A global simulation for laser driven MeV electrons in 50µm-diameter fast ignition targets C. Ren1, M. Tzoufras2, J. Tonge3, W. B. Mori2,3, F. S. Tsung3, M. Fiore4, R. A. Fonseca5,4, L. O. Silva4, J.-C. Adam6, and A. Heron6 1. Department of Mechanical Engineering, Department of Physics and Astronomy, 6 and Laboratory for Laser Energetics, 0 University of Rochester, Rochester, NY 14627. 0 2. Department of Electrical Engineering, 2 University of California, Los Angeles, CA 90095. n 3. Department of Physics & Astronomy, a University of California, Los Angeles, CA 90095. J 4. GoLP/Centro de Fisica dos Plasmas, 5 Instituto Superior Tecnico, 1049-001 Lisboa, Portugal. 2 5. DCTI, Instituto Superior de Cincias do Trabalho e da Empresa, 1649-026, Lisboa, Portugal. ] 6. Ecole Polytechnique, France. h (Dated: February 2, 2008) p - The results from 2.5-dimensional Particle-in-Cell simulations for the interaction of a picosecond- m longignitionlaserpulsewithaplasmapelletof50-µmdiameterand40criticaldensityarepresented. s The high density pellet is surrounded by an underdensecorona and is isolated by a vacuum region a fromthesimulationboxboundary. Thelaserpulseisshowntofilamentandcreatedensitychannels l p onthelaser-plasmainterface. Thedensitychannelsincreasethelaserabsorptionefficiencyandhelp . generate an energetic electron distribution with a large angular spread. The combined distribution s of the forward-going energetic electrons and the induced return electrons is marginally unstable to c i the current filament instability. The ions play an important role in neutralizing the space charges s induced by the the temperature disparity between different electron groups. No global coalescing y ofthecurrentfilamentsresultedfrom theinstability isobserved,consistent withtheobservedlarge h angular spread of the energetic electrons. p [ 1 INTRODUCTION v 0 0 The fast ignition concept, first proposedmore than a decade ago[1], provides a possible alternative path to achieve 2 ignition in inertial confinementfusion with the potential of reducing drivers energy andincreasing gains significantly. 1 In this concept, the ignition is first achieved in a small region (L 20µm) within a compressed pellet by a 10 ps 0 ∼ ∼ intense flux of MeV electrons. The MeV electrons are produced by the absorption of a petawatt (PW)-class laser 6 at the edge of the pellet. Recent experimental results in Japan[2] showed 3 orders of magnitude increase of neutron 0 / yield with a 0.5-PW ignition laser and consequently spurred intense new research activities in this area. One of the s research foci is to predict the ignition laser energy needed to ignite a certain target. This energy is determined by a c i number of highly nonlinear and dynamic processes such as the laser propagationin the underdense corona, the laser s y hole-boring and electron heating near the critical surface, the transport of the energetic electrons in the overdense h plasma, and the subsequent heating of the ignition region by these electrons. No theory exists and simulations and p experiments are the only tools to study these processes. : v Particle-in-Cell (PIC) simulations provide a first-principle, detailed description for the plasma region of the pellet, i whichcomprisesthe majorityofthe pellet exceptfor the verydense core(electrondensity n >1023/cm3). Since the X e PIC simulations are computationally intensive, currently it is not possible to perform full scale simulations for the r a entirepellet. Figure1showsthene andelectrontemperature(Te)informationofacryogenictarget,implodedbya25- kJ laser,atits peak compression[3]. The T informationis representedbythe plasmaparameter(number ofelectrons e in a Debye cube, nλ3 ) and the ratio of the electron-ion collisionality to the electron plasma frequency, ν /ω . In D ei pe the regionwhere n <1023/cm3, n λ3 >>1 whichsatisfiesthe usualdefinitionofa plasma. Notice heren λ3 is not e e D e D overlylargeeither,whichindicatesthatthediscreteparticleeffectsareimportant. Inthisregionν /ω <<1,which ei pe showsthatplasmabehaviorisnotdominatedbythecollisions. AllthesemakethePICmodeltheidealsimulationtool for this region. However,evenif we wereto simulatea plasmaslabof250µm-longand50µm 50µmincrosssection × and resolve the smallest relevant physical