Jump to content

Maths

English
From Study Outlet, Bocconi course notes

First partial: 21 October 2026. The 30319 syllabus defines the first partial as the material taught from the beginning of the course up to the point reached when the exam takes place. It does not specify a last lecture or a fixed chapter cutoff. These notes provisionally follow syllabus items 1–17: linear algebraVectors, operations and linear (in)dependenceA vector is an ordered list of real numbers. Its dimension is the number of components. A column vector stacks its components vertically; a row vector places them horizontally. The position of a component carries its meaning, so a vector of policy… and differential calculusFunctions of one real variableA function assigns one output to each admissible input. For a function of one real variable, both the input and output are numbers. The domain is the set of admissible inputs; the graph collects the points (x,f(x)). In social-science applications,…, through constrained optimizationConstrained optimization and applicationsConstrained optimization chooses the best value of an objective function among inputs satisfying a restriction. A resource constraint prevents the independent variables from being varied freely. The source introduces Lagrange multipliers through a…. The final topic must be confirmed against the forthcoming class notes. Integral calculus and dynamic systems are outside this provisional block.

The course presentation confirms the 21 October date. Both partials contain open-answer questions and closed-answer questions. The 2026/2027 syllabus assigns each partial 50% of the final grade. The older first partials have different coverage: October 2015 includes Leontief systemsLinear algebra exercisesThe supplied past exams combine matrix operations, determinants, inversion and economic interpretation. Each question should be split into a mathematical condition and a statement about the model. A singular matrix can invalidate a claimed inverse… and one-variable differentiationFunctions exercisesThe functions exercises require the same sequence of decisions used in theory: specify the domain, differentiate, solve the stationary equations, classify the candidates and state the answer in the model’s terms. The supplied textbook exercises…, while October 2016 contains transition matrices and another Leontief model. Historical questions are practice evidence, not a guarantee of the 2026 cutoff.

Definitions, matrix operations, derivativesDerivativesThe derivative measures the local slope of a function. The source introduces it through a local affine approximation: close to a point, a differentiable function behaves like its tangent line. Its increment consists of a term proportional to the…, worked examples and optimization arguments come from the supplied textbook, vector slides and past exams. Citations use PDF page numbers, counting the first file page as page 1. The textbook’s printed page numbers differ: Chapter 1 generally has a ten-page offset and Chapter 2 a nine-page offset. A calculation is marked explicitly wherever the supplied source prints a conflicting result.

Vectors, operations and linear (in)dependence

A vector is an ordered list of real numbers. Its dimension is the number of components. A column vector stacks its components vertically; a row vector places them horizontally. The position of a component carries its meaning, so a vector of policy decisions or election shares cannot be reordered without changing the interpretation. Transposition changes its orientation and preserves its components.

Representation in real coordinate spaces

The set Rn contains all vectors with n real components. A two-dimensional vector can be represented by a point in the Cartesian plane or an arrow from the origin. A three-dimensional vector uses three coordinates. The zero vector has every component equal to zero. Coordinates are numbers; the dimension counts how many coordinates there are.

x=[x1x2⋮xn]∈ℝn
[1−15]T=[1−15]

Addition and scalar multiplication

Vector addition is defined for vectors of the same dimension and combines corresponding components. Geometrically, addition gives the diagonal of the parallelogram constructed on the two arrows. Scalar multiplication multiplies each component by the same real number. A positive scalar preserves direction, a negative scalar reverses it, and zero produces the zero vector.

(x+y)i=xi+yi
(αx)i=αxi
  • Commutativity: the order of the two vectors in a sum does not matter.
  • Associativity: grouping successive additions does not change the result.
  • The zero vector is the identity for addition; the opposite vector cancels a vector.
  • Distributivity connects addition with scalar multiplication. Multiplication by one leaves a vector unchanged.
v+w=w+v
(u+v)+w=u+(v+w)
v+0=v
v+(−v)=0
α(v+w)=αv+αw
(α+β)v=αv+βv
α(βv)=(αβ)v
1v=v

Worked example: quarterly shop expenses

The slides list quarterly expenses in the order rent, goods purchases and salaries. Annual expenses require vector addition, because the three entries describe the same categories in each quarter. The quarterly values are 20.5, 24.3, 19.3; 19.8, 23.9, 19.5; 21.2, 22.4, 19.4; and 20.5, 28.5, 25.9.

q1+q2+q3+q4=[20.5+19.8+21.2+20.524.3+23.9+22.4+28.519.3+19.5+19.4+25.9]
qyear=[8299.184.1]

Worked example: combining election districts

Duckburg has an East district with 2,000 voters and a West district with 3,000. Parties A, B and C have East shares of 38%, 43% and 19%, and West shares of 36%, 22% and 42%. Percentages alone do not give voter counts: party A has a larger share in East, but more voters in West. Combine the districts by multiplying each share vector by its district population, adding the resulting count vectors, and dividing by the combined population.

vE=2000[0.380.430.19]=[760860380]
vW=3000[0.360.220.42]=[10806601260]
vtotal=vE+vW=[184015201640]
ptotal=vtotal2000+3000=[0.3680.3040.328]

Linear combinations and dependence

A linear combination multiplies given vectors by scalar coefficients and adds the products. A combination can encode a portfolio of contracts, a weighted aggregation of election results, or a production plan. The coefficients need not be positive and need not sum to one. Those additional restrictions belong to particular applications rather than to the definition.

v=∑h=1kαhvh

Vectors are linearly independent when no one of them is a linear combination of the others. They are linearly dependent otherwise. Two vectors are dependent exactly when one is a scalar multiple of the other. The slides also state that more than n vectors in Rn must be dependent. The computational test writes a linear combination equal to the zero vector and asks whether a nonzero choice of coefficients is possible.

αv+βw=0 ⟹ α=β=0
[1−805]=2[2−131]−3[122−1]

The slides contrast the dependent triple with (1,2,1,−1), (2,1,3,2) and (4,5,5,1). They label that triple linearly independent. It can be checked by forming a matrix of columnsGaussian elimination, linear (in)dependence and rankGaussian elimination simplifies a matrix through elementary row operations. It provides a common method for finding rank, testing linear independence, solving linear systems and computing a matrix inverse. Each operation changes the displayed rows… and reducing it: the first two component equations give α = −2γ and β = −γ, and the third then forces γ = 0.

α[121−1]+β[2132]+γ[4551]=0

Bases and coordinates

A basis of Rn generates every vector in that space by a unique linear combination. The coefficients of the unique combination are the vector’s coordinates relative to the chosen basis. A basis consists of n linearly independentLinear combinations and dependenceA linear combination multiplies given vectors by scalar coefficients and adds the products. A combination can encode a portfolio of contracts, a weighted aggregation of election results, or a production plan. The coefficients need not be positive… vectors. The standard basis uses a one in one coordinate and zeros everywhere else. Changing basis changes the coordinates, while the represented vector remains the same.

x=x1e1+x2e2+⋯+xnen
[34]=103[11]+13[−12]
  1. The new basis vectors are (1,1) and (−1,2). They are not proportional, so they span the plane.
  2. Equating the first components gives α − β = 3; equating the second gives α + 2β = 4.
  3. Subtracting the equations gives 3β = 1. Then β = 1/3 and α = 10/3.
  4. The new coordinate axes need not be perpendicular. Linear independence makes the representation unique.

Worked example: contracts and arbitrage

The textbook’s two-date contracts are u = (−100,120) and v = (−100,115). A negative first component is a payment today; a positive second component is a receipt tomorrow. The target portfolio pays 1,000 at each date. Solving for the activation levels gives 430 units of the first contract and −440 of the second. The example illustrates arbitrage arising from inconsistent interest rates; its mathematical point is that nonaligned vectors generate any target in the plane.

α[−100120]+β[−100115]=[10001000]
−100α−100β=1000
120α+115β=1000
430u−440v=[10001000]

Replacing the second contract with z = (50,−60) makes it proportional to u. The same target then leads to an inconsistent system: substituting the first equation into the second yields −1200 = 1000. The printed prose labels the proportionality as z = −2u, but the displayed vectors imply z = −u/2 and u = −2z. Both the source statement and the corrected calculation are recorded here.

z=−12u

Matrices and operations

A matrix is a rectangular arrangement of numbers with m rows and n columns. The entry aij is at row i and column j. Matrix dimensions determine which operations are possible. A matrix can also be read as a collection of column vectors or row vectors, making it the natural language for linear systemsLinear systems and applicationsA linear system consists of equations that are linear in the unknown components of a vector. It is written Ax = b, with A the coefficient matrix, x the unknown vector and b the vector of known terms. A solution is the entire vector that satisfies… and data tables.

Matrix types

A=(aij)m×n
  • A square matrix has the same number of rows and columns; its order is that number.
  • A diagonal matrix is square and has zero entries off the main diagonal.
  • The identity matrix has ones on the main diagonal and zeros elsewhere.
  • A zero matrix has all entries equal to zero.
  • An upper triangular matrix has zeros below the main diagonal; a lower triangular matrix has zeros above it.
I3=[100010001]

Transposition, addition and scalar multiplication

Transposition interchanges rows and columns. An m × n matrix becomes an n × m matrix. Transposing twice gives the original matrix. A symmetric matrix is square and equal to its transposeTransposition, addition and scalar multiplicationTransposition interchanges rows and columns. An m × n matrix becomes an n × m matrix. Transposing twice gives the original matrix. A symmetric matrix is square and equal to its transpose. Matrix addition is defined only when both matrices have the…. Matrix addition is defined only when both matrices have the same dimensions; it adds corresponding entries. Scalar multiplication multiplies every entry by the given number.

(AT)ij=aji
(AT)T=A
(A+B)ij=aij+bij
(αA)ij=αaij
(αA)T=αAT
(A+B)T=AT+BT

Matrix multiplication and conformability

The matrix product AB exists when the number of columns of A equals the number of rows of B. If A is m × n and B is n × p, then AB is m × p. Each entry is a row-by-column product: multiply the corresponding entries of a row of A and a column of B, then add them. The repeated dimension disappears from the output shape.

(AB)ij=∑k=1naikbkj

Matrix multiplication is associative and distributes over addition when the dimensions fit. It is generally not commutative: AB and BA can differ, and one product can exist while the other does not. For a square matrix, multiplying by the identity matrix leaves it unchanged. Transposing a product reverses the order of its factors.

(AB)C=A(BC)
A(B+C)=AB+AC
(A+B)C=AC+BC
(AB)T=BTAT
A2=AA

A product as a linear combination

If A has columns a1, …, an, multiplying A by a column vector b combines those columns with the components of b as coefficients. This connects matrix multiplication to linear combinationsLinear combinations and dependenceA linear combination multiplies given vectors by scalar coefficients and adds the products. A combination can encode a portfolio of contracts, a weighted aggregation of election results, or a production plan. The coefficients need not be positive…. Each column of AB can be calculated independently by multiplying A by the corresponding column of B.

Ab=b1a1+b2a2+⋯+bnan
A[b1⋯bp]=[Ab1⋯Abp]
[13−52−46][132231]=[−84124]

Worked example: temporary housing

Two architects propose mixes of small, medium and large houses. Four construction firms quote the price of each house type in thousands of euros. The proposal matrix uses rows for architects and columns for house types. The price matrix uses rows for house types and columns for firms. Their product gives the total cost of each architect–firm combination. Reading the labels explains why the dimensions must fit and what each product entry means.

