Power series with Taylor coefficients of sum-product type and algebraic differential equations /SP-series and the interplay between their resurgence and differential properties Shweta Sharma 2 1 0 2 n Contents a J 5 1 Introduction 2 2 1.1 Definitions and basics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 ] 1.2 Definition and calculation of the inner generators. . . . . . . . . . . . . . 3 A 1.3 Definition and calculation of the outer generators. . . . . . . . . . . . . . 4 C 1.4 Singularities inferred from Taylor coefficient analysis. . . . . . . . . . . . 4 . h t a 2 Differential aspects 6 m 2.1 Variable and covariant linear homogeneous polynomial ODEs . . . . . . 7 [ 2.2 Method for calculating the differential equation . . . . . . . . . . . . . . 8 1 2.2.1 The Differential Equations: . . . . . . . . . . . . . . . . . . . . . . 9 v 2.2.2 Dimensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 6 6 2.2.3 The equivalence: (∂ ,−ν) ⇐⇒ (n−1,∂ ) . . . . . . . . . . . . . . 10 ν n 2 2.2.4 Compressing the covariant ODEs . . . . . . . . . . . . . . . . . . 11 5 . 2.2.5 Basic polynomials f(x) and p(ν) . . . . . . . . . . . . . . . . . . . 11 1 0 2.2.6 Basic symmetric functions . . . . . . . . . . . . . . . . . . . . . . 11 2 2.3 Covariant ODEs and leading covariant polynomials . . . . . . . . . . . . 11 1 : 2.3.1 Covariant ODEs . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 v 2.3.2 Leading covariant polynomials Λ(ν) . . . . . . . . . . . . . . . . . 12 i X 2.3.3 Characterisation of the invariance and covariance . . . . . . . . . 12 r a 2.3.4 Invariants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 2.3.5 Canonical covariant polynomials of degree 1 . . . . . . . . . . . . 13 2.4 The global resurgence picture for the polynomial inputs f . . . . . . . . . 13 3 Rational inputs F : the resurgence pattern and connection matrices. 13 3.1 Matrice entries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 3.2 T and γ polynomials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23 4 ODEs: numerical calculations 28 4.1 Likely non-existence of ODEs for general rational inputs . . . . . . . . . 28 1 5 References. 29 1 Introduction The present article is anelaboration with numerical details and some explicit calculations of the chapter 6 of the long paper [ES]. The article is essentially based on the differential properties of the inner generators that occur while handling the SP series (sum product series, a reminder of the definition is given below). The inner generators introduced in [ES] not always but often satisfy linear homogeneous ordinary differential equation. The article is structured as follows: First there are just some reminders about basic definitions used in SP series. This serves as a preliminary and a way to settle notation and not the least in the exposition of the problem at hand. Section 2 is about the two types of differential equation - variable and covariant. We give a schematic description of the existence of such ODEs and recommend [ES] for a more formalistic view. The next section is about resurgence and connection matrices that appear in our treatment of the problem. Some of the matrices are explicitely written down for different values of p, the parameter in our driving function f. In passing we give a list of the T and γ polynomials which are related to ODEs of infinite order. The last section talks about the main numerical result of the article i.e. the non existence of differential equations for rational F inputs in a global perspective. In this series, I’m writing another article [SS] which would be soon finished. It would be essentially involved in treating examples of arbitrary functions F of various types which always fall in the category of the series of the type sum-product (SP) to be treated theoretically as well as numerically, the study of