Chebyshev Polynomials of the First Kind

• Chebyshev polynomials of the first kind are a family of orthogonal polynomials defined by Tₙ(x) = cos(n arccos x), playing a pivotal role in approximation theory.
• They exhibit remarkable numerical stability and optimal extremal properties, which underpin their wide adoption in spectral methods and minimax approximations.
• Their rich structure, including three-term recurrences, hypergeometric forms, and orthogonality relations, connects them to diverse polynomial families and computational techniques.

Chebyshev polynomials of the first kind, denoted $T_n(x)$, are a prototypical family of classical orthogonal polynomials defined on the interval $[-1,1]$. They exhibit rich algebraic, analytic, and computational properties, with deep connections to approximation theory, numerical analysis, spectral methods, special function theory, and even combinatorics. For $x \in [-1,1]$ and integer $n \geq 0$, the standard trigonometric definition is $T_n(x) = \cos(n \arccos x)$. These polynomials are also uniquely characterized by their three-term recurrence, explicit power and hypergeometric expansions, optimal extremal properties, and exceptional numerical stability under recurrence-based evaluation schemes. Their applications also extend to advanced topics such as multivariate generating functions, SDP-based minimax constructions, endpoint-mass generalizations, and new connection formulae to other polynomial families.

1. Fundamental Definitions and Key Properties

The classical definition of $T_n(x)$ uses the cosine multiple-angle identity: for any $\theta \in \mathbb{R}$, if $x = \cos \theta$, then

$T_n(x) = \cos(n\theta)$

This trigonometric view immediately yields basic properties:

• Three-term recurrence:

$$T_0(x) = 1, \quad T_1(x) = x, \quad T_{n+1}(x) = 2x\,T_n(x) - T_{n-1}(x)$$

This recurrence uniquely determines the sequence from the initial terms (Smoktunowicz et al., 2013).

• Explicit algebraic (power sum) form:

$$T_n(x) = \sum_{k=0}^{\lfloor n/2 \rfloor} (-1)^k \, \frac{n}{n-k} \binom{n-k}{k} (x^2-1)^k x^{n-2k}$$

This expansion appears in several results and is instrumental in both theoretical and computational contexts (Kronenburg, 2020).

• Orthogonality:

$$\int_{-1}^1 T_m(x) T_n(x) (1-x^2)^{-1/2} dx = \begin{cases} 0, & m \neq n \ \pi, & m=n=0 \ \pi/2, & m=n \geq 1 \end{cases}$$

This reflects optimality for minimal $L^\infty$ norm and plays a central role in spectral approximation (Anshelevich, 2011, Smoktunowicz et al., 2013).

• Zeros and extrema: $T_n(x)$ has $n$ simple roots at $x_k^{(n)} = \cos\bigl( \frac{2k-1}{2n}\pi \bigr)$ $(k = 1,\dots,n)$, and extrema at $x'_k = \cos\bigl( \frac{k\pi}{n} \bigr)$ with values $(-1)^k$ (Abrarov et al., 2016).
• Generating function:

$$\sum_{n=0}^\infty T_n(x) t^n = \frac{1-x t}{1-2x t + t^2}, \quad |t| < 1$$

This power series encodes closed-form relationships and arises both in classical and modern characterizations (Anshelevich, 2011, Mesk et al., 2016).

2. Structural Identities and Representations

Chebyshev polynomials admit an array of alternative expansions and characterizations, reflecting their centrality in the Boas–Buck classification (Anshelevich, 2011):

• Hypergeometric form:

$T_n(x) = {}_2 F_1(-n, n; \tfrac12; \tfrac{1-x}{2})$

This representation arises from the expansion of $\cos(n \arccos x)$ and enables connection formulae and explicit combinatorial evaluations (Bedratyuk et al., 2019, Abd-Elhameed et al., 2015).

• Jacobi polynomial connection:

$T_n(x) = P_n^{(-1/2, -1/2)}(x)$

Where $P_n^{(\alpha, \beta)}(x)$ denotes the Jacobi polynomials (Bedratyuk et al., 2019).