H=[304010254510]
P=[202218213028273045465046]
HP=[22502240212022902300227021652335]

Gaussian elimination, linear (in)dependence and rank

Gaussian elimination simplifies a matrix through elementary row operationsElementary row operationsThe elimination method selects a nonzero entry, moves its row into position if necessary, scales it to a convenient pivot and clears other entries in the pivot column. Continue with the remaining rows and columns. Zero rows give no new independent…. It provides a common method for finding rank, testing linear independence, solving linear systemsLinear systems and applicationsA linear system consists of equations that are linear in the unknown components of a vector. It is written Ax = b, with A the coefficient matrix, x the unknown vector and b the vector of known terms. A solution is the entire vector that satisfies… and computing a matrix inverseInverse, singularity and adjoint notationThe inverse matrix A−1 satisfies multiplication by A on either side equal to the identity matrix. Such an inverse is unique. In this course’s square-matrix setting, invertibility is equivalent to independent rows and columns, full rank and a nonzero…. Each operation changes the displayed rows but preserves the information needed for those tasks. A pivot identifies an independent row or a determined unknown.

Elementary row operations

  1. Interchange two rows.
  2. Multiply a row by a nonzero scalar.
  3. Add a multiple of one row to another row.
Ri↔Rj
Ri←αRi(α≠0)
Ri←Ri+αRj

The elimination method selects a nonzero entry, moves its row into position if necessary, scales it to a convenient pivot and clears other entries in the pivot column. Continue with the remaining rows and columns. Zero rows give no new independent information. When a system is being solved, the same operations must be performed on the augmented matrix, including its right-hand side.

Rank and full rank

The rank of a matrix is the maximum number of linearly independent rows that can be extracted from it. The same number is obtained from the columns. The equality of row rank and column rank makes ρ(A) a single quantity. Rank cannot exceed either matrix dimension. A matrix has full rank when its rank equals the smaller dimension.

0≤ρ(A)≤min(m,n)
ρ(A)=min(m,n)for a full-rank matrix

For A = [(α,β);(kα,kβ)], the second row is k times the first. If both α and β are zero, its rankRank and full rankThe rank of a matrix is the maximum number of linearly independent rows that can be extracted from it. The same number is obtained from the columns. The equality of row rank and column rank makes ρ(A) a single quantity. Rank cannot exceed either… is zero; otherwise its rank is one. The textbook distinguishes a specified algebraic matrix from a generic empirical matrix: its discussion of typical full rank does not replace actual elimination or determinant calculationsSquare matrices, determinants and inverseA determinant assigns a real number to a square matrix. It diagnoses whether the matrix is singular, supports matrix inversion, and identifies independent rows and columns. Minors and cofactors extend the calculation to larger matrices. The same… for an exam matrix.

A=[αβkαkβ]

Worked example: rank by reduction

The textbook’s 4 × 3 example begins with a zero row. Move that row to the bottom, eliminate the first component of the remaining rows, and observe that two new rows coincide. Subtracting them leaves two independent nonzero rows. Thus the matrix has rank two, despite having three columns and four rows.

A=[000123456579]
A~[123456579000]
A~[1230−3−60−3−6000]
A~[10−1012000000]
ρ(A)=2

Rank by minors and independence

A nonzero minor of order k proves that the rankRank and full rankThe rank of a matrix is the maximum number of linearly independent rows that can be extracted from it. The same number is obtained from the columns. The equality of row rank and column rank makes ρ(A) a single quantity. Rank cannot exceed either… is at least k. The textbook’s Kronecker method starts from a nonzero minor and tries to extend it by one row and one column. When every possible bordering of the selected nonsingular submatrix gives zero, the rank has been found. In the example, a nonzero minor of order two exists, while all possible extensions to order three vanish.

det[1245]=−3≠0
det[123456579]=0

To test linear independence of given vectors, place them as columns of a matrix and solve the homogeneous system whose unknowns are the combination coefficients. If the only solution is the zero coefficient vector, the columns are independent. In the rankRank and full rankThe rank of a matrix is the maximum number of linearly independent rows that can be extracted from it. The same number is obtained from the columns. The equality of row rank and column rank makes ρ(A) a single quantity. Rank cannot exceed either… example, the first two columns form an independent pair. The third is a linear combinationLinear combinations and dependenceA linear combination multiplies given vectors by scalar coefficients and adds the products. A combination can encode a portfolio of contracts, a weighted aggregation of election results, or a production plan. The coefficients need not be positive… of them, so adding it cannot increase the rank.

a3=−a1+2a2

Square matrices, determinants and inverse

A determinant assigns a real number to a square matrix. It diagnoses whether the matrix is singular, supports matrix inversion, and identifies independent rows and columns. Minors and cofactors extend the calculation to larger matrices. The same determinant machinery also appears in Leontief viabilityLeontief balance and Hawkins–Simon viabilityIn a Leontief economy, each sector produces one good and uses the goods of other sectors as inputs. The entry ars is the amount from sector r required per unit produced in sector s. Production must cover both intermediate uses and final uses. The… and Hessian testsHessian matrix and optimizationThe Hessian matrix collects all second partial derivatives. Its diagonal contains pure second derivatives, taken twice with respect to the same input; the off-diagonal entries contain mixed second derivatives. For a smooth function satisfying the….

Determinants of orders one, two and three

det[a]=a
det[abcd]=ad−bc
det[1234]=1·4−2·3=−2

For a 2 × 2 matrix, the absolute value of the determinantDeterminants of orders one, two and threeFor a 2 × 2 matrix, the absolute value of the determinant is the area of the parallelogram determined by its columns. A zero determinant corresponds to aligned columns and zero area. For a 3 × 3 matrix, the analogous magnitude is a parallelepiped… is the area of the parallelogram determined by its columns. A zero determinant corresponds to aligned columns and zero area. For a 3 × 3 matrix, the analogous magnitude is a parallelepiped volume. The sign depends on orientation; it is the magnitude that measures area or volume.

detA=a11a22a33+a12a23a31+a13a21a32−a13a22a31−a11a23a32−a12a21a33
det[123456789]=45+84+96−105−48−72=0

Minors, cofactors and Laplace expansion

A submatrix is obtained by deleting rows and columns. A minor is the determinant of a square submatrix. The principal north-west submatrix of order k retains the first k rows and first k columns. Its determinant is a principal north-west minor. These particular minors recur in the source’s Hawkins–Simon conditionLeontief balance and Hawkins–Simon viabilityIn a Leontief economy, each sector produces one good and uses the goods of other sectors as inputs. The entry ars is the amount from sector r required per unit produced in sector s. Production must cover both intermediate uses and final uses. The… and second-order conditionsHessian matrix and optimizationThe Hessian matrix collects all second partial derivatives. Its diagonal contains pure second derivatives, taken twice with respect to the same input; the off-diagonal entries contain mixed second derivatives. For a smooth function satisfying the….

Mk=detAk

The complementary minor Mij deletes row i and column j. Its cofactor multiplies that minor by an alternating sign. The sign is positive when i + j is even and negative when it is odd. Laplace expansion computes the full determinantDeterminants of orders one, two and threeFor a 2 × 2 matrix, the absolute value of the determinant is the area of the parallelogram determined by its columns. A zero determinant corresponds to aligned columns and zero area. For a 3 × 3 matrix, the analogous magnitude is a parallelepiped… using any row or any column. Choosing a line with many zeros reduces the amount of arithmetic.

Cij=(−1)i+jMij
detA=∑j=1naijCij
detA=∑i=1naijCij
det[10−3−11−4124]=0·0+1·7+2·7=21

Inverse, singularity and adjoint notation

The inverse matrix A−1 satisfies multiplication by A on either side equal to the identity matrix. Such an inverseInverse, singularity and adjoint notationThe inverse matrix A−1 satisfies multiplication by A on either side equal to the identity matrix. Such an inverse is unique. In this course’s square-matrix setting, invertibility is equivalent to independent rows and columns, full rank and a nonzero… is unique. In this course’s square-matrix setting, invertibility is equivalent to independent rows and columns, full rank and a nonzero determinantDeterminants of orders one, two and threeFor a 2 × 2 matrix, the absolute value of the determinant is the area of the parallelogram determined by its columns. A zero determinant corresponds to aligned columns and zero area. For a 3 × 3 matrix, the analogous magnitude is a parallelepiped…. A matrix lacking an inverse is singular. A square shape by itself is insufficient: the all-ones 2 × 2 matrix is singular.

AA−1=A−1A=I
detA≠0 ⟺ ρ(A)=n ⟺ A−1 exists

The textbook calls the matrix of cofactors A* the adjoint, and transposes that matrix when constructing the inverseInverse, singularity and adjoint notationThe inverse matrix A−1 satisfies multiplication by A on either side equal to the identity matrix. Such an inverse is unique. In this course’s square-matrix setting, invertibility is equivalent to independent rows and columns, full rank and a nonzero…. Retain this source convention: its A* is the cofactorMinors, cofactors and Laplace expansionA submatrix is obtained by deleting rows and columns. A minor is the determinant of a square submatrix. The principal north-west submatrix of order k retains the first k rows and first k columns. Its determinant is a principal north-west minor.… matrix, and (A*)T is the matrix used in the inverse formula. The computation first finds the determinantDeterminants of orders one, two and threeFor a 2 × 2 matrix, the absolute value of the determinant is the area of the parallelogram determined by its columns. A zero determinant corresponds to aligned columns and zero area. For a 3 × 3 matrix, the analogous magnitude is a parallelepiped…, then all cofactors, then their transposeTransposition, addition and scalar multiplicationTransposition interchanges rows and columns. An m × n matrix becomes an n × m matrix. Transposing twice gives the original matrix. A symmetric matrix is square and equal to its transpose. Matrix addition is defined only when both matrices have the…, and finally divides every entry by the determinant.

A−1=1detA(A*)T
[abcd]−1=1ad−bc[d−b−ca]

Worked example: the inverse of a two-by-two matrix

A=[1234]
A*=[4−3−21]
(A*)T=[4−2−31]
A−1=[−213/2−1/2]

Elimination gives an alternative inverse algorithm. Append the identity matrix to A, and apply each row operation to both halves of the augmented matrixExistence, uniqueness and degrees of freedomA reduced row of the form 0 = β with β ≠ 0 makes the system impossible. A row 0 = 0 adds no restriction. For a possible underdetermined system, choose the free unknowns and express the pivot unknowns in terms of them. The textbook prints β2 in its…. If the left half can be turned into identity, the right half becomes the inverse. If A is singular, that identity cannot be completed. This method uses the same row operationsElementary row operationsThe elimination method selects a nonzero entry, moves its row into position if necessary, scales it to a convenient pivot and clears other entries in the pivot column. Continue with the remaining rows and columns. Zero rows give no new independent… as Gaussian eliminationGaussian elimination, linear (in)dependence and rankGaussian elimination simplifies a matrix through elementary row operations. It provides a common method for finding rank, testing linear independence, solving linear systems and computing a matrix inverse. Each operation changes the displayed rows….

[A∣I]~[I∣A−1]

Linear systems and applications

A linear system consists of equations that are linear in the unknown components of a vector. It is written Ax = b, with A the coefficient matrix, x the unknown vector and b the vector of known terms. A solution is the entire vector that satisfies every equation simultaneously. The number of equations and unknowns alone does not determine whether a solution exists or is unique.

