Beyond the Exogenous Mask: Learning Feedback, Policy Thresholds, and Sensitive Intervention Points in Energy System Models

Energy systems are the complex networks that generate and deliver the electricity powering modern life. To plan how we move away from fossil fuels toward cleaner sources like wind and solar, scientists build computer models. These models act as virtual laboratories, testing different policies to see which ones will successfully lower carbon emissions without breaking the economy. A central question in these models is how technology gets cheaper. For decades, the standard approach has been to assume that costs drop automatically over time, as if technology improves on a fixed schedule regardless of how much of it is actually being built. This method is computationally simple, but it ignores a powerful real-world force: the idea that making and using more of a technology actually drives its price down. This phenomenon, known as learning by doing, means that every new solar panel or battery installed helps make the next one cheaper. When models ignore this feedback loop, they miss a crucial dynamic that shapes how energy transitions happen.

A new study by Nadav Mantel at Reichman University challenges the way these models are built. The research argues that by treating cost reductions as a simple, automatic function of time, current models smooth over the messy, non-linear reality of how technology adoption actually works. The paper suggests that when you include the feedback loop where deployment drives down costs, the relationship between policy and outcome changes fundamentally. Instead of a smooth curve where a little more money leads to a little more progress, the system can develop sharp tipping points. In these specific zones, a tiny change in policy can trigger a massive, self-sustaining explosion in deployment, or conversely, a tiny drop in support can cause the entire effort to collapse. The study uses a simplified computer model of the Great Britain power system to demonstrate that these tipping points exist and are invisible to the standard models used by policymakers today.

The researchers constructed a streamlined version of the British electricity grid, stripping away complex details like transmission bottlenecks or local building delays to focus purely on the interaction between cost and deployment. They ran this model in two different ways. In the first version, they used the traditional method where costs fall automatically over time, independent of how much capacity is built. In the second version, they enabled the learning-by-doing feedback, so that every unit of new capacity built immediately lowered the cost for all future units. They then tested various policy scenarios, such as carbon taxes and subsidies for solar panels and batteries, to see how the system responded. The results revealed a stark difference between the two approaches. The traditional model showed a steady, predictable increase in renewable energy as policies became more aggressive. The model with learning-by-doing, however, showed that the system could get stuck in a low-deployment trap. In this scenario, even with the same policies, solar and battery technologies failed to take off because they never reached the scale necessary to drive their own costs down.

The most striking finding was the discovery of a sensitive intervention point, a specific threshold where the system flips from failure to success. When the researchers tested subsidies for battery storage, they found that under the learning-by-doing rules, the system remained stuck with very little battery capacity as long as the subsidy was below a certain level. However, once the subsidy crossed a sharp line near £513 per kilowatt, the deployment suddenly accelerated. A change in the subsidy of less than one percent was enough to push the system across this threshold, leading to order-of-magnitude differences in how many batteries were installed. Below this line, the policy was ineffective; above it, the policy became self-reinforcing, with the falling costs making further deployment profitable without needing as much help. This sharp transition was completely absent in the traditional model, which showed a smooth, gradual increase in batteries regardless of the subsidy level.

The study also uncovered hidden relationships between different technologies that standard models miss. When the researchers looked at solar power and battery storage together, they found that these two technologies are deeply linked in a way that creates a barrier to entry. In the learning-by-doing model, solar and batteries were stuck in a low-deployment state because neither could become cheap enough on its own to compete without the other. Solar needs batteries to store its energy, and batteries need cheap solar to provide the surplus energy they store. The traditional model, which treats costs as independent, failed to see this lock-in. The new approach showed that a coordinated subsidy for both solar and batteries could break this deadlock. By supporting both simultaneously, the total cost of the intervention could be reduced by about six percent compared to subsidizing batteries alone, because the two technologies helped each other cross the cost threshold.

Furthermore, the research highlighted the limitations of relying solely on carbon pricing to launch new technologies. In the model that accounted for learning-by-doing, a high carbon price alone was not enough to get emerging technologies like batteries off the ground. The price of carbon would need to be set at an extremely high level, far beyond what is typically considered politically feasible, to trigger the initial deployment needed to start the cost-reduction cycle. This suggests that carbon pricing might be a poor tool for initiating the very first steps of a transition for technologies that have not yet descended their learning curves. The study implies that direct subsidies are often necessary to push a technology over the initial hump, after which the market can take over.

The author emphasizes that their work is not a prediction of exactly how much money should be spent or exactly which technologies will win. The specific numbers, such as the £513 threshold, are results of a simplified model and would shift if real-world factors like construction limits or local market conditions were added. However, the core insight is robust: the existence of these tipping points and the non-linear nature of the transition are real features of the system that current models ignore. The study proposes a new diagnostic tool for policymakers. Before committing to a major policy, they could run a simplified version of their model that includes learning-by-doing to check if their proposed policy sits near a tipping point. If it does, the policy is highly sensitive, and a small miscalculation could lead to failure. If it is far from a tipping point, the policy is likely more robust.

This approach offers a way to avoid two common pitfalls. The first is false sufficiency, where a policy looks good in a standard model because it seems to just barely meet the targets, but in reality, it sits below the tipping point and results in almost no deployment. The second is false redundancy, where a policy looks wasteful because it seems to be spending far more than necessary, when in fact it is only just above the threshold and is essential for keeping the system moving forward. By identifying these sensitive zones, the method helps regulators understand that the intensity of early support may matter more than its duration. A constant subsidy set just below the threshold will fail indefinitely, while a slightly higher push can unlock a self-sustaining transition.

The research concludes that the choice of how to model technology costs is not just a technical detail but a decision that shapes policy outcomes. By smoothing over the non-linear dynamics of learning, traditional models may be giving a false sense of security about how easily the energy transition can be managed. The new method does not replace the complex, detailed models used for long-term planning but serves as a necessary check. It acts as a pre-deployment screen to flag where the system is fragile and where small changes in policy can lead to disproportionately large results. In a world where the stakes of the energy transition are incredibly high, understanding these hidden thresholds is essential for designing policies that actually work. The study suggests that with the right tools to identify these sensitive intervention points, it may be possible to accelerate the shift to clean energy while reducing the total public expenditure required to get there.