The [[Differentiation Rules#Chain Rule|Chain Rule]] for a univariate function $f(x)$ where $x$ depends on $t$ states: $ \frac{df}{dt} = \frac{df}{dx} \cdot \frac{dx}{dt}$ Now, suppose if $f$ is a function of multiple variables, where each $x_i$ depends on the same parameter $t$. $ f\big(x_1(t), \dots, x_n(t)\big)$ The *multivariate Chain rule* means to apply the Chain rule to each variable $x_i$ and summing all contributions. The outcome is the *Total Derivative* of $f$ with respect to $t$. $ \frac{df(x_1,\dots,x_n)}{dt}= \left(\frac{\partial f}{\partial x_1}*\frac{\partial x_1}{\partial t}\right) + \cdots + \left(\frac{\partial f}{\partial x_n}*\frac{\partial x_n}{\partial t}\right) $ Interpretation: - $\frac{\partial f}{\partial x_i}$ measures how $f$ responds to changes in the variable $x_i$. - $\frac{\partial x_i}{\partial t}$ measures how $x_i$ changes as $t$ changes. Multiplying these and summing over all $i$ gives the "net effect" of $t$ on $f$. **Vector notation:** If we arrange the [[Partial Derivative|partial derivatives]] $\frac{\partial f}{\partial x_i}$ into a [[Jacobian#Jacobian Vector|Jacobian vector]] (row vector by convention), and gather the rates ​$\frac{\partial x_i}{\partial t}$ into a column vector, we get: $ \frac{df}{dt}= \begin{bmatrix} \frac{\partial f}{\partial x_1} &\cdots & \frac{\partial f}{\partial x_n} \end{bmatrix} *\begin{bmatrix} \frac{\partial x_1}{\partial t} \\[4pt] \frac{\partial x_2}{\partial t} \\\vdots \\ \frac{\partial x_n}{\partial t} \end{bmatrix} $ The Total Derivative becomes: $ \frac{df}{dt} = \frac{df}{dx} \cdot \frac{dx}{dt} = J_f \cdot \frac{dx}{dt}$ > [!note:] > We have assumed all $x_i$ depend on the same parameter $t$ for simplicity, but this is not necessary. If each $x_i$ depends on different parameters, the Total Derivative can still be defined by considering how $f$ changes w.r.t. each parameter individually and then combining these changes. ## Example Consider a function $f(x(u(t)))$, where: + $f(x)$ is scalar-valued, depending on $x = (x_1, x_2)$. + $x(u)$ is vector-valued, depending on $u=(u_1, u_2)$. + $u(t)$ is vector-values, depending on scalar $t$. $ f(x) = f(x_1, x_2), \quad x(u) = \begin{bmatrix} x_1(u_1, u_2) \\ x_2(u_1, u_2) \end{bmatrix}, \quad u(t) = \begin{bmatrix} u_1(t) \\ u_2(t) \end{bmatrix} $ The chain rule for this composition can be written as: $ \frac{df}{dt} = \frac{\partial f}{\partial x} \cdot \frac{ \partial x}{\partial u} \cdot \frac{\partial u}{ \partial t} $ In matrix form: $ \frac{df}{dt} = \underbrace{ \begin{bmatrix} \frac{\partial f}{\partial x_1} & \frac{\partial f}{\partial x_2} \end{bmatrix}}_{J_f} \quad \underbrace{ \begin{bmatrix} \frac{\partial x_1}{\partial u_1} & \frac{\partial x_1}{\partial u_2} \\[4pt] \frac{\partial x_2}{\partial u_1} & \frac{\partial x_2}{\partial u_2} \end{bmatrix}}_{J_x} \quad \underbrace{ \begin{bmatrix} \frac{\partial u_1}{\partial t} \\[4pt] \frac{\partial u_2}{\partial t} \end{bmatrix} }_{\frac{\partial u}{\partial t}} $ Simplified: $ \frac{df}{dt}= J_f*J_x*\frac{\partial u}{\partial t} $