mobiu-q 1.5.0

Soft Algebra Optimizer for Quantum & Complex Optimization

Mobiu-Q

Soft Algebra Optimizer for Quantum Computing

Mobiu-Q achieves +23% on VQE and +46% on QAOA compared to Adam optimizer, using a novel Soft Algebra approach that's resilient to quantum noise.

Installation

pip install mobiu-q

Quick Start

VQE (Molecular Simulation)

from mobiu_q import MobiuQCore, Demeasurement
import numpy as np

# Initialize optimizer for VQE
opt = MobiuQCore(
    license_key="your-license-key",
    problem="vqe"         # Uses Trust Ratio, lr=0.01
)

# Optimization loop
params = np.random.uniform(-np.pi, np.pi, n_params)
for step in range(60):
    energy = energy_fn(params)
    gradient = Demeasurement.finite_difference(energy_fn, params)
    params = opt.step(params, gradient, energy)

opt.end()

QAOA (Combinatorial Optimization)

from mobiu_q import MobiuQCore
import numpy as np

# Initialize optimizer for QAOA
opt = MobiuQCore(
    license_key="your-license-key",
    mode="noisy",         # For SPSA gradients
    problem="qaoa",       # Uses Super-Equation Δ†
    base_lr=0.1           # Important! Default is 0.02
)

# SPSA gradient function (built-in to your workflow)
def spsa_gradient(fn, params, c=0.1):
    delta = np.random.choice([-1, 1], size=len(params))
    E_plus = fn(params + c * delta)
    E_minus = fn(params - c * delta)
    grad = (E_plus - E_minus) / (2 * c) * delta
    energy = (E_plus + E_minus) / 2
    return grad, energy

# Optimization loop
params = np.random.uniform(-np.pi, np.pi, n_params)
for step in range(150):
    gradient, energy = spsa_gradient(qaoa_energy_fn, params)
    params = opt.step(params, gradient, energy)

opt.end()

Multi-Seed Experiments (Counts as 1 Run)

opt = MobiuQCore(license_key="your-license-key", problem="vqe")

for seed in range(10):
    opt.new_run()  # Reset optimizer state, keep session
    np.random.seed(seed)
    params = np.random.uniform(-np.pi, np.pi, n_params)
    
    for step in range(60):
        gradient = Demeasurement.finite_difference(energy_fn, params)
        params = opt.step(params, gradient, energy_fn(params))

opt.end()  # All 10 seeds count as 1 run!

Validate Your Results

Run our validation script to see the improvement on your machine:

# Download validation script
curl -O https://raw.githubusercontent.com/mobiuai/mobiu-q/main/examples/customer_validation.py

# Edit LICENSE_KEY in the file, then run:
python customer_validation.py

Expected output:

TEST 1: VQE - H2 Molecule (60 steps)
📊 IMPROVEMENT: ~23% better accuracy with Mobiu-Q

TEST 2: QAOA - MaxCut Ising (150 steps)  
📊 IMPROVEMENT: ~46% (Mobiu-Q wins 10/10 seeds)

Benchmarks

VQE (Quantum Chemistry)

Molecule	Improvement vs Adam	p-value
H₂	+23%	< 0.05
LiH	+50.6%	< 10⁻¹²
HeH⁺	+68%	< 10⁻⁴⁰

QAOA (Combinatorial, noise=10%)

Problem	Depth	Improvement	Win Rate
MaxCut	p=5	+46%	10/10
Vertex Cover	p=5	+35.77%	55/60
Max Independent Set	p=5	+30.62%	52/60

Hardware Validation

Tested on IBM Quantum Fez (156-qubit):

• Adam: Crashed to non-physical energy
• Mobiu-Q: Converged to 99.6% accuracy

Parameters

Parameter	Default	Description
problem	"vqe"	"vqe" or "qaoa"
mode	"standard"	"standard" or "noisy"
base_lr	auto	VQE: 0.01, QAOA: use 0.1

Recommended Settings

Use Case	Settings
VQE (simulator)	problem="vqe"
VQE (hardware)	problem="vqe", mode="noisy"
QAOA	problem="qaoa", mode="noisy", base_lr=0.1

How It Works

VQE: Trust Ratio

For smooth energy landscapes (molecular chemistry), Mobiu-Q uses the Trust Ratio:

φ = |S.real| / (|S.real| + |S.soft| + ε)

High trust = stable gradient = larger learning rate.

QAOA: Super-Equation Δ†

For rugged combinatorial landscapes, Mobiu-Q uses the Super-Equation:

Δ† = |Du[sin(πS)]| · g(τ,α) · Γ(a,β) · √(b·g(τ,α))

This identifies optimal "emergence points" where the optimization should act aggressively.

Pricing

• Free: 20 runs/month
• Pro: $19/month unlimited

Get your license at app.mobiu.ai

Support

• Email: [e-mail]

License

Proprietary - All rights reserved.

---

Made with ❤️ by Mobiu Technologies