Large-size expansion for triangular Wilson loops in confining gauge theories

The asymptotic behavior of Wilson loops in the large-size limit ($L\rightarrow\infty$) in confining gauge theories with area law is controlled by effective string theory (EST). The $L^{-2}$ term of the large-size expansion for the logarithm of Wilson loop appears within EST as a two-loop correction. Ultraviolet divergences of this two-loop correction for polygonal contours can be renormalized using an analytical regularization constructed in terms of Schwarz-Christoffel mapping. In the case of triangular Wilson loops this method leads to a simple final expression for the $L^{-2}$ term.

(with large flat contour C bounding a region with area S (C)) play an important role as models of the heavy-quark confinement and are often briefly called confining gauge theories (although the problem of confinement in real QCD is more complicated). In our modern understanding of confining gauge theories, the area law is considered as a secondary manifestation of a much deeper phenomenon of effective string formation. This phenomenon is described by effective string theory (EST) which provides a detailed information about the asymptotic behavior of Wilson loops in the limit of large size of contour C (denoted informally as |C| → ∞ in eq. (1.1)). EST assumes that the asymptotic behavior of Wilson loops in the limit of large size of contour C can be described by a functional integral over surfaces Σ bounded by contour C: 3) The idea that Wilson loops W (C) can be approximated by functional integral (1.3) has a long history. Starting from qualitative and heuristic arguments [2], one can try to justify the stringy approach to Wilson loops using various limits and expansions: large size, large number of colors [3]- [6], large number of space-time dimensions [7]- [10], Regge limit [10]- [12]. Here we are interested in EST understood as an effective theory describing the large-size limit. The first steps in this direction were made by M. Lüscher, G. Münster, K. Symanzik and P. Weisz [15]- [17]. In the computation of the first terms of the large-size expansion, one can approximate S EST [Σ] by Nambu action S Nambu (Σ) = σ 0 S (Σ) (1.4) where S (Σ) is the area of surface Σ and σ 0 is the bare string tension (different from the renormalized physical string tension σ appearing in area law (1.1 )). If one wants to use EST for the construction of higher terms of the large-size expansion then S EST [Σ] must be understood as an infinite series containing all possible terms compatible with the symmetries of the problem. The theoretical work in EST has been proceeding in various directions including -derivation of general constraints on terms appearing in EST action S EST [Σ] [18]- [27], -computation of loop corrections in EST for rectangular Wilson loops [26], [28]- [31] and for other closely related quantities like correlation functions of Polyakov lines and spectra of closed and open strings [18], [20]- [25], -analysis of string finite-width effects [17], [32], [33]. The predictions of EST have been successfully verified by many lattice Monte Carlo tests (see [26], [34]- [40] and references therein).
In the case when large-size limit |C| → ∞ is implemented as a uniform rescaling of contour C by a large factor λ, EST predicts the following structure of the large-size expansion for polygonal contours C: (1.5) Here f k (C) (k = ln, −2, −1, 0, . . .) are functionals of contour C. EST provides much information on f k (C) but some properties of f k (C) depend on the underlying microscopic gauge theory (MGT). Two leading terms of expansion (1.5) are controlled by functionals associated with area S (C) and perimeter L (C): Here σ is the string tension appearing in eq. (1.1). Parameters σ, ρ are determined by MGT and cannot be computed in EST. Parameter ρ depends on the renormalization scheme used for Wilson loops in MGT whereas string tension σ is renormalization invariant in MGT. Some properties of functionals f k (C) may be derived on general grounds without a direct involvement of EST. For example, the compatibility of expansion (1.5) with transitivity W ((λ 1 λ 2 ) C) = W (λ 1 (λ 2 C)) leads to relation

Simple example
The primary subject of this paper is functional f 2 (C). In the case of rectangular contours, f 2 (C) has been already computed (see Sec. 1.3). Our aim is to provide a general computational scheme (including regularization and renormalization) for the case of f 2 (C) with arbitrary polygons C and to perform an explicit calculation for the simplest case of triangles C. Although term λ −2 f 2 (C) is strongly suppressed in expansion (1.5), this term plays a crucial role in lattice tests of EST. The computation of this correction for general polygons is also important for understanding the structure of renormalization in EST (in the sense of renormalization in effective theories).
As for more dominant terms of expansion (1.5) associated with f ln (C) and f 0 (C), these terms are well known. A brief review of their properties may be found in Appendices B, C.
In this introductory section we would like to concentrate on functional f 2 (C) and to minimize the involvement of terms f ln (C) and f 0 (C). Therefore it makes sense to start from an expression combining several Wilson loops in such a way that the role of term f 2 (C) is enhanced whereas some of dominant terms f k (C) cancel. An instructive example is ratio taken in the limit λ → ∞ at fixed ν 1 , ν 2 . Using properties (1.9) -(1.12) of expansion (1.5), we find (1.14) The LHS is a renormalization invariant combination of Wilson loops (which can be computed, e.g. by lattice Monte Carlo simulations). On the RHS we have an asymptotic series in inverse powers of λ → ∞. Note that terms f −1 (C) and f 0 (C) cancel on the RHS of (1.14). The explicit expression for f ln (C) is given by eq. (B.2). Expression (1.14) provides a good illustration of the role of the λ −2 f 2 (C) correction. On the other hand, one can construct more complicated renormalization invariant ratios of Wilson loops whose large-size expansions involve also f 0 (C): see Appendices B, C.

