Systemic Carbon Lock-In Dynamics and Optimal Sustainable Reduction Pathways for a Just Industrial Transition in South Africa

Abstract

South Africa’s manufacturing sector, a driving force for sustainable development, faces a profound challenge in decarbonizing without deindustrializing. This study provides an optimized, scenario-based assessment of the sector explicitly aligned with its Just Energy Transition Partnership (JETP) objectives. A novel framework is applied, integrating an extended Kaya–Logarithmic Mean Divisia Index (Kaya–LMDI) decomposition with scenario forecasting and Genetic Algorithm (GA) optimization. The decomposition disaggregates a conventional carbon intensity (CI) driver to include Electrification Share (ELE), Renewable Share (REN), and a newly defined Residual Carbon Factor (RCF) that captures direct fossil fuel use for industrial process heat. Historical analysis (2002–2022) shows that emissions growth was primarily driven by the RCF (224.1 MtCO₂, 160%) and Economic Activity (187.5 MtCO₂, 134%), partly offset by gains in Energy Intensity (−141.8 MtCO₂, 101.35%) and REN (−202.2 MtCO₂, −144.53%). Carbon emissions projections to 2040 reveal a critical sustainability trilemma: the Just Transition accelerated scenario (JTAS), despite achieving rapid renewable deployment, increases emissions by 469% as economic growth overwhelms decarbonization efforts. Conversely, the mathematically optimal (GA) pathway achieves a 90.8% reduction but only through structural contraction that implies socially unsustainable deindustrialization. This tension exposes the systemic limits of incremental decarbonization and underscores that a truly sustainable pathway requires transcending this binary choice by directly addressing the fossil fuel substrate of industrial production.

1. Introduction

Energy remains the backbone of economic progression, but sustained growth has long relied on fossil fuels, driving CO₂ emissions and intensifying climate risks [1]. Decoupling economic growth from emissions is now a pressing global agenda [2], especially for emerging economies such as South Africa, where industrialization, developmental aspirations, and energy transitions converge. As the largest emitter on the African continent, South Africa’s economy is still heavily dependent on coal, which provides 77% of primary energy and 85% of electricity [3].

In response, the country has committed to ambitious international climate agreements, culminating in the landmark Just Energy Transition Partnership (JETP). Through the JETP and its Investment Plan (JET-IP), South Africa aims to align climate action with economic revitalization, driving green industrialization while reducing fossil dependence [4]. Achieving these dual objectives hinges on accelerated progress across three decarbonization pillars: expanding renewable energy, improving energy efficiency, and electrifying industrial processes. However, the long-term impact of these interventions on the manufacturing sector’s emissions trajectory remains highly uncertain.

A closer examination of the literature within South Africa and globally reveals three significant and interconnected gaps that this study aims to address, all of which are central to the challenge of achieving sustainable industrial development.

Globally, decomposition analyses involving Logarithmic Mean Divisia Index (LMDI) combined with scenario forecasting have become standard tools for understanding emission drivers and projecting future pathways within sustainability science [5,6,7,8,9,10]. Similarly, Genetic Algorithms (GAs) are well-established for solving complex optimization problems in sustainable energy systems [11,12,13,14,15,16,17]. However, a systematic review highlights a critical methodological disconnect: these components are rarely integrated into a unified framework capable of navigating the sustainability trilemma. For instance, studies such as Zhang et al. [9] and Jiang et al. [10] merged decomposition with scenario projection, but fall short of identifying a mathematically optimal pathway within a realistic policy space, leaving trade-offs between growth and decarbonization insufficiently explored. Conversely, GA-based studies often prioritize techno-economic optimization of energy systems [12,16] without grounding their models in a historical, driver-level decomposition of sectoral emissions, overlooking the socio-technical inertia fundamental to sustainability transitions. This fragmentation limits the development of policy benchmarks that are simultaneously evidence-based, optimized for environmental performance, and aligned with socio-economic constraints.

Within the South African context, previous decomposition studies [18,19,20,21,22,23,24] have been invaluable, consistently identifying economic activity as the primary driver of emissions and energy intensity as the key mitigator. However, these studies exhibit two overarching limitations from a sustainable transition perspective. First, they were conducted before the landmark JETP, a framework explicitly designed to align climate action with socio-economic development, and thus do not incorporate its transformative policy assumptions, particularly the accelerated drive for industrial electrification and a decarbonized grid. Second, and more crucial, their analytical frameworks rely on conventional decomposition identities that overlook a crucial source of industrial emissions: direct fossil fuel consumption for process heat. By disaggregating this factor into an aggregate carbon intensity term, existing studies obscure the specifics and persistent “carbon lock-in” embedded in industrial thermal processes. This omission represents a significant blind spot for sustainable industrial policy, as the decarbonization of industrial heat remains one of the most challenging and unresolved issues in the global sustainability literature, and a major barrier to achieving net-zero ambitions for hard-to-abate sectors [11,25].

Consequently, a critical knowledge gap persists at the intersection of methodological innovation and sustainable policy design. No study has yet applied an integrated LMDI-GA framework to produce a quantitative, optimized assessment of an industrial sector that (i) is explicitly aligned with the sustainable development objectives of a major international JETP and (ii) separately identifies fossil-based process heat as a distinct driver of persistent carbon lock-in. This study fills these gaps by introducing a novel framework and the Residual Carbon Factor (RCF) metric, providing a replicable approach for diagnosing and overcoming carbon lock-in in industrializing economies.

Relative to existing literature, this study makes three core contributions to both South Africa and global discourse on sustainable industrial decarbonization:

First, it presents the first scenario-based, optimized sustainability assessment of an industrial sector explicitly linked to a major international JETP. To our knowledge, it is the first study to (i) project long-term manufacturing emissions under JET pathways and (ii) demonstrate quantitatively that a narrow reliance on renewables and efficiency; the core pillars of “green growth”, is insufficient for sustainability, as these gains are overtaken by entrenched carbon inertia. This finding challenges the feasibility of genuine decoupling without confronting the fundamental unsustainability of the industrial energy substrate [26,27].

Second, it bridges technical decarbonization modelling with the social dimensions of sustainability by introducing the RCF, a novel metric that directly quantifies the lock-in of unsustainable production patterns. By decomposing and foregrounding this previously conflated driver, the study moves beyond the limitations of conventional analyses and exposes the systemic, substrate-level barrier to sustainable transition that is often overshadowed by power-sector-centric strategies. This metric provides a precise target for the deep, structural transformations required to achieve long-term environmental sustainability.

Third, it advances the field methodologically by shifting from diagnosis to sustainable pathway recalibration. The study integrates an extended Kaya–LMDI decomposition, scenario forecasting, and GA optimization into a unified framework that explicitly quantifies the trade-offs and synergies between environmental integrity and socio-economic development. The results demonstrate that a truly sustainable pathway requires mandatory fuel switching and structural transformation, rather than incremental efficiency gains alone, to achieve the absolute decoupling necessary for operating within planetary boundaries. Collectively, these contributions establish both a methodological advancement for sustainability science and a policy-relevant benchmark for redirecting South Africa’s manufacturing sector toward a genuinely sustainable and just future under the JETP.

The rest of the paper is organized as follows. Section 2 present the literature review. Section 3 introduces the materials and methods. Section 4 explains and analyzes the results. Section 5 provides conclusion and suggestions and research limitations and future directions.

2. Literature Review

2.1. Energy-Related Emission Accounting

Globally, advances in emission accounting and decarbonization modelling provide a strong analytical foundation. Index Decomposition Analysis (IDA), particularly the LMDI, has become a standard approach for attributing emission changes to drivers such as economic activity, energy intensity, and carbon intensity [5,6,7,8,9,10]. In parallel, emission reduction pathways are widely applied to examine how economic growth can be decoupled from environmental pressures. Integrating decomposition analysis with pathway modelling, as demonstrated in several international studies, offers critical insights into both the mechanisms and the limits of emission–economic decoupling [9,10].

Table 1. Summary of South African related studies analyzing energy-related CO₂ emissions at different scale.

Author	Region/Sector	Method	Data Span	Key Findings and Suggestions
Kana et al. [18]	South Africa/ Manufacturing	Index decomposition analysis (IDA) and historical trends analysis	1999–2015	Highlighted energy consumption patterns in the manufacturing sector; identified process inefficiencies and opportunities for targeted energy savings
Inglesi-Lotz [19]	South Africa/ National level	LMDI decomposition	1990–2014	Decomposed CO2 emissions in a BRICS context; noted potential rebound effects and emphasized the importance of energy efficiency in mitigating emissions
Olanrewaju [20]	South Africa/ Industrial	LMDI decomposition	1970–2016	Activity significantly promoted industrial energy consumption increase and emphasized the role of energy efficiency and structural changes in reducing emissions.
Zhang et al. [21]	South Africa	LMDI decomposition	1993–2011	Identified energy intensity and structural factors as key drivers of CO2 emissions.
Lin et al. [22]	South Africa	OECD decoupling + Kaya identity	1990–2012	South Africa achieved weak to strong decoupling of CO2 emissions from economic growth, with energy efficiency and renewable energy adoption identified as critical drivers
Beidari et al. [23]	South Africa	LMDI	1990–2013	Improvement of the efficiency of existing electricity power generation plants and expanding more of their renewable energy sources (nuclear included).
Inglesi-Lotz and Pouris [24]	South Africa	Index decomposition analysis (IDA) and historical trends analysis	1993–2006	The energy intensity of energy use has contributed to the declining trend in energy efficiency. The need to implement differentiated pricing policies to support a more effective energy efficiency strategy is crucial

Source: Authors.

summarizes South Africa–specific decomposition and decoupling studies across different scales. The varying time ranges in Table 1 reflect the respective studies and the data availability at those times. This heterogeneity underscores why the current research provides a unique and updated analysis covering the most recent two-decade period (2002–2022), which includes critical recent developments such as the inception of the REIPPPP and the policy context of the JETP.

To move beyond descriptive insights, predictive methods have been used to forecast CO₂ emissions, enabling structured comparisons of alternative pathways under varying policy assumptions [28,29,30]. Although real-world energy systems are complex and cannot be fully captured by a single model, mathematical models offer an objective and quantifiable framework for analyzing emissions within this complexity [31]. However, while valuable for framing policy options, its limitation lies in examining only a set of predefined trajectories without exploring the broader solution space for optimal outcomes. Addressing this gap requires more flexible approaches. GA is particularly suited to this challenge, as they can efficiently navigate complex, non-linear, and constrained search spaces to identify near-optimal solutions where traditional analytical methods fall short [11,12].

