ValenceForgeCatalysis

Modalities

Catalysis

Catalysts accelerate chemical reactions and can favour useful products. Their performance depends on both the material and the conditions in which it operates.

For solid catalysts, activity depends on the working surface. Adsorption, restructuring and loss of active sites can change behaviour over time, making preparation and operating history essential to understanding performance.

AI Agent Potential

AI agents can connect experimental design with quantitative analysis. They can compare kinetic models, write reproducible analysis code and identify a change in conditions that would distinguish competing explanations.

The harder question is whether a model remains useful beyond the measurements used to fit it. A convincing steady-state fit may misrepresent recovery after a feed change, while narrow uncertainty intervals can conceal an incomplete physical description.

Dependable assistance requires scientific judgement throughout the workflow. An agent should preserve sample history and units, reserve measurements for independent validation, and account for variation between preparations before recommending a material.

Interpreting the Evidence

Activity needs a clear physical basis. A rate per gram of catalyst differs from a turnover frequency, which requires an estimate of active-site count. Comparing materials also requires calibration and checks for heat or mass-transfer limitations.

A changing signal can have several causes. Slow recovery might reflect surface chemistry, but gas transport and detector response can also delay the measurement. A useful experiment separates these contributions before assigning a mechanism.

Example Experiment

The catalyst, scenario and all plotted data are invented.

An agent must investigate why a fictional carbon monoxide oxidation catalyst loses activity in wet feed. Two explanations fit the steady response: reversible inhibition or lasting loss of active sites. The aim is to choose a measurement that distinguishes them.

Choose a feed history to compare predictions, then reveal the synthetic measurement. Temperature, pressure, reactant feeds and total flow stay fixed. Water inhibits this invented catalyst; its effect on real materials varies.

Fictional CO oxidation · synthetic responses

1 Define a controlled comparison

  1. Controlled feed

    CO + O₂ + balance gas.
    Vary water only.

  2. Catalyst bed

    Fixed temperature and pressure.
    History can change activity.

  3. Product measurement

    CO₂ downstream.
    Sampling adds a delay.

Illustrative CO + ½O₂ → CO₂ reaction. Inert balance gas compensates for the humidity change; the reactant feeds and total flow stay fixed.

A · Reversible inhibition

Water blocks a fraction of the available sites. Drying allows activity to recover.

B · Lasting activity loss

Activity remains depressed and does not recover during drying.

2 Choose an experiment

Synthetic humidity experiment: Wet steady statePredictions include the sampling delay. The steady response does not distinguish the models. The product response is normalised to an original dry reference; it is not conversion. Model A is reversible inhibition; model B is lasting activity loss. All points are invented. Synthetic observations are shown. Humidity at the catalyst Wet Relative CO₂ response 0.00.40.81.0 036912 Time (min)
Synthetic humidity experiment: Wet steady statePredictions include the sampling delay. The steady response does not distinguish the models. The product response is normalised to an original dry reference; it is not conversion. Model A is reversible inhibition; model B is lasting activity loss. All points are invented. Synthetic observations are shown. Humidity at the catalyst Wet Relative CO₂ response 0.00.40.81.0 036912 Time (min)

Model AModel BSynthetic observation

Response 1 = original dry reference. The model curves overlap in the wet steady state.

3 · Assess the evidence

The steady response does not distinguish the models.

Both explanations fit this condition. More precision at the same steady state cannot separate them.

Next step

Choose a dry/rewet cycle to test whether activity recovers.

Synthetic observations follow model A. Recovery supports a reversible contribution, but does not exclude some lasting damage or establish a unique mechanism.
About the illustration

Model A describes reversible site blocking; model B retains a fixed loss of activity. The model assumes negligible reactant depletion and heat or mass-transfer limitations, with humidity changing instantly at the catalyst. Response is normalised to the original dry reference, not to complete conversion.

The white points are model A’s output after a first-order sampling delay, with a small illustrative perturbation. Switching off the delay changes only the predictions. In practice, feed switching and the measurement response need independent calibration; these curves do not establish a molecular mechanism or statistical confidence.

This is purely an example task and is not part of GDPx.