GREEN: A Multi-Domain Sustainability Framework

• GREEN is a cross-domain sustainability framework that integrates practices in business, software, AI, and urban planning to optimize environmental, economic, and social performance.
• It formalizes sustainability using measurable indices like the Green Quotient and lifecycle impact assessments to transform vague environmental claims into explicit metrics.
• Practical applications include green computing, green marketing, green job creation, and urban governance strategies that balance long-term viability with reduced environmental burden.

In contemporary research, “GREEN” denotes a family of sustainability-oriented constructs rather than a single invariant category. In business, it refers to long-term viability organized around the triple bottom line of people, planet, and profit rather than short-term profit alone (Kabiraj et al., 2010). In software and AI, it denotes lower energy consumption, lower associated carbon emissions, and explicit optimization of efficiency alongside functional performance (Mehra et al., 2022, Schwartz et al., 2019). In labor markets, industrial policy, urban governance, and energy planning, it refers to green jobs, green qualifications, low-impact infrastructures, and multi-objective transition paths that reduce environmental harm without ignoring economic and social constraints (Sulich et al., 2021, Salis et al., 16 Jul 2025, Schnidrig et al., 2024). This suggests that GREEN is best understood as a cross-domain framework for reducing environmental burden while preserving, and sometimes improving, technical, organizational, and economic performance.

1. Conceptual foundations

In business research, GREEN is defined as more than environmental friendliness. It is tied to business sustainability, efficient use of energy and materials, responsible treatment of waste and emissions, and transparent environmental communication. A central formulation is the triple bottom line, expressed as performance in terms of people, planet, and profit, while green computing is defined as the efficient and eco-friendly use of computing resources across design, manufacture, use, and disposal or recycling (Kabiraj et al., 2010).

In marketing research, green marketing strategy is defined as activities designed to create and facilitate exchanges so that satisfying human needs and wants involves minimal harmful and destructive environmental impacts. This definition ties GREEN to the marketing mix through green product, green promotion, green distribution, and green pricing, and places environmental management within ordinary managerial practice rather than outside it (Nohekhan et al., 2024).

In labor-market research, GREEN appears as green jobs, green careers, and green qualifications. Green jobs are treated as elements of the economy’s transformation toward a green economy; green qualifications are the skills, knowledge, and competencies that translate sustainable development into business practice. In this setting, GREEN links work content, organizational reputation, and individual career choice (Sulich et al., 2021).

In AI and software, GREEN is explicitly operational rather than symbolic. Green AI is defined as research that yields novel results without increasing computational cost, and ideally reduces it, by making efficiency a first-class evaluation criterion alongside accuracy (Schwartz et al., 2019). For software projects, greenness refers to energy consumption and associated carbon emissions of the software system and its supporting infrastructure and processes, together with the adoption of greener practices across the software development lifecycle (Mehra et al., 2022).

In energy-system and urban-governance research, GREEN is broader than carbon alone. Swiss energy-system planning defines green performance through economic cost together with life-cycle indicators such as Carbon Footprint, Fossil and Nuclear Energy Use, Water Scarcity Footprint, Remaining Human Health Damage, and Remaining Ecosystem Quality Damage (Schnidrig et al., 2024). Urban green governance treats green spaces as strategic infrastructure—the “green lungs” of the city—managed through IoT systems, GIS, and decision-support workflows (Salis et al., 16 Jul 2025).

2. Formalization and measurement

A recurring feature of GREEN research is the replacement of vague environmental language with explicit lifecycle variables, composite indices, or constrained optimization. In business transformation, greenness is organized around a business life cycle of inputs, processes, outputs, environmental externalities, and marketing, with criteria such as energy intensity, resource intensity, renewable sources, recycled materials, greenhouse gas emissions, waste, green products, green services, green labels, and voluntary standards (Kabiraj et al., 2010).

Software-project greenness is formalized by the Green Quotient, a composite score on a continuous scale between 0 and 1. The Project Green Quotient decomposes into technology, process, metrics, and team dimensions, and can be written as a weighted composite index, $GQ = \sum_{d \in D} w_d \cdot GQ_d$, where each $GQ_d$ is itself an aggregation of normalized question scores (Mehra et al., 2022). This metric converts otherwise diffuse sustainability practices into an auditable SDLC-level assessment.

