A 501(c)(3) non-profit organization Applied AI research for public benefit
CheAI Research Inc.

Across four kinetic mechanisms, ammonia/hydrogen flames reach the burning velocity of methane at hydrogen fractions of 0.443 to 0.528

We scored four open ammonia/hydrogen mechanisms against measured flames and ignition delays, then computed with each the blend that burns as fast as methane.

CheAI Research Inc. | Research article | December 2025

Abstract

Hydrogen is blended into ammonia to raise its low burning velocity. The blend that matches methane is a first design number for burner conversion, and designers usually compute it with a kinetic mechanism. We ran four open NH3/H2 mechanisms unmodified in Cantera and compared them with 218 laminar burning velocities and 115 shock-tube ignition delays. A selection rule fixed before the comparison picks Zhang 2021, with a 3.56 % mean absolute percent error near stoichiometry at atmospheric pressure, ahead of Stagni 2020 at 6.77 %. Otomo 2018 and Okafor 2018 under-predict the burning velocity by 13-18 % on average at H2 fractions of 0.2 and above. The blend reaches the methane value of 36.3 ± 0.6 cm/s at an H2 mole fraction of 0.443-0.528 across the four mechanisms. This spread is about eleven times the width of the interval produced by the methane uncertainty. Ignition delay ranks the mechanisms differently, and Zhang 2021 ignites early by a factor of about 1.5.

1. Introduction

Ammonia is liquid under modest pressure, carries hydrogen in a dense form and burns without producing CO2 [1], but it burns slowly. A stoichiometric NH3/air flame at ambient conditions has a laminar burning velocity of 6-7 cm/s [2], [3], against about 36 cm/s for methane [2]. Blended hydrogen raises the burning velocity steeply [2], [3].

Burner conversion needs the hydrogen fraction at which the blend burns as fast as methane. At φ = 1 and atmospheric pressure, the finely spaced data series stops just short of methane [2], and the other series reach it in steps of 0.1-0.2 [3], [4]. Designers therefore compute it with a mechanism. Earlier assessments found the predictions of NH3 mechanisms scattered [5] and none satisfactory across all conditions [6]. We asked how far this number moves when one mechanism replaces another. It moves from 0.443 to 0.528, and the mechanism selected on the data gives 0.460.

2. Methods

We compared the four open mechanisms that loaded in Cantera without hand edits, Stagni 2020 [7], Otomo 2018 [8], Zhang 2021 [9] and Okafor 2018 [10]. No parameter was changed. Burning velocities came from adiabatic, freely propagating one-dimensional flames in Cantera 3.2.0 [11], with multicomponent transport and Soret diffusion. Ignition delays came from constant-volume reactors, or from the measured volume history for the Shu shock tube.

The burning-velocity data are four primary sets, from the heat-flux burners of Han [2] and Wang [4] and the spherical flames of Lhuillier [3] and Gotama [12]. A digitized set by Ichikawa [13] was simulated and plotted but kept out of every score. The declared scope is 218 points at 298 K, 0.1-0.5 MPa, equivalence ratio φ = 0.7-1.4 and H2 fractions of 0-0.6, where

xH2 = XH2 / (XNH3 + XH2).

The ignition data comprised 115 shock-tube points at φ = 1 from Chen [14], Mathieu [15] and Shu [16].

The burning-velocity error of each data set is the mean absolute percent error,

MAPE = (100/n) Σ |SLsim - SLexp| / SLexp.

The score is the equal-weight mean of these per-set values. The ignition error is the mean of |log10(τsim/τexp)|. Before solving any flame we fixed the rule. The mechanism with the lowest score in the design band (76 points at 298 K, atmospheric pressure, φ = 0.9-1.1) gives the design number. Ignition does not enter the selection.

At φ = 1, 1 atm and 298 K, the design number xH2* is the H2 fraction that gives the CH4/air value measured on the Han burner, 36.3 ± 0.6 cm/s [2]. The ±0.6 cm/s has no stated confidence level, so its end points served as a sensitivity. Two other CH4 values, 35.03 cm/s [17] and 36.5 cm/s at 300 K [18], gave a wider sensitivity. The second was taken from its ReSpecTh database record, and its primary value was not checked.

3. Results

3.1 Two mechanisms follow the data and two burn too slowly

At φ = 1 and atmospheric pressure, Zhang 2021 and Stagni 2020 follow the measured burning velocity at every H2 fraction where data exist (Figure 1). Otomo 2018 and Okafor 2018 fall below the data from xH2 ≈ 0.2 upward.

Burning velocity against hydrogen fraction.
Figure 1. Burning velocity of stoichiometric NH3/H2/air at atmospheric pressure. Symbols are measurements with stated uncertainty, open triangles the unscored Ichikawa set, and the grey line the CH4/air value of the Han burner.

