Cheat sheet on derivatives and optimization September 11-15, 2023 The concept of the derivative The meaning of the derivative Imagine a graph of distance vs. time! Now the distance over time (the slope of our function) is the velocity. If we pick a point on the graph, and draw a ray from the origin to the graph like this: 1 the slope will measure the average speed up till that moment. The average speed is quite useful – e.g. when we want to estimate how much time we will need to cover a similar distance. In other cases, however we are more interested in our momentary speed – e.g when we want to check whether we exceeded the speed limit at a certain point. The graphical representation of this would be a tangent to our function: 2 The deraivative is nothing else but the slope of this tangent. Why use derivatives in economics We are going to use derivatives a lot, and I do mean a lot. the reasons for this are twofold. Economist heavily rely on marginal thinking. When you determine how many workers to employ at your firm, you do not ask yourself the question: ”Should I employ 340 people?” Rather you ask: ”Should I hire one more person?” The choices we make are more about small adjustments, then all-out decisions. Rationality was about finding what is the best of the available options. Most of the cases we are trying to model behavior by striving to achieve a maximum: people want to maximize their welfare, firms want to maximize their profit, etc. That means that the mathematical form of many of our exercises will be some kind of optimization problem. Optimization methods – as we will soon see – rely on derivatives. Calculating the derivative Without going deeply into the math here, I am going to give you a handful of rules. If you memorize these simple rules, you are going to be able to handle the overwhelming majority of the functions that you encounter in this course. The constant function Rule: f (x) = c ⇒ f ′ (x) = 0 Example: f (x) = 4 f ′ (x) = 0 The linear function Rule: f (x) = k · x + c ⇒ f ′ (x) = k Example: f (x) = 2x + 5 f ′ (x) = 2 3 The generalized power function Rule: f (x) = k · xa ⇒ f ′ (x) = k · a · xa−1 Example: f (x) = 2x5 f ′ (x) = 2 · 5 · x4 = 10 · x4 Rule for sums and differences Rule: f (x) ± g(x) ⇒ f ′ (x) ± g ′ (x) Example: f (x) = x3 g(x) = 2x − 3 h(x) = x3 + (2x − 3) h′ (x) = 3 · x2 + 2 *Bonus material: the natural logarithm function Rule: f (x) = k · ln x ⇒ f ′ (x) = k x Example: f (x) = 3 · ln x f ′ (x) = 3 x *Bonus material: composite functions Rule: h(x) = f (g(x)) ⇒ h′ (x) = f ′ (g(x)) · g ′ (x) Example: f (x) = x3 g(x) = 5x − 3 h(x) = (5x − 3)3 h′ (x) = 3 · (5x − 3)2 · 5 = 15 · (5x − 3)2 4 More on derivatives Higher-order derivatives We get functions as results, so we can take their derivatives as well. f (x) = x3 − 4x2 + 3x − 10 f ′ (x) = 3 · x2 − 8x + 3 (first-order derivative) f ′′ (x) = 6x − 8 (second-order derivative) f ′′′ (x) = 6 (third-order derivative) and so on. Increasing and decreasing functions By calculating the value of the derivative at certain points we can learn more about the shape of a function. If the first-order derivative is positive, the function goes upwards, thus the function at this point is increasing. If the first-order derivative is negative, the function goes downwards, thus the function at this point is decreasing. *Bonus material: Convex and concave functions Some functions ”bend” upwards (convex functions), some ”bend” downwards (concave functions).The second-order derivative indicates which is the case. If the second-order derivative is positive, the function is convex at that point. Let’s take the function f (x) = x2 as an example! f ′ (x) = 2x f ′′ (x) = 2 That is positive (everywhere), so this function ”bends” upward. If the second-order derivative is√negative, the function is concave at that point. Let’s take the function f (x) = x = x0.5 as an example! f ′ (x) = 0.5 · x−0.5 1 4 · x1.5 That is negative (for every positive number), so this function ”bends” downward. f ′′ (x) = −0.25 · x−1.5 = − 5 Optimization Maxima and minima If we are looking for the (local) maximum of a function, it cannot be on an increasing part of a function, since then moving along the function would bring us higher. It cannot be on a decreasing part either, because then moving back would get us higher. Hence at the maximum, the first-order derivative can neither be positive, nor negative. This leaves us with one option: at the maximum, the first-order derivative is zero. (This is often referred to as the first-order condition, abbreviated as FOC.) We can use a similar reasoning for (local) minima as well. The minimum cannot be on a decreasing part of a function, since then moving along the function would bring us lower. It cannot be on an increasing part either, because then moving back would get us lower. Hence at the