Module 14 of 15

Multi-objective Optimization

When policies trade off goals that cannot all be maximised at once.

AMathematical Foundation

Multi-objective optimisation considers several objectives f₁, …, fₖ at once. Option a Pareto-dominates option b if a is at least as good on every objective and strictly better on at least one. Options not dominated by any other form the Pareto frontier. Mathematics can eliminate dominated options, but choosing among frontier options requires additional judgments: weights (maximise w·f₁ + (1 − w)·f₂), constraints (maximise f₁ subject to f₂ ≥ c), or other decision rules.

a\succ b\iff f_j(a)\ge f_j(b)\;\forall j\;\text{ and }\;f_k(a)>f_k(b)\text{ for some }k

In words: a dominates b if a is at least as good as b on every objective and strictly better on at least one.

$a, b$: two options (for example, two redevelopment plans)
$f_j$: score of an option on objective j, oriented so that higher is better

\max_{x}\; w\,\tilde f_1(x)+(1-w)\,\tilde f_2(x)\quad\text{s.t.}\quad f_2(x)\ge c

In words: Maximise a weighted sum of the rescaled objectives, subject to a minimum level of the second objective.

$w$: weight on objective 1 (a value judgment)
$\tilde f$: objective rescaled to a common 0–1 range
$c$: minimum acceptable level of objective 2 (a constraint)

Assumptions

Objectives can be measured for each option.
Weighted sums can only reach frontier points on the convex hull; constraints can reach others.

BIntuitive Explanation

A city choosing a redevelopment plan wants more homes and wants original residents to stay. A plan that delivers fewer homes and displaces more people than another plan is simply worse and can be discarded. Among the remaining plans, each extra home costs some displacement. How many homes are worth one family's displacement is not a mathematical question.

CConnection to Debate

Many debates are disputes about weights disguised as disputes about facts. Making the frontier explicit helps: if both sides agree an option is dominated, it drops out; the real clash is over where on the frontier to be. It also exposes false dilemmas: sometimes a design (one-for-one replacement, phased construction) moves the frontier outward and makes both objectives better.

DReal-World Application

Redevelopment and the right to return
Competing interpretation
London's estate regeneration guidance asks schemes involving demolition to increase affordable housing and give social tenants a full right to return. In the U.S. HOPE VI programme, grantees expected 46% of relocated residents to return, with large variation by site (GAO, 2003). In optimisation terms, a right-to-return guarantee is a constraint on housing supply, cost and speed: critics argue it shrinks the feasible set of schemes, while supporters argue that unconstrained schemes treat displacement as costless.
Sources
- Better Homes for Local People: The Mayor's Good Practice Guide to Estate RegenerationGreater London Authority, 2018-02
- Public Housing: HOPE VI Resident Issues and Changes in Neighborhoods Surrounding Grant Sites (GAO-04-109, Highlights)U.S. General Accounting Office, 2003-11
Luxury group strategy
Competing interpretation
When LVMH completed its acquisition of Tiffany in 2021 it appointed Louis Vuitton executives to lead the company, said the deal would deeply transform its jewellery division, and announced that Tiffany's chief artistic director would leave. Chanel's métiers d'art ateliers are owned by a Chanel subsidiary but encouraged to work for other brands and expected to be profitable. Both illustrate a trade-off between scale and coordination and the distinctiveness and creative autonomy of individual houses; consolidation debates are partly about which objective should constrain the other.
Sources
- LVMH completes the acquisition of Tiffany & Co. (press release, SEC Form 8-K exhibit)LVMH Moët Hennessy Louis Vuitton / Tiffany & Co., 2021-01-07
- Chanel gathers its métiers d'art ateliers under one roofAmy Serafin (Wallpaper*), 2021-10

These examples describe real policies and research. The numbers in the interactive below and in the practice questions are hypothetical teaching data unless a question says otherwise.

EInteractive Visualization

Move options on two objectives, see which are Pareto-dominated, and watch weights and constraints choose different frontier options.

Illustrative example — hypothetical data.Hypothetical redevelopment options scored on two objectives: net new homes, and the share of original residents able to stay or return.

Edit an option

F: net new homes500

F: residents who stay40%

F is dominated by C, G: another option is at least as good on both objectives and better on one.

Weight on new homes (vs. residents staying)50%

Constraint: at least this share must stay0%

Frontier options

A, B, C, D, E

Chosen under these weights

Mathematics can rule out dominated options (F, G). It cannot choose among frontier options: that choice depends on how much displacement you are willing to accept for each additional home, which is an ethical judgment expressed here as a weight or a constraint.

FDebate Practice

GLimitations

Optimisation needs measurable objectives, but some values (dignity, belonging, rights) resist measurement, and forcing them onto a scale can distort them. Weights and constraints are ethical choices, not outputs of the method; presenting a weighted score as objective hides the value judgment.