Known results for f 2 (C)
Dimensional arguments and scaling property (1.9) lead to the following structure of f 2 (C): where σ is the string tension appearing in area law (1.1) and g 2 (C) is a dimensionless function depending only on the geometry of polygon C.
For rectangular contours, f 2 (C) was computed in Refs. [28], [29] with an arithmetic error which was later detected and corrected in Refs. [30], [31]. For rectangle C rectangle (L 1 , L 2 ) with sides L 1 , L 2 one has where D is the space-time dimension and E n (z) are Eisenstein series. In this work we compute f 2 (C) for the case of triangular loops C. For triangle C triangle (θ 1 , θ 2 , θ 3 ; S) with interior angles θ 1 , θ 2 , θ 3 and with area S we find According to eq. (1.15), the full explicit expression for f 2 (C triangle ) is

Outline of the work
The main part of this work concentrates on the computation of f 2 (C) in largesize expansion (1.5). Working with ratios (1.14) of Wilson loops, one can eliminate more dominant term f 0 (C) from the consideration. Nevertheless for completeness a review of well-known results for terms f ln (C) and f 0 (C) is provided in Appendices A, B, C. In Sec. 2 we sketch the basic formulas of loop expansion in EST. This part of the work is common for our triangular case and for the rectangular case studied in Refs. [28], [29]. Correction f 2 (C) is given by a figure-eight-like Feynman diagram for a two-dimensional field theory of (∂φ) 2 + (∂φ) 4 type (see eq. (2.7)) in the region bounded by contour C with zero boundary conditions. We use notation S 4 (see eqs. (2.11), (2.17)) for the integral associated with this Feynman diagram. In principle, S 4 represents the result for f 2 (C) (up to the sign). However, this Feynman diagram has ultraviolet divergences.
The deviation of the current work from Refs. [28], [29] starts when it comes to the regularization of ultraviolet divergences in S 4 . Unfortunately the regularization used in Refs. [28], [29] can be applied only to the case of rectangular contours C. In our work we suggest another regularization which can be used for arbitrary polygons. In Sec. 3 we describe the difference between the regularization used in Refs. [28], [29] and our regularization. Our renormalization procedure operationally consists of two steps: Here S 4 ren is completely free of ultraviolet divergences and provides the result for f 2 (C) (see eq. (2.24)) whereas the intermediate quantity S 4 ren,1 is only partially renormalized and still has ultraviolet divergences.
Our calculation is based on Schwarz-Christoffel (SC) mapping for polygon C and on other powerful methods of complex analysis. In Sec. 4 we introduce a complexified representation for S 4 and use it in order to perform the first step of renormalization S 4 → S 4 ren,1 (with technical details placed in Appendix E).
In Sec. 5 we rewrite the derived expression for S 4 ren,1 in terms of SC mapping.
In Sec. 6 we describe the second step of renormalization procedure S 4 ren,1 → S ren based on analytical regularization of the SC representation for S 4 ren,1 assuming the general case of arbitrary polygons C.
In Sec. 7 we apply the general formula derived for S ren with an arbitrary polygon to the case of triangles. We show that in the triangular case all integrals involved in the computation of S ren can be computed explicitly and derive final expression (1.19 ) for f 2 (C) = − S ren . The technical details of the computation are placed in Appendices F, G, H. For completeness we also provide information about f 0 (C) with triangular contours C in Appendix C.2.

