An industrial CO₂-capture unit rarely receives a fixed feed. The CO₂ concentration and flow rate of flue gas change with plant load, fuel composition, upstream operation, and combustion conditions. The required capture rate or CO₂ production target can also change. The absorber–stripper system must therefore operate while both the supply of CO₂ and the demand for captured CO₂ are moving.
Most solvent-screening and process-design studies simplify this problem to one steady-state condition. They select an inlet CO₂ concentration, gas flow rate, and capture target, then determine the absorber and stripper design and the operating point that minimize reboiler duty. The resulting solvent ranking is often treated as fixed. This is useful for establishing a nominal design, but it does not show whether the same solvent remains efficient when the feed and production target move away from that point.
That gap is the problem examined by this paper. A CO₂-capture solvent should not be ranked only by its minimum reboiler duty at one nominal design point. The location of that minimum can move as flue-gas composition, throughput, and capture rate change. Even when the external condition is fixed, a plant rarely holds the solvent circulation rate exactly at its calculated optimum.
The economic scale is already visible from the paper’s regeneration-energy calculation. Using its steam price of 6 USD/GJ, the nominal duties correspond to about 22.93 USD/tCO₂ for MEA and 19.97 USD/tCO₂ for MDEA/PZ 20/20. The nominal difference is therefore about 2.96 USD per tonne of captured CO₂. At 100 tCO₂/day, continuous operation for 365 days would make the difference about 108,000 USD/year in regeneration steam alone. Yet a ±10% L/G deviation raises the duty of MDEA/PZ 20/20 by roughly 8–12%. Under the same simplifying assumptions, that penalty is about 58,000–87,000 USD/year.
Three quantities therefore need to be separated:
- the minimum specific reboiler duty at the nominal condition;
- the change in the best attainable duty when the external condition changes;
- the energy penalty caused by operating away from the best liquid-to-gas ratio, or L/G.
Lim et al. compare 40 wt% MEA, 20/20 wt% MDEA/PZ, and 30/10 wt% MDEA/PZ with a rate-based Aspen Plus absorber–stripper model. MDEA/PZ 20/20 performs best on the first two measures. It has the lowest nominal duty and generally the smallest response to the tested changes in CO₂ concentration and production rate. Yet its L/G–duty curve is the sharpest near the optimum. A small circulation set-point error can erase part of its energy advantage.
That apparent contradiction is the useful engineering result: the solvent that is least sensitive to external load changes can still be the most sensitive to an internal operating decision.
The nominal ranking
The reference condition is 13 mol% CO₂, 100 tonnes per day of captured CO₂, and a 90% capture rate. The reported optima are:
| Solvent | Optimal L/G | Minimum reboiler duty |
|---|---|---|
| MEA 40 wt% | 3.695 | 3.821 GJ/tCO₂ |
| MDEA/PZ 20/20 wt% | 2.352 | 3.328 GJ/tCO₂ |
| MDEA/PZ 30/10 wt% | 2.942 | 3.591 GJ/tCO₂ |
At this point, MDEA/PZ 20/20 uses about 12.9% less regeneration energy than MEA. Its lower optimal L/G also means that the target capture rate is reached with less solvent circulation.
The result is chemically plausible. MDEA offers favorable regeneration through a bicarbonate-dominated pathway, while PZ accelerates CO₂ absorption. Raising the PZ fraction from the 30/10 blend to the 20/20 blend helps when gas-phase driving force or contact time is limited. The faster reaction and higher effective cyclic capacity reduce the solvent flow required for the same capture target, which in turn reduces the sensible heat needed to warm the circulating liquid.
This explanation has a boundary. The model does not directly include PZ precipitation, volatilization, aerosol emissions, long-term degradation, or every viscosity-related hydraulic penalty. The result establishes an energy ranking inside the modeled process, not a complete solvent-life or environmental ranking.
Why the L/G curve has a minimum
The U-shaped relation between L/G and reboiler duty is the physical core of the comparison.
At very low L/G, a small amount of solvent must carry a large CO₂ load. To maintain the specified capture rate, more CO₂ must be removed from each unit of solvent in the stripper so that the returning lean solvent has a lower CO₂ loading. This requires additional reboiler heat and stripping steam, although it does not necessarily require a higher stripper-bottom temperature.
