Akademik tadqiqotlar, tushunarli til

Verianla | O‘zbekcha akademik tadqiqotlar va ilm-fan

17 Noyabr 2026, Seshanba
VERİANLA Mustaqil ilmiy nashriyot
Menyuni ochish yoki yopish
...
Bosh sahifa / Amaliy fanlar / Matematika / Kvant Algoritmlari uchun Variatsion Dissipativ Doira
Matematika

Kvant Algoritmlari uchun Variatsion Dissipativ Doira

Variatsion dissipativ doira — standart parametrik kvant sxemalaridagi faqat unitar holat evolyutsiyasiga ancilla-yordamidagi va o‘rgatiladigan nonunitary kvant kanallarini qo‘shadigan kvant algoritmi dizayn doirasidir.

13/11/2026  Veri Anla  163 marta ko‘rildi
Kvant Algoritmlari uchun Variatsion Dissipativ Doira

Variatsion dissipativ doira — standart parametrik kvant sxemalaridagi faqat unitar holat evolyutsiyasiga ancilla-yordamidagi va o‘rgatiladigan nonunitary kvant kanallarini qo‘shadigan kvant algoritmi dizayn doirasidir. Asosiy tizim kubitlari parametrik system-only variatsion blok tomonidan qayta ishlanar ekan, alohida ancilla kubitlari tizim bilan boshqariladigan tarzda bog‘lanadi; birgalikdagi tizim–ancilla unitaridan so‘ng ancilla quyi tizimi partial trace orqali chiqariladi va qayta \(|0\rangle\) holatiga tayyorlanadi. Shu tariqa asosiy tizimda Completely Positive Trace-Preserving (CPTP), biroq umuman olganda nonunitary o‘zgarish olinadi. Tadqiqot ushbu mexanizmni ikki xil vazifada qo‘llaydi: asosiy holat energiyasini pasaytiruvchi dissipative Variational Quantum Eigensolver (VQE) va shovqinli tayyorlangan kvant holatining maqsad holatga fidelity-sini oshiruvchi dissipative state recovery.

Tadqiqotning asosiy g‘oyasi dissipatsiyani faqat kvant apparatining istalmagan yon ta’siri sifatida ko‘rish o‘rniga, to‘g‘ri tizim–muhit o‘zaro ta’siri loyihalansa, optimallashtiriladigan variatsion resurs sifatida ishlatishdir. Standart VQE sof boshlang‘ich holatga faqat unitar amallarni qo‘llaganda sof holatlar manifoldida qoladi. Ancilla bilan o‘zaro ta’sir va keyinchalik ancilla-ni tizimdan chiqarish esa asosiy tizimning mixed state-larga o‘ta olishini hamda faqat tizim ichidagi unitar evolyutsiya bilan erishib bo‘lmaydigan umumiyroq CPTP o‘zgarishlarini variatsion qidiruv fazosiga qo‘shishga imkon beradi.

Sonli VQE tajribalarida model ochiq chegara shartli nearest-neighbor spin Hamiltonianlarida baholangan. Namunaviy tajribada \(J_x=1.0\), \(h_z=0.3\), boshqa koeffitsiyentlar nol, learning rate \(lr=0.2\), dissipative round soni \(T=1\) va depolarizing-noise strength \(p=0.1\) ishlatilgan. \(n=3\) va \(n=5\) asosiy tizimlarda ancilla soni ortgani sari energiya dastlabki iteratsiyalarda tezroq kamaygan va final energiya ancilla-siz VQEga nisbatan pastroq darajalarga yetgan. Exact ground-state energy qiymatlari, boshqacha ko‘rsatilmagan bo‘lsa, tegishli Hamiltonian matritsalarining exact diagonalizationidan olingan.

State-recovery tajribalarida W state, uniform superposition state va dressed cluster state bo‘lgan uch xil maqsad tuzilma ishlatilgan. \(n=3\), \(m=3\), \(T=3\), \(lr=0.8\) va \(p=0.1\) sharoitidagi target-state taqqoslashida uchala maqsad uchun ham final fidelity 100 ta optimallashtirish iteratsiyasi ichida %99 dan yuqoriga ko‘tarilgan. Biroq resurs miqdori cheksiz oshirilganda uzluksiz foyda kuzatilmagan: mualliflar aksariyat testlarda \(T=3\) ni yaxshi muvozanat deb belgilagan va maqsad holatga bog‘liq ancilla saturation xatti-harakatini kuzatgan.

Tadqiqotning barcha natijalari sonli simulyatsiyalardan olingan; haqiqiy kvant protsessorida eksperimental tasdiqlash amalga oshirilmagan. Bundan tashqari, B ilovadagi “conditional exponential advantage” natijasi haqiqiy apparat wall-clock tezlashuvi yoki shartsiz kvant ustunligi emas. Bu natija dissipative objective muayyan global Polyak–Łojasiewicz shartini bajarishi va ordinary unitary VQAning optimallashtirish trayektoriyasi exponentially flat bo‘lib qolishi kabi kuchli farazlar ostida gradient-descent iteratsiya murakkabligining nazariy taqqoslanishidir.

Variatsion Dissipativ Kvant Doirasi Nima?

Variatsion dissipativ kvant doirasi parametrik unitar sxemalarni ancilla-yordamidagi, o‘rgatiladigan ochiq-tizim kvant kanallari bilan birlashtirib, asosiy tizimning variatsion holat fazosini faqat sof holatlar bilan cheklangan unitar ansatzdan tashqariga kengaytiradigan usuldir; tizim–ancilla o‘zaro ta’siri unitar bo‘lsa ham, ancilla-ning keyinchalik partial trace orqali chiqarilishi asosiy tizimda samarali ravishda nonunitary CPTP o‘zgarish hosil qiladi.

Standart VQE avval nima qiladi?

Variational Quantum Eigensolver (VQE) muayyan Hamiltonianning ground-state energy qiymatini taxminan topish uchun parametrik kvant sxemasini klassik optimallashtirish bilan birlashtiradi. Boshlang‘ich holat odatda \(|0\rangle^{\otimes n}\) ko‘rinishida bo‘ladi va \(U(\theta)\) parametrik unitari bu holatni trial statega aylantiradi.

\[ \rho_\theta = U(\theta) |0\rangle\langle0|^{\otimes n} U^\dagger(\theta). \tag{1} \]

\(\rho_\theta\) variatsion trial statening density matrix ifodasidir. \(n\) — asosiy tizimdagi kubitlar soni; \(\theta\) — parametrik gatelarning optimallashtiriladigan parametrlari. Boshlang‘ich holat sof va \(U(\theta)\) unitar bo‘lgani sabab Eq. (1) dagi holat sof bo‘lib qoladi.

VQEning energiya objective-i:

\[ E(\theta) = \operatorname{Tr} \left[ H\rho_\theta \right]. \tag{2} \]

\(H\) — ground state qidirilayotgan Hamiltonian; \(\operatorname{Tr}\) — trace amali. Bu ifoda trial statening Hamiltonian expectation valuesini hisoblaydi.

Variational principle:

\[ E(\theta) = \operatorname{Tr} \left[ H\rho_\theta \right] \geq E_0 \tag{3} \]

tengsizlikni beradi. \(E_0\) — haqiqiy ground-state energy. Shu sababli VQE \(E(\theta)\) ni kamaytirishga urinadi; ideal holatda trial state ground statega yaqinlashgan sari energiya \(E_0\) ga yaqinlashadi, biroq variational principle ostida undan pastga tushmaydi.