the various singularities that occur, how they could be retrieved using the machinery that is detailed later (using the symmetry between the series and the Taylor coefficient expansion asymptotics) their placement and the resurgence equation and the resurgence that helps us to retrieve the Riemann surface in its complete description using the alien derivatives over the series and the Taylor expansion. We go through some of the analytic properties that we would exploit. For a polynomial f or a rational function F (resp. a trigonometric polynomial) and Taylor coefficients J(n) defined only by pure products the series in the ζ-plane would be of hypergeometric (q-hypergeometric) type. Hence, the theory of SP series links two Q important theories. This problem started with a problem of the study of the series associated to knots[CG3] 1.1 Definitions and basics The sum-product type series are essentially Taylor series of the form j(ζ) := J(n)ζn n≥0 X whose coefficients are of the sum product (SP) type: k k J(n) := F( ) = exp − f( ) n n ! ǫ≤m≤nǫ≤k≤m ǫ≤m≤n ǫ≤k≤m X Y X X 2 for ǫ ∈ {0,1} The two driving functions F and f are related as follows: F ≡ exp(−f). The summation always starts at ǫ = 0 unless F(0) ∈ {0,∞} (here it starts at ǫ = 1). F is meromorphic and special care is to be taken in case the function is holomorphic A few things that need attention are the outer/inner dichotomy and the ingress factor [ES]. The SP series lead to two different types of singularities (generators) as under the alien derivation this generates the resurgence algebra of the SP series. The generators that occur in this problem are of the following types: • the inner generators, so called as they occur indefinetely under alien derivation but never produce the outer generators. • The outer generators, as they occur one, never twice (they never occur under re- peated alien derivation) • The other two generators are less important for the problem, they are the original generators which is the SP series itself and the exceptional generators which do not appear naturally, but which are a very useful tool for the exploration of the Riemann surface, since (i)they produce all inner generators under alien derivation (ii) their base point can be moved around at will and taken as close as one wishes to the base point of any inner generator. 1.2 Definition and calculation of the inner generators. The nir-transform The inner generators are given by the double integral: nir: f → h 1 c+i∞ dn ∞ h(ν) = exp(nν) exp#(−f⇑β∗(n,τ))dτ 2πi n Zc−i∞ Z0 1 c+i∞ dn ∞ n−κτκ+1 = exp(nν) exp(−f )exp (−f↑β∗(n,τ))dτ 2πi n κ κ+1 # Zc−i∞ Z0 where exp (X) (resp. exp#(X)) denotes the exponential expanded as a power series of # X (resp. minus its leading term). β∗ = Bk+1(τ+21) for the standard case, where B is the k k+1 k kth Bernoulli polynomial and f⇑β∗(n,τ) and f↑β∗(n,τ) are defined as follows: f(x) = f xk (κ ≥ 1, f 6= 0) k k k≥κ X τ ∂ f⇑β∗ := β(∂ )f( ) with ∂ := τ τ n ∂τ n−κτk+1 := f n−kβ∗(τ) := f +f↑β∗(n,τ) k k κ k +1 k≥κ X In the integral expression for h(ν), we first perform the term by term ramified Laplace integration in τ followed by a term by term Borel integration in n. 3 Theinner generators alsoadmitamorecomplicatedanalyticalexpression, asasucces- sion of nine successive operations, all of which are more or less elementary (though often highly non-linear) except for the crucial step, the mir-transform, which is an integro- differential operator of infinite order (with rational coefficients): see section 4.2 [ES]. 1.3 Definition and calculation of the outer generators. The nur-transform The outer generators can be viewed as infinite sums of inner generators and the inner generators (see section 1.2) but they are also capable of a direct construction. Altogether, we have the choice between three equivalent definitions. 