• Product and addition formulae:

$$T_m(x) T_n(x) = \tfrac12 [T_{m+n}(x) + T_{|m-n|}(x)]$$

These identities generalize angle addition and are vital for spectral algorithms.

• Rodrigues-type formula:

$$T_n(x) = \frac{(-1)^n \sqrt{\pi}}{2^{n-1} \Gamma(n+\tfrac12)} (1-x^2)^{\frac12-n} \frac{d^n}{dx^n}(1-x^2)^{n-\frac12}$$

Allowing derivative-based constructions and explicit error bounds (Mesk et al., 2016).

• Multivariate generating functions: Closed forms for series like

$$\sum_{j \ge 0} \rho^j T_{j+m}(x) = \frac{T_m(x) - \rho T_{m-1}(x)}{1 - 2\rho x + \rho^2}$$

Generalize to rational expressions in multiple Chebyshev and associated polynomials via Szabłowski’s method (Szabłowski, 2017).

• Cayley element identity (from derivation theory):

$$T_n(x) + \sum_{k=1}^{n}(-2)^k \binom{n}{k} \sum_{i=0}^{\lfloor \frac{n-k-1}{2} \rfloor} \binom{n-i-1}{k-1} \binom{k+i-1}{k-1} T_{n-k-2i}(x) [T_1(x)]^k = (-1)^n$$

Illustrates hidden combinatorial structure via kernel polynomials (Bedratyuk et al., 2019).

3. Computational Methods and Numerical Stability

Accurate and stable computation of $T_n(x)$ is essential in scientific computing. Smoktunowicz et al. (Smoktunowicz et al., 2013) provide detailed analyses:

• Algorithm I: Forward three-term recurrence.
  • Complexity: $O(N)$
  • Backward stability: Yes (with error $O(\epsilon N^2)$ except $N$ odd with $|x|<S_N$).
  • **Recommended for general-purpose evaluation, stable up to $N\approx512$.
• Algorithm II: Fast doubling (for $N=2^p$).
  • Complexity: $O(\log N)$
  • Backward stability: Yes, error $O(\epsilon N^2)$.
• Algorithm III: Trigonometric evaluation $\cos(N\arccos x)$).
  • Complexity: $O(1)$
  • Backward stability: No (subject to large relative errors if $\arccos$ or $\cos$ not accurate near $\pm1$).
• Algorithm IV: Horner’s scheme on monomial expansion.
  • Complexity: $O(N)$
  • Stability: No (coefficient growth makes it highly unstable for $N\gtrsim16$).
• Clenshaw’s algorithm: Highly stable for general Chebyshev expansions (based on the three-term recurrence).

Algorithm	Work	Error bound	Backward stable
I	$O(N)$	$\epsilon O(N^2)$	Yes (with caveats)
II	$O(\log N)$	$\epsilon N^2$	Yes
III	$O(1)$	Can be $\gg \epsilon$	No (generally)
IV	$O(N)$	Grows as $2^N$	No

Numerical evidence confirms Algorithm I as optimal for most applications.

4. Extremal Properties, Generalizations, and SDP Methods

Chebyshev polynomials of the first kind are extremal for uniform approximation over $[-1,1]$, minimizing maximal deviation among degree-$n$ monic polynomials (Foucart et al., 2019): $$T^*_n(x) = \arg\min_{p \in P_n} \max_{x \in K} |p(x)|$$ For general compact $K = \bigcup_\ell [a_\ell, b_\ell] \subset [-1,1]$, modern construction uses semidefinite programming (SDP) via nonnegativity certificates in the Chebyshev basis.

• SDP constraints: $t \pm p(x) \ge 0$ on each interval, with nonnegativity proven via LMI constraints involving PSD matrices $Q, R$ expanded in Chebyshev basis.
• Endpoint-restricted Chebyshev polynomials: Roots restricted to $K$ via additional SDP blocks.