Genetic Algorithms (GAs)

GAs, introduced by John Holland at the University of Michigan, are global optimization methods inspired by natural selection and genetic processes [13]. They simulate mechanisms such as selection, crossover, and mutation to identify optimal solutions within complex search spaces. GA is robust, stochastic, and well-suited for parallel processing, making it highly effective for solving multidimensional, non-linear optimization problems. By generating a population of candidate solutions and iteratively evaluating them through a fitness function, GA converge toward solutions that best satisfy predefined objectives [14,15]. In the context of CO₂ emissions modelling, GAs can identify the optimal combination of driver growth rates that minimizes cumulative emissions while respecting realistic socio-technical constraints. Their application to environmental and energy policy is well established, where conflicting objectives such as cost, energy efficiency, and environmental impact, offering a structured pathway toward deep decarbonization [16,17]. Despite their proven effectiveness in environmental and energy policy applications, the use of GAs for guiding decision-making in optimal emission reduction scenarios remains relatively underexplored, highlighting an important inclusion in this study.

2.2. The Challenge of Sustainable Energy Transitions

The transition from fossil fuels is a defining 21st-century challenge, driven by climate urgency and the unsustainability of current socio-economic systems [32]. While its necessity is clear, the pace, direction, and political feasibility remain contested [33]. For developing nations, the core tension lies in balancing growth and equity with climate goals, as emphasized by the Paris Agreement and global frameworks such as the SDGs. This review synthesizes these perspectives to explore the trade-offs shaping sustainable energy pathways.

2.2.1. The Empirical Theoretical Limits of Green-Growth

The dominant policy narrative to resolve the growth–climate tension is “green growth”, which holds that technological innovation and efficiency gains can deliver absolute decoupling of GDP from environmental pressures, enabling continued expansion within planetary boundaries [26,34]. Notwithstanding, the empirical record offers little support; evidence shows that while relative decoupling is common, emissions rising more slowly than GDP, absolute decoupling is rare, short-lived, geographically limited, and far too weak to meet Paris targets [27,35]. Since 2000, the global economy has even re-materialized, with resource use once again outpacing GDP growth. Even in the best-performing countries, observed CO₂ reductions fall an order of magnitude short of Paris-compatible pathways [35].

As a result, mainstream mitigation scenarios that maintain growth increasingly depend on speculative negative-emission technologies to close the carbon budget later this century [27,36]. This reliance exposes a widening gap between policy ambition and empirical reality, leading critics to label green growth as an “irresponsible bet” [26] and a “misguided objective” [27]. The shortfall is not simply due to weak policy but reflects structural and biophysical barriers like rising energy costs of resource extraction [37]; rebound effects, where efficiency gains spur greater consumption [11]; problem-shifting that creates new environmental burdens; and the material intensity underpinning even service-based economies. For essential, non-substitutable resources such as energy and raw materials, many scholars argue that permanent absolute decoupling is physically impossible due to thermodynamic constraints [26,34].

2.2.2. The Rebound Effect (Jevons Paradox), Demand-Side Solutions, and the Degrowth Imperative

A key barrier to absolute decoupling is the rebound effect, or Jevons Paradox, where efficiency gains are offset by higher consumption [31]. Efficiency lowers costs, which can lead to greater use of the same service (direct rebound), spending on other energy-intensive goods (indirect rebound), or economy-wide growth that raises overall demand. Evidence shows these rebounds are often above 50%, meaning efficiency alone cannot deliver absolute reductions, especially in green-growth models where savings flow back into carbon-intensive activities. This makes limits on consumption essential.

With supply-side measures constrained, attention is shifting to demand-side strategies that reduce energy and material use while maintaining well-being [11]. The Avoid–Shift–Improve (ASI) framework structures these: Avoid cuts demand entirely (e.g., compact cities, telework); Shift replaces high-impact practices with lower-impact ones (public transport over cars, plant-based diets); Improve boosts efficiency within existing systems (appliances, building retrofits). The IPCC estimates such measures could cut global emissions 40–70% by 2050. A more radical alternative, degrowth, questions GDP growth itself. Defined as an “equitable downscaling of throughput with a concomitant securing of wellbeing” [36], it argues that in high-income countries, growth is incompatible with ecological limits, requiring planned reductions in production and consumption [27]. Degrowth scenarios consistent with 1.5 °C show lower feasibility risks than technology-heavy IPCC pathways, avoiding reliance on extreme decoupling, large-scale carbon removal, or rapid renewable expansion [36]. While politically contested, degrowth highlights the need to redefine prosperity beyond growth, focusing instead on sufficiency and well-being within planetary boundaries [37].

2.2.3. The Political Economy of Carbon Lock-In and the Post-Growth Imperative

Transitioning to a low-carbon society is not just a technical or economic challenge; it is fundamentally political, involving conflicts over who bears costs and who benefits. The persistence of high-carbon systems is explained by carbon lock-in, a path-dependent process reinforced by interacting infrastructural, institutional, and behavioural factors that generate strong inertia [33,38]. Powerful incumbent actors, fossil fuel companies, utilities, and allied unions actively defend the status quo, using instrumental power (lobbying), material power (sunk investments), and discursive power (framing debates around “energy security” or “clean coal”) [39]. These actors often form a cohesive “techno-institutional complex” with policymakers, creating alliances that protect existing socio-technical regimes. A clear example is Poland’s coal sector, where the government, state-owned enterprises, and unions have collectively resisted phase-out efforts [38]. Thus, the central challenge of decarbonization is twofold: to support green niche innovations while simultaneously destabilizing entrenched fossil fuel regimes [25].

Fundamentally, the persistence of carbon-intensive manufacturing in South Africa occurs against a backdrop of broader sectoral fragility. As shown in

Table A1. Contextual Challenges in South African Manufacturing (2020–2022) [57].

Indicator	Value	Implication for JETP
Sector GDP Contribution	13% (down from pre-pandemic)	Limited fiscal space for transition
Employment Trend	1.4 million (down from 1.8 m in 2019)	Job preservation concerns paramount
Steel Capacity Utilization	23.5% (February 2022)	Carbon-intensive sectors already distressed
Key Constraints	Energy reliability, high costs, skills gaps	Reinforces carbon lock-in without intervention
Investment Impact	8% employment from 10% investment increase	Shows potential with right policy mix

, key sub-sectors face multiple constraints including energy reliability issues, high operational costs, and under-utilization of capacity (e.g., steel at 23.5% capacity utilization), creating a context where environmental considerations risk being deprioritized without careful policy design.

3. Materials and Methods

This study employs an integrated analytical framework, depicted in Figure 1, which combines historical decomposition analysis, scenario forecasting, and GA-optimization to assess decarbonization pathways for South Africa’s manufacturing sector. The methodology consists of four main components: (1) Historical energy consumption data collection and emission estimation, (2) an extended Kaya-LMDI decomposition analysis to identify historical drivers, (3) scenario definition and projection modelling, and (4) GA-optimization to identify a benchmark optimal pathway.

3.1. Emission Estimation and Formulation of Distinctive Indicators

According to IPCC [40] guidelines, the manufacturing carbon emissions were derived by converting consecutive national aggregated energy balance data [41] into CO₂ equivalents, using the emission factors listed in

Table 2. South Africa–specific and Default IPCC Emission Factors.

Energy Source	Emission Factor	Carbon Content (kg CO2/TJ)	Conversion Factor (tCO2/TJ)
Coal	2.2988 kgCO2/kg	94,600.82	94.601
Petroleum	2.2637 kgCO2/L	66,190.06	66.190
Natural Gas	2.0086 kgCO2/m3	48,990.24	48.990
Electricity (purchased)	0.9264 kgCO2/kWh	257,333.33	257.333
Renewables	0	0.0	Neutral

Source: [18,40].

, as shown in Equation (1).

$$\mathrm{C}{\mathrm{O}}_2^t=\underset{Energy-C{O}_2 Accounting Model}{\underbrace{\sum _{i\in fuels}\left(TE{U}_{it}· E{F}_i\right)+\left(EL{E}_t· E{F}_t^{grid} \right)}}$$

(1)

where ${\mathrm{C}\mathrm{O}}_2^{\mathrm{t}}$ denotes emissions at time $t$ (MtCO₂). $TE{U}_{it}$ represents the total energy use (TJ) of fuel $i$, at the time $t$; ${EF}_i$ is the emission factor (tCO₂/TJ). Non-fossil energy sources (i.e., renewables) are assumed to be carbon-neutral. This formulation ensures sectoral consistency and explicitly distinguishes fossil and non-fossil sources. Thereby, capturing electrification effects and renewable substitution.

The Kaya identity defines the mathematical relationships between factors such as population, energy use, economic activity, and CO₂ emissions [42]. Its straightforward structure and ease of application make it a powerful tool for structurally decomposing CO₂ emissions and for exploring the links between emissions and their driving factors. The Kaya identity is adapted to analyze manufacturing sector emissions through structural decomposition analysis. This approach separates economy-wide effects from sector-specific dynamics as shown in Equation (2):

$$C_t=GD{P}_t·{\displaystyle \frac{MV{A}_t}{GD{P}_t}}·{\displaystyle \frac{TE{U}_t}{MV{A}_t}}·{\displaystyle \frac{C{O}_2{,}_t}{TE{U}_t}}$$

(2)

These expresses manufacturing CO₂ emissions as the product of economic scale, structural importance of manufacturing, sectoral energy intensity, and aggregate carbon intensity.

To explicitly capture energy-transition dynamics, the aggregate carbon-intensity term is decomposed into electricity share (ELE), renewable share (REN), and a residual carbon factor (RCF). Total manufacturing energy use is partitioned into purchased electricity $\left(E{L}_t\right)$ and other energy carriers, while purchased electricity is further divided into renewable ($R_t$) and non-renewable sources. Unlike conventional decompositions, this framework treats ELE, REN, and RCF as independent drivers while preserving an exact algebraic identity: the product of all factors precisely equals the aggregate carbon intensity $({CO}_{2,t}/TE{U}_t)$, ensuring that each contribution is fully attributable to a distinct driver without residual error. This provides a rigorous and interpretable decomposition that explicitly isolates the effects of energy transition.