1. The first expression: nur: f → h 1 c+i∞ dn ∞ h(ν) = exp(nν) exp#(−f⇑β∗(n,m)) 2πi n Zc−i∞ m=X12+N 1 c+i∞ dn ∞ n−κmκ+1 = exp(nν) exp(−f )exp (−f↑β∗(n,m)) 2πi n κ κ+1 # Zc−i∞ m=X12+N is analogous to the double integral in the expression of the nir-transform, but the second integral is “discretised”, i.e. replaced by an infinite series. exp (X) (resp. exp#(X)) denotes the exponential function expanded as a power # series of X (resp.X minus the leading term of X, which remains within the ex- ponential). β∗ = Bk+1(τ+12) for the standard case, where B is the kth Bernoulli k k+1 k polynomial and the function f⇑β∗(n,τ) and f↑β∗(n,τ) are defined as follows: τ ∂ f⇑β∗ := β(∂ )f( ) with ∂ := τ τ n ∂τ n−κτk+1 =: f +f↑β∗(n,τ) κ k +1 2. The second, more analytical expression of nur, consists in a succession of four (rather than nine) steps, three of which are elementary, but the fourth (the first, in fact) not at all, see section 5.2 [ES]. 3. Then again nur can be expressed as an infinite sum of nir-transforms. nur(f) = (−1)pnir(2πip+f) p∈Z X 1.4 Singularities inferred from Taylor coefficient analysis. When dealing with a SP series we are led to calculate the singularities using the asymp- totics of its Taylor coefficients. Keeping the symmetry between the SP series φ and its Taylor coefficients J intact, we look at them as resurgent functions in variables z and n 4 and ζ and ν for different models that are explained in [ES]. For further details on this section [SS] The idea is to follow the correspondence between the triplets: {φ˜(z),φ(ζ),φ(z)} ↔ {J˜(n),J(ν),J(n)} and that of the alien derivatives that are linked to them. Retrieving the closest singularities: ˆ Consider that our series φ has one singularity at the boundary of the disk of convergence, which we denote by ζ . 0 φˆ(z) = (n+1)!J(n)z−n−1(dνt) 0≤n X on applying the Borel transform gives φˆ(ζ) = J(n)ζn 0≤n X To write down the closest singularity of φˆ from the closest singularity of Jˆ , we write J(n) as a cauchy integral on a circle |ζ| = |ζ |−ǫ. 0 1 J(n) = φˆ(ζ)ζ−n−1dζ 2πi I By a contour deformation : 1 = φˆ(ζ)ζ−n−1dζ +o(ζ−n) 2πi 0 ZΓ putting ζ := ζ expν = expν+ν0, we obtain 0 1 J(n) = φˆ(ζ expν)exp−nνdζ +o(exp−nν0) 0 2πi ZΓ∗ ˆ ˇ Using identities between φ and φ 1 J(n) = φˇ (ζ −ζ expν)exp−nν dζ +o(exp−nν0) 2πi ζ0 0 0 ZΓ∗ 1 J(n) = φˇ (ζ −ζ exp−ν)expnνdζ +o(exp−nν0) 2πi ζ0 0 0 ZΓ∗ ⋄ and in the case φ is integrable (that is to say in the case when the singularity at ζ is ζ0 0 entirely recoverable from its minor i.e. free from Laurent terms), we arrive at: c φˆ (ζ expν −ζ )exp−nνdζ +o(exp−nν0) ζ0 0 0 Z0 In the above calculation we consider the contour Γ which is the deformation of the circle |ζ| = |ζ |−ǫ to meet the larger circle |ζ| = |ζ |+ǫ avoiding the interval [ζ,ζ +ǫ]. 0 0 Γ is the contour Γ after undergoing ζ = ζ expν. The other distant singularities ∗ 0 could be obtained by extending this procedure. An important remark here is that if Jˆis endlessly continuable, its resurgence pattern can be very precisely obtained from that of ˆ the φ. 5 2 Differential aspects The inner generators for some driving functions (for instance all polynomial f and all monomialF)satisfyordinarydifferential equationoflinearhomogeneoustypeswithpoly- nomial coefficients of usually high order. These equations can be particularily interesting as they give a rather explicit explanation of the inner generators which occur in our prob- lem in an algebraic manner. They enable a detailed understanding of their behaviour at ∞ in the ν plane or at the origin in the Borel ζ-plane. What is a bit more appealing than the rest is the occurence of the phenomenon of rigid resurgence which involves discrete Stokes constants in the asymptotic expansions. The reason it might be called rigid is that it exhibits a certain stability to the changing parameters of the problem: contrary to what happens with most