Tutorial 3a: Multiple Substances with AI

Motivation

In ABC, we have more than just domestic refrigeration! In this tutorial, we'll expand our model to capture more of the broader national profile, helping us understand where policy may put its focus. Specifically, this tutorial builds directly on our Tutorial 2 Domestic Refrigeration model, adding Domestic AC. We'll also introduce multiple refrigerant substances with different Global Warming Potentials (GWPs), demonstrating how volume and climate impact interact.

All that said, all of this can take quite a bit of clicking. So, we will see if AI can help us out to speed up the process. Specifically, we will demonstrate the loop of giving AI a task and checking its output.

Preparing for AI

In this tutorial series, we have used Claude. However, in practice, most AI assistants can help with Kigali Sim using a standard called llms.txt. So, to get AI ready to go, please create a new chat session. Then, tell it to look up information about Kigali Sim through a message like this:

Hello! I would like help with building a Kigali Sim simulation. Please read https://kigalisim.org/llms-full.txt?v=20260128 to learn more. Please stick to only features compatible with the UI editor.

Need a little cheat? Download the Tutorial 2 file here. Also, for more capable assistants like Claude, this is enough. However, some assistants cannot access the full internet or won't know how to work with this kind of file out of the box. If your assistant is having issues, instead, attach the Kigali Sim llms-full.txt file as an attachment!

When your assistant is ready, next give it your Tutorial 2 file. This can be done by going to the Save File button at the top of the Kigali Sim application and adding it to the chat as an attachment with a message like this:

Great! The attached is the simulation I started working on.

Note: More details and troubleshooting steps specific to individual AI assistants are available in Tutorial 11.

Add a new substance

Let's start with one new substance record. Here's a prompt:

Please add R-600a to the simulation. It will be in domestic refrigeration like HFC-134a with a GWP of 3 kgCO2e / kg and initial charge of 0.07 kg / unit. Please use retirement of 5% / year and 10% recharged per year at 0.07 kg / unit. Please have 100000 units of prior equipment and 2 mt / year of domestic manufacture. Thanks!

After the AI is done, download the resulting file and click "Load File" in the UI and upload the AI output. If you are comfortable, you can also copy from the AI assistant chat into the editor tab.

Check the work

Using what you learned in Tutorial 2, go ahead and click on edit next to the new consumption record. Is it what you expected? These kinds of simple prompts are almost always successful with more capable AI assistants like Claude but be careful to double check the output of any AI.

Note that, if Kigali Sim reports that a simulation requires use of the code editor, see Tutorial 7 or remind the AI assistant to only stick to UI editor compatible features.

Adding multiple substances

This is a great start but let's try to be more brief in our next message:

Great! Let's add two more substances...

HFC-32:

 - Application: Domestic AC
 - GWP: 675 kgCO2e / kg
 - Initial Charge: 0.85 kg / unit
 - Retirement: 7%
 - Recharge: 15% at 0.85 kg / unit
 - Prior Equipment: 40000 units
 - Domestic Manufacture: 15 mt / yr

R-410A:

 - Application: Domestic AC
 - GWP: 2088 kgCO2e / kg
 - Initial Charge: 1.00 kg / unit
 - Retirement: 7%
 - Recharge: 15% at 1.00 kg / unit
 - Prior Equipment: 20000 units
 - Domestic Manufacture: 5 mt / yr

Don't get lazy! Again, please double check the AI's work after loading its resulting simulation into Kigali Sim.

Interpreting Multi-Application Results

As you work, the simulation will update automatically.

Examine the results to understand how multiple applications and substances add together. You can do this by looking at results by selecting the Application or Substances radio buttons. To get a complete picture with the Emissions radio button, try clicking "configure custom" under emissions and combining both end-of-life and recharge emissions. This represents the total leakage throughout the equipment lifetime.

Before concluding, let's also pause to understand if these results make sense. First, the custom emissions which combines both end of life and recharge emissions is higher than either alone. Second, consider that the HFC-134a has higher volume and higher GWP than R-600a. Therefore, focusing on HFC-134a, we notice that these two factors intersect through a larger gap to R-600a in emissions relative to consumption when we have selected the Substances radio button.

Conclusion

You've successfully expanded ABC Country's model to include multiple applications and substances! To get there, AI helped speed up the process quite a bit. All that said, we considered:

• Multi-application modeling: Different sectors with distinct equipment characteristics and service patterns.
• Multi-substance analysis: Comparing different refrigerants within and across applications.
• GWP diversity: Understanding how different substances have varying climate impacts.
• Equipment population dynamics: How different applications scale and behave over time.

The model now provides a foundation for understanding how substance choice and application type interact to determine overall consumption and climate impact patterns.

Download the completed tutorial result at tutorial_03.qta which contains the complete multi-application and multi-substance model. It differs from the prior tutorial result in that it adds new substances and applications.

Next Steps

Tutorial 4a will add economic growth projections and business-as-usual forecasting to your multi-application model. You'll learn to model how economic expansion drives consumption changes over time, creating realistic baseline scenarios for policy comparison.

Previous: Tutorial 2 | Return to Guide Index | Next: Tutorial 4a