Polynomial Closed-Form Model (PCFM)

• PCFM is a framework that expresses complex analytical problems in a polynomially tractable closed form across diverse fields.
• In optical communications, PCFM approximates spatial power profiles to analytically solve nonlinear interference integrals, significantly enhancing computational speed.
• In finance and logic, PCFM underpins models with invariant polynomial processes, enabling explicit closed-form evaluations of moments and model counts.

Searching arXiv for the cited PCFM papers and related uses of the term. Polynomial Closed-Form Model (PCFM) is a field-dependent label used in several arXiv literatures for model classes in which a difficult analytical object is replaced by a polynomially parameterized or polynomially tractable closed form. The most recent and technically specific usage is in optical communications, where PCFM denotes a closed-form reformulation of the Generalized Nonlinear (GN) model obtained by approximating each channel’s span-wise spatial power profile with a low-order polynomial and then evaluating the nonlinear-interference integrals analytically (Jiang et al., 23 Jan 2026, Poggiolini et al., 29 Aug 2025). Earlier usages attach the same label, or an equivalent expository description, to polynomial-process models in stochastic portfolio theory and power prices, and to polynomial-size closed forms for weighted first-order model counting in the two-variable fragment FO² (Cuchiero, 2017, Filipovic et al., 2017, Malhotra et al., 2020, Malhotra et al., 2021).

1. Scope of the term

The term appears in multiple, otherwise unrelated, research areas. This indicates that PCFM is not a single universally standardized framework, but a recurrent naming pattern for constructions in which polynomial structure yields analytically explicit formulas.

Domain	Core object	Closed-form mechanism
Optical fiber transmission	GN/EGN nonlinear-interference integrals	Polynomial spatial power profiles along each span
Stochastic portfolio theory	Joint process of market weights and capitalizations	Generator preserves polynomial spaces
Power-price modeling	Polynomial diffusion factors and spot-price map	Finite-dimensional generator matrix exponential
FO² weighted model counting	WFOMC over lifted interpretations and type counts	Polynomial-size sum over count vectors
Polynomial equations	Roots of a univariate polynomial	Contour-integral representation in elementary functions

In the optical-communications usage, the model is designed for wavelength-division multiplexed transmission over links with distributed Raman amplification, lumped losses, hybrid schemes, and ultrawide-band operation with inter-channel stimulated Raman scattering (ISRS) (Jiang et al., 23 Jan 2026). In the finance and energy literatures, the defining feature is the polynomial-process property, namely invariance of polynomial spaces under the generator or semigroup (Cuchiero, 2017, Filipovic et al., 2017). In the logic literature, PCFM refers to a closed-form expression for weighted first-order model counting whose size is polynomial in the domain cardinality for fixed FO² formulas (Malhotra et al., 2020, Malhotra et al., 2021).

2. Optical-communications PCFM: formulation within the GN model

In optical transmission, the propagation of WDM signals is limited by nonlinear interference (NLI) arising from the Kerr effect. Within the GN model, the self-channel, cross-channel, and multi-channel interference integrals become cumbersome when realistic span-by-span power evolution is included. The optical PCFM addresses this by representing, within each span, the spatial power profile of each channel as a polynomial in the fiber-length coordinate (Jiang et al., 23 Jan 2026).

The basic approximation is

$P_k(z)=\sum_{n=0}^{N} a_{k,n}\,z^n,$

or, in normalized form,

$$p_i^{(n_s)}(z)\approx \sum_{k=0}^{K} a_{i,k}^{(n_s)} z^k,\qquad a_{i,0}^{(n_s)}=1.$$

The coefficients are determined from the known gain/loss profile and launched powers, either by fitting or by analytic expansion. The truncation order is chosen to balance approximation error against algebraic complexity; one formulation states that in practice $N\le 4\text{–}6$ suffices even for strongly varying profiles, while the UWB formulation reports $K=4\text{–}8$ as effective in typical C+L+S spans with backward Raman pumps and ISRS (Jiang et al., 23 Jan 2026, Poggiolini et al., 29 Aug 2025).

