Thermal-runaway off-gas is not a single fixed mixture
When a lithium-ion battery undergoes abnormal heating, electrolyte evaporation and decomposition, breakdown of electrode interphase layers, reactions between the charged anode and electrolyte, and thermal decomposition of the cathode can overlap to produce gases and vapours. Often called thermal-runaway gas in the field, the actual effluent can include not only permanent gases such as CO₂, CO, and H₂, but also hydrocarbons such as methane, ethylene, ethane, and propylene; vaporised carbonate-based electrolytes; fine particles; and fluorinated substances. The concentration of a single component therefore cannot explain every cell or every stage of failure.
The study of 51 large automotive cells by Koch et al. identified CO₂, CO, H₂, C₂H₄, CH₄, C₂H₆, and C₃H₆ as the seven most common components. Yet that same study found that cell capacity, energy density, and case format affected emission quantity, thermal-runaway onset temperature, and mass loss. The important conclusion is that, although a representative list of components can be compiled, a universal composition ratio is difficult to establish. Facility hazard assessment requires actual test data for the cells in use.
Venting and thermal runaway are not the same event
Venting is a mechanical event in which internal cell pressure exceeds the design limit and a safety vent or rupture disc opens, releasing internal materials. Thermal runaway is a thermal event in which the rate of exothermic reactions exceeds heat-rejection capacity and temperature rises with self-acceleration. Venting may occur before thermal runaway, but a venting event does not necessarily mean that the cell has entered thermal runaway. Conversely, depending on the damage mode and cell construction, the interval between the two events can be so short that field measurements show them as nearly simultaneous.
In experiments by Abd-El-Latif et al. combining ARC with online mass spectrometry, an interval of about 50 minutes was observed between venting and thermal runaway in new NCA cylindrical cells. However, the separation between venting and vigorous reaction was much less distinct in aged cells with lithium plating from low-temperature cycling and in overcharged cells. Venting temperature and thermal-runaway onset temperature also varied with cell history in the same study. These results show both that gas alarms may provide a useful leading signal and that the available response time is not a fixed value.
Where the principal components originate
CO₂ can form when the solid electrolyte interphase (SEI) on the anode surface becomes unstable, when carbonate-based electrolyte reacts with its decomposition products, or when carbonate species on electrode surfaces decompose. It is detected relatively often even in early abnormality stages and is readily distinguishable from the atmospheric background, so it has been widely studied as a candidate for early monitoring. However, the non-flammability of CO₂ itself must not be used to conclude that the entire gas mixture is safe. Flammable vapours and toxic components may coexist with CO₂, and oxygen displacement must also be considered in enclosed spaces.
CO is formed principally when organic electrolyte is incompletely oxidised or decomposed under oxygen-deficient conditions. Even if oxygen is supplied within the cell during thermal runaway, reaction zones are not uniform, and composition changes again after discharge as the gas mixes with air or passes through flame. Directly comparing the CO fraction sampled in a sealed inert atmosphere with that measured in an open fire test can therefore be misleading. Because CO is both toxic and flammable, it must be included in assessments of both life safety and explosion risk.
H₂ can be generated through several pathways, including reactions between the charged anode and electrolyte, high-temperature reactions involving binder or solvent, and side reactions involving trace moisture. Its low molecular mass allows rapid diffusion, but it may accumulate in stagnant spaces at ceilings or above battery racks. In particular, it cannot be assumed that H₂ is low simply because the chemistry is LFP. In some comparative tests, H₂ represented a large molar fraction of LFP off-gas. By contrast, some nickel-based cells have shown a more pronounced share of CO or CO₂ and more intense high-temperature discharge.
Hydrocarbons and electrolyte vapours must also be considered. Ethylene can arise from decomposition pathways of ethylene carbonate and the SEI; methane, ethane, propylene, and related compounds can arise from chain reactions involving organic solvents and electrodes. Low-boiling electrolytes such as DMC and EMC may be discharged first as vapours before they decompose. A 2025 time-resolved measurement study reported that DMC, EC, and HF were detected first immediately after the vent opened in NMC and LFP cylindrical cells, and noted that condensable electrolyte components must also be included in flammability calculations for the mixed gas. The sequence in that study is an observation for those cells and heating conditions, not a fixed sequence for all batteries.
What differs by cell chemistry
Cathode chemistry affects thermal stability, oxygen release, reaction temperature, and total off-gas volume. Because its phosphate framework is comparatively stable, LFP has a lower tendency than layered oxides to release oxygen from the cathode. This does not mean that off-gas hazards disappear. LFP cells also contain electrolyte and a charged anode and can generate H₂, CO, CO₂, and hydrocarbons; if these accumulate in an enclosed space before ignition, they can form a mixture capable of deflagration.
Nickel-based layered oxides such as NMC and NCA can become structurally unstable at high state of charge and temperature, accompanied by oxygen release. Reactions between released oxygen and electrolyte can accelerate heat release and gas generation. However, nickel content alone cannot predict final composition, because electrolyte formulation and additives, anode material, separator, cell size, and pressure-relief design all act together.