Using this structure, carbon intensity becomes:

$$C{I}_t={\displaystyle \frac{C{O}_{2,t}}{TE{U}_t}}={\displaystyle \frac{E{L}_t}{\underset{EL{E}_t}{\underbrace{TE{U}_t}}}}·{\displaystyle \frac{R_t}{\underset{RE{N}_t}{\underbrace{E{L}_t}}}}·{\displaystyle \frac{C{O}_{2,t}}{\underset{RC{F}_t}{\underbrace{R_t}}}}$$

(3)

This yields an exact identity for carbon intensity:

$$C{I}_t=EL{E}_t·RE{N}_t·RC{F}_t$$

(4)

Notably, the RCF captures the portion of emissions that remains after accounting for electrification and renewables integration. It reflects carbon intensity arising from non-renewable electricity and direct fossil-fuel combustion.

Table 3. LMDI formula for decomposing CO₂ emissions determinants in the manufacturing sector.

Notation	Driver	Structure	Definition
Ct	CO2 emissions	CO2	Total CO2 emissions from manufacturing
EA	Economic activity	$\Delta C_{EA}=W·In\left({\displaystyle \frac{GD{P}_{t+1}}{GP{D}_t}}\right)$	The scale of the overall economy
ST	Structural shift	$\Delta C_{ST}=W·In\left({\displaystyle \frac{{\left(\frac{MVA}{GDP}\right)}_{t+1}}{{\left(\frac{MVA}{GDP}\right)}_t}}\right)$	The economic importance of the manufacturing sector relative to the whole economy
EI	Energy intensity	$\Delta C_{EI}=W·In\left({\displaystyle \frac{{\left(\frac{TEU}{MVA}\right)}_{t+1}}{{\left(\frac{TEU}{MVA}\right)}_t}}\right)$	The energy required per unit of manufacturing value added
ELE	Electrification share	$\Delta C_{ELE}=W·In\left({\displaystyle \frac{{\left(\frac{EL}{TEU}\right)}_{t+1}}{{\left(\frac{EL}{TEU}\right)}_t}}\right)$	The proportion of total energy used by purchased electricity.
REN	Renewable share	$\Delta C_{REN}=W·In\left({\displaystyle \frac{{\left(\frac{R}{EL}\right)}_{t+1}}{{\left(\frac{R}{EL}\right)}_t}}\right)$	The proportion of electricity consumption derived from renewable sources
RCF	Residual CO2 factor	$\Delta C_{RCF}=W·In\left({\displaystyle \frac{{\left(\frac{C{O}_2}{R}\right)}_{t+1}}{{\left(\frac{C{O}_2}{R}\right)}_t}}\right)$	The proportion of the energy system’s carbon intensity that remains unaffected by electrification and renewable integration. Practically, it reflects the combined carbon intensity of non-renewable electricity and direct fossil fuel combustion in industrial processes, capturing the residual emissions not yet addressed by decarbonization efforts.

Source: Authors.

details the corresponding LMDI formulas for each factor and their interpretation. This decomposition strategy uses GDP as the final economic activity term rather than Manufacturing Value Added (MVA). This approach isolates the structural shift effect (ST = MVA/GDP) as a distinct driver, allowing changes in manufacturing emissions to be separated into Economic scale effects, Structural shift effects, and Sector-specific efficiency effects. The decomposition analysis, however, requires this structural separation to identify economy-wide versus sector-specific drivers.

The LMDI quantifies the contribution of each driver to emission changes between two periods $\left(\mathrm{t}\to \mathrm{t}+1\right)$. Thus, $\Delta \mathrm{C}$ can be decomposed into the following determinant factors:

$$\Delta \mathrm{C} = {\mathrm{C}}_{\mathrm{t}+1}- {\mathrm{C}}_t= \Delta {\mathrm{C}}_{\mathrm{EA}}+ \Delta {\mathrm{C}}_{\mathrm{ST}}+ \Delta {\mathrm{C}}_{\mathrm{EI}}+ \Delta {\mathrm{C}}_{\mathrm{ELE}}+ \Delta {\mathrm{C}}_{\mathrm{REN}}+ \Delta {\mathrm{C}}_{\mathrm{RCF}}$$

(5)

Here, ΔC represents the total change in carbon emissions, which can be further broken down into specific contributing indicators. Accordingly, the contribution of each factor $\Delta C_x$ in Equation (5) is calculated using the standard LMDI formula [43], where for a driver $x$, its contribution is:

$$\Delta C_x={\displaystyle \sum _{i=5}}W\left(C_{t+1},C_t\right)· In\left({\displaystyle \frac{x_{t+1}}{x_t}}\right)$$

(6)

where $i=5,$ denotes the energy mixes (coal, purchased electricity, natural gas, petroleum and renewables); and $W\left(C_{\mathrm{t}+1,}{C}_t\right)$ is the logarithmic mean of $C_{\mathrm{t}+1}$ and $C_t$, defined as $W\left(a,b\right)={\displaystyle \frac{\left(a-b\right)}{\underset{Log \mathrm{mean} \mathrm{weight}}{\underbrace{In\left(a-b\right)}}}}$.

A comprehensive nomenclature of the mathematical symbols and parameters used in the decomposition analysis and genetic algorithm optimization is provided in

Table A2. Nomenclature of mathematical symbols and parameters.

Symbol	Description	Unit/Domain
$t$	Time index (a specific year)	Year
$t_0$	Base year for projections and decomposition (2022)	Year
$T$	Projection period length (18 years, 2023–2040)	Year
$\Delta$	Change in a variable between two periods	-
$W(\cdot )$	Logarithmic mean weight function for LMDI	-
i	Index for energy fuel types (coal, petroleum, natural gas, electricity, renewables)
Emission and Energy Variables
$C_t$	Total $C{O}_2$ emissions from the manufacturing sector	Mt$C{O}_2$
$C{O}_2$	Total $C{O}_2$ emissions (accounting model)	Mt$C{O}_2$
$TE{U}_t$	Total Energy Use by the Manufacturing Sector	TJ
$TE{U}_{it}$	Total Energy Use of fuel $i$ at time $t$	TJ
$E{F}_i$	Emission factor for fuel $i$	t$C{O}_2$/TJ
$E{F}_t^{grid}$	Emission factor for grid electricity at time $t$	t$C{O}_2$/TJ
Kaya-LMDI Drivers (Levels and Changes)
$E{A}_t$	Economic Activity (Gross Domestic Product, GDP)	ZAR (constant)
$S{T}_t$	Structural Shift (Manufacturing Value Added/GDP)	Dimensionless
$E{I}_t$	Energy Intensity (Total Energy Use/MVA)	TJ/ZAR
$EL{E}_t$	Electrification Share (Share of electricity in total final energy)	Dimensionless
$RE{N}_t$	Renewable Share (Share of renewables in electricity consumption)	Dimensionless
$RC{F}_t$	Residual Carbon Factor (Carbon intensity from direct fossil fuel use)	t$C{O}_2$/TJ
$\Delta C_{EA}$	Contribution of Economic Activity to emission change	Mt$C{O}_2$
$\Delta C_{ST}$	Contribution of Structural Shift to emission change	Mt$C{O}_2$
$\Delta C_{EI}$	Contribution of Energy Intensity to emission change	Mt$C{O}_2$
$\Delta C_{ELE}$	Contribution of Electrification Share to emission change	Mt$C{O}_2$
$\Delta C_{REN}$	Contribution of Renewable Share to emission change	Mt$C{O}_2$
$\Delta C_{RCF}$	Contribution of the Residual Carbon Factor to emission change	Mt$C{O}_2$
Scenario Projection Model
$g_x$	Vector of constant annual driver growth rates $(g_{EA},g_{ST}, g_{EI},g_{ELE}, g_{REN}, g_{RCF})$	%/year
$g_{EA}$	Annual growth rate of Economic Activity ($\Delta \alpha$)	%/year
$g_{ST}$	Annual growth rate of Structural Shift ($\Delta \beta$)	%/year
$g_{EI}$	Annual growth rate of Energy Intensity ($\Delta \gamma$)	%/year
$g_{ELE}$	Annual growth rate of Electrification Share ($\Delta \delta$)	%/year
$g_{REN}$	Annual growth rate of Renewable Share ($\Delta \tau$)	%/year
$g_{RCF}$	Annual growth rate of Residual Carbon Factor ($\Delta \phi$)	%/year
Genetic Algorithm (GA) Optimization
$\mathrm{g}$	Optimal vector of growth rates from GA	%/year
${\mathrm{g}}_x$	Vector of constant annual driver growth rates, $(g_{EA},g_{ST},g_{EI},g_{ELE},g_{REN},g_{RCF})$	%/year
$J(\mathrm{g})$	Objective function (cumulative emissions, 2023–2040)	M$C{O}_2$
$\mathrm{Z}$	Matrix of time-varying annual growth rates (detailed formulation)	%/year
$\mathrm{L},\mathrm{U}$	Lower and upper bound matrices/vectors for growth rates	%/year
$N$	GA population size	-
$p_c$	GA crossover probability	-
$p_m$	GA mutation probability (per gene)	-
$e$	Fraction of population preserved via elitism (e.g., 0.05)	-
${ϵ}_{ST},{ϵ}_{RCF}$	Small positive constants, lower bounds for ST and RCF to avoid undefined math	Dimensionless, t$C{O}_2$/TJ

for clarity and reference.

3.2. CO₂ Emissions Projection

The projection model is based on the extended Kaya identity and incorporates the six key drivers identified in the decomposition analysis. The core projection formula defines emissions in a future year $t+1$ as a function of their compounded growth from the base year $t$. Using the transformed form of Equation (5), the model forecasts future CO₂ emissions in the manufacturing sector by accounting for the compounded growth of these drivers [9] as follow:

$$C_{t+1}=C_t·{\left(1+g_{EA}\right)}^{\left(t+1-t\right)}·{\left(1+g_{ST}\right)}^{\left(t+1-t\right)}·{\left(1+g_{EI}\right)}^{\left(t+1-t\right)}·{\left(1+g_{ELE}\right)}^{\left(t+1-t\right)}·{\left(1+g_{REN}\right)}^{\left(t+1-t\right)}·{\left(1+g_{RCF}\right)}^{\left(t+1-t\right)}$$

(7)

where $t$ is the base year (2022), and $g_x$ represents the constant annual growth rate of driver $x$.

This multiplicative model is consistent with the additive LMDI framework. To estimate the annual additive change in emissions $\left(\Delta C\right)$ and the contribution of each driver $\left(\Delta C_x\right)$ from the growth rates, a scaling factor $W_T$ is used. The contribution of each driver $x$ to the emission change over a period is calculated as:

$$\Delta C_x=W_T·In\left(1+g_x\right)$$

(8)

In Equation (8), $W_T$ is a composite scaling factor that ensures perfect decomposition between the multiplicative projection and the additive framework. Its value, which varies each year, is derived by ensuring the total change $\Delta C=C_{t+1}-C_t$ is exactly equal to the sum of the contributions from all drivers. Intuitively, $W_T$ represents the general scale of the emissions system in a given year, acting as a weighting factor that translates the logarithmic growth rates of each driver into their absolute contribution (in MtCO₂) to the total annual change in emissions.

