomaly matching conditions for the family symmetries listed in Table 6. We have already seen that in the ( = 0, r = k−1) models with even k, some of the operators in F become redundant in the U (1) B preserving vacuum. If this pattern continues in the ( , r) models with = 0 and r = 0, then we would expect that the set F obeys the anomaly matching conditions only if and r are even. If either or r is odd, we expect that some operators in F become redundant if U (1) A × U (1) B is preserved in the vacuum.

Equations of Motion
In the limit where Λ 0 is small, the theory reduces to two copies of F = N susy QCD with product group extensions. According to [10], the fields obey the following constraints: If is odd-valued, then the sum of neighbor contractions includes a constant term, (Λ b 1 Λ b 3 . . . Λ b ); if is even, then all terms include some power of Z i . The same relationship holds for r and det M R . As in the SU (N ) k models, we expect that the distinction between even and odd and r determines which of the operators in F are redundant when U (1) A and U (1) B are conserved in the vacuum. When G 0 confines, {M L , A, M R } form the following hadrons: with the dynamically-generated superpotential for some Λ b 0 consistent with the anomalous symmetries. We show the charges of the composite fields in Table 8.
The equations of motion from U 1 , K , and Z L produce the following constraints: These equations are not all independent, but contain N cons = 8 independent constraints. Table 8: After all of the gauge groups confine, the infrared degrees of freedom are described by the hadrons shown above. Their U (1) A and U (1) B charges depend on and r, respectively.
If we introduce Lagrange superfields λ L and λ R , the quantum modified constraints relating {Z L , Z R } to {Z i , Z j } as a superpotential: Redundant Operators: In this section we use the equations of motion to study the operator basis F. In the U (1) A preserving vacuum with Z i = 0, the expectation value of Z L depends heavily on whether is even or odd. If is even, then The same pattern holds for r and Z j when U (1) B is preserved.
It is simplest to consider the case in which both and r are even. Expanding about the Z i = Z j = 0 vacuum to first order in Z i and Z j , we find that every term in Eq. (3.10) contains a product of at least two fields, so that none of the operators in the set F are redundant. This is consistent with the fact that all of the anomaly coefficients from SU (4) L × SU (4) R × U (1) A × U (1) B × U (1) R match the ultraviolet theory when r and are even. This is not true if is odd. In this case the equations of motion for K r Z L and U 1 Z L can be rewritten as .
Similarly, the equation of motion for det M r becomes , (3.14) which can be recast into a linear constraint equation for any one of the Z even fields. Taken together, Eqs. (3.13) and (3.14) imply that the operators {K r , U 1 , Z even } should be removed in the U (1) A × U (1) B preserving vacuum if is odd and r is even. In the even , odd r case it is the operators {K , U 1 , Z even } which become redundant, and Z R rather than Z L remains large in the Z j = 0 vacuum. If both and r are odd, then the origin of moduli space is no longer a solution to the equations of motion: To satisfy this constraint, either M = 0, Z even Z odd = 0, or Z even Z odd = 0. Different family symmetries are broken in each case, leaving different sets of independent operators. In the M = 0 vacuum where M i j is proportional to δ i j , SU (4) L × SU (4) R is broken to its diagonal subgroup SU (4) d . The fields Q and Q r transform under SU (4) d as and , respectively, while the meson M decomposes as In the U (1) A × U (1) B preserving vacuum with Z i = Z j = 0, it is possible to write K r and U 1 either in terms of K and M r , or K and U 1 in terms of K r and M r . Therefore, we can either remove the set {K , U 1 , Tr M } or {K r , U 1 , Tr M }. This degeneracy is related to the fact that K and K r have the same transformation properties under SU If instead M = 0 and Z even Z odd = 0, only U (1) A is broken in the vacuum. One "(Z even + Z odd )" linear combination determined by the ratio of the expectation values becomes massive, and all sixteen M i j degrees of freedom remain independent. The operator K r is not redundant in this vacuum: the Z L K r equation of motion includes a term Z even Z odd K r which is not small. The set of redundant operators is {K , U 1 , (Z even + Z odd )}.