After frequency-domain manipulations, the remaining object is a core integral of the form

$$K_x=\int_{f_1}\int_{f_2}\left|\int_0^L p_x(z)e^{j\Phi(f_1,f_2)z}\,dz\right|^2 df_1df_2,$$

with $p_x(z)$ equal to the product of the relevant channel power profiles and $\Phi(f_1,f_2)=4\pi^2\beta_{2,\mathrm{eff}}f_1f_2$. Substituting the polynomial expansion yields

$$K_x=\sum_{n=0}^{N}\sum_{m=n}^{N}(2-\delta_{nm})\,p_{n,x}p_{m,x}\,I_{nm},$$

where $\delta_{nm}$ is the Kronecker delta and $I_{nm}$ reduces to closed-form expressions involving frequency kernels $$p_i^{(n_s)}(z)\approx \sum_{k=0}^{K} a_{i,k}^{(n_s)} z^k,\qquad a_{i,0}^{(n_s)}=1.$$0 and special functions including sine integrals, cosine integrals, Beta functions, and, depending on the case, $$p_i^{(n_s)}(z)\approx \sum_{k=0}^{K} a_{i,k}^{(n_s)} z^k,\qquad a_{i,0}^{(n_s)}=1.$$1 (Jiang et al., 23 Jan 2026).

Once $$p_i^{(n_s)}(z)\approx \sum_{k=0}^{K} a_{i,k}^{(n_s)} z^k,\qquad a_{i,0}^{(n_s)}=1.$$2 is available for each SCI, XCI, and MCI island, the per-span NLI power spectral density follows from the standard GN prefactors:

$$p_i^{(n_s)}(z)\approx \sum_{k=0}^{K} a_{i,k}^{(n_s)} z^k,\qquad a_{i,0}^{(n_s)}=1.$$3

Summation over all islands and spans yields the total NLI PSD at the channel under test. The 2026 paper emphasizes that the result is a generic closed-form expression for all self-, cross-, and multi-channel contributions (Jiang et al., 23 Jan 2026).

3. Assumptions, coefficient determination, and computational profile in the optical model

The optical PCFM retains intra-span coherent accumulation and full frequency-dependent phase evolution, explicitly contrasting itself with approximations that neglect intra-span coherence or use infinite-series expansions. Its assumptions are local to the derivation rather than global to the architecture. The effective dispersion $$p_i^{(n_s)}(z)\approx \sum_{k=0}^{K} a_{i,k}^{(n_s)} z^k,\qquad a_{i,0}^{(n_s)}=1.$$4 is assumed constant within each island rectangle, with higher-order dispersion piecewise incorporable by splitting islands. The original derivation assumes rectangular channel spectra and fixed channel spacings, but the polynomial-profile approach can be combined with more general spectral shapes through modified frequency-integration limits (Jiang et al., 23 Jan 2026).

The polynomial coefficients $$p_i^{(n_s)}(z)\approx \sum_{k=0}^{K} a_{i,k}^{(n_s)} z^k,\qquad a_{i,0}^{(n_s)}=1.$$5 may be obtained by a truncated Taylor expansion of the exact Raman-ODE solution or by a least-squares fit of the numerical spatial power profile over the span. For Raman-amplified spans, closed-form expressions for the first few coefficients can be written in terms of forward-pump power $$p_i^{(n_s)}(z)\approx \sum_{k=0}^{K} a_{i,k}^{(n_s)} z^k,\qquad a_{i,0}^{(n_s)}=1.$$6, backward-pump power $$p_i^{(n_s)}(z)\approx \sum_{k=0}^{K} a_{i,k}^{(n_s)} z^k,\qquad a_{i,0}^{(n_s)}=1.$$7, the Raman gain spectrum, and the fiber loss $$p_i^{(n_s)}(z)\approx \sum_{k=0}^{K} a_{i,k}^{(n_s)} z^k,\qquad a_{i,0}^{(n_s)}=1.$$8. For EDFA-only spans, $$p_i^{(n_s)}(z)\approx \sum_{k=0}^{K} a_{i,k}^{(n_s)} z^k,\qquad a_{i,0}^{(n_s)}=1.$$9 is exactly represented by an infinite series, and truncation at $N\le 4\text{–}6$0 is reported to yield excellent accuracy (Jiang et al., 23 Jan 2026).