Existence, uniqueness and degrees of freedom

  • A system is possible if it has at least one solution. It is impossible if it has no solution.
  • A possible system is determined when its solution is unique.
  • A possible system is underdetermined when it has infinitely many solutions.
  • The augmented matrix contains the coefficient matrix and the known-term column. Comparing its rank with the coefficient rank diagnoses consistency.
Ax=b is possible ⟺ ρ(A)=ρ([A∣b])
ρ(A)=ρ([A∣b])=n
ρ(A)=ρ([A∣b])<n
degrees of freedom=n−ρ(A)

A reduced row of the form 0 = β with β ≠ 0 makes the system impossible. A row 0 = 0 adds no restriction. For a possible underdetermined system, choose the free unknowns and express the pivot unknowns in terms of them. The textbook prints β2 in its page-78 formula for the pivot block, although the preceding reduced matrix places the pivot right-hand side in β1. The corrected block calculation is shown here.

xpivot=β1−Cxfree

Worked example: solving by elimination

The source solves x1 + 2x2 = 10 and 3x1 + 4x2 = 10. Its coefficient matrix has determinantDeterminants of orders one, two and threeFor a 2 × 2 matrix, the absolute value of the determinant is the area of the parallelogram determined by its columns. A zero determinant corresponds to aligned columns and zero area. For a 3 × 3 matrix, the analogous magnitude is a parallelepiped… −2, so the system has a unique solution. Gaussian elimination and multiplication by the inverse matrix must produce the same vector. The following reduction performs the row operationsElementary row operationsThe elimination method selects a nonzero entry, moves its row into position if necessary, scales it to a convenient pivot and clears other entries in the pivot column. Continue with the remaining rows and columns. Zero rows give no new independent… on the known-term column as well.

[A∣b]=[12103410]
[A∣b]~[12100−2/3−20/3]
[A∣b]~[12100110]
[A∣b]~[10−100110]
x*=A−1b=[−1010]

Leontief balance and Hawkins–Simon viability

In a Leontief economy, each sector produces one good and uses the goods of other sectors as inputs. The entry ars is the amount from sector r required per unit produced in sector s. Production must cover both intermediate uses and final uses. The production vector x, technical-coefficient matrixLinear systems and applicationsA linear system consists of equations that are linear in the unknown components of a vector. It is written Ax = b, with A the coefficient matrix, x the unknown vector and b the vector of known terms. A solution is the entire vector that satisfies… A and final-use vector c therefore satisfy a balance equation. Matrix inversion converts the desired final uses into required gross production when I − A is invertible.

x=Ax+c
(I−A)x=c
x=(I−A)−1c

Invertibility and viability answer different questions. An inverseInverse, singularity and adjoint notationThe inverse matrix A−1 satisfies multiplication by A on either side equal to the identity matrix. Such an inverse is unique. In this course’s square-matrix setting, invertibility is equivalent to independent rows and columns, full rank and a nonzero… means that the balance equations have a unique mathematical solution. Viability requires economically admissible production for the final-use choices in the model. The textbook’s Hawkins–Simon condition states that a Leontief system is viable exactly when every principal north-west minor of I − A is positive. A negative or zero relevant minor prevents the stated condition from holding.

Mk(I−A)>0(k=1,…,n)

The textbook compares A = [(0.2,2);(0.2,0.4)] with A = [(0.2,3);(0.2,0.3)]. The first has M1 = 0.8 and a positive determinantDeterminants of orders one, two and threeFor a 2 × 2 matrix, the absolute value of the determinant is the area of the parallelogram determined by its columns. A zero determinant corresponds to aligned columns and zero area. For a 3 × 3 matrix, the analogous magnitude is a parallelepiped…, while the second has determinant −0.04. On PDF page 62 the first determinant is printed as 0.88. Direct multiplication of the displayed entries gives 0.08. Its sign remains positive, so the viabilityLeontief balance and Hawkins–Simon viabilityIn a Leontief economy, each sector produces one good and uses the goods of other sectors as inputs. The entry ars is the amount from sector r required per unit produced in sector s. Production must cover both intermediate uses and final uses. The… conclusion survives the arithmetic discrepancy.

det[0.8−2−0.20.6]=0.48−0.4=0.08
det[0.8−3−0.20.7]=0.56−0.6=−0.04

Policy instruments, objectives and solvability

The source applies linear systems to economic policy: an instrument vector x affects an objective vector y through y = Ax. Reaching a target means solving the corresponding system. Rank measures how many independent objective directions the instruments can generate. If there are more independently specified objectives than effective instrument directions, an arbitrary target cannot generally be attained. A target that lies in the generated column space can still be attainable.

y=Ax

The textbook also studies the family B(λ) = λI − A. Values of λ that make its determinantDeterminants of orders one, two and threeFor a 2 × 2 matrix, the absolute value of the determinant is the area of the parallelogram determined by its columns. A zero determinant corresponds to aligned columns and zero area. For a 3 × 3 matrix, the analogous magnitude is a parallelepiped… zero make the family member singular. A nonzero vector satisfying Ax = λx is an eigenvector and the corresponding number is an eigenvalue. This is supplementary square-matrix material in the supplied chapter; its exam relevance remains subject to the class cutoff.

det(λI−A)=0
Ax=λx(x≠0)

Linear algebra exercises

The supplied past exams combine matrix operations, determinants, inversion and economic interpretation. Each question should be split into a mathematical condition and a statement about the model. A singular matrix can invalidate a claimed inverse even before any multiplication is attempted. A nonzero determinant alone does not prove a Leontief economy viable.

October 2015: a parameter-dependent Leontief system

The first partial of October 2015 uses A = [(0.1,α);(5,0.1)] and final uses c = (500,500). The question asks whether I − A is invertible for every positive α, and whether the economy is always viable. Compute the determinantDeterminants of orders one, two and threeFor a 2 × 2 matrix, the absolute value of the determinant is the area of the parallelogram determined by its columns. A zero determinant corresponds to aligned columns and zero area. For a 3 × 3 matrix, the analogous magnitude is a parallelepiped… once, then apply it separately to invertibility and the Hawkins–Simon conditionLeontief balance and Hawkins–Simon viabilityIn a Leontief economy, each sector produces one good and uses the goods of other sectors as inputs. The entry ars is the amount from sector r required per unit produced in sector s. Production must cover both intermediate uses and final uses. The….

I−A=[0.9−α−50.9]
det(I−A)=0.81−5α
α*=0.815=0.162
0<α<0.162
(I−A)−1=10.81−5α[0.9α50.9]
x=10.81−5α[450+500α2950]

The source prints 2680 as the second production numerator. The displayed matrix and final-use vector instead give 5 × 500 + 0.9 × 500 = 2950. This discrepancy is preserved explicitly. The corrected vector can be checked by multiplying it by I − A and recovering c. The claims that an inverseInverse, singularity and adjoint notationThe inverse matrix A−1 satisfies multiplication by A on either side equal to the identity matrix. Such an inverse is unique. In this course’s square-matrix setting, invertibility is equivalent to independent rows and columns, full rank and a nonzero… exists for every positive α and that every such system is viable are both false.

October 2016: transition-matrix multiplication

The October 2016 partial uses three voter groups: Blue, Red and Uncertain. Its transition matrix uses columns for starting groups and rows for destination groups. Each column sums to one, so multiplying a population vector preserves the total count. Only the finite matrix calculations in this historical question are included here; the later dynamic systems chapter is outside the provisional scope.

A=[0.50.10.40.10.60.40.40.30.2]
x0=[11001000900]
x1=Ax0=[10101070920]
A2=[0.420.230.320.270.490.360.310.280.32]
x2=A2x0=[9801111909]

The matrix I − A is singular in this question, so its inverseInverse, singularity and adjoint notationThe inverse matrix A−1 satisfies multiplication by A on either side equal to the identity matrix. Such an inverse is unique. In this course’s square-matrix setting, invertibility is equivalent to independent rows and columns, full rank and a nonzero… does not exist. The source gives a zero determinant as the reason. This task illustrates that matrix powers and inversion test different properties. Every population vector here totals 3,000, providing a quick consistency check on the arithmetic.

det(I−A)=0

October 2016: invertibility versus viability

The second question uses A(α) = [(0.6,α);(α,0.6)] and c = (150,50). It asks separately when A is singular, when I − A has an inverseInverse, singularity and adjoint notationThe inverse matrix A−1 satisfies multiplication by A on either side equal to the identity matrix. Such an inverse is unique. In this course’s square-matrix setting, invertibility is equivalent to independent rows and columns, full rank and a nonzero…, and when the Leontief system is viable. For positive α, the thresholds are different: 0.6 for singularity of A and 0.4 for singularity of I − A.

detA=0.36−α2
det(I−A)=0.16−α2
(I−A)−1=10.16−α2[0.4αα0.4]
x=10.16−α2[60+50α20+150α]

The source prints negative off-diagonal entries in its intermediate inverseInverse, singularity and adjoint notationThe inverse matrix A−1 satisfies multiplication by A on either side equal to the identity matrix. Such an inverse is unique. In this course’s square-matrix setting, invertibility is equivalent to independent rows and columns, full rank and a nonzero… display, then uses positive contributions in the final production vector. The inverse formula requires the positive signs shown above. For positive α, invertibility requires α ≠ 0.4, whereas viabilityLeontief balance and Hawkins–Simon viabilityIn a Leontief economy, each sector produces one good and uses the goods of other sectors as inputs. The entry ars is the amount from sector r required per unit produced in sector s. Production must cover both intermediate uses and final uses. The… requires 0 < α < 0.4. Thus a matrix can remain invertible beyond the viable range.

Functions of one real variable

A function assigns one output to each admissible input. For a function of one real variable, both the input and output are numbers. The domain is the set of admissible inputs; the graph collects the points (x,f(x)). In social-science applications, the meaningful domain can be smaller than the mathematical one: a population, production quantity or vote share may have its own restrictions.

Intervals and domain

  • The closed interval [a,b] includes both endpoints.
  • The open interval (a,b) excludes both endpoints.
  • The half-open intervals [a,b) and (a,b] include one endpoint.
  • Unbounded intervals extend toward positive or negative infinity; infinity is not an included endpoint.
[a,b]={x∈ℝ:a≤x≤b}
(−∞,+∞)=ℝ

Determining the domain comes before taking derivativesDerivativesThe derivative measures the local slope of a function. The source introduces it through a local affine approximation: close to a point, a differentiable function behaves like its tangent line. Its increment consists of a term proportional to the…. Exclude zeros of a denominator, negative arguments of even roots and nonpositive arguments of logarithms. A restriction from one part of a formula cannot be ignored because another part is well defined. After finding the mathematical domain, apply the restrictions of the model, such as x ≥ 0 or a vote share between zero and one.

Affine and quadratic functions

An affine function has a straight-line graph. Its slope gives the change in the output for one unit of input, and its intercept is the output at zero. A quadratic function has a parabola as its graph. The sign of its leading coefficient determines whether it opens upward or downward. Its discriminant determines the number of real zeros.

f(x)=mx+q
f(x)=3x−5
f(x)=ax2+bx+c(a≠0)
Δ=b2−4ac
x1,2=−b±Δ2a
V=(−b2a,−Δ4a)
  • If a > 0, the parabola is convex and the vertex is a minimum. If a < 0, it is concave and the vertex is a maximum.
  • If Δ > 0, the graph crosses the horizontal axis at two distinct points.
  • If Δ = 0, it touches the axis at one repeated root.
  • If Δ < 0, it has no real intersection with the axis.

The source’s example f(x) = x² − 4x − 5 has vertex (2,−9) and discriminant 36. Its printed root calculation uses −4 in place of −b = 4 and reports −5 and 1, while the following sentence reports axis crossings −1 and 5. Direct substitution confirms −1 and 5. The disagreement is explicit here; the correct factorization is shown below.

x2−4x−5=(x−5)(x+1)

Power and exponential functions

A power function has a variable base and a fixed exponent. An exponential function has a fixed positive base and a variable exponent. These different structures produce different rates of growth. The source introduces xα for nonnegative inputs and positive exponents: exponents above one give growth that becomes steeper, while exponents between zero and one give growth that becomes flatter. The later derivative rule extends powers to their appropriate real domains.

f(x)=xα
f(x)=ax(a>0)
  • If a > 1, the exponential function increases as x increases.
  • If 0 < a < 1, it decreases.
  • If a = 1, it is constant at one.
  • For every positive base, a0 = 1 and the exponential output is positive.

