Double Soft Theorems in Gauge and String Theories

We investigate the tree-level S-matrix in gauge theories and open superstring theory with several soft particles. We show that scattering amplitudes with two or three soft gluons of non-identical helicities behave universally in the limit, with multi-soft factors which are not the product of individual soft gluon factors. The results are obtained from the BCFW recursion relations in four dimensions, and further extended to arbitrary dimensions using the CHY formula. We also find new soft theorems for double soft limits of scalars and fermions in N=4 and pure N=2 SYM. Finally, we show that the double-soft-scalar theorems can be extended to open superstring theory without receiving any alpha' corrections.


Introduction
Recently there has been a resurrection of interest in studying various low energy limits of scattering amplitudes. Of particular interest are situations which exhibit universal behavior; that is, when the limiting behavior of an amplitude factors into a product of a universal "soft factor" times a lower-point amplitude independent of the soft particles. Such cases are called "soft theorems", the most famous of which may be Weinberg's classic soft (photon, gluon, or graviton) theorems [1]. Other theorems include [2][3][4][5] as well as, much more recently, the subleading and sub-subleading graviton theorems of Cachazo and Strominger [6] (see [7] for further developments and applications).
Strominger and collaborators [8] have argued that all of the known soft and subleading soft theorems may be understood as consequences of large gauge transformations. That is, transformations which fall off sufficiently rapidly at infinity such that they must be considered consistent with the asymptotic boundary conditions defining the theory, while sufficiently slowly that they act nontrivially on asymptotic scattering states. In the case of gravity, the relevant "gauge transformations" are of course diffeomorphisms, and the relevant asymptotic symmetry group (in four-dimensional Minkoswki space) is the Bondi, van der Burg, Metzner, Sachs group [9,10]. It has been shown using the CHY scattering equations [11] that the subleading and sub-subleading graviton soft theorems hold for tree-level graviton amplitudes in any number of space-time dimensions, suggesting that an analog of the BMS symmetry should be relevant more generally [8,12]. Perturbative theories at null infinity realizing these symmetries have been proposed in [13]. The issues regarding possible loop corrections to the subleading soft theorems were studied in [14].
Double-soft limits (where two particles are taken to have very low energy) have also received a lot of attention in the literature, both in the earlier works [15,16] and more recently. For example, Arkani-Hamed et. al. [17] have shown that the double soft limit of scalars in N = 8 supergravity exhibit the expected E 7 (7) symmetry of the scalar moduli space, in a manner analogous to the classic soft-pion theorem of [15,18]. This result was recently extended to the four-dimensional supergravity theories with N < 8 supersymmetry, and the N = 16 supergravity in three dimensions in [19]. Furthermore, supergravity amplitudes in both four and three dimensions with two soft fermions were studied in [20], and new soft theorems were proposed. New double-soft leading and subleading theorems for scalars (and leading for photons) were also studied in various theories such as DBI, Einstein-Maxwell-scalar, NLSM, and Yang-Mills-scalar in [21]. Kac-Moody structure has been found for the four dimensional Yang-Mills at null infinity [22], where double soft limits play another important role.
In this paper we derive several new soft theorems for tree-level scattering amplitudes in gauge and string theories with more than one soft particle. We derive the universal behavior of amplitudes with two or three soft gluons. It is known that when the soft gluons have identical helicities, the result can be obtained simply by setting the gluons to be soft one by one, thus we focus on the non-trivial cases when the soft gluons have different helicities. Indeed we find that for these cases the soft factors are a product of the individual soft-gluon factors with certain non-trivial corrections. We first derive theorems from the BCFW formula in four dimensions [23], and further extend our results with double-soft gluons for gauge theories in any number of dimensions by using CHY formula [11]. We check that our results are consistent with the fact that if the soft limit is taken in order then the soft factors reduce to a product of the single-soft factors given by Weinberg. We also note that, in contrast to the gluon case, amplitudes with multiple soft gravitons can always be obtained by simply taking the gravitons to be soft one by one.
