Mathematical Foundations II: Variational Calculus

CHAPTER 2Math Foundations

Mathematical Foundations II: Variational Calculus

Physical Intuition and Historical Context

To understand the Finite Element Method (FEM) at a graduate level, one must move beyond the elementary engineering approach of assembling spring equations and enter the domain of Variational Calculus. While classical calculus focuses on finding the specific points (numbers) that make a function stationary (minimum, maximum, or saddle points), variational calculus is concerned with finding the entire functions (curves, surfaces, or fields) that render a functional stationary.

Historically, this branch of mathematics arose from questions of optimization in physics. In 1696, Johann Bernoulli posed the famous Brachistochrone problem: what is the shape of a frictionless curve joining two points at different heights such that a bead slides down it under the influence of gravity in the shortest possible time? The answer is not a straight line, but a cycloid. Solving this problem required a new mathematical framework where the independent variable is not a coordinate $x$, but a path $y(x)$, and the dependent quantity is the total travel time, which is an integral of $y(x)$ and its derivative.

This concept was generalized by Leonhard Euler and Joseph-Louis Lagrange into a formal system. In classical mechanics, this culminated in Hamilton's Principle of Least Action, which states that the actual path taken by a dynamic system is the one that minimizes (or makes stationary) the integral of the Lagrangian (the difference between kinetic and potential energy) over time. In solid mechanics, this manifests as the Principle of Minimum Potential Energy: of all possible displacement fields that a structure could experience, the actual displacement field that satisfies equilibrium is the one that minimizes the total potential energy functional.

For a mechanical engineer, variational calculus is the bridge that connects the physical laws of nature (expressed as differential equations) to the computer-solvable algebraic systems of FEM. When we look at a deformed beam, a heated engine block, or a fluid flowing through a nozzle, nature is continuously solving a variational minimization problem. By mimicking this process mathematically, FEM discretizes the continuous variational problem into a finite set of algebraic equations.

Definition and Structure of Functionals

A functional is a mathematical operator that maps a function (or a set of functions) to a single real scalar value. Symbolically, if $v(x)$ is a function belonging to a class of admissible functions $V$, then a functional $I$ is denoted as:

$$I: V \to \mathbb{R} \quad \text{or} \quad I[v] = \int_{a}^{b} F(x, v(x), v'(x)) \, dx$$

Here, the integrand $F(x, v, v')$ is a function of three variables: the independent coordinate $x$, the dependent function $v(x)$, and its first derivative $v'(x) = \frac{dv}{dx}$. The domain $V$ of the functional is not a subset of real numbers, but a function space. To be mathematically rigorous, the function space $V$ must be defined with care. For example, it could be the space $C^2[a,b]$ of functions that are twice continuously differentiable on the interval $[a,b]$, or a Hilbert space $H^1(a,b)$ of functions whose first derivatives are square-integrable.

Let us look at three classic examples of functionals to solidify this definition:

The Arc Length Functional: The length $S$ of a curve $y(x)$ connecting $(x_0, y_0)$ to $(x_1, y_1)$ is given by the integral:

$$S[y] = \int_{x_0}^{x_1} \sqrt{1 + \left(\frac{dy}{dx}\right)^2} \, dx$$

Here, the functional $S$ takes any curve $y(x)$ as input and outputs its physical length. The goal of variational calculus is to find the function $y(x)$ that minimizes $S[y]$.

The Brachistochrone Functional: The total time $T$ taken by a particle to slide down a curve $y(x)$ from $(0,0)$ to a point $(x_1, y_1)$ under gravity is:

$$T[y] = \int_{0}^{x_1} \frac{\sqrt{1 + (y')^2}}{\sqrt{2gy}} \, dx$$

This functional maps the shape of the slide $y(x)$ to the scalar time $T$.

Total Potential Energy of a 1D Elastic Bar: In structural mechanics, the total potential energy $\Pi$ of a bar under axial load consists of the internal strain energy $U$ minus the work done by external forces $W$:

$$\Pi[u] = U[u] - W[u] = \int_{0}^{L} \frac{1}{2} A E \left( \frac{du}{dx} \right)^2 \, dx - \int_{0}^{L} q(x) u(x) \, dx - P u(L)$$

Here, $u(x)$ is the axial displacement field, $A$ is the cross-sectional area, $E$ is Young's modulus, $q(x)$ is a distributed axial force, and $P$ is a point load at the free end $x=L$. The functional $\Pi[u]$ maps the displacement function $u(x)$ to a scalar energy value. The equilibrium state is the function $u(x)$ that minimizes $\Pi[u]$.

The Concept of Variations

In classical calculus, to find the minimum of a function $f(x)$, we introduce an infinitesimal change in the coordinate $dx$ and set the differential $df = f'(x)dx = 0$. In variational calculus, we introduce an infinitesimal change in the *shape* of the function itself, which we call the variation (or virtual change) and denote by $\delta$.

