--- title: 'Paired Functions: A Conceptual Overview' url: https://www.emergentmind.com/topics/paired-functions type: topic --- # Paired Functions: A Conceptual Overview “Paired functions” is not a single universal notion. In contemporary literature, the expression is used for several non-equivalent constructions: function pairs linked by addition laws or translations, functions related by the Legendre transformation, reproducing pairs of weakly measurable functions, Boolean “paired” functions on the slice, pairs of functions arising in functional data analysis, and pairing functions governing fermionic many-body states and their entanglement spectra [2110.14519][2208.05043][1505.04187][2510.02804][2509.24170][1105.4808]. The common feature is the imposition of a structured relation between two functions or between a function and a canonical partner, but the underlying algebraic, analytic, combinatorial, statistical, and physical meanings differ substantially. ## 1. Pairable functions and Cauchy pairs In the functional-equation literature, a function \(f\) is called pairable with \(g\) if there exists a period function \(T:I^2\to\mathbb{R}\) such that \(g(x)=f(x+T)\) and the pair \((f,g)\) satisfies the sine-addition-type equation \[ f(x+y)=f(x)g(y)+f(y)g(x), \] for all \(x,y\in I\), where \(I\subset\mathbb{R}\) is an interval closed under addition [2110.14519]. The associated cosine-type relation is \[ g(x+y)=g(x)g(y)-f(x)f(y). \] A pair \((f,g)\) is called a Cauchy pair with respect to a given Cauchy functional equation when \(f\), and in some cases \(g\), is a regular solution of an additive, multiplicative, exponential, or logarithmic Cauchy equation [2110.14519]. This framework extends the classical trigonometric model. The paper explicitly identifies classical paired functions such as \((\sin,\cos)\) as instances where the period is constant, while “pairability” permits a non-constant period function and therefore a broader class of solutions [2110.14519]. The sine and cosine representers are given by \[ f_S(x)=g(x)=\frac{f(2x)}{2f(x)}, \qquad f_C(x)=\pm\sqrt{(g(x))^2-g(2x)}. \] The representers are described as uniquely determined away from zeros [2110.14519]. The explicit examples in this literature emphasize that the period function need not be constant. For the additive case \(f(x)=cx\), \[ T(x,y)=\frac{1}{c}-\frac{2xy}{x+y}, \] while for the exponential case \(f(x)=a^x\), pairability is possible only with constant period \[ T=-\log_a 2. \] The paper also states that Euler’s identity is interpreted as a property of certain pairable functions and finite cyclic groups, with the Gamma function and reflection formulas providing the relevant structural link [2110.14519]. This suggests a broad functional-equation viewpoint in which “paired” behavior is encoded by addition theorems and generalized periodicity rather than by a fixed named partner alone. ## 2. Legendre pairs and pair-sharing in complex analysis A second major use of function pairing is Legendre duality. A Legendre-transformation pair consists of functions \(f(x)\) and \(g(m)\) linked by \[ g(m)=\sup_x\{mx-f(x)\}, \] for convex \(f\), or, when \(f'\) is invertible, \[ g(m)=m(f')^{-1}(m)-f((f')^{-1}(m)). \] The paper gives the equivalent parametric representation \[ m=f'(x), \qquad d=xf'(x)-f(x), \] and describes the involution property \[ L\{L\{f(x)\}(m)\}(x)=f(x) \] as a general theorem [2208.05043]. The tabulated examples include quadratic, power, exponential, logarithmic, trigonometric, and special-function pairs. Among the standard entries are \[ f(x)=\frac{x^2}{2}\Rightarrow g(m)=\frac{m^2}{2}, \] \[ f(x)=\frac{x^p}{p}\Rightarrow g(m)=\frac{m^q}{q}, \qquad \frac{1}{p}+\frac{1}{q}=1, \] \[ f(x)=e^x \Rightarrow g(m)=m\ln m-m, \] and \[ f(x)=x\ln x \Rightarrow g(m)=e^{m-1}. \] The same source provides a practical authentication criterion: for a candidate pair, the quantity \[ y=xf'(x)-f(x)-g(f'(x)) \] should vanish identically on the dual domain [2208.05043]. A