We then proceed to study amplitudes in N = 4 and pure N = 2 Super Yang-Mills theory (SYM) with two soft scalars or two soft fermions. We find that the double soft behavior is governed by R-symmetry generators acting on a lower-point amplitude, resembling the results of supergravity theories found in [17,19,20], although the vacuum structure of SYM is quite different from that of supergravity theories. Finally, we consider double-soft scalars in the open superstring theory. Unlike the double-soft-scalar theorem in N = 8 supergravity, which would receive α corrections if one tried to extend it to closed superstring theory, we find that open superstring amplitudes satisfy exactly the same double-soft-scalar theorem of SYM at α = 0. Given the similarity of the double-soft theorems of SYM and those of supergravity theories, it would be very interesting to understand if any of these theorems could have an interpretation as hidden symmetries.
The paper is organized as follows: In section 2 we derive the double-soft-gluon theorem for tree-level amplitudes in gauge theories using the BCFW recursion relations formula (2.10), which may be recast into a different form (2.12), and we further extend the results to arbitrary dimensions resulting in formula (2.28). Then, in section 3 amplitudes with three soft gluons are considered. In the following section 4 we comment on multi-soft gravitons. Subsequently, in section 5, we explore the universal behavior of amplitudes with two soft scalars or two soft fermions in supersymmetric gauge theories, including N = 4 SYM as well as pure N = 2 SYM, with main results given by (5.13) and (5.28). Finally, in section 6, we prove that the newly discovered double-soft-scalar theorem in SYM can be extended to the open superstring theory without any α corrections.
Note added: After finishing this work, we became aware of a related work by Klose, McLoughlin, Nandan, Plefka and Travaglini, which has some overlap with our paper [24].

Double-soft gluons from BCFW recursions
We start by considering color-stripped amplitudes in gauge theories with two adjacent gluons taken to be soft. It is straightforward to see that if the two gluons have the same helicity, then the two gluons may be taken soft one at a time. Moreover it is evident from that the result is independent of the order in which the two gluons are taken soft. A similar simple calculation shows that if the two gluons have different helicities, then the result cannot be given by a product of two single soft factors obtained by taking the gluons to be soft one by one. Therefore this is the non-trivial case we are interested in, namely we would like to study the amplitude A(1 + , 2 − , 3, . . . , n) in the double-soft limit We will use the standard spinor-helicity formalism for the four-dimensional massless particles throughout this paper: and realize the soft limit by taking Using the BCFW recursion relations [23], it is straightforward to see that the two dominant diagrams that contribute in the limit at hand arē (2.5) with the following BCFW shifts λ1 = λ 1 + zλ n ,λn =λ n − zλ 1 . (2.6) Let us now analyse the two contributions separately. First for diagram (a) we have where we have used the fact thatP → p 3 in the limit; hence this result is independent of whether particle 3 has positive or negative helicity (in the above calculation we have chosen it to be positive). Now, the second diagram (b) gives [nP ][P 3] A n−2 (3, . . . , n) , (2.8) where in the second expression we have used the fact that pn = p n in the limit, and we also applied the single-soft theorem for the soft legP . After some simplification, we find Adding the contributions from the two diagrams together, we obtain the final result for two soft gluons having different helicities, As typical for amplitudes computed from the BCFW recursion relations, the result contains a spurious pole 1 n|1+2|3] . We will show that it indeed cancels out between the two terms at leading order of the soft limit, as should be the case. Now, if on the other hand we take the soft limit in succession, namely say take p 1 to be soft first, then the first term in the soft factor is subleading, and the second term simplifies to , (2.11) which is precisely the product of two soft factors of a positive gluon and a negative gluon, with the positive gluon p 1 being taken soft first. Although the above result (2.10) is very compact and nicely reduces to a product of two soft factors, if we take the soft limits in succession, it is specific to four dimensions and as a natural property of using the BCFW recursion it contains a spurious pole. In the next section we will use the CHY formula for pure Yang-Mills tree level scattering amplitudes [11] to derive a further formula for the universal double-soft-gluon factor. This result will be valid in any dimension, for any helicity combination of the soft gluons, and it will be manifestly free of unphysical poles. When the two soft gluons have opposite helicity, the comparison of the result obtained from the CHY formula and (2.10) will yield agreement and provide us with the intuition to recast the above into the following equivalent form lim p 1 ∼p 2 →0 A(1 + , 2 − , 3, . . . , n) → n2 n1 12 [13] [12] [23] 1 + n1 [13] 32 [23] s n12 [13] A n−2 . (2.12) Therefore, we see that the alternating helicity double soft gluon factor is composed of the product of two single soft gluon factors plus a non-trivial correction.