Dissipativ kengaytma nimani o‘zgartiradi?

Standart unitar ansatz o‘rniga parametrli dissipative quantum channel ishlatilganda trial state:

\[ \rho_\theta = \mathcal{D}_\theta \left( |0\rangle\langle0|^{\otimes n} \right) \tag{4} \]

ko‘rinishida yoziladi. \(\mathcal{D}_\theta\) — parametrli CPTP kanal. Energy objective Eq. (2) bo‘lib qoladi; o‘zgaradigan narsa optimallashtirish erisha oladigan holatlar oilasidir. Asosiy tizim endi faqat sof holatlarni emas, ancilla bilan o‘zaro ta’sirdan keyin hosil bo‘lishi mumkin bo‘lgan mixed state-larni ham ifodalay oladi.

Bu nuqta “dissipative VQE albatta standard VQEdan yaxshiroq natija beradi” degani emas. Tenglama faqat qidiruv fazosi umumiyroq quantum-channel oilasiga kengayganini ko‘rsatadi; haqiqiy foyda sonli optimallashtirish natijalari bilan alohida sinovdan o‘tkazilgan.

Sof holat bilan mixed state orasidagi farq

Sof kvant holati bitta state vector \(|\psi\rangle\) bilan to‘liq ta’riflanadi va density matrixi \(\rho=|\psi\rangle\langle\psi|\) ko‘rinishida bo‘ladi. Mixed state esa turli sof holatlarning klassik yoki muhitdan kelib chiqqan noaniqlikni o‘z ichiga olgan statistik aralashmasini ifodalaydi va umuman bitta state vector bilan ifodalanmaydi.

Ancilla bilan asosiy tizim o‘zaro chirmashgandan so‘ng ancilla ma’lumotini e’tiborga olmaslik asosiy tizimning o‘zi mixed state ko‘rinishiga olib kelishi mumkin. Bu dissipativ doirani shunchaki chuqurroq unitar circuit sifatida talqin qilmaslik kerakligining asosidir.

Ochiq kvant tizimi nima?

Ochiq kvant tizimi — muhitidan to‘liq izolyatsiya qilinmagan va tashqi degrees of freedom bilan o‘zaro ta’sirlasha oladigan kvant tizimi. Bu o‘zaro ta’sir coherence yo‘qolishiga, entropy almashinuviga va axborotning tizimdan muhitga oqishiga olib kelishi mumkin.

Born–Markov yondashuvi ostida bunday reduced dynamics uchun manba Gorini–Kossakowski–Lindblad–Sudarshan (GKLS) master equationni beradi:

\[ \frac{d\rho_S}{dt} = -i[H_S,\rho_S] + \sum_k \gamma_k \left( L_k\rho_S L_k^\dagger - \frac{1}{2} \left\{ L_k^\dagger L_k,\rho_S \right\} \right). \tag{5} \]

\(\rho_S\) — asosiy tizimning density matrixi; \(H_S\) — tizim Hamiltoniani; \(L_k\) — Lindblad jump operator; \(\gamma_k\) — tegishli dissipative processning rate parametri. \([A,B]=AB-BA\) commutator, \(\{A,B\}=AB+BA\) anticommutator.

Birinchi had yopiq-tizim unitar Hamiltonian evolyutsiyasini, yig‘indi ichidagi hadlar esa muhit bilan o‘zaro ta’sirdan kelib chiqadigan dissipative o‘zgarishni ifodalaydi.

Muhim farq: Maqolaning sonli sxema usuli Eq. (5) dagi GKLS differential equationni bevosita vaqt-integratsiyasi bilan yechmaydi. GKLS tenglamasi ochiq-tizim fizikasining nazariy fonini tushuntiradi; taklif qilingan haqiqiy circuit construction ancilla-yordamidagi CPTP kanallari orqali qurilgan.

Stinespring Dilation Kvant Kanalini Qanday Hosil Qiladi?

Stinespring dilation asosiy tizim ustidagi istalgan CPTP kvant kanalini kattaroq tizim–ancilla Hilbert fazosida unitar evolyutsiya va so‘ng ancilla-ni partial trace orqali chiqarish shaklida ifodalash mumkinligini aytadi; ushbu tadqiqot bu matematik ekvivalentlikdan o‘rgatiladigan tizim–ancilla interactionni loyihalash uchun foydalanadi.

Stinespring ifodasi

\[ \mathcal{E}(\rho_S) = \operatorname{Tr}_A \left[ V_{SA} \left( \rho_S\otimes|0\rangle\langle0|_A \right) V_{SA}^\dagger \right]. \tag{6} \]

\(\mathcal{E}\) — asosiy tizimda samarali quantum channel; \(V_{SA}\) — system \(S\) va ancilla \(A\) ustidagi birgalikdagi unitar; \(|0\rangle_A\) — ancilla boshlang‘ich statei; \(\operatorname{Tr}_A\) — ancilla ustidagi partial trace amali.

Partial trace “ancilla fizik jihatdan yo‘q qilindi” degani emas. Matematik jihatdan tizimning reduced density matrixini olishda ancilla degrees of freedom ma’lumotini natijadan chiqarishdir. Sxema darajasida bu g‘oya ancilla-ni discard qilish yoki reset qilib qayta ishlatishga mos kelishi mumkin.

CPTP nima degani?

Completely Positive Trace-Preserving (CPTP) map — yaroqli density matrixlarni yaroqli density matrixlarga aylantiradigan standart fizik quantum-channel sinfi. Trace-preserving bo‘lishi jami ehtimollikning 1 bo‘lib qolishini, complete positivity esa tizim kattaroq tayanch tizim bilan entangled bo‘lsa ham o‘zgarish fizik jihatdan yaroqli bo‘lib qolishini ta’minlaydi.

Kraus ifodasi

Ancilla-ning orthonormal \(\{|i\rangle_A\}\) bazisidagi proyeksiyasi:

\[ K_i = \langle i|_A V_{SA} |0\rangle_A \]

Kraus operatorlarini beradi va kanal:

\[ \mathcal{E}(\rho_S) = \sum_i K_i\rho_S K_i^\dagger \tag{7} \]

ko‘rinishida yozilishi mumkin. Stinespring dilation bilan Kraus representation ikki xil fizik jarayon emas; bir xil CPTP o‘zgarishning ikki matematik ifodasidir.

Maqoladagi o‘rgatiladigan dissipative channel

Taklif qilingan blok Stinespring shaklini parametrik holga keltiradi:

\[ \mathcal{D}_{\theta}(\rho_S) = \operatorname{Tr}_A \left[ V_{SA}(\theta) \left( \rho_S \otimes |0\rangle\langle0|_A^{\otimes m} \right) V_{SA}^{\dagger}(\theta) \right]. \tag{8} \]

\(m\) — ancilla kubitlar soni. Trainable parameters \(\theta\) bevosita system–ancilla interaction ichida joylashadi. Ancilla har bir qo‘llashdan oldin \(|0\rangle^{\otimes m}\) holatiga tayyorlanadi va qo‘llash oxirida trace-out/reset qilinadi.

