This paper uses recent theoretical work to determine the best configurations for cross-channel micromixers in optimizing mixing between two fluids. Insight into the positioning, widths, and flow protocols within the lateral channels is provided.Microfluidic devices ͑in which dimensions are typically of the order of millimeters, and which handle fluid volumes of the order of nanoliters͒ have seen a recent explosion of interest, with significant potential applications in drug delivery and monitoring, cell culture, gene profiling, chemical synthesis, "lab-on-a-chip" printing, and protein analysis. 1-6 The inevitable low Reynolds number limit in such devices means that turbulent mixing is suppressed. Diffusion by itself is not sufficiently effective in obtaining well-mixed solutions in some applications, leading to an interest in deterministic chaos as a mixing mechanism ͓see also the entire issue of 7-16 Philos. Trans. R. Soc. London, Ser. A 362, 923 ͑2004͔͒.A variety of diagnostics are frequently used as measures of mixing, such as Lyapunov exponents, effective diffusivities, transition or escape rates, etc. ͑see, for example, Refs. 15-21͒. Nevertheless, these often provide little insight into the best mixing protocols that enhance mixing, resulting in very few available investigations on mixing optimization. 12,13,20,22 Recent theoretical work, 22,23 which utilizes lobe dynamics and Melnikov techniques 24,25 to precisely quantify flux across flow separatrices, provides a method to do so. This is particularly relevant to crosschannel micromixers, 7-10 a schematic illustration of which is presented in Fig. 1. Fluids A and B, entering the micromixer from two sides, tend not to mix across the dashed line, which acts as a flow separatrix. How and where should perturbing fluid channels be placed to maximize mixing?A brief description of the theoretical basis upon which this question is to be answered will be first presented. Any perturbed flow expressible asis amenable to this analysis. Here H͑x , y͒ is the Hamiltonian for the unmixed incompressible flow ͑such as that in Fig. 1͒, and 0 ϽӶ1. The perturbing term g = ͑g 1 , g 2 ͒ is to be chosen to optimize mixing. The = 0 flow must possess a trajectory ⌫ that connects two hyperbolic stagnation points; it is across this that chaotic flux is to be assessed. Suppose ⌫ is given by the heteroclinic trajectory (x͑t͒ , ȳ͑t͒͒, thereby providing a parametrization for ⌫ with t R. By determining leading-order perturbations of this separatrix using a Melnikov analysis, 25,26 computing the area of the lobes created, 24,25 and then rationalizing the chaotic flux directly as a volume of fluid transferred per unit time, 22,23,27 it is possible to express directly the chaotic flux across ⌫ as a perturbative expansion s͑͒ + O͑ 2 ͒ in whichwhere F is the standard Fourier transform ͑see Refs. 22 and 23͒. For a given g, Eq. ͑2͒ is a computationally powerful method for determining the flux for a given perturbation and frequency, particularly so given the prevalence of Fourier transform softwar...