A1. 40Ar/39Ar Techniques
 Pure mineral separates were prepared from crushed material using standard heavy liquid and magnetic methods. In all cases, material was handpicked to ensure sample homogeneity (>99%). Mineral separates were washed in distilled water, acetone, and ethanol prior to packaging in Al foil for irradiation. Individual packets were loaded in aluminum disks, shielded with Cd foil, and irradiated in the core position of the research reactor at McMaster University, Ontario, Canada.
 K and Ca production factors during irradiation were established by analyzing reagent grade K2SO4 and CaF2 included in each package. Isotopic correction factors are as follows: 36Ca/37Ca, 3.38 × 10−4; 38Ca/37Ca, 2.04 × 10−4; 39Ca/37Ca, 9.54 × 10−4; 36Cl/38Cl, 1481; 37K/39K, 2.07 × 10−4; 38K/39K, 1.16 × 10−2; 40K/39K, 5.00 × 10−3. Fast neutron flux was monitored using Fish Canyon sanidine (27.95 Ma) [Cebula et al., 1986] and McClure Mountain (MMhb-1) hornblende (520.4 Ma) [Samson and Alexander, 1987]. Monitors were analyzed by total fusion of 1–5 crystals with an Ar-ion laser. Flux gradients are typically negligible within a disk, but the irradiation parameter, J, may vary by up to 2% along the length of a package [Hodges et al., 1994]. We assigned the mean J calculated for a disk to all samples in that disk. We assume a conservative 2% uncertainty in J (at 2σ) for all samples in order to account for potential uncertainties in interpolation of J between monitor positions and in potential heterogeneities in monitor materials. Ages stated in the text and figures include this propagated uncertainty.
 Resistance furnace gas extraction was accomplished using a double-vacuum assembly in the Massachusetts Institute of Technology (MIT) Noble Gas Laboratories. Details of the extraction line are given by Hodges et al. . Temperatures were continuously monitored using a Re-W thermocouple; this system provides ∼5°C control on temperature over the course of a heating increment. All samples were allowed to equilibrate with the ambient furnace temperature (250°C) for 5 min prior to analysis.
 Operational blanks for the resistance furnace extraction line are dominated by the furnace and are strongly temperature dependent. Furnace system blanks were measured as a function of temperature prior to sample analysis. Blank corrections were generally small; signal sizes were typically 2 to 3 orders of magnitude larger than the furnace system blank.
 After corrections were made for interfering isotopes, mass discrimination, and blank, 39Ar/40Ar data were analyzed in a variety of ways (see supporting data Table A1). The model age for each increment of gas extraction was calculated assuming an initial 40Ar/36Ar value of 295.5 and is assigned a 2σ uncertainty that reflects propagated errors in all correction factors and the J parameter. Release spectra illustrate model ages for incremental heating analyses as a function of the amount of 39ArK in each step. Age plateaus determined from the incremental release spectra are defined as the error-weighted mean age of contiguous steps that define 50% or more of the total 39ArK released, and they are statistically indistinguishable at the 2σ confidence level, exclusive of uncertainty in the J value. For the biotites in this study, none of the spectra define plateaus, and we use a weighted mean of selected contiguous steps as our best estimate of the bulk closure age of the sample. All uncertainties associated with the ages are reported at the 2σ level and include the uncertainty in the J factor but not uncertainties in the ages of the flux monitors.
 For the K-feldspar analyses, the heating schedule was varied in order to facilitate retrieval of kinetic information for each sample [Lovera et al., 1991]. Heating schedules are presented along with 40Ar/39Ar data in supporting data Table A1. Activation energies for the samples were obtained by fitting an unweighted linear regression to the initial, low-temperature diffusion data on Arrhenius diagrams. Domain size distributions were modeled from fits to the 39Ar release data via log(r/ro) plots [Lovera et al., 1989Lovera et al., 1991; Lovera, 1992]. The model does not consider gas evolved above the temperature for incongruent melting of K-feldspar (∼1150°C). For our samples we were typically able to model between 60 and 80% of the gas.