Finally, if the nonzero expectation value is Z even Z odd , then U (1) B is broken. As we would expect from the left-right symmetry, the redundant operators are {K r , U 1 , (Z even + Z odd )} in this vacuum. It is also possible to break a linear combination of U (1) A and U (1) B if Z even Z odd = 0 and Z even Z odd = 0.
Anomaly Matching: We have discussed six distinct cases with maximal symmetry in the vacuum, based on and r. Below, we show a summary of our results for each case: Broken symmetry Redundant operators (even, even) None None (odd, even) None For the remaining symmetries and operators in each case, we have verified that the anomaly coefficients match the UV theory. There are 21 matching conditions for each of the first three cases, 17 for the fourth case, and 12 each for the final two cases. Although some of these coefficients are related to each other via the left-right symmetry, the explicit calculation is lengthy and not very illuminating.
Let us also consider points on the moduli space with nonzero Z i or Z j , where none of the operators in the set F are redundant. In these vacua U (1) A × U (1) B is spontaneously broken, and the infrared operators should obey anomaly matching conditions for the remaining symmetries.
For the odd , even r case, U (1) A is broken by Z i = 0 for some Z i . After U (1) A is broken, {U 1 , Z even } form an anomaly-neutral pair: their U (1) B,R charges are opposite, so all of the U (1) 3 and gravitational U (1) anomalies cancel. The fermionic part of K r is neutral under U (1) B ×U (1) R , and it is in a real representation of SU (4) R : therefore, K r contributes nothing to the remaining anomaly coefficients. Thus, the t' Hooft anomaly matching conditions are also satisfied in the Z i = 0 vacuum where the operators {K r , U 1 , Z even } are independent degrees of freedom.
In the even-, odd-r models, the operators {K , U 1 , Z even } are restored as independent degrees of freedom when Z j = 0 and U (1) B is spontaneously broken. Applying the left-right transformation to the above results, the introduction of {K , U 1 , Z even } has no net effect on the anomaly coefficients once U (1) B is removed. Finally, when Z i = 0 and Z j = 0 in the odd-, odd-r models, the operators {K , U 1 , Z even } are restored as independent degrees of freedom without contributing to the anomaly coefficients of the remaining symmetries. Both U (1) A and U (1) B are broken in this case.  Table 9, so that different models are labelled by the sum ( + r). Models of this type appear in deconstructions of 5d gauge theories, as in [7]. Table 9: Above, we show the matter fields of the SU (4) ring extension to the A + 4Q + 4Q model.
Although the baryon operators Pf A and det Q i are unaffected by the ringlike nature of the product gauge group, there is now only one gauge-invariant meson operator: Tr M = Tr (Q 0 Q 1 . . . Q ). For any group G i , the adjoint operator is a degree of freedom in the limit where G i is weakly gauged, and can be used to create gaugeinvariant operators of the type Tr (M iMi ) and Tr (M 3 i ). In this notation, Q −1 = Q for the i = 0 case.
Even when these operators have large expectation values, the gauge group is not completely broken. It has been shown [23] in the SU (N ) k extension to F = N susy QCD that at an arbitrary point on the moduli space has a remaining U (1) 3 gauge group. In the A + 4Q + 4Q model it is also possible to set Pf A Λ 0 , so that SU (4) 0 is broken to Sp(4). This reduces the rank of the group by one, but is not sufficient to break U (1) 3 completely. Therefore, the SU (4) ring extension has a Coulomb branch, and is not s-confining.

