Multivariate Taylor Series Expansions

• Multivariate Taylor series expansions express functions of several variables using partial derivatives and explicit remainder terms to capture local behavior.
• Advanced methods replace traditional unknown remainders with explicit polynomial and logarithmic expressions, enabling exact expansions under compact support constraints.
• Measure-based generalizations extend these expansions to non-analytic or discontinuous functions, unifying classical Taylor theory with modern functional analytic techniques.

A multivariate Taylor series expansion expresses a real-valued function of several variables in terms of its value, partial derivatives, and higher-order derivatives at a reference point, capturing local behavior through polynomial or closely related terms involving the increments from that point. Contemporary developments extend this concept to explicit remainder terms and to measure-based expansions, enabling exact representations for a wider class of functions, including non-differentiable and merely measurable cases (Alghalith, 2015, Micheas, 14 Aug 2025).

1. Classical Multivariate Taylor Expansion and Remainder

Let $f : \mathbb{R}^n \to \mathbb{R}$ be $(n+1)$-times continuously differentiable on an open set $U$, and fix a reference point $a \in U$. Express $x = (x_1,\dots,x_n)$ and use the multi-index notation $\alpha = (\alpha_1,\dots,\alpha_n)$, $|\alpha| = \sum_i \alpha_i$, $(x-a)^\alpha = \prod_i (x_i - a_i)^{\alpha_i}$, and $$\partial^\alpha f(a) = \frac{\partial^{|\alpha|} f}{\partial x_1^{\alpha_1} \cdots \partial x_n^{\alpha_n}}(a)$$. The classical expansion to order $n$ reads

$$f(x) = \sum_{|\alpha| \leq n} \frac{\partial^\alpha f(a)}{\alpha!} (x-a)^\alpha + \sum_{|\alpha| = n+1} \left[ (n+1) \int_0^1 (1-t)^n\, \partial^\alpha f(a + t(x-a))\,dt \right] \frac{(x-a)^\alpha}{\alpha!}$$

where the remainder involves derivatives evaluated along the segment from $a$ to $x$ (Micheas, 14 Aug 2025). The remainder term is generally unknown a priori except for its form and integral bounds.

2. Exact Multivariate Expansions without Unknown Remainder

A new class of exact Taylor expansions replaces the traditional remainder with fully explicit terms. For $f \in C^1$ on a compact set and reference $c \in \mathbb{R}^n$, the order-one expansion is

$$\boxed{ f(x) = A_0 + \sum_{i=1}^n A_i (x_i - c_i) + A_1 \sum_{i=1}^n (x_i - c_i + a_{n+1}) \ln(x_i - c_i + a_{n+1}) }$$

with fixed coefficients $A_0 = f(c)$, $A_i = \frac{\partial f}{\partial x_i}(c)$, $a_{n+1} = \max_i \{|x_i - c_i|\} + 0^+$ (Alghalith, 2015). No unknown, implicitly defined remainder appears. The last (logarithmic) term results from explicitly integrating the difference of sequential Taylor remainders.

The approach generalizes to higher order by recursively evaluating differences of Taylor remainders of consecutive order, yielding a sum of explicit polynomial and logarithmic terms at each step, so that after $N$ steps the expansion is exact and contains only explicit expressions (Alghalith, 2015). The requirement is that $f$ be $C^1$ with compact support to justify mean-value and multiple integration steps.

3. Measure-Based Generalization: Taylor Measures

The Taylor-measure framework further extends the multivariate expansion to measurable, non-analytic, or non-differentiable functions (Micheas, 14 Aug 2025). A Taylor measure at $a \in \mathbb{R}^d$ is any finite (signed) measure $\mu_a$ defined on $(\mathbb{N}^d, \mathcal{B}(\mathbb{N}^d))$. For a measurable coefficient map $\alpha \mapsto A_\alpha(x)$ and generator $g:\mathbb{R}^d \to \mathbb{R}^d$,

$$T_{g(x),A(x)}(B) = \sum_{\alpha\in B} A_\alpha(x) \frac{g(x)^\alpha}{\alpha!}$$

so that any function $f(x)$ can be expressed as $$T_{g(x),A(x)}(\mathbb{N}^d) = \sum_\alpha A_\alpha(x) \frac{g(x)^\alpha}{\alpha!}$$. The classical Taylor expansion is recovered by taking atomic measures and partial derivatives for $A_\alpha$ (Micheas, 14 Aug 2025).

This measure-based expansion allows exact representation even when $f$ is not differentiable or analytic, by suitable measure and generator choices.