The textbook contrasts a square room’s area, p(x) = x², with a population example r(x) = 2.5x. Increasing a room side from 3 to 4 changes the area from 9 to 16; adding a unit to the population model’s input multiplies its value by 2.5. The number e is the exponentialPower and exponential functionsA power function has a variable base and a fixed exponent. An exponential function has a fixed positive base and a variable exponent. These different structures produce different rates of growth. The source introduces xα for nonnegative inputs and… base whose curve has slope one at the origin. Natural exponentials are written ex or exp(x).

p(3)=32=9
p(4)=42=16
r(1)=2.5
r(2)=2.52=6.25
e≈2.71828

Logarithms and their identities

A logarithm inverts an exponential function. The logarithm of x to base a is the exponent needed to obtain x from a. Its argument must be positive, and its base must be positive and different from one. Logarithm values can be negative: log10(0.1) = −1. Natural logarithms use base e and are written ln(x).

y=logax ⟺ ay=x
ln(pq)=lnp+lnq
ln(p/q)=lnp−lnq
ln(pq)=qlnp
ln(ex)=x
elnx=x(x>0)
ax=exlna

The source rewrites a population growing by 10% a year from P(t) = P(0)1.1t to a natural-exponentialPower and exponential functionsA power function has a variable base and a fixed exponent. An exponential function has a fixed positive base and a variable exponent. These different structures produce different rates of growth. The source introduces xα for nonnegative inputs and… representation. The conversion rate is δ = ln(1.1) ≈ 0.09531. This is an algebraic change of representation: 1.1t and eδt describe the same values. The later dynamical-system analysis of such expressions is outside these notes.

P(t)=P(0)1.1t=P(0)etln1.1

Continuity and thresholds

A continuous function has nearby outputs for inputs sufficiently close to a given point. The source’s cost model shows why this can fail: a fixed-cost increase occurs when production passes capacity. Below or at 1,000 units, total cost is 5,000 plus three times production. Above that threshold it is 7,000 plus three times production. The jump is caused by the change in fixed costs, rather than by a new variable-cost slope.

f(x)={5000+3x0≤x≤10007000+3xx>1000
f(1000)=8000
f(1001)=10003

Small-o notation and comparisons

The textbook develops differential calculus using small-o notation instead of a full theory of limits. A quantity is o(h) when its size divided by h becomes negligible in the specified limiting situation. The limiting direction matters: a quadratic term is negligible relative to a linear one near zero, but dominates it at infinity. The symbol describes relative size; it does not denote one fixed formula.

r(h)=o(h) ⟺ |r(h)||h|⟶0
4h2h=4h⟶0(h→0)
4h2+h3=o(h)(h→0)
1000hh2/1000=1000000h⟶0(h→+∞)

A finite linear combination of negligible remainders stays negligible. A product of two o(h) quantities is o(h²). Their quotient has no fixed behavior, because the relative rates may differ. Continuity at x can be expressed by saying the increment is o(1) as h approaches zero. Differentiability will strengthen that statement by isolating a linear term and a smaller remainder.

αo(h)+βo(h)=o(h)
o(h)o(h)=o(h2)
h3h2=h⟶0(h→0)
h2h3=1h
f(x+h)−f(x)=o(1)(h→0)
  • Near zero, a larger positive power vanishes faster than a smaller positive power.
  • At positive infinity, a larger power grows faster than a smaller power.
  • Among exponential bases above one, a larger base grows faster.
  • At positive infinity, every positive power grows more slowly than an exponential with base above one, and every positive power of the logarithm grows more slowly than a positive power of the argument.
hβ=o(hα)(β>α>0, h→0)
hβ=o(hα)(α>β>0, h→+∞)
hα=o(ah)(α>0, a>1, h→+∞)
(lnh)α=o(hβ)(α,β>0, h→+∞)

Derivatives

The derivative measures the local slope of a function. The source introduces it through a local affine approximation: close to a point, a differentiable function behaves like its tangent line. Its increment consists of a term proportional to the input change and a smaller remainder. The coefficient of the proportional term is the derivative, and the term itself is the differentialDifferentials and approximationThe differential is the linear part of a function’s increment. It turns a small input change into an approximate output change. The derivative is a slope; the differential multiplies that slope by the actual input change. The neglected remainder is….

Increment, differentiability and tangent line

Δf=f(x+h)−f(x)
f(x+h)−f(x)=f′(x)h+o(h)
f′(x)=limh→0f(x+h)−f(x)h
y=f(x0)+f′(x0)(x−x0)

For one-variable functions, existence of the difference-quotient derivative is equivalent to the source’s increment definition of differentiability. Differentiability implies continuity, but continuity alone does not guarantee differentiability. For several variables, the existence of individual partial derivativesFunctions of two variables, partial derivatives and gradientA function of two real variables assigns one number to an admissible pair of inputs. Its graph is a surface in three-dimensional space, and a level curve collects input pairs giving the same output. In social-science models, multiple independent… is a weaker statement, so the distinction becomes important again.

Worked example: extracting a derivative from an increment

(x+h)2−x2=2xh+h2
f′(x)=2xfor f(x)=x2
(x+h)3−x3=3x2h+3xh2+h3
Δ(2x2+x)=(4x+1)h+2h2
D(2x2+x)=4x+1

The absolute-value function f(x) = |x| is continuous at zero, but its increment equals h for positive h and −h for negative h. No single slope approximates both sides. The right slope is one and the left slope is minus one. This example separates continuity from differentiability using the actual local behavior, rather than the appearance of the graphThe source’s graph-sketching procedure alone.

|h|−|0|h={1h>0−1h<0

Elementary derivatives

The source’s differentiation table covers constants, affine functions, powers, exponentials and the natural logarithm. Its power rule applies on the appropriate domain, including negative and fractional exponents. For ln(x), the domain is x > 0. The natural exponential is its own derivative; a general positive-base exponential acquires a factor equal to the natural logarithm of its base.

Dk=0
D(mx+q)=m
D(xα)=αxα−1
D(ax)=axlna
D(ex)=ex
D(lnx)=1x
D(x)=12x(x>0)
D(x−37)=−37x−38

Differentiation rules

Linearity of differentiation means that fixed coefficients remain attached to their functions’ derivatives. The sum rule, difference rule and constant-multiple rule are special cases. This applies to a linear combination of functions, rather than to arbitrary products or compositions. Those structures require their own rules.

D(αf)=αf′
D(f+g)=f′+g′
D(f−g)=f′−g′
D(αf+βg)=αf′+βg′

The product rule differentiates one factor at a time and adds the two terms. The quotient rule places the square of the denominator below a difference of cross-products. The denominator must be nonzero. In the chain rule, differentiate the outer function at the inner function’s value, then multiply by the derivative of the inner function. The source’s “onion rule” describes this outer-to-inner order.

D(fg)=f′g+fg′
D(fg)=f′g−fg′g2(g≠0)
D[f(g(x))]=f′(g(x))g′(x)

Worked examples of the rules

D(5x2)=10x
D(x2+x3)=2x+3x2
D(x2−x3)=2x−3x2
D(5x2−x3)=10x−3x2
D(x3ex)=3x2ex+x3ex
D(x3ex)=x2(3+x)ex
D(x3ex)=3x2ex−x3exe2x
D(x3e−x)=x2(3−x)e−x
D(ex2)=2xex2

In its explanation of linear combinations, PDF page 139 prints 10x − 3x³ for the derivative of 5x² − x³. The correct exponent in the second term is two, as the source states elsewhere on PDF page 133. The formula above records the checked power-rule calculation and retains this note about the disagreement.

Second derivatives, monotonicity and concavity

A second derivative is the derivative of the first derivative. It describes how the local slope changes. The first derivative diagnoses monotonicity and stationary points; the second diagnoses convexity, concavity and some local extrema. A stationary point is a candidate for an extremum rather than proof of one. The sign on intervals carries more information than a zero at an isolated point.

Computing second derivatives

f″(x)=D[f′(x)]
D2(ex2)=2ex2+4x2ex2
D2(ex2)=2ex2(1+2x2)

Derivative sign and monotonicity

On an interval where f′ is positive, the function is increasing: larger inputs have larger outputs. Where f′ is negative, the function is decreasing. Factoring a derivative is useful because it turns a sign problem for a sum into signs of factors. ExponentialPower and exponential functionsA power function has a variable base and a fixed exponent. An exponential function has a fixed positive base and a variable exponent. These different structures produce different rates of growth. The source introduces xα for nonnegative inputs and… factors are positive, so they often do not affect the sign diagnosis.

f′(x)>0 ⟹ f is increasing on the interval
f′(x)<0 ⟹ f is decreasing on the interval

Stationary points and local extrema

A local maximum is a point whose value is at least as large as values sufficiently near it. A local minimum reverses that comparison. At an interior extremum where the function is differentiable, its derivative must be zero. This is a necessary condition, not a sufficient one. A function can keep increasing through a horizontal tangent. Points where the derivative does not exist and boundary points require their own checks.

f′(x*)=0
f(x*)≥f(x*+h)for sufficiently small |h|
  • A sign change of f′ from positive to negative gives a local maximum.
  • A sign change of f′ from negative to positive gives a local minimum.
  • If the derivative does not change sign at its zero, the zero need not be an extremum.
  • A local statement compares nearby values. A global extremum compares all values in the relevant domain.

For f(x) = x²e−x, the derivative factors as x(2 − x)e−x. It is negative for x < 0, positive for 0 < x < 2 and negative for x > 2. Thus zero is a local minimumStationary points and local extremaA local maximum is a point whose value is at least as large as values sufficiently near it. A local minimum reverses that comparison. At an interior extremum where the function is differentiable, its derivative must be zero. This is a necessary… and two is a local maximumStationary points and local extremaA local maximum is a point whose value is at least as large as values sufficiently near it. A local minimum reverses that comparison. At an interior extremum where the function is differentiable, its derivative must be zero. This is a necessary…. The exponentialPower and exponential functionsA power function has a variable base and a fixed exponent. An exponential function has a fixed positive base and a variable exponent. These different structures produce different rates of growth. The source introduces xα for nonnegative inputs and… factor is always positive, so the two linear factors determine the changes of sign.

f′(x)=x(2−x)e−x

Concavity, convexity and inflection

A concave function lies above the straight chords joining points on its graphThe source’s graph-sketching procedure. A convex function lies below them. A negative second derivativeConcavity, convexity and inflectionA concave function lies above the straight chords joining points on its graph. A convex function lies below them. A negative second derivative makes the slope decrease and gives concavity on the interval; a positive second derivative makes the slope… makes the slope decrease and gives concavity on the interval; a positive second derivative makes the slope increase and gives convexity. Increasing or decreasing and concave or convex describe different aspects of behavior, so either slope sign can occur with either curvature sign.

f″(x)>0 ⟹ f is convex on the interval
f″(x)<0 ⟹ f is concave on the interval

An inflection point is a point where concavity changes. Solving f″ = 0 finds possible candidates, but a zero without a sign change does not establish an inflection. At a stationary pointStationary points and local extremaA local maximum is a point whose value is at least as large as values sufficiently near it. A local minimum reverses that comparison. At an interior extremum where the function is differentiable, its derivative must be zero. This is a necessary…, a strictly positive second derivativeConcavity, convexity and inflectionA concave function lies above the straight chords joining points on its graph. A convex function lies below them. A negative second derivative makes the slope decrease and gives concavity on the interval; a positive second derivative makes the slope… gives a local minimumStationary points and local extremaA local maximum is a point whose value is at least as large as values sufficiently near it. A local minimum reverses that comparison. At an interior extremum where the function is differentiable, its derivative must be zero. This is a necessary…, and a strictly negative one gives a local maximumStationary points and local extremaA local maximum is a point whose value is at least as large as values sufficiently near it. A local minimum reverses that comparison. At an interior extremum where the function is differentiable, its derivative must be zero. This is a necessary…. If the second derivative is zero, these particular tests do not decide the point’s nature.

f′(x*)=0, f″(x*)>0 ⟹ x* is a local minimum
f′(x*)=0, f″(x*)<0 ⟹ x* is a local maximum

Worked examples of curvature

D(xex)=ex(x+1)
D2(xex)=ex(x+2)

For f(x) = xex, the second derivative changes from negative to positive at x = −2, so that point is an inflection point. For g(x) = ln(x + 1), the domainIntervals and domainDetermining the domain comes before taking derivatives. Exclude zeros of a denominator, negative arguments of even roots and nonpositive arguments of logarithms. A restriction from one part of a formula cannot be ignored because another part is well… is x > −1, its first derivativeDerivative sign and monotonicityOn an interval where f′ is positive, the function is increasing: larger inputs have larger outputs. Where f′ is negative, the function is decreasing. Factoring a derivative is useful because it turns a sign problem for a sum into signs of factors.… is positive and its second derivative is negative. It increases and remains concave throughout that domain, with no inflection point.