singular ODEs, here the resurgence coefficients are discrete; they don’t vary continously with the real or complex parameters inside the ODE. More detail can be found in [ES] In the developpement of the differential properties of the inner generators, we would talk about the so called “variable” and the “covariant” differential equations. The “vari- able” equation applies to the whole nir-transform, whereas the “covariant” equation ap- plies to the non-entire part (usually semi-entire part) of the nir-transform, which alone has intrinsic geometric meaning: it describes the singularity present at this point. As a consequence, the “variable” equation is specific to inner generator, whereas the “co- variant” equation applies equally to all inner generators under a shift of base point. In what follows, we would follow closely the sketch drawn for the treatment of this problem in chapter 6 of [ES]. The notations stay the same and to illustrate the problem and to be able to handle this section efficiently, we would like to write down a reminder of the basics. Let’s start with writing down the four types of shift operators β(∂ ) to settle notation: τ trivial choice β(τ) := τ−1 standard choice β(τ) := (expτ/2 −exp−τ/2)−1 odd choice β(τ) := τ−1 + β τ2s+1 2s+1 s≥0 X general choice β(τ) := β τ s s s≥0 X Consider the driving function f. It can be subjected to thenir-transform provided f(0) = 6 0 and more particularily f′(0) 6= 0. f(x) := f xs s 1≤s≤r X τ 1τ2 φ(n,τ) := β(τ)f( ) = f +... 1 n 2 n φ(n,τ) := φ+(n,τ)+φ−(n,τ) φ±(n,±τ) := ±φ±(n,±τ) ∞ k(n) := [ exp#(φ(n,τ))dτ] singular Z0 ∞ := exp#(φ(n,τ))cosh (φ−(n,τ))dτ (iff 6= 0) # 1 Z0 1 c+ι∞ k(n) := [ k(n)expνndn] = h′(ν) formal 2πι Zc−ι∞ 1 c+ι∞ dn k(n) := [ k(n)expνn ] = h(ν) formal 2πι n Zc−ι∞ 2.1 Variable and covariant linear homogeneous polynomial ODEs The general form of such equations : variable ODE : P (n,−∂ )ktotal(n) = 0 ⇔ P (∂ ,ν)ktotal(ν) = 0 v n v ν covariant ODE : P (n,−∂ )ksingular(n) = 0 ⇔ P (∂ ,ν)ksingular(ν) = 0 c n c ν where: P (n,ν) = dv npνψ(n,τ)q v p,q k 0≤p≤d0≤qδ X X P (n,ν) = dc npνq c p,q 0≤p≤d0≤qδ X X These polynomials are of degree d and δ in the variables n, ν. Here n and ν should be treated as non-commutative variables subject to the following relation [n,ν] = 1. The relation of covariance is as follows: Pǫf(n,ν −η) ≡ Pf(n,ν) for all ǫ c c with: ǫf(x) = f(x+ǫ) and η := ǫf(x)dx 0 Here we give a little note about the existence and the calculation of these ODEs. The R details can be found [ES]. The polynomials ϕ and ψ in n−1 and τ for any s ∈ N can be s s written down using the following identities: ∞ ∂sktotal(n) = ϕ (n,τ)exp#(ϕ(n,τ))dτ n s Z0 7 with ϕ (n,τ) ∈ C[∂ ϕ,∂2ϕ,...,∂sϕ] ∈ C[n′ −1,τ] s n n n ∞ ∞ ds(τkexp#(ϕ(n,τ))) = ψ (n,τ)exp#(ϕ(n,τ))dτ = 0 τ s Z0 Z0 with ψ (n,τ) = τs∂ ϕ(n,τ)+sτs−1 ∈ C[n−1,τ] s τ For larger degrees the polynomials ϕ and ψ become linearly independant on C[n−1]. s s A (n)ϕ(n,τ)+ B (n)ϕ(n,τ) = 0 for A(n),B(n) ∈ C[n] s s 0X≤s≤δ 0≤Xs≤δ′ and to each such relation there corresponds a linear ODE for ktotal : ( A (n)∂s)ktotal(n) = 0 s n 0≤s≤δ X The expressions given here work for f(0) = 0 also. In this case when we apply the nir transform to the corressponding driving function, we obtain powers that are integer as well as fractional and hence the use of the term ktotal which implies that one considers all powers for an overall analysis of the inner generators. In this sction we talk about the polynomial f inputs for which k and ktotal in both x and ν plane, satisfy ODEs. The ones satisfied by ktotal are called variable as they strictly depend on a proper base point. Whereas the ones satisfied by k are called covariant: for a change of a proper base point x to x , the covariant differential equation satisfied in the ν plane undergoes i j a translation in the ν plane of ν = xif(x)dx. Here x is proper base point means