Manbaning numerical implementationida dissipative module tegishli kubitlarda parametrik single-qubit rotations va system–ancilla \(i\mathrm{SWAP}\) couplingsdan foydalanadi. Bog‘lanish fully connected: har bir system qubit har bir ancilla qubit bilan ulanadi. Shuning uchun \(n\) system qubit va \(m\) ancilla uchun bitta dissipative round ichida:

\[ N_{i\mathrm{SWAP}} = nm \]

system–ancilla \(i\mathrm{SWAP}\) interaction mavjud. Bu tenglik maqolaning izohidan bevosita keladi; umumiy circuit gate count bundan kattaroq, chunki single-qubit rotations va system-only variational layers alohida mavjud.

Parametrik iSWAP nima qiladi?

Manba Qo‘shimcha 8-rasmda ishlatilgan \(i\mathrm{SWAP}\)-turidagi couplingni quyidagicha ko‘rsatadi:

\[ i\mathrm{SWAP}(\phi) = \begin{pmatrix} 1 & 0 & 0 & 0\\ 0 & \cos(\phi/2) & i\sin(\phi/2) & 0\\ 0 & i\sin(\phi/2) & \cos(\phi/2) & 0\\ 0 & 0 & 0 & 1 \end{pmatrix}. \]

Bu gate \(|01\rangle\) va \(|10\rangle\) komponentlarini parametrik tarzda bir-biri bilan aralashtiradi; \(|00\rangle\) va \(|11\rangle\) bloklari ko‘rsatilgan matritsa ichida o‘zgarmas qoladi. Maqoladagi ilmiy roli asosiy tizim bilan ancilla o‘rtasida boshqariladigan state/information exchange hosil qiluvchi trainable coupling ta’minlashdir.

Variatsion dissipativ doiraning umumiy mexanizmi. Asosiy tizim va ancilla birgalikda unitar tarzda qayta ishlanadi; ancilla trace-out/resetdan so‘ng asosiy tizim samarali CPTP va umuman nonunitary kanaldan o‘tgan bo‘ladi. Sxema manba 1, 8 va 9-rasmlarining ilmiy munosabatlaridan mustaqil ravishda qayta yaratilgan.

Dissipative VQE bilan Standart VQE O‘rtasidagi Asosiy Farq Nima?

Standart VQE sof boshlang‘ich holatni tizimda faqat parametrik unitarlar bilan o‘zgartirsa, dissipative VQE ushbu unitar qatlamlar orasiga ancilla-yordamidagi CPTP kanallarni qo‘shadi; shu tariqa energiyani minimallashtirish coherent system-only parametrlarni ham, tizim–ancilla interaction parametrlarini ham birgalikda optimallashtiradi va asosiy tizimning reduced statei mixed state-larni ham o‘z ichiga olgan kengroq variatsion oilada harakatlana oladi.

Dissipative VQEning rekursiv ta’rifi

Boshlang‘ich asosiy tizim statei:

\[ \rho_{\mathrm{in}} = |0\rangle\langle0|^{\otimes n} \]

olinadi. Birinchi system-only variational transformation:

\[ \rho_S^{(0)} = U_{\mathrm{vqe}}^{(0)} \rho_{\mathrm{in}} U_{\mathrm{vqe}}^{(0)\dagger}. \tag{9} \]

So‘ng \(t=1,\ldots,T\) dissipative rounds uchun:

\[ \rho_S^{(t)} = \operatorname{Tr}_A \left[ U_{\mathrm{dis}}^{(t)} \left( U_{\mathrm{vqe}}^{(t)} \rho_S^{(t-1)} U_{\mathrm{vqe}}^{(t)\dagger} \otimes |0\rangle\langle0|_A^{\otimes m} \right) U_{\mathrm{dis}}^{(t)\dagger} \right]. \tag{10} \]

Har bir round ichida avval asosiy tizimning coherent variational transformationi, so‘ng system–ancilla dissipative block qo‘llanadi. Ancilla keyin trace-out qilinadi, \(|0\rangle^{\otimes m}\) ga reset qilinadi va keyingi round uchun qayta ishlatiladi.

Final loss:

\[ \mathcal{L}_{\mathrm{VQE}} = \operatorname{Tr} \left[ H\rho_S^{(T)} \right]. \tag{11} \]

bo‘ladi. Ya’ni algoritmning maqsadi hanuz energiyani minimallashtirishdir; yangilik trial stateni hosil qiladigan o‘zgarishlar oilasidadir.

State recovery nega boshqa objective ishlatadi?

State-recovery vazifasida maqsad ground-state energy qidirish emas. Shovqin sabab target statedan uzoqlashgan \(\rho_{\mathrm{in}}\) holatini yana maqsadga yaqinlashtirishdir. Shu sababli system-only VQE qatlamlari olib tashlanadi va ayni ancilla-assisted dissipative block bevosita recovery channel sifatida o‘qitiladi.

Target state sof olinadi:

\[ \rho_{\mathrm{tar}} = |\psi_{\mathrm{tar}}\rangle \langle\psi_{\mathrm{tar}}|. \tag{12} \]

Boshlanishi:

\[ \rho_0 = \rho_{\mathrm{in}}. \tag{13} \]

Har bir recovery round:

\[ \rho_t = \operatorname{Tr}_A \left[ U_{\mathrm{dis}}^{(t)} \left( \rho_{t-1} \otimes |0\rangle\langle0|_A^{\otimes m} \right) U_{\mathrm{dis}}^{(t)\dagger} \right], \qquad t=1,\ldots,T. \tag{14} \]

bilan ta’riflanadi.

Fidelity nimani o‘lchaydi?

Fidelity ikki quantum statening qanchalik yaqinligini o‘lchaydigan o‘lchamsiz o‘xshashlik metrikasidir. Umumiy mixed-state shakli:

\[ F_T = F(\rho_T,\rho_{\mathrm{tar}}) = \left( \operatorname{Tr} \sqrt{ \sqrt{\rho_{\mathrm{tar}}} \rho_T \sqrt{\rho_{\mathrm{tar}}} } \right)^2. \tag{15} \]

Target state sof bo‘lganda bu ifoda:

\[ F_T = \operatorname{Tr} \left( \rho_T\rho_{\mathrm{tar}} \right) \tag{16} \]

ko‘rinishiga soddalashadi. \(F_T=1\), final state bilan sof target state o‘rtasidagi maksimal fidelityni anglatadi.

Recovery loss:

\[ \mathcal{L}_{\mathrm{rec}} = 1-F_T \tag{17} \]

deb ta’riflanadi. Fidelity oshgan sari loss kamayadi.

Dissipatsiya bilan recovery nega bir xil tuzilma ichida ko‘rib chiqilishi mumkin?

Ayni system–ancilla joint unitary turlicha qayta ishlanishi mumkin. Ancilla e’tiborga olinmay trace-out qilinsa, asosiy tizimda unconditional dissipative channel olinadi. Ancilla o‘lchanib, measurement outcome conditional correction uchun ishlatilsa, ayni kengaytirilgan tizim recovery amali sifatida ham talqin qilinishi mumkin. Bu tadqiqot o‘z recovery protokolida ancilla-ni trace-out/reset qilib o‘rgatiladigan channel yondashuvini ishlatadi.