The factor $W_T$ is formulated as the logarithmic mean of the projected emissions $C_{t+1}$ and the base year emissions $C_t$, which is the standard weighting function in LMDI, applied here to future projections to maintain additive consistency. It consolidates the non-linear interaction of all six drivers’ growth rates and is calculated as:

$$W_T={\displaystyle \frac{C_t·\left[\prod x\left(1+g_x\right)-1\right]}{In\left(\prod x\left(1+g_x\right)\right)}}$$

(9)

where $\prod x\left(1+g_x\right)$ represents the product of the growth factors for all six drivers $\left(1+g_{EA}\right)\times \left(1+g_{ST}\right)\times \left(1+g_{EI}\right)\times \left(1+g_{ELE}\right)\times \left(1+g_{REN}\right)\times \left(1+g_{RCF}\right)$, which is equivalent to ${\displaystyle \frac{C_{t+1}}{C_t}}$ in Equation (7).

This formulation provides the direct link between the assumed annual growth rates of the drivers ($g_x$) and their cumulative impact on emissions, which forms the basis for the scenario projections and GA optimization. The calculated annual values of $W_T$ for the Baseline (BAS), Stagnation and Crisis (SCS), and Just Transition Accelerated (JTAS) scenarios are provided in Supplementary Table S1; the main text reports only the consolidated results. Accordingly,

Table 4. Summary Statistics of Scenario Data (2023–2040).

Statistic	Emissions (MtCO2)	EA (GDP)	ST (MVA/GDP)	EI (TEU/MVA)	ELE (Share)	REN (Share)	RCF (tCO2/TJ)
GA-Optimized
Mean	70.700	682,900	0.096	1.520	0.326	0.118	0.002
St. Dev.	47.800	108,200	0.018	0.500	0.015	0.054	0.000
Minimum	15.400	538,426	0.067	0.760	0.303	0.039	0.002
Maximum	146.100	877,646	0.122	2.190	0.349	0.204	0.002
2040 Value	15.400	877,646	0.067	0.760	0.349	0.039	0.002
BAS
Mean	169.100	772,500	0.108	1.570	0.300	0.133	0.004
St. Dev.	1.400	168,400	0.012	0.400	0.000	0.048	0.002
Minimum	167.000	544,412	0.089	1.020	0.300	0.067	0.002
Maximum	171.100	1,070,909	0.123	2.220	0.300	0.210	0.008
2040 Value	171.100	1,070,909	0.089	1.020	0.300	0.067	0.008
SCS
Mean	97.200	610,200	0.095	1.870	0.300	0.104	0.003
St. Dev.	38.100	48,600	0.017	0.240	0.000	0.048	0.001
Minimum	44.000	531,019	0.066	1.540	0.300	0.034	0.002
Maximum	154.900	683,963	0.121	2.270	0.300	0.202	0.004
2040 Value	44.000	683,963	0.066	1.540	0.300	0.034	0.004
JTAS
Mean	391.500	832,900	0.123	1.420	0.390	0.698	0.002
St. Dev.	165.300	239,500	0.001	0.480	0.062	0.327	0.000
Minimum	184.200	549,330	0.121	0.690	0.310	0.252	0.001
Maximum	608.400	1,259,073	0.126	2.180	0.512	1.000	0.002
2040 Value	574.900	1,259,073	0.120	0.690	0.512	1.000	0.001

Source: Author.

presents the summary statistics for CO₂ emissions across all scenarios for the period 2023–2040. The statistics reveal a fundamental decarbonization trade-off. The GA-optimized pathway reduces emissions by 15.4 MtCO₂ by 2040 but does so through a contraction in ST by 6.7% and a declining renewable share (REN: 3.9%), relying primarily on direct fuel switching. In contrast, JTAS reaches a fully renewable power mix and sustains strong economic growth, yet emissions rise by 574.9 MtCO₂ because expansionary growth outweighs its moderate RCF reductions. BAS remains largely stagnant (~171 MtCO₂) with a continued rise in RCF, while SCS lowers emissions to 44.0 MtCO₂ solely through economic decline, reinforcing its characterization as a crisis-driven pathway.

3.2.1. Policy Scenario and Parameter Setting

Scenario parameters are derived from historical CO₂ emission drivers (2002–2022) quantified through LMDI decomposition. The Baseline (BAS) extends observed compound annual growth rates (CAGR). The SCS and JTAS adjust these outcomes using policy documents, industry roadmaps, and expert assessments to ensure plausibility.

The Baseline (BAS) scenario assumes continuity of past trends: EA (4.06%), ST (−1.93%), EI (−4.50%), ELE (0.0%), REN (−6.54%), and RCF (7.82%). The Stagnation and Crisis (SCS) scenario reflects economic instability and stalled transition: EA slows (1.5%), ST declines further (−3.5%), EI improves modestly (−2.25%), ELE stagnates (0.0%), REN falls faster (−10.0%), while the RCF moderates (4.0%).

The Just Transition Accelerated (JTAS) scenario represents an ambitious but feasible decarbonization pathway aligned with the country’s JETP. Assumptions include strong growth in EA (5.0%), near-stable ST (−0.25%) driven by policy stabilization and green reindustrialization [4], accelerated EI gains (−6.5%) through mandated adoption of best-available technologies [44], rapid ELE (3.0%) via strategic deployment in industrial processes [45], large-scale REN expansion (12.0%) aligned with least-cost capacity expansion plans [46], and declining RCF (−2.5%) through direct fuel switching supported by green hydrogen and biomass roadmaps [47].

Table 5. Quantified annual average change rates for each scenario driver (2023–2040).

Driver	BAS (2002–2022) (%)	SCS (2022–2040) (%)	JTAS (2022–2040) (%)	Key Data Source and Rationale for JTAS
EA (Δα)	4.06	1.5	5.0	JET-IP target of green investment-led manufacturing growth, [4].
ST (Δβ)	−1.93	−3.5	−0.25	JET-IP: Policy-driven stabilization and green reindustrialization, [41].
EI (Δγ)	−4.50	−2.25	−6.5	Aggressive efficiency gains from mandated BAT adoption, [44].
ELE (Δδ)	0.00	0.00	3.0	Strategic push for productive electrification of industrial processes [45].
REN (Δ$\tau$)	−6.54	−10.0	12.0	Accelerated deployment aligned with the least-cost capacity expansion plans, [46].
RCF (Δ$\phi$)	7.82	4.00	−2.5	Fuel switching mandated by climate policy and supported by green H2/biomass roadmaps [47].

Source: Authors.

summarizes the annual average change rates (2022–2040) and key sources for each scenario.

3.2.2. GA-Optimization Scenario

To identify the most optimal decarbonization pathway within realistically achievable limits, the scenario assumptions outlined in Table 3 are further used to define the constraint parameters for a GA-optimization model. The lower and upper bounds for the six drivers’ growth rate $\left(g_{EA}, g_{ST}, g_{EI}, g_{ELE}, g_{REN},g_{RCF}\right)$ were set using the SCS and JTAS, respectively, creating a probable solution space for the algorithm to explore [14,17,48]. The GA’s objective function is to minimize the cumulative CO₂ emissions from 2023 to 2040. This integration allows for a comparison between policy-defined pathways (BAS, SCS, JTAS) and a mathematically optimized (GA-optimized) pathway that operates within the same realistic constraints.

3.2.3. Formulation of Optimal Reduction Trajectory

The core of the model is the extended Kaya identity. The value of each driver $i$ in a future year $t+1$ is determined by its compounded growth from the base year $C_t$ = 2022. The most comprehensive formulation defines the decision variable as a matrix of time-dependent growth rates:

$$z=\left[\begin{array}{c}{g}_{EA,2023} g_{EA,2024} \text{---} g_{EA,2040}\\ g_{ST,2023} g_{ST,2024} \text{---} g_{ST,2040}\\ g_{EI,2023} g_{EI,2024} \text{---} g_{EI,2040}\\ g_{ELE,2023} g_{ELE,2024} \text{---} g_{ELE,2040}\\ g_{REN,2023} g_{REN,2024} \text{---} g_{REN,2040}\\ g_{RCF,2023} g_{RCF,2024} \text{---} g_{RCF,2040}\end{array}\right]$$

(10)

where each element $g_{i,t}$ represents the annual growth rate of the driver $i$ in a year $t+1$. The value of each driver in the year $t+1$ is thus given by Equation (11).

$$i_{t+1}=i_t·\prod _{T=t}^{t+1} \left(1+g_{i,T}\right)$$

(11)

where $i_t$ represents $E{A}_{t+1}, S{T}_{t+1}, E{I}_{t+1}, EL{E}_{t+1}, RE{N}_{t+1}, \mathrm{and} RC{F}_{t+1}$. The Total Energy Used $\left(TE{U}_{t+1}\right)$ and CO₂ emissions $\left(C_{t+1}\right)$ are then calculated as:

$$TE{U}_{t+1}=E{A}_{t+1}· E{I}_{t+1}$$

(12)

The final CO₂ emissions in the year $t_{+1}$ are given by the identity:

$$C_{t+1}=TE{U}_{t+1}· EL{E}_{t+1}· RE{N}_{t+1}RC{F}_{t+1}$$

(13)

3.2.4. Optimization Problem

The formal optimization problem is to find the matrix Z that minimizes the cumulative emissions over the projection period T = 18 years, as shown in Equation (14).

$$\underset{Z}{min}J(Z)=\sum _{t_{+1}=1}^T{C}_t$$

(14)

subject to bound constraints for each element of the matrix:

$$L\le \mathrm{Z} \le \mathrm{U}$$

(15)

where L and U are matrices of lower and upper bounds derived from scenario analysis.

