Differential Transformation Method (DTM)

• Differential Transformation Method is an iterative, semi-analytical technique that constructs local Taylor series to solve linear and nonlinear differential equations.
• It employs algebraic recurrence relations and Bell polynomials to systematically transform variable-coefficient and composite nonlinear terms without explicit higher-order differentiation.
• The method offers rapid convergence and automation potential for diverse differential and differential-algebraic systems, though its accuracy is inherently local.

The Differential Transformation Method (DTM) is an iterative, semi-analytical technique for constructing power-series solutions—most often local Taylor expansions—to linear and nonlinear ordinary and partial differential equations, as well as coupled systems and differential-algebraic equations (DAEs). DTM works by systematically generating Taylor coefficients through simple algebraic recurrences, sidestepping explicit calculation of higher-order derivatives or symbolic integration. The essential innovation of DTM is the formalization of a set of operational rules that map nonlinear, composite, and variable-coefficient terms into algebraic relations for the sequence of series coefficients, making it feasible to automate or implement in symbolic computation environments.

1. Formal Definition and Core Principles

Let $y(t)$ be analytic at $t_0$. The differential transform of order $k$ is defined by

$$Y(k)[t_0] = \frac{1}{k!} \left. \frac{d^k y(t)}{dt^k} \right|_{t=t_0},$$

and the inverse transform reconstructs $y(t)$ through the (truncated) Taylor expansion: $y_N(t) = \sum_{k=0}^N Y(k)[t_0](t - t_0)^k.$ All operations within DTM proceed via this sequence-based representation, and approximate solutions are formed by truncating at finite $N$. DTM extends directly to systems, higher dimensions, and problems with initial or boundary data by applying analogous definitions to each variable and differentiating with respect to multiple independent variables as needed (Srivastava et al., 2023, Narendranath, 2016).

2. Operational Rules and General Recurrence Relations

The algebraic structure is encapsulated by specific transformation rules:

• Linearity: $f(t) + g(t) \leftrightarrow F(k) + G(k)$.
• Derivative: $$\frac{d^n f}{dt^n} \leftrightarrow (k+1)\cdots(k+n) F(k+n)$$.
• Product: $f(t)g(t) \leftrightarrow \sum_{i=0}^k F(i)G(k - i)$.

For nonlinear compositions such as $h(t) = f(g(t))$, the Faà di Bruno–style recurrence is employed. If $g(t) \leftrightarrow G(k)$ and $f(\tau) \leftrightarrow F(k)$ (with $\tau = g(t_0)$), then the transform of $h$ is given by

$$H(k) = \begin{cases} F(0) & k = 0, \ \sum_{l=1}^k F(l)\, \hat B_{k, l}(G(1),\ldots,G(k-l+1)) & k \geq 1, \end{cases}$$

where $\hat B_{k, l}$ denote the partial ordinary Bell polynomials. The Bell polynomial structure enables handling arbitrary nonlinearities without computing symbolic derivatives or repeated compositions (Srivastava et al., 2023).

When transforming a differential-algebraic or higher-order system of the form

$$w^{(m)}(t) = f(w(t), u(t), t), \quad g(w(t), u(t), t) = 0,$$

the transformed relations are

$$(k+1)(k+2)\cdots(k+m) W(k+m) = F(W(k), U(k), t_0), \quad 0 = G(W(k), U(k), t_0),$$

together with initial data mapped directly as $W(i) = \eta_i/i!$ for $i=0,\ldots, m-1$.

3. Handling Nonlinearities and Composite Functions

A distinguishing feature of DTM, particularly in the non-autonomous and nonlinear-DAE context, is the use of discrete Faà di Bruno formulas and partial Bell polynomials for the transform of composite functions. For a nonlinearity $h(t) = f(g(t))$, with transforms as above, one avoids symbolic differentiation by constructing the transform coefficients via

$\hat B_{k,l}(\hat x_1, \ldots, \hat x_{k-l+1})$

defined through

$$\sum_{k\geq l} \hat B_{k,l}(\hat x_1, \ldots, \hat x_{k-l+1}) t^k = \left( \sum_{m\geq 1} \hat x_m t^m \right)^l,$$

and the recurrence

$$\hat B_{k,l} = \sum_{i=1}^{k-l+1} \frac{i l}{k} \hat x_i \hat B_{k-i, l-1}, \quad \hat B_{0,0} = 1, \quad \hat B_{k,0} = 0 \ (k \geq 1).$$

This algebraic approach generalizes to arbitrarily deep function compositions, exponentials, powers, and similar constructs, with per-case specialization given for $f(x) = e^x$, powers, and logarithms (Srivastava et al., 2023).