length c/ω (∆x = 1/3c/ω ) for n = 5 1022/cm3, the number of cells pe pe e × needed would be 1.3 1012. Assuming 4 particles/cell for each species (electrons and ions), the number of particles × neededwouldbe1013. Thetotalmemoryrequiredtostoretheparticlemomentumandpositioninformationwouldbe 2 560 Tera-Bytes (TB). Assuming a time step limited by Courant condition of ∆t=1/(3√3)ω−1 and total simulation pe time of 10 ps, the total number of steps needed would be 6.7 105. The total number of floating-point operations × (FLOPs)wouldbe2.7 1020(assuming40FLOPs/particle-step). Thiswouldtake10monthsonacomputercapableof × 10TeraFLOPs/second(TFLOPS).Thiskindofsimulationscanberoutinelyperformedonlywhen100-1000TFLOPS computerswithlargememoryarewidelyavailable. MostrecentfastignitionPICsimulationsusedscaled-downtargets withreducedsizesand/orintwo-dimension(2D)insteadofthree-dimension(3D)[4,5,6,7,8]. Manyoftheimportant issues, such as hole-boring, laser absorption, and current filamentation, have been studied. However,some keyphysics depends critically onglobalelectronpaths inthe targetandthus depends onthe target size and the boundary conditions used in the simulations. For example, to absorb30%of the energyin a 1020W/cm2 laser for 1 ps, the thickness of a plasma slab with n = 1022/cm3 needs to be 190µm, assuming every electron in e the slab leaves with an energy increase of 1 MeV. However, the laser cannot sweep this distance in 1 ps through hole-boring. Even if it could, the ion layer formed from electron depletion would pull the electrons back. The actual absorption is achieved through continuous replenishing of cold electrons from the bulk of the target into a narrow region in contact with the laser front. If the simulated target is too small and cannot provide enough needed cold electrons,theMeVelectronsmaynotleavetheinteractionregionandmaybeheatedfurthermoretoachieveadifferent distribution. In this paper we present results from a series of 2.5D (with two space dimensions but all three components of particle velocities and electromagnetic fields) PIC simulations with the PIC code OSIRIS[9] in a simulation box of 100λ 100λ with a round target of 51λ diameter. (λ is the laser wavelength and also the basic length unit the × simulationsarescaledto.) Thesetup(Fig.2)allowsthesimulationoftheplasma-vacuumboundarywhichprovidesan important alternative pathway for the MeV electrons. The target consists of a proton-electron plasma with T =7.4 e keV and T = 1 keV. The target has a uniform density core with n = 40n within a diameter of 32λ and a coronal i e c ring with the density linearly decreasing from 40n to 0 in 9.5λ. Here n is the critical density which corresponds c c to 1.1 1021cm−3 for λ = 1µm. The laser is launched from the left boundary with peak intensities of I = 1020 and × 1021(1µm/λ)2W/cm2 and a rise time of 19 laser period, after which the laser amplitude is kept constant. The laser transverse profile is a Gaussian with a diffraction limited spot size of w = 7.5λ. Both s- and p-polarization are used to infer 3D effects. A typical simulation duration is 309 laser period, which corresponds to about 1 ps for λ=1µm. To resolve the skin depth for n = 40n , c/ω λ/40, we use a grid size of ∆xω /c = ∆yω /c = 0.33 with c pe pe pe ≈ 12032 12032 grid cells. Therefore ∆x 3λ . We reduce numerical self-heating[10] by smoothing the current. The D × ≈ timestepusedis∆tω =0.23. Weuse4particleseachfortheelectronsandtheprotons(2.4 108totalparticles)and pe × their charges are weighted accordingto the initial local density. The effective finite size particle collisionfrequency is roughlyν /ω 0.035[11]. Theparticlesarereemittedwiththeirinitialtemperatureswhentheyhitthex-boundary ei pe ≈ and are subject to periodic boundary condition in the y-direction. The boundary conditions for the electromagnetic fields are open in the laser propagation direction (x) and periodic in the transverse direction (y). A typical run consists of 1.6 1013 particle-steps. × The rest of the paper is devoted mainly to two issues: the laser absorption and electron heating near the laser- plasmainterface(Sec.)andthefilamentationoftheelectronfluxinsidethetarget(Sec.). Adiscussionandasummary are provided in Sec.. LASER ABSORPTION AND ELECTRON HEATING