The formulation in Equation (10) defines a high-dimensional problem with 6 × 18 = 108 decision variables. To enhance computational tractability and convergence of the GA, while maintaining policy relevance (as long-term plans often assume constant annual improvement rates), we simplify the problem. We assume each driver grows at a constant annual rate throughout the projection period. This reduces the decision variable matrix Z to a vector $g$ of only six parameters:

$$g=\left(C_{EA}, C_{ST}, C_{EI}, C_{ELE}, C_{REN}, C_{RCF}\right)$$

(16)

The projection equations simplify to:

$$i_{t+1}= i_t·{\left(1 + g_i\right)}^{\left(t_{+1}-t\right)}$$

(17)

The physical and mathematical constraints are applied as follows:

$$E{A}_{t+1}= E{A}_t·{\left(1 + g_{EA}\right)}^{t_{+1}-t}$$

(18)

$$S{T}_{t+1}= \max \left(S{T}_t\right.·{\left(1+g_{ST}\right)}^{\left(t_{+1}-t\right)},\left.{\in }_{ST}\right)$$

(19)

$$E{I}_{t+1}=E{I}_t·{\left(1+g_{EI}\right)}^{\left(t_{+1}-t\right)}$$

(20)

$$EL{E}_{t+1}= \min \left(\max \right.\left(EL{E}_t\right.·{\left(1+g_{ELE}\right)}^{\left(t_{+1}-t\right)},\left.\left.0\right),1\right)$$

(21)

$$RE{N}_{t+1}= \min \left(\max \right.\left(RE{N}_t\right.·{\left(1+g_{REN}\right)}^{\left(t_{+1}-t\right)},\left.\left.0\right),1\right)$$

(22)

$$RC{F}_{t+1}= \max \left(RC{F}_t\right.·{\left(1+g_{RCF}\right)}^{\left(t_{+1}-t\right)},\left.{\in }_{RCF}\right)$$

(23)

where $\in$ are small positive constants to avoid undefined mathematical operations.

$$\underset{g}{min}J(g)=\sum _{t_{+1}=1}^T{C}_t\left(g\right) \mathrm{subject} \mathrm{to} \mathrm{L}\le g\le U$$

(24)

The bounds L and U are vectors defined by the extreme values of the SCS and JTAS:

$$L=\left(\stackrel{min}{g_{EA}}, \stackrel{min}{g_{ST}}, \stackrel{min}{g_{EI}}, \stackrel{min}{g_{ELE}}, \stackrel{min}{g_{REN}}, \stackrel{min}{g_{RCF}}\right)=0.015, 0.035, -0.065, 0.00, -0.10, -0.025$$

(25)

$$U=\left(\stackrel{max}{g_{EA}}, \stackrel{max}{g_{ST}}, \stackrel{max}{g_{EI}}, \stackrel{max}{g_{ELE}}, \stackrel{max}{g_{REN}}, \stackrel{max}{g_{RCF}}\right)=0.050,-0.0025, -0.225, 0.03, 0.12, 0.04$$

(26)

3.2.5. GA Implementation Steps

The high-dimensional, non-linear, and constrained nature of this problem makes a GA an appropriate optimization tool. The GA is implemented with the following components:

• Chromosome Encoding: Each candidate solution is a “chromosome” represented by the vector $g=\in R^6$, containing the six growth rates.
• Fitness Function: The fitness of a chromosome is evaluated by computing the negative cumulative emissions:

$$Fitness (g)=-J(g)=-\sum _{t_{+1}=1}^T{C}_t(g)$$

(27)

The GA is configured to maximize this fitness function, thereby minimizing J(g).

• Operators and Parameters:
  • Selection: Roulette wheel selection, where the probability $P_i$ of selecting chromosome $i \mathrm{is} P_i={\displaystyle \frac{Fitnes{s}_i}{{\sum }_{j=1}^{N}Fines{s}_j}}$
  • Crossover: Arithmetic crossover with probability $p_c=0.8$ creates an offspring $g^{child}$ from two parents $g^1, {\mathrm{g}}^2$:

    $$g^{child}=\mu g^1+\left(1-\mu \right)g^2, \mu \sim v\left[\mathrm{0,1}\right]$$

    (28)
  • Mutation: Uniform mutation with probability $p_m=0.1$ per gene replaces $g_i$ with a value drawn uniformly from $\left[L_i, U_i\right]$.
  • Elitism: The top $k=2$ solutions are preserved in each generation.
  • Population and Termination: A population size of $N=300$ evolves for a maximum of 200 generations, halting early if the best fitness does not improve for 50 generations.

The algorithm outputs the optimal growth vector:

$$g=\mathit{arg} \underset{g}{min} J\left(g\right) subject to L\le g\le U$$

(29)

3.2.6. Output and Scenario Comparison

The emissions trajectory resulting from $g$ defines the GA-Optimized scenario. This pathway is compared against the three predefined policy scenarios (BAS, SCS, JTAS), which are generated by passing their respective fixed growth vectors $g^{BAS}, {\mathrm{g}}^{SCS}, {\mathrm{g}}^{JTAS}$ through the same projection model.

The comparison is performed on two primary metrics:

• The cumulative $C{O}_2$ emissions over the projection period; $J(g)$
• The annual emission trajectory; $C_t(g)$

This LMDI-GA framework provides a powerful, mathematically robust method to derive a benchmark optimal pathway, against which the feasibility and ambition of policy-based scenarios can be rigorously evaluated.

3.3. Sensitivity Test

A sensitivity test is further conducted to assess the robustness of the JTAS projections and to identify which driver exerts the greatest influence on long-term emissions. The analysis focused on the JTAS, representing the most ambitious policy pathway. This involved varying the annual growth rates of the six key drivers: EA, ST, EI, ELE, REN, and RCF. The sensitivity analysis was performed in R programming language (version 4.5.1) [49] by varying each driver’s growth rate by ±20% and re-running the forecasting model. For each adjustment, the model was re-run to project manufacturing CO₂ emissions from 2023 to 2040. Results were compared with the baseline JTAS forecast, with particular attention to emissions in 2040. Outcomes were presented using tables and graphical plots, illustrating the potential range of emission changes attributable to each driver. This approach provided a clear comparison of driver sensitivities and highlighted the levers most critical for achieving or potentially undermining the goal of absolute decoupling.

3.4. Data Sources

Data for 2002–2022 were sourced from successive South African Energy Balance and Energy Reports [41], providing official statistics on energy production and consumption by fuel type (coal, petroleum, natural gas, electricity, and renewables) and by subsector. Macroeconomic indicators, including GDP (2015 constant prices) and manufacturing value-added, were obtained from Statistics South Africa [50]. Country-specific emission factors, following IPCC guidelines [40], were applied to improve the CO₂ estimate accuracy. These datasets are widely used for national planning, NDC reporting, and the Integrated Resource Plan [51], lending both credibility and policy relevance for examining long-term CO₂ drivers in the manufacturing sector. Since utility-scale renewable consumption in manufacturing was not systematically reported before 2010, a bounded exponential back-casting procedure was applied to create a continuous time series compatible with LMDI decomposition (see Appendix A for methodological details,

Table A3. Reconstructed renewable energy consumption values (TJ) for South African manufacturing sector, 2002–2009.

Year	2002	2003	2004	2005	2006	2007	2008	2009	2010
Renewable Consumption (TJ)	363,046.75	363,046.78	363,046.82	363,046.85	363,046.89	363,046.93	363,046.96	363,047	363,047.00 (Actual)

Note: The 2010 REN value reflects the level prior to the full launch of the REIPPPP in 2011. Values for 2002–2009 represent a smooth transition to reported data. To validate the robustness of our findings against uncertainties in this reconstruction, we conducted sensitivity analyses with alternative growth rates (r = 0.1% and r = 0.2%), as reported in Section 4.5.1. The results confirm that the relative contributions of key drivers remain stable across these assumptions.

). This method captures the non-linear trajectory of early renewable adoption and addresses the LMDI method’s sensitivity to zero values [52]. The resulting trend of renewable energy consumption, showing both the reconstructed (2002–2009) and reported (2010–2022) periods, is presented in Figure 2. The manufacturing sector’s renewable data for 2002–2009 were reconstructed using this approach to provide a continuous time series, consistent with the scaling-up trajectory observed after the launch of the REIPPPP in 2011.

4. Results and Analysis

4.1. Analysis of Energy-Related CO₂ Emission Features and Carbon Intensity

South Africa’s manufacturing sector is central to the nation’s economy and its just energy transition ambitions. An analysis of emissions from 2002 to 2022 reveals that the sector has experienced significant volatility without achieving absolute decoupling over the past two decades. As depicted in Figure 3a, sectoral emissions demonstrated remarkable instability, beginning at 165.98 MtCO₂ in 2002 and, after considerable fluctuation, finishing at 166.81 MtCO₂ in 2022, a net change of only 0.50%. This overall inertia masks periods of intense variation, with annual growth rates fluctuating from −32.0% (2006) to 40.0% (2007 and 2021). The average absolute annual growth rate was 10.4%, underscoring the extreme volatility inherent in the sector’s emission trajectory, largely driven by economic cycles and energy supply crises [3,18]. The trend reveals distinct phases: the highly volatile pre-financial crisis expansion (2002–2007), a period of relative stability during the global economic downturn and recovery (2007–2012), a surge of 18.4% in 2013 reflecting continued carbon intensity during economic expansion [12,24], and the COVID-19 induced collapse (−29.6% in 2020) followed by a record 40.3% rebound in 2021. This pattern demonstrates the fragile, recession-led nature of emission reductions in the sector, with emissions rapidly returning to pre-crisis levels [3].

Complementing absolute emissions, the carbon intensity of energy use, measured as fossil-derived CO₂ emissions per unit of primary energy consumption (MtCO₂/TJ), provides further insight (Figure 3b). From 2002 (0.00012 MtCO₂/TJ) to 2022 (0.00013 MtCO₂/TJ), intensity showed no sustained decline, despite a temporary low of 0.000107 MtCO₂/TJ in 2013. This stagnation indicates that the energy powering South Africa’s factories has not been structurally decarbonized, even with policy efforts and renewable energy investments. Short-term improvements, such as those during the initial rollout of the Renewable Energy Independent Power Producer Procurement Programme (REIPPPP), were offset by increased reliance on carbon-intensive backup generation during grid crises. These patterns underscore that meaningful decarbonization requires not only the greening of the grid but also the direct substitution of fossil fuels used for industrial process heat.

