Calculus III: Tangent Planes and Linear Approximation

Understanding how to approximate a complex, curvy surface with a simple, flat plane is a cornerstone of engineering mathematics. It allows you to model real-world systems—like the stress on a beam or the output of a chemical reactor—locally, using manageable linear calculations instead of unwieldy nonlinear ones. This process of linearization is indispensable for error analysis, optimization, and simplifying complex differential equations in your engineering work.

The Tangent Plane: A Multivariable Foundation

In single-variable calculus, you approximate a curve near a point using its tangent line. For a surface defined by $z=f(x,y)$, the analogous concept is the tangent plane. Just as the tangent line's slope is given by the derivative $f^{\prime }(a)$, the orientation of the tangent plane is determined by the rates of change in the $x$ and $y$ directions—the partial derivatives.

If a function $f$ is differentiable at a point $(x_0,y_0)$—meaning it is smooth and has no sharp edges or corners there—then the equation of the tangent plane is: $$L(x,y)=f(x_0,y_0)+f_x(x_0,y_0)(x-x_0)+f_y(x_0,y_0)(y-y_0).$$ Here, $f_x$ and $f_y$ are the first partial derivatives. The term $f(x_0,y_0)$ is the "anchor point" on the surface. The partial derivatives $f_x$ and $f_y$ act as slopes, telling you how much $z$ changes per unit change in $x$ (holding $y$ constant) and in $y$ (holding $x$ constant), respectively. Visually, you can think of the tangent plane as the unique flat surface that just "grazes" the more complex surface at the point of tangency.

Linearization: The Local Approximation Formula

The function $L(x,y)$ defined above is also called the linearization of $f$ at $(x_0,y_0)$. This is our working approximation tool. The core idea is that near the point $(x_0,y_0)$, the complicated surface $z=f(x,y)$ is very close to the simple plane $z=L(x,y)$.

Example: Approximate the value of $\sqrt{(3.02{)}^2+(3.97{)}^2}$. Let $f(x,y)=\sqrt{x^2+y^2}$. We choose a nearby point easy to compute: $(x_0,y_0)=(3,4)$. Note $f(3,4)=\sqrt{9+16}=5$. The partial derivatives are $f_x=x/\sqrt{x^2+y^2}$ and $f_y=y/\sqrt{x^2+y^2}$. At $(3,4)$, $f_x(3,4)=3/5=0.6$ and $f_y(3,4)=4/5=0.8$. Our changes are $\Delta x=3.02-3=0.02$ and $\Delta y=3.97-4=-0.03$. The linearization gives the approximation: $$L(3.02,3.97)=f(3,4)+f_x(3,4)(0.02)+f_y(3,4)(-0.03)=5+(0.6)(0.02)+(0.8)(-0.03).$$ This simplifies to $5+0.012-0.024=4.988$. The true value is about $4.9881$, confirming an excellent approximation.

The Total Differential: Measuring Infinitesimal Change

While linearization gives the new approximate function value $L(x,y)$, the total differential $dz$ measures the corresponding approximate change in the output from the anchor point. It is defined as: $$dz=f_x(x_0,y_0)dx+f_y(x_0,y_0)dy.$$ Here, $dx$ and $dy$ represent small changes (or errors) in the independent variables $x$ and $y$. The total differential $dz$ then estimates the resulting change in $z$.

Geometrically, if you move from $(x_0,y_0,f(x_0,y_0))$ on the surface to a nearby point on the tangent plane, the vertical change in the plane's height is exactly $dz$. The actual change in the function, $\Delta z=f(x_0+dx,y_0+dy)-f(x_0,y_0)$, is approximated by $dz$. The formula shows how sensitivity to input errors is weighted by the respective partial derivatives.

Error Estimation Using Differentials

This concept is directly applied in engineering for propagation of measurement error. You often measure quantities $x$ and $y$ with known possible errors or tolerances $|\Delta x|$ and $|\Delta y|$. The differential provides a way to estimate the maximum possible error in a calculated quantity $z=f(x,y)$.