The computational advantage is central. One formulation states that direct numerical evaluation of the four-fold GN integral scales as $N\le 4\text{–}6$1, whereas PCFM reduces the calculation to $N\le 4\text{–}6$2 algebraic terms plus $N\le 4\text{–}6$3 special-function calls per island, with a near-$N\le 4\text{–}6$4 speedup in typical high-baud-rate, 80–100-spans links. The application scope listed for the 2026 formulation includes 10 THz of contiguous spectrum with ISRS, backward Raman, lumped and distributed gain; rapid per-span NLI accumulation for network-planning “what-if” studies; and real-time re-evaluation in software-defined networks using look-up-table data for the polynomial coefficients (Jiang et al., 23 Jan 2026).

The 2025 UWB predecessor provides explicit validation figures. In single-span C+L+S tests on 100 km standard SMF with backward Raman pumps, ISRS across 150 × 100 GBaud channels, and lumped splice losses at arbitrary positions, $N\le 4\text{–}6$5 yields per-channel SPP errors $N\le 4\text{–}6$6. With $N\le 4\text{–}6$7, the PCFM error distribution for $N\le 4\text{–}6$8 has $N\le 4\text{–}6$9 dB and systematic bias $K=4\text{–}8$0 dB; after applying the standard ML-based correction $K=4\text{–}8$1, the bias vanishes and $K=4\text{–}8$2 shrinks further. Under multiple lumped losses, predictions remain within $K=4\text{–}8$3 dB of the numerically integrated EGN benchmark, and full UWB NLI estimation is reported in $K=4\text{–}8$4 ms rather than seconds (Poggiolini et al., 29 Aug 2025).

4. Polynomial-process PCFM in stochastic portfolio theory and power prices

A distinct PCFM tradition is built on polynomial processes. In Cuchiero’s stochastic-portfolio framework, a process $K=4\text{–}8$5 on a closed state space $K=4\text{–}8$6 is polynomial if, for each $K=4\text{–}8$7, the space $K=4\text{–}8$8 of polynomials of total degree $K=4\text{–}8$9 is invariant under the semigroup, equivalently if the extended generator maps $$K_x=\int_{f_1}\int_{f_2}\left|\int_0^L p_x(z)e^{j\Phi(f_1,f_2)z}\,dz\right|^2 df_1df_2,$$0 into itself. This yields linear drift and quadratic diffusion characteristics in the diffusion case and corresponding polynomial conditions with jumps (Cuchiero, 2017).

The market model uses absolute capitalizations $$K_x=\int_{f_1}\int_{f_2}\left|\int_0^L p_x(z)e^{j\Phi(f_1,f_2)z}\,dz\right|^2 df_1df_2,$$1, relative weights $$K_x=\int_{f_1}\int_{f_2}\left|\int_0^L p_x(z)e^{j\Phi(f_1,f_2)z}\,dz\right|^2 df_1df_2,$$2, and total capitalization $$K_x=\int_{f_1}\int_{f_2}\left|\int_0^L p_x(z)e^{j\Phi(f_1,f_2)z}\,dz\right|^2 df_1df_2,$$3. The joint process $$K_x=\int_{f_1}\int_{f_2}\left|\int_0^L p_x(z)e^{j\Phi(f_1,f_2)z}\,dz\right|^2 df_1df_2,$$4, or equivalently $$K_x=\int_{f_1}\int_{f_2}\left|\int_0^L p_x(z)e^{j\Phi(f_1,f_2)z}\,dz\right|^2 df_1df_2,$$5, is polynomial and Markov if and only if $$K_x=\int_{f_1}\int_{f_2}\left|\int_0^L p_x(z)e^{j\Phi(f_1,f_2)z}\,dz\right|^2 df_1df_2,$$6 is a polynomial diffusion on the simplex $$K_x=\int_{f_1}\int_{f_2}\left|\int_0^L p_x(z)e^{j\Phi(f_1,f_2)z}\,dz\right|^2 df_1df_2,$$7 and $$K_x=\int_{f_1}\int_{f_2}\left|\int_0^L p_x(z)e^{j\Phi(f_1,f_2)z}\,dz\right|^2 df_1df_2,$$8 is an independent one-dimensional polynomial diffusion on $$K_x=\int_{f_1}\int_{f_2}\left|\int_0^L p_x(z)e^{j\Phi(f_1,f_2)z}\,dz\right|^2 df_1df_2,$$9. The weights follow a multivariate-Jacobi, Wright–Fisher-type diffusion with an admissible simplex parameter set, and $p_x(z)$0 solves an affine-CIR or pure Black–Scholes equation (Cuchiero, 2017).