Double-soft gluons from CHY
As we mentioned earlier, in this section we will reconsider the double-soft-gluon limit making use of the CHY formula for tree-level scattering amplitudes in pure Yang-Mills, valid in arbitrary dimensions [11]. The CHY formula for an n-point gluon scattering amplitude is given by Upper and lower indices on the matrix Ψ denote removed columns and rows respectively. The indices p, q, r, i, j, k, m and w can be fixed arbitrarily without changing the result. The 2n × 2n dimensional matrix Ψ is given by where the n × n dimensional sub-matrices are Here k µ a are external leg momenta, and the µ a are parts of corresponding polarization tensors. The product of delta functions enforces the scattering equations and saturates all integrals. With this the integration reduces to a sum over all solutions to the scattering equations.
We want to make the external gluon momenta k µ 1 and k ν 2 soft by substituting k µ 1 → τ k µ 1 and k ν 2 → τ k ν 2 and considering τ → 0. It is essential to send both momenta to zero simultaneously in order to capture the double soft factor structure. We choose not to erase indices (1) and (2). With this we have to isolate the extra terms in A n as compared to A n−2 and integrate out σ 1 and σ 2 . While doing so we will only keep the leading contribution in the τ → 0 limit, to obtain the leading double soft gluon factor.
First we notice that at leading order in τ the entire σ 1 and σ 2 dependence in A n apart from the pfaffian Pf (Ψ m,w m,w ) is contained in Another σ n,3 term in the denominator is suppressed, which will help restore the proper Parke-Taylor factor for the (n − 2)-point amplitude case. As in [21], we can make the convenient variable transformation (2.14) and immediately integrate out δ(f 1 − f 2 ) using the variable ξ. This will introduce a summation over all solutions ξ for the equation f 1 − f 2 = 0, and an overall factor of 1/F (ξ), where Here we used that on the support of f 1 − f 2 = 0 we can always substitute Making use of this, (2.13) becomes Before we rewrite dρδ(f 1 + f 2 ) as a contour integral over poles and deform the contour as usual, we should also extract the extra terms depending on ρ and ξ from the pfaffian factor Pf (Ψ m,w m,w ) in order to reduce it to the (n − 2)-point amplitude case. To do that, we will use the recursive definition of a pfaffian for an anti-symmetric 2n × 2n matrix A: where a ij is an element of matrix A, θ(x) is the Heaviside step function, and index i can be chosen arbitrarily. If rows and/or columns are missing from matrix A before the expansion is applied, the respective indices have to be skipped in the summation. Since we are ultimately interested in the leading double soft gluon factor, for convenience we will only keep the leading in τ terms in the expansion of Pf (Ψ m,w m,w ). In order to isolate the leading terms, we recall that the summation over the solutions ξ in (2.17) features two types of solutions: non-degenerate solutions for which ξ = O(1), and a unique degenerate solution for which ξ = O(τ ) [21].