The design-band scores are 3.56 % for Zhang 2021, 6.77 % for Stagni 2020, 17.20 % for Otomo 2018 and 20.39 % for Okafor 2018. At xH2 ≥ 0.2 across all in-scope points, Zhang 2021 is fast by 2-4 % on average and Stagni 2020 by 8-9 %. Otomo 2018 is slow by 13-18 % and Okafor 2018 by 16-18 %.

3.2 The selection holds against point scatter but not always against calibration error

Zhang 2021 is selected with a margin of 3.21 percentage points (pp) over Stagni 2020, not a tie under the declared 1 pp rule. Dropping any one data set leaves Zhang 2021 first, with margins of 2.55-4.71 pp. We then perturbed the measured design-band values in 105 random draws, reading the stated uncertainty as one or two standard deviations. With independent errors per point, Zhang 2021 was first in every draw. With one shared error per data set and the one-standard-deviation reading, the margin fell below 1 pp in 11.9 % of draws, and the order reversed in 3.0 %. The perturbations add scatter to measurements that already contain error, so these results describe sensitivity and are not confidence intervals on the margin.

3.3 The mechanism moves the design number far more than the methane reference

The flame reaches 36.3 cm/s at xH2* = 0.443 with Stagni 2020, 0.460 with Zhang 2021, 0.499 with Otomo 2018 and 0.528 with Okafor 2018 (Figure 2). The spread is 0.085. Moving the target to the ends of the methane uncertainty shifts each root by 0.003-0.005, and the mechanism spread is about eleven times the mean width of that interval (0.0078). Refining the grid at each root changed its burning velocity by at most 0.134 %.

Methane-equivalent hydrogen fractions by mechanism.
Figure 2. Methane-equivalent H2 fraction at φ = 1, 1 atm and 298 K. Panel a shows the simulated burning velocity near the crossing, with the methane value 36.3 ± 0.6 cm/s shaded. Panel b shows the roots for 36.3 cm/s (dots), with thick bars for 35.7-36.9 cm/s and thin bars for 35.03-36.5 cm/s. The dashed line is the linear interpolation of the measured series. †Shrestha 2021 [19], not validated against burning-velocity data.

Linear interpolation of the measured series gives a mean of 0.442. If the true curve is convex, as every mechanism predicts, this estimate is low, but the data alone do not establish that bias. The finely spaced Han series measures 33.9 cm/s at xH2 = 0.45, so its data place the crossing above 0.45.

3.4 Ignition delay ranks the mechanisms differently

Otomo 2018 is closest to the shock-tube data, with a mean |log10| error of 0.105, followed by Stagni 2020 (0.121), Zhang 2021 (0.190) and Okafor 2018 (0.363). Zhang 2021 ignites early on average and late at only 5 of the 115 points. Its mean signed error of -0.188 corresponds to a geometric-mean delay about 1.5 times too short.

4. Discussion

The mechanism, more than the methane target or the numerics, sets the methane-equivalent H2 fraction. The spread comes from the two slow mechanisms. At the local slope, a deficit of 13-18 % takes 0.03-0.05 more H2 to recover, the same order as their 0.04-0.07 offset from Zhang 2021. Discarding the slow pair on the data, as the selection rule does, removes most of the spread.

The 0.085 spread agrees with earlier reports of scattered mechanism predictions [5], [6]. Zhang 2021 has the lowest near-stoichiometric score, and Stagni 2020 is a credible alternative, because a set-wide calibration error of the stated size ties or reverses their order in about one draw in eight. For φ = 1, 1 atm and 298 K we therefore recommend the interval of their central predictions, 0.44-0.46, with 0.460 from the selected mechanism. For applications governed by autoignition, such as engines or reheat combustors, the choice should rest on ignition data, where Otomo 2018 or Stagni 2020 is the better starting point.

5. Limitations

Only mechanisms that load unmodified in Cantera were compared. Glarborg 2018 [20] needs hand edits, and Shrestha 2021 was too slow, so its root in Figure 2 is not validated. The grid check covers the comparison flames and the 20 design roots, but not the Shrestha 2021 roots. Stagni 2020 is the 2020 CRECK release, not the 2023 revision, and newer mechanisms may do better.

On the ignition side, the H2-blend data come from one group and stop at xH2 = 0.3, so the ignition ranking is mostly a neat-NH3 ranking. The Chen delay definition rests on a later paper from the same facility, and the Chen records give no facility pressure-rise rate. An illustrative correction borrowed from a later study of a different mixture may not transfer. The Shu delays depend in part on a volume held constant after the measured trace ends. On the burning-velocity side, the stated uncertainties have different and partly unstated bases. The perturbation test brackets them but does not resolve them. Lean flames at 0.5 MPa can also become cellular, and Gotama flags those points [12].

The spread across four mechanisms is not a confidence interval. The mechanisms are not independent samples, and some were validated against data that overlap with the sets used here [9]. The recommended interval carries neither measurement nor kinetic-parameter uncertainty. The design number is for one condition, φ = 1, 1 atm and 298 K. Equal burning velocity is also only one part of fuel interchangeability, and we did not examine flame thickness, stretch response, ignition behaviour or Wobbe index.

