Partial Derivatives in Economics: Marginal Products, Multivariable Optimisation and the Chain Rule
A firm chooses labour and capital. A consumer eats many goods. Demand depends on price, income and substitutes at once. The ordinary derivative handles one variable; economics almost always has several, and the partial derivative is how you hold the world still enough to see one at a time.
Single-variable calculus is a simplification we teach first. Total cost usually depends on more than output; utility depends on every good you consume, not one. The real world is multivariable. When you want the effect of changing just one thing, you need to isolate it — to ask how output responds to labour if capital stays put. That is the partial derivative: the ordinary derivative with a discipline attached — differentiate with respect to one variable while treating all others as constants.
The core idea: hold everything else constant
The notation switches from d to the curly partial symbol. The partial of Q with respect to L means the rate of change of Q as L changes, with every other variable frozen.
The one rule you need. To take a partial derivative with respect to a variable, differentiate as normal for that variable and treat every other variable as a number. If you can differentiate 3x squared, you can differentiate 3x squared times y — with respect to x, the y is just a constant multiplier, giving 6xy. That is the entire technique.
Every rule you know still applies — power, product, chain. The only new instruction is: pick a variable, freeze the rest, proceed as usual.
A first example
Let a production function be Q = 5 times L squared times K + 3 times L times K squared + 100.
Partial with respect to L (treat K as constant): 10LK + 3K squared. Partial with respect to K (treat L as constant): 5 times L squared + 6LK. Same function, different frozen variable, different derivative — each answers a different economic question.
Marginal products, properly this time
The marginal product of labour is the extra output from one more worker, holding capital fixed — which is exactly the partial of Q with respect to L. For Cobb-Douglas Q = A times L to the power alpha times K to the power beta, the marginal product of labour is alpha times A times L to the power (alpha minus 1) times K to the power beta.
Notice something single-variable analysis misses: the marginal product of labour depends on how much capital there is. Give a worker more machines and they produce more. That interaction is invisible without partial derivatives — and it is why capital investment raises wages, and why the same worker earns more in a capital-rich country.
The utility case: marginal utility and the MRS
If utility is U(X, Y), the marginal utility of X is the partial with respect to X. The marginal rate of substitution — how much Y a consumer gives up for one more X while staying equally happy — is the ratio of the two partials:
MRS = MU of X over MU of Y = (partial of U with respect to X) over (partial of U with respect to Y)
The slope of the indifference curve, expressed through partial derivatives.
Example. Let U = X to the 0.5 times Y to the 0.5. Then MU of X = 0.5 times X to the power minus 0.5 times Y to the 0.5, and MU of Y = 0.5 times X to the 0.5 times Y to the power minus 0.5. Their ratio is Y over X. So willingness to trade Y for X depends on the ratio in which the goods are held — the more X held relative to Y, the less an extra X is worth. Diminishing MRS, in three lines.
Consumer theory, production theory and optimisation all lean on partial derivatives. The Economics Made Simple bundle builds these up from scratch with worked indifference-curve and isoquant problems, and the Case Studies collection shows the same maths applied to real consumer and firm data.
Second-order partials and the cross-partial
Differentiate again. The second-order partial of Q with respect to L twice tells you whether marginal product is rising or falling — a negative value is diminishing returns. More interesting is the cross-partial: differentiate the partial of Q with respect to L, then by K. It measures how the marginal product of labour changes as capital changes.
What the cross-partial tells you. If it is positive, labour and capital are complements in production — more capital makes each worker more productive. If negative, they are substitutes in the productivity sense. This single number captures whether investing in machines raises or lowers the value of workers — central to the economics of automation and wages.
Young’s theorem says the order of differentiation does not matter: differentiating by labour then capital equals capital then labour. It always holds for well-behaved functions, and is occasionally worth a mark to state.
Multivariable optimisation: the first-order conditions
At an unconstrained optimum, every partial derivative must be zero simultaneously.
partial of profit w.r.t. L = 0 and partial of profit w.r.t. K = 0
At a peak, the surface is flat in every direction — so each partial vanishes.
Imagine profit as a hill over a landscape whose directions are labour and capital. At the summit, the ground is level whichever way you look. Level in every direction is precisely every partial equals zero.
Worked example. Profit is π = 100L + 80K − L squared − K squared − LK. First-order conditions: partial w.r.t. L = 100 − 2L − K = 0; partial w.r.t. K = 80 − 2K − L = 0. From the first, K = 100 − 2L. Substitute into the second: 80 − 2(100 − 2L) − L = 0, so 3L = 120, L = 40. Then K = 20. The optimum is L = 40, K = 20. The minus-LK term made the conditions interdependent — optimal labour depends on capital and vice versa, which single-variable calculus could never capture.