Tadqiqot Usuli va Natijalari

Benchmark Hamiltoniani

Dissipative VQE ochiq chegara shartli nearest-neighbor spin Hamiltonianida sinovdan o‘tkazilgan:

\[ H = \sum_{i=1}^{n-1} \left( J_x X_iX_{i+1} + J_y Y_iY_{i+1} + J_z Z_iZ_{i+1} \right) + \sum_{i=1}^{n} \left( h_x X_i + h_y Y_i + h_z Z_i \right). \tag{18} \]

\(X_i\), \(Y_i\) va \(Z_i\) — \(i\)-kubitdagi Pauli operators. \(J_x,J_y,J_z\) nearest-neighbor interaction strengths; \(h_x,h_y,h_z\) local-field coefficients. Manba bu koeffitsiyentlar yoki energiya qiymatlari uchun fizik joule/eV turidagi birlikni belgilamaydi; shuning uchun Verianla ularga manbada bo‘lmagan fizik birlikni bermaydi.

Birinchi yig‘indi qo‘shni spin juftlarining interactionini, ikkinchi yig‘indi esa har bir spin ustidagi local fieldsni ifodalaydi. Exact ground-state energy tayanchlari, boshqacha ko‘rsatilmagan bo‘lsa, Hamiltonian matritsasining exact diagonalizationi bilan hisoblangan.

Birinchi VQE benchmarki

Namunaviy tajriba:

\[ J_x=1.0, \qquad h_z=0.3 \]

tanlovi bilan bajarilib, boshqa coefficients nolga o‘rnatilgan. Shu tariqa XX-type interaction bilan transverse Z field birgalikda ishlatilgan.

ParametrManbada ishlatilgan qiymatIlmiy roli
Asosiy tizim o‘lchami\(n=3\) va \(n=5\)Ikki xil muammo o‘lchami
Ancilla sonlari\(m=0,1,3,5,7\)Dissipative channel resource skani
Learning rate\(lr=0.2\)Optimallashtirish qadam kattaligi
Dissipative round\(T=1\)Bitta dissipative update round
NoiseDepolarizing, \(p=0.1\)Shovqinli optimallashtirish sharti
Kuzatilgan tendensiyaYuqoriroq \(m\) → odatda tezroq energiya pasayishi va pastroq final energiyaAncilla-assisted channelning optimallashtirishga ta’siri

Manba grafigida x o‘qi optimallashtirish iteratsiyasini, y o‘qi energy qiymatini ko‘rsatadi. \(n=3\) misolida egri chiziqlar ancilla soni bilan bosqichma-bosqich ajralsa, \(n=5\) misolida farq yaqqolroq bo‘ladi. Bu kuzatuv kattaroq \(m\) ko‘rilgan sharoitlarda yaxshiroq natija berganini ko‘rsatadi; barcha \(m\) qiymatlari uchun cheksiz monoton yaxshilanish kafolati emas.

State-recovery maqsad holatlari

Manba uch xil state structureni tanlagan.

1. W state:

\[ |\psi_W\rangle = |W_n\rangle = \frac{1}{\sqrt{n}} \sum_{i=1}^{n} |0\cdots010\cdots0\rangle. \tag{19} \]

W state computational basis states ichida aynan bitta excitationning barcha \(n\) joy bo‘yicha teng amplitude bilan superpozitsiyasidir. \(1/\sqrt{n}\) koeffitsiyenti state normalizationni ta’minlaydi.

2. Uniform superposition state:

\[ |\psi_+\rangle = |+\rangle^{\otimes n} = \frac{1}{\sqrt{2^n}} \sum_{x\in\{0,1\}^n} |x\rangle. \tag{20} \]

Bu yerda barcha \(2^n\) computational-basis bitstring bir xil amplitude bilan mavjud.

3. Dressed cluster state:

\[ |\psi_{\mathrm{DC}}\rangle = \prod_{d=1}^{D} \left( \prod_{i=1}^{n-1} CZ_{i,i+1} \right) \left( \bigotimes_{i=1}^{n} R_y \left( \alpha_i^{(d)} \right) \right) |+\rangle^{\otimes n}. \tag{21} \]

Manba:

\[ D=3, \qquad \alpha_i^{(d)}=\frac{\pi}{4} \]

qiymatlarini barcha \(i\) va \(d\) uchun ishlatadi. \(CZ_{i,i+1}\) neighboring qubits orasidagi controlled-Z entangling gate; \(R_y\) esa Y o‘qi atrofidagi parametrik single-qubit rotation.

State recoveryda ancilla ortishi

Manba 3-rasmda ancilla-free \(m=0\) baseline bilan \(m=1,2,3\) dissipative configurations taqqoslanadi. X o‘qi iteration, y o‘qi fidelity. Dissipative ancilla qo‘shilgandan keyin fidelity noisy-input baselinedan yuqoriga chiqadi va umuman olganda kattaroq \(m\) ham tezroq recovery, ham yuqoriroq final fidelity hosil qiladi.

Turli noise channellarda chidamlilik

Manba uch asosiy noise turini sinaydi:

  • Amplitude Damping (AD): energiya relaxationga o‘xshash \(|1\rangle\rightarrow|0\rangle\) jarayonlarini ifodalovchi standart noise modeli.
  • Bit-Flip (BF): computational basis statelarni Pauli-X xatosi bilan o‘zgartira oladigan noise modeli.
  • Depolarizing (DP): stateni random Pauli/mixed-state tomonga buzadigan standart noise modeli.

Bu uch qisqa izoh shovqin belgilarining ma’nosini o‘rgatish uchun berilgan; maqolaning yangi ilmiy topilmasi emas.

Manbada “fully noisy setting” deyilganda noise faqat asosiy system variational circuitga emas, dissipative blockslarga ham qo‘llanadi. Shunga qaramay dissipative VQE ancilla-free baselinega qaraganda tezroq yaqinlashgan, kamroq tebrangan va pastroq final energiesga erishgan.

Bir jumlada “DP, BP, AD” deb yozilganiga qaramay, maqolaning avvalgi ta’rifi va tegishli grafik legendi Bit-Flip uchun “BF”ni ishlatadi. Verianla buni alohida “BP” noise modeli sifatida talqin qilmaydi; manba ichidagi notatsiya nomuvofiqligi sifatida saqlaydi.

Uch xil Hamiltonian ustidagi robustness testi

Manba quyidagi uch Hamiltonianni ham taqqoslaydi:

\[ H_1 = \sum_{i=1}^{n-1} X_iX_{i+1} + 0.3 \sum_{i=1}^{n} Z_i, \tag{22} \]
\[ H_2 = \sum_{i=1}^{n-1} Z_iZ_{i+1} + 0.3 \sum_{i=1}^{n} X_i, \tag{23} \]
\[ H_3 = \sum_{i=1}^{n-1} X_iX_{i+1} + \sum_{i=1}^{n-1} Y_iY_{i+1} + 0.3 \sum_{i=1}^{n-1} Z_iZ_{i+1}. \tag{24} \]

\(H_1\) transverse Z field ichidagi XX-type model; \(H_2\) transverse-field Ising model; \(H_3\) XXZ-type model sifatida ta’riflanadi. Bu taqqoslash fully noisy settingda va depolarizing noise \(p=0.01\) ostida o‘tkazilgan.

Uch Hamiltonianning barchasida dissipative VQE ancilla-free baselinega qaraganda tezroq convergence, smooth training trajectory va exact ground-state energyga yaqinroq final energy ko‘rsatgan. Bu natija usul unumdorligi faqat bitta energy landscapega xos emasligini qo‘llab-quvvatlaydi; barcha Hamiltonian sinflari uchun universal ustunlikni isbotlamaydi.