Dln(x+1)=1x+1
D2ln(x+1)=−1(x+1)2

Sketching a function and applications

Graph sketching combines the domain, endpoint behavior, zeros, signs, monotonicity and curvature of a function. Each calculation answers a different question about the graph. An axis crossing comes from f = 0, a stationary pointStationary points and local extremaA local maximum is a point whose value is at least as large as values sufficiently near it. A local minimum reverses that comparison. At an interior extremum where the function is differentiable, its derivative must be zero. This is a necessary… from f′ = 0, and an inflection candidate from f″ = 0. Confusing these three equations produces a graph with the wrong locations and shape.

The source’s graph-sketching procedure

  1. Find the domain and restrict it to the model’s meaningful inputs.
  2. Determine behavior at the domain’s boundaries, using the source’s elementary comparisons or numerical inspection.
  3. Find the vertical-axis intersection when zero belongs to the domain, and solve f(x) = 0 for horizontal-axis intersections.
  4. Determine where f(x) is positive and negative.
  5. Factor f′, find its zeros and signs, and classify stationary points.
  6. Compute f″, determine its sign and verify whether each candidate is an inflection point.
f(x)=0

Worked example: a fifth-degree polynomial

The source studies f(x) = x⁵ − 3x⁴ + 2x³. Its domainIntervals and domainDetermining the domain comes before taking derivatives. Exclude zeros of a denominator, negative arguments of even roots and nonpositive arguments of logarithms. A restriction from one part of a formula cannot be ignored because another part is well… is the whole real line, and the fifth-degree term determines the behavior at both ends. Factoring the function itself gives zeros at 0, 1 and 2. PDF page 153 instead prints the two nonzero stationary-point coordinates as the axis crossings. The factorization and direct substitution establish the correct zeros; the derivative calculation uses the different coordinates below.

f(x)=x5−3x4+2x3=x3(x−1)(x−2)
f(x)→−∞(x→−∞)
f(x)→+∞(x→+∞)
  • The function is negative for x < 0.
  • It is positive for 0 < x < 1.
  • It is negative for 1 < x < 2.
  • It is positive for x > 2.
  • It is zero at x = 0, x = 1 and x = 2.
f′(x)=5x4−12x3+6x2=x2(5x2−12x+6)
xmax=6−65≈0.7101
xmin=6+65≈1.6899

The factor x² is nonnegative on both sides of zero. Therefore the derivative remains positive through the stationary pointStationary points and local extremaA local maximum is a point whose value is at least as large as values sufficiently near it. A local minimum reverses that comparison. At an interior extremum where the function is differentiable, its derivative must be zero. This is a necessary… zero: it is not an extremum. The quadratic factor makes the derivative positive outside the two nonzero roots and negative between them. The smaller nonzero root is a local maximum, and the larger is a local minimum. The function has neither a global maximum nor a global minimum over the whole real line because its values are unbounded in both directions.

f″(x)=20x3−36x2+12x=4x(5x2−9x+3)
xinflection,1=0
xinflection,2=9−2110≈0.4417
xinflection,3=9+2110≈1.3583
  • For x < 0, the second derivative is negative: the function is concave.
  • Between zero and the smaller positive inflection coordinate, it is positive: the function is convex.
  • Between the two positive inflection coordinates, it is negative: the function is concave.
  • After the larger positive inflection coordinate, it is positive: the function is convex.

The supplied Table 2.8 places a positive sign in its product column for the middle positive interval, while showing the concave shape there. The factored second derivative is negative on that interval. The corrected sign list above preserves the curvature indicated by the source and records the printed sign error.

Demand elasticity and expenditure

The source’s demand function x(p) decreases as price rises. Revenue or total expenditure is price times quantity demanded. Its derivative combines the direct effect of a higher price with the indirect effect of lower demand. The source defines elasticity as a nonnegative magnitude, using the absolute value of the demand derivative. This convention should be kept distinct from a signed elasticity used in some other course material.

I(p)=px(p)
I′(p)=x(p)+px′(p)
ε(p)=p|x′(p)|x(p)
ε(p)=p|ddplnx(p)|
  • If ε < 1, demand is inelastic and increasing price increases expenditure locally.
  • If ε > 1, demand is elastic and increasing price decreases expenditure locally.
  • If ε = 1, the two first-order effects cancel; the source calls this case “anelastic”.

For constant-elasticity demand, x(p) = A/pα with positive A and α, taking logarithms separates the coefficient and exponent. Differentiation gives elasticityDemand elasticity and expenditureThe source’s demand function x(p) decreases as price rises. Revenue or total expenditure is price times quantity demanded. Its derivative combines the direct effect of a higher price with the indirect effect of lower demand. The source defines… α at every positive price. The source’s example uses A = 1000 and α = 0.8. Its formula on PDF page 155 prints x in the denominator, while the explicit example on PDF page 156 uses the price variable p; the latter agrees with the demand interpretation.

x(p)=Ap−α
lnx(p)=lnA−αlnp
ε(p)=α
E(p)=Ap1−α
x(p)=1000p0.8

Differentials and approximation

The differential is the linear part of a function’s increment. It turns a small input change into an approximate output change. The derivative is a slope; the differential multiplies that slope by the actual input change. The neglected remainder is small compared with the change in the independent variable, rather than necessarily small compared with every other quantity in the problem.

Linear approximation and its remainder

df=f′(x)h
f(x+h)≈f(x)+f′(x)h
dy=dydxdx

The approximation uses the tangent line at the starting point. It becomes better in the source’s precise small-o sense as the input change shrinks. It is a local approximation and does not say the function is globally affine. At a stationary point the first differentialLinear approximation and its remainderThe approximation uses the tangent line at the starting point. It becomes better in the source’s precise small-o sense as the input change shrinks. It is a local approximation and does not say the function is globally affine. At a stationary point… is zero, although a higher-order change may remain. For x² at x = 0, the exact increment is h², despite the zero differential.

Worked example: the square function

(x+h)2=x2+2xh+h2
df=2xhfor f(x)=x2
f(1+h)≈1+2h

Differentials with several inputs

For a function of several variables, add the first-order effects of every component change. The gradientFunctions of two variables, partial derivatives and gradientA function of two real variables assigns one number to an admissible pair of inputs. Its graph is a surface in three-dimensional space, and a level curve collects input pairs giving the same output. In social-science models, multiple independent… contains the partial slopes, and its product with the increment vector gives the differentialLinear approximation and its remainderThe approximation uses the tangent line at the starting point. It becomes better in the source’s precise small-o sense as the input change shrinks. It is a local approximation and does not say the function is globally affine. At a stationary point…. A partial derivative by itself describes a change in one coordinate while holding the others fixed; the full differential combines simultaneous changes. Differentiability requires the remaining error to be negligible relative to the norm of the increment vector.

df=∇f(x)h=∑i=1nfxi(x)hi
f(x+h)−f(x)=∇f(x)h+o(‖h‖)

For f(x1,x2) = 2x1² + 3x2² + 10, the source moves from (1,1) to (1.05,0.95). The gradientPartial derivatives and gradientA partial derivative differentiates with respect to one input, holding every other input fixed. The usual differentiation rules still apply, but a variable held fixed behaves as a constant during that calculation. The gradient collects all first… at the start is (4,6) and the increment vector is (0.05,−0.05). The differentialLinear approximation and its remainderThe approximation uses the tangent line at the starting point. It becomes better in the source’s precise small-o sense as the input change shrinks. It is a local approximation and does not say the function is globally affine. At a stationary point… is −0.1, while the exact change is −0.0875. The gap 0.0125 is the quadratic remainder, giving a concrete measure of the approximation error.

∇f(1,1)=[46]
df=4(0.05)+6(−0.05)=−0.1
Δf=2(1.05)2+3(0.95)2+10−15=−0.0875

Functions of two variables, partial derivatives and gradient

A function of two real variables assigns one number to an admissible pair of inputs. Its graphThe source’s graph-sketching procedure is a surface in three-dimensional space, and a level curve collects input pairs giving the same output. In social-science models, multiple independent variables can represent separate policy instruments, media expenditures or attributes of a candidate. The source extends the same ideas to vectors of any dimension, but the syllabus emphasizes the two-variable case.

Domain, graph and level curves

f:ℝ2⟶ℝ
f(x1,x2)=k

For f(x1,x2) = x1x2, nonzero level curves are hyperbolas. Positive levels lie in the first and third quadrants; negative levels lie in the second and fourth. Fixing k and solving for one coordinate gives a way to draw the curve. Several such curves form a contour map, analogous to height contours on a geographic map.

x1x2=k
x2=kx1(x1≠0)

Norm and simultaneous increments

The Euclidean norm measures the size of a vector. In two dimensions it is the distance from the origin given by Pythagoras. A multivariable input change is itself a vector, so its norm measures how small that simultaneous change is. A norm is zero precisely when every component is zero. For a one-component vector it agrees with absolute value.

‖x‖=x12+⋯+xn2
‖(3,6)‖=32+62=35
‖(4,4,2)‖=42+42+22=6
‖(1,−2,3,−4)‖=30≈5.4772

Partial derivatives and gradient

A partial derivative differentiates with respect to one input, holding every other input fixed. The usual differentiation rulesDifferentiation rulesLinearity of differentiation means that fixed coefficients remain attached to their functions’ derivatives. The sum rule, difference rule and constant-multiple rule are special cases. This applies to a linear combination of functions, rather than to… still apply, but a variable held fixed behaves as a constant during that calculation. The gradient collects all first partial derivatives. The textbook stores it as a row vector, matching its use in the product with a column increment vector.