4. Functional Analytic Structure: The Multivariate Taylor Measure Function Space

The set of all real-valued functions on $\mathbb{R}^d$ admitting a measure-based expansion forms the multivariate Taylor measure function space ("MTMF space", Editor's term) (Micheas, 14 Aug 2025): $f(x) = T_{g(x),A(x)}(B)$ with measurable $A_\alpha(x)$, Borel $B \subset \mathbb{N}^d$. The space admits a natural inner product: $$\langle f_1, f_2 \rangle = \sum_{\alpha \in B} \int_{\mathbb{R}^d} A_\alpha^{(1)}(x) A_\alpha^{(2)}(x) \frac{(g_1(x)g_2(x))^\alpha}{\alpha!}\,dx$$ making it a separable Hilbert space (under mild integrability), which is also a Banach space with induced norm. Classes $C^n(U)$, analytic, simple, $L^p$-, Lipschitz, and uniformly continuous functions all embed into this MTMF space by suitable choices of the index set, coefficient maps, and generator.

A plausible implication is that this space enables Hilbert space analysis—orthonormal bases, reproducing kernels, etc.—for function classes unreachable by classical Taylor theory.

5. Explicit Examples and Applications

Reaction–convection–diffusion PDE: For

$$V_t + r x V_x + \frac{1}{2} \sigma^2 x^2 V_{xx} - r V = 0,$$

the explicit expansion

$$V(x, t) = a_1 + a_2 x + a_3 t + a_2 (x + a_3) \ln(x + a_3)$$

converts the differential equation into an algebraic system in parameters $(a_1, a_2, a_3)$. Solving this system yields closed-form solutions (Alghalith, 2015).

Nonlinear PDE in two variables:

$V_t - k(V_{xx} + V_{yy}) + \alpha x V_x V_y = 0$

has the ansatz

$$V(x, y, t) = a_0 + a_1 x + a_2 y + a_3 t + a_1(x + a_4) \ln(x + a_4) + a_2(y + a_4) \ln(y + a_4),$$

reducing solution to an algebraic problem (Alghalith, 2015).

Portfolio model (HJB PDE): For

$$V_t + r x V_x + \sup_\pi [ (\mu - r)\pi V_x + \tfrac{1}{2} \sigma^2 \pi^2 V_{xx} ] = 0$$

the same explicit expansion yields closed-form expressions for the value function $V(x, t)$ and optimal allocation $\pi^*(x, t)$, with the optimal dollar allocation a logarithmic-affine function of wealth (Alghalith, 2015).

Measure-based exact expansions beyond classical differentiability:

Function	Measure–Based Expansion Features	Classical Taylor?
Simple function $f(x)=\sum c_k \mathbf{1}_{C_k}(x)$	Discontinuous, non-analytic, finite support	Not representable
Indicator $f(x)=\mathbf{1}_{\{\|x\|\leq 1\}}(x)$	Expansion with $B=\{0,1\}$, $g(x)=1$	Not representable
$f(x) = |x|$	Expansion via measure splitting	Not $C^1$ at $0$

These expansions illustrate the ability of the Taylor-measure framework to exactly represent functions outside the traditional domain of convergence of Taylor series (Micheas, 14 Aug 2025).

6. Recovery of Classical Results and Unifying Role

When $f \in C^{n+1}$, the Taylor measure $\mu_x$ collapses to atomic masses equal to partial derivatives, and the measure-based expansion coincides with the classical Taylor formula including its integral remainder. If one takes counting measure on the multi-index set with $A_\alpha(x) = D^\alpha f(a)$, the series recovers the textbook multivariate Taylor expansion. For analytic functions, the expansion holds at all orders (Micheas, 14 Aug 2025).

This suggests the Taylor-measure framework subsumes classical analytic, smooth, and elementary cases, while enabling a unified treatment across mathematics. The “measure perspective” replaces single pointwise derivatives with finite signed measures on multi-indices, allowing broader applicability with precise remainder control (Micheas, 14 Aug 2025).

7. Significance and Ongoing Directions

Recent advances demonstrate both explicit exact expansions for classical $C^1$ functions and a generalized theory accommodating nonsmooth, non-analytic, or discontinuous functions—all using unified multi-index or measure notation (Alghalith, 2015, Micheas, 14 Aug 2025). The resulting MTMF space, with its Hilbert and Banach structure, provides powerful instruments for functional-analytic and PDE applications, numerical schemes, and further generalizations of approximation theory.

A plausible implication is that these expansions enable new explicit solution strategies for multi-dimensional PDEs and optimal control, as well as systematic embedding of irregular or merely measurable functions into analytic frameworks. The elimination of unknown remainder terms or the explicit identification of measure-theoretic coefficients marks a significant structural generalization of Taylor expansion theory.