Expansion of Nambu action
The first steps of our work are identical to those of Refs. [28], [29]. For the computation of the first orders of the large-size expansion of Wilson loops (including f 2 (C)), one can replace full EST action S EST [Σ] by Nambu action (non-Nambu terms of effective action S EST [Σ] become essential starting from the computation of correction f 3 (C) generated by the boundary action [18], [21], [26]). Thus for our aim we can approximate with S Nambu (Σ) given by eq. (1.4). Assuming parametrization X a x 1 , x 2 of surface Σ, one has In EST (like in the theory of fundamental boson strings) reparametrization invariance must be treated as a gauge symmetry. In the case of flat Wilson contours C lying in the X 1 , X 2 plane, it is convenient to use the planar gauge (sometimes called static gauge): Expanding up to quartic terms, we find in this gauge (2.6) We use notation U (C) for the flat region in x 1 , x 2 plane bounded by Wilson contour C. Thus Here S (C) is the area of region U (C) bounded by flat contour C. One can rewrite this expansion in the form (2.10) (2.11)

Loop expansion in EST
The large-size expansion for Wilson loops is constructed by applying steepestdescent method to functional integral (2.1). Expanding in small fluctuations of Σ near the minimal surface spanned on contour C, one finds in one-loop approximation [15]- [17]: Here Det reg [−∆ (C)] is the functional determinant of Laplace operator (referred later as Laplace determinant for brevity) with zero (Dirichlet) boundary conditions on contour C with properly regularized ultraviolet divergences. In this oversimplified approach, at intermediate stages of the calculation we formally deal with the loop expansion in small parameter 1/σ 0 → 0 at fixed contour C but in the end this formal 1/σ 0 expansion can be rearranged into large-size expansion for Wilson loops at fixed physical (renormalized) string tension σ.
The real situation is more complicated because -loop corrections of EST have ultraviolet divergences which must be renormalized (in the sense of renormalization in effective theories), -microscopic gauge theory (MGT) has its own ultraviolet divergences and renormalization.
Fortunately, these complications play a limited role if one is interested in the EST computation of the first large-size terms of expansion (1.5).
For the computation of f 2 (C) one needs only the Nambu term of the full effective action. Loop expansion of EST with Nambu action naively goes in even powers of 1/σ 0 which results in even terms λ −2n f 2n (C) of large-size expansion (1.5). This naive argument explains the vanishing of odd term f 1 (C) (1.12) but a serious proof of (1.12) requires more effort since divergences and their renormalization in EST may generate odd and non-analytical terms like f −1 (C) and f ln (C).
Anyway the concept of naive and formal loop expansion in small parameter 1/σ 0 may be helpful (at least operationally) for the computation of f k (C) in EST and within this framework one has the following correspondence between loop counting and f k (C) computation (2.14) The main subject of this paper is two-loop correction f 2 (C). Technically (but not conceptually) the computation of f 2 (C) can be well separated from the computation of one-loop terms f ln (C) ln λ+f 0 (C). Therefore in subsequent sections we concentrate on f 2 (C) whereas a detailed discussion of one-loop terms is placed to Appendices B, C.
In one-loop approximation (2.12) there appears Laplace determinant which has area, perimeter and cusp divergences (see Appendices A.1, A.2 ). Therefore the completion of one-loop computation (i.e. the computation of f ln (C), f 0 (C)) includes two parts: 1) renormalization of Laplace determinant (Appendix A.2) using heat-kernel expansion (Appendix A.1), 2) computation of the renormalized Laplace determinant for arbitrary polygons (Appendix C) using Schwarz-Christoffel (SC) mapping.
Both parts work are well described in literature so that we simply combine all known pieces together in Appendices A, B, C.
An explicit expression for f ln (C) can be obtained already at the first step (renormalization of Laplace determinants) -see eq. (B.2). As for f 0 (C), the situation is more subtle because single Wilson loops W (C) and corresponding functionals f 0 (C) are not renormalization invariant quantities in MGT. where contours C a and coefficients m a must obey certain balance conditions (B.5), (B.6), (B.8). Once these balance conditions are satisfied, linear combination (2.15) is given by eq. (C.29) derived in Appendix C in terms of parameters of SC mapping. In the special case when all polygons C a in linear combination (2.15) are of triangles one can derive a simpler expression for (2.15) directly in terms of geometric parameters of triangles -see eq. (C.32).

Two-loop EST correction
Using action (2.8), we compute the functional integral of EST in the 2-loop approximation: (2.16) Here Thus in the 2-loop approximation Comparing this with the 1-loop approximation In fact, one should use a renormalized version of this formula: Comparing this with expansion (1.5), we see that Using expression (2.11) for S 4 , we compute Gaussian integral (2.17) Now Wick theorem yields a figure-eight-like Feynman diagram: and is Green function of Laplace operator with Dirichlet boundary condition: Now we insert eq. (2.28) into eq. (2.27) One should keep in mind that the RHS is plagued by divergences. Indeed, nondiagonal matrix element (2.30) has a singularity at y → x so that quantity G µν (x) is divergent for any x. This means that on the RHS of eq. (2.36) the integrand is divergent at any point x. Nevertheless eq. (2.36) provides a good starting point for the discussion of the renormalization.

