After the final paper was submitted, we realized that the thresholds for the iterative solve of PAM bilinear and PAM trilinear in Table 1 were not comparable. We removed the iterative solution and now simply use a direct solve for both PAM bilinear and PAM trilinear. We also rewrote some of the code that was done in a stupid way to improve the memory usage/speed. These results made it into the presentation, but not the paper. For the same machine, the table should read |-------------------------------------------------| | | 64 | 128 | 256 | 512 | 1024 | |-------------------------------------------------| |Box | 0.000 | 0.000 | 0.001 | 0.002 | 0.010 | |Tent | 0.001 | 0.006 | 0.029 | 0.131 | 0.584 | |Lanczos3 | 0.032 | 0.151 | 0.695 | 3.140 | 13.99 | |-------------------------------------------------| |PAM Box | 0.001 | 0.002 | 0.006 | 0.021 | 0.078 | |PAM bi. | 0.111 | 0.304 | 0.971 | 3.906 | 21.40 | |PAM tri. | 0.296 | 0.880 | 3.101 | 12.57 | 61.21 | |-------------------------------------------------|