Asthelaserpropagatesthroughtheshortunderdenseregionhereandreachesthecriticalsurface,themostprominent feature observed from the simulations is the filamentation of the laser accompanied by the rippling of the target surface (Fig.3). The observed filament width is roughly an optical wavelength. A laser can filament due to either relativisticmassorponderomotiveforceeffects. Fortheparametersofthefastignition,bothmechanismswilloccuron timescalesthatareveryshortcomparedtothelaserpulselength. Therefore,itisnotasimportanttostudytheinitial growthphaseasitistounderstandthe eventualnonlinearsteadystate. Thenon-relativistictheoryofponderomotive filamentation predicts that the mode number with the fastest growth rate increases with plasma density and the growth rate is higher in the direction perpendicular to the polarization than in the polarization direction[12]. These conclusions are verified in Fig.3 which shows that the s-polarized cases (laser electric field E out of the simulation plane) display stronger laser filamentation than the p-polarized cases (E in the simulation plane). With their higher intensity the laser filaments dig density channels at the laser-plasma interface via the ponderomotive force (Fig.3, right column). These micro density channels in turn focus the laser filaments like converging waveguides which dig deeper channels. Therefore, the laser-plasma interface is unstable to any initial transverse modulation of the laser intensity, which can be viewed as filamentation in the overdense plasma[13]. 3 The density channels on the laser-plasma interface can enhance the laser absorption significantly. Heating mech- anisms for normal-incident lasers include the normal coupling of the laser oscillation and the plasma wave at the critical surface[14] and the J B heating[15]. The density channels provide locations where the laser E-field can × be in the density gradient direction. This provides additional heating mechanisms such as the enhanced resonant heating[16] and Brunel heating[17]. In our 2.5D simulations, the additional heating mechanisms are present only for the p-polarization cases. Therefore, we expect that the laser absoprtion is larger in the p-cases than in the s-cases. Figure 4a showsthe laserabsorptionrate for the electrons,defined as the ratioof the electronkinetic energyincrease to the incoming laserenergy within a certaintime interval∆t, η(t)=[KE(t+∆t) KE(t)]/(P ∆t), for both s- and L − p-polarizations and verifies this conclusion. In addition, the p-cases also show that η increases with time, consistent with the fact that as the channels becomes deeper these additional heating mechanisms become more efficient. On thecontraryη inthes-caseremainsalmostflat. Theabsorptiondoesnotchangesignificantlywhenthelaserintensity changes from 1020W/cm2 to 1021W/cm2. However, the electron energy composition changes significantly (Fig.4b). In the I = 1021W/cm2 case, more than half of the absorbed laser energy goes to the electrons with energy greater than 10 MeV. These super-hot electrons may be too energetic to be stopped in the target core and be useful to the fast ignition. The electron spectrum in the I =1020W/cm2 cases can be approximatedby a power law, which begins at E 0.2 MeV and falls off as E−(2−3)[8]. ∼ Theadditionalheatingmechanisms[16,17]allowtheelectronstobeaccelerateddirectlyinthelaserE-fielddirection. Thus the energetic electrons are expected to to have a large transverse momentum (P ). This can be seen from the y P P phase space plot for the electrons localized in front of the laser (Fig.5a). It is also clear from the phase space x y plot that the energetic electrons do not form a beam with narrow emittance. The root-mean-square angle in the xy-plane for the above-1 MeV electrons is plotted in Fig.5b, which also shows the large angular spread for both 1020W/cm2 and 1021W/cm2 cases. The large angular spread is one of the most important characteristics of the energetic electron distribution in these simulations and must be taken into account in studying their subsequent transport and in accessing the fast ignition feasibility. CURRENT FILAMENT INSTABILITY Ifleftalone,thelaser-generatedenergeticelectronswoulddeposittheirenergyoveralargeareawhentheyreachthe dense core region due to their large angular spread. However,the