different complex-analytic notion of pairing appears in Nevanlinna theory. Two meromorphic functions \(f\) and \(g\) are said to share a pair \((a,b)\) if \(f-a\) and \(g-b\) have the same zeros, with the usual modifications when one of the values is \(\infty\) [1411.7199]. The paper classifies pairs \((f,g)\) that share four pairs \((a_\nu,b_\nu)\), \(1\le \nu\le 4\), and a fifth pair \((a_5,b_5)\) counting multiplicities, under the additional condition \[ m(r,1/(f-a_5))+m(r,1/(g-b_5))=S(r). \] Under these hypotheses, either \(f\) and \(g\) are Möbius transformations of each other, or there exist Möbius transformations \(S\) and \(T\) and a non-constant entire function \(h\) such that \[ f=T\circ \hat f\circ h,\qquad g=S\circ \hat g\circ h, \] where \(\hat f,\hat g\) are Gundersen’s prototype functions [1411.7199]. The same work connects the classification to the algebraic curve \[ 4x^3+2xy+y-8x=0 \] and records as open whether the additional analytic condition can be dropped [1411.7199]. Here the paired structure is not duality but shared value-pair geometry. ## 3. Pairing bijections on discrete sets In combinatorics and theoretical computer science, a pairing function is a bijection \(f:\mathbb{N}\times\mathbb{N}\to\mathbb{N}\). One paper describes two general mechanisms for producing infinite families of such bijections: one based on \(n\)-adic valuations, and one based on characteristic functions of subsets of \(\mathbb{N}\) [1301.0129]. The \(n\)-adic mechanism yields a countable family of distinct pairing bijections, while the characteristic-function mechanism yields \(2^\mathbb{N}\) distinct pairing bijections [1301.0129]. The paper also notes that these constructions can be combined to generate families of permutations of \(\mathbb{N}\), and gives the classical \(2\)-adic example \[ z=2^x(2y+1)-1. \] A distinct line of work constructs a symmetric invertible binary pairing function on positive integers, \[ F(m,n)=\frac{1}{4}\left[(m+n-1)^2-(m+n-1)\bmod 2\right]+\min(m,n), \] with the property \(F(m,n)=F(n,m)\) [2105.10752]. The symmetry follows from dependence on \(m+n\) and \(\min(m,n)\), and the paper gives a complete proof of bijectivity together with an explicit inverse formula [2105.10752]. The stated motivation is the encoding of unordered pairs, in contrast with Cantor’s pairing function, which is not symmetric [2105.10752]. Binary proportional pairing functions generalize binary perfect pairing functions to settings in which the two inputs have lengths differing by a fixed proportion [1809.06876]. The base-\(n\) length is defined by \[ \operatorname{len}_n(x)=\lceil \log_n(x+1)\rceil. \] The paper presents a general construction from any non-decreasing unbounded function \(g\), with pairing map \[ \phi_g(x,y)= \begin{cases} y\cdot g^+(y)+x & \text{if } y>g(x),\\ x\cdot(g(x)+1)+y & \text{otherwise}, \end{cases} \] where \(g^+(y)\) is the smallest \(x\) such that \(g(x)\ge y\) [1809.06876]. Specializing to \[ g_{a,b}(x)=(\lfloor x^{1/a}\rfloor+1)^b-1 \] produces the proportional pairing function \[ p_{a,b}(x,y)= \begin{cases} y\,\lfloor y^{1/b}\rfloor^a+x & \text{if } \lfloor y^{1/b}\rfloor>\lfloor x^{1/a}\rfloor,\\ x\,(\lfloor x^{1/a}\rfloor+1)^b+y & \text{otherwise}. \end{cases} \] The paper identifies binary perfect pairing as the special case \(a=b=1\) [1809.06876]. These constructions treat “paired functions” not as a relation between two pre-existing functions, but as explicit encodings of pairs into one variable. ## 4. Reproducing pairs of measurable functions In frame theory and functional analysis, a reproducing pair is a pair \((\psi,\phi)\) of weakly measurable functions from a measure space \((X,\mu)\) into a Hilbert space \(\mathcal H\) such that the sesquilinear form \[ \Omega_{\psi,\phi}(f,g)=\int_X \langle f,\psi_x\rangle \langle \phi_x,g\rangle\,d\mu(x) \] is well defined and bounded, and the associated operator \[ S_{\psi,\phi}f=\int_X \langle f,\psi_x\rangle \phi_x\,d\mu(x) \] belongs to \(GL(\mathcal