Skip to content

Multi-Objective Optimization

How ASCICat handles multiple objectives.

The Challenge

Catalyst selection involves competing objectives:

  • Maximize activity
  • Maximize stability
  • Minimize cost

No single catalyst optimizes all simultaneously.

Approaches

1. Weighted Sum (ASCICat)

\[\phi = \sum_i w_i \cdot S_i\]

Advantages:

  • Single interpretable metric
  • Reproducible rankings
  • Explicit trade-offs

Limitations:

  • Non-convex Pareto regions missed
  • Weight selection required

2. Pareto Optimization

Find non-dominated solutions where no other solution is better on all objectives.

Advantages:

  • No weight specification needed
  • Complete trade-off information

Limitations:

  • Returns set, not ranking
  • Requires subjective final selection

3. Goal Programming

Minimize deviation from targets:

\[\min \sum_i |S_i - S_i^{target}|\]

Advantages:

  • Aspiration-based
  • Handles constraints

Limitations:

  • Targets often arbitrary

Why Weighted Sum?

ASCICat uses weighted sum because:

  1. Interpretability - Weights directly reflect priorities
  2. Reproducibility - Same weights = same ranking
  3. Comparability - Enables cross-study comparison
  4. Simplicity - Easy to understand and implement

Weight Selection

Default: Equal Weights

\[w_a = w_s = w_c \approx 0.33\]

Unbiased starting point for exploration.

Application-Specific

Application Suggested Weights
Fundamental research (0.5, 0.3, 0.2)
Industrial catalyst (0.35, 0.40, 0.25)
Large-scale deployment (0.30, 0.25, 0.45)

Sensitivity Analysis

Don't trust single weight choice - explore the weight space!

Mathematical Properties

Convex Combination

For \(w_i \geq 0\) and \(\sum w_i = 1\):

  • \(\phi \in [\min S_i, \max S_i]\)
  • Linear in scores
  • Continuous in weights

Pareto Optimality

Weighted sum solutions are Pareto-optimal when:

  • All weights positive
  • Pareto frontier is convex

Trade-off Analysis

For two catalysts A and B:

\[\Delta \phi = w_a(S_a^A - S_a^B) + w_s(S_s^A - S_s^B) + w_c(S_c^A - S_c^B)\]

A is preferred when \(\Delta \phi > 0\).

References

  • Marler, R. T. & Arora, J. S. Struct. Multidiscip. Optim. 26, 369 (2004)
  • Hwang, C.-L. & Masud, A. S. M. Multiple Objective Decision Making (Springer, 1979)