The approximate absolute error in $z$ is $|dz|$. A more practical, worst-case estimate is given by: $$\Delta z\approx |f_x(x,y)|\cdot |\Delta x|+|f_y(x,y)|\cdot |\Delta y|.$$ This formula comes from taking the absolute value of each term in the differential sum, providing a conservative bound.

Engineering Scenario: You calculate the power dissipated in a resistor using $P=I^{2}R$. You measure current as $I=2.00\text{ A}\pm 0.01\text{ A}$ and resistance as $R=100.0\Omega \pm 0.2\Omega$. Estimate the maximum absolute and relative error in the calculated power.

First, $P(2,100)=(4)(100)=400$ W. The function is $P(I,R)=I^{2}R$. Partial derivatives: $P_I=2IR$ and $P_R=I^2$. At the measured point: $P_I(2,100)=400$ and $P_R(2,100)=4$. Tolerances: $|\Delta I|=0.01$, $|\Delta R|=0.2$.

Maximum absolute error: $$\Delta P\approx |P_I|\cdot |\Delta I|+|P_R|\cdot |\Delta R|=|400|\cdot 0.01+|4|\cdot 0.2=4+0.8=4.8\text{ W}.$$ Thus, $P=400\text{ W}\pm 4.8\text{ W}$.

The relative error is $\frac{\Delta P}{P}\approx \frac{4.8}{400}=0.012=1.2\%$. This analysis shows which measurement error ($I$ or $R$) contributes more to the uncertainty in $P$, guiding where to improve instrumentation.

Common Pitfalls

1. Applying the formula to non-differentiable points: The tangent plane and linear approximation only exist where the function is differentiable. A common red flag is a point where a partial derivative does not exist or the surface has a cusp or ridge. Always check that $f_x$ and $f_y$ exist and the function is smooth at your point of interest.
2. Confusing the differential $dz$ with the actual change $\Delta z$: The differential $dz$ is an approximation of the true change $\Delta z$, based on the tangent plane. They are equal only for a linear function. For nonlinear functions, $dz$ is a good estimator only for very small changes $dx$ and $dy$.
3. Misapplying the absolute error formula: When calculating maximum possible error for error propagation, you must use the absolute values of the partial derivatives multiplied by the absolute values of the input errors ($|f_x|\cdot |\Delta x|$). Simply plugging in the differential $f_x\Delta x+f_y\Delta y$ can allow positive and negative errors to cancel, giving a non-conservative (and potentially dangerous) underestimate of worst-case error.
4. Forgetting the anchor point in linearization: A common computational error is to use the general partial derivative formulas $f_x(x,y)$ and $f_y(x,y)$ instead of evaluating them at the specific anchor point $(x_0,y_0)$. Remember, the coefficients in the tangent plane equation are numbers, not functions, at the point of linearization.

Summary

• The tangent plane to a surface $z=f(x,y)$ at a point $(x_0,y_0)$ provides the best linear approximation to the surface nearby. Its equation is $L(x,y)=f(x_0,y_0)+f_x(x_0,y_0)(x-x_0)+f_y(x_0,y_0)(y-y_0)$.
• The function $L(x,y)$ is the linearization of $f$ and is used to approximate difficult function values near a known point.
• The total differential $dz=f_{x}dx+f_{y}dy$ estimates the change in the function's output ($\Delta z$) resulting from small changes $dx$ and $dy$ in its inputs.
• Differentials are directly applied in engineering for error estimation and propagation of measurement error. The maximum possible error in a calculated quantity is approximated by $\Delta z\approx |f_x|\cdot |\Delta x|+|f_y|\cdot |\Delta y|$.
• These techniques are powerful tools for sensitivity analysis, simplifying models, and understanding how uncertainties in measurements affect final results in your designs and calculations.