With two decisions x and y, ask one question at a time: how does f change if I move x and keep y fixed? That rate is the partial derivative \partial f / \partial x. It is an ordinary derivative in which the other variable is treated as a constant. At a smooth unconstrained optimum both partial derivatives are zero.
\frac{\partial f}{\partial x}(x, y) = \lim_{h \to 0} \frac{f(x + h, y) - f(x, y)}{h} In words: the slope of f along the x direction, with y held still; \partial f/\partial y is the same with the roles swapped.
f_{xy} = \frac{\partial}{\partial y}\Bigl(\frac{\partial f}{\partial x}\Bigr) = f_{yx} In words: the mixed second partial says how the x-slope changes as y moves; for smooth functions the order does not matter.
f_x(x^*, y^*) = 0, \qquad f_y(x^*, y^*) = 0 In words: at an interior optimum no single variable can improve f on its own.
A workshop makes x tables and y chairs a week, with profit P(x, y) = 60x + 50y - x^2 - xy - y^2 dollars. The marginal profits are P_x = - 2 x - y + 60, \qquad P_y = - x - 2 y + 50 \quad \text{(dollars per table, per chair)} . In words: each extra table earns $60, less a growing amount as more tables and chairs compete for the same shop. At (10, 10): P_x = 30 and P_y = 20 dollars per unit, so more of both still pays. The mixed partials agree: P_{xy} = -1 and P_{yx} = -1. Setting both marginal profits to zero gives x = \frac{70}{3}, y = \frac{40}{3} (fractions of a unit; the continuous model allows them), with profit \frac{3100}{3} dollars. The kernel agrees:
maximize 60*x + 50*y - x^2 - x*y - y^2
# x = tables, y = chairs per week; profit in dollars
x = 23.3333, y = 13.3333, profit 1033.3333 dollars (rounded to 4 decimals), status optimal.
Partial derivatives are the entries of the gradient, which every smooth solver uses, and in the answer they read as marginal values: dollars per extra unit. Whether the zero-slope point is a maximum is a second-derivative question, answered by the Hessian (see the Hessian refresher).