that xj 0 f(x ) = 0. Another important thing about the covariant ODEs is that there exists a 0 R unique extension of the covariant ODE to a non proper base point x . This evidently l does not coincide with the variable ODE. 2.2 Method for calculating the differential equation A rather schematic version of the entire problem. For details for explicit calculations with the technical aspects, the reader is invited to refer to [ES]. There are detailed reports on schemes on the existence and calculation of variable ODEs and the covariant ODEs (f(0) = 0 and f(0) 6= 0). The integrals for the inner generators: h, h′ and hh are as follows: 1 +i∞ dn ∞ h(ν) = enν eφ(n,τ)dτ (singular germ) 2πi n Z−i∞ Z0 1 +i∞ ∞ h′(ν) = enνdn eφ(n,τ)dτ (singular germ) 2πi Z−i∞ Z0 ∞ hh(n) = eφ(n,τ)dτ (divergent power series) Z0 8 for φ(n,τ) := −β(∂ ).f(τ). τ n The four choices for β, the shift operator, are as has been mentioned before in the article, namely trivial, standard, odd and general. Now one can write down the series for h(ν), h′(ν) and hh(n): h(ν) = h νm m mX∈Nk h′(ν) = h mνm−1 m mX∈Nk hh(n) = h m!n−m m mX∈Nk with N := − κ + 1 N k 1+κ 1+κ 2.2.1 The Differential Equations: The inner generators h, h′ and hh verify the differential equations D, D′ and DD re- spectively. (These differential equations are of linear and homogeneous type and fall into both, the variable and covariant category). D = D νd∂δ d,δ ν 0≤d≤d∗, 0≤Xδ≤δ∗ D′ = D′ νd∂δ d,δ ν 0≤d≤d∗, 0≤Xδ≤δ∗ DD = D n−δ∂d = D (−∂ )dn−δ d,τ n d,δ n 0≤d≤d∗, 0≤d≤d∗, 0≤Xδ≤δ∗ 0≤Xδ≤δ∗ The last equality uses the equivalence (∂ ,−ν) ↔ (n−1,∂ ) ν n One observes the following: ∞ (∂ )khh(n) ≡ φ (n,τ)eφk(n,τ)dτ (∀k ∈ N) n k Z0 ∞ ∞ 0 ≡ ∂ [τkeφk(n,τ)]dτ = ψ (n,τ)eφk(n,τ)dτ (∀k ∈ N∗) τ k Z0 Z0 where ψ (n,τ) := τk∂ φ (n,τ)+kτk−1 k τ k Then for large enough values of the d∗ and δ∗, the elementary functions φ (n,τ) for k (k = 0,1,...,d∗) and ψ (n,τ) for (k = 0,1,...,δ∗), are no longer linearly independant. k We use this fact to eliminate the ψ (n,τ) to get an ODE that is satisfied by hh(n). k Check: To verify that a certain test function satisfies an ODE, the way we go about it is to explicitely calculate the nir transform for the function and solve for a differential 9 system with a convenient number of successive derivatives. This is the method that we have used extensively in the course of this article to prove the non existence of elementary (i.e. low-order and low degree) differential equations (even algebraic) for certain class of driving functions f (or F). 2.2.2 Dimensions Dimensions of the spaces of variable ODEs. For r := def(f) and for each pair (x,y) with x ∈ {v,c} = variable,covariant y ∈ {t,s,o,g} = {trivial,standard,odd,general} It turns out that in each case, the dimension of the space of the ODEs is of the form: dim (r,d,δ) ≡ (d−A (r))(δ −B (r))−C (r) x,y x,y x,y x,y forδ andd, thedifferential orderoftheODEsinthen-variable(anddfortheν -variable). The extremal pairs (d, −δ) and (−d,δ) for d = 1+A (r) x,y d = 1+A (r)+C (r) x,y x,y δ = 1+B (r) x,y δ = 1+B (r)+C (r) x,y x,y (d and δ is minimal, d and δ is cominimal), the corresponding dimension is always exactly one. The details for the dimensions of the variable and covariant ODEs in the various cases (trivial, standard, odd and general) are specified in the article [ES]. 2.2.3 The equivalence: (∂ ,−ν) ⇐⇒ (n−1,∂ ) ν n The differential operator for both the variable ODEs and the covariant ODEs, can be expressed as polynomials P(n,ν) with degree (d,δ). They are written down as functions in (n,ν). (n,ν) are a non commutating pair of variables with [n,ν] = 1. Hence there are two equivalent parametrisations possible (consistent with the relation of non-commutation) (n,ν) =⇒ (n,−∂ ) (⋆) n (n,ν) =⇒ (∂ ,ν) (⋆⋆) ν The corresponding ODE equivalence can be written as: ˆ P(n,−∂ )k(n) = 0 ⇐⇒ P(∂ ,ν)k(n) = P(∂ ,ν)∂ h(ν) = 0 n ν ν ν 10