4.2. Additive Decomposition Results of CO₂ Emissions Drivers

The results, shown in Figure 4, reveal a complex interplay of economic forces, policy shifts, and systemic shocks shaping South Africa’s manufacturing emissions over the past two decades. Between 2002 and 2022, total CO₂ emissions rose by ~1.40 MtCO₂, but this apparent stability masks significant volatility and the counteracting influence of key drivers. The Residual Carbon Factor (RCF) was the largest positive driver (224.1 MtCO₂; 160.14%), reflecting the persistent carbon intensity of fossil-based energy use, especially from direct coal and fossil fuel combustion. Economic Activity (EA) was another major driver that promoted emission growth by 187.5 MtCO₂ (134.03%), confirming that manufacturing growth remains a strong engine of emissions. In contrast, Energy Intensity (EI) strongly mitigated emissions by −141.8 MtCO₂ (101.35%), demonstrating substantial efficiency gains, while the Renewable Share (REN) had an even larger suppressive effect of −202.23 MtCO₂ (−144.53%), highlighting the critical role of grid decarbonization. Structural Shift (ST) further dampened emissions by −64.58 MtCO₂ (−46.15%), consistent with the declining share of manufacturing in GDP amid deindustrialization pressures [53]. Electrification Share (ELE), however, had a negligible net impact −1.59 MtCO₂ (−1.13%), indicating that the gradual move toward electrification was largely offset by the carbon intensity of the grid during this period.

4.2.1. EA and RCF Effects

The consistent positive contribution of the EA effect aligns with established literature for South Africa, which identifies economic growth as the principal driver of energy consumption and emissions [20,21]. As shown in Figure 5. EA proved to be the most reliable positive driver, contributing in 17 of the 20 intervals with an average annual additive effect of 9.38 MtCO₂. Its influence was especially pronounced during periods of robust growth, most notably in 2012–2013, when EA alone added 68.01 MtCO₂ to an overall sectoral increase of 26.30 MtCO₂, underscoring its overwhelming force in shaping emissions outcomes [18]. In contrast, the RCF, emerged as both the most unstable and the largest overall positive driver, contributing a cumulative 224.07 MtCO₂ across the two decades. Its swings were extreme, ranging from −246.66 MtCO₂ in 2016–2017 to 241.12 MtCO₂ in 2014–2015, quantitatively confirming the sector’s entrenched dependence on direct fossil fuel combustion, particularly coal, for process heat. The dominance of the RCF effect effectively nullified the gains achieved through efficiency improvements and renewable integration for many years. Together, EA and RCF reveal the structural lock-in of South Africa’s manufacturing energy metabolism, highlighting the inadequacy of strategies that prioritize energy efficiency and grid decarbonization without explicitly addressing industrial fuel switching to low-carbon alternatives such as green hydrogen or biomass, as envisioned in the JETP [4,47].

4.2.2. EI and REN Effects

The progress in decarbonization over the two decades is almost entirely attributable to improvements in EI and the REN. However, their effects were highly variable, as indicated in Figure 6. EI was the most stable mitigating factor, contributing a cumulative of −141.81 MtCO₂ with an average annual reduction of −7.09 MtCO₂. Key intervals of improvement, such as −73.44 MtCO₂ in 2005–2006 and −52.53 MtCO₂ in 2010–2011, underscore the significance of efficiency gains, which were likely achieved through technological modernization and a gradual reallocation towards less energy-intensive sub-sectors. REN provided an even larger cumulative contribution of −202.23 MtCO₂, but its pattern was far more erratic. Its strongest impact was felt in 2016–2017, when the REN effect alone reduced emissions by −246.66 MtCO₂, a shift closely tied to the integration of new renewable capacity under the REIPPPP [54]. This effect was so substantial that it drove the net emissions change for that year to −9.00 MtCO₂, despite ongoing economic growth. However, the fragility of this progress was laid bare in 2017–2018, when renewable generation faltered due to grid instability and a “renewables drought” [55], producing a sharply positive REN effect of 240.73 MtCO₂. This reversal forced greater reliance on carbon-intensive backup generation, particularly open-cycle gas turbines [46], resulting in a net sectoral increase of 7.78 MtCO₂. Together, EI and REN illustrate both the promise and the precarity of South Africa’s decarbonization trajectory: efficiency gains delivered consistent suppression of emissions, while renewable rollouts produced breakthroughs that remain vulnerable to systemic instability.

4.2.3. ELE and ST Effects

The two drivers, ELE and ST, highlight the challenges of aligning industrial transformation with a just transition. As shown in Figure 7, the ELE effect had a negligible net contribution of −1.59 MtCO₂ over the period, but this masks large annual swings, from a sharp increase of 60.88 MtCO₂ during 2005–2006 to a substantial decline of −49.40 MtCO₂ in 2006–2007. These fluctuations reveal a crucial insight: electrification, on its own, is not a decarbonization strategy. Its climate benefit is entirely conditional on the carbon intensity of the national electricity grid. When the grid remained highly carbon-intensive (high RCF), additional electrification inconsistently raised emissions, as seen in 2005–2006. In contrast, the ST effect demonstrated a persistent downward pull, with a cumulative reduction of −64.58 MtCO₂ with an average annual effect of −3.23 MtCO₂. While this appears to reduce emissions, it is not the outcome of productive decarbonization, but rather of structural decline. The shrinking relative share of manufacturing in GDP reflects deindustrialization and economic vulnerability [54], rather than genuine efficiency gains. If left unchecked, this trend risks undermining employment, output, and industrial competitiveness.

4.3. Scenario Analysis

The scenario analysis quantifies the tension between growth, decarbonization, and justice. Results (

Table 6. Comparative Performance of Decarbonization Pathways.

Scenario	Baseline 2022 (MtCO2)	Target Year 2040 (MtCO2)	Change from Baseline (%)	Cumulative Emissions (2023–2040) (MtCO2)
BAS	166.81	120.47	−27.78	2539.88
JTAS	166.81	950.05	469.54	8502.02
SCS	166.81	23.18	−86.10	1239.54
GA-Optimized	166.81	15.36	−90.79	1068.9
JTAS under GA-Optimized	166.81	574.87	−70.98	7603.50

Source: Authors.

and Figure 8) show divergent trajectories, underscoring that current policies remain insufficient to secure a just transition.

4.3.1. Baseline (BAS) Scenario

As shown in Table 6, the BAS, which is parameterized by extending the historical compound annual growth rates (CAGR) observed from 2002 to 2022, projects emissions decline from 166.81 MtCO₂ in 2022 to 120.47 MtCO₂ in 2040, representing a 27.8% decline (Figure 8a). While this appears encouraging, it is achieved largely through efficiency gains and deindustrialization rather than structural transformation. With cumulative emissions of 2539.9 MtCO₂ (2023–2040), BAS illustrates a path of relative decoupling that contradicts the reindustrialization and inclusive growth goals of the National Development Plan and JETP.

4.3.2. Stagnation and Crisis (SCS) Scenario

The SCS shows the sharpest decline (Table 6, Figure 8a), with emissions declining to 23.18 MtCO₂ by 2040, an 86.1% reduction, and cumulative emissions of 1239.5 MtCO₂. This outcome stems from economic contraction (−3.5% structural shift), stalled efficiency (−2.25%), and a collapse in renewable growth (−10.0%). The reductions are therefore crisis-driven, not policy-led, reflecting degrowth through systemic failure. While ecologically significant, this pathway would be socially and politically untenable, highlighting the risks of a disorderly transition. This pathway exemplifies what theorists describe as “ill-being-led degrowth” [56], where ecological gains come at the expense of social and economic well-being. It echoes the deepest concerns of trade unions and civil society about a transition that sacrifices workers and communities in pursuit of emission targets [54].

4.3.3. Just Transition Accelerated (JTAS) Scenario

The JTAS, aligned with the country’s JETP ambitions, assumes strong economic growth (5.0% MVA), deep efficiency gains (−6.5%), rapid electrification (3.0%), and major renewable expansion (12.0%). However, emissions rise to 950.05 MtCO₂ by 2040, as shown in Figure 8a, a stunning 469.5% increase, with cumulative emissions of 8502.0 MtCO₂. This paradox reflects the country’s structural carbon inertia, where demand growth outpaces grid decarbonization, converting “green growth” into carbon growth. JTAS thus reveals the critical necessity of accelerating fossil phase-out alongside renewables if green industrialization is to deliver genuine decarbonization.

4.3.4. Genetic Algorithm Optimization Pathway

As shown in Table 6 above, the emission trajectory produced by the GA-optimized pathway demonstrates substantial decarbonization potential. Emissions declined from 166.81 MtCO₂ in 2022 to 15.36 MtCO₂ by 2040, a 90.8% reduction, the deepest reduction achieved across all scenarios. The corresponding annual optimal growth vector (g*) is presented in

Table 7. Optimal Growth Vector Derived from Genetic Algorithm Optimization. Source: Authors.

Driver	Variable	GA-Optimized Growth Rate (gx) (Annual)	Policy Scenario Bounds (%)
Economic Activity (EA)	$g_{EA}=\Delta \alpha$	2.15%	SCS: 1.5 to JTAS: 5.0
Structural Shift (ST)	$g_{ST}=\Delta \beta$	−3.41%	SCS: −3.5 to JTAS: −0.25
Energy Intensity (EI)	$g_{EI}=\Delta \gamma$	−6.04%	SCS: −2.25 to JTAS: −6.5
Electrification Share (ELE)	$g_{ELE}=\Delta \delta$	0.82%	SCS: 0.0% to JTAS: 3.0
Renewable Share (REN)	$g_{REN}=\Delta \tau$	−9.25%	SCS: −10.0 to JTAS: 12.0
Residual Carbon Factor (RCF)	$g_{RCF}=\Delta \phi$	−1.01%	SCS: 4.0 to JTAS: −2.5

. The algorithm prioritizes aggressive improvements in Energy Intensity ($g_{EI}$ = −6.04%) as the dominant mitigation lever and identifies a necessary, albeit modest, reduction in the RCF ($g_{RCF}$ = −1.01%). However, the pathway attains this level of decarbonization only by inducing a sharp annual decline in the manufacturing sector’s economic share ($g_{ST}$ = −3.41%), signalling substantial structural contraction. Adding to this, the model assigns a negative growth rate to the Renewable Share ($g_{REN}$ = −9.25%), a counterintuitive outcome when interpreted against the broader requirements of system-wide decarbonization. These specific results, particularly the reliance on structural contraction and the reduction in renewable share, highlight both the strengths and limitations of the optimization approach. Their full implications for transition feasibility and justice are discussed in detail in the Socio-Economic Trade-offs (Section 4.4.2).

4.4. Comparative Analysis

4.4.1. Policy Pathways vs. Mathematical Optimum

The comparative analysis between policy-defined scenarios (BAS, JTAS, SCS) and the mathematically GA-optimized pathway reveals critical limitations in current decarbonization approaches. As shown in Table 6, the GA-optimized pathway demonstrates superior environmental performance across all metrics compared to the policy scenarios. The emission trajectories in Figure 8b highlight three key insights.

