Quantum Scheduling: The next frontier?
Analysis of Grover Search for scheduling
Insert Quantum Pun Here · Intermountain Quantum Competition 2026
Meet Master Planner
Who they are
A scheduling outsourcer that builds staff schedules for businesses across many industries
What they deliver
Constructed schedules for clients within given parameters
Our sample for this briefing: staff scheduling at a care facility
What Master Planner has been hearing
“Quantum computers try every schedule in parallel”
“Grover search runs faster and cheaper”
“O(1) computation time”
“Get in early or fall behind”
Quantum Overview
Bits and qubits
Bit
0
or
1
Always exactly one value
Qubit
α|0⟩
+
β|1⟩
A superposition of both; measuring gives 0 or 1 with probabilities |α|² and |β|²
n qubits hold amplitudes for all 2ⁿ combinations at once: 8 qubits = 256 schedules
This scales exponentially - more qubits, more possibilities
Grover in four steps
01
Superposition
Initializing all schedules equally likely
02
Oracle
Flip the sign of every valid schedule
03
Diffuser
Flip every amplitude around the average
04
Measure
Read out a schedule and check it
Each iteration rotates toward the valid schedules
Valid schedules
Invalid
θ
Start: almost all invalid
Valid schedules
Invalid
3θ
oracle flip
1 iteration: oracle flips, diffuser reflects: +2θ
Valid schedules
Invalid
5θ
~(π/4)√(N/M) iterations: on target. One more overshoots
Searching with a checker
Starting with a test that can recognize a valid schedule.
Brute force
O(N)
Grover
O(
N
)
Our sample: one day at a care facility
Staff | Day | Night |
Ana (RN) | q0 | q1 |
Cara (RN) | q2 | q3 |
Ben | q4 | q5 |
Dev | q6 | q7 |
1 qubit per decision: 1 = works that shift
Coverage · 2
2 on Day, 1 on Night
Safety · 2
An RN on every shift
Wellbeing · 4
No double shifts
Availability · 1
Ben can’t work nights
Conflict · 1
Ana and Dev apart on Day
Budget · 1
Day at most $85/hour
2⁸ = 256 schedules
11 rules
3 valid (1.2%)
Writing business rules as data
Rule('Ben_unavailable_night', 'pattern',
(bit(2, NIGHT),), {'pattern': [1]})
Rule('day_headcount', 'count',
day_bits, {'lower': 2, 'upper': 2})
Rule('day_budget_85_per_hour', 'weighted',
day_bits, {'weights': [52, 48, 30, 28],
'lower': 0, 'upper': 85})
# written in qiskit
A VALID SCHEDULE · 01000110
Shift | Staff | Cost/h |
Day | Cara (RN), Dev | $76 |
Night | Ana (RN) | $52 |
Off | Ben | – |
Headcount 2 + 1 · RN on both shifts · no doubles · Ben off nights · Ana and Dev apart · Day $76 ≤ $85
Passes all 11 rules
The oracle: count broken rules and flip if zero
for rule in problem.rules:
compute.append(gate, args) # flag = broken?
add_flag_to(counter) # counter += flag
compute.append(gate.inverse(), args)
qc.x(counter)
phase_all_ones(qc, counter) # sign flip if 0
qc.x(counter)
qc.compose(compute.inverse()) # clean up
The Grover loop
qc.h(data) # 1 superposition
for _ in range(iterations):
qc.compose(oracle.circuit) # 2 flip valid
qc.h(data); qc.x(data) # 3 diffuser:
phase_all_ones(qc, data) # reflect about
qc.x(data); qc.h(data) # the average
qc.measure_all() # 4 measure
known_iterations(8, 3) # → 7
Valid schedules
Invalid
3θ
oracle flip
In a noiseless system, Grover works
Measured over 2,000 shots: 94.5%, 99.95% and 99.7% success, versus 12.5%, 3.1% and 1.2% for a random guess
The classical opponents
def search(i, partial):
if not all(rule_still_possible(r, partial)
for r in problem.rules):
return None # prune
if i == problem.n:
return partial # valid schedule
for value in (0, 1): ... # try both
# plus brute force and Google OR-Tools CP-SAT
https://xkcd.com/835/
Fewer checks than brute force
Checks to find a valid schedule (8 qubits, 256 options)
Brute force | 64 |
Backtracking | 14 |
Grover | 8 |
CP-SAT | 0 (≈1 ms) |
But a “check” isn’t the same size for each method…
Reason 1: each quantum check is huge
One Grover check ≈ 2,300 two-qubit gates · 11 rules: 16,537 gates in total, before any error correction
Reason 2: noise on hardware drowns the answer
8 qubits at 1-in-1,000 errors: 1.2% success, the same as guessing. It takes about 1-in-100,000 errors to reach 87%
Reason 3: classical already wins at real size
17 ms
30 staff × 7 days × 3 shifts: 630 decisions, 2⁶³⁰ possible schedules
Grover would need 630+ qubits and about 2³¹⁵ iterations
Pass.
Grover’s speedup is real but small, today’s hardware drowns it in noise, and classical solvers already handle real schedules in milliseconds.
Questions?
Insert Quantum Pun Here
Joseph Davis
Andrew Nicholls
Case Watkins
Calum Sommer