Constraints on the Delayed Transition to Detonation in Type Ia Supernovae A.M. Lisewski, W. Hillebrandt 0 0 Max-Planck-Institut fu¨r Astrophysik, Karl-Schwarzschild-Str. 1, 85740 Garching, Germany 0 2 n S.E. Woosley a J UCO/Lick Observatory,University of California Santa Cruz, Santa Cruz, CA 95064,USA 4 2 Received ; accepted 2 v 6 5 0 0 1 9 9 / h p - o r t s a : v i X r a – 2 – ABSTRACT We investigate the possibility of a delayed detonation in a type Ia supernova under the assumption that the transition to detonation is triggered by turbulence only. Our discussion is based on the Zeldovich mechanism and suggests that typical turbulent velocities present during the explosion are not strong enough to allow this transition to occur. Although we are able to show that in carbon-rich matter (e.g., X(12C) = 0.75) the possibility of a deflagration to detonation transition (DDT) is enhanced, even in this case the turbulent velocities needed are larger than the expected value of u′(L) ≈ 107cm s−1 on a length-scale of L ≈ 106 cm. Thus we conclude that a DDT may not be a common event during a thermonuclear explosion of a Chandrasekhar-masswhite dwarf. Subject headings: supernovae: general – shock waves – turbulence – 3 – 1. Introduction The question whether a detonation occurs during the thermonuclear explosion of a white dwarf of Chandrasekhar mass has already been discussed extensively. Many existing models assume that such a transition does occur in certain situations during the explosion. Some of these models, the so-called delayed-detonation models, assume that at a critical value of a model parameter (usually the density) a previously subsonic burning front turns into a detonation. In comparison with observables, like lightcurve shape, peak brightnessorelement abundances,the delayed-detonationscenarioseemedto comply well(e.g., Ho¨flich et al. 1997). However, a conclusive answer whether this transition is physically possible has not been given yet. In a thermonuclear supernova, the explosion energy is produced by a moving thermonuclear flame that burns carbon and oxygen into higher mass elements. According to our present understanding at least at highdensities, i.e. in the inner core(ρ>108g cm−3)of the white dwarf,the burning mode ofthis flame is a deflagration,where the burning frontcanlocally be described asa laminar flame. The properties oflaminar flames in white dwarf matter are well known (Timmes & Woosley 1992). One general feature is that with lower density the speed of a laminar flame decreases rapidly while at the same time its thickness increases. On the other hand, numerical models of type Ia supernovae taylored to reproduce lightcurves and spectra require a flame velocity that is nearly independent of the density in the unburned material. Thus there must be a mechanism that leads to an effective flame velocity being independent of its laminar properties. It is commonly believed that turbulence provides this mechanism. However, quantitative predictions of the turbulent flame velocity are not easy to obtain because they depend on many (often very uncertain) parameters such as density, nuclear compositions, turbulent properties, etc. It is not even clear if the flame speed remains always subsonic or if it might turn into a detonation. Leaving out scenarios like the prompt initial detonation(see, for example, Imshennik et al. 1999andreferences therein) andthe pulsating delayed detonation(e.g., Khokhlovet al. 1997),we focus onthe problemwhether turbulence aloneis able to trigger a supersonic burning wave accompanied by a hydrodynamic shock. In order to achieve this goal, we introduce a method to represent the structure of a flame moving through a turbulent medium. For a given physical state of the unburned matter, and for a specified turbulent energy, the model adopts the most optimistic assumptions for the transition from a deflagration to a detonation. We then investigate if such a transition happens under these special prerequisites. Cases where no transition is observed therefore provide quantitative limits on some relevant physical parameters, – 4 – like nuclear composition or turbulence intensity. These limits constrain the regime where a detonation triggered by turbulence might occur. This letter is structured as follows: First we briefly recall some properties of carbon-flames at densities around 107g cm−3 where according to our present understanding (see, e.g. Khokhlov et al. 1997 and Niemeyer & Woosley 1997) the chances of a DDT are the highest. Then we introduce our method to represent turbulence effects on subsonic flames. In this way we obtain some turbulent flame profiles which we eventually implement as initial conditions in a gas-dynamic system to study their temporal evolution. Finally, we discuss some implications of our results. 