H)\) [1505.04187]. When \(\psi=\phi\), this reduces to the standard definition of a continuous frame [1505.04187]. A central theorem states that each reproducing pair generates two Hilbert spaces, \(V(\phi,\mu)\) and \(V(\psi,\mu)\), that are conjugate dual to each other with respect to \[ \langle \xi,\eta\rangle_\mu=\int_X \xi(x)\overline{\eta(x)}\,d\mu(x) \] [1505.04187]. The construction proceeds through spaces of measurable coefficient functions \(\xi\) for which the map \[ g\mapsto \int_X \xi(x)\langle \phi_x,g\rangle\,d\mu(x) \] defines a bounded conjugate linear functional, and similarly for \(\psi\) [1505.04187]. The associated synthesis map is \[ T_\phi \xi=\int_X \xi(x)\phi_x\,d\mu(x). \] The later extension to partial inner product spaces treats the case in which the measurable functions take values in a PIP-space rather than strictly in a Hilbert space [1610.03420]. The paper states that the reproducing-pair machinery lifts to lattices or scales of Banach or Hilbert spaces, including examples built from weighted sequence spaces, \(L^p\)-lattices, and Sobolev scales [1610.03420]. In this setting, reproducing pairs generalize continuous frames while allowing the analysis operators to land in spaces more general than \(L^2(X,\mu)\) [1610.03420]. The same source emphasizes that \(S_{\psi,\phi}\) need not be self-adjoint or positive [1610.03420]. ## 5. Paired Boolean functions on the slice A highly specific meaning of “paired functions” appears in the Boolean analysis of the slice \(\binom{V}{k}\), also described as the Johnson scheme [2510.02804]. For \(I\subseteq J\subseteq V\), the basic function \(f_{I,J}\) is the characteristic function of the set of all \(K\in\binom{V}{k}\) with \(I\subseteq K\subseteq J\) [2510.02804]. For disjoint \(I,J\subseteq V\), the paired function is defined by \[ p_{I,J}^{(V,k)}=f_{I,J^\complement}+f_{J,I^\complement}. \] The paper states that paired functions are Boolean except for the trivial case \(I=J=\varnothing\), when the function is identically \(2\) [2510.02804]. The main theorem determines the exact degree of every paired function in terms of \(i=|I|\) and \(j=|J|\): \[ \deg p_{I,J}^{(V,k)}= \begin{cases} i+j-1 & \text{if } i+j \text{ is odd and } i+j \leq \min(k,n-k),\\ k-1 & \text{if } k \text{ is odd, } n=2k, \text{ and } i+j \geq k,\\ \min(i+j,k,n-k) & \text{otherwise}. \end{cases} \] The proof is described as elementary and as not involving any spectral methods [2510.02804]. This construction is noteworthy because, in certain cases, the degree is strictly smaller than the elementary upper bound for the sum of functions [2510.02804]. The paper therefore identifies paired functions as good candidates for fixed-degree Boolean functions of small support size. In the middle layer \(n=2k\), for even degree \(t\notin\{0,k\}\), paired functions provide the smallest known non-zero Boolean functions, surpassing the \(t\)-pencils; for \(p_{T,\emptyset}\) with \(|T|=t+1\), the support size is \[ 2\binom{2k-t-1}{k} \] [2510.02804]. A plausible implication is that the paired construction isolates a cancellation mechanism that is invisible from the naive degree bound on a sum of basic functions. ## 6. Paired functional data and nonparametric inference In statistics, “pairs of functions” arise when each observational unit contributes two related curves. One paper develops sign and signed rank tests for paired functional data, where the basic object is the subject-level difference function \[ D_i(s)=X_{i1}(s)-X_{i0}(s) \] evaluated on a grid \(\mathcal S=\{s_1,\ldots,s_S\}\) [2509.24170]. The functional sign test uses \[ G_i(s)=\operatorname{sign}[D_i(s)] \] and the sufficient-statistic summary \[ t_r[g_i(\mathcal S)]=\sum_{s=1}^S g_i(s), \] or, with zeros allowed, a weighted version \(t_c\) [2509.24170]. After subjectwise summarization, the univariate sign statistic \[ U_r^+=\sum_{i=1}^n 1\!