Turli target state va preparation noise natijalari

ParametrQiymat
System qubit\(n=3\)
Ancilla\(m=3\)
Dissipative round\(T=3\)
Learning rate\(lr=0.8\)
Noise strength\(p=0.1\)
Noise comparisonDP, BF va AD
Target-state comparisonW, uniform superposition va dressed cluster

Uch target statening \(p=0.1\) depolarizing-noise bilan tayyorlangan inputlarida dissipative recovery final fidelityni 100 optimization iteration ichida %99dan yuqoriga ko‘targan.

Bu %99 natija bevosita uch target-state experimentga tegishli. Haqiqiy quantum hardware fidelity emas va boshqa system sizelarga avtomatik umumlashtirilmasligi kerak.

Ko‘proq Ancilla yoki Ko‘proq Dissipative Round Har Doim Yaxshimi?

Yo‘q. Maqolaning resource scanlari unumdorlik avval ancilla va dissipative-round ortishidan foyda olishini, biroq ma’lum nuqtadan keyin qo‘shimcha resource faqat kichik foyda yoki ko‘proq training/circuit overhead hosil qilishini ko‘rsatadi; recovery vazifasidagi ancilla saturation nuqtasi target statega bog‘liq va ko‘pchilik round taqqoslashlarida \(T=3\) muvozanatli tanlov sifatida ishlatilgan.

Dissipative round sonining ta’siri

VazifaRasmda berilgan asosiy shartlarAsosiy kuzatuv
Dissipative VQE\(n=3\), \(m=3\), \(p=0.013\)Juda kam round yetarli emas; namunaviy testda \(T=3\) eng yaxshi umumiy energiya muvozanatini beradi.
State recovery\(n=2\), \(m=3\), \(p=0.019\)\(T=4\) yoki \(T=5\) taxminan 55 iteratsiyadan keyin \(T=3\)ni oz miqdorda ortda qoldirishi mumkin; erta convergence zaifroq va training uzoqroq.

Bu ikki panel bir xil fizik/hyperparameter sharoitlarini ishlatmaydi; manba ularni bevosita bir-biriga ustunlik taqqoslash uchun emas, o‘xshash resource trade-off ikkala vazifada ham mavjudligini ko‘rsatish uchun beradi.

Ancilla saturation

Target stateTaxminiy to‘yinishTalqin
W state\(m\approx2\)Qo‘shimcha ancilla-lar bundan keyin juda cheklangan qo‘shimcha fidelity gain beradi.
Uniform superposition\(m\approx5\)Ko‘rilgan uch state ichida yuqoriroq ancilla ehtiyojini ko‘rsatadi.
Dressed cluster\(m\approx3\)O‘rta darajadagi ancilla to‘yinishi kuzatiladi.

Manba to‘yinishni faqat target statening entanglement miqdori bilan tushuntirib bo‘lmasligini ta’kidlaydi. Zarur ancilla resource preparation circuit tuzilishi, noise-induced error modes va tanlangan dissipative ansatzning shu error modes bilan expressive compatibilitysiga birgalikda bog‘liq.

Manba 7-rasmda shaded region statistical confidence interval emas. Captionga ko‘ra, u har bir target state uchun turli noise strengths ostida olingan eng yuqori va eng past fidelity curves orasidagi envelope. Shu sababli soyali hududni standart og‘ish yoki %95 ishonch oralig‘i sifatida talqin qilish noto‘g‘ri.

Manba 6, 7 va 11-rasmlarda xabar qilingan resource trade-off va ancilla-saturation munosabatlaridan tayyorlangan konseptual Verianla sxemasi. Egri chiziqlarning xom sonli ma’lumotlari berilmagani uchun grafik rekonstruksiyasi qilinmagan.

Noise strength skani

Qo‘shimcha 10-rasm ordinary VQE bilan besh ancilla ishlatadigan dissipative VQEni turli depolarizing-noise strengths ostida taqqoslaydi. Tasvirda:

\[ p\in \{ 0.00,\, 0.02,\, 0.04,\, 0.06,\, 0.08,\, 0.10 \} \]

qiymatlar ko‘rsatilgan. Ancilla-free VQEda noise strength oshgani sari converged energy bilan theoretical/exact ground-state value o‘rtasidagi farq kattalashadi. \(m=5\) dissipative VQEda esa ayni noise darajalarida convergence tezroq qoladi va final energies exact referencega yaqinroq bo‘ladi.

Bu natija “dissipative circuit shovqindan ta’sirlanmaydi” degani emas. Manbaning da’vosi nisbiy robustness: ko‘rilgan sharoitlarda ayni noise ostida dissipative model ordinary baselinedan pastroq energy gap qoldiradi.

W-state fidelity gain va “best 5 seeds” izohi

Qo‘shimcha 11-rasm W-state recovery uchun \(F_m-F_0\) fidelity gainni noise strengthga nisbatan ko‘rsatadi va \(m\ge2\)dan keyin gain egri chiziqlarining katta darajada ustma-ust tushishini ancilla saturation dalili sifatida ishlatadi. Rasm panellarida “The Best 5 Seeds” iborasi mavjud; biroq taqdim etilgan asosiy PDF matni jami random-seed sonini va bu besh seedni tanlash qoidasini batafsil statistical protocol sifatida tushuntirmaydi. Shu sababli ushbu rasmdan confidence interval yoki population-level uncertainty hosil qilinmagan.

Statistik hisobot chegarasi

Tadqiqot klassik gipoteza testi dizayniga ega emas. Asosiy matnda p-value, confidence interval yoki effect-size testi xabar qilinmaydi. Bu yerdagi \(p\) belgisi ko‘pincha noise strength / error probability ma’nosini anglatadi; statistik p-qiymat emas.

Grafiklardagi tendensiyalar optimallashtirish trajectories orqali baholanadi va ground-state reference exact diagonalizationdan olinadi. Shu sababli Verianla natijalarga manbada bo‘lmagan “statistik ahamiyatlilik” yorlig‘ini qo‘shmaydi.

Reproducibility nuqtai nazaridan xabar qilinmagan jihatlar

Taqdim etilgan PDF ayrim muhim hyperparameterlarni beradi: learning rate, qubit/ancilla soni, dissipative round soni, noise strength, Hamiltonian coefficients va target-state tayyorlash tuzilmalari. Biroq asosiy numerical section ishlatilgan software framework, optimizerning to‘liq nomi, gradient-evaluation usuli, random initialization distribution va asosiy natija grafiklarining barcha seed/repetition protocolini batafsil xabar qilmaydi.

B ilova nazariy jihatdan gradient descentni tahlil qiladi; buni asosiy simulyatsiyalarda ishlatilgan optimizer albatta B ilovadagi idealized gradient-descent modelining aynan o‘zi bo‘lgan deb talqin qilmaslik kerak.

Dissipative VQA uchun “Shartli Eksponensial Afzallik” Nimani Anglatadi?

Maqoladagi shartli eksponensial afzallik haqiqiy kvant protsessorida o‘lchangan wall-clock speedup emas; muayyan smoothness, global Polyak–Łojasiewicz va trajectory-uniform flatness farazlari ostida gradient descent dissipative VQA objectiveni polinom sonli iteratsiyalarda kamaytira olsa, ordinary unitary VQA ayni target gapga yetish uchun eksponensial sonli iteratsiya talab qilishi mumkinligini ko‘rsatuvchi nazariy iteration-complexity ajralishidir.