The comparison of SOC and overcharge in commercial LFP and NCA 18650 cells by Golubkov et al. showed that state of charge, as well as chemistry, substantially changes thermal-runaway intensity and off-gas volume. In addition, a 2023 Cell Reports Physical Science study comparing four initiation modes and two cell formats found that the principal components overlapped as CO, CO₂, H₂, C₂H₄, and CH₄, but pressure, gas quantity, and debris differed among overcharge, lateral heating, oven heating, and nail penetration; overcharge presented the greatest hazard. Comparisons between chemistries must always state SOC and initiation conditions together.
Read sequence as a stage model, not a fixed order
In practice, the following conditional stage model is useful. It is not a sequence chart claiming that a particular molecule always precedes another; it is a framework for understanding how pressure, temperature, and discharge rate can change.
During the abnormal-reaction and internal-gas-accumulation stage, side reactions involving the SEI and electrolyte begin and internal cell pressure rises. Gas may not yet be detected outside the cell because, although it has already formed inside the sealed can, the vent has not opened.
During the first venting stage, vaporised solvents such as DMC, EMC, and EC, CO₂, small amounts of hydrocarbons, and fluorinated species may be discharged. Some tests detect gas before changes in temperature or voltage, but sensor placement and transport delay can make the reverse appear true.
During self-heating acceleration and full thermal runaway, chain reactions involving electrodes and electrolyte intensify, and CO₂, CO, H₂, ethylene, methane, and other substances are expelled at high flow rates. Particles and droplets are released together, and the safety vent may reopen or the case may rupture.
During ignition and post-reaction, off-gas mixes with air and burns. Depending on whether ignition occurs and on the ventilation rate, CO may oxidise to CO₂, or CO and soot may increase through incomplete combustion. Measurements at this stage reflect not only products generated inside the cell but also changes caused by external combustion.
Even these four stages are not always clearly separated. Ageing and lithium plating, high SOC, and abrupt internal short circuits compress the interval between first venting and thermal runaway. Pouch cells may lack a defined safety vent like that of cylindrical cells and may leak continuously at the seal. In modules, off-gas from one cell heats neighbouring cells, causing multiple venting events to overlap. Declaring a universal molecular order such as CO₂ followed by H₂ and then CO therefore makes a claim stronger than the current experimental evidence.
Why measured values differ
When reading off-gas studies, first check the sampling boundary. Gas collected in a bag after the test has lost temporal information, whereas real-time FTIR or mass spectrometry near a high-temperature discharge outlet shows transient changes but may miss components that condensed or reacted in the sampling line. GC is effective for quantifying major permanent gases and hydrocarbons, but its sampling time and target analytes are limited. Filters protect instruments, but substances attached to droplets and particles may be excluded from the analysis.
The denominator also matters. The mole fraction excluding nitrogen in a nitrogen-filled chamber, the actual volume fraction in a chamber containing air, and the moles generated per cell capacity are different values. It is also necessary to record whether off-gas passed through flame, whether water spray or extinguishing agent was used, and whether the sample was dried. Full-scale fire experiments by Larsson et al. confirmed that fluorinated toxic substances such as HF and POF₃ can be released in battery fires and showed that release quantities vary with chemistry, SOC, and test configuration. A system that measures only CO₂, CO, and H₂ may capture a substantial part of the explosion hazard but cannot represent the entire toxicity risk.
Implications for ESS monitoring and response
Gas monitoring may provide an earlier signal than heat or smoke detection, but a threshold from a single sensor must not be interpreted immediately as confirmation of thermal runaway. CO₂ is a relatively consistent candidate, but human respiration, outside air, and fire-suppression systems can also change its concentration. H₂ and CO sensors may be cross-sensitive to other reducing gases, and although VOC sensors can respond quickly to electrolyte vapours, they do not directly identify which cell has failed.
Accordingly, in ESS applications it is safer to examine rate of change and spatial distribution rather than a single gas species, and to cross-check against temperature, voltage, smoke, pressure, or acoustic information. Sensors should be located with expected flow paths, stagnant areas above racks, and ventilation inlets in mind. Alarm logic should be staged, for example as caution, suspected venting, multi-signal confirmation, and emergency response; for each stage, the sequence for stopping charge/discharge, isolating the area, remote verification, ventilation control, and notifying fire services should be defined in advance. Ventilation that simply discharges flammable gas outdoors must also be designed with ignition sources and outlet safety in view.
The most reliable basis is test data for installed cells and modules under conditions close to actual use. Suppliers should be asked not only for the chemistry name, but also for total off-gas quantity and composition by SOC, the timing of first venting and thermal runaway, test atmosphere and analytical method, and variation across repeat tests. Where data are unavailable, literature values should be considered conservatively as a range rather than adopted as a single design value.
Key takeaways
Thermal-runaway off-gas generally includes CO₂, CO, H₂, and various hydrocarbons, while vaporised electrolyte and fluorinated substances are also important. Venting is a pressure-relief event, whereas thermal runaway is a self-accelerating exothermic event, so the two must be distinguished. Multiple experiments have observed a stage in which solvent vapours and some decomposition gases emerge during first venting, followed by increased discharge rate and a broader component range during thermal runaway; however, a universal generation order for particular molecules has not been demonstrated. Chemistry, SOC, ageing, initiation mode, cell format, atmosphere, and analytical method all change the result. Sound safety interpretation does not start with memorising a single sequence. It starts by confirming which cell was measured, under what conditions, and by what method, then validating alarm and response criteria with multiple signals and actual product testing.