Common error — solving the conditions separately. The two first-order conditions form a simultaneous system and must be solved together. Students often set the partial w.r.t. L to zero and solve for L alone, ignoring that K appears in it. Because the variables interact, you must solve jointly.
The chain rule with several variables
Suppose output depends on labour and capital, and both depend on time as the firm grows. The multivariable chain rule adds the effect through each channel:
dQ/dt = (partial of Q w.r.t. L)(dL/dt) + (partial of Q w.r.t. K)(dK/dt)
Total rate of change = effect through labour + effect through capital. Sum the paths.
This is the engine behind growth accounting: output growth decomposes into the contribution from labour growth (weighted by its marginal product) plus capital growth (weighted by its marginal product) plus productivity growth. The chain rule is why growth can be split into sources at all.
Case Study — Comparative Statics and the Slutsky Equation
When the price of a good falls, a consumer buys more for two reasons. First, it is now cheaper relative to others, so they substitute toward it — the substitution effect. Second, their real purchasing power has risen, so they can afford more of everything — the income effect. Both come from one price change, and separating them needs partial derivatives.
The Slutsky equation formalises this, writing the total response of demand to a price change as the sum of a pure substitution term and an income term, each itself a partial derivative.
Why it matters: it explains Giffen goods — the paradox where a price rise increases quantity demanded. For a strongly inferior good on which the poor spend a large budget share, the income effect can be large, negative, and outweigh the substitution effect. Without partial derivatives to separate the two channels, the Giffen paradox is inexplicable; with them, it falls out of the algebra as a specific condition. This is comparative statics at its most powerful — using calculus to predict a counterintuitive real-world response.
Research Spotlight — Automation and the Cross-Partial
The question of the decade: does investing in machines and algorithms raise or lower the value of human workers? The answer lives in a cross-partial derivative, and economists have fought over its sign.
The optimistic view holds that capital and labour are complements: the cross-partial is positive. More capital raises the marginal product of labour, and therefore wages. Historically this often held — the tractor made farmers more productive, the computer made analysts more productive, and pay rose.
Daron Acemoglu and Pascual Restrepo challenged the simple story. They distinguished capital that complements labour from capital that displaces it by automating tasks humans used to do. When automation shifts tasks from workers to machines, the effect on the marginal product of the displaced worker can be negative, even as total output rises. The aggregate cross-partial masks very different effects across different kinds of worker.
Why this matters for you: the entire debate about robots, AI and jobs can be framed as an argument about the sign and heterogeneity of a cross-partial derivative. When an exam asks whether technology helps or harms workers, the sophisticated frame is: it depends on whether new capital complements labour (positive cross-partial) or displaces it.
Bringing it together
| Concept | Partial derivative | Economic meaning |
|---|---|---|
| Marginal product of labour | partial of Q w.r.t. L | Extra output per worker, capital fixed |
| Marginal utility of X | partial of U w.r.t. X | Extra satisfaction per unit of X |
| Marginal rate of substitution | ratio of the two partials | Slope of the indifference curve |
| Diminishing returns | second partial w.r.t. L < 0 | Marginal product falling |
| Complementarity of inputs | cross-partial > 0 | More K raises MP of L |
| Optimum (two variables) | both partials = 0 | Flat in every direction |
Every core object in production and consumer theory is a partial derivative wearing an economic name.
AP & Cambridge A-Level Exam Technique
1. Say which variable you are holding constant. State holding K constant before differentiating — it signals understanding and stops you differentiating the wrong variable.
2. Other variables are constants, including as coefficients. In 5 times L squared times K, differentiating by L gives 10LK. The commonest error is trying to differentiate the frozen variable too.
3. Reduce the exponent on the active variable only. For A times L to alpha times K to beta by L, the L exponent drops to alpha minus 1; K is untouched.
4. Solve first-order conditions simultaneously. Two partials set to zero form a system; solve them together, especially with an interaction term like LK.
5. Use the cross-partial for complementarity questions. Compute it and state its sign. Positive means complements.
6. Form the MRS as a ratio of partials. Watch which good is in the numerator — the good on the horizontal axis goes on top by convention.
7. Interpret, do not just compute. After finding the marginal product of labour, add: so it rises with capital — workers are more productive with more machines.
Practice Questions
Question 1 — Basic partials (5 marks)
A production function is Q = 6 times L squared times K cubed + 4 times L times K + 50. (a) Find the partial w.r.t. L. (b) Find the partial w.r.t. K. (c) Evaluate the marginal product of labour at L = 2, K = 1.