B ilova nega asosiy sonli tajribalardan alohida o‘qilishi kerak?

B ilova ordinary VQA bilan dissipative VQEni abstraktroq optimization modelida taqqoslaydi. Bu yerda dissipative VQA periodic non-unital reset operationsni o‘z ichiga olgan reset-assisted quantum channel sifatida ta’riflanadi. Bu tuzilma asosiy matndagi ancilla-assisted dissipative g‘oyasi bilan ayni ochiq-tizim motivatsiyasini bo‘lishadi; biroq theoremlar asosiy sonli grafiklarning shunchaki qayta formulalangan shakli emas.

Ordinary VQA ta’rifi: Tenglama (25)–(26)

Parametrik unitary ansatz:

\[ U(\theta_{\mathrm{uni}}) = \prod_j \exp \left( -i\theta_{\mathrm{uni}}H_j \right) \]

bilan ordinary VQA state:

\[ \rho_{\mathrm{uni}} (\theta_{\mathrm{uni}}) = U(\theta_{\mathrm{uni}}) \rho_0 U^\dagger(\theta_{\mathrm{uni}}). \tag{25} \]

Cost:

\[ C_{\mathrm{uni}} (\theta_{\mathrm{uni}}) = \operatorname{Tr} \left[ O \rho_{\mathrm{uni}} (\theta_{\mathrm{uni}}) \right] = \operatorname{Tr} \left[ O U(\theta_{\mathrm{uni}}) \rho_0 U^\dagger(\theta_{\mathrm{uni}}) \right]. \tag{26} \]

\(O\) — optimallashtiriladigan observable/cost operator. VQE kontekstida \(O\) Hamiltonian bo‘lishi mumkin.

Unital va non-unital channel farqi

Unital quantum channel identity operatorni identity sifatida saqlaydigan kanaldir. Non-unital channel buni qilishga majbur emas. Reset amali non-unital va tizimni muayyan sof holatga yo‘naltira olgani sabab entropy-ni tashqariga olib chiqadigan mexanizm bera oladi.

B ilova reset channelni:

\[ \mathcal{A}_{\mathrm{nr}}(\rho) = (1-q)\rho + q \operatorname{Tr}_R[\rho] \otimes |0\rangle\langle0|_R, \qquad q\in(0,1] \]

ko‘rinishida ta’riflaydi va asosan \(q=1\) holatini ko‘rib chiqadi. \(q=1\), tanlangan \(R\) subsystemning to‘liq \(|0\rangle\) holatiga reset qilinishiga mos keladi.

Dissipative VQA kanal kompozitsiyasi: Tenglama (27)–(30)

\[ \Phi_{\theta_{\mathrm{dis}}}^{M} := \prod_{k=1}^{L} \left( V^k \circ \mathcal{N} \circ U_{\theta_{\mathrm{dis}}}^{k} \circ \mathcal{A}_{\mathrm{nr}}^{\chi(k)} \right), \tag{27} \]

bu yerda:

\[ \chi(k) = \begin{cases} 1, & k\equiv0\pmod L,\\ 0, & \text{aksi halde}. \end{cases} \tag{28} \]

\(U_{\theta_{\mathrm{dis}}}^k\) — parametrli two-qubit layer; \(V^k\) — fixed single-qubit gates; \(\mathcal{N}\) — contractive unital single-qubit noise channels tensor producti; \(\mathcal{A}_{\mathrm{nr}}\) — non-unital reset map.

Dissipative VQA state:

\[ \rho_{\mathrm{dis}} (\theta_{\mathrm{dis}}) = \Phi_{\theta_{\mathrm{dis}}}^{M} (\rho_0), \tag{29} \]

va cost:

\[ C_{\mathrm{dis}} (\theta_{\mathrm{dis}}) = \operatorname{Tr} \left[ O \rho_{\mathrm{dis}} (\theta_{\mathrm{dis}}) \right] = \operatorname{Tr} \left[ O \Phi_{\theta_{\mathrm{dis}}}^{M} (\rho_0) \right]. \tag{30} \]

Manbaning nazariy bayonida ordinary unitary map bilan dissipative map o‘rtasidagi muhim farq non-unital reset-assisted o‘zgarish hisoblash davomida entropy-ni tizimdan uzoqlashtira olishi.

Gradient descent: Tenglama (31)–(32)

\[ \theta_{A,t+1} = \theta_{A,t} - \eta_A \nabla C_A (\theta_{A,t}), \qquad A\in\{\mathrm{uni},\mathrm{dis}\}. \tag{31} \]

\(\eta_A\) — learning rate / step size.

Bir iteratsiyadagi cost kamayishi va optimality gap:

\[ \Delta C_{A,t} := C_A(\theta_{A,t}) - C_A(\theta_{A,t+1}), \]
\[ \Delta_{A,t} := C_A(\theta_{A,t}) - C_A^\ast, \qquad C_A^\ast := \inf_{\theta} C_A(\theta). \tag{32} \]

\(\Delta C_{A,t}\) — single-step improvement; \(\Delta_{A,t}\) — joriy costning optimal costdan uzoqligi.

Faraz 1: Smoothness — Tenglama (33)

\[ \| \nabla C_A(\theta) - \nabla C_A(\phi) \|_2 \le \beta_A \| \theta-\phi \|_2. \tag{33} \]

\(\beta_A\)-smoothness gradient parametr fazosida cheksiz keskinlik bilan o‘zgarmaydi deb faraz qiladi. Gradient descent uchun:

\[ 0<\eta_A\le\frac{1}{\beta_A} \]

tanlanadi. Manba shuningdek \(\beta_A=O(\operatorname{poly}(n))\) farazidan foydalanadi.

Faraz 2: Unitary two-design

Initializationda parametrik unitaries har bir gate ustida unitary 2-design hosil qiladigan distributiondan keladi deb faraz qilinadi. Natijada boshlanishida:

\[ \mathbb{E} [ \partial_\mu C_A (\theta_{A,0}) ] = 0 \]

va shuning uchun:

\[ \mathbb{E} [ (\partial_\mu C_A)^2 ] = \operatorname{Var} [ \partial_\mu C_A ] \]

munosabati ishlatiladi.

Faraz 3: Dissipative VQA uchun PL sharti — Tenglama (34)

\[ \frac{1}{2} \| \nabla C_{\mathrm{dis}}(\theta) \|_2^2 \ge \mu_{\mathrm{dis}} \left( C_{\mathrm{dis}}(\theta) - C_{\mathrm{dis}}^\ast \right). \tag{34} \]

Polyak–Łojasiewicz (PL) sharti gradient normasini optimumga bo‘lgan cost gap bilan bog‘laydi. Katta gap turganda gradientning ixtiyoriy tarzda nolga tushib ketishining oldini oluvchi kuchli optimization-landscape farazidir. Theorem 1ning global contraction natijasi shu shartga bog‘liq; maqola bu barcha dissipative quantum objectives uchun avtomatik ravishda o‘rinli ekanini isbotlamaydi.