transport of the energetic electrons is not a simple free-streaming process but a highly nonlinear one during which their interactions with emerging magnetic fields play a very important role. For example, we have observed that the laser-heated electrons can generate intense magnetic fieldsthroughthe T neffect[18]onthetargetsurface(notshownhere). Thesemagneticfields,togetherwiththe ∇ ×∇ surfaceradialelectric fields,causea fractionofthe energeticelectronsto movealongthe surface throughE B-drift, × whichactuallypreventstheelectronsfromhittingthesimulationboxboundary. Themotionoftheenergeticelectrons that are shot forward into the interior of the target is also influenced by magnetic fields, only in this case it is the current filament instability that provides these fields. To properly study the current filament instability in the fast ignition one needs to add space charge effects due to temperature disparity between different groups of electrons to the standard analysis of Weibel instability[19, 20]. The 100 MA current carried by the MeV electrons in the fast ignition greatly exceeds the Alfven current limit[21], ∼ which is set by the pinching force on the beam electrons from the self-magnetic field and is 30 kA for the MeV ∼ electrons. A return current in the background plasma must be induced to neutralize the magnetic field of the beam and render the whole system nearly current neutral. Unlike the previously studied system of two identical counter- propagating electron clouds[20], the forward energetic electrons have a much higher transverse temperature than the background plasma electrons carrying the return current. Therefore, the forward current and return current pinch to different degrees when filamenting and a space charge imbalance will develop. The electric field from the space charge is along the wave vector (or the y direction here) and resists further filamenting. The space charge effects − are especially important when the system is in the marginally unstable regime which is the typical case for the fast ignitionsimulationshere. Furthermore,theionswillrespondtotheelectricfieldandtheirmotionbecomesimportant in the marginally unstable case where the growth rate is comparable to the ion plasma frequency[8, 22]. With a totally different origin from that in Ref.[23], the space charge effects discussed here exist even when the system is stable to the two-streaminstability. The detailed analysis of the current filament instability with the space charge effects is presented in a sep- arated paper[24]. Here we illustrate the basic space charge effects on the filament instability by considering a plasma system with 3 species: one species of hot energetic electrons moving in the +xˆ-direction with a drift- ing velocity V , another species of cool return electrons moving in the xˆ-direction with a drifting velocity V , d1 d2 − 4 and a cold ion species. Specifically, the electron equilibrium distribution functions are assumed to be f (v,x) = l0 nl/(2π√vtylvtxl)exp(−vy2/2vt2yl)exp[−(vx −Vdl)2/2vt2xl], where nl is the density and vtxl and vtyl are the thermal velocitiesinthe xˆ-andyˆ-directionsforeachelectronspecies,(l =1,2). Underthe relevantgeometryandpolarization v is a negligible variable. The ion equilibrium distribution function is f =n δ(v )δ(v ). Both charge and current z 30 3 x y densitiesarezerointheequilibrium, 3 q n =0andn V +n V =0. Hereq =q = earetheelectroncharge Pl=1 l l 1 d1 2 d2 1 2 − and q the ion charge. Consequently, there is no equilibrium electric or magnetic field. 3 Tostudythelinearstabilityofthissystemagainstthefilamentmode,weassumethatthe modepropagationvector is in the y-direction, k=kyˆ, and all perturbations have spatial and temporal dependence of exp(γt iky). For each − species, the perturbed distribution function f can be found from the linearized Vlasov equation, l1 (q /m ) v ∂f f =i l l (E+ B) l0, (1) l1 ω kv c× · ∂v y − where m is the mass of the particle in the l-th species. The relevant perturbed fields are E = E xˆ +E yˆ and l x y B = B zˆ. These can be related to the charge density ρ and current density j using the Maxwell’s equations. The z x linear analysis yields the following two coupled equations, ρ{1+X2 k2ωvp22l [1+ξlZ(ξl)]− ωωp223}+ jcx X2 (ω2ωp2klV2dcl2c)v2 [1+ξlZ(ξl)]=0, (2) l=1 tyl l=1 − tyl 2 ωp2lVdl jx 2 ωp2l(Vd2l+vt2xl) 3 ωp2l ρXk2v2 c[1+ξlZ(ξl)]+ c {1+X(ω2 k2c2)v2 [1+ξlZ(ξl)]−Xω2 k2c2}=0, (3) l=1 tyl l=1 − tyl l=1 − where ωpl ≡ p4πql2nl/ml is the plasma frequency of each species, ξl ≡ω/(√2kvtyl), and Z is the plasma dispersion function[25]. The analysis of the dispersion relation from Eqs.2-3 reveals two regimes of instability[24]. In the relative fast growth regime, (m/M)kv <<γ <<kv , the ions can be treated as immobile (ion mass M and ion plasma ty2 ty2 →∞ frequency ω =0) and the instability threshold is pI 2 V2 +v2 2 (P2l=1 ωvp2l2Vdl)2 Xl=1ωp2l dlvt2yltxl >k2c2+Xl=1ωp2l+ k2+P2l=t1ylvω2p2l . (4) tyl The third termonthe righthandside of Eq.4comes fromthe off-diagonaltermin Eqs.2-3 that couplesthe perturba- tions in ρ and j and therefore represents the space charge effects. This term is positive definite, indicating that the x spacechargesalwaysraisethe instabilitythreshold. Ifthe twoelectronspecieshavethe sametransversetemperature, v = v , the space charge term vanishes since the current neutrality condition implies ω2V = 0. Therefore, ty1 ty2 P pl dl the space charges originate from the temperature disparity between the two electron groups. In general, the space charge term decreases the growth rate and reduces the range of k of the instability[24]. Inthe slowgrowthregime,γ <<(m/M)kv ,the ionresponsemustbe considered. Thenthe instability threshold ty2 turns out to be 2 V2 +v2 2 Xωp2l dlv2 txl >k2c2+Xωp2l+ωp2I. (5) l=1 tyl l=1 Compared to Eq.4, we can see in this marginal instability regime, the growth rate is so small that the ions always have time to reactto cancelany potential space charge. Therefore,the instability thresholdis the same as that when no space charge effect is considered[20]. The actual electron distribution in our PIC simulations (see Fig.5a) cannot be accurately approximated by two counter-propagating maxwellians. However, the above analysis can be extended for a general electron distribution in the form of f (P ,P ) = N δ(P P )exp( P2/2P2 ), basically breaking the distribution into N drifting e0 x y Pl=1 x − dl − y tyl beamlets. ThethresholdsinEqs.4-5canbemodifiedwiththesumnowextendedtoallN beamlets. Wehaveexamined theelectrondistributionsintwolocationsinoursimulations,oneintheshockregionandtheotherinthetargetinterior (shown by the small boxes in Fig.6a), for their stability property. Specifically, we plot ω2(P2/P2 1), which is pl dl tyl − basically Eq.5 when v and k are set to zero and ω is neglected, for each beamlets in Fig.6c and 6d. A positive txl pI dot contributes to the instability and a negative one to the stability. The sum of all the dots needs to be positive for 5 the instability to occur. We found thatthe shockregionis unstable (Fig.6c) butthe interiorregionis stable (Fig.6d). Furthermore, if the space charge effect is included as in Eq.4 even the shock region becomes stable. Therefore, the electrondistributionsinthesefastignitionsimulationsaremarginallyunstableduetotheirlargeangularspread. The ions play an important role to neutralize any possible space charges. This is also supported by the observation that the iondensity alwaysdisplaythe samefilamentstructureasthe electrons. Itisalsoworthwhileto pointoutthatthe return current beamlets (those with negative P ) contribute more than the forward beamlets toward the instability dl becauseoftheirrelativelysmallerangularspread. Anycurrentfilamentstabilityanalysiswithoutconsideringtherole of the return current would be inaccurate. Due to dilution of the energetic electron relative density in the interior region, the electrons there are actually stable to the filament instability. There the magnetic pinching force cannot overcome the thermal pressure due to the large transverse momentum spread. The filament structure observed there is due to the electron streaming from the filaments in the shock region. It is therefore no surprise that no global coalescing of these filaments into a single strong filament is observed here (Fig.6b), contrary to the previous simulations with smaller transverse box size[5, 6], since the coalescing requires that the magnetic pinching force is larger than the thermal pressure. Even without the global coalescing the magnetic field from the filament