perf(project_euler): replace problem_092/sol1 with digit DP (~1500x speedup) (#15141)

* perf(project_euler): replace problem_092 sol1 with digit DP

The previous solution iterated all 10,000,000 values in a Python-level
for-loop (~6 s on GitHub Actions CI).

The new approach uses digit DP over the decimal digits of (number-1):
counts in O(k * 568 * 10) ~ 40k ops how many integers in [0, number-1]
have each digit-square sum, then multiplies by a precomputed lookup of
whether each sum eventually reaches 89.
Running time on the default input drops from ~6 s to ~0.004 s (~1500x).

Closes #8594

* perf(project_euler): replace problem_092/sol1 with digit DP

* chore: remove workflow artifact (.oss-upstream)
This commit is contained in:
Ankur Kataria
2026-08-31 23:37:40 +02:00
committed by GitHub
parent 7421e02e05
commit 9391f546d6
+77 -75
View File
@@ -9,93 +9,95 @@ For example,
Therefore any chain that arrives at 1 or 89 will become stuck in an endless loop.
What is most amazing is that EVERY starting number will eventually arrive at 1 or 89.
How many starting numbers below ten million will arrive at 89?
References:
- https://en.wikipedia.org/wiki/Digital_root
- https://en.wikipedia.org/wiki/Digit_DP
"""
DIGITS_SQUARED = [sum(int(c, 10) ** 2 for c in i.__str__()) for i in range(100000)]
def next_number(number: int) -> int:
"""
Returns the next number of the chain by adding the square of each digit
to form a new number.
For example, if number = 12, next_number() will return 1^2 + 2^2 = 5.
Therefore, 5 is the next number of the chain.
>>> next_number(44)
32
>>> next_number(10)
1
>>> next_number(32)
13
def solution(number: int = 10_000_000) -> int:
"""
Returns how many starting numbers below `number` will arrive at 89
in the digit-square chain.
sum_of_digits_squared = 0
while number:
# Increased Speed Slightly by checking every 5 digits together.
sum_of_digits_squared += DIGITS_SQUARED[number % 100000]
number //= 100000
Uses digit DP so the count is computed in O(k * d_max * 10) time —
roughly 40 000 operations for number = 10^7 — instead of iterating
all `number` values explicitly.
return sum_of_digits_squared
# There are 2 Chains made,
# One ends with 89 with the chain member 58 being the one which when declared first,
# there will be the least number of iterations for all the members to be checked.
# The other one ends with 1 and has only one element 1.
# So 58 and 1 are chosen to be declared at the starting.
# Changed dictionary to an array to quicken the solution
CHAINS: list[bool | None] = [None] * 10000000
CHAINS[0] = True
CHAINS[57] = False
def chain(number: int) -> bool:
"""
The function generates the chain of numbers until the next number is 1 or 89.
For example, if starting number is 44, then the function generates the
following chain of numbers:
44 → 32 → 13 → 10 → 1 → 1.
Once the next number generated is 1 or 89, the function returns whether
or not the next number generated by next_number() is 1.
>>> chain(10)
True
>>> chain(58)
False
>>> chain(1)
True
"""
if CHAINS[number - 1] is not None:
return CHAINS[number - 1] # type: ignore[return-value]
number_chain = chain(next_number(number))
CHAINS[number - 1] = number_chain
while number < 10000000:
CHAINS[number - 1] = number_chain
number *= 10
return number_chain
def solution(number: int = 10000000) -> int:
"""
The function returns the number of integers that end up being 89 in each chain.
The function accepts a range number and the function checks all the values
under value number.
Key observations:
1. For any n < number, digit_square_sum(n) ≤ num_digits * 81,
so chain endpoints can be precomputed for that small range only.
2. A digit DP over the decimal digits of (number - 1) counts how many
integers in [0, number-1] have each possible digit-square sum,
grouping by whether the prefix is still bounded ("tight") or free.
Integers whose digit-square sum equals 0 are exactly 0 itself.
>>> solution(100)
80
>>> solution(10000000)
>>> solution(10_000_000)
8581146
"""
for i in range(1, number):
if CHAINS[i] is None:
chain(i + 1)
num_digits = len(str(number - 1)) if number > 1 else 1
limit = num_digits * 81 + 1 # max possible digit-square sum + 1
return CHAINS[:number].count(False)
def digit_square_sum(n: int) -> int:
total = 0
while n:
total += (n % 10) ** 2
n //= 10
return total
# Precompute whether each value 1..limit-1 eventually reaches 89.
# All intermediate chain values stay below limit because the digit-square
# sum of any k-digit number is at most k * 81 = limit - 1.
ends_at_89 = bytearray(limit)
for i in range(1, limit):
n = i
while n not in (1, 89):
n = digit_square_sum(n)
ends_at_89[i] = n == 89
# Digit DP over the decimal digits of (number - 1).
# Treating shorter numbers as zero-padded strings (e.g. 7 → "0000007")
# is safe because 0^2 = 0 contributes nothing to the digit-square sum.
# dp_tight[s] / dp_free[s] = count of digit sequences whose running
# digit-square sum is s and whose prefix is still ≤ / already < the
# corresponding prefix of (number - 1).
digits = [int(d) for d in str(number - 1)] if number > 1 else [0]
dp_tight: dict[int, int] = {0: 1}
dp_free: dict[int, int] = {}
for lim in digits:
new_tight: dict[int, int] = {}
new_free: dict[int, int] = {}
for dss, cnt in dp_tight.items():
for d in range(lim + 1):
new_val = dss + d * d
if new_val < limit:
if d == lim:
new_tight[new_val] = new_tight.get(new_val, 0) + cnt
else:
new_free[new_val] = new_free.get(new_val, 0) + cnt
for dss, cnt in dp_free.items():
for d in range(10):
new_val = dss + d * d
if new_val < limit:
new_free[new_val] = new_free.get(new_val, 0) + cnt
dp_tight, dp_free = new_tight, new_free
# Sum counts for all digit-square sums that end at 89.
# dss == 0 corresponds to the number 0, which is excluded.
return sum(
cnt
for dss, cnt in (*dp_tight.items(), *dp_free.items())
if 0 < dss < limit and ends_at_89[dss]
)
if __name__ == "__main__":