Green-washing detection research uses a different ratio. DeepGreen first counts green claims and then evaluates whether each claim is implemented. Its core variable is

$$GreenImplement = \frac{\sum_{k=1}^m LLM_B(G_I \mid u_k^{(\tau_k)})}{GreenKeywords},$$

so that the numerator is the number of green claims judged to be implemented and the denominator is the total number of green keywords detected in the statement (Xu et al., 10 Apr 2025). This makes GREEN measurable as the proportion of implemented rather than merely declared sustainability content.

Energy-system planning formalizes GREEN at the level of full life-cycle impacts. For each impact category $i$, total annual impact is defined as

$\mathrm{LCIA_{tot}(i)} = \sum_{\text{tec}} \big( \mathrm{LCIA_{stat}(i,\text{tec}) + \mathrm{LCIA_{var}(i,\text{tec})} \big),$

thereby integrating construction, operation, and end-of-life effects into energy-system optimization (Schnidrig et al., 2024). This is precisely the formal setting in which “burden shifting” can be observed: reducing one impact, such as climate change, while worsening others.

AI research uses yet another formal lens. Green AI highlights the simple cost model

$\text{Cost}(R) \propto E \cdot D \cdot H,$

where $E$ is per-example processing cost, $D$ is dataset size, and $H$ is the number of hyperparameter experiments. It further recommends reporting Floating Point Operations as a hardware-agnostic measure of computational work and, by extension, a proxy for energy and carbon intensity (Schwartz et al., 2019).

Wireless semantic communication makes the trade-off explicit through a semantic compression ratio,

$$\Omega_k = \frac{\sum_{j=1}^Q d_{kj}}{\sum_{k=0}^K u_k},$$

which links computation and communication energy. Higher semantic compression reduces transmission load but increases compression and inference overhead, so total energy is optimized jointly rather than at one layer alone (Xu et al., 2024).

3. Firms, disclosure, and market interfaces

Business-oriented GREEN research emphasizes that the shift is driven by rising energy costs, climate concerns, corporate social responsibility, regulation, consumer preference, and capital-market change. The business literature highlights renewable electricity standards, carbon-pricing schemes such as the European Emissions Trading Scheme and the Regional Greenhouse Gas Initiative, and the increasing use of carbon neutrality and CSR as strategic markers (Kabiraj et al., 2010). It also insists that going green must rest on “good business sense,” not charity.

The same literature presents concrete operational examples. IBM is reported to have saved 4.6 billion kWh of electricity since 1990 and avoided about 3 million metric tons of CO$GQ_d$0, while VMware is reported to have achieved \$5 million in annual savings through a “green IT” data center using virtualization and energy-efficient design (Kabiraj et al., 2010). These examples anchor GREEN in cost avoidance and operational efficiency rather than solely in symbolic commitment.

Green marketing research links environmental orientation to brand outcomes. In a survey of 182 employees and managers from food exporting companies, green marketing strategy had a strong positive Pearson correlation with brand awareness, $GQ_d$1, while green product, green promotion, green distribution, and green pricing showed positive correlations of $GQ_d$2, approximately $GQ_d$3, $GQ_d$4, and $GQ_d$5, respectively (Nohekhan et al., 2024). Within that sample, promotion and pricing were the strongest correlates of brand awareness.

At the same time, both business-sustainability and disclosure research warn against equating GREEN with labels. The business literature notes that proliferating green labels and standards can lose meaning when used as marketing devices rather than as evidence of actual environmental improvement (Kabiraj et al., 2010). DeepGreen addresses exactly this problem by distinguishing implemented green practice from repeated green language in 204 financial statements of 68 Chinese A-share firms over three years (Xu et al., 10 Apr 2025). Its empirical results show that green implementation significantly boosts asset return rate, but that this effect is heterogeneous: for small and medium-sized firms, the contribution to asset return is limited, which the authors interpret as a stronger motivation for green-washing (Xu et al., 10 Apr 2025).

4. Software, AI, and digital systems

In software research, GREEN is motivated by the claim that ICT is already responsible for an estimated 2–7% of global greenhouse gas emissions, with projections up to 14% by 2040 (Mehra et al., 2022). This makes software design decisions—not only hardware procurement—environmentally material. The Green Quotient framework therefore evaluates technology choices, development processes, metrics practices, and team behavior together, rather than treating energy efficiency as a narrow code-level issue (Mehra et al., 2022).

