Amogh N P
✦ In loving memory of Amogh N P — Architect · Designer · Visionary ✦
A building massing study with a coloured daylight-analysis gradient washing from cool blue to hot red across its surfaces: form judged by measured performance, not by eye alone.
Unit IVComputational Design Process

Performance & Optimisation

Simulation in the loop, the fitness function, and the Pareto front.

≈ 45 min + studio task

Performance-driven design closes a feedback loop: the parametric model is coupled to ANALYSIS so that measured performance feeds back into form. To let a computer search for better designs, 'good' must become a number — a fitness function. Learn simulation-in-the-loop, objectives vs constraints, single- vs multi-objective optimisation and the Pareto front, evolutionary/genetic algorithms, and design-space exploration. And the honest caveat: 'optimised' is only as meaningful as the fitness function — what isn't measured isn't optimised.

Learning objectives

By the end of this lesson, you will be able to — mapped to the course outcomes for Computational Design Process:

1
CO4 · Understand

Explain simulation-in-the-loop and how performance feeds back into form.

2
CO4 · Apply

Define a fitness function and distinguish objectives from constraints.

3
CO4 · Analyse

Explain single- vs multi-objective optimisation, the Pareto front, and genetic algorithms.

4
CO6 · Evaluate

Judge the limits of 'optimised' — what the fitness function fails to capture.

Defining 'good' numerically

Simulation in the loop & fitness

Couple the model to analysis so performance feeds back into form; the fitness function turns the brief into a number, and objectives must be kept distinct from hard constraints.[1]

Simulation in the loop parametric model analysisdaylight · energy · structure fitness score generate variant → ↑ performance feeds back into form Form is judged by MEASURED behaviour, and that measurement drives the next iteration.
DiagramSimulation in the loop — the parametric model coupled to analysis so performance feeds back into form

Performance feeds back into form

Performance-driven design couples the parametric model to ANALYSIS so performance feeds back into form. Instead of design → hand-off → 'it's too hot/dark/heavy', the designer embeds analysis inside the design environment — daylight, solar, energy, structure, CFD/airflow, acoustics evaluated on every variant. Form is judged by MEASURED behaviour, and that measurement drives the next iteration.[1]

The Pareto front minimise heat gain → maximise daylight → Pareto front non-dominated trade-offs You can't improve one objective without worsening another — the optimiser maps the front; the DESIGNER picks a point.
DiagramA Pareto front — trade-off-optimal solutions for two competing objectives
Genetic search, then judgement

Optimisation & its honest limits

Genetic algorithms search a huge space toward good solutions; design-space exploration builds intuition; but 'optimised' is only as meaningful as the fitness function — what isn't measured isn't optimised.[3, 1]

How a genetic algorithm searches population evaluate fitness select fittest crossover + mutate→ next generation ↻ iterate toward GOOD (not guaranteed optimal) solutions Evolution-inspired search (Holland, 1975); Galapagos single-objective, Octopus multi-objective.
DiagramHow a genetic algorithm searches — generate a population, evaluate fitness, select, crossover and mutate

How the search works

When the design space is huge and non-smooth, metaheuristics search intelligently. Genetic algorithms (GAs, Holland 1975) mimic evolution: encode each design as a 'genome' (its parameters), generate a population, evaluate fitness, then SELECT the fittest, CROSSOVER and MUTATE them to breed the next generation. Galapagos (in Grasshopper) is the emblematic solver; Octopus extends it to multi-objective/Pareto. GAs don't guarantee the global optimum — they converge toward GOOD solutions.[3, 4]

Single- vs multi-objective

At a glance

AspectSingle-objectiveMulti-objective
Number of goalsSingle-objective: oneMulti-objective: two or more (often competing)
ResultSingle: one best valueMulti: a Pareto front of trade-offs
Designer's roleSingle: accept/validate the optimumMulti: choose a point on the front by judgement
Typical solverSingle: Galapagos (GA / annealing)Multi: Octopus (Pareto-based)
RiskSingle: optimises one thing, ignores restMulti: harder to interpret; still model-bound
Vocabulary

Key terms

Fitness function

A numeric score of a design's performance, driving the search.

Objective

A quantity to be maximised or minimised.

Constraint

A hard condition a valid design must satisfy.

Pareto front

The set of non-dominated, trade-off-optimal solutions.

Genetic algorithm

A metaheuristic search inspired by selection, crossover and mutation.

Design-space exploration

Generating/comparing many options to understand trade-offs.

Apply it

Studio task

For a façade or a roof, write a fitness function in words — what you would measure (e.g. daylit area, solar heat gain) and the constraints (code, budget). Identify two objectives that genuinely conflict, sketch the Pareto front you'd expect, and pick a point on it with a reason. Finally, name one important value your fitness function does NOT capture — and that the optimiser would therefore silently sacrifice.

Check your understanding

Self-assessment

1. On a Pareto front, every solution is one where —

2. A genetic algorithm's core cycle is —

3. 'This design is optimised' most accurately means —

In a nutshell

Recap

Performance-driven design couples the model to simulation so measured behaviour feeds back into form.
The fitness function turns 'good' into a number — it is the brief made computable, and the riskiest thing to get right.
Distinguish objectives (maximise/minimise) from constraints (hard rules); a well-posed problem states both.
Multi-objective optimisation yields a Pareto front of trade-offs; genetic algorithms search toward good solutions.
'Optimised' means optimal only for the stated objective — what isn't measured isn't optimised.
The evidence

References & further reading

  1. [1]Robert Woodbury, Elements of Parametric Design (2010) — links parametric models to search/exploration.
  2. [2]Wassim Jabi, Parametric Design for Architecture (2013) — coupling models to evaluation.
  3. [3]John H. Holland, Adaptation in Natural and Artificial Systems (1975) — origin of genetic algorithms.
  4. [4]Galapagos (single-objective) & Octopus (multi-objective/Pareto) evolutionary solvers for Grasshopper (verify versions).
  5. [5]Vilfredo Pareto (c. 1900) — Pareto optimality, the trade-off concept.

Further reading

  • Robert Woodbury — Elements of Parametric Design.
  • Wassim Jabi — Parametric Design for Architecture.
  • Branko Kolarevic — Architecture in the Digital Age.

Sources gathered and fact-checked June 2026. Published values vary by source, sample and method — treat as indicative and confirm against the cited standard before structural use.