(a) 12 times L times K cubed + 4K. [2]
(b) 18 times L squared times K squared + 4L. [2]
(c) At L = 2, K = 1: 12(2)(1) + 4(1) = 28. [1]
(b) 18 times L squared times K squared + 4L. [2]
(c) At L = 2, K = 1: 12(2)(1) + 4(1) = 28. [1]
Question 2 — Marginal rate of substitution (6 marks)
A consumer has U = X to the 0.3 times Y to the 0.7. (a) Find the marginal utility of each good. (b) Derive the MRS. (c) At X = 10, Y = 20, how many units of Y is the consumer willing to give up for one more X?
(a) MU of X = 0.3 times X to minus 0.7 times Y to 0.7; MU of Y = 0.7 times X to 0.3 times Y to minus 0.3. [2]
(b) MRS = (0.3 over 0.7) times (Y over X) = (3/7)(Y/X). [2]
(c) At X = 10, Y = 20: (3/7)(2) = 6/7, about 0.86. [2]
(b) MRS = (0.3 over 0.7) times (Y over X) = (3/7)(Y/X). [2]
(c) At X = 10, Y = 20: (3/7)(2) = 6/7, about 0.86. [2]
Question 3 — Multivariable optimisation (7 marks)
Profit is π = 120L + 90K − 2 times L squared − 3 times K squared − 2LK. Find the profit-maximising L and K.
Conditions: partial w.r.t. L = 120 − 4L − 2K = 0; partial w.r.t. K = 90 − 6K − 2L = 0. [2]
Solve: from the first, K = 60 − 2L. Substitute: 90 − 6(60 − 2L) − 2L = 0, so 10L = 270, L = 27. Then K = 6. [4]
Optimum: L = 27, K = 6. [1] The minus-2LK term links the conditions, so solve jointly.
Solve: from the first, K = 60 − 2L. Substitute: 90 − 6(60 − 2L) − 2L = 0, so 10L = 270, L = 27. Then K = 6. [4]
Optimum: L = 27, K = 6. [1] The minus-2LK term links the conditions, so solve jointly.
Question 4 — Evaluation (10 marks)
Explain how the cross-partial derivative captures the relationship between capital and labour, and evaluate what it implies for whether automation raises or lowers wages.
Mechanism: [4] The marginal product of labour is the partial of Q w.r.t. L. The cross-partial measures how that changes as capital changes. Positive means complements — more machines raise each worker’s productivity and (since competitive wages equal the marginal product) wages. Negative means substitutes — more capital lowers the marginal product of labour and depresses wages.
Evaluation: [6] Optimistic case: historically capital largely complemented labour (tractors, computers), so if the cross-partial stays positive, automation keeps raising wages. Displacement case: Acemoglu and Restrepo argue automation of human tasks can push the effective cross-partial negative for displaced workers — output rises but their wages fall. Heterogeneity: the aggregate number hides very different effects — complementing high-skill workers while displacing routine-task workers, widening inequality. Time horizon: short-run displacement may give way to new tasks restoring complementarity. Limits: real automation happens in discrete jumps with adjustment costs the smooth derivative ignores. Judgement: the cross-partial frames the question — automation raises wages if new capital complements labour more than it displaces it — but the outcome is empirical and distributional, depending on the type of capital, type of worker, and horizon.
Summary
Partial derivatives handle the fact that almost nothing in economics depends on one variable. The technique is a single habit: differentiate with respect to one variable, treat the rest as constants. From it come the marginal products, marginal utilities and the marginal rate of substitution — each carrying the holding-everything-else-fixed their definitions demand.
Second-order partials add curvature (diminishing returns when negative) and, through the cross-partial, tell you whether inputs are complements or substitutes — the crux of the automation debate. Multivariable optimisation sets every partial to zero at once, solved as a system, and the multivariable chain rule decomposes total change into contributions from each channel, exactly as growth accounting does. See the partial hiding behind marginal, and the multivariable economy becomes legible.
References
- Acemoglu, D. and Restrepo, P. (2019) Automation and new tasks, Journal of Economic Perspectives, 33(2), pp. 3-30.
- Chiang, A.C. and Wainwright, K. (2005) Fundamental Methods of Mathematical Economics. 4th edn. New York: McGraw-Hill.
- Slutsky, E. (1915) Sulla teoria del bilancio del consumatore, Giornale degli Economisti, 51, pp. 1-26.
- Simon, C.P. and Blume, L. (1994) Mathematics for Economists. New York: W.W. Norton.
- Varian, H.R. (2014) Intermediate Microeconomics. 9th edn. New York: W.W. Norton.