instability is still strong enough (reaching 300 MG in the ∼ shock region for 1µm-laser)so that the gyro radius of MeV electrons is less than the laser wavelength. The electrons are confined by the self-generated magnetic field near the laser spot. The difference regarding the filament coalescing embodies the difference in the electron distributions between differentfastignitionsimulations. Inthesimulations,theelectrondistributioncanbeaffectedbyanumberoffactors. One factor is the total heat capacity of the target, which is determined by the target size and density used in the simulation. Given the amount of laser energy absorbed, the fraction of energetic electrons in the distribution will be determined by the heat capacity. The danger of using too small a target lies in that eventually all electrons in the simulation acquire MeV energy, a situation totally different from the actual fast ignition scheme. Another factor is the recycling of the energetic electrons, which is determined by the target size and the boundary conditions. If the transverse box size is small and no plasma-vacuum boundary is present, the electrons can recirculate through the laser spot many times transversely under the periodic boundary condition before they eventually leave in the longitudinal direction. The plasma-vacuum interface used here combined with the surface magnetic field prevents suchtransverserecirculation. However,thelongitudinalrecirculationisstillpresentbutoccurslaterthaninasmaller sizedtarget. Stillanotherfactoristhetrue2Dnatureofthesimulationaffordedbythelargetransversesizeusedhere. This allows the realistic simulation of the laser filaments and density channels which are important for the heating mechanisms[16, 17] which produce the large angular spread observed here. SUMMARY AND DISCUSSION Inthispapertheresultsfrom2.5DPICsimulationsfortheinteractionofapicosecond-longignitionlaserpulsewith a plasma pellet of 50-µm diameter and 40-n are presented. The high density pellet is surrounded by an underdense c coronaandisisolatedbyavacuumregionfromthesimulationboxboundary. Thelaserpulseisshowntofilamentand createdensitychannelsonthelaser-plasmainterface. Thedensitychannelsincreasethelaserabsorptionefficiencyand helpgenerateanenergeticelectrondistributionwithalargeangularspread. Thecombineddistributionoftheforward- going energetic electrons and the induced return electrons is marginally unstable to the current filament instability. The ions play an important role in neutralizing the space charges induced by the the temperature disparity between different electron groups. No global coalescing of the current filaments resulted from the instability is observed, consistent with the observed large angular spread of the energetic electrons. The initialtargetT usedhereis 7.4keV,whichis higherthanthe actualtargetbulkT 1 keV.This T ischosen e e e ≈ mainly to avoid numerical self-heating[10]. Recently we have implemented a second-order-spline current deposition scheme in OSIRIS which when combined with the current smoothing allows a grid size of ∆x=12λ with virtually D no self-heating. We repeat the simulations with T = 1.1 keV and see no change in the main conclusions presented e here. Whilethegenerationoftheenergeticelectronsisstudiedhereinconsiderabledetails,theirtransportfromthebirth place to n = 1023/cm3 needs further study. The highly nonlinear nature of the interaction between the energetic electrons and emerging magnetic fields require simulations with targets of 100µm-length and up to 100n . The c ∼ longitudinalrecyclingoftheenergeticelectronsshouldalsobeeliminatedthroughsuitableboundaryconditions. Only then can a reliable estimate of the energy flux reaching the n=1023/cm3 surface be obtained. Fortunately, with the advance of computer speed and PIC code improvement this type of simulations is within reach now. For the very dense region of n > 1023/cm3, the collisional effects become important and the electron transport 6 and its final stopping will be affected by collisional scattering[26]. PIC or particle-fluid hybrid type of codes need to include a proper collision model to correctly simulate the physics there. Eventually the simulations need to be compared with experiments. This work was supported by the US Department of