The same work provides highly specific examples of green software choices. Adding dark mode can reduce UI-related energy by about 64%; switching remote collaboration from video to audio can reduce meeting-related emissions by about 96%; and choosing Java instead of Python can yield about 97.4% energy savings in some scenarios (Mehra et al., 2022). The prototype was piloted with six software projects, and 83.33% of respondents answered positively when asked whether the Green Quotient metric and breakdown allowed the team to gauge overall sustainability (Mehra et al., 2022).

A complementary line of work treats GREEN as a full software lifecycle concern. HADAS identifies recurring runtime energy hotspots such as storage, communication, compression, security, and data access, models them as variability-aware concerns, and uses AspectJ-based self-adaptation to switch between greener implementations at runtime (Gamez et al., 2016). In its Media Store case study, switching from LAME to Vorbis for large locally stored files saves 48% energy at 128 MB and up to 65% at 512 MB; under remote storage, switching to Speex yields savings up to 52% relative to LAME for files of at least 64 MB and up to 81% for large files (Gamez et al., 2016).

In AI, GREEN emerged as a response to rapid compute escalation. Training compute for leading deep models increased by about 300,000× from 2012 to 2018, with cost doubling every few months, and Green AI argues that efficiency, price tag, and computational work should be reported alongside task metrics (Schwartz et al., 2019). More recent inference-time work extends this logic from model development to deployment: dynamic model selection through cascading and routing achieves energy savings up to about 25% while retaining up to about 95% of the accuracy of the most energy-greedy solution (Cruciani et al., 24 Sep 2025).

5. Networks, infrastructures, and urban-environmental systems

GREEN computing extends beyond individual models and programs into distributed infrastructures. In green cloudlet networks for mobile cloud computing, the GEAR strategy migrates user “Avatars” among cloudlets to align load with local renewable generation while respecting an SLA on propagation delay. In simulation, GEAR reduces on-grid energy consumption by about 35% relative to the delay-minimizing FAR strategy (Sun et al., 2015). The same principle appears in GRASP, an SDN platform that schedules jobs to geographically distributed data centers according to available solar power; under favorable load conditions it achieves roughly 15–16% higher green usage ratio than round robin (Grigoryan et al., 2018).

Cloud-continuum deployment research formalizes GREEN as adaptive orchestration under explicit carbon-aware constraints. In that setting, avoidNode(d(s,f), n) is generated when $GQ_d$6, while affinity(d(s,f), d(z,_)) is generated when $GQ_d$7, so placement and co-location decisions are driven directly by monitored service energy, inter-service communication, and node carbon intensity (D'Iapico et al., 20 Feb 2026). This converts sustainability from a retrospective KPI into a scheduler input.

Energy-system planning makes the same point at national scale. In Switzerland, optimizing key environomic indicators can reduce system costs by 15% to 47%; a system optimized solely for economic efficiency still achieves a 63% reduction in carbon footprint compared to 2020, but the study shows a potential risk of burden shifting to other environmental issues (Schnidrig et al., 2024). More extreme carbon-footprint minimization reaches an 81% reduction relative to 2020, while minimizing water scarcity footprint can reduce that indicator by up to 76% but with higher cost (Schnidrig et al., 2024). The central lesson is that GREEN cannot be reduced to low CO$GQ_d$8 alone.

Urban green governance applies similar logic at municipal scale. In Campobasso, a cloud-based decision-support platform integrates 20 Tree Talker devices, 2 weather stations, 2 air-quality sensors, 4 electro sensors, 4 hydro sensors, LoRaWAN gateways, 5G backhaul, GIS, and remote sensing. The platform computes vegetation indices such as

$GQ_d$9

tracks soil moisture, salinity, sap flow, trunk stability, and related variables, and issues alerts when thresholds are exceeded (Salis et al., 16 Jul 2025). The case study reports seasonal NDVI values from about $$GreenImplement = \frac{\sum_{k=1}^m LLM_B(G_I \mid u_k^{(\tau_k)})}{GreenKeywords},$$0 to $$GreenImplement = \frac{\sum_{k=1}^m LLM_B(G_I \mid u_k^{(\tau_k)})}{GreenKeywords},$$1, soil salinity typically between $$GreenImplement = \frac{\sum_{k=1}^m LLM_B(G_I \mid u_k^{(\tau_k)})}{GreenKeywords},$$2 and $$GreenImplement = \frac{\sum_{k=1}^m LLM_B(G_I \mid u_k^{(\tau_k)})}{GreenKeywords},$$3 dS/m, and successful intervention when abnormal trunk movement was detected and corrected through targeted pruning (Salis et al., 16 Jul 2025).