\left[t_r\{g_i(\mathcal S)\}>0\right] \] has \(\operatorname{Binomial}(n,1/2)\) null law under \(H_0\) [2509.24170]. The signed doubly ranked test instead summarizes \[ t_d[d_i(\mathcal S)]=\frac{1}{S}\sum_{s=1}^S \operatorname{sign}[d_i(s)]\,R|d_i(s)| \] and then applies a Wilcoxon signed rank statistic \[ W=\frac{1}{2}\sum_{i=1}^n \operatorname{sign}\!\left[t_d\{d_i(\mathcal S)\}\right]R\!\left|t_d\{d_i(\mathcal S)\}\right|. \] The simulation study varied sample size \(n=15,30,60\), grid size \(S=40,120,360\), Gaussian and \(t\)-processes, and paired-observation correlations \(\rho=0.5,0.75\) [2509.24170]. The reported findings are that the signed doubly ranked test consistently maintained nominal type I error and was more powerful than either functional sign test [2509.24170]. In the randomized crossover heart-health study, with 34 subjects having sufficient data for both conditions, the signed doubly ranked test gave \(p=0.0299\) for increased heart rate under the flight condition, while both sign tests gave \(p\approx 0.058\) [2509.24170]. A complementary model-based approach addresses sparsely observed paired functional data through a reduced-rank mixed-effects model \[ Y_i(t)=\mu(t)+\sum_{j=1}^{k_\alpha} f_j(t)\alpha_{ij}+\epsilon_i(t), \qquad Z_i(t)=\nu(t)+\sum_{j=1}^{k_\beta} g_j(t)\beta_{ij}+\xi_i(t), \] with the association of the two functional variables modeled through the association of the principal component scores [2209.00784]. The score vectors and measurement errors are modeled by multivariate scale mixtures of normals, the mean and principal component functions are represented by splines with roughness penalties, and fitting is performed by an EM algorithm [2209.00784]. The simulation study is reported to show that the proposed method outperforms an existing method that is not designed for robust estimation, and the Type Ia supernovae application records a lower cross-validation error for the robust method in the I-band light curves, \(0.58\) versus \(0.80\) for the normal model [2209.00784]. In this statistical sense, paired functions are not a new function class but a data structure together with inferential procedures that exploit within-subject dependence. ## 7. Pairing functions in many-body physics In fermionic many-body physics, the pairing function is the central object in BCS-type paired states. For fully gapped complex paired superfluids in two dimensions, one paper studies the entanglement spectrum using the BCS form of the ground-state wavefunction on a cylinder, with the pairing function \[ g_{\mathbf k}=\frac{\xi_{\mathbf k}-\sqrt{\xi_{\mathbf k}^2+|\Delta_{\mathbf k}|^2}}{\overline{\Delta}_{\mathbf k}} \] and model pairing functions such as \(\Delta_{\mathbf k}=\widehat{\Delta}(k_x-i k_y)^{|\ell|}\) [1105.4808]. In the \(p+ip\) model case \((\ell=-1)\), the low-lying entanglement spectrum consists of a single chiral Majorana fermion mode per allowed \(k_y\), with pseudo-energy \[ \varepsilon(k_y)=2\ln\!\left(\sqrt{1+(k_y/m^*\widehat{\Delta})^2}+k_y/(m^*\widehat{\Delta})\right), \] and pseudo-Hamiltonian \[ H_{ES}=\sum_{k_y\ge 0}\varepsilon(k_y)\eta_{-k_y}\eta_{k_y} \] [1105.4808]. More generally, in the weak-pairing phase of \(\ell\)-wave paired spinless fermions with odd \(\ell\), the universal low-lying part of the entanglement spectrum consists of \(|\ell|\) chiral Majorana fermion modes, while spin-singlet even-\(\ell\) states yield \(2|\ell|\) modes [1105.4808]. The same paper states that the entanglement gap is infinite for the model pairing functions and diverges logarithmically as the model pairing function is approached [1105.4808]. A closely related construction appears in paired composite-fermion wave functions for the second Landau level [2309.17003]. The BCS ansatz is written as \[ |\Psi_{\mathrm{BCS}}\rangle = \exp\!\left( \frac{1}{2}\sum_{\mathbf k} g_{\mathbf k}\, c^\dagger_{-\mathbf k}c^\dagger_{\mathbf k} \right)|0\rangle, \] and the paired composite-fermion trial state on the sphere is \[ \Psi_{\mathrm{MS}} = \mathrm{Pf}\!\left[g(\mathbf r_i,\mathbf r_j)\right] \prod_{i