Let us first consider the non-degenerate (nd) solutions. In this case it is straightforward to see that the only leading term in the pfaffian expansion is 19) where for convenience we define the abbreviation Combining (2.17) with (2.19), writing dρδ(f 1 + f 2 ) as a contour integral 21) and deforming the contour to wrap around all other poles in ρ instead, immediately reveals that there is no pole at infinity and the only residues come from the two poles at (ρ + ξ/2 − σ 3 ) → 0 and (ρ − ξ/2 − σ n ) → 0 due to the term σ n,3 /(σ n,1 σ 2,3 ) remaining from the Parke-Taylor factor. Keeping (2.19) in mind, this tells us that for any of the non-degenerate solutions ξ (nd) , at leading order in τ these two residues will always be proportional to 2 · k 3 and 1 · k n respectively. Therefore, we select the following polarization gauge for the external legs going soft 2 · k 3 = 0 , 1 · k n = 0. (2.22) In this gauge all the non-degenerate solution contributions to the leading double soft gluon factor vanish, such that we can concentrate on the degenerate solution only. Now we compute the degenerate (d) solution contribution. Using (2.16) we can straightforwardly expand the degenerate solution ξ (d) to leading order All the terms appearing in (2.17) are expanded to leading order in τ analogously. The expansion of the pfaffian features three leading terms in this case: , and we used the abbreviation (2.20). Again, we combine (2.17) with (2.24), write dρδ(f 1 + f 2 ) as a contour integral (2.21) and deform the contour to wrap around all other poles in ρ instead. Analogously to the non-degenerate case we see that there is no pole at infinity, and the only two contributing residues come from poles at ρ − σ 3 → 0 and ρ − σ n → 0. Dropping Pf Ψ m,w m,w , which is part of the (n − 2)-point amplitude and not the double soft gluon factor, both these residues are of the following type at leading order in τ : With this we conclude that the leading double soft gluon factor for legs i and i + 1 going soft is given by Despite the appearance, the double-soft factor (2.26) is not manifestly anti-symmetric under 3 ↔ n, since this symmetry is broken by the gauge choice (2.27). This is consistent with the results of the previous section. Therefore, the particular computation above for legs 1 and 2 going soft in an n-point amplitude gives the following factorization in the double soft gluon limit valid in the gauge (2.22). Here we emphasize again that since the above result is obtained from the CHY formula, it holds in arbitrary dimension and for all helicity combinations of the two soft gluons. Let us now compare (2.28) to the result (2.10) obtained from BCFW. Specifying to four dimensions and selecting (1 + , 2 − ) helicities for the soft gluons, we use the following standard dictionary to translate R 3 1,2 and R n 1,2 into spinor helicity formalism: . (2.29) Here we have selected proper reference spinors to account for the gauge (2.22). Anticipating that R 3 1,2 and R n 1,2 roughly correspond to the two terms that are summed in (2.10), we notice that R 3 1,2 already features an s 123 ≈ 2k 3 · (k 1 + k 2 ) and R n 1,2 an s n12 ≈ 2k n · (k 1 + k 2 ) in the denominator. So in both cases we introduce an extra factor of n|1 + 2|3] in numerator and denominator, and expand the numerators. The Schouten identity then yields a slight simplification such that the terms in the numerators separate into an expected part and a part proportional to s 123 or s n12 in the two cases respectively. Finally, subtracting the resulting R n 1,2 from R 3 1,2 displays some cancellation and we are left with exactly the terms appearing in (2.10). 1 Similarly, we can show that selecting the soft gluons to be of the same helicity, i.e. (1 + , 2 + ), the double soft gluon factor (2.28) reduces to the product of two single soft factors. Here we also use: In this case no strategic term manipulations are needed. R 3 1,2 directly reduces to half of the expected result and R n 1,2 to minus half of it, so that (R 3 1,2 − R n 1,2 ) properly gives what we expect.

Triple-soft gluons
With results of the double-soft limit at hand, we can go on to study the universal behavior of scattering amplitudes with multiple gluons being soft. Here we will take a look at the triple-soft limit, which is a natural next step beyond the double-soft limit. Again the nontrivial cases occur when all soft gluons are adjacent. Beside the straightforward case of all soft gluons having the same helicity, there are two helicity configurations of interest: . .) and A(1 + , 2 − , 3 + , . . .), where 1, 2 and 3 are the soft legs.