2. The model The structure of a flame moving through a turbulent medium crucially depends on the characteristic turbulent lengthscales, like the scale where the biggest turbulent motions are produced and the dissipation scale where turbulent motionis smearedoutby microscopicdiffusion, as well as onthe amountof turbulent kinetic energy that drives the turbulent motion. Of course, in case of turbulence driven by inertia these scales become the inertial scale L and the Kolmogorovscale l . On the other hand, it is important to know k how these typical turbulent entities relate to the given chemical and thermodynamical properties that are responsible for the burning process itself. Here we set up a model that represents this relation in a certain way. Consider a flame – laminar or turbulent–intheincompressiblecase,thatisaflamewithnopressurejumpacrossit. Thentheburnedand unburned states of matter are given by the density, temperature and nuclear composition of the unburned state and by the available amount of energy released by nuclear reactions under these conditions. The only difference between a turbulent and a laminar flame is the spatial and temporal distribution of density, temperature and nuclear abundances within the interface between fuel and ashes, i.e. the flame profile. For instance, in a laminar flame temperature is a monotonically decreasing function in the direction of the unburned material, whereas in the turbulent case fluctuations can cause this function to have sharp spikes and dips. Turbulence has two main impacts on the burning process. First, there is an enlargement of the flame surface caused by turbulent stretching and wrinkling leading to a higher effective flame speed. Second, in cases of higher turbulence intensity, vortices can even mix burned or burning material with unburned matter. This happens when they are fast enough to carry reactive material away from the flame before this – 5 – material is burned completely. As a direct consequence one observesa broadening of the flame profile. Now the flame structure is superposed by fluctuations coming from turbulent motion. However, in this work we neglecta directrepresentationoffluctuationsandinsteadconcentrateonone neteffectofthe latter,namely the enhanced heat and mass transport that actually causes the broadening of the flame. The resulting spatial structure of the flame and in particular its size play a crucial role in the DDT problem. The profile of an incompressible laminar flame is described by the set of functions T(x;t), ρ(x;t), l l Y (x;t), which give the spatio-temporal distributions of temperature, density and nuclear number density i,l of the ith species at a constant pressure P. These functions are solutions of the following system of equations: ∂Y i,l = −Y Y λ (i)+Y Y λ (i), (1) i,l k,l jk i,l k,l kj ∂t Xj,k ∂T 1 ∂ ∂T ∂ 1 1 l = σ l −P + S˙, (2) ∂t ρc ∂x(cid:18) ∂x(cid:19) ∂tρ c p p dY S˙ =N i,lB . (3) A i dt Xi Herein, c = c (T,ρ,Y ) is the heat capacity, σ = σ(T,ρ,Y ) the thermal conductivity, B the binding p p l l i,l l l i,l i energy and Y is the number density of the ith nucleus while λ denotes the the reaction rate of nuclei i i,l ij and j. Our reaction network consists of seven species, viz. 4He, 12C, 16O, 20Ne, 24Mg, 28Si, and 56Ni. The direction of the flame is chosen to be such that the completely burned state is at x = −∞. In a steady state the flame represented by these equations propagates with a constant velocity s. Thus in a l co-movingframe the functions T, ρ and X are constantin time and it is convenient to place the originof l l i,l the co-moving coordinate system at the point of maximum energy generation. The flame profile defines also a spatialdistribution of the nuclear reactiontimescale of the ith element, τ (i), by m Y i τ (i)= , (4) m (cid:18)Y˙ (cid:19) i T,ρ where the time derivative is taken at constant temperature and density. In this work we are particulary interested in C+O matter at densities around 107g cm−3, where by far the most relevant nuclear reaction is 12C+ 12C. Burning of heavier elements takes place on timescales much longer than τ (12C) and releases m much less energy. We will therefore always refer to carbon-burning in the following. Note that the equation (4) oversestimatesthe nuclear burning