Rectangular case: method of Dietz and Filk
The naive non-renormalized expression for the 2-loop correction (2.36) allows for arbitrary polygons C and in this sense is common for the current work concentrating on triangle and for Refs. [28], [29] devoted to rectangles. But starting from eq. (2.36) the paths of this work and Refs. [28], [29] diverge. The main reason is that Refs. [28], [29] use an ultraviolet regularization specific for rectangular case which cannot be generalized to other polygons. The regularization of Refs. [28], [29] are based on the explicit diagonalization of Laplace operator acting in rectangle with Dirichlet boundary conditions. The spectral problem − ∆ψ mn = λ mn ψ mn (3.38) has the obvious solution The method of Refs. [28], [29] is based on 1) formal spectral decomposition of G µν (2.29) in terms of λ mn , ψ mn ignoring divergence of G µν , 2) insertion of this formal expression for G µν into the RHS of eq. (2.36), 3) formal integration over x in eq. (2.36).
As a result, one obtains an expression for S 4 in terms of divergent sums m1m2...mn where R a (m 1 , m 2 , . . . , m n ) are some polynomials of integer m k (initiating from parameters m, n of eigenvalues λ mn ). Formal sums (3.41) are divergent. Refs.
[28], [29] use analytical regularization for numerators R 1 so that the problem reduces to the computation of sums in convergence region of α k -space with a subsequent analytical continuation in α k to integer values corresponding to monomials of the original numerator The power of this approach is that it allows for a complete analytical computation of f 2 (C) for rectangles C with result (1.16). However, the method is based on the explicit expressions for spectrum (3.39) and for eigenfunctions (3.40) and therefore is limited to the case of rectangles.

General polygons: Renormalization based on SC mapping
As was discussed above, the analytical regularization of Refs. [28], [29] was designed for the special case of polygons. Our aim is to construct a regularization (and renormalization) procedure for f 2 (C) with arbitrary polygons. We also want this regularization method to be efficient for practical computations. The first hint for the construction of this regularization procedure comes from the lessons of 1-loop corrections. As discussed in Appendix C, the explicit analytical computation of 1-loop correction f 0 (C) is based on SC mapping for polygon C.
This observation tells us that we should try to construct the regularization and renormalization for f 2 (C) also in terms of SC representation. The next observation is that the naive divergent expression (2.36) has two types of divergences: 1) The integrand is divergent at any point x because G µν (x) is divergent at any x. Therefore first one must renormalize the integrand on the RHS of (2.36).
2) After this step of renormalization the integrand on the RHS of (2.36) becomes finite in the internal part of the integration region but singular on the boundary C so that the integral is still divergent. At this moment we will need the second step of the renormalization.
This two-step renormalization was already announced -see eq. (1.20). Note that in the case of Refs. [28], [29] both types of divergences were renormalized using one procedure based on analytical regularization (3.42) of formal divergent sums: discrete summation (3.41) of the spectral representation keeps interior and boundary divergences in one pot. But our method based on SC representation requires a separate treatment of interior and boundary divergences.
Thus we have a two-stage renormalization procedure. A detailed description of the two steps will be given below. But already now it makes sense to announce that the second step (renormalization of boundary divergences) will be implemented via a sort of analytical regularization for divergent integrals in SC representation. This SC version of analytical regularization (applicable to arbitrary polygons C) is different from the analytical regularization of Refs. [28], [29] designed for rectangles and dealing with discrete sums.
3.3 First step of renormalization: from S 4 to S 4 ren,1 has a singularity in the diagonal limit y → x: This divergence is eliminated by a renormalization of string tension σ 0 . Operationally this renormalization corresponds to the replacement of G µν by its finite part and by the replacement of bare string tension σ 0 with renormalized string tension σ appearing in area law (1.1). After the replacement eq. (2.36) takes the form (3.49) We use label ren,1 in S 4 ren,1 because the integral on the RHS still contains additional divergences: -boundary divergences which appear in K µν (x) when x approaches the boundary of region U , -extra cusp divergences appearing in polygonal regions U (C) when x approaches a vertex of the polygon.
Thus, the first step S 4 → S 4 ren,1 of our renormalization program (1.20) has made the integrand finite in the internal part of U (C) on the RHS of (3.49) but the integral is still divergent because of boundary and cusp singularities of the integrand.