Let us begin with the first case, A(1 + , 2 − , 3 − , . . .). It is easy to see that the following BCFW diagrams are dominant in the soft limit Since the calculation is similar to that of the double-soft limit, we will be brief here. The contribution from diagram (a) to the soft factor gives , where we have used the fact thatP is soft, as well as the result of the double-soft limit with two negative-helicity gluons. SpecifyingP in terms of external momenta, the soft factor simplifies to 1 One should keep in mind that in spinor helicity formalism factors of √ 2 from the amplitude are absorbed into the coupling constant in front. In case of the double soft gluon factor this amounts to an overall extra factor of 2 which is suppressed in (2.10). Summing over the three contributions, we obtain the universal behavior of amplitudes with three adjacent soft gluons Now we go on to consider the second case of interest, A(1 + , 2 − , 3 + , . . .). The result is given by the same set of BCFW diagrams, but now with the helicity of gluon 3 changed where we have applied the alternating helicity double-soft gluon theorem (2.10) to the right sub-amplitude in the BCFW diagram. After some further simplifications taking the soft limit into account, we obtain Note that it is not allowed to discard the soft momenta k 1 , k 2 and k 3 in 2|K n1 K n123 |4 and 2|K 34 K 1234 |4 to further simplify the above expressions in the soft limit. For the diagram (b) we find . (3.11) In conclusion, we obtain the following soft theorem for three adjacent soft gluons with alternating helicities nicely reduce to a product of a single-soft factor and a double-soft factor if we take any one of the three soft gluons to be soft first. Finally, we note that all the unphysical poles appear in pairs, and we have checked numerically that they all precisely cancel at leading order in the soft limit.

Multi-soft gravitons
In this section we comment that, unlike in the case of two soft gluons, the double-softgraviton limit is simply given by the product of two single-soft gravitons, independent of their helicity configuration. For instance, let us consider soft gravitons of opposite helicity g ++ 1 and g −− 2 . Similar to the case of double-soft gluons from BCFW recursion, one needs to consider the following three diagrams: In fact, a simple analysis of three-and four-point amplitudes reveals that only the diagram (a) will contribute at leading order in the double-soft limit. A simple way to obtain the result for diagram (a) is to view it as an "inverse-soft" diagram [25], where leg 1 + is considered as being added to an (n−1)-point amplitude making use of Here the soft factor S 1 + is defined as In this diagram the shifted legs are p i and p n , which are given bỹ In the soft limit we simply have p i → p i and p n → p n . Since p 2 is soft as well, it follows from the single-soft graviton theorem that the above expression reduces to (4.6) for any choices of x and y. Considering that M (a) n is the dominant diagram at leading order, we have replaced it by the full tree-level amplitude M n . Finally, we note that the terms S 1 + (2) and S 2 − (1), which are missing in the summation in (4.5), are subleading in the limit. Thus the result can be alternatively written as being simply the product of two single-soft factors. As mentioned earlier, this confirms that the leading double-soft-graviton limit can be obtained by taking the gravitons to be soft in succession, in either order, unlike the case of double-soft gluons. Given the result of double-soft gravitons, it is straightforward to see that it can be extended to the case of multiple soft gravitons, such that the soft factor of multiple-soft gravitons should be given by the product of multiple single-soft-graviton factors for any number of soft gravitons.

Double-soft limits in supersymmetric gauge theories
In this section, we move on to study the universal behavior of scattering amplitudes in supersymmetric gauge theories (in particular N = 4 SYM and pure N = 2 SYM) in the limit with the momenta of two scalars or two fermions being soft. The double-soft-scalar limit was first studied in N = 8 supergravity in [17], where the 70 scalar fields in the theory parametrize the coset space E 7(7) /SU (8). Thus these scalar fields behave as "pions". As pointed out in [17], amplitudes in this theory vanish in the single-soft-scalar limit consistent with the famous "Adler's zero" [18], and behave universally in the double-soft-scalar limit in a manner analogous to the soft-pion theorem where (R i ) I J is the generator for SU(8) rotations on particle i Recently, this result was extended to more general supersymmetric gravity theories [19], including 4 ≤ N < 8 supergravity theories in four dimensions as well as N = 16 supergravity in three dimensions. Soft-scalar theorems have been very useful in determining the UV counter terms in supergravity theories [19,28]. It is known that for supersymmetric gauge theories (in particular N = 4 SYM), a generic vacuum has mostly massive particles, and the massless S-matrix only exists at the origin of moduli space. Thus one should not expect that the scalars would behave as "pions". Indeed it is easy to see that the amplitudes in N = 4 SYM do not vanish in the single-soft-scalar limit, in contrast to supergravity theories. However, as in [20], one can still ask whether the amplitudes in SYM exhibit some universal behavior in certain soft limits. This is what we will explore in this section.