time, because burning a small fraction of the givencarbonalreadysignificantlyraisesthetemperature,whichinturnincreasesthereactionrate. Equation (4) assumes that temperature remains constant during the reaction, neglecting any self-accelerating effects – 6 – of nuclear burning. Thus to obtain a more realistic estimate on the nuclear burning time we will refer to the so-called induction time of nuclear burning. It is known (e.g., Khokhlov 1991) that the process of explosive nuclear burning can be subdivided into two temporal stages. During the first phase after ignition, at times shorter than the induction time τ, the temperature increases slowly (linearly) in time, and the i corresponding energy production is relatively small. After this, duringthe explosivephase,energyproductionevolvesquasi-exponentiallyintime. Thusonly the induction time is actually needed to burn a fluid element completely, because τ is nearly equal to the time i it takes to burn most of the fuel. Self-acceleration of nuclear burning leads to the relation τ ≪ τ . The i m ratio of the induction and the constant temperature timescale can approximately be given by an analytic expression known as the Frank-Kamenetskii factor. However, for our purposes we calculate τ numerically i using the equation 2τi Y˙12Cdt=γY12C(t=0), (5) Z 0 where γ denotes the total fraction of burned material. Based on equation (5), Figure 1 represents the time required to burn a fraction (1−γ) of the initial fuel amount and it clearly shows that for any value γ ∈[0,2,0.9] this time turns out to be alomst the same. We therefore use equation (5) – with an arbitrary choice of γ ∈[0,2,0.9] – to define the induction time τi,12C as half of the time it takes to burn most of the fuel. This definition ensures that for times smaller than τ the burning process evolves almost linearly. i For a phenomenological description of turbulence we assume that the turbulent velocities obey Kolmogorov’s scaling law. In particular, this implies the following scaling of the r.m.s. turbulent velocity fluctuations u′(l): 1/3 l ′ ′ u(l)=u(L) , (6) (cid:18)L(cid:19) with l =LRe−3/4 <l <L, where L is the integral length-scale and Re is the turbulent Reynolds number. k In order to model the effect of turbulent motions on a flame front we make two additional assumptions: 1. Turbulent vortices are viewed as a transportmechanism of heat and mass. Thus a fluid element can ′ be carried by a vortex over a distance l after a time τ =1/2(l/u(l)), which is half of the eddy’s turn-over l time. In case of a one-dimensional and co-moving representation of a flame, this mechanism transports heat and nuclear species located at x>0 in the positive direction, while fluid elements placed at x<0 are transported in the negative one. 2. Since the spatial distribution of the induction time τ(x) is a single-valued function in the regions i x>0 and x<0, every fluid element has its unique value of τ(x). Then the distance l′ over which a fluid i τ – 7 – element located at x can be moved is determined by the equation τ =τ leading to l i (2u′(L)τ)3/2 ′ i l (x)= . (7) τ L1/2 In this approachl′(x) is not allowedto exceed the integral length, L, neither should it get smaller than the τ Kolmogorov length, l 1. Furthermore, because the flame itself moves on during the time τ (at least with k i ′ its laminar velocity s), we have to subtract the distance sτ from the original value of l (x). This results l l i τ in the expression ′ ′ ′ ′ l (x)=H(χ (l )l −sτ) χ (l )l −s τ , (8) τ [lk,L] τ τ l i [lk,L] τ τ l i (cid:2) (cid:3) where χ is the characteristic function of the interval [l ,L] (This is, χ (x) = 1forx ∈ [l ,L], and [lk,L] k [lk,L] k 0 otherwise.) for giving a cut-off for scales smaller than l and larger than L. H is Heaviside’s function k representing the obvious fact that negative values of l (x) are not admissable. τ Using equation (8), the turbulent flame profile is given by {ρ ,T ,Y }(x+l (x)) x>0 t t t τ ={ρ,T,Y}(x) (9) l l l {ρ ,T ,Y }(x−l (x)) x<0  t t t τ  Our first assumption was that there is additional heat/mass-transportdue to turbulent eddies leading to a broadening of the flame. But it ignores the fact that turbulence is actually isotropic, which means that eddies occur in any direction without a preferred one. However, as long as the flame’s reaction zone is relatively well localized there will always be a flame surface defined by the points of maximum energy production. Thus there is a preferred direction locally defined by the normal vector of the flame surface. Turbulence randomly drags and