Complex representation 4.1 Definitions and conventions
In order to proceed, it is convenient to pass from Cartesian 2D coordinates x to complex coordinates. Our conventions for this complexification are We use the metric tensor with components For symmetric tensor 14) When writing real and complex representations for the Green function we assume the equivalence of various notations for the same coordinate state.
With the above conventions we can rewrite eq. (3.46) in the complex form:

Conformal mapping to semiplane
If one knows a conformal mapping of region U (bounded by Wilson contour C) to the upper complex semiplane then functions K ζζ , K ζζ * can be expressed via this conformal mapping. The computation is done in Appendix E. The result is {z, ζ}, (4.20) Here ζ is a complex coordinate in region U and z (ζ) is the conformal mapping of U to the upper complex semiplane. We use notation {z, ζ} for Schwarz derivative (D.1).
In expression (4.24) for I 2 we use property (D.3) of Schwarz derivative Thus Here the integration runs over upper complex semiplane C 2 In the case of polygons, the conformal mapping ζ (z) is known as SC mapping. SC mapping ζ (z) is defined by differential equation Parameters β k determine interior angles of the polygon: Angles θ k satisfy geometric conditions so that parameters β k obey constraints Sometimes it is convenient to use parameters For SC mapping (5.1), Schwarz derivative {ζ, z} (D.1) can be easily computed: (5.8) Now eqs. (4.26), (4.27) lead to where Q 1 (z, z * ) = |Im z| −4 , (5.12) (5.13)

SL(2,R) symmetry
It is well known that linear fractional transformations

SC mapping with one vertex at infinity
The freedom of linear fractional transformations (5.18) allows for taking one SC vertex to infinity. Let us put vertex z nv to infinity: When taking this limit, one should tune parameter A = A (z nv ) in eq. (5.1) so that This equation does not contain parameter β nv but it still can be defined by eq. (5.6) and obeys condition (5.5) together with other parameters β k : The expression for Schwarz derivative (5.8) of mapping (5.21) can be obtained by taking limit (5.19) in eq. (5.8) (5.24) Now we apply general formulas (5.11) -(5.12) to SC mapping (5.21): (5.27)

SC vertex at infinity: pro and contra
Thus we have two versions of SC representation for I a . Representation (5.11) is based on SC mapping with finite vertices z k whereas representation (5.25) assumes that SC vertex z nv is taken to infinity. Both integrals (5.11), (5.25) are divergent. These integrals must be renormalized (by the second step S 4 ren,1 → S 4 ren of our renormalization program (1.20)) and our plan is to use analytical regularization. In principle, our procedure of analytical regularization can be formulated for both versions of SC mapping with the same final renormalized result for I ren a . But when it comes to the practical work, the SC representation with one vertex at infinity seems to be preferable. One advantage of this approach is obvious: keeping one SC vertex at infinity we significantly simplify all integrals. The second argument against SC mapping with finite vertices is that parameters β k must obey constraint (5.6). This constraint is critical at some steps of analytical regularization so that one gets into the problem of analytical continuation in β k under constraint (5.6) which leads to some technical (solvable but annoying) problems.
As was discussed in Sec. 5.2, in the case of SC mapping with all SC vertices being finite we have a powerful constraint of SL(2,R) symmetry on allowed renormalization procedures. Our choice to construct renormalization in terms of SC mapping with one vertex at infinity makes SL(2, R) symmetry rather implicit. We are left with the freedom of choice: which vertex of the polygon is associated with the SC vertex taken to infinity. The result of the renormalization must be independent of this choice.
whereas for parameters γ k we allow arbitrary complex values. We use label naive in M (n)naive P because the integral on the RHS of eq. (6.1) may be divergent. Expression (5.25) can be rewritten for I 1 in the form For I 2 we find where P 2 (z, z * ) is a polynomial defined via holomorphic polynomial One can construct analytical regularization and renormalization of integrals (6.1) by analogy with the theory of Euler B-function  The transition to eqs. (6.9), (6.10), (6.11) corresponds to the second step S 4 ren,1 → S 4 ren of our renormalization program (1.20).