The closed-form content is finite-dimensional and semigroup based. For every multi-index $p_x(z)$1 of total degree $p_x(z)$2,

$p_x(z)$3

where $p_x(z)$4 is the restriction of the generator to $p_x(z)$5. This gives exact conditional mixed moments and supports closed-form analysis of capital-distribution statistics. The same framework also yields explicit conditions for local-martingale deflators, boundary non-attainment, and strong relative arbitrage. In particular, strong relative arbitrage over $p_x(z)$6 exists if and only if there is an index $p_x(z)$7 with $p_x(z)$8 somewhere on the face $p_x(z)$9 and

$\Phi(f_1,f_2)=4\pi^2\beta_{2,\mathrm{eff}}f_1f_2$0

and optimal strategies admit an explicit matrix-exponential representation (Cuchiero, 2017).

In energy-price modeling, the same polynomial-process principle is applied under the pricing measure $\Phi(f_1,f_2)=4\pi^2\beta_{2,\mathrm{eff}}f_1f_2$1. A one-factor or two-factor polynomial diffusion $\Phi(f_1,f_2)=4\pi^2\beta_{2,\mathrm{eff}}f_1f_2$2 is mapped into spot prices by a polynomial or affine function,

$\Phi(f_1,f_2)=4\pi^2\beta_{2,\mathrm{eff}}f_1f_2$3

and forward prices satisfy

$\Phi(f_1,f_2)=4\pi^2\beta_{2,\mathrm{eff}}f_1f_2$4

For one-factor Jacobi models fitted to Alberta power prices, the best Bayesian-Information-Criterion score is reported at degree 5 in 1998–2000 and degree 4 in 2010–2014. Two-factor variants, including regime-switching and double-Jacobi specifications, achieve significantly better BIC scores than the one-factor case and separate “spike” and “base” regimes more naturally (Filipovic et al., 2017).

5. PCFM as a closed form for weighted first-order model counting in FO²

In the logic literature, PCFM denotes a reformulation of weighted first-order model counting,

$\Phi(f_1,f_2)=4\pi^2\beta_{2,\mathrm{eff}}f_1f_2$5

as a polynomial-size expression in the domain size $\Phi(f_1,f_2)=4\pi^2\beta_{2,\mathrm{eff}}f_1f_2$6 for fixed sentences in the two-variable fragment FO². The objective is domain-liftability: computation in time polynomial in $\Phi(f_1,f_2)=4\pi^2\beta_{2,\mathrm{eff}}f_1f_2$7 rather than exponential enumeration over interpretations (Malhotra et al., 2020, Malhotra et al., 2021).

The construction is based on lifted interpretations and type counts. For a universal FO² sentence $\Phi(f_1,f_2)=4\pi^2\beta_{2,\mathrm{eff}}f_1f_2$8, domain elements are grouped by unary types, with counts $\Phi(f_1,f_2)=4\pi^2\beta_{2,\mathrm{eff}}f_1f_2$9, and ordered or unordered pairs are grouped by admissible binary patterns. The unweighted universal-fragment formula has the form

$$K_x=\sum_{n=0}^{N}\sum_{m=n}^{N}(2-\delta_{nm})\,p_{n,x}p_{m,x}\,I_{nm},$$0

where $$K_x=\sum_{n=0}^{N}\sum_{m=n}^{N}(2-\delta_{nm})\,p_{n,x}p_{m,x}\,I_{nm},$$1 and $$K_x=\sum_{n=0}^{N}\sum_{m=n}^{N}(2-\delta_{nm})\,p_{n,x}p_{m,x}\,I_{nm},$$2 for $$K_x=\sum_{n=0}^{N}\sum_{m=n}^{N}(2-\delta_{nm})\,p_{n,x}p_{m,x}\,I_{nm},$$3. In the weighted setting, symmetric predicate weights multiply each term, and the later formulation generalizes further to arbitrary nonnegative weight functions $$K_x=\sum_{n=0}^{N}\sum_{m=n}^{N}(2-\delta_{nm})\,p_{n,x}p_{m,x}\,I_{nm},$$4 on the global count vector (Malhotra et al., 2020, Malhotra et al., 2021).