∇f(x1,x2)=[fx1(x1,x2)fx2(x1,x2)]

A stationary point of a multivariable function makes every first partial derivativePartial derivatives and gradientA partial derivative differentiates with respect to one input, holding every other input fixed. The usual differentiation rules still apply, but a variable held fixed behaves as a constant during that calculation. The gradient collects all first… zero. A single zero component of the gradientPartial derivatives and gradientA partial derivative differentiates with respect to one input, holding every other input fixed. The usual differentiation rules still apply, but a variable held fixed behaves as a constant during that calculation. The gradient collects all first… is insufficient. At a stationary point the full first differential vanishes for every increment direction. The point’s classification needs second-order informationHessian matrix and optimizationThe Hessian matrix collects all second partial derivatives. Its diagonal contains pure second derivatives, taken twice with respect to the same input; the off-diagonal entries contain mixed second derivatives. For a smooth function satisfying the… or another argument: a stationary point can be a maximum, minimum or saddle.

∇f(x*)=0

Worked example: a three-variable gradient

The source uses f(x1,x2,x3) = x1² + x1x2 − x2x3². In the first derivativeDerivative sign and monotonicityOn an interval where f′ is positive, the function is increasing: larger inputs have larger outputs. Where f′ is negative, the function is decreasing. Factoring a derivative is useful because it turns a sign problem for a sum into signs of factors.…, treat x2 and x3 as constants; in the second, treat x1 and x3 as constants; in the third, treat x1 and x2 as constants. Each partial has a different structure because the input occupies a different role in the formula.

fx1=2x1+x2
fx2=x1−x32
fx3=−2x2x3
∇f=[2x1+x2x1−x32−2x2x3]
  1. The equation −2x2x3 = 0 requires x2 = 0 or x3 = 0.
  2. If x2 = 0, the first equation gives x1 = 0 and the second then gives x3 = 0.
  3. If x3 = 0, the second equation gives x1 = 0 and the first then gives x2 = 0.
  4. Both cases produce the same stationary point: the origin. It is stationary because the gradient vanishes there, rather than because its coordinates happen to be zero.
x*=(0,0,0)

Separate changes versus joint changes

The source’s Boolean consensus function gives consensus only when both candidate attributes exceed their benchmark values. Moving along one coordinate alone does not gain consensus, while moving in a joint direction can. Thus separate one-coordinate behavior does not automatically determine joint behavior. This motivates the stronger increment definition of differentiability using the full vector change and its norm.

c(x1,x2)={1x1>x1+ and x2>x2+0otherwise

Hessian matrix and optimization

The Hessian matrix collects all second partial derivatives. Its diagonal contains pure second derivatives, taken twice with respect to the same input; the off-diagonal entries contain mixed second derivatives. For a smooth function satisfying the source’s regularity conditions, mixed derivatives are equal when their order is reversed, giving a symmetric matrix. Evaluating this matrix at a stationary point supports second-order classification.

Constructing the Hessian

Hf=∇2f=[fx1x1fx1x2fx2x1fx2x2]
fx1x2=fx2x1

To construct the Hessian, first find the gradientPartial derivatives and gradientA partial derivative differentiates with respect to one input, holding every other input fixed. The usual differentiation rules still apply, but a variable held fixed behaves as a constant during that calculation. The gradient collects all first…, then differentiate each of its components with respect to every variable. The first row is the gradient of the first partial derivativePartial derivatives and gradientA partial derivative differentiates with respect to one input, holding every other input fixed. The usual differentiation rules still apply, but a variable held fixed behaves as a constant during that calculation. The gradient collects all first…; the next row is the gradient of the second partial derivative. Distinguish the matrix-valued expression Hf(x) from the numerical matrix Hf(x*) obtained by inserting a candidate point.

Hf(x)=[21010−2x30−2x3−2x2]

First- and second-order conditions

For an interior unconstrained extremum, the first-order condition is a zero gradient. It finds candidates. The source’s second-order conditions inspect the principal north-west minors of the evaluated Hessian. All positive leading minors give a local minimum; alternating signs beginning with a negative minor give a local maximumStationary points and local extremaA local maximum is a point whose value is at least as large as values sufficiently near it. A local minimum reverses that comparison. At an interior extremum where the function is differentiable, its derivative must be zero. This is a necessary…. Strict inequalities provide a sufficient conclusion. Cases containing zero require more analysis.

H1>0, H2>0,…,Hn>0 ⟹ local minimum
(−1)kHk>0(k=1,…,n) ⟹ local maximum

For two variables, only the first diagonal entry and the determinantDeterminants of orders one, two and threeFor a 2 × 2 matrix, the absolute value of the determinant is the area of the parallelogram determined by its columns. A zero determinant corresponds to aligned columns and zero area. For a 3 × 3 matrix, the analogous magnitude is a parallelepiped… are needed for the strict cases. A positive determinant and positive first diagonal entry give a local minimumStationary points and local extremaA local maximum is a point whose value is at least as large as values sufficiently near it. A local minimum reverses that comparison. At an interior extremum where the function is differentiable, its derivative must be zero. This is a necessary…. A positive determinant and negative first diagonal entry give a local maximumStationary points and local extremaA local maximum is a point whose value is at least as large as values sufficiently near it. A local minimum reverses that comparison. At an interior extremum where the function is differentiable, its derivative must be zero. This is a necessary…. A negative determinant gives a saddle point, where the function moves upward in some directions and downward in others. A zero determinant makes this strict test inconclusive.

D=detHf=fx1x1fx2x2−fx1x2,2
  • D > 0 and fx₁x₁ > 0: local minimum.
  • D > 0 and fx₁x₁ < 0: local maximum.
  • D < 0: saddle point.
  • D = 0: no conclusion from this strict second-order test.

Worked example: quadratic cases and zero minors

f(x1,x2)=ax12+bx22
∇f=[2ax12bx2]
Hf=[2a002b]
H1=2a
H2=4ab
  • If a > 0 and b > 0, the origin is a minimum.
  • If a < 0 and b < 0, the origin is a maximum.
  • If ab < 0, the origin is a saddle point.
  • If one coefficient is zero, the Hessian test has a zero minor. Direct inspection of this special quadratic shows a weak minimum when the other coefficient is positive and a weak maximum when it is negative.
  • A zero minor in an arbitrary function cannot be given the same conclusion without examining that function.

Worked example: a negative second minor

The three-variable gradient example has only the stationary pointStationary points and local extremaA local maximum is a point whose value is at least as large as values sufficiently near it. A local minimum reverses that comparison. At an interior extremum where the function is differentiable, its derivative must be zero. This is a necessary… at the origin. Its evaluated HessianConstructing the HessianTo construct the Hessian, first find the gradient, then differentiate each of its components with respect to every variable. The first row is the gradient of the first partial derivative; the next row is the gradient of the second partial… has leading minors 2, −1 and 0. The negative second minor is incompatible with both strict minimum and strict maximum patterns. In this example, varying the first two coordinates already reveals both increasing and decreasing directions. The source classifies the origin as nonextremal.

Hf(0)=[210100000]
H1=2
H2=−1
H3=0

Optimization and applications

Optimization starts with an objective function and its meaningful domain. A stationary point is useful only after it is checked against that domain and classified. The supplied examples include monopoly profit, NGO efficiency and ordinary least squares. These examples link the mathematics to the quantities being maximized or minimized, and explain how changing a parameter affects the optimal decision.

Monopoly profit and marginal conditions

A monopolist selling quantity x uses an inverse demand function p(x). Revenue is quantity times unit price. Profit subtracts total cost from revenue. At an interior differentiable optimum, marginal revenue equals marginal cost. This equality is a first-order condition, so the objective still needs classification and the admissible domainIntervals and domainDetermining the domain comes before taking derivatives. Exclude zeros of a denominator, negative arguments of even roots and nonpositive arguments of logarithms. A restriction from one part of a formula cannot be ignored because another part is well… must still be checked.

I(x)=xp(x)
P(x)=I(x)−C(x)
P′(x)=I′(x)−C′(x)
I′(x*)=C′(x*)

The source specifies p(x) = B/xβ and C(x) = F + vx, with positive parameters. If 0 < β < 1, marginal profit changes from positive to negative and there is a unique positive optimum. The second derivativeConcavity, convexity and inflectionA concave function lies above the straight chords joining points on its graph. A convex function lies below them. A negative second derivative makes the slope decrease and gives concavity on the interval; a positive second derivative makes the slope… is negative throughout the positive domainIntervals and domainDetermining the domain comes before taking derivatives. Exclude zeros of a denominator, negative arguments of even roots and nonpositive arguments of logarithms. A restriction from one part of a formula cannot be ignored because another part is well…, making that optimum global there. Increasing the unit variable cost v reduces the optimal output. For β ≥ 1, profit is decreasing over positive outputs; the source interprets this as an incentive to stop production, rather than an interior positive optimum.

P(x)=Bx1−β−F−vx
P′(x)=(1−β)Bx−β−v
x*=((1−β)Bv)1/β(0<β<1)
P″(x)=−β(1−β)Bx−β−1<0(0<β<1)

NGO average cost and efficient scale

An NGO with fixed cost F and variable unit cost v(x) incurs total cost C(x) = F + xv(x). Average cost divides that amount by output. Fixed cost per unit falls with scale, while the variable unit cost can rise. The efficient scale balances those effects. At a differentiable minimum of average cost, average cost equals marginal cost.

C(x)=F+xv(x)
A(x)=C(x)x=Fx+v(x)
A′(x)=−Fx2+v′(x)
v′(x*)=F(x*)2
M(x)=C′(x)=v(x)+xv′(x)
A(x*)=M(x*)

For a linear unit variable cost v(x) = α + βx, with positive α and β, the average-cost function falls initially because F/x is large, then rises because βx grows. The only positive stationary quantity is √(F/β). The minimized average and marginal costs both equal α + 2√(Fβ). This makes the optimum’s dependence on the fixed investment explicit.

C(x)=F+αx+βx2
A(x)=Fx+α+βx
A′(x)=−Fx2+β
x*=Fβ
A(x*)=α+2Fβ
M(x)=α+2βx
M(x*)=α+2Fβ

Ordinary least squares as minimization

The source models observed consumption Ct as an affine function of GNP Yt plus an error. Adding signed errors can conceal large positive and negative misses. Ordinary least squares instead minimizes the sum of squared errors over the intercept and slope. Its stationary equations are a linear system, connecting linear algebraLinear systems and applicationsA linear system consists of equations that are linear in the unknown components of a vector. It is written Ax = b, with A the coefficient matrix, x the unknown vector and b the vector of known terms. A solution is the entire vector that satisfies… with multivariable optimization.

Ct=a+bYt+εt
f(a,b)=∑t=1T(Ct−a−bYt)2
fa=−2∑t=1T(Ct−a−bYt)
fb=−2∑t=1TYt(Ct−a−bYt)
C―−a−bY―=0
YC―−aY―−bY2―=0
b*=YC―−Y―C―Y2―−(Y―)2
a*=C―−b*Y―

The Hessian of this objective is constant. Its first leading minor is positive and its determinantDeterminants of orders one, two and threeFor a 2 × 2 matrix, the absolute value of the determinant is the area of the parallelogram determined by its columns. A zero determinant corresponds to aligned columns and zero area. For a 3 × 3 matrix, the analogous magnitude is a parallelepiped… is positive whenever the observed GNP values are not all identical. Under that condition, the stationary pair gives a global minimum. The denominator in the slope formula is the GNP variance, and the numerator is the covariance of GNP and consumption. If the variance is zero, the displayed division does not identify a unique slope.

Hf=[2T2TY―2TY―2TY2―]
detHf=4T2(Y2―−(Y―)2)

Constrained optimization and applications

Constrained optimization chooses the best value of an objective function among inputs satisfying a restriction. A resource constraint prevents the independent variables from being varied freely. The source introduces Lagrange multipliers through a production-resource interpretation, then applies them to an electoral budget and a consumption budget. The constraint changes both the candidate equations and the second-order classification.