6. Conclusions

Four open NH3/H2 mechanisms disagree on the methane-equivalent H2 fraction mainly because two of them burn too slowly. Zhang 2021 and Stagni 2020 fit the near-stoichiometric data best and place the blend at 0.46 and 0.44 H2 mole fraction in the fuel at φ = 1, 1 atm and 298 K. Otomo 2018 and Okafor 2018 need 0.50-0.53. A mechanism chosen on burning velocity should not be trusted for ignition, because the order of the four changes and the selected mechanism ignites about 1.5 times too early.

References

  1. Valera-Medina A, Xiao H, Owen-Jones M, David WIF, Bowen PJ. Progress in Energy and Combustion Science 69, 63-102, 2018. DOI 10.1016/j.pecs.2018.07.001
  2. Han X, Wang Z, Costa M, Sun Z, He Y, Cen K. Combustion and Flame 206, 214-226, 2019. DOI 10.1016/j.combustflame.2019.05.003
  3. Lhuillier C, Brequigny P, Lamoureux N, Contino F, Mounaïm-Rousselle C. Fuel 263, 116653, 2020. DOI 10.1016/j.fuel.2019.116653
  4. Wang S, Wang Z, Elbaz AM, Han X, He Y, Costa M, Konnov AA, Roberts WL. Combustion and Flame 221, 270-287, 2020. DOI 10.1016/j.combustflame.2020.08.004
  5. da Rocha RC, Costa M, Bai X-S. Fuel 246, 24-33, 2019. DOI 10.1016/j.fuel.2019.02.102
  6. Girhe S, Snackers A, Lehmann T, Langer R, Loffredo F, Glaznev R, Beeckmann J, Pitsch H. Combustion and Flame 267, 113560, 2024. DOI 10.1016/j.combustflame.2024.113560
  7. Stagni A, Cavallotti C, Arunthanayothin S, Song Y, Herbinet O, Battin-Leclerc F, Faravelli T. Reaction Chemistry & Engineering 5, 696-711, 2020. DOI 10.1039/c9re00429g
  8. Otomo J, Koshi M, Mitsumori T, Iwasaki H, Yamada K. International Journal of Hydrogen Energy 43, 3004-3014, 2018. DOI 10.1016/j.ijhydene.2017.12.066
  9. Zhang X, Moosakutty SP, Rajan RP, Younes M, Sarathy SM. Combustion and Flame 234, 111653, 2021. DOI 10.1016/j.combustflame.2021.111653
  10. Okafor EC, Naito Y, Colson S, Ichikawa A, Kudo T, Hayakawa A, Kobayashi H. Combustion and Flame 187, 185-198, 2018. DOI 10.1016/j.combustflame.2017.09.002
  11. Goodwin DG, Moffat HK, Schoegl I, Speth RL, Weber BW. Cantera, version 3.2.0. Zenodo, 2025. DOI 10.5281/zenodo.17620923
  12. Gotama GJ, Hayakawa A, Okafor EC, Kanoshima R, Hayashi M, Kudo T, Kobayashi H. Combustion and Flame 236, 111753, 2022. DOI 10.1016/j.combustflame.2021.111753
  13. Ichikawa A, Hayakawa A, Kitagawa Y, Somarathne KDKA, Kudo T, Kobayashi H. International Journal of Hydrogen Energy 40, 9570-9578, 2015. DOI 10.1016/j.ijhydene.2015.04.024
  14. Chen J, Jiang X, Qin X, Huang Z. Fuel 287, 119563, 2021. DOI 10.1016/j.fuel.2020.119563
  15. Mathieu O, Petersen EL. Combustion and Flame 162, 554-570, 2015. DOI 10.1016/j.combustflame.2014.08.022
  16. Shu B, Vallabhuni SK, He X, Issayev G, Moshammer K, Farooq A, Fernandes RX. Proceedings of the Combustion Institute 37, 205-211, 2019. DOI 10.1016/j.proci.2018.07.074
  17. Xiouris C, Ye T, Jayachandran J, Egolfopoulos FN. Combustion and Flame 163, 270-283, 2016. DOI 10.1016/j.combustflame.2015.10.003
  18. Varghese RJ, Kumar S. International Journal of Hydrogen Energy 45, 3216-3232, 2020. DOI 10.1016/j.ijhydene.2019.11.134
  19. Shrestha KP, Lhuillier C, Barbosa AA, Brequigny P, Contino F, Mounaïm-Rousselle C, Seidel L, Mauss F. Proceedings of the Combustion Institute 38, 2163-2174, 2021. DOI 10.1016/j.proci.2020.06.197
  20. Glarborg P, Miller JA, Ruscic B, Klippenstein SJ. Progress in Energy and Combustion Science 67, 31-68, 2018. DOI 10.1016/j.pecs.2018.01.002

Related studies