From divergent integrals to meromorphic functions
The suggested scheme of analytical renormalization raises many questions if one wants a mathematically impeccable implementation of this program. On the other hand, in practical calculations one can typically (but not always) perform formal mathematical operations with divergent integrals without wasting time for a careful justification of these naive formal manipulations. A careful mathematical theory of functions M (n) P for arbitrary n and arbitrary polynomials P can be constructed. But in the framework of the current work devoted to triangular Wilson loops we need functions M (n) P only for the special case n = 2 and with rather simple polynomials P . In this simple case many intermediate integrals can be computed analytically and the problem of analytical continuation can be (at least partly) solved using these explicit expressions rather than invoking the general theory of functions M (n) P with arbitrary n and arbitrary P . Therefore the complete description of the theory of functions M (n) P will be presented in a separate work devoted to Wilson loops with arbitrary polygonal contours. Here only a list of main results will be given without proofs and derivations.
When one starts with the construction of the rigorous theory of functions M (n) P one meets many questions: 1) Is the region of parameters {γ k } n k=0 where integral (6.1) is convergent non-empty? In other words, do we have a starting region for the construction of the analytical continuation?
2) Does the analytical continuation depend on the path? 3) Is the result of the analytical continuation regular at final physical points appearing on the RHS of eqs. (6.9), (6.10)?
The brief answers are: 1) The region of parameters {γ k } n k=0 where integral (6.1) is convergent is non-empty for any fixed {z k } n k=1 and fixed P . A careful proof of these statements for arbitrary n ≥ 2 (i.e. for polygons with arbitrary number of vertices n v ≥ 3) requires some work. We postpone the proof till a separate work devoted to arbitrary polygons. In the case of triangles, i.e. for n = n v − 1 = 2 many of the above general properties can be seen directly from explicit expressions computed in Appendices G, H. P we distinguish between naive integral representation corresponding to (6.1)

Functions Π
and functions Π (n) which come from the analytical continuation of (6.14) with technical subtleties discussed in Sec. 6.2.
In terms of functions Π (n) P relations (6.9), (6.10) take the form 7 Calculation of f 2 (C) for triangles 7

.1 Plan
In the case of triangles eqs. (6.15), (6.16) become Below we will see that function Π 1 (α, {γ 1 , γ 2 } , {z 1 , z 2 }) with arbitrary arguments can be easily computed analytically, which will immediately lead to the result for I ren 1 . As for I ren 2 , one can compute Π In Appendix G we derive expression (G.10) for function Π 1 . Applying this result to our case (7.1), we find Next we express parameter Ã via area S (C) of the triangle using eq. (F.11) and simplify the result using expression (F.6) for β 3 :

Final result
Our final results for I ren 1 (7.6) and I ren 2 (7.23) are symmetric with respect to permutations of parameters β 1 , β 2 , β 3 whereas the intermediate steps of the computation were asymmetric. The symmetry of the final result is a good test of the consistency of our renormalization procedure. In fact, it is a test of the compatibility of our analytical renormalization with SL(2, R) symmetry discussed in Sec. 5.4. In our case of triangular contours, the compatibility of renormalization procedure with SL(2, R) was demonstrated via the explicit computation of I ren 1 and I ren 2 . In principle, one can prove that our analytical renormalization respects SL(2, R) for arbitrary polygons but this is a subject of a separate work devoted to arbitrary polygons. Now we insert results (7.6), (7.23) into the basic expression (6.11) for S 4 ren : and θ k are interior angles (5.2) of the triangle. Inserting eq. (7.25) into eq. (2.24), we obtain which completes the derivation of eq. (1. 19).
Note that in the case of triangles general polygonal constraint (5.5) is enhanced to 0 < β k < 1. (7.28) Obviously f 2 (C) (7.27) is regular in this region of β k .

Conclusions
Using EST, we have computed term f 2 (C) of the large-size expansion (1.5) for triangular Wilson loops. The result is given by expression (1.19). The success of the calculation is based on the new version of analytical regularization using SC mapping. This regularization can be applied to arbitrary polygons.

A Properties of Laplace determinants A.1 Heat-kernel expansion for Laplace operator
For Laplace operator acting in the region bounded by contour C one has the following heat-kernel expansion [15], [42]- [46]: For polygonal contours C Here θ k are interior angles of polygon C, n v (C) is the number of vertices and For function ξ(θ), an integral representation was derived by M. Kac in Ref. [43]. Explicit expression (A.3) for ξ(θ) was obtained by D. B. Ray (the derivation is described in Ref. [47]). In ζ-regularization method, one first defines the regularizing ζ-function at Re s > 2

A.2 Renormalization of Laplace determinants
Then one performs the analytical continuation to s = 0 and computes the derivative Laplace determinant in the ζ-regularization is given by and has the property where δ (C) is given by eq. (A.2).