This approach enables design of polynomials optimal in varied domains, with explicit coefficients for classical and disconnected supports.

5. Classical and Generalized Orthogonality, Bernstein Basis, Endpoint Masses

Classically, $T_n(x)$ is orthogonal on $[-1,1]$ with respect to $w(x) = (1-x^2)^{-1/2}$. AlQudah (AlQudah, 2015) extends this to generalized Chebyshev-type polynomials $\mathscr{T}_n^{(M,N)}(x)$: $$\mathscr{T}_n^{(M,N)}(x) = c\,T_n(x) + M\,Q_n(x) + N\,R_n(x) + M N S_n(x)$$ with $M$, $N$ introducing mass points at $-1$ and $+1$. These admit closed-form expansions in Bernstein polynomials $B_k^n(x)$, supporting geometric design algorithms and yielding diagonal moment matrices for least-squares approximation.

• Bernstein expansion:

$$\mathscr{T}_n^{(M,N)}(x) = \sum_i c_{i,n} B_i^n(x) + \sum_k A_k \sum_j c_{j,k} B_j^k(x)$$

• Modified orthogonality:

$$\int_{-1}^{1} \mathscr{T}_n^{(M,N)}(x) \mathscr{T}_m^{(M,N)}(x) w_{M,N}(x)\,dx = 0, \text{ for } n \neq m$$

This generalization enables endpoint control and hybrid spectral-geometric methods.

6. Connections to Other Polynomial Families and Special Functions

Chebyshev polynomials of the first kind are uniquely singled out in generating function classifications (Mesk et al., 2016, Anshelevich, 2011):

• Only families with a generating function of the form $F(x z - \alpha z^2)$ and satisfying three-term recursion are (up to rescaling) the monomials, ultraspherical/Gegenbauer, Hermite, and Chebyshev-first-kind polynomials.
• Chebyshev corresponds to the logarithmic case:

$F(z) = 1 + \ln \frac{1}{1-z}$

yielding

$$\sum_{n=0}^\infty T_n(x)\,z^n = \frac{1 - x z}{1 - 2x z + z^2}$$

Hypergeometric connection formulae relate Chebyshev and Fibonacci polynomials: $$T_j(x) = j \sum_{m=0}^{\lfloor j/2 \rfloor} (-1)^m \binom{j-m}{j-2m} \frac{2^{j-2m-1}}{j-m} {}_2F_1(-m, j-m; j-2m+2; -4) F_{j-2m+1}(x)$$ and

$$F_{j+1}(x) = \sum_{m=0}^{\lfloor j/2 \rfloor} \alpha_{m} T_{j-2m}(x)$$

demonstrating deep combinatorial interplay (Abd-Elhameed et al., 2015).

7. Advanced Identities, Multivariate Expansions, and Modern Applications

Szabłowski (Szabłowski, 2017) advances the use of multivariate generating functions encoding products of Chebyshev polynomials, leading to closed-form rational functions for sums over products of $T_n$ and $U_n$. Kibble–Slepian-type expansions further generalize these to covariance-matrix-indexed expressions, relevant to random matrix theory and free probability. These closed forms facilitate exact integration of rational functions weighted by Chebyshev kernels.

Nontrivial orthogonality-type identities—such as new four-term relations among Chebyshev polynomials of mixed orders—furnish combinatorial structure useful in advanced polynomial algebra. Identities involving higher derivatives of $\arctan x$ connect Chebyshev polynomials to differential calculus via closed-form representations (Kronenburg, 2020).

Chebyshev polynomials appear in reformulations of Viète’s formula for $\pi$, with nested radical expressions and product-to-sum transformations (Abrarov et al., 2016), bridging numerical analysis with classical mathematical constants.

---

Chebyshev polynomials of the first kind thus form a cornerstone of classical and contemporary analysis, combining explicit structure, optimality, deep connections, and robust numerical methods. Their role as minimal/maximal polynomials, their algebraic and analytic identities, and their extensions through generating function theory and numerical optimization position them as foundational elements for both theory and practice in computational mathematics.