Constraint, Lagrangian and first-order conditions

maxx1,x2f(x1,x2)subject to g(x1,x2)=b

The Lagrangian adds the objective to a multiplierShadow price and budget sensitivityThe shadow price is the optimal multiplier under the source’s sign convention. If the optimal objective value is y*(b), then a small resource increase h changes it approximately by λ*h. The multiplier therefore measures the marginal value of… times the difference between the available resource and its use. The source uses the sign convention L = f + λ(b − g). Holding λ fixed allows resource overuse or underuse to be priced. Making λ a variable restores the constraintConstraint, Lagrangian and first-order conditionsThe Lagrangian adds the objective to a multiplier times the difference between the available resource and its use. The source uses the sign convention L = f + λ(b − g). Holding λ fixed allows resource overuse or underuse to be priced. Making λ a… through its derivative. At a candidate optimum, solve all derivatives of the Lagrangian simultaneously.

L(λ;x1,x2)=f(x1,x2)+λ[b−g(x1,x2)]
Lλ=b−g(x1,x2)=0
Lx1=fx1−λgx1=0
Lx2=fx2−λgx2=0

The multiplier is an auxiliary unknown, so one equality constraintConstraint, Lagrangian and first-order conditionsThe Lagrangian adds the objective to a multiplier times the difference between the available resource and its use. The source uses the sign convention L = f + λ(b − g). Holding λ fixed allows resource overuse or underuse to be priced. Making λ a… in two variables gives three equations in three unknowns. These are first-order conditions, and solving them does not by itself prove a maximum or minimum. The source describes the LagrangianConstraint, Lagrangian and first-order conditionsThe Lagrangian adds the objective to a multiplier times the difference between the available resource and its use. The source uses the sign convention L = f + λ(b − g). Holding λ fixed allows resource overuse or underuse to be priced. Making λ a…’s constrained optimum as a saddle rather than an unconstrained maximum in all its variables. Candidate feasibility and second-order behavior must both be checked.

Worked example: a quadratic objective and a linear constraint

f(x1,x2)=10−x12−x22
x1+x2=10
L=10−x12−x22+λ(10−x1−x2)
∇L=[10−x1−x2−2x1−λ−2x2−λ]
x1*=5
x2*=5
λ*=−10

Subtracting the two input equations forces x1 = x2. The budget equation then gives five units of each. The multiplierShadow price and budget sensitivityThe shadow price is the optimal multiplier under the source’s sign convention. If the optimal objective value is y*(b), then a small resource increase h changes it approximately by λ*h. The multiplier therefore measures the marginal value of… follows from either input equation. Substituting into the objective gives −40. Eliminating one input by the constraintConstraint, Lagrangian and first-order conditionsThe Lagrangian adds the objective to a multiplier times the difference between the available resource and its use. The source uses the sign convention L = f + λ(b − g). Holding λ fixed allows resource overuse or underuse to be priced. Making λ a… provides an independent check: the remaining one-variable quadratic is concave and peaks at five.

f(5,5)=10−25−25=−40
f(x1,10−x1)=−2(x1−5)2−40

Shadow price and budget sensitivity

The shadow price is the optimal multiplierShadow price and budget sensitivityThe shadow price is the optimal multiplier under the source’s sign convention. If the optimal objective value is y*(b), then a small resource increase h changes it approximately by λ*h. The multiplier therefore measures the marginal value of… under the source’s sign convention. If the optimal objective value is y*(b), then a small resource increase h changes it approximately by λ*h. The multiplier therefore measures the marginal value of relaxing the resource amount. Its sign is meaningful: an equality constraintConstraint, Lagrangian and first-order conditionsThe Lagrangian adds the objective to a multiplier times the difference between the available resource and its use. The source uses the sign convention L = f + λ(b − g). Holding λ fixed allows resource overuse or underuse to be priced. Making λ a… can require a change in allocation that lowers the objective when its right-hand side rises.

dy*db=λ*(b)
y*(b+h)−y*(b)=λ*(b)h+o(h)

For the electoral campaign objective V(x1,x2) under x1 + x2 = b, each medium’s marginal votes equal the optimal multiplierShadow price and budget sensitivityThe shadow price is the optimal multiplier under the source’s sign convention. If the optimal objective value is y*(b), then a small resource increase h changes it approximately by λ*h. The multiplier therefore measures the marginal value of…. The source interprets that number as extra votes per small increase in the budget. PDF page 186 repeats the first subscript in the equality; the first-order conditionsFirst- and second-order conditionsFor an interior unconstrained extremum, the first-order condition is a zero gradient. It finds candidates. The source’s second-order conditions inspect the principal north-west minors of the evaluated Hessian. All positive leading minors give a… on page 185 show that the comparison is between the first and second media derivatives.

Vx1(x*)=Vx2(x*)=λ*

The bordered Hessian

With one equality constraint and two variables, the bordered Hessian is the HessianConstructing the HessianTo construct the Hessian, first find the gradient, then differentiate each of its components with respect to every variable. The first row is the gradient of the first partial derivative; the next row is the gradient of the second partial… of the LagrangianConstraint, Lagrangian and first-order conditionsThe Lagrangian adds the objective to a multiplier times the difference between the available resource and its use. The source uses the sign convention L = f + λ(b − g). Holding λ fixed allows resource overuse or underuse to be priced. Making λ a… with variables ordered as (λ,x1,x2). Its first entry is zero because the Lagrangian is linear in the multiplierShadow price and budget sensitivityThe shadow price is the optimal multiplier under the source’s sign convention. If the optimal objective value is y*(b), then a small resource increase h changes it approximately by λ*h. The multiplier therefore measures the marginal value of…. The constraint derivatives form the border, and the input–input block contains the second derivatives of the Lagrangian. Under the source’s regular constraint framework, a positive determinantDeterminants of orders one, two and threeFor a 2 × 2 matrix, the absolute value of the determinant is the area of the parallelogram determined by its columns. A zero determinant corresponds to aligned columns and zero area. For a 3 × 3 matrix, the analogous magnitude is a parallelepiped… gives its sufficient maximum condition.

HL=[0−gx1−gx2−gx1Lx1x1Lx1x2−gx2Lx2x1Lx2x2]
detHL=2Vx1x2−Vx1x1−Vx2x2when g=x1+x2

For n inputs and m regular equality constraints, the source uses the last n − m principal north-west minors of the bordered HessianConstructing the HessianTo construct the Hessian, first find the gradient, then differentiate each of its components with respect to every variable. The first row is the gradient of the first partial derivative; the next row is the gradient of the second partial…. The first relevant order is 2m + 1 and the last is m + n. For its strict maximum test, their signs alternate beginning with (−1)m+1. The two-input, one-constraintConstraint, Lagrangian and first-order conditionsThe Lagrangian adds the objective to a multiplier times the difference between the available resource and its use. The source uses the sign convention L = f + λ(b − g). Holding λ fixed allows resource overuse or underuse to be priced. Making λ a… case has only one relevant minor, the order-three determinantDeterminants of orders one, two and threeFor a 2 × 2 matrix, the absolute value of the determinant is the area of the parallelogram determined by its columns. A zero determinant corresponds to aligned columns and zero area. For a 3 × 3 matrix, the analogous magnitude is a parallelepiped…. This supplementary general rule should be studied only if covered before the partial.

δ=n−m
k=2m+1,…,m+n

Worked example: allocating an electoral budget

The source specifies votes as V = a x1α₁x2α₂, with positive parameters and positive spending. Maximizing its natural logarithmLogarithms and their identitiesA logarithm inverts an exponential function. The logarithm of x to base a is the exponent needed to obtain x from a. Its argument must be positive, and its base must be positive and different from one. Logarithm values can be negative: log10(0.1) =… gives the same allocation, because ln is increasing. The logarithm turns the product into a sum, making the first-order equations simple. The budget is divided in proportion to the two response elasticities, rather than necessarily in equal halves.

V(x1,x2)=ax1α1x2α2
W=lnV=lna+α1lnx1+α2lnx2
L=lna+α1lnx1+α2lnx2+λ(b−x1−x2)
α1x1=λ
α2x2=λ
λ*=α1+α2b
x1*=α1α1+α2b
x2*=α2α1+α2b
HL=[0−1−1−1−α1/x120−10−α2/x22]
detHL=α1x12+α2x22>0
ε1=x1Vx1V=α1

The allocation that maximizes V also maximizes ln(V), but their shadow prices have different units. The multiplierShadow price and budget sensitivityThe shadow price is the optimal multiplier under the source’s sign convention. If the optimal objective value is y*(b), then a small resource increase h changes it approximately by λ*h. The multiplier therefore measures the marginal value of… just calculated measures change in log votes per budget unit. Multiplying it by the optimal vote total gives the marginal vote gain for the original objective. Changing the objective by an increasing transformation preserves the chosen inputs, while changing the meaning of its multiplier.

Worked example: optimal consumption and a source discrepancy

The source uses Cobb–Douglas utility u(a,b) = aαb1−α, with 0 < α < 1, and budget pAa + pBb = β. The printed partial derivatives on PDF page 187 omit the multiplicative factors involving the other good. The corrected derivatives of the stated objective are shown below. They imply the same optimal spending shares that the source reports, but a different shadow-price expression from its printed one.

u(a,b)=aαb1−α
b=c1/(1−α)aα/(α−1)
L=aαb1−α+λ(β−pAa−pBb)
ua=αaα−1b1−α
ub=(1−α)aαb−α
uaub=αb(1−α)a=pApB
a*=αβpA
b*=(1−α)βpB
λ*=αα(1−α)1−αpAαpB1−α

The source instead prints αα(1−α)1−α(pB/pA)α for the multiplierShadow price and budget sensitivityThe shadow price is the optimal multiplier under the source’s sign convention. If the optimal objective value is y*(b), then a small resource increase h changes it approximately by λ*h. The multiplier therefore measures the marginal value of…. The corrected expression above follows by substituting the reported optimal basket into ua/pA. This explicitly retains the disagreement and keeps the marginal utility of money consistent with the stated utility function and prices.

Functions exercises

The functions exercisesFunctions exercisesThe functions exercises require the same sequence of decisions used in theory: specify the domain, differentiate, solve the stationary equations, classify the candidates and state the answer in the model’s terms. The supplied textbook exercises… require the same sequence of decisions used in theory: specify the domain, differentiate, solve the stationary equations, classify the candidates and state the answer in the model’s terms. The supplied textbook exercises include one-variable optimization, elasticity, gradients and equality constraints. The selected worked solutions below are derived from their supplied statements; the steps are calculations from those statements rather than additional academic sources.

October 2015: a logarithmic objective

The 2015 first partial studies f(x) = (x + 2)ln(x + 2). The logarithm requires x > −2. The product rule gives a logarithm plus one; setting it to zero and undoing the logarithm produces the only stationary pointStationary points and local extremaA local maximum is a point whose value is at least as large as values sufficiently near it. A local minimum reverses that comparison. At an interior extremum where the function is differentiable, its derivative must be zero. This is a necessary…. The second derivative is positive throughout the domainIntervals and domainDetermining the domain comes before taking derivatives. Exclude zeros of a denominator, negative arguments of even roots and nonpositive arguments of logarithms. A restriction from one part of a formula cannot be ignored because another part is well…, so the stationary point is both a local and a global minimum.

f(x)=(x+2)ln(x+2)
domf=(−2,+∞)
f′(x)=ln(x+2)+1
ln(x+2)=−1
x*=e−1−2≈−1.6321
f″(x)=1x+2>0(x>−2)

Textbook exercises 2.4.1–2.4.3

Exercise 2.4.1 asks for extrema of x − ln(x), 4x + 1/x and ln(x) − √x. The first and third require positive inputs, while the second excludes zero. Each case is handled by the same stationarity condition, with its own domainIntervals and domainDetermining the domain comes before taking derivatives. Exclude zeros of a denominator, negative arguments of even roots and nonpositive arguments of logarithms. A restriction from one part of a formula cannot be ignored because another part is well… and sign classification. Keeping the domain visible prevents using an algebraic root that is not admissible.