Cardinality constraints are incorporated by restricting the admissible count vectors. Existential quantifiers in Scott normal form are handled by inclusion–exclusion after introducing fresh unary predicates for “bad” witnesses, yielding an alternating sum of universal-case counts. The 2021 extension also handles counting quantifiers $$K_x=\sum_{n=0}^{N}\sum_{m=n}^{N}(2-\delta_{nm})\,p_{n,x}p_{m,x}\,I_{nm},$$5 by a gadget construction with fresh predicates and additional cardinality constraints, together with a division by $$K_x=\sum_{n=0}^{N}\sum_{m=n}^{N}(2-\delta_{nm})\,p_{n,x}p_{m,x}\,I_{nm},$$6 to remove ordering of witness predicates (Malhotra et al., 2021).

The complexity statement is explicit: for any fixed FO² sentence, the numbers of unary types, binary types, Skolem predicates, and counting gadgets are constants independent of $$K_x=\sum_{n=0}^{N}\sum_{m=n}^{N}(2-\delta_{nm})\,p_{n,x}p_{m,x}\,I_{nm},$$7. Accordingly, the number of summation indices is polynomial in $$K_x=\sum_{n=0}^{N}\sum_{m=n}^{N}(2-\delta_{nm})\,p_{n,x}p_{m,x}\,I_{nm},$$8, each term is computable with constant-size table lookups and arithmetic, and the full closed form is polynomial-time in the domain size. The 2020 paper further emphasizes that the resulting weight-function family is strictly larger than symmetric weights, because one may assign arbitrary nonnegative values directly to the global count vectors $$K_x=\sum_{n=0}^{N}\sum_{m=n}^{N}(2-\delta_{nm})\,p_{n,x}p_{m,x}\,I_{nm},$$9 (Malhotra et al., 2020, Malhotra et al., 2021).

6. Other usage and the meaning of “closed form”

A further expository usage applies the label PCFM to the contour-integral representation of roots in “Closed-form solution of polynomial equations.” There the roots of a univariate polynomial with simple zeros are represented via Cauchy’s theorem and contour deformation, after shifting the polynomial so that all zeros lie in $\delta_{nm}$0, introducing branch cuts, and defining single-valued functions on the slit plane. Each root is then expressed through residues at auxiliary poles together with convergent real-axis integrals built from elementary functions (Kheyfits, 2018).

Across these usages, the quantity rendered “closed form” differs substantially. In optical communications, it is the GN/EGN nonlinear-interference integral, expressed through elementary functions and special functions such as $\delta_{nm}$1, $\delta_{nm}$2, Beta functions, and $\delta_{nm}$3 (Jiang et al., 23 Jan 2026, Poggiolini et al., 29 Aug 2025). In stochastic portfolio theory and power prices, it is the conditional moment or forward-price semigroup, realized by a finite-dimensional matrix exponential (Cuchiero, 2017, Filipovic et al., 2017). In FO² model counting, it is a finite sum over type-count vectors whose size grows polynomially with the domain cardinality (Malhotra et al., 2020, Malhotra et al., 2021). In the polynomial-equation setting, it is an integral representation rather than a radical formula in general degree (Kheyfits, 2018).

This suggests that the unifying content of the acronym is methodological rather than ontological: polynomial structure is used to compress an otherwise high-dimensional analytical problem into an explicit finite-dimensional representation. The optical PCFM currently provides the most concrete contemporary meaning of the term, especially because it is explicitly named as such in recent arXiv work and developed as a practical tool for SCI, XCI, and MCI evaluation in ultra-wideband optical systems (Jiang et al., 23 Jan 2026, Poggiolini et al., 29 Aug 2025).