Let $y(x)$ be the actual function that minimizes the functional $I[y]$. We define an family of perturbed curves:

$$Y(x) = y(x) + \epsilon \eta(x)$$

where $\epsilon$ is an extremely small scalar parameter, and $\eta(x)$ is an arbitrary, well-behaved function representing the shape of the perturbation. The variation of the function $y$, denoted by $\delta y$, is defined as the difference between the perturbed path and the actual path at a fixed coordinate $x$ for an infinitesimal $\epsilon$:

$$\delta y = Y(x) - y(x) = \epsilon \eta(x)$$

Similarly, the variation of the derivative $y'$ is:

$$\delta (y') = Y'(x) - y'(x) = \epsilon \eta'(x) = \frac{d}{dx}(\delta y)$$

This indicates that the variational operator $\delta$ commutes with the differential operator $\frac{d}{dx}$:

$$\delta \left( \frac{dy}{dx} \right) = \frac{d}{dx}(\delta y)$$

For a variation to be admissible, the perturbed function $Y(x)$ must satisfy the exact same geometric constraints as the original function $y(x)$. For instance, if the curve is fixed at its endpoints, meaning $y(x_0) = y_0$ and $y(x_1) = y_1$, then the perturbed function must also satisfy $Y(x_0) = y_0$ and $Y(x_1) = y_1$. Substituting these constraints into our perturbation formula yields:

$$Y(x_0) = y(x_0) + \epsilon \eta(x_0) \implies y_0 = y_0 + \epsilon \eta(x_0) \implies \eta(x_0) = 0$$

$$Y(x_1) = y(x_1) + \epsilon \eta(x_1) \implies y_1 = y_1 + \epsilon \eta(x_1) \implies \eta(x_1) = 0$$

Thus, the variation $\delta y(x)$ must vanish at any boundary where the value of the function is pre-determined:

$$\delta y(x_0) = 0, \quad \delta y(x_1) = 0$$

This is a critical concept in finite elements: virtual displacements (or test functions) must be zero at all boundaries where displacements are prescribed.

Boundary Conditions: Dirichlet vs. Neumann

Boundary conditions (BCs) are the mathematical representations of physical constraints on the boundaries of our domain. In variational calculus, boundary conditions are classified into two distinct types based on how they interact with the function space:

Essential (Dirichlet) Boundary Conditions: These conditions prescribe the value of the field variable itself on the boundary.

•

*Physical Examples*: A fixed support of a beam where displacement $u = 0$, a wall maintained at a specific temperature $T = T_0$, or a fluid boundary with a fixed velocity $v = v_0$.

•

*Mathematical Significance*: These conditions directly restrict the space of admissible functions. Any candidate function $u(x)$ must satisfy these conditions to even be considered a potential solution. In the variational process, because the field variable is fixed at these boundaries, the variation of the variable must vanish there (e.g., $\delta u = 0$).

Natural (Neumann) Boundary Conditions: These conditions prescribe the derivative or flux of the field variable at the boundary.

•

*Physical Examples*: A tensile force applied at the end of a bar ($A E \frac{du}{dx} = P$), an insulated boundary in heat transfer where heat flux is zero ($-k \frac{dT}{dx} = 0$), or a shear force applied to the end of a beam.

•

*Mathematical Significance*: These conditions do not restrict the space of admissible functions. Candidates for the solution do not need to satisfy them a priori. Instead, these boundary conditions emerge "naturally" out of the variational minimization process when setting the first variation of the functional to zero. They represent the natural balance of forces or fluxes at the boundary.

Mixed (Robin) Boundary Conditions: These prescribe a linear relationship between the boundary value of the field variable and its normal derivative. A classic example is convective heat transfer at a boundary, where the heat flux leaving the surface is proportional to the difference between the surface temperature and the ambient fluid temperature:

$$-k rac{dT}{dx} = h (T - T_{infty})$$

In mechanical systems, this corresponds to an elastic spring support at the boundary, where the force in the spring is proportional to the boundary displacement.

Strong vs. Weak Forms

The governing physics of a continuous medium can be written in two mathematically equivalent but structurally different formats: the Strong Form and the Weak Form. Understanding the transition between these two forms is the core mathematical foundation of the Finite Element Method.

The Strong Form

The strong form consists of a differential equation and a set of boundary conditions that must be satisfied locally at every single point within the domain $\Omega$ and on its boundary $\Gamma$.

•

*Example (1D Bar)*: $-AE \frac{d^2u}{dx^2} = q(x)$ for all $x \in (0, L)$.

•

*Continuity Requirement*: To solve this equation directly, the displacement function $u(x)$ must be twice continuously differentiable ($u \in C^2$). If $u(x)$ has a discontinuous slope (which happens at points where the cross-sectional area changes or where a point force is applied), the second derivative $\frac{d^2u}{dx^2}$ becomes undefined in the classical sense, and the strong form fails.

The Weak Form

The weak form is an integral statement that enforces the governing physics globally on average over the entire domain. It is derived by multiplying the strong form differential equation by an arbitrary weight function $v(x)$ (also called a test function or virtual displacement) and integrating the product over the domain.

