Replica-Exchange Optimization Control for Faster Ground-State Search

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current optimization methods, such as simulated annealing, face challenges in efficiently finding the ground state due to slow temperature reduction, leading to prolonged calculation times and increased risk of getting stuck in local solutions.

Innovation Solution

The optimization device employs a replica-exchange method with multiple search units operating at different temperatures, allowing for temperature exchanges based on an exchange probability, which speeds up the search for the ground state by adjusting the exchange count and temperature settings.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If temperature is reduced slowly to ensure theoretical convergence, then the reliability of obtaining the ground state is improved, but the calculation time becomes excessively long

Engineering Contradiction:
Improvereliability of obtaining ground stateVSAvoidcalculation time
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The system divides the search process into multiple parallel chains, each operating at different temperature levels. This segmentation allows simultaneous exploration of both high-temperature (diverse) and low-temperature (precise) regions, resolving the contradiction between thorough search (reliability) and fast convergence (time) by distributing the computational workload across multiple temperature segments

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent introduces a temperature dimension to the search space, creating a multi-dimensional optimization landscape. By adding this thermal dimension and allowing exchanges between different temperature levels, the system transforms a single-timescale problem into a multi-timescale process, enabling both rapid convergence and thorough exploration to occur simultaneously at different temperature levels

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

2Loss of time

If temperature is reduced too fast, then the calculation time is shortened, but the system gets stuck in local solutions

Engineering Contradiction:
Improvecalculation timeVSAvoidprobability of obtaining good solution
Core Design Contradiction:
Loss of timeVSReliability

Solution Approach 1:

The system performs preliminary exploration at high temperatures before committing to low-temperature convergence. By allowing chains to operate at elevated temperatures first, the system pre-explores the solution landscape and identifies promising regions, then transitions to lower temperatures with better initial knowledge, avoiding premature convergence to poor local optima

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

High-temperature chains act as intermediaries that facilitate escape from local optima. When a low-temperature chain becomes trapped, it can exchange states with high-temperature chains that have broader exploration capability, using these high-temperature states as intermediary steps to escape local minima and continue the search more effectively

Inventive Principle:
Principle #24Intermediary (Mediator)

3Productivity

If multiple search units operate in parallel, then the search efficiency is improved, but the device complexity increases

Engineering Contradiction:
Improvesearch efficiencyVSAvoidnumber of search units
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

Each search unit is designed as a universal module capable of operating at any temperature level and participating in exchanges with any other unit. This multi-functionality allows the same hardware structure to serve multiple purposes across different temperature regimes, improving search efficiency through parallelism while minimizing the increase in device complexity through structural uniformity

Inventive Principle:
Principle #6Universality (Multi-functionality)

Data Source

PatentEP3739473B1Optimization device and method of controlling optimization device
Publication Date: 2022.09.28 FUJITSU LTD
  • EP3739473B1 patent drawingFigure 1
  • EP3739473B1 patent drawingFigure 2
  • EP3739473B1 patent drawingFigure 3

AI summary

An optimization device includes: a plurality of search units that determines a plurality of energy values for a plurality of combinations of two state variables of a plurality of state variables, based on a plurality of weight values each indicating strength of connection between the two state variables, and values of the two state variables, determines values of the plurality of state variables, based on results of comparison between a plurality of evaluation values obtained by adding a noise value corresponding to a temperature to each of the plurality of energy values, and a threshold value, and are set to be equal to each other in connection states between the plurality of state variables based on the plurality of weight values, and different from each other in temperature; an exchange count determination unit that determines an exchange count, based on a difference between a temperature set for a search unit corresponding to the lowest energy value of the plurality of energy values determined by the plurality of search units, and a minimum temperature that is the lowest temperature of temperatures set for the plurality of search units; and a temperature control unit that sets a temperature for each of the plurality of search units, and exchanges the set temperatures between the plurality of search units for the exchange count.