As L/G increases, each unit of solvent carries less CO₂. Deep regeneration is no longer necessary, so reboiler duty initially falls. Beyond the optimum, however, the plant is heating and pumping more liquid than it needs. Sensible heat becomes dominant and duty rises again.
The minimum alone does not describe this curve. Its local curvature matters. The paper measures this by perturbing L/G by ±10% around the optimum. The approximate duty increases are 2–3% for MEA, 8–12% for MDEA/PZ 20/20, and 6–9% for MDEA/PZ 30/10.
MEA therefore has the highest minimum duty but the flattest neighborhood. MDEA/PZ 20/20 has the lowest minimum but the narrowest neighborhood. MDEA/PZ 30/10 lies between them. In operational terms:
- MDEA/PZ 20/20 rewards accurate L/G control;
- MEA sacrifices energy efficiency but tolerates more circulation error;
- MDEA/PZ 30/10 offers an intermediate compromise.
This is a local, one-factor sensitivity metric. Stripper pressure, lean loading, and other operating variables are held fixed rather than reoptimized together. A multivariable controller could recover some of the apparent penalty. The ±10% result should not be read as a closed-loop control test.
External variability is a different axis
The study also varies three supply-and-demand conditions: inlet CO₂ concentration, CO₂ production rate, and capture rate. These scenarios ask a different question from the L/G perturbation. They ask how the process response changes when the plant is given a different task.
For CO₂ concentrations of 11, 13, and 15 mol%, the gas flow is adjusted to maintain 100 tonnes per day of CO₂ production. MDEA/PZ 20/20 shows an approximate duty variation of 3–4%, compared with roughly 8–10% for MEA; MDEA/PZ 30/10 responds non-monotonically. The higher PZ content is helpful at low CO₂ partial pressure because fast reaction kinetics compensate for weaker gas-phase driving force.
This scenario is not a clean concentration-only experiment. Lowering CO₂ concentration while holding CO₂ production fixed requires more flue gas. The result combines a partial-pressure effect with changes in gas velocity, contact time, and column hydraulics. “Concentration–throughput coupled sensitivity” is the more exact description.
The size of this coupling is not small. Relative to 13 mol%, maintaining the same captured-CO₂ product and capture rate at 11 mol% requires about 18.2% more molar flue-gas flow; at 15 mol%, it requires about 13.3% less. This is a meaningful demand-following scenario if a downstream process requires a fixed CO₂ product rate and the capture unit can adjust the fraction of flue gas it treats. It is not, however, a generic representation of a power plant disturbance. In a single plant, load, fuel composition, excess air, and air leakage jointly determine both flue-gas flow and CO₂ concentration. A load decrease will often reduce total flue-gas flow while concentration changes less, although the exact correlation is plant-specific.
A more discriminating study would separate three cases: fixed gas flow with concentration varied, fixed concentration with gas flow varied, and joint concentration–flow trajectories taken from plant data. It would then evaluate a factorial grid or realistic joint scenarios with capture rate, inlet temperature, and production target varied together. The paper tests several parameters, but mainly one at a time or through the constant-product coupling above. That is useful steady-state scenario mapping, but it cannot identify nonlinear interactions among simultaneous disturbances or establish performance over the actual joint operating distribution.
When CO₂ production changes from 90 to 110 tonnes per day, MDEA/PZ 20/20 again has the smallest reported duty change, about 1–2%, compared with 8–10% for MEA and 3–4% for MDEA/PZ 30/10. This supports a limited claim: the 20/20 blend is relatively insensitive within the tested ±10% throughput range. It does not establish performance over the full load-following range of a power plant, and the steady-state model contains neither solvent inventory dynamics nor thermal transients.
Raising capture rate is more punishing. The duties at 80% and 90% capture are relatively close, whereas all three solvents become much more energy-intensive at 99%. The reported curves place the 99% duties at roughly 5.6 GJ/tCO₂ for MEA and 4.5 GJ/tCO₂ for both MDEA/PZ blends. Near the absorber top, the remaining gas-phase CO₂ partial pressure becomes very small. Achieving the last percentage points requires a much lower lean loading, which means removing more CO₂ from the solvent with additional reboiler heat and stripping steam.