Wireless systems research adds a further layer: green semantic communication. There, a probabilistic knowledge graph, semantic compression ratio, and RSMA are jointly optimized to minimize total computation and communication energy under semantic accuracy and latency constraints (Xu et al., 2024). The system outperforms variants using SDMA or NOMA, indicating that GREEN in communication engineering increasingly means semantic-level, not merely bit-level, efficiency (Xu et al., 2024).

6. Labor markets, industrial transformation, and macroeconomic transition

In labor-market analysis, GREEN is tied to career choice under sustainable development. A survey of 1,045 students and graduates at Wroclaw University of Technology identified 11 key criteria for career-path choice, including possibility of career development, earnings, job–study match, skills and interests in green jobs, sustainable reputation of the company, and green experience possible to gain in the organization (Sulich et al., 2021). The corresponding Bellinger weights place career development first at $$GreenImplement = \frac{\sum_{k=1}^m LLM_B(G_I \mid u_k^{(\tau_k)})}{GreenKeywords},$$4, skills and interests in green jobs at $$GreenImplement = \frac{\sum_{k=1}^m LLM_B(G_I \mid u_k^{(\tau_k)})}{GreenKeywords},$$5, earnings at $$GreenImplement = \frac{\sum_{k=1}^m LLM_B(G_I \mid u_k^{(\tau_k)})}{GreenKeywords},$$6, sustainable reputation at $$GreenImplement = \frac{\sum_{k=1}^m LLM_B(G_I \mid u_k^{(\tau_k)})}{GreenKeywords},$$7, and green experience at $$GreenImplement = \frac{\sum_{k=1}^m LLM_B(G_I \mid u_k^{(\tau_k)})}{GreenKeywords},$$8 (Sulich et al., 2021). In the hypothetical ranking exercise, job position $$GreenImplement = \frac{\sum_{k=1}^m LLM_B(G_I \mid u_k^{(\tau_k)})}{GreenKeywords},$$9 is optimal with $i$0, and career development—not education match—emerges as the strongest driver (Sulich et al., 2021).

At the level of industrial structure, green expansion is neither purely incremental nor purely disruptive. Across 65 countries and 2007–2017 trade data, countries expand their green production baskets both through path dependency and through what the authors call high investment structural jumps (Talebzadehhosseini et al., 2019). The paper concludes that countries do not only add green products predicted by their existing capabilities; they also expand into products that path dependency does not predict by investing in innovating and developing new environmental-related technologies (Talebzadehhosseini et al., 2019). China is presented as the primary case of this mixed dynamic.

Macroeconomic work makes the transition logic more explicit. Green technology news shocks, identified from the economic value of green patents in a Bayesian VAR, decompose into a common technological component and an idiosyncratic green component (Jaulin et al., 24 Jul 2025). The common component behaves like a standard technology news shock with long-run productivity effects, whereas the idiosyncratic green component generates inflationary pressures and stock price reductions, which the authors interpret as a green transition news mechanism associated with expectations of stricter carbon policies or environmental standards (Jaulin et al., 24 Jul 2025).

A related evolutionary-competition model of firm transition shows that increasing transition risk, for example by threatening stricter environmental regulation, effectively incentivizes the green transition (Radi et al., 2024). The paper recommends maintaining high transition risk regardless of the industry’s level of greenness; it also finds that subsidizing the costs of adopting green technologies can reduce the risk of a failed green transition, while advances in green technologies amplify but do not eliminate that risk (Radi et al., 2024). Across these analyses, GREEN appears not as a frictionless equilibrium state but as a contested transition shaped by capabilities, incentives, adjustment costs, and the credibility of policy.

Across the literatures considered here, a consistent conclusion emerges: GREEN is substantive only when it survives measurement. In firms, this means implementation rather than labels; in software and AI, energy and compute reporting rather than accuracy alone; in infrastructures, lifecycle and carbon-aware constraints rather than narrow utilization targets; and in economic transition, multi-objective governance rather than carbon reduction pursued in isolation.