Sp(2m) with (2m + 4) quarks
In Section 2.3, we found that the SU (N ) k extension of the A + 4Q + N Q model flows to an Sp(2m) × SU (2m) k−1 theory. In the limit where Sp(2m) is much more strongly coupled than the SU (2m) groups, the (2m + 4) quarks confine to produce the operator M = (Q 2 ), which transforms in the representation under the approximate SU (2m + 4) family symmetry.
The fields Q and M have the following charges: A dynamically generated superpotential reproduces the classical constraints on the Q i fields. In the product gauge group model shown in Table 10, an SU (2m) subgroup of the family symmetry is gauged and new bifundamental fields are added to cancel the anomalies. The family SU (2m+4) is explicitly broken to SU (2m)×SU (4)×U (1), under which the meson M decomposes as (3.20) and the dynamically generated superpotential becomes Including the bifundamental field Q 1 , the SU (2m) 1 charged matter in the confined phase of Sp(2m) is M A + 4M Q + 2mQ 1 , which is expected to s-confine. This model can also be derived using the deconfinement technique of Berkooz [1], by treating the matter field A as a bound state of two quarks transforming in the fundamental representation of a new Sp(N ).

SUSY QCD
A product group extension to F = N + 1 susy QCD can be derived from the N = 3 case of A + 4Q + N Q. In SU (3), the representation is the same as , so that the G 1 matter is effectively 4 + 4 . By gauging the SU (3) family symmetry of the Q and adding a sequence of bifundamental fields Q i , we have found a product group extension to susy QCD.
For larger values of N , let us gauge an SU (N ) subgroup of the SU (N + 1) R family symmetry as shown below: After SU (N ) 1 confinement, the hadrons are (Qq), (QQ 1 ), (Q N ), (Q ), which transform under SU (N ) 2 and the family symmetries as: Under SU (N ) 2 there are (N + 1)( + ) matter fields, which is consistent with the index constraint for s-confinement.
For this theory to be s-confining, it must be shown that the dynamically generated superpotential from SU (N ) 1 does not prevent the operators (QQ 1 ) and (qQ N −1 1 ) from varying independently; that the infrared operators obey the appropriate anomaly matching conditions; and that the origin is on the moduli space. The additional gauge groups are likely to introduce quantum-modified constraints between some of the operators, which may induce chiral symmetry breaking in some cases.
This theory can also be extended by gauging an SU (N ) subgroup of SU (N + 1) L , so that the most general product group extension is SU (N ) × SU (N ) 0 × SU (N ) r . Based on the behavior of the ( , r) A + 4Q + 4Q model for odd and r, we expect that some of the ( , r) susy QCD models also break chiral symmetry.
Alternating Gauge Groups: The F = N +1 model can also be extended by gauging the entire SU (N + 1) family symmetry. In this case, the gauge group has the alternating form SU (N ) × SU (N + 1) × SU (N ) × SU (N + 1) × . . ., with a series of bifundamental fields: The matter content is simpler in this case, as all of the fields are SU (N + 1) × SU (N ) bifundamentals. When SU (N ) 1 confines, we are left with Under SU (N + 1) 2 , there are (N + 1) flavors of + which is expected to confine with chiral symmetry breaking. Many of the G 2 singlets we would naïvely construct, such as (QQ 1 )(Q N 1 ), are set to zero by the equations of motion, so G 2 confinement leaves the following charged fields: After G 1 × G 2 confinement, the low energy theory is simply F = N + 1 susy QCD with some gauge singlet fields. Both product group models based on susy QCD have the potential to be s-confining, and may be promising directions for future study.

Other Models
Of the s-confining theories listed in [5], there are only a few models possessing non-Abelian family symmetries larger than the gauge group. We have already discussed the SU (N ) models with A + 4Q + N Q and (N + 1)(Q + Q), as well as the Sp(2m) model with (2m + 4)Q. There are two remaining cases based on Sp(2m) with A + 6Q [24,25]. If m = 2 or m = 3, an SU (4) or SU (6) subgroup of the family symmetry can be gauged. In this section, we show that the product group extensions do not exhibit s-confinement.
Sp(6) with A + 6Q: Consider the m = 3 case with just one extra product group. Below, we show the matter fields above and below the Sp(6) confinement scale: In the confined phase of Sp (6), the SU (6) index sum becomes j µ j − µ G = 3 · (6 − 2) + 6 · 1 − 2 · 6 = +6, (3.22) so the product group does not s-confine. It may be possible to remove some of the degrees of freedom by adding a nonzero tree-level superpotential, but this is outside the scope of the current study.

