Jorge Quitério Search
Technology

Figures of merit and their theoretical limits

A figure of merit is meaningful only in relation to its theoretical limit. A simple calculation for melting scrap steel shows how to derive that limit, and a calculator shows how much improvement remains.

Claims that a technology is "ten times more efficient" appear regularly. An effective way to assess such a claim is to ask what physics allows: what is the best performance possible at all?

This approach is one of the most useful lessons I took from Olivier de Weck's MIT programme on Management of Technology: Roadmapping & Development. Every technology can be described by a few figures of merit, quantities that capture how well it does its job: energy per tonne, bits per second, kilometres per charge. Those figures improve over time, but physics sets a floor or a ceiling they can never cross. The distance between a technology and that limit indicates how much progress remains possible and whether a claimed breakthrough is plausible.

This article applies the question to one technology: how close is a real electric arc furnace to the thermodynamic minimum for melting steel?

The same reasoning works in a kitchen. Bringing one litre of water from 15 °C to the boil requires about 4.19 kJ per kilogram and degree over 85 degrees, roughly 356 kJ or 0.099 kWh. A kettle that uses 0.12 kWh for the task operates at about 82% of the ideal, so no new kettle, however it is marketed, can reduce that energy use by more than about 18%. The steel calculation below follows the same steps at industrial scale.

De Weck's Technology Roadmapping and Development (2022) sets out figures of merit (FOMs) and contrasts them with business KPIs: a KPI describes how an organisation performs, a FOM how a technology performs, in units physics can reason about. The classic example of a hard physical limit is the Shockley-Queisser analysis (1961), which caps the efficiency of a single-junction solar cell at about 33%. Foster's Innovation: The Attacker's Advantage (1986) popularised the S-curve picture of technologies slowing as they approach their limits, although de Weck's course cautions that empirical evidence for ubiquitous performance S-curves is thinner than textbooks suggest.

Figures of merit and physical limits

Market share or customer satisfaction cannot be compared with a law of nature. Kilowatt-hours per tonne can. This is the main advantage of a FOM: it has a theoretical limit to be measured against.

Example: melting scrap steel

Electric arc furnaces melt scrap steel to make new steel. Melting a mass mm from room temperature requires heating it to its melting point and then supplying the latent heat of fusion:

Q=m c ΔT+m L. Q = m\,c\,\Delta T + m\,L .

With a specific heat of c=0.466c = 0.466 J/(g·°C), heating from 24 °C to 1500 °C (ΔT=1476\Delta T = 1476 °C) and a latent heat of L=276L = 276 J/g:

Qm=0.466×1476+276=687.8+276=963.8 J/g. \frac{Q}{m} = 0.466 \times 1476 + 276 = 687.8 + 276 = 963.8 \ \text{J/g}.

That is 963.8 MJ per tonne, or about 268 kWh per tonne: the thermodynamic floor. No furnace, whatever its design, can melt steel with less. The calculation is simplified: the specific heat of steel changes with temperature, steel melts over a range rather than at a point, and real melts are heated beyond the melting point before tapping, so the true floor is somewhat higher. The order of magnitude is sufficient for this purpose.

Interpreting the gap

The course material puts typical arc furnaces of the 2000s at around 400 kWh per tonne, so

268400≈0.67: \frac{268}{400} \approx 0.67 :

those furnaces operated at about two thirds of the theoretical ideal. Roughly a third of the energy went to losses: heat through walls and roof, off-gases, electrical losses, slag. This shows an engineer where improvement is possible (in the losses, not the melting) and gives a manager a realistic expectation of progress. A technology at 67% of its limit can still improve meaningfully. One at 95% cannot, however much is invested.

Approaching the limit

Technologies near their limits tend to improve more slowly; each further gain costs more. Solar cells illustrate this: laboratory silicon cells have passed 26% against the 33% Shockley-Queisser ceiling for a single junction, so research has moved to multi-junction cells with a higher ceiling. Where a technology sits on the way to its limit has to be measured, not assumed.

The calculator below shows that the floor follows from two lines of physics, and the share of the ideal achieved is simply floor ÷ actual. Any claim below the floor is impossible.

  1. Reference case. With steel, no superheat and 400 kWh/t, the readouts show 268 kWh/t and 67%.
  2. Superheat. Add 100 °C of superheat, typical before tapping. The floor rises and the furnace looks a little better than the simple calculation suggested.
  3. Physical impossibility. Drag the actual consumption below the floor. The readout flags this, because it would require energy to be created.
  4. Other metals. Switch to aluminium or copper. Aluminium melts far lower than steel but has a high specific heat and latent heat, so its floor is almost the same as steel's. Copper's is lower. Intuition about melting temperature is therefore not a substitute for the calculation.

Unit conversion. Energy per gram is c ΔT+Lc\,\Delta T + L in J/g. One tonne is 10610^6 g and 1 kWh is 3.6×1063.6 \times 10^6 J, so J/g divided by 3.6 gives kWh per tonne: 963.8 / 3.6 ≈ 268.

When a breakthrough is claimed, I ask for the figure of merit, its theoretical limit and today's value. A few lines of physics often say more about how much room is left than a market report does, and they immediately exclude any claim that would violate thermodynamics.

References

  1. W. Shockley and H. J. Queisser, "Detailed balance limit of efficiency of p-n junction solar cells", Journal of Applied Physics, 32(3), 510-519, 1961.
  2. R. N. Foster, Innovation: The Attacker's Advantage, Summit Books, 1986.
  3. O. L. de Weck, Technology Roadmapping and Development: A Quantitative Approach to the Management of Technology, Springer, 2022.