B One-loop EST corrections
where ξ (θ) given by eq. (A.3). Now let us turn to f 0 (C). Wilson loop W (C) is not a renormalization invariant quantity so that the single functional f 0 (C) is not a physical quantity and any attempt to write an explicit expression for single functional f 0 (C) will result in an ugly combination of scheme dependent parameters. However, simple elegant formulas can be written for linear combinations (2.15) associated with renormalization invariant combinations of Wilson loops -see eqs. (B.13) and (C.29).
The simplest examples of renormalization invariant ratios are combination (1.13) and Creutz ratios Here L (C a ) is the perimeter of contour C a . In addition, one must obey a vertex-balance condition. Roughly speaking, each vertex angle value must appear in the numerator of (B.4) as many times as in the denominator. In order to make a more careful formulation of this constraint, let us first define Θ = {θ A } as a set of all vertex angle values appearing in polygons C a so that each θ A is included in Θ only once (whatever often it may appear in polygons C a ): (B.7) Next, let p aA be the number of occurrences of angle value θ A among vertices of polygon C a (so that possible values are p aA = 0, 1, 2, . . .). Then the vertexbalance constraint reads It is well known that Laplace determinants in 2D regions can be computed using the conformal anomaly [48]- [51]. In order to compute Laplace determinant Det (−∆ (C)) for the region bounded by contour C, one has to construct a conformal mapping of this region to some standard region (semiplane or circle). In the case of polygons C this conformal mapping is known as Schwarz-Christoffel (SC) mapping. Thus combining the conformal anomaly and SC mapping, one can compute Laplace determinants for arbitrary polygons. However, on this way one must solve two problems: -the anomaly-based representation for Laplace determinants has an integral form and this integral must be computed, -this integral representation has cusp divergences which must be renormalized.
These two problems were successfully solved by E. Aurell and P. Salomonson in Ref. [41] where the renormalized functional determinant (in ζ-regularization) of two-dimensional Laplace operator with Dirichlet boundary condition for an arbitrary polygon region was computed and expressed via parameters of SC mapping.
In this Appendix we compute the combinations of Laplace determinants appearing on the RHS of eqs. In order to simplify the extraction of final expressions for Laplace determinants from Ref. [41], we provide a small dictionary relating the notation used in the current work (LHS) and notation adopted Ref. [41] (RHS with label AS).
Polygonal boundary C of the region where Laplace operator acts: Area of the region bounded by contour C: Complex coordinate in the polygon plane: Function Z ′ 1−β (0) (do not confuse it with quantity Z ′ P (0) given by eq. (C.5)) playing an important role in Ref. [41] but irrelevant for our work: For angles of polygons Ref. [41] uses the same parameters θ k , β k , α k as in this work (but with Greek indices) -see eqs. (5.2), (5.7).

C.1.2 SC mapping from unit circle to polygon
One should keep in mind that Ref. [41] uses Schwarz-Christoffel (SC) mapping of the unit circle |u| ≤ 1 in complex u-plane to the polygon in complex ζ-plane u − e iφ k −β k (C.8) (the original notation of Ref. [41] uses z instead of ζ). Here e iφ k are points on the boundary of the unit circle which are mapped to vertices ζ k of polygon C. λ 0 is a real constant controlling the size of the polygon.

C.1.3 Result for Laplace determinant
The general result for the determinant of Laplace operator defined in polygon C with Dirichlet boundary condition in ζ-regularization (A.6) -(A.8) is given by eq. (62) of Ref. [41]: The RHS contains already defined circular SC parameters λ 0 , {φ k }, {β k } and function h (β) (C.7) whose explicit form plays no role for our work.

C.1.4 SC mapping: from unit circle to semiplane
One can easily establish a connection between semiplane version of SC mapping (5.1) and unit-circle version of SC mapping (C.8). Indeed, one can map unit circle in the u-plane to the upper semiplane of complex variable z using linearfractional transformation with the inverse transformation This maps points e iφ k on the boundary of the unit circle in the u-plane to points z k on the real axis of the z-plane 14) one finds that in terms of the z-semiplane parametrization, SC mapping (C.8) takes form (5.1).

C.1.5 Laplace determinants in terms of semiplane parameters of SC mapping
Using relation (C.12), we find ln e iφj − e iφ k = ln |z k − z j | + ln cos 1 2 φ j + ln cos 1 2 φ k + ln 2. (C.15) It follows from eq. (C.14) that Combining the last two equations and using constraint (5.6), one can derive relation Now eq. (C.9) takes the form This provides an expression for Laplace determinant via parameters z k , β k , A of SC mapping in semiplane form (5.1 ). The RHS of (C.18) contains function h (β) (C.7) which is computed in Ref. [41] but cancels in our final formulas.