First, the ambition gap: despite JTAS representing the most ambitious policy vision, emissions rise 469.5% by 2040, vastly exceeding the GA-optimized reduction of 90.8%, a 560.3% divergence. This gap begins around 2025 and widens rapidly, showing that even aggressive renewable deployment (12.0%) and efficiency gains (−6.5%) cannot offset the carbon intensity of industrial growth when RCF remains largely unaddressed.

Second, the efficiency gap: while the SCS achieves deep emission cuts (−86.1%), it does so through economically destructive contraction. The GA-optimized pathway surpasses SCS by 4.7% while sustaining higher economic activity (EA = 2.15% vs. 1.5%), demonstrating that coordinated decarbonization levers can yield superior environmental outcomes without sacrificing economic performance. The cumulative emissions difference of 170.64 MtCO₂ reflects avoided emissions achieved through strategic coordination rather than contraction.

Third, the coordination deficit: the GA-optimized pathway outperforms because it simultaneously balances all six drivers, rather than treating them in isolation. JTAS, for example, excels in individual parameters (REN = 12.0%, EI = −6.5%) but fails to counteract critical forces such as RCF (−2.5%). This shows that moderate, harmonized improvements across all drivers outperform scenarios with uneven progress and critical weaknesses. BAS, extending historical trends, illustrates the limits of incremental change, achieving only a 27.8% reduction by 2040.

4.4.2. Socio-Economic Trade-Offs

The GA-optimized pathway, while a mathematical benchmark, reveals profound socio-economic trade-offs that challenge the feasibility of a just transition. Its success is predicated on structural contraction (ST = −3.41%), which would reduce manufacturing’s GDP share to 6.7% by 2040. This represents accelerated “deindustrialization-led decoupling,” a sectoral equivalent to the ‘degrowth’ scenarios proposed in global sustainability science [34,35]. It demonstrates that deep emissions reductions are mathematically feasible through reduced economic throughput, challenging the green-growth paradigm [31,36].

However, this mathematical optimum translates into a socio-economic crisis. This prescription occurs against a backdrop of pre-existing structural decline, with the country’s manufacturing employment having already plummeted from 1.8 million to 1.4 million between 2019 and 2021 alone [57]. Pursuing this pathway would accelerate this trend, constituting a form of ‘ill-being-led degrowth’ [55] that directly contradicts the JETP’s objectives of justice, inclusion, and industrial development [4].

Fundamentally, such a trajectory is politically untenable. The concentrated job losses in industrial hubs would deepen severe pre-existing vulnerabilities, inevitably provoking the kind of socio-political resistance observed in other fossil-dependent regions undergoing disorderly transitions [37,58]. This underscores a central insight from transition theory: decarbonization is not just a technical challenge but a profoundly political one [30,39]. A pathway that sacrifices workers and communities on the altar of emission targets is likely to be blocked by the very forces of carbon lock-in it seeks to overcome.

This theme of model limitations is further illustrated by the counterintuitive result for the Renewable Share ($g_{REN}$ = −9.25%). This is a crucial outcome of the single-sector optimization boundary. Within the narrow objective of minimizing manufacturing emissions directly, the model “chooses” to deprioritize investments in grid decarbonization; a shared, economy-wide resource in favour of more direct, within-sector levers. It is essential to state that this should not be misconstrued as a real-world policy recommendation. In reality, grid decarbonization (REN) is a non-negotiable prerequisite for enabling low-carbon electrification (ELE) and for the overall energy transition. This result powerfully highlights a core limitation of single-sector optimization: its inability to capture essential, economy-wide synergies.

Therefore, the primary value of the GA-optimized pathway is not as a policy blueprint, but as a stark benchmark. It highlights both the untapped mitigation potential and the non-negotiable imperative to pursue decarbonization through green structural transformation rather than simple economic contraction.

4.5. Sensitivity Analysis

4.5.1. Robustness of Decomposition Results Under Renewable Data Uncertainty

Table 8. Sensitivity of emission drivers to renewable energy data assumptions (MtCO₂).

Period	Description	EA	ST	EI	ELE	REN	RCF	Total CO2 Change
2002–2022	r = 0.001	187.54	−64.58	−141.81	−1.59	−202.23	224.07	1.40
2002–2022	r = 0.002	187.54	−64.58	−141.81	−1.59	−202.23	203.82	1.40
2010–2022	REIPPPP	142.23	−45.20	−113.78	−7.20	−198.72	226.18	3.51

Source: Authors.

presents a sensitivity analysis addressing uncertainties in pre-2010 renewable energy consumption. While the magnitude of individual driver contributions shifts under alternative assumptions, their hierarchy and directional influence remain consistent, confirming the robustness of the findings.

The baseline scenario (r = 0.001), which assumes conservative renewable growth, shows only a modest net emissions increase over two decades (1.40 MtCO₂), masking strong opposing forces. Economic Activity (EA: 187.54 MtCO₂) and the Residual Carbon Factor (RCF: 224.07 MtCO₂) dominate, underscoring manufacturing growth and fossil fuel reliance, especially coal for process heat, as the main drivers. Offsetting these pressures are Energy Intensity improvements (EI: −141.81 MtCO₂) and Renewable Share gains (REN: −202.23 MtCO₂), which act as critical brakes on emissions. A higher back-casting rate (r = 0.002), assuming faster renewable uptake in the early 2000s, produces a nearly identical trajectory (Table 8), with RCF varying by only 0.10%. This confirms that renewable growth assumptions minimally affect the relative contributions of other drivers.

4.5.2. Hierarchical Influence of Decarbonization Levers

The sensitivity analysis reinforces the GA-optimization driver hierarchy, confirming REN and EI as the most influential decarbonization levers. By quantifying each driver’s impact, the analysis provides a precise hierarchy of systemic leverage points (

Table 9. Sensitivity of JTAS 2040 emissions to a ±20% change in driver assumptions.

Driver	Emission Range (MtCO2)	Key Interpretation
REN	748.23	The most potent lever. Grid decarbonization is the paramount external factor.
EI	479.71	A critical internal lever; efficiency gains are a non-negotiable prerequisite.
EA	327.07	Confirms economic growth as a fundamental, upward pressure on emissions.
RCF	175.60	Highlights the critical barrier of fossil fuel lock-in in industrial process heat.
ELE	199.54	Efficacy is conditional and contingent on a decarbonized grid (REN).
ST	17.14	A weak lever; broad structural change is ineffective for short-term decarbonization.

Source: Authors.

; Figure 9). A ±20% variation in annual growth rates revealed a pronounced efficacy gradient for 2040 emissions. REN emerges as the dominant lever, with an influence range of 748.23 MtCO₂, highlighting the grid’s carbon intensity as the largest external driver; a 20% shortfall raises emissions by 441.5 MtCO₂, emphasizing risks from delays in programmes like the REIPPPP.

EI ranks second (479.71 MtCO₂), underscoring the importance of sustained efficiency gains. Strong sensitivities to EA and RCF reveal a core challenge: green industrialization increases emissions unless RCF is reduced through mandatory fuel switching. ELE has a moderate impact, effective only with rapid grid decarbonization, while ST is negligible through 2040. Figure 9 illustrates these divergences from the baseline JTAS trajectory under ±20% driver variations.

As shown in Figure 9, each subplot illustrates the effect of perturbing a single driver (labelled) while holding all others constant at their JTAS values. The solid black line in each panel represents the baseline JTAS trajectory. The panels for REN and EI display the widest bands of outcomes, confirming that these are the most influential levers for the long-term emission trajectory. The resulting range of emissions in 2040 for each driver is summarized in Figure 10, which ranks drivers hierarchically by their influence. This analysis demonstrates that REN and EI are the most potent levers, whereas ST has a negligible impact.

4.5.3. Evaluation of Policy Intervention Timing

To test whether intensifying conventional strategies could shift the sector’s emissions trajectory, three JTAS-based interventions were modelled (

Table 10. Comparison of 2040 emissions under policy intervention scenarios.

Scenario	2040 Emissions (MtCO2)	Core Intervention
Enhanced Renewables and Efficiency	1115.1	25% increase in REN and EI rates from 2023.
Maximum Feasible Intervention	1024.0	Aggressively improved rates for EI, ELE, REN, and RCF.
Delayed Enhanced Intervention	1002.2	25% increase in REN and EI rates from 2035.
JTAS (Baseline)	950.1	Ambitious policy based on current NDC ambitions.

Source: Authors.

). The results show that accelerating REN and EI gains without transforming industrial energy metabolism (RCF) is not only insufficient but counterproductive. The Enhanced Renewables and Efficiency scenario performs worst, adding 165 MtCO₂ above the JTAS baseline by 2040. By boosting output while leaving fossil-based process heat intact, such measures effectively “supercharge” emissions, an instance of policy resistance where well-intentioned actions worsen the problem.

Even the Maximum Feasible Intervention scenario, representing the upper bound of technology deployment, fails to achieve an emissions peak, demonstrating that absolute “green growth” is unviable within the modelled timeframe. The system’s carbon inertia is too great for incremental improvements alone.

A comparison of early versus delayed action (

Table 11. Quantifying the penalty of early but misdirected action.

Metric	Value	Key Interpretation
Avoided Emissions	−112.97 MtCO2	The delayed intervention results in lower emissions than the Early intervention.
Avoided Percent	−11.27%	Early action with a flawed strategy is more detrimental than a delayed version of the same strategy.

Source: Authors.

) adds a counterintuitive insight: the Delayed Enhanced Intervention scenario outperforms the early version, yielding negative “Avoided Emissions.” This does not mean delay is beneficial, but rather that misaligned action taken too soon locks the system into a higher-emission pathway. A strategy focused solely on REN and EI is, therefore fundamentally flawed.

5. Discussion, Key Findings and Their Proposition

The findings of this study reveal a counterintuitive reality for South Africa’s manufacturing sector: the most ambitious policy-driven transition (JTAS) results in a sharp rise in emissions, while the mathematically optimal pathway (GA) achieves deep reductions only through socially and economically costly deindustrialization. Although Energy Intensity (EI) has historically acted as a key mitigator, its future potential may be overestimated. The aggressive efficiency improvements modelled under JTAS are highly susceptible to the rebound effect (Jevons Paradox), whereby energy and cost savings are reinvested into additional carbon-intensive economic activities. Sorrell [59] underscores that while EI improvements are necessary, they are not sufficient on their own, as unmanaged rebound effects can ultimately amplify emissions in growth-oriented scenarios. This caution aligns with the systematic review by Haberl et al. [35], which shows that efficiency gains are frequently overwhelmed by scale effects, preventing absolute decoupling. These outcomes are robust quantitative evidence of a broader global challenge, as observed across national and sectoral studies as summarized in

Table A4. Key findings and their proposition.