Conclusion
For several s-confining theories, we find product gauge group models with the following properties: • All infrared degrees of freedom are gauge invariant composite fields; • The infrared physics is described by a smooth effective theory, which is valid everywhere on the moduli space (including the origin); • There is a dynamically generated superpotential.
This allows confinement without symmetry breaking, even when the quantum and classical moduli spaces are different. In particular, this behavior may be found in the following models: In this paper we argue that the A + 4Q + N Q and Sp(2m) : (2m + 4)Q product group models s-confine. Based on less rigorous arguments we suggest two product group extensions of susy QCD which may also be s-confining, but a more detailed analysis is required. It is also entirely possible that there are many other s-confining product group theories unrelated to the models considered in this paper.
In the A + 4Q + N Q model with N = 4, we consider a set of product group extensions of the form G L × G 0 × G r R . When and r are both odd, the chiral symmetry is necessarily broken in the vacuum, so the theory is not s-confining. If instead the sum ( + r) is odd, then the origin remains on the quantum-deformed moduli space, and some of the infrared operators become redundant in the symmetry-enhanced vacua. Finally, if and r are both even, we find that all of the operators are interacting degrees of freedom in the neighborhood of the origin. In each case, there is a dynamically generated superpotential.
One feature of the product group models is the lack of small gauge-invariant operators, which has a promising phenomenological application to composite axion models. After lifting some of the flat directions, a Peccei-Quinn U (1) symmetry may be dynamically broken when the gauge group confines, producing a light composite axion. If the product gauge group is suitably large, the Peccei-Quinn symmetry is protected against the explicit symmetry breaking effects which would otherwise be induced by higher-dimensional operators. We explore this option in an upcoming paper [26].
Another promising direction for future study is to treat the product gauge groups as k site decompositions of 5d susy theories. Exact calculations in N = 2 susy may provide us with a better understanding of the 4d N = 1 models considered in this paper.

A.1 D-Flat Directions
The auxiliary gluon scalar fields have interactions from the Kähler potential given by V = 1 2 D a D a , where Ground state solutions are given by D a D a = 0. Equation (A.1) can be simplified by replacing T and T with T a : With this substitution, we may write D a as The indices i and j refer to SU (4) L and SU (N ) R , respectively, while α, β and γ correspond to the gauge group. The generators T a span the set of traceless N × N matrices, so if the fields satisfy for any constant ρ, then D a = 0. It is useful to define the matrices d,d, and d A as follows: (d A ) α β = Diag(|σ 1 | 2 , |σ 1 | 2 , |σ 2 | 2 , |σ 2 | 2 , . . . , |σ m | 2 , |σ m | 2 , 0). (A.8) The Pfaffian, Pf A, is not defined for odd-dimensional matrices. It is not generally possible to simultaneously diagonalize d A , d, andd. This is a departure from susy QCD: in this case, ifd is diagonal, then d α β =d α β + ρδ α β must also be diagonal. Once d A is added, this condition is relaxed.

A.2 Special Cases
In this section we consider the φ Λ limit along particular flat directions in which d A , d, and d happen to be diagonal. Let us begin with the N = 2m case: where we define 16) for i = 1 . . . m.
In the N = 2m + 1 case we add a row and column to A, with A α,N = A N,β = 0 for all α and β. The form of Q is left unchanged, but we add a nontrivial N th row to Q i N with entries q i = 0. With these modifications, the matrices d A , d andd become d A = Diag |σ 1 | 2 , |σ 1 | 2 , |σ 2 | 2 , |σ 2 | 2 , . . . , |σ m | 2 , |σ m | 2 , 0 (A.17) d = Diag |v 1 | 2 , |v 2 | 2 , |v 3 | 2 , |v 4 | 2 , 0, . . . , 0, and the gauge-invariant operators are , (A.20) Classical constraints The dynamically generated superpotential has the form W ∼ A N −2 Q 4QN . For odd N , there are three ways to contract the gauge indices: while for even N there are five terms: The relationships between the coefficients are determined by matching the equations of motion from W d to the classical constraints on the operators. In the classical limit for even N , it follows from Eq. (A.15) that for example, so that Applying this technique to other products of gauge invariant operators, we find  which when matched to the corresponding products of J and K imply that In both cases the overall constant α has no effect on the equations of motion, and cannot be calculated from the classical constraints.