•

*Example (1D Bar)*: $\int_0^L AE \frac{du}{dx} \frac{dv}{dx} \, dx = \int_0^L q v \, dx + P v(L)$ for all admissible $v(x)$.

•

*Continuity Requirement*: In the weak form, the derivatives are split between the solution $u(x)$ and the weight function $v(x)$ through integration by parts. As a result, $u(x)$ only needs to have a square-integrable first derivative ($u \in H^1$). This "weakens" the continuity requirements on the approximation, allowing us to use simple piecewise linear functions that have discontinuous derivatives at element boundaries.

Metric	Strong Form	Weak Form
:---	:---	:---
Mathematical Type	Differential Equation	Integral Equation
Enforcement	Pointwise (Local)	Domain-averaged (Global)
Continuity Requirement	High (e.g., $u \in C^2$ for second-order PDEs)	Low (e.g., $u \in H^1$ for second-order PDEs)
Handling of Natural BCs	Must be explicitly stated alongside PDE	Automatically included in the boundary terms
Suitability for FEM	Poor (requires complex smooth approximations)	Perfect (piecewise continuous polynomials suffice)

Physical Meaning of Energy States

In computational mechanics, variational formulations are deeply rooted in energy principles. The functional we minimize is typically the Total Potential Energy of the system:

$$\Pi = U - W$$

where:

•

Internal Strain Energy ($U$): The energy stored inside the elastic body due to deformation. For a 1D linear elastic material, the strain energy density is $\frac{1}{2} \sigma \epsilon = \frac{1}{2} E \left(\frac{du}{dx}\right)^2$. Integrating this over the volume gives $U = \int_V \frac{1}{2} E \left(\frac{du}{dx}\right)^2 \, dV$.

•

Work Potential ($W$): The potential energy of the external loads. It is the work done by body forces, surface tractions, and point loads as the structure deforms.

The principle of minimum potential energy is a statement of physical stability. The first variation of potential energy, $\delta \Pi$, represents the change in energy due to a virtual displacement $\delta u$. For the system to be in equilibrium, the energy state must be stationary, which means:

$$\delta \Pi = 0$$

This is equivalent to the Principle of Virtual Work: the internal virtual work done by stresses under a virtual strain must equal the external virtual work done by loads under virtual displacements:

$$\delta W_{int} = \delta W_{ext}$$

To determine if this equilibrium state is stable, unstable, or neutral, we must examine the second variation of the potential energy, $\delta^2 \Pi$, which is analogous to the second derivative in classical calculus:

Stable Equilibrium (Local Minimum): If $\delta^2 \Pi > 0$ for all admissible variations, the energy state is a local minimum. Any displacement away from equilibrium requires an input of external work, and the structure will naturally spring back to its equilibrium state.

Unstable Equilibrium (Local Maximum): If $\delta^2 \Pi < 0$, the energy state is a local maximum. A tiny perturbation will cause the system to release energy and move further away from the equilibrium state. This is what occurs physically during structural buckling or material failure.

Neutral / Critical Equilibrium: If $\delta^2 \Pi = 0$, the system is at a bifurcation point. Multiple equilibrium paths may exist, which is the starting point for non-linear stability analysis.

Extended chapter reference

Energy functionals, admissible fields, stationarity, and weak statements

Variational calculus turns a differential boundary-value problem into an energy statement. This exposes the natural boundary conditions and provides the mathematical bridge from continuum mechanics to element matrices.

Construct a functional from strain energy and external work.

Distinguish admissibility, compatibility, equilibrium, and stationarity.

Apply integration by parts to reduce continuity requirements.

Interpret a finite element solution as a stationary point in an admissible trial space.

Visual map

From formulation to verified result

Each stage produces evidence for the next. A reliable result preserves traceability from the physical idealization to the final engineering check.

Equation sheet

Governing relationships

Total potential

$$\Pi(\mathbf{u})=U(\mathbf{u})-W(\mathbf{u})$$

Balances stored energy and external work.

Stationarity

$$\delta\Pi=0$$

Defines equilibrium for admissible virtual displacements.

Ritz approximation

$$u_h=\sum_{i=1}^{n}N_i q_i$$

Restricts the field to a finite trial space.

Engineering comparison

Selection and interpretation table

Statement	Derivative order	Boundary treatment	FEM role
Strong form	Highest	All conditions explicit	Defines governing PDE
Weak form	Reduced by integration	Natural terms emerge	Primary FEM starting point
Minimum potential	Energy variation	Essential conditions on trial field	Linear elastic shortcut
Hamilton principle	Space and time variation	Includes end-time conditions	Transient dynamics

Verification and application

Checks before accepting the result

Trial field satisfies essential BCs
Units of every energy term match
Boundary work sign is consistent
Second variation indicates stability

Industry application

For a tapered cantilever, an admissible polynomial can predict tip deflection without solving the variable-coefficient beam equation directly. Increasing the polynomial order should lower the potential energy and approach the continuum solution from the stiff side.