Faraz 4: Ordinary VQAning trajectory-uniform flatnessi — Tenglama (35)

\[ \mathbb{E} \left[ \| \nabla C_{\mathrm{uni}} (\theta_{\mathrm{uni},t}) \|_2^2 \right] \le G_{\mathrm{uni}}(n) := mB_{\mathrm{uni}}b^{-n}, \qquad b>1. \tag{35} \]

Bu yerdagi \(m\) — variational parameter soni; bu \(m\) asosiy matndagi ancilla-qubit belgisi bilan ayni harfning boshqa kontekstdagi ishlatilishidir. Shu sababli belgi ma’nosi bo‘lim kontekstiga ko‘ra o‘qilishi kerak.

\(b^{-n}\) omili gradient system size \(n\) bilan exponentially kichrayishi mumkinligini anglatadi. Faraz faqat initializationga emas, target gap \(\varepsilon\) ga yetgunga qadar barcha tegishli optimization trajectoryga qo‘llanadi; shuning uchun juda kuchli shartdir.

Lemma 1: Har bir step uchun minimal kamayish — Tenglama (36)–(37)

\[ f(x)-f(x^+) \ge \frac{\eta}{2} \| \nabla f(x) \|_2^2. \tag{36} \]

Bu natija \(\beta\)-smooth \(f\) va \(0<\eta\le1/\beta\) uchun o‘rinli.

Smoothnessdan:

\[ f(x^+) \le f(x) - \eta \| \nabla f(x) \|_2^2 + \frac{\beta\eta^2}{2} \| \nabla f(x) \|_2^2 \tag{37} \]

olinadi va step-size chegarasi ishlatilganda Eq. (36) kelib chiqadi.

Lemma 2: Har bir step uchun maksimal kamayish — Tenglama (38)–(40)

\[ f(x)-f(x^+) \le \frac{3\eta}{2} \| \nabla f(x) \|_2^2. \tag{38} \]

Manba buni Hessian normasi tegishli segment bo‘ylab \(\beta\) bilan cheklangan degan faraz ostida beradi.

\[ f(x^+) \ge f(x) - \eta \| \nabla f(x) \|_2^2 - \frac{\beta\eta^2}{2} \| \nabla f(x) \|_2^2. \tag{39} \]
\[ f(x)-f(x^+) \le \eta \left( 1+\frac{\beta\eta}{2} \right) \| \nabla f(x) \|_2^2 \le \frac{3\eta}{2} \| \nabla f(x) \|_2^2. \tag{40} \]

Lemma 3: PL contraction — Tenglama (41)–(44)

\[ \frac{1}{2} \| \nabla f(x) \|_2^2 \ge \mu \left( f(x)-f^\ast \right). \tag{41} \]

PL inequality va smoothness birgalikda gradient descent gapning exponential contractionini beradi:

\[ f(x_t)-f^\ast \le e^{-\eta\mu t} \left( f(x_0)-f^\ast \right). \tag{42} \]

Oraliq qadam:

\[ f(x_{t+1})-f^\ast \le f(x_t)-f^\ast - \frac{\eta}{2} \| \nabla f(x_t) \|_2^2. \tag{43} \]

PL inequality qo‘llanganda:

\[ f(x_{t+1})-f^\ast \le (1-\eta\mu) \left( f(x_t)-f^\ast \right) \le e^{-\eta\mu} \left( f(x_t)-f^\ast \right). \tag{44} \]

Lemma 4: Flat ordinary VQA uchun iteratsiyaning quyi chegarasi — Tenglama (45)–(47)

Boshlang‘ich expected gap:

\[ \mathbb{E} [ \Delta_{\mathrm{uni},0} ] \ge \delta_0 > 0 \]

va maqsad:

\[ \mathbb{E} [ \Delta_{\mathrm{uni},T} ] \le \varepsilon < \delta_0 \]

bo‘lsa:

\[ T \ge \frac{ 2(\delta_0-\varepsilon) }{ 3\eta_{\mathrm{uni}} G_{\mathrm{uni}}(n) } = \Omega \left( \frac{ b^n }{ \eta_{\mathrm{uni}}m } \right). \tag{45} \]

Har bir step uchun expected decrease:

\[ \mathbb{E} [ \Delta C_{\mathrm{uni},t} ] \le \frac{ 3\eta_{\mathrm{uni}} }{2} \mathbb{E} \left[ \| \nabla C_{\mathrm{uni}} (\theta_{\mathrm{uni},t}) \|_2^2 \right] \le \frac{ 3\eta_{\mathrm{uni}} }{2} G_{\mathrm{uni}}(n). \tag{46} \]

Iteratsiyalar jamlanganda:

\[ \mathbb{E} \left[ C_{\mathrm{uni}} (\theta_{\mathrm{uni},0}) - C_{\mathrm{uni}} (\theta_{\mathrm{uni},T}) \right] \le \frac{ 3\eta_{\mathrm{uni}} }{2} T G_{\mathrm{uni}}(n). \tag{47} \]

Gradient energy exponentially kichik bo‘lib qolsa, bitta iteratsiyada olinadigan expected cost improvement ham exponentially kichik bo‘ladi; shuning uchun doimiy optimality-gap kamayishi ko‘p iteration talab qiladi.

Theorem 1: Dissipative VQAning shartli global convergencei — Tenglama (48)–(52)

Dissipative cost:

\[ C_{\mathrm{dis}}(\theta) = \operatorname{Tr} \left[ O \Phi_\theta^M (\rho_0) \right]. \tag{48} \]

Assumption 1 va PL Assumption 3 ostida \(\eta_{\mathrm{dis}}=1/\beta_{\mathrm{dis}}\) tanlansa:

\[ \Delta_{\mathrm{dis},t} \le e^{ -\mu_{\mathrm{dis}}t/ \beta_{\mathrm{dis}} } \Delta_{\mathrm{dis},0}. \tag{49} \]

\(\Delta_{\mathrm{dis},T}\le\varepsilon\) uchun kerakli iteration soni:

\[ T_{\mathrm{dis}} (\varepsilon) \le \frac{ \beta_{\mathrm{dis}} }{ \mu_{\mathrm{dis}} } \log \left( \frac{ \Delta_{\mathrm{dis},0} }{ \varepsilon } \right). \tag{50} \]

Agar:

\[ \frac{ \beta_{\mathrm{dis}} }{ \mu_{\mathrm{dis}} } \le \operatorname{poly}(n) \]

va logarithmic accuracy factor ham polinom chegarada bo‘lsa:

\[ T_{\mathrm{dis}} (\varepsilon) \le \operatorname{poly}(n). \tag{51} \]

Eq. (50),

\[ e^{ -\mu_{\mathrm{dis}}T/ \beta_{\mathrm{dis}} } \Delta_{\mathrm{dis},0} \le \varepsilon \tag{52} \]

tengsizlikni \(T\) uchun yechishdan olinadi.

Theorem 2: Shartli iteration-complexity separation — Tenglama (53)–(56)

Dissipative tomonda:

\[ \frac{ \beta_{\mathrm{dis}} }{ \mu_{\mathrm{dis}} } \le \operatorname{poly}(n) \tag{53} \]

va ordinary VQA tomonida:

\[ \eta_{\mathrm{uni}}m \le \operatorname{poly}(n), \qquad \mathbb{E} [ \Delta_{\mathrm{uni},0} ] \ge \delta_0 > 0 \tag{54} \]

deb faraz qilinsa, dissipative VQA uchun:

\[ T_{\mathrm{dis}} (\varepsilon) \le \operatorname{poly}(n) \log \left( \frac{ \Delta_{\mathrm{dis},0} }{ \varepsilon } \right), \tag{55} \]

ordinary trajectory-flat VQA uchun esa:

\[ T_{\mathrm{uni}} (\varepsilon) \ge \Omega \left( \frac{ b^n }{ \operatorname{poly}(n) } \right). \tag{56} \]

olinadi.