Double-soft scalars in N = 4 SYM
The on-shell fields in N = 4 SYM can be nicely packaged into a superfield [26], where g + is the positive-helicity gluon, ψ A is the spin +1/2 gluino, and so on. In this section, we will consider the limit with two scalars φ AB becoming soft. First of all, as we mentioned previously, it is easy to see that amplitudes in N = 4 SYM behave as O(τ 0 ) in the single-soft-scalar limit.
Let us now consider the double-soft-scalar limit. First we note that when the two soft scalars are not adjacent, the amplitude is not singular, and thus it cannot behave universally under the soft limit. So we will only consider the case where the two soft scalars are adjacent, which is singular and therefore universal. To be precise, we take p 1 and p 2 to be soft. Furthermore, if two scalars have no common SU(4) index, they form a singlet and the leading singular result should simply be given by the single-soft gluon limit. However, as pointed out in [19] for supergravity theories, one can extract interesting information about this case by introducing suitably anti-symmetrised amplitudes. This is particularly relevant to pure N = 2 SYM where two scalars can only form a singlet, which will be discussed in the next section. Here we will focus on the case where two scalars do not form a singlet, as was considered in [17] for N = 8 supergravity. For this configuration it is easy to see that the leading contribution arises when two soft scalars have one and only one common SU(4) index. In terms of the BCFW representation of the amplitude, there are two leading contributions in the double-soft limit: After integrating out ηP , the contribution (a) is given by where the integration over η's selects the soft legs 1 and 2 to be scalars. Note that, as mentioned above, we are interested in the case where two scalars have one common SU (4) index. We have applied the super BCFW recursion relations [17,27], with shifts chosen as λ1 = λ 1 − zλ n ,λn =λ n + zλ 1 , ηn = η n + zη 1 .
Finally, we have written the shifts in A n−2 in an exponentiated form and only kept the leading terms. There are two possible ways to get a leading contribution above, one is by expanding η 2 from exp − 1 2 1P η 2 ∂ ∂η 3 , and another one is by expanding η 1 from exp z P η 1 ∂ ∂ηn . In the first case we get one η 2 from the exponent, thus from the fermionic delta-function δ (4) we have one η 2 , two η 1 's and one η 3 . Thus we obtain, where an extra minus due to the fermionic integral has been included. In the soft limit the above expression simplifies to (5.8) Analogously, we obtain the second contribution, which is given by where we used the on-shell solution z P = − s 123 n|3+2|1] ∼ − s 123 n3 [31] . Let us now consider the diagram (b), for which a similar consideration leads to Now, using the fact thatP is also soft, one can apply the supersymmetric single-soft theorem to A n−1 (P , . . . , n). Thus we have 11) Substituting this result into eq. (5.10), integrating out η P , and selecting the scalar components we find (5.12) We observe that the unphysical pole cancels out. In particular, the first term in A (b) cancels A (a),2 , and we obtain the double soft-scalar theorem in N = 4 SYM 13) where φ BC = ABCD φ DA . Note the appearance of the R-symmetry generators η B ∂ η D . As mentioned earlier, although scalars in SYM are not Goldstone bosons, we find that our result very much resembles what has been found in N = 8 supergravity. Furthermore, as we will see, the double-soft-scalar theorem is exact even when we consider amplitudes in open superstring theory, meaning that it does not receive any α corrections from string theory. Finally, we remark that the subleading order of this limit will be finite and thus not universal, since general BCFW diagrams start to contribute. This is the same for the double-soft limit of scalars in N = 8 supergravity.

Double-soft scalars in pure N = 2 SYM
In this section we consider the double-soft-scalar limit for pure N = 2 SYM. Due to the fact that it is not a maximally supersymmetric theory, the on-shell fields in N = 2 SYM are separated into two distinct mulitplets. These multiplets can be nicely obtained from N = 4 SYM by SUSY truncation [31], 14) where A N =4 (η) is the superfield in N = 4 SYM that we defined in the previous section. Therefore, we see that the scalar in A N =2 (η) corresponds to φ 12 in N = 4 SYM, while the scalar inĀ N =2 (η) corresponds to φ 34 in N = 4 SYM. Thus they form a singlet.