shifts this surface but in a locally co-moving frame turbulent motion increases the thermal and nuclear transport in directions normal to it. Heuristically, this can be seen by introducing a local eddy-diffusivity D (x) = u′(x)x, where x denotes the distance from the flame surface. t Thus turbulence imposes an enhanced heat flux ahead of the flame’s surface. In the reaction region the induction times are so short that turbulent mixing hardly changes its structure. Well behind the thin reaction zone almost all of the carbon is already burned. Thus turbulent motions in regions behind this 1 The reason why we do not consider length-scales larger than L is that we actually do not know how the turbulent spectrum of velocity fluctuations looks like at long wavelengths. Rayleigh-Taylor unstable structures (bubbles) in the non-linear stage of this instability may cause additional effects on the turbulent behavior (see, e.g., Niemeyer & Woosley 1997) leading to significant deviations from isotropy and homogeneity. – 8 – zone actuallystirrspure ashes. In the contextofthe presentworkitis thereforesufficient to considereffects of turbulence only for x>0. One may argue that in reality there is not just one flame profile along the normal direction of a flame surface, but there is a random superposition of flames, fuels and ashes, giving rise to strong fluctuations in temperature and nuclear abundances. However, as was stated before, we do not intend to give a full descriptionofthe structureofa turbulentflame. Itis ouraimto modela situationwhichis mostfavourable for carbon detonation. In order to trigger a detonation based on the Zeldovich mechanism (Zeldovich et al. 1970) one needs a region of a certain critical size with a rather uniform temperature and fuel fraction. To be more specific, a non-uniform spatial region of induction times must be present such that within this region spontaneous burning reaches a critical velocity threshold. Let σ(τ ) be the resulting variance of the i induction times. Then this region must be at least of size σ(τ )a , where a is the local sound velocity i s s (see, Khokhlov 1991). In the presence of temperature and fuel fluctuations a successful formation of such a regionis stronglysuppressed,because the rate and the amplitude offluctuations growwith scale. Therefore one expects that such homogeneous regions are already torn apart into smaller sections before they can ever reach the critical size (Niemeyer 1999). Our model aims at describing the unlikely situation that such strong fluctuations are not present and therefore we could find detonations which in reality would not happen. On the other hand, if certain physical conditions described by a set of parameters ρ, s,Re,L,l , l k etc., make a successful detonation impossible in our model, it indicates that under the same prerequisites a detonation in a real turbulent flame would not happen either. In order to determine a relation between turbulence intensity and the resulting shape of the flame we make one more assumption. The main motivation for it is the desire to relate an undisturbed (laminar) flame profile with a time setting a maximum limit for the turbulent transport. In fact, the maximum time for which a fluid element with a significant amount of fuel can be transportedby turbulent motion is τ, i.e. i itsinductiontime. Duringthisperioditcoversadistancel givenbythesizeofacertaineddy,c.f. equation τ (7) 2 . Considering this for every fluid element within a laminar flame, one can derive a formula for the shape of the turbulent flame, which is exactly equation (9). This equation already involves a simplification because we use the induction time in order to estimate l . As mentioned above, the thermodynamic state τ 2It should be remarked that l , given by equation (7), is the maximum distance for turbulent transport τ at a given τ. That means that bigger eddies are not able to carry a fluid element further out than l . This τ can be easily verified when Kolmogorovscaling is assumed. – 9 – of a fluid element changes only slightly after a period of time given by τ, i.e. T(t+τ) ≈ T(t) and i l i l Y12C,l(t+τi) ≈ Y12C,l(t). Thus the state of the fluid element after being transported over a distance lτ is nearly the same as its original state. However, taking into account significantly longer times than τ i would lead to strong spatial fluctuations in temperature and composition. For instance, assume that a fluid element is transported over a distance l where τ ≈ 2τ. Furthermore consider a small fluctuation in time τ i