4. Implementation Workflow and Algorithmic Procedure

The stepwise DTM workflow for solving a (possibly DAE) initial value problem is as follows:

1. Initialization: Set the transforms of all variables from initial/boundary data. For $m$-th order equations, set $W(k) = w^{(k)}(0)/k!$ for $k = 0, \ldots, m-1$.
2. Nonlinear/composite terms: For each nonlinear operation (composition, product, etc.), use the discrete Faà di Bruno–Bell polynomial expansions (or other operational rules) to compute the transform of that term at order $k$.
3. Recurrence update: At each $k$, solve the system

$$(k+1)\cdots(k+m) W(k+m) = F_\text{transformed}(W(k), \ldots), \qquad 0 = G_\text{transformed}(W(k), \ldots)$$

for the next set of unknowns (often $W(k+m)$ and similarly for other variables).

1. Series assembly: After reaching the prescribed truncation order $N$, construct the solution approximation

$w_N(t) = \sum_{k=0}^N W(k) (t - t_0)^k.$

2. Error metrics (optional): Evaluate $E_N(v) = |w(v) - w_N(v)|$ or supremal error over the interval as needed (Srivastava et al., 2023).

5. Error Estimates, Convergence, and Analytic Properties

Truncation after $N$ terms yields a residual (Taylor remainder)

$$R_N(t) = y(t) - \sum_{k=0}^N Y(k)[t_0](t - t_0)^k = \frac{y^{(N+1)}(\xi)}{(N+1)!} (t-t_0)^{N+1}, \quad \xi \in [t_0, t].$$

This bounds the error as

$$|R_N(t)| \le \frac{\max_{\xi \in I} |y^{(N+1)}(\xi)|}{(N+1)!} |t-t_0|^{N+1},$$

demonstrating rapid, exponential convergence as $N \to \infty$ for analytic solutions. Tables in (Srivastava et al., 2023) show $E_{N, \infty} \to 0$ with increasing $N$.

6. Advantages, Limitations, and Contextual Role

Advantages

• No symbolic derivatives: The method, via Bell polynomials, avoids explicit computation of high-order derivatives for nonlinearities.
• Automatable algebraic recursion: Each coefficient is constructed through fixed operational rules and recurrences, making the approach well-suited for computational implementation.
• Generality: The formalism is compatible with DAEs, stiff problems, and nonlinearity beyond polynomial type, including arbitrary compositions.
• Accuracy: Rapid convergence in the analytic domain, with error controlled by the proximity to the nearest singularity.

Limitations

• Radius of convergence: Limited by the analyticity of the solution; series may diverge or converge slowly near singularities.
• Computational resources: For high order or strong nonlinearity, the dimensionality and complexity of recurrences grow rapidly, though the Bell polynomial formalism alleviates some bottlenecks.
• Local nature: The truncated series is accurate locally; for global domains, a multistep or piecewise analytic continuation approach may be required.
• Dependence on initial data and function regularity: For DAEs, the well-posedness and smoothness of solutions are prerequisite for successful application (Srivastava et al., 2023).

7. Illustrative Example: Nonlinear DAE with Bell-Polynomial-Based DTM

Consider the index-1 semiexplicit DAE system: $$\begin{aligned} &w_1' = 2 w_3, \ &w_2' = 2 w_4, \ &w_3' = -2w_3 + e^{w_2 + w + \phi_1(v)}, \ &w_4' = 2w_4 + e^{w_1 + w + \phi_2(v)}, \ &\text{subject to the constraint} \quad w_1 + w_2 = \phi_3(v), \end{aligned}$$ with prescribed initial data (Srivastava et al., 2023). The transform for composite exponential terms, such as $h_1(v) = e^{w_2(v) + w(v) + \phi_1(v)}$, is

$$H_1(k) = \sum_{l=1}^k \frac{1}{l!} \hat B_{k, l}(W_2(1), \ldots, W_2(k-l+1)), \quad H_1(0) = 1,$$

with analogous expressions for other terms. The full DTM system yields, after iteration, power-series approximations matching the analytic solution to machine precision with moderate truncation order ($N\sim 20$), demonstrating both the efficiency and accuracy of the approach.

---

In summary, the Differential Transformation Method, with Bell polynomial and Faà di Bruno-based nonlinear algebra, provides a systematic, symbolic-free, and power-series-based computational framework for solving a broad class of nonlinear ordinary and partial differential equations, as well as DAEs, without reliance on symbolic differentiation or linearization. Its convergence, error properties, and algebraic foundations make it a foundational component in modern semi-analytical solution toolkits for applied mathematics and engineering (Srivastava et al., 2023).