minimum, the first-order derivative can neither be positive, nor negative, only zero. How to decide whether we have a minimum or a maximum? In most exercises in this course, you’ll be told whether you should look for a maximum or a minimum. Bonus material: However, if you want to check it for yourself, the secondorder derivative is going to be helpful. If it’s negative, we have a maximum. Recall that a negative second-order derivative means that the first-order derivative is decreasing, thus turning from positive to zero to negative. We have an increasing part, the maximum, then a decreasing part. If the second-order derivative is positive, we have a minimum. (A decreasing part, a minimum, then an increasing part.) What happens if the second-order derivative is zero? Then it’s neither a minimum, nor a maximum: it is a point of inflection. Here the function turns from convex to concave or vice versa. Examples: min f (x) = x2 − 4x + 10 F OC : f ′ (x) = 2x − 4 = 0 Solving that, we find that the minimum is at: x=2 [Double-checking that we truly have a minimum: f ′′ (x) = 2, so we are good.] max f (x) = −x2 + 10x + 2 F OC : f ′ (x) = −2x + 10 = 0 6 Solving that, we find that the maximum is at: x=5 [Double-checking that we truly have a maximum: f ′′ (x) = −2, so everything is fine.] Multivariate functions Many functions have more than one variable, e.g. z(x, y) = 3x + 2xy + y 2 . How can we handle functions like that? We will introduce the concept of partial derivatives. I.e. we are going to pretend that we have only one variable. So the partial derivative of z with respect to dz ) is taking the derivative of z while pretending that y is a x (denoted as dx constant. dz = 3 + 2y + 0 dx Similarly, when we calculate the partial derivative of z w.r.t. y, we pretend that x is a constant. dz = 0 + 2x + 2y dy Optimization and multivariate functions We are going to proceed like we did previously: we are going to use the first-order condition to find the maximum (or minimum) point of a function. However, we are going to use more than one FOC, one or each partial derivative. Example: max z(x, y) = 60x + 34y − 4xy − 6x2 − 3y 2 + 5 x,y z = 60 − 4y − 12x = 0 x z F OCy : = 34 − 4x − 6y = 0 y F OCx : x∗ = 4, y ∗ = 3 Note that this time again we could check whether we have found a maximum or a minimum point. However, the methods are too advanced for our discussion, so the interested reader is referred to any college textbook on calculus. 7 Optimization with a constraint Many times, it is impossible to maximize (or minimize) a function without constraint. (It is worth to also ponder the philosophical implications of the above statement.) Would you be able to find an x and a y that maximizes f (x, y) = x + y? Also, in real life almost all of our choices are constrained one way or the other. I cannot spend more than my budget, I cannot allocate more than 24 hours per day for activities, etc. Let us look at an example to deal with the mathematical formulation of this problem: max x · y x,y subject to the constraint: x + y = 20. Our first step is to use the constraint to express one of the variables: y = 20 − x. Secondly, we plug this back to our original function: max x · (20 − x). x We killed two birds with one stone! We have only one variable and thus one FOC, and more importantly, we don’t have to worry about the constraint: it’s built into our new function! The final steps of the solution: max x · (20 − x) = 20x − x2 x F OC : 20 − 2x = 0 x = 10, y = 20 − x = 10 Bonus material: The Lagrange method An alternate way to optimize multivariate functions with a constraint is the Lagrange method. Here we find a different way to work the constraint into our function. Let’s solve the previous problem again! Our first step is to rearrange the constraint to zero: x + y − 20 = 0 We multiply the left-hand side by λ and subtract the result from our original function, thereby obtaining our Lagrange function: L = x · y − λ · (x + y − 20) 8 Wait, what? Now we have three variables! That means we need three FOCs to solve this problem. F OCx : y − λ = 0 F OCy : x − λ = 0 F OCλ : x + y − 20 = 0 Easy calculation yields: x = 10, y = 10, λ = 10 We got the previous result back and also some extra info. What does it mean that λ = 10? It roughly means that if we increase our constraint by 1 (x + y = 21), then our maximum value will increase (approximately) by 10. (You can try to check this.) So while this method requires more work, it rewards as with some built in sensitivity check. Practice problems a) Maximize √ u(x, y) = 4 x + y s.t. the constraint x + 2y = 100 b) Minimize T C(K, L) = 4K + 2L s.t. √ K · L = 10 Parting words Even though this whole stuff might sound intimidating, don’t worry! Practice makes perfect and you’ll get the hang of it soon. Moreover, next week you will see, why this was so important... 9