D(x−lnx)=1−1x
x*=1for x−lnx
D2(x−lnx)=1x2>0
D(4x+1/x)=4−1x2
x*=±12for 4x+1/x
D2(4x+1/x)=2x3

Thus 4x + 1/x has a local minimumStationary points and local extremaA local maximum is a point whose value is at least as large as values sufficiently near it. A local minimum reverses that comparison. At an interior extremum where the function is differentiable, its derivative must be zero. This is a necessary… at 1/2 on the positive branch and a local maximumStationary points and local extremaA local maximum is a point whose value is at least as large as values sufficiently near it. A local minimum reverses that comparison. At an interior extremum where the function is differentiable, its derivative must be zero. This is a necessary… at −1/2 on the negative branch. Neither is a global extremum over the complete domainIntervals and domainDetermining the domain comes before taking derivatives. Exclude zeros of a denominator, negative arguments of even roots and nonpositive arguments of logarithms. A restriction from one part of a formula cannot be ignored because another part is well… R excluding 0, because the other branch and the behavior near zero produce unbounded values. For the third function, compare the derivative’s numerator 2 − √x on the positive domain.

D(lnx−x)=1x−12x
x*=4for lnx−x
D2(lnx−x)|x=4=−132

Exercise 2.4.2 uses f(x) = ln(x) − (ln(x))². Its domainIntervals and domainDetermining the domain comes before taking derivatives. Exclude zeros of a denominator, negative arguments of even roots and nonpositive arguments of logarithms. A restriction from one part of a formula cannot be ignored because another part is well… is positive inputs. The derivative vanishes when ln(x) = 1/2, so x* = √e. Writing the objective as 1/4 − (ln(x) − 1/2)² shows directly that this is a global maximum. The second derivativeConcavity, convexity and inflectionA concave function lies above the straight chords joining points on its graph. A convex function lies below them. A negative second derivative makes the slope decrease and gives concavity on the interval; a positive second derivative makes the slope… agrees with that classification.

f′(x)=1−2lnxx
x*=e
f(x)=14−(lnx−12)2
f″(x)=2lnx−3x2

Exercise 2.4.3 uses f(x) = xαe−x with positive α and nonnegative inputs. For positive inputs, its derivative has the sign of α − x. It increases before α and decreases after α, giving a global maximum at x = α. The boundary zero is included in the original exercise and should be considered separately when discussing minima.

f′(x)=xα−1e−x(α−x)
x*=α
f″(α)=−αα−1e−α<0

Textbook exercise 2.4.4: a cubic cost function

The exercise supplies C(x) = a + bx³, fixed cost 500,000 and cost 600,000 at output ten. Substitute the first datum at zero to obtain a, then use the second to obtain b. Average costNGO average cost and efficient scaleAn NGO with fixed cost F and variable unit cost v(x) incurs total cost C(x) = F + xv(x). Average cost divides that amount by output. Fixed cost per unit falls with scale, while the variable unit cost can rise. The efficient scale balances those… divides the resulting total cost by output. The efficient positive quantity balances the derivative of the reciprocal term with the derivative of the quadratic term.

a=C(0)=500000
500000+1000b=600000
b=100
m(x)=500000x+100x2
m′(x)=−500000x2+200x
x*=25003≈13.5721
m″(x)=1000000x3+200>0(x>0)

Textbook exercise 2.4.9: a logarithmic two-variable function

The exercise states f(x1,x2) = x1² ln(x2) − x2. Its domainIntervals and domainDetermining the domain comes before taking derivatives. Exclude zeros of a denominator, negative arguments of even roots and nonpositive arguments of logarithms. A restriction from one part of a formula cannot be ignored because another part is well… requires x2 > 0. The first partial derivativePartial derivatives and gradientA partial derivative differentiates with respect to one input, holding every other input fixed. The usual differentiation rules still apply, but a variable held fixed behaves as a constant during that calculation. The gradient collects all first… is a product of x1 and ln(x2); the second contains a quotient. Solving both partial equations gives stationary points (1,1) and (−1,1). A zero first coordinate cannot satisfy the second equation.

∇f=[2x1lnx2x12/x2−1]
x2=x12
x2=1
x*=(1,1) or (−1,1)
f(2,e)=4−e

Textbook exercise 2.4.11: checking a candidate

The source asks for the Lagrangian and stationary points of f = 4x1x2 − 2x1² + x2² under 3x1 + x2 = 5. Solving the Lagrange equations gives x1 = −5 and x2 = 20. Substitution into the constraintConstraint, Lagrangian and first-order conditionsThe Lagrangian adds the objective to a multiplier times the difference between the available resource and its use. The source uses the sign convention L = f + λ(b − g). Holding λ fixed allows resource overuse or underuse to be priced. Making λ a… reduces the objective to a one-variable concave quadratic, verifying that this candidate is a maximum on the equality line.

L=4x1x2−2x12+x22+λ(5−3x1−x2)
Lx1=4x2−4x1−3λ=0
Lx2=4x1+2x2−λ=0
Lλ=5−3x1−x2=0
(x1*,x2*,λ*)=(−5,20,20)
f(x1,5−3x1)=50−(x1+5)2

Review exercises for the first partial

The first partial assesses the material actually reached in class. This review therefore separates the confirmed date from the provisional topic boundary. The syllabus permits both open-answer questions and closed-answer questions, so preparation requires definitions, efficient calculations and reasons for each conclusion. The examples in this article provide practice for those different forms; the supplied historical papers cannot fix the current exam’s coverage.

Review: linear algebra

  • For given vectors, test linear independence using a zero linear combination or the rank of their column matrix.
  • For each matrix product, state the input and output dimensions before calculating.
  • For a rank question, reduce the matrix and count independent pivot rows, or justify a nonzero minor and the failure of larger minors.
  • For an inverse, check square shape and nonsingularity, then calculate and verify by multiplication.
  • For a linear system, compare coefficient and augmented ranks before stating whether the solution is unique, free-parameter or impossible.
  • For a Leontief system, distinguish existence of the mathematical inverse from viability.
ρ(A)≠ρ([A∣b]) ⟹ no solution

Review: calculus and optimization

  1. State the domain, including any restrictions of the application.
  2. Identify whether the derivative requires a sum rule, product rule, quotient rule or chain rule.
  3. Solve the stationarity equations and keep only admissible candidates.
  4. Use derivative signs or appropriate second-order conditions to classify each candidate.
  5. For a constrained optimum, enforce the constraint alongside the marginal equations.
  6. State whether the conclusion is local or global, and explain the mathematical reason.
∇f(x*)=0 is necessary, not by itself sufficient

Formula sheet

The formulas below collect the principal computational rules already derived in this article. They retain the same source citations and notation. Each rule has assumptions: matrix multiplication requires conformable dimensions, an inverse requires nonsingularity, logarithmic derivatives require a valid domain, and second-order tests require the relevant differentiability and stationarity conditions. The sheet supplements the worked sections rather than replacing those assumptions.

(x+y)i=xi+yi
(αx)i=αxi
v=∑h=1kαhvh
(AT)ij=aji
(A+B)ij=aij+bij
(αA)ij=αaij
(AB)ij=∑k=1naikbkj
0≤ρ(A)≤min(m,n)
det[abcd]=ad−bc
detA=a11a22a33+a12a23a31+a13a21a32−a13a22a31−a11a23a32−a12a21a33
Cij=(−1)i+jMij
detA=∑j=1naijCij
AA−1=A−1A=I
detA≠0 ⟺ ρ(A)=n ⟺ A−1 exists
A−1=1detA(A*)T
[abcd]−1=1ad−bc[d−b−ca]
[A∣I]~[I∣A−1]
Ax=b is possible ⟺ ρ(A)=ρ([A∣b])
ρ(A)=ρ([A∣b])=n
ρ(A)=ρ([A∣b])<n
degrees of freedom=n−ρ(A)
x*=A−1b=[−1010]
(I−A)x=c
x=(I−A)−1c
Mk(I−A)>0(k=1,…,n)
f(x)=mx+q
f(x)=ax2+bx+c(a≠0)
Δ=b2−4ac
x1,2=−b±Δ2a
f(x)=ax(a>0)
y=logax ⟺ ay=x
ln(pq)=lnp+lnq
ln(p/q)=lnp−lnq
ln(pq)=qlnp
ln(ex)=x
elnx=x(x>0)
ax=exlna
r(h)=o(h) ⟺ |r(h)||h|⟶0
f(x+h)−f(x)=f′(x)h+o(h)
f′(x)=limh→0f(x+h)−f(x)h
y=f(x0)+f′(x0)(x−x0)
Dk=0
D(mx+q)=m
D(xα)=αxα−1
D(ax)=axlna
D(ex)=ex
D(lnx)=1x
D(αf)=αf′
D(f+g)=f′+g′
D(f−g)=f′−g′
D(αf+βg)=αf′+βg′
D(fg)=f′g+fg′
D(fg)=f′g−fg′g2(g≠0)
D[f(g(x))]=f′(g(x))g′(x)
f″(x)=D[f′(x)]
f′(x)>0 ⟹ f is increasing on the interval
f′(x)<0 ⟹ f is decreasing on the interval
f′(x*)=0
f″(x)>0 ⟹ f is convex on the interval
f″(x)<0 ⟹ f is concave on the interval
f′(x*)=0, f″(x*)>0 ⟹ x* is a local minimum
f′(x*)=0, f″(x*)<0 ⟹ x* is a local maximum
ε(p)=p|x′(p)|x(p)
df=f′(x)h
f(x+h)≈f(x)+f′(x)h
df=∇f(x)h=∑i=1nfxi(x)hi
‖x‖=x12+⋯+xn2
∇f(x1,x2)=[fx1(x1,x2)fx2(x1,x2)]
∇f(x*)=0
Hf=∇2f=[fx1x1fx1x2fx2x1fx2x2]
H1>0, H2>0,…,Hn>0 ⟹ local minimum
(−1)kHk>0(k=1,…,n) ⟹ local maximum
D=detHf=fx1x1fx2x2−fx1x2,2
I′(x*)=C′(x*)
L(λ;x1,x2)=f(x1,x2)+λ[b−g(x1,x2)]
Lλ=b−g(x1,x2)=0
Lx1=fx1−λgx1=0
Lx2=fx2−λgx2=0
dy*db=λ*(b)
HL=[0−gx1−gx2−gx1Lx1x1Lx1x2−gx2Lx2x1Lx2x2]

Sources and provisional coverage

The 2026/2027 syllabus provides topic order, the course textbook and the current rule for splitting the exam. The course presentation supplies the 21 October date. The textbook provides theory and examples; the vector slides supply current vector definitions and numerical applications; the past exams supply historical practice. Class attachments and notes are still forthcoming, so the endpoint of the 2026 first partial remains provisional.

  • 30319-syllabus-2026: syllabus.pdf, PDF pages 1–4, academic year 2026/2027.
  • 30319-course-presentation: Course presentation.pdf, PDF pages 1–3.
  • 30319-textbook: Peccati, D’Amico and Cigola, Maths for Social Sciences, 2018; supplied 353-page PDF.
  • 30319-vectors-slides: LA1(1).pdf, PDF pages 1–6.
  • 30319-past-exams: Exercises_past_exams.pdf, 19 pages; first-partial questions on PDF pages 1–2 and 10–11.