Binomial MLMC: code artifact

Abstract

Software and hardware sources for multilevel Monte Carlo (MLMC) pricing of an arithmetic-average Asian call with Binomial MLMC: exact Binomial-count increments on a Cox–Ross–Rubinstein lattice, drawn from alias tables, with prices read from a ROM, so that the per-step kernel needs no multiplier.Contents: - Python MLMC level routines and drivers for Binomial MLMC, Nested MLMC (approximate 10-bit table Normals with a correction family) and Gaussian MLMC (exact and Euler stepping), with the experiments that produce the step counts and bias budget. - VHDL kernels, fleet nodes and fleet tops for Binomial MLMC and log-space Nested MLMC; GHDL testbenches; table generators; bit-accurate Python models and bit-exact replay of the RTL random streams. - Vivado 2025.1 Tcl scripts for single-kernel and full-device fleet builds, the fleet-search plan, and the Vivado summaries of the final fleet searches on six AMD UltraScale parts (XCKU035, XCVU065, XCVU125, XCVU160, XCVU190, XCVU440), with the script that regenerates the device table from them.Tests: ./run_tests.sh --quick (about 5–10 minutes) or ./run_tests.sh (about 1–2 hours) runs the software smoke tests, the reduced-scale experiments, and the hardware checks (GHDL kernel and fleet simulations matched bit for bit against the models). It needs Docker and Python 3.10 or newer; Vivado is needed only for place and route. See README.md.Built from commit a907495d53e4a10fe512f242a28ebd065e4d0469