C.1.6 Taking one SC vertex to infinity
Sometimes it is convenient to keep one SC vertex at infinity and to work with SC representation (5.21). Taking the limit |z nv | → ∞ on the RHS eq. (C.18) with constraint (5.20) and using relation (5.22), we find the expression for Laplace determinant in terms of SC mapping (5.21) In the case of SC mapping (5.21) with vertex z nv at infinity we derive from eq.
Now we can compute the combination of Laplace determinants appearing on the RHS of eq. (B.13). For each polygon C a appearing in eq. (B.13) we introduce SC mapping (5.1) Using decomposition (C.20), we obtain (C.28) Remember that this result was derived assuming vertex-balance condition (B.8).

C.2.1 Result
The primary subject of this paper is the computation of 2-loop correction f 2 (C) for triangles. For completeness it makes sense to compute also 1-loop terms f ln (C) and f 0 (C) for triangles. Quantity f ln (C) is given by simple formula (B.2) which can be used for any polygon including triangle. As for f 0 (C), we must compute the RHS of eq. (C.28) assuming balance conditions. If all polygons C a are triangles and vertex-balance condition (B.8) holds then one can derive relation where in agreement with eq. (5.7) . (C.32)

C.2.2 Nontriviality of vertex-balance condition for triangles
Before deriving relation (C.32) it makes sense to check whether this relation is of any use. The problem is that in the case of triangles the combination of vertex-balance condition (B.8) and angle sum rules (5.4) for each triangle is very restrictive. One can easily obey balance conditions using simplest renormalization invariant combination (1.13) but the for the computation of large-size expansion (1.14) for this combination it is sufficient to know function f ln (C) whereas f 0 (C) does not appear in (1.14) at all. In other words, for combination One can wonder whether using only triangle contours C a we can satisfy balance condition in a non-trivial way so that for each triangle C a . An example of the solution of this problem can be constructed using Wilson loops w (n a1 , n a2 , n a3 ) for triangles C a with interior angles θ ak = π 9 n ak (C. 37) with angle sum rule n a1 + n a2 + n a3 = 9.

C.2.3 Derivation
Now we turn to the derivation of eq. (C.30). In principle, one can derive eq. (C.30) from general equation (C.28). On the other hand, one can profit from the explicit expression computed in Ref. [41] for Laplace determinant with triangle contour C. One can read this explicit expression from eqs. (68) and (69) of Ref. [41]. In the native notation of Ref. [41] the result reads where parameters α k are defined by eq. (5.7) and label T stands for triangle. Using relations of Appendix C.1.1 (with P = T ), we can translate eq. (C.40) to the notation adopted in the current work: .

D Schwarz derivative
Schwarz derivative of function z (w) is defined by expression where the prime stands for derivative d/dw: Schwarz derivative has the property

E.1 Green function
It is well known that Green function of the two-dimensional Laplace operator in simply connected region U with Dirichlet boundary conditions can be expressed via the conformal mapping of the upper complex semiplane to region U . We denote this Laplace operator ∆ (C) where C is the boundary of U . Let be a representation of this conformal mapping in terms of real coordinates w 1 , w 2 on the semiplane and real coordinates x 1 , x 2 in region U . Then Green function x ′ | [∆ (C)] −1 |x for region U and Green function for the semiplane w ′ | ∆ −1 semiplane |w are connected by relation Green function for the semiplane can be constructed by the well-known image method from the Green function on the plane: Here w R is the reflection of point w with respect to w 1 -axis: This result can be rewritten using complex coordinates on the upper semiplane of complex z and for region U . Then where ζ = f (z) , (E.10) and f is conformal mapping from the semiplane to U corresponding to mapping f formulated in terms of points with real coordinates.
be the inverse mapping. Then eq. (E.9) takes the form Differentiating eq. (E.13), we find (E.14) The second differentiation gives (E.15) Next we set and expand the RHS in powers of η This can be expressed via Schwarz derivative (see eq. (D.1) in Appendix D ): Since φ describes conformal mapping z = φ (ζ), we can identify Combining this with eq. (4.18), we obtain eq. (4.20).

F SC mapping for triangles
In this Appendix we apply SC equation (5.21) with one vertex at infinity to the case of triangles and derive some useful relations between geometric parameters and SC parameters. The case of triangles corresponds to setting n v = 3 in eq. This result can be easily generalized to the case of arbitrary z 1 , z 2 : . Using the representation for the factor In this Appendix we compute function Π P (α, {γ 1 , γ 2 } , {z 1 , z 2 }) for the special value of argument α = 1. For simplicity we choose z 1 = 0, z 2 = 1. We will perform a formal calculation ignoring divergences of some intermediate integrals but still leading to the correct final result. A rigorous justification of this formal calculation requires methods whose discussion makes sense in the context of the theory of functions Π (n) P with arbitrary n, i.e. in the case of the theory of analytical regularization for arbitrary polygons C.
We start from integral