Energy through grants DE-FG02-03ER54721, DE- FG02- 03NA00065, and DE-FC02-04ER54789,by ILSA at LLNL under W-07405-ENG48and by NSF un- der PHY032345. Simulations were done on the DAWSON Cluster and at NERSC. The work of Prof. L. O. Silva, Prof. R. A. Fonseca and M. Fiore was partially supported by FCT (Portugal) through grants PDCT/FP/FAT/50190/2003,and POCI/FIS/55905/2004. [1] M. Tabak, J. Hammer, M. E. Glinsky, W. L. Kruer, S. C. Wilks, J. Woodworth, E. M. Campbell, M. D .Perry and R.J. Mason, Phys.Plasmas 1, 1626 (1994). [2] R.Kodama et al., Nature 412, 801 (2001), ibid. 418, 933 (2002). [3] R.Betti, privatecommunication. [4] S.C. Wilks, W. L. Kruer,M. Tabak, and A.B. Langdon, Phys. Rev. Lett.69, 1383 (1992). [5] A.Pukhov and J. Meyer ter Vehn,Phys. Rev.Lett 79, 2686 (1997). [6] B. F. Lasinski, A.B. Langdon, S.P. Hatchett,M. H.Key,and M. Tabak, Phys.Plasmas 6, 2041 (1999). [7] Y.Sentoku,K. Mima, S.Kojima, H.Ruhl,Phys. Plasmas 7, 689 (2000). [8] C. Ren,M. Tzoufras, F. S.Tsung, W.B. Mori, S. Amorini, R. A. Fonseca, L.O. Silva, J. C. Adam, and A.Heron, Phys. Rev.Lett. 93, 185004 (2004). [9] R.G. Hemker, UCLA PhD thesis (2000); R. A.Fonseca et al., LCNS 2331, 342 (Springer-Verlag, Heidelberg, 2002). [10] C. K. Birdsall and A. B. Langdon, Plasma Physics via Computer Simulation, Institue of Physics Publishing, Bristol and Philadelphia, 1991. [11] R.W. Hockney,J. Comput. Phys. 8, 19 (1971). [12] P.Kaw, G. Schmidt,and T. Wilcox, Phys. Fluids 16, 1522 (1973). [13] E. J. Valeo and K.G. Estabrook Phys. Rev.Lett. 34, 1008 (1975). [14] P.K. Kaw and J. M. Dawson, Phys.Fluids 12, 2586 (1969). [15] W. L.Kruer and K. Estabrook, Phys.Fluids 28, 430 (1985). [16] K.Estabrook and W. L. Kruer, Phys.Rev.Lett. 40, 42 (1978). [17] F. Brunel, Phys.Rev. Lett. 59, 52 (1987). [18] D.W. Forslund and J. U. Brackbill, Phys.Rev.Lett. 48, 001614 (1982). [19] E. S.Weibel, Phys.Rev.Lett. 2, 83 (1959). [20] R.C. Davidson,D. A.Hammer, I. Haber, and C. E. Wagner, Phys. Fluids 15, 317 (1972). [21] H.Alfven, Phys.Rev. 55, 425 (1939). [22] M. Honda, J. Meyer-ter-Vehn,and A. Pukhov,Phys.Rev.Lett. 85, 2128 (2000); Phys. Plasmas 7, 1302 (2000). [23] A.Bret, M.-C. Firpo, and C. Deutsch,Phys.Rev. E 70, 046401 (2004); Phys.Rev. Lett 94, 115002 (2005). [24] M.Tzoufras, C.Ren,F.S.Tsung,J.W.Tonge,W.B.Mori, M.Fiore,R.A.Fonseca,andL.O.Silva,submittedtoPhys. Rev.Lett. [25] B. F. Fried and S.Conte, The Plasma Dispersion Function, Academic, New York,1961. [26] C. K. Li and R.D. Petrasso, Phys. Rev.E 70, 067401 (2004). 7 FIG. 1: (color online) Electron density and temperature (in terms of nλ3D and νei/ωpe) profiles of a typical target at its peak compression. FIG. 2: (color online) 2D PIC simulation setup. 8 FIG. 3: (color online) Laser intensityE2 (left column, in unit of (mcω/e)2) and electron charge density(right column,in unit y of nc) for the s- (top row) and p-polarization (bottom row) cases at t=648/ω. Here ω is the laser frequency and distance is in unit of c/ω. FIG.4: (coloronline)Fractionoflaserenergyabsorbedbytheelectronsfordifferentlaserintensitiesandpolarizations (a)and the electron energy composition in the p-polarization case for two laser intensities at t=510 fs (b). All laser intensities and time are calculated assuming the laser wavelength is 1 µm. 9 FIG. 5: (color online) (a) The PxPy (in unit of mc) phase space of the electrons localized near the laser-plasma interface for the I =1020W/cm2 and p-polarization case (t=454 fs). (b) The rms angle (in degree) of those electrons with energy above 1 MeVfor twolaserintensities(bothp-polarization). Alllaser intensitiesandtimearecalculated assumingthelaser wavelength is 1 µm. FIG. 6: (color online) (a) The electron charge density (in unit of nc) at t = 1080c/ω. The two boxes indicate where the stability analysis in (c) and (d) are carried out. (b) The magnetic field Bz in unit of mcω/e, which is 107 MG for 1µm-laser, att=1296c/ω. Thedistancesin(a)and(b)areinunitofc/ω. (c)Thecurrentfilamentstabilityanalysisfortheshockregion (indicated by the left box in (a)). (d) The same stability analysis for the target interior (indicated by the left box in (a)). A dot abovethehorizontal axis indicates itscontribution totheinstability. Seethetext for details. Thecase shown hereis with I =1020W/cm2 and p-polarization.