The 99% case uses equipment designed around the 90% reference condition. It therefore estimates the penalty of pushing that equipment to 99%, not the minimum energy of a new process designed specifically for 99% capture. Packing height, heat-exchanger area, circulation capacity, pressure level, intercooling, or split-flow design could change that result.
What “robust” means here
The word “robust” in the title should be interpreted empirically, not as a mathematical guarantee. The paper maps steady-state Aspen Plus responses across selected off-design scenarios. It does not define an uncertainty set or probability distribution, solve a min–max or chance-constrained problem, examine simultaneous worst cases, or simulate a feedback controller under temporal disturbances.
A low steady-state energy sensitivity does not imply short settling time, low overshoot, freedom from actuator saturation, or closed-loop stability. Likewise, identifying the lowest point on a sampled response curve does not prove a joint global optimum over L/G, stripper pressure, column design, and every disturbance.
The most defensible reading is narrower: within the modeled equipment, tested operating ranges, and converged sensitivity cases, MDEA/PZ 20/20 maintains a low reboiler duty across several external scenarios. Near its own L/G optimum, however, its energy performance deteriorates faster than MEA’s when the circulation ratio is displaced.
Model and comparison boundaries
The Aspen Plus model uses an electrolyte thermodynamic description and rate-based columns to represent reaction-enhanced gas–liquid mass transfer. That level of detail is needed to distinguish the kinetic effect of PZ from the slower MDEA response. The model is based on Aspen examples and parameter sets validated in prior literature, but the paper does not report independent pilot-scale validation for every solvent composition, column size, and 99% off-design case studied here. It is better described as a literature- and database-based rate process model than as a plant-validated digital twin.
The equipment also differs by solvent. The reported absorber heights are 15 m for MEA, 20 m for MDEA/PZ 20/20, and 25 m for MDEA/PZ 30/10, with differences in stripper geometry as well. This is a comparison of separately sized solvent–process systems, not a same-column solvent swap. That is reasonable for comparing operating energy in new designs, but insufficient for a retrofit decision.
These equipment dimensions are fixed during the sensitivity studies. The columns are sized separately for each solvent around the nominal basis of 13 mol% CO₂, 100 tonnes per day of captured CO₂, and 90% capture, after which diameter, height, packing, and associated equipment parameters are carried into the off-design cases. L/G and stripper pressure are the main operating variables that are swept; the equipment design is not resized or co-optimized for each disturbance.
It is therefore too strong to describe the equipment as a mathematically optimal design. The paper does not formulate a joint design problem in which column dimensions, packing, heat-exchanger area, and operating policy are decision variables optimized over all scenarios or total annual cost. “Solvent-specific nominal sizing followed by off-design operating analysis” is the more accurate description. The reported solvent ranking is conditional on those fixed designs. A different column geometry or a design chosen explicitly for 99% capture or wide load-following could produce a different energy ranking and operating window.
The economic calculation mainly converts steam duty at 6 USD/GJ. It excludes column and packing capital, pump and blower electricity, solvent makeup, corrosion, emissions control, precipitation management, instrumentation, maintenance, and downtime. “Lowest regeneration-steam cost” is supported; “lowest total cost” is not.
Design implication
The practical lesson is to replace a single solvent-ranking number with a response surface. At minimum, a screening study should report the minimum duty, the optimal L/G, the local curvature around that optimum, and performance under coupled concentration and throughput changes. It should then check whether the low-energy region remains feasible under hydraulic, thermal, degradation, and control constraints.
For MDEA/PZ 20/20, that suggests pairing the solvent choice with accurate solvent-flow measurement, sufficient pump turndown, and an L/G control or real-time optimization strategy that tracks a moving optimum. For MEA, the control requirement is less sharp, but the baseline steam penalty remains. The plant decision is therefore not “which solvent has the lowest point?” It is “which solvent–equipment–control combination keeps an acceptably low duty over the conditions the plant will actually visit?”
The most efficient solvent is not necessarily the easiest solvent to operate.
Reference
Lim, S., Yun, S. H., Oh, J., Yoon, Y. I., Jang, J. T., & Park, J. (2026). Robust process design and operation for efficient CO₂ capture under variable supply-and-demand conditions. Journal of Environmental Chemical Engineering, 14, 124308. https://doi.org/10.1016/j.jece.2026.124308