δτ ≪τ. Then we have lτ±δτ ≈lτ but T(τ ±δτ) might be much bigger or smaller then T(τ) depending on the sign of the fluctuation. Of course, spatial fluctuations in carbon composition would be even stronger in amplitude. Again, based on the same arguments as given above, these fluctuations would make the formation of a detonation much more unlikely. Therefore for the purpose of this letter it is sufficient to represent flame-turbulence interaction in the linear regime characterizedby the induction time scale. Two of the resulting turbulent profiles are shown in Figure 2. One sees that the innermost region of the flame is hardly disturbed by turbulent motions even at turbulent velocities of u′(L) = 107cm s−1. In contrast, regions of slow reactions are widely extended causing a huge preheated zone ahead of the flame. 3. Testing for Detonations In this section we use the results stated above to implement them as initial conditions for the solution of the fully compressible hydrodynamical equations. The question then is under which conditions the flame evolves into a detonation. Similar investigations have already been done (Khokhlov 1991, Niemeyer & Woosley 1997). In these studies the initial conditions were parametrized by the width of a non-uniformly burning region. It was shown that there exists – for given density and fuel composition – a minimum (critical) length scale such that non-uniform burning on scales larger than this leads to a detonation. However,inthese investigationsthe importantquestionofhowsucharegionmightformwasnotaddressed. Only weak necessary conditions limiting the strength of the turbulence-flame-interaction have been given. For instance, the authors demand a breaking up of the flame (Khokhlov et al. 1997) or burning in the distributed regime (Niemeyer & Woosley 1997). Our model establishes a direct relation between turbulence intensity and the shape of the flame profile. Using this relation we are able to give necessary conditions for a transition to detonation. The system of conservation laws that we solve numerically is ∂ρ +∇·(ρU)=0, (10) ∂t – 10 – ∂ρU +∇·(ρUU)+∇P =0, (11) ∂t ∂E +∇·[(E+P)U]=0, (12) ∂t ∂Y i +(U ·∇)Y + Y Y λ (i)−Y Y λ (i)=0, (13) i i k jk i k kj ∂t Xj,k where U is the fluid velocity and E = ǫ+ρU2/2+ρN B Y is the total energy density defined as the A i i P sum of the internal, the kinetic and the binding energy term. These equations are coupled to the same equation of state and nuclear reaction network as already used in the reactive diffusion equations (1) and (2). They are solved in a one-dimensional planar geometry with outflow boundary conditions as well as in spherical geometry. In the latter case we use a reflecting inner boundary and an outflow condition on the outside. As a numerical scheme we used PROMETHEUS (Fryxell et al. 1989), a second order explicit in time solver for reactive hydrodynamics. We consider three different densities, namely 2.3,1.3 and 0.8×107g cm−3, and three different compositions of the fuel: X(12C)/X(16O) = 1,1/3 and 3. The turbulent velocity fluctuations in an exploding Chandrasekhar-masswhite dwarf are (see, Niemeyer & Woosley 1997, Khokhlov et al. 1997) believed to be u′(L) ≈ 106−107cm s−1 on an integral lengthscale of L ≈ 106cm. In all our numerical computations we keep the integral scale fixed at this particular value and the only turbulent variable remains the amplitude turbulent velocity fluctuations u′(L). Given initial conditions as described above we need to find a natural time limit for evolving the hydrodynamics. Since the flame dynamics is embedded into the global dynamics of the star, we adopt this limit to be the hydrodynamical timescale, τ , of the exploding white dwarf. After this time has elapsed the h stellar density has dropped significantly and our originally assumed value of the density in the unburned material is no longer valid. Thus if no detonation is observed after an elapsed time τ it is fair to conclude h that for the conditions investigated no detonation can occur. We use a value of τ = 0.02 s which is h consistentwith estimates from directnumericalcalculations of exploding Chandrasekhar-masswhite dwarfs (Reinecke et al. 1999). In Figure 3 we see how a spontaneous wave is initiated (at a density of 2.3 × 107g cm−3, a fuel-composition of X(12C) = X(16O) = 1/ and a turbulent velocity of u′(L) = 0.2×108cm s−1) and 2 propagatesalong the temperature gradient. However,after anelapsed time of 0.02s=τ it does not pile up h to a detonation. Instead it moves with a constant velocity at constant pressure.