Since the scattering amplitudes in pure N = 2 SYM can be obtained from amplitudes in N = 4 SYM via SUSY reduction, we will use the same strategy as in [19]: instead of studying the amplitudes in N = 2 SYM directly we will study the relevant amplitude in N = 4 SYM first, and then reduce it to N = 2 SYM via the SUSY reduction. Now, in contrast with the case we studied in the previous section, here we are interested in precisely the amplitudes with the two soft scalars forming a singlet A((φ 1 ) 12 , (φ 2 ) 34 , . . .), and with the following anti-symmetrization as introduced in [19]: (5.15) Let us focus on A((φ 1 ) 12 , (φ 2 ) 34 , . . .). As before, in the soft limit the dominant contributions are given by the diagrams shown in Fig.(5.4). The diagram (a) is given by a similar expression to that used above, but now we select different species of scalars Since we select different scalars, the leading term now seems to come from taking two η 1 's as well as two η 2 's from the fermionic delta-function. However, this vanishes after the anti-symmetrization (5.15). Thus we will focus on subleading terms, and the calculation proceeds as outlined in the previous section. Therefore, we only quote the results (5.17) Similarly from diagram (b), we find ( 5.18) Summing over all contributions, we find that after the anti-symmetrization we end up with 19) where the double-soft factor S N =4 ij;k is defined as (5.20) Now we have to project this to N = 2 SUSY. The soft factor S N =4 12;k is unchanged, while S N =4 34;k depends on whether particles 3 and n are in the A N =2 (η) or theĀ N =2 (η) multiplet. If they are in A N =2 (η), then the contribution from S N =4 34;k should be discarded, since we set η 3 and η 4 to 0. If they are in A N =2 (η), then integrating out η 3 and η 4 the contribution from S N =4 34;k simply reduces to 2. The result can be summarized as where the U(1) generator R N =2 i is defined as 22) which precisely correspond to the U(1) part of the R-symmetry generators in pure N = 2 SYM.

Double-soft fermions in N = 4 and pure N = 2 SYM
In a similar fashion one can study the limit with two soft fermions in N = 4 SYM as well as pure N = 2 SYM. As before, the interesting case occurs when the two fermions are adjacent. Because the (anti)-symmetrization procedure does not work for the double-soft fermions since they have different helicities [20], we will only consider the case when two fermions do not form a singlet. Thus the leading singular terms arise from adjacent fermions having one and only one common SU(4) index. To be precise we take soft particles as (ψ 1 ) D and (ψ 2 ) BCD . The calculation in terms of BCFW recursion relations is very similar to the case of double-soft scalars, and again the relevant BCFW diagrams are shown in Fig.(5.4). Let us quote them here for conveniencen As before, any other generic BCFW diagrams are subleading, since they are diagrams with a single-soft fermion and behave as 1/ √ τ in our soft limit. In contrast, the dominant diagrams above behave as 1/τ . The contribution from diagram (a) is given by Now the integration on η's is such that the soft legs 1 and 2 are the soft fermions of interest. Following the analysis of double-soft scalars, we find two kinds of contributions from diagram (a). One of them is given by 25) and the other contribution is Similarly, from diagram (b) we find  (5.27) Adding the results of the two diagrams together, we finally obtain the double soft-fermion theorem in N = 4 SYM, (5.28) Unlike the case of double-soft scalars, the cross term η A 3 ∂ ∂η D n does not cancel anymore. However, all the unphysical poles cancel out manifestly. Note that the fermions in N = 2 SYM are not required to form a singlet like the scalars. The extension to the fermions in N = 2 SYM is straightforward via SUSY truncation as we discussed in the previous section.

Double-soft limit in open superstring theory
It is known that the soft-scalar theorems in N = 8 supergravity are violated in the closed superstring theory if α corrections are included [28]. It is then natural to ask whether the newly established double-soft-scalar theorems in SYM would receive any α corrections for scattering amplitudes in open superstring theory. We find remarkably that amplitudes in open superstring theory satisfy exactly the same double-soft-scalar theorems as in SYM theory.