19965

17238

كارشناسي ارشد

ماده چگال

اصفهان : دانشگاه صنعتي اصفهان

1402

چهارده، 62ص. :مصور، جدول، نمودار

1403/08/29

كتابنامه

فيزيك

فيزيك

1403/09/05

23086430

ﺩﺭ ﺳﺎﻝﻫﺎﯼ ﺍﺧﯿﺮ ﺑﺮﺍﯼ ﺣﻞ ﻣﺴﺌﻠﻪ ﮐﻨﺘﺮﻝ ﺑﻬﯿﻨﻪ ﺳﯿﺴﺘﻢﻫﺎﯼ ﻧﻮﺳﺎﻧﮕﺮ⁃ﮐﯿﻮﺑﯿﺖ ﺩﺭ ﺍﻧﻮﺍﻉ ﻣﺨﺘﻠﻒ ﭘﺮﺩﺍﺯﻧﺪەﻫﺎﯼ ﮐﻮﺍﻧتمي ،ﺭﺍەﺣﻞﻫﺎﯼ ﺯﯾﺎﺩﯼ ﺑﺎ ﺍﺳﺘﻔﺎﺩﻩ ﺍﺯ ﺗﮑﻨيكﻫﺎﯼ ﯾﺎﺩﮔﯿﺮﯼ ﻣﺎﺷﯿﻦ ﺍﺭﺍﺋﻪ ﺷﺪەﺍﺳﺖ ﺍﻣﺎ ﺍﻏﻠﺐ ﺁﻥﻫﺎ ﺩﺭ ﺁﺯﻣﺎﯾﺶ ﻭﺍقعي ﻗﺎﺑﻞ ﺍﺟﺮﺍ ﻧﯿﺴﺘﻨﺪ. ﺩﺭ ﺍﯾﻦ ﭘﺎﯾﺎﻥﻧﺎﻣﻪ ﺭﻫﯿﺎﻓتي ﺭﺍ ﺟﻬﺖ ﮐﻨﺘﺮﻝ ﺟﺎمع ﻧﻮﺳﺎﻧﮕﺮ ﺩﺭ ﻣﺪﺍﺭ ‫ﺍﻟ‬كﺘﺮﻭﺩﯾﻨﺎﻣيك‬ ﮐﻮﺍﻧﺘمي (𝑐𝑄𝐸𝐷) ﺑﺎ ﺟﻔﺖﺷﺪگي ﭘﺮﺍﮐﻨﺪﻩ ﻣﻄﺎﻟﻌﻪ ميﮐﻨﯿﻢ ﮐﻪ ﺩﺭ ﺁﻥ ﻣﺴﺌﻠەﯼ ﮐﻨﺘﺮﻝ ﮐﻮﺍﻧﺘﻤ ﺭﺍ ﺑﻪ ﺻﻮﺭﺕ يك ﻓﺮﺁﯾﻨﺪ ﺗﺼﻤﯿﻢﮔﯿﺮﯼ ﻣﺎﺭﮐﻮﻑ ﻣﺸﺎﻫﺪەﭘﺬﯾﺮ ﮐﻮﺍﻧﺘﻤ ﻓﺮﻣﻮﻝﺑﻨﺪﯼ ﻣ ﮐﻨﯿﻢ ﻭ ﺑﺎ ﺍﺳﺘﻔﺎﺩﻩ ﺍﺯ ﺭﻭﺵ ﻣﻮﻧﺖ ﮐﺎﺭﻟﻮ ﺩﺭ ﯾﺎﺩﮔﯿﺮﯼ ﺗﻘﻮﯾﺖﺷﺪﻩ ﻭ ﺑﺪﻭﻥ ﺍﺳﺘﻔﺎﺩﻩ ﺍﺯ ﻫﯿﭻ مدلي ﺑﻪ ﺣﺎﻟﺖ ﻣﻮﺭﺩ ﻫﺪﻑ ﺳﯿﺴﺘﻢ ﺑﻪ ﺻﻮﺭﺕ ﺑﻬﯿﻨﻪ ميﺭﺳﯿﻢ . ﺩﺭ ﺍﯾﻦ ﺭﻫﯿﺎﻓﺖ ﯾﺎﺩﮔﯿﺮﻧﺪەﯼ ﯾﺎﺩﮔﯿﺮﯼ ﺗﻘﻮﯾﺖﺷﺪﻩ يك ﺑﺮﻧﺎﻣﻪ ﮐﻼﺳيك ﯾﺎﺩﮔﯿﺮﯼ ﻣﺎﺷﯿﻦ(ﺷبكه ﻋﺼﺒﯽ) ﺍﺳﺖ ﮐﻪ ﺍﺯ ﻃﺮﯾﻖ ﻣﺪﺍﺭ ﭘﺎﺭﺍﻣﺘﺮﯼﺷﺪﻩ ﮐﻨﺘﺮﻝ ﺑﺎ ﻣﺤﯿﻂ ﮐﻮﺍﻧﺘمي (ﻧﻮﺳﺎﻧﮕﺮ⁃ﮐﯿﻮﺑﯿﺖ) ﺗﻌﺎﻣﻞ ﺩﺍﺭﺩ. ﺍﯾﻦ ﺭﻭﺵ ﺑەﻃﻮﺭ ﮐﺎﻣﻞ ﺑﺪﻭﻥ ﺍﺳﺘﻔﺎﺩﻩ ﺍﺯ ﻣﺪلي ﺑﺮﺍﯼ ﺳﯿﺴﺘﻢ ﺍﺳﺖ ﻭ ﻗﺎﺑﻠﯿﺖﻫﺎﯼ ﮐﻨﺘﺮﻝ ﮐﻮﺍﻧﺘمي ﺭﺍ ﺩﺭ ﺯﻣﯿﻨەﻫﺎﯾﯽ ﮐﻪ ﺭﻭﺵﻫﺎﯼ ﺑﺮﺍﺳﺎﺱ ﺷﺒﯿەﺳﺎﺯﯼ ﺑﺮﺍﯼ ﺁﻥﻫﺎ ﮐﺎﺭﺁﻣﺪ ﻧﯿﺴﺘﻨﺪ، ﮔﺴﺘﺮﺵ مي ﺩﻫﺪ. ﺍﯾﻦ ﺭﻫﯿﺎﻓﺖ ﮐﻪ ﯾﺎﺩﮔﯿﺮﻧﺪﻩ ﻓﻘﻂ ﺍﺯ ﻧﺘﺎﯾﺞ ﺍﻧﺪﺍﺯەﮔﯿﺮﯼ ﮐﯿﻮﺑﯿﺖ ﺍﺳﺘﻔﺎﺩﻩ ميﮐﻨﺪ ﺭﺍ ﺑﺮﺍﯼ ﺁﻣﺎﺩەﺳﺎﺯﯼ ﺣﺎﻟﺖﻫﺎﯼ ﻋﺪﺩﯼ ﻭ ﺣﺎﻟﺖ ﺩﻟﺨﻮﺍﻩ ﻧﻮﺳﺎﻧﮕﺮ ﺍﺟﺮﺍ مي ﮐﻨﯿﻢ ﻭ ﻧﺸﺎﻥ ﺩﺍﺩەﺷﺪەﺍﺳﺖ ﮐﻪ ﺍﯾﻦ ﺭﻫﯿﺎﻓﺖ، ﺩﺭ ﺁﺯﻣﺎﯾﺶ ﻭ ﺩﻧﯿﺎﯼ ﻭﺍقعي ﺍﺟﺮﺍﺷﺪني ﺍﺳﺖ.

In recent years, numerous machine learning techniques have been proposed to address the optimal control of oscillator-qubit systems in various types of quantum processors; however, many of these solutions are not feasible for real-world implementation. In this thesis, we explore an approach for comprehensive oscillator control in a circuit quantum electrodynamics (cQED) with dispersive coupling. We formulate the quantum control problem as an quantum observable Markov decision process and achieve the target system state optimally through a model-free reinforcement learning method using Monte Carlo techniques. In this approach, the reinforcement learning agent is a classical machine learning program (neural network) that interacts with the quantum environment through a parametrized control circuit. This method is entirely model-free for the system, extending quantum control capabilities to domains where simulation-based methods are ineffective. We apply this approach—where the agent relies just on qubit measurement outcomes— for the preparation of number states and arbitrary states of the oscillator. It has been demonstrated that, this method is feasible in experimental and real-world settings.

فرهاد فضيله

كيوان آقابابائي ساماني

حميده شاكري پور , اسماعيل عبدالحسيني سارسري