 Excess argon is often present in the first few percent of gas released and may obscure age information in the low-temperature portion of the experiment [Harrison et al., 1993]. Recent experiments have shown that in many cases this excess argon is associated with the decrepitation of Cl-rich fluid inclusions [Harrison et al., 1993, 1994]. In order to facilitate retrieval and interpretation of age information from the early portion of the gas release we subjected all samples to duplicate isothermal heating. In all cases, the first step at a given temperature increment yielded anomalously old ages, while the second step typically yielded younger ages, presumably permitting us to see through the effects of Cl-related excess argon and to obtain a geologically meaningful age. We are thus able to model the thermal history of the samples to fairly low temperature (in some cases as low as ∼120°C).
 Once the domain structure and diffusion parameters are obtained, their thermal histories were explored using an automated inversion model [Zeitler, 1993]. The automated inversion model is based on the controlled random search algorithm (CRS), previously applied to modeling fission track data [Willett, 1997]. The CRS algorithm is used to find thermal histories, that when fed into a finite difference diffusion model, produce model age spectra (typically a set of 100) that match the observed spectrum derived from laboratory incremental heating. In describing the cooling history of our samples we used the mean temperature history for the 100 histories. In the advent of an episode of reheating, a unique solution for the thermal history is not possible. In contrast, solutions allowing only monotonic cooling tend to cluster tightly within the closure window for argon diffusion in the K-feldspar.
A2. (U-Th)/He Techniques
 Apatite was separated from crushed samples using standard magnetic and heavy liquid techniques at MIT. Mineral separates were hand picked to ensure sample homogeneity; apatites were selected on the basis of morphology, size, and the absence of visible defects and inclusions (helium associated with U- and Th-bearing silicate inclusions can produce anomalously old ages [House et al., 1997; Spotila et al., 1998]). Samples typically consisted of 10–20 crystals of apatite between 60 and 90 μm in diameter. Helium was degassed in a high-vacuum furnace and measured on a quadrupole mass spectrometer in the Noble Gas Laboratory of the California Institute of Technology (see Wolf et al.  for details of the extraction line). Analytical precision is typically 6–8% (2σ), based on reproducibility of intralaboratory standards [Wolf, 1997]. Following Farley et al. , measured He concentrations were corrected for alpha ejection. This was accomplished using a geometric factor, FT, derived from the size and shape of crystals in each sample [Farley et al., 1996]. FT values ranged from 0.58 to 0.79. Estimated uncertainty in this parameter is ∼4–6% (2σ), based on repeat measurements [Spotila et al., 1998]. After helium extraction, samples were dissolved in HNO3 and measured for U and Th contents by isotope dilution on an inductively coupled plasma mass spectrometer (ICP-MS) at California Institute of Technology (Caltech). Typical analytical precision is ∼2% [Spotila et al., 1998]. Together, these individual uncertainties propagate to yield a 2σ uncertainty of ±10% for individual ages [Spotila et al., 1998]. On the basis of replicate analyses we adopt uncertainties (2σ) for our samples of 6% (Table 1). However, for samples with low helium yield (97-14 and 94-3) we adopt uncertainties of 15% (2σ).
 Zircon ages were measured in the WSU (U-Th)/He dating lab. Helium extraction was performed using a laser heating method adapted for zircons and similar to that of House et al. . For each sample, three crystals were picked from heavy mineral separates and loaded into ∼1 mm Mo foil envelopes. Mo foils were heated in a Cu sample planchet under high vacuum by a ∼10 W, 1 mm diameter, CO2 laser beam, projected through a ZnS window. Each aliquot was heated to bright incandescence (estimated temperature ∼1500°–1700°C) for ∼5–8 min. 4He blanks measured on empty foil envelopes by this method were typically <0.02 ncc STP 4He. 4He was measured by 3He isotope dilution, cryogenic purification, and quadropole mass spectrometry, similar to the procedure at Caltech; these measurements have an estimated analytical uncertainty of ∼1%. Following helium measurements, crystals were retrieved from foil and spiked with 229Th and 233U. Crystals were dissolved in a two-step procedure involving 3 days in HF-HNO3 at 200°C, followed by 1 day in HCl at 200°C. U and Th measurements were made by isotope dilution using an HP4500-Plus quadropole mass spectrometer. U and Th measurements have an estimated analytical uncertainty of 1–2%. Alpha ejection effects were modeled using the method of Farley et al. , modified for zircon morphology, density, and stopping distances. On the basis of replicate analyses of Fish Canyon Tuff and other intralab zircon standards during the period in which these samples were run, we estimate an analytical uncertainty of 8% (2σ).