, a mass of 0.89 M ʘ , a composition of equal parts by mass of carbon and oxygen and an initial temperature of 5× 10 5 K. The initial orbit is circular with a period of 28s. This corresponds to the state when tidal forces deform the white dwarfs sufficiently to make the system unstable (prior evolution is driven by gravitational wave emission and is not simulated). This progenitor system is set up as the initial conditions for a simulation using a smoothed particle hydrodynamics code 12 (SPH). In the initial conditions, marginal asymmetries were deliberately introduced since perfect symmetry is not expected in Nature. In this first simulation, we follow the 3 inspiral and subsequent merger of the binary system (Figure 1). After two orbits one of the two white dwarfs is disrupted. This unequal evolution of the white dwarfs originates from the symmetry-breaking in the initial conditions. The disrupted white dwarf violently merges with the remaining white dwarf and material is heated by compression.In the hottest regions carbon burning begins and releases additional energy which further heats the material. A hotspot, which is resolved by several SPH particles, forms with a temperature of 2.9× 10 9 K in high-density material (3.8× 10 6 g cm -3). Highresolution small-scale simulations 13 show that under such conditions a detonation ignites.In the second step of our simulation sequence, we impose the triggering of a detonation at the hottest point and follow it with a grid-based hydrodynamics code 14,15 as it crosses the merged object. The energy release from the nuclear burning in the detonation disrupts the system (see Figure 1) Finally, we use the structure of the ejecta and the detailed chemical abundances to calculate synthetic light curves and spectra using a Monte Carlo radiative transfer code 19 as required to quantitatively test this model against observations. Owing to the small 56 Ni mass synthesized during the nuclear burning, the synthetic light curves (Figure 2) are faint and decline rapidly compared to those of normal SNe Ia, despite the large total ejecta mass of our simulation (1.8 M ʘ ). Given that there has been no fine-tuning of the explosion model, the light curves agree remarkably well with those of the 1991bg-like SNe Ia -in both absolute magnitude and colour evolution. Moreover, our model naturally predicts the lack of secondary maxima in the near-infrared (J, H and K) light curves which is a peculiarity of 1991bg-like objects compared to normal SNe Ia.In detail, however, there are some discrepancies between our model light curves and the observational data. Comparing the difference in brightness at B band maximum and 15 days thereafter we find values between 1.4 and 1.7 -depending on the line-ofsight. This is less than typically observed for 1991bg-like objects (1.9), but at worst similar to the fastest declining normal SNe Ia and substantially faster than for objects which have previously been claimed as possible super-Chandrasekhar explosions (e.g. The total mass of the system is essential...