What does Detours and the Depth of Thinking conclude, in numbers, about “The Limits of Greed: Inside and Outside the Matroid”?

Answer: Chapter 3 sets greed against dynamic programming on matroids and beyond; in the knapsack trap greed obtains 0.0002 of the optimum.

Scope of the numbers: Parameters are illustrative and all calculations are generated by the book's programs. This is not investment advice or advice on any particular business decision, and it promises no return.

中文版

Data table

The numbers below are printed verbatim by the chapter's Python program and can be reproduced by running it. standard-library Python onlyprinted values that match the answer: 0.0002

A. Trap instance: greedy / optimum
budget10: greedy 2 optimum 10 ratio 0.2000
budget100: greedy 2 optimum 100 ratio 0.0200
budget1,000: greedy 2 optimum 1,000 ratio 0.0020
budget 10,000: greedy2 optimum 10,000 ratio 0.0002
B. 1,000 random instances (12 items, budget 60)
greedy optimal in 598 (59.8%)
mean ratio 0.9828; 1st percentile 0.8663; worst 0.7410
Show the full program
# Chapter 3. p32: greedy by value density on a budget - how bad can it be?
# 0/1 knapsack (not a matroid). Greedy takes items by value per unit of
# budget while they fit; dynamic programming gives the exact optimum.
# Part A: the textbook trap - item 1 (value 2, cost 1) and item 2
# (value C, cost C) under budget C: greedy keeps item 1, then item 2
# no longer fits, so greedy / optimum = 2 / C, unbounded as C grows.
# Part B: 1,000 random instances (12 items, costs 1..30, values 1..60,
# budget 60) - typical rather than worst case. Synthetic data.
import math
import random


def greedy(items, cap):
    left, got = cap, 0
    for v, c in sorted(items, key=lambda it: (-it[0] / it[1], it[1])):
        if c <= left:
            left -= c
            got += v
    return got


def exact(items, cap):
    best = [0] * (cap + 1)
    for v, c in items:
        for b in range(cap, c - 1, -1):
            best[b] = max(best[b], best[b - c] + v)
    return best[cap]


print("A. Trap instance: greedy / optimum")
for C in (10, 100, 1000, 10000):
    items = [(2, 1), (C, C)]
    g, o = greedy(items, C), exact(items, C)
    print("   budget {:>6,}: greedy {:>3,}  optimum {:>6,}  ratio {:.4f}"
          .format(C, g, o, g / o))
rng = random.Random(38432)
ratios, hit = [], 0
for _ in range(1000):
    items = [(rng.randint(1, 60), rng.randint(1, 30)) for _ in range(12)]
    g, o = greedy(items, 60), exact(items, 60)
    ratios.append(g / o)
    hit += g == o
ratios.sort()
print("B. 1,000 random instances (12 items, budget 60)")
print("   greedy optimal in {} ({:.1%})".format(hit, hit / 1000))
print("   mean ratio {:.4f}; 1st percentile {:.4f}; worst {:.4f}".format(
    math.fsum(ratios) / 1000, ratios[10], ratios[0]))

Single source and citation

Fan, Y. (2026). When the Shortest Path Fails: Brachistochrone Detours, Second-Order Consequence Mapping and Holographic Deep Thinking (Version 1.0.0), Chapter 3: The Limits of Greed: Inside and Outside the Matroid. 5D Library. DOI pending

DOI pending · monograph page goes live after publication · 5DL-B-384 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026384,
  author    = {Fan, Yafu},
  title     = {When the Shortest Path Fails: Brachistochrone Detours, Second-Order Consequence Mapping and Holographic Deep Thinking},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

Other chapter answers in this book