Key Findings from This Study	Discussion and Interpretation	Synergy with Related Literature
The Just Transition Accelerated (JTAS) scenario causes a 469% rise in emissions by 2040.	The rapid economic growth (Δα = 5.0%) and slow fossil fuel phase-out (Δφ = −2.5%) overwhelm the positive contributions from renewables and efficiency. This illustrates the limits of “green growth” in a carbon-intensive system, where absolute decoupling remains unattainable.	Hickel and Kallis [27]; Parrique et al. [26]: These studies provide foundational critiques of green growth, arguing that absolute decoupling is empirically rare and insufficient. This finding offers a rigorous sectoral case study of that phenomenon, reflecting the broader challenge of achieving decarbonization within a growth-oriented framework.
The mathematically optimal (GA-optimized) pathway requires deindustrialization ($g_{ST}$ = −3.41%).	This represents “deindustrialization-led decoupling,” a socio-economically devastating outcome that contradicts the development and justice imperatives of South Africa’s JETP. The model’s suggestion to cut the renewable share ($g_{REN}$ = −9.25%) indicates that from a pure emissions calculus, direct fossil fuel substitution is a more powerful short-term lever than grid greening.	Keyßer and Lenzen [36]; Jackson [37]: This sectoral result is an analogue to global “degrowth” and “demand-side mitigation” scenarios. It quantifies the potent yet socially painful emissions savings from reducing material throughput. Reference [36] elaborates on the economic models for a ‘post-growth’ challenge, providing the socio-economic theory behind such mathematical findings.
Historical volatility was driven by a “fierce interaction” between the REN and RCF effects.	This highlights that the core of the decarbonization challenge lies not in power generation alone, but in industrial process heat. The fierce opposition between RCF and REN quantifies the devastating impact of grid instability: when load-shedding forces reliance on carbon-intensive backup generation, it instantly erases gains from renewable energy investments. This reveals a critical interdependence between the electricity transition and the industrial fuel transition.	This finding empirically confirms that carbon lock-in and regime resistance as theorized by Geels [25] have a precise, measurable variable: the RCF. The volatility of the RCF effect is a numerical signature of the political and economic battles over the future of coal in South Africa’s industrial heartland, demonstrating how power and politics directly shape emission trajectories.
The Residual Carbon Factor (RCF) is the dominant and most volatile driver.	The carbon intensity of direct fossil fuel use for process heat represents the critical challenge. Any successful transition strategy must address this factor directly and promptly.	This aligns with the latest IPCC assessment by Creutzig et al. [11], which emphasizes the need for radical technological and demand-side shifts in hard to abate sectors.

in Appendix A.

The integrated findings of this study point to a fundamental systemic challenge: the deep carbon inertia embedded within South Africa’s industrial socio-technical system. This inertia arises from a structural dependence on fossil fuels for industrial process heat, creating a lock-in effect that undermines decarbonization efforts focused solely on efficiency and the power sector. The analysis shows that even substantial mitigating impacts from renewable energy penetration and efficiency gains are systematically offset by the carbon intensity of direct fuel combustion during periods of economic expansion. These dynamics can be interpreted through a thermodynamic analogy of the industrial energy system. In this conceptualization, the manufacturing sector functions as a complex energy conversion system. EA acts as the system’s throttle, dictating total throughput and energy demand. EI represents the thermodynamic efficiency of the core processes, the useful output per unit of energy input. The REN reflects the gradual decarbonization of one energy input vector, electricity. Meanwhile, RCF constitutes the dominant carbon-intensive substrate, representing fossil fuels combusted directly within industrial processes. The critical insight from our scenario analysis is that the current policy paradigm, as embodied in the JTAS, prioritizes improving system efficiency (EI) and greening one input vector (REN) while expanding throughput (EA), but neglects the fundamental transformation of the primary energy substrate (RCF). This is analogous to making a combustion engine more efficient and supplementing its fuel with a less carbon-intensive additive, while increasing its revolutions per minute, without replacing the petroleum base. The system’s overall carbon emissions remain tethered to the carbon content of its core feedstock.

Consequently, the projected 469% surge in emissions under the ambitious JTAS pathway is not an anomaly but a direct outcome of this systemic misalignment. The strategy boosts economic output atop an industrial metabolism that remains fundamentally fossil-based. This explains the historical ineffectiveness of ELE, whose climate benefit is entirely contingent on grid decarbonization (REN). It also highlights the limitations of modelled policy interventions, which intensified renewables and efficiency without addressing RCF, measures that merely improve the performance of a system whose underlying carbon logic remains unchanged. Therefore, achieving a definitive emissions peak requires a foundational intervention that precedes and enables these efforts: the mandated replacement of fossil fuels in industrial process heat with zero-carbon alternatives such as green hydrogen or biomass. Without directly addressing this source of carbon inertia, policies focused solely on efficiency and renewable electricity will continue to be overwhelmed by the entrenched fossil fuel dependency of the industrial sector.

5.1. Conclusions and Policy Implications

5.1.1. Conclusions

This study provides a stress-test, scenario-based assessment of South Africa’s manufacturing decarbonization, revealing a critical irony in the country’s just energy transition. Even under an accelerated JTAS, emissions are projected to rise by 469% by 2040. This surge occurs because the carbon intensity of direct fossil-fuel use in industrial process heat captured by the RCF overwhelms gains from renewable energy and efficiency improvements. The core finding highlights deep industrial carbon inertia: structural dependence on fossil fuels for process heat creates a lock-in effect that grid greening and efficiency measures alone cannot break. Historical trends and model results further show that electrification and renewable adoption, while essential, are insufficient on their own; the electricity transition and the industrial fuel transition are interdependent, and progress in one is nullified without the other. Although the GA identifies a pathway to a 91% emissions reduction by 2040, this mathematically optimal solution depends on socially damaging deindustrialization, directly contradicting the country’s JETP goals of justice, inclusion, and industrial development.

The scenario analysis, therefore, presents a stark choice between two unsustainable outcomes: the JTAS pathway, which drives an alarming rise in emissions, and the GA-optimized pathway, which achieves deep decarbonization only through socioeconomically costly deindustrialization. This tension is resolved by synthesizing a viable “middle-ground” pathway, logically implied by the driver hierarchy. JTAS fails because rapid EA growth overwhelms mitigation gains, while the GA-optimized pathway suppresses EA unsustainably. A viable alternative must therefore consciously avoid the extremes of both scenarios:

• From JTAS, it adopts the goal of industrial development but tempers the unrealistically high EA growth to a moderate, sustainable level, reducing the mitigation burden.
• From the GA-optimal pathway, it adopts the imperative of deep decarbonization but achieves it not through contraction, but by channeling political and financial resources toward the core of the problem: aggressive, mandatory action on the RCF.

A practical and equitable pathway would thus accept slightly lower but stable EA growth to reduce the mitigation burden, while channeling policy and finance toward mandatory industrial fuel switching, reflected in a sharply declining RCF. This requires prioritizing green hydrogen, biomass, and other zero-carbon heat substitutes, supported by strong, concurrent improvements in EI and REN. Such a synthesized pathway recognizes that absolute decoupling ultimately depends on a managed transformation of the industrial substrate itself.

5.1.2. Policy Recommendations

Grounded in these findings, targeted interventions are proposed to align with the key drivers of emissions and ensure that the JETP delivers credible decarbonization. The recommendations address the critical challenges identified in the RCF, the interconnected dynamics of ELE and REN, and the management of EA and ST to prevent a destructive trade-off:

Mandate Industrial Fuel Switching with Financial De-Risking

First, a Green Hydrogen/Biomass Procurement Mandate should be introduced for high-temperature industrial processes such as iron and steel, and chemicals. To ensure viability, this mandate must be paired with a Carbon Contract for Difference (CCfD) scheme, which guarantees a fixed carbon price for investors by bridging the cost gap between fossil fuels and green alternatives. By de-risking capital investment, this intervention directly addresses the largest driver identified in this study: the Residual Carbon Factor (RCF). This approach is critical to overcome the implementation barriers identified in transition regions. As Xaba [60] demonstrates, without such place-based financial de-risking mechanisms, even well-designed national policies fail to gain traction in coal-dependent industrial areas. Rather than relying on uncertain market-led change, this strategy proactively creates a market for green industrial fuels and aligns with the Hydrogen Society Roadmap advanced by the Department of Science and Innovation [45].

Link Grid Investment to Industrial Electrification Through “Green Industrial Hubs”

New renewable energy capacity and grid stability investments should be explicitly tied to the establishment of Green Industrial Clusters. These hubs would provide manufacturers with preferential, long-term green electricity tariffs and guaranteed supply, conditional on commitments to electrify production processes. This approach resolves the historical ineffectiveness of electrification by directly coupling grid greening (REN) with industrial electrification (ELE), ensuring that renewable expansion displaces fossil fuel use within industry. This is particularly urgent given the institutional context. Phalatse’s [61] historical analysis of Eskom underscores that decades of underinvestment and governance challenges necessitate ring-fenced, dedicated infrastructure channels to ensure transition investments are effective and protected from systemic risks. By creating these guaranteed demand clusters, we address both the technical imperative and the institutional reality of South Africa’s energy landscape.

Implement Differentiated Industrial Policy to Foster Low-Carbon Growth

Industrial policy should focus on supporting cleaner, competitive sectors instead of treating manufacturing as a single block. Low-energy industries like automotive, food and beverages, and green technology manufacturing can be boosted with tax breaks, R&D grants, and export support. At the same time, carbon-intensive sectors should face gradually tighter carbon limits under the Climate Change Act to push them toward fuel switching and innovation. This differentiated approach prevents the “deindustrialization-led decoupling” identified in our GA-optimized scenario and instead builds an industrial base that grows while cutting emissions. This strategy aligns with the empirical finding that targeted manufacturing investment can generate significant employment and GDP multipliers, particularly in sectors like agro-processing and automotive [57].

However, this study also has inherent limitations that point toward necessary future work. The sector-level focus obscures important sub-sectoral heterogeneity. Therefore, future research should pursue disaggregated, techno-economic analyses of decarbonization pathways for energy-intensive industries such as steel and chemicals, while also integrating the socio-political feasibility considerations highlighted by [60]. This will better bridge the gap between modelling exercises and the realities of a just transition.