Physical Intuition and Analogies

In Finite Element Analysis (FEA), we frequently encounter the need to evaluate integrals over element domains. The most prominent example is the computation of the element stiffness matrix $\mathbf{k}^e$, which is defined as:

where $\mathbf{B}$ is the strain-displacement matrix, $\mathbf{D}$ is the material constitutive matrix, and $V^e$ is the volume of the element. For any element beyond the simplest 1D bar or constant-strain triangle, finding a closed-form analytical solution to this integral is practically impossible.

To build physical intuition, imagine you are attempting to calculate the average elevation of a highly undulating, rugged landscape. One approach is to measure the height at equally spaced grid points and compute their average (similar to Newton-Cotes integration methods like the Trapezoidal or Simpson's rules). However, if your sampling grid happens to align with the peaks and misses the valleys, your average will be highly inaccurate.

Alternatively, suppose you could choose exactly *where* to place your measuring poles and *how much weight* to assign to each measurement. By strategically placing a small number of poles—some in the deep valleys, some on the high peaks, and others on the slopes—you could calculate the exact average volume of the terrain with remarkably few measurements.

This is the essence of Gauss Quadrature. It is a mathematical framework that treats both the sampling locations (integration points or roots) and the importance of each sample (weights) as variables to be optimized. By placing these integration points at the roots of orthogonal polynomials, we maximize the precision of our integration scheme, allowing us to integrate complex polynomials exactly with a minimal number of function evaluations.

Core Concepts

1. Why Numerical Integration is Preferred in FEA

In practical FEA, numerical integration is not merely a computational convenience; it is a fundamental mathematical necessity due to the following reasons:

2. Newton-Cotes vs. Gauss Quadrature

To understand why Gauss Quadrature is the gold standard in finite elements, we compare it to the more familiar Newton-Cotes formulas:

By allowing the coordinate locations of the integration points to be variables, Gauss Quadrature doubles the highest degree of polynomial that can be integrated exactly. For example, a 2-point Gauss Quadrature rule can integrate a cubic polynomial ($a_0 + a_1 x + a_2 x^2 + a_3 x^3$) exactly, whereas a 2-point Newton-Cotes rule (the Trapezoidal rule) can only integrate a linear polynomial ($a_0 + a_1 x$) exactly.

3. Legendre Polynomials and Gauss Points

The optimal integration points for Gauss Quadrature on the standard interval $[-1, 1]$ are the roots of the Legendre polynomials $P_n(x)$. Legendre polynomials are a sequence of orthogonal polynomials defined on $[-1, 1]$ that satisfy the orthogonality condition:

The first few Legendre polynomials, derived using Rodrigues' formula or recurrence relations, are:

For an $n$-point Gauss Quadrature rule, the sampling points $x_i$ are the roots of $P_n(x) = 0$. These roots:

The corresponding weights $w_i$ represent the "influence area" of each point and are calculated using:

where $P'_n(x_i)$ is the derivative of the $n$-th Legendre polynomial evaluated at the root $x_i$. Because the roots are symmetric, the weights are also symmetric about the center of the interval.

4. Coordinate Mapping to the Parent Domain

Gauss-Legendre integration is strictly defined over the symmetric parent interval $[-1, 1]$. However, physical finite elements exist in arbitrary global coordinate spaces (e.g., a 1D bar stretching from $x = a$ to $x = b$, or a 2D quadrilateral with arbitrary corner coordinates).

To apply Gauss Quadrature, we must map the physical domain to the natural (parent) coordinate system. This requires a coordinate transformation:

This mapping scales and translates the integration domain, and the differential elements of length, area, or volume are scaled by the determinant of the Jacobian matrix, which acts as a local magnification factor.

5. Integration Limits and Order Selection in FEA: Full vs. Reduced Integration

Choosing the number of Gauss points (the integration order) is a critical design decision in FEA that directly affects the behavior of elements:

Full Integration

Full integration uses a quadrature rule that is of sufficient order to integrate all terms of the element stiffness matrix exactly for a regular, undistorted element geometry.

Reduced Integration

Reduced integration uses a lower-order quadrature rule than what is required for full integration. For instance, using a $1 \times 1$ Gauss Quadrature rule (a single point at the center $\xi = 0, \eta = 0$ with weight $w = 4.0$) for a 4-node quadrilateral element.