Bu mualliflarning “conditional exponential advantage in gradient-descent iteration complexity” natijasidir. Bu yerdagi “advantage”ning uch muhim chegarasi bor:

  • haqiqiy quantum-device runtime o‘lchovi emas;
  • dissipative objective PL conditionni bajarishi shart;
  • ordinary VQA maqsadga yetguncha trajectory-uniform exponentially flat bo‘lib qolishi shart.

Manba buni alohida ravishda ochiq cheklaydi: avvalgi barren-plateau-avoidance natijasi o‘z-o‘zicha dissipative VQA uchun global convergence isbotlamaydi. Polinomik global iteration bound faqat qo‘shimcha PL va scaling farazlari bilan yuzaga keladi.

Tadqiqot bevosita qo‘llab-quvvatlaydigan natijalar

Ancilla-assisted trainable dissipation standart system-only unitary ansatzga qo‘shimcha quantum-channel degree of freedom berishi matematik jihatdan quriladi.

Ko‘rilgan numerical VQE benchmarklarida dissipative VQE ancilla-free baselinega qaraganda tezroq energy descent va pastroq final energy ko‘rsatadi.

Bu nisbiy afzallik depolarizing, amplitude-damping va bit-flip noise sharoitlarida hamda uch xil representative Hamiltonianda saqlanadi.

State recoveryda ayni dissipative block turli preparation-noise kanallari va turli target-state structures ostida final fidelityni oshiradi.

Ko‘rilgan target-state taqqoslashida W, uniform-superposition va dressed-cluster state uchun final fidelity 100 optimization iteration ichida %99dan yuqoriga chiqarilgan.

Ancilla va dissipative-round resurslarida task-dependent saturation mavjudligi ko‘rsatiladi; ko‘proq resource har doim mutanosib qo‘shimcha foyda bermaydi.

B ilova muayyan mathematical assumptions ostida dissipative va ordinary VQA o‘rtasida gradient-descent iteration complexity nuqtai nazaridan shartli eksponensial ajralish hosil qiladi.

Tadqiqot qo‘llab-quvvatlamaydigan natijalar

Tadqiqot haqiqiy kvant apparatida speedup, quantum advantage yoki fault-tolerant operation ko‘rsatmaydi.

Tadqiqot dissipatsiya fizik noiseni butunlay yo‘qotishini ko‘rsatmaydi.

Tadqiqot har bir Hamiltonian va har bir system size uchun dissipative VQE ordinary VQEdan ustun ekanini isbotlamaydi.

Tadqiqot ancilla soni cheksiz oshgani sari unumdorlik cheksiz oshishini ko‘rsatmaydi; aksincha saturation kuzatadi.

Tadqiqot B ilovadagi PL sharti taklif qilingan barcha dissipative circuits uchun avtomatik ravishda bajarilishini isbotlamaydi.

Tadqiqot ordinary VQA barcha haqiqiy optimallashtirish muammolarida trajectory-uniform exponential flatness ko‘rsatishini isbotlamaydi.

Theorem 2dagi exponential iteration-complexity separation bevosita end-to-end computational speedup yoki hardware execution-time advantage emas.

Asosiy uslubiy cheklovlar

Faqat simulyatsiya: Fizik kvant protsessorida hardware validation yo‘q.

Kichik tizimlar: Ko‘rsatilgan asosiy numerical examples bir necha kubitli spin tizimlari va target-state recovery muammolaridir. Kattaroq system-size scaling kelajakdagi ish mavzusi sifatida qoldirilgan.

Optimizer tafsiloti: Asosiy numerical section learning rate qiymatlarini beradi, biroq ishlatilgan software/optimizer implementation tafsilotlari to‘liq xabar qilinmaydi.

Seed/statistical protocol: Asosiy grafiklarning hammasi uchun mustaqil repetition va uncertainty protocol aniq berilmaydi. Qo‘shimcha 11-rasm “best 5 seeds” iborasini ishlatadi, biroq umumiy seed poolni batafsil bayon qilmaydi.

Resource overhead: Har bir dissipative round fully connected system–ancilla architectureda \(nm\) iSWAP coupling talab qiladi. Ko‘proq ancilla va round circuit depth va trainable parameter yukini oshiradi.

Theoretical assumptions: B ilovaning kuchli complexity ajralishi PL va trajectory-flatness farazlariga bog‘liq.

Noise realism: Ishlatilgan AD, BF va DP modellari nazorat qilinadigan numerical noise modellaridir. Haqiqiy qurilmadagi coherent calibration errors, crosstalk, leakage, non-Markovian effects va device-specific correlations alohida tasdiqlanmagan.

Manba va Usul Haqida Izoh

Asl sarlavha: A Variational Dissipative Framework for Quantum Algorithms

Mualliflar: Yuan Yao, Ruipeng Xing, Yongjian Gu, Yiming Huang, Xiao Yuan.

Teng hissa: Taqdim etilgan manbada teng hissa/teng birinchi muallif belgisi yo‘q.

Muassasa 1: Center on Frontiers of Computing Studies, Peking University, Beijing 100871, China.

Muassasa 2: School of Computer Science, Peking University, Beijing 100871, China.

Muassasa 3: Faculty of Information Science and Engineering, Ocean University of China, Qingdao 266100, China.

Muassasa 4: Institute of High Energy Physics, Chinese Academy of Sciences, Beijing 100049, China.

Muassasa 5: China Center of Advanced Science and Technology, Beijing 100190, China.

Corresponding author: PDF bitta corresponding authorni aniq belgi bilan ajratmaydi. Manba Yuan Yao, Yongjian Gu, Yiming Huang va Xiao Yuan uchun aloqa e-pochta manzillarini beradi.

Manbada ko‘rinadigan aloqa izohi: Yongjian Guning e-pochta manzili PDFda `yjgu@.ouc.edu.cn` ko‘rinishida turadi. Bu ko‘rinadigan shakl ehtimoliy yozuv/formatlash muammosini o‘z ichiga olgani sabab Verianla tomonidan jim ravishda tuzatilmagan.

Manba turi: Preprint.

Platforma: arXiv.

Soha: quant-ph — Quantum Physics.

arXiv identifikatori: arXiv:2605.25841v1.

Dastlabki yuborish sanasi: 25 may 2026.

PDFda berilgan sana: 25 may 2026.

DOI: 10.48550/arXiv.2605.25841.

Taqriz holati: 13 noyabr 2026 sanadagi tekshiruvda alohida taqrizlangan journal version of record tasdiqlanmagan; tadqiqot Verianlada preprint sifatida ko‘rilishi kerak.

Jurnal / jild / son: Tasdiqlangan journal publication mavjud bo‘lmagani sabab yo‘q.

Rasmiy havola: https://arxiv.org/abs/2605.25841

Litsenziya: Yuklangan PDFda ochiq Creative Commons qayta foydalanish litsenziyasi yo‘q. arXivdagi mavjudlik manba figuralaridan cheksiz qayta foydalanishga ruxsat sifatida baholanmagan.

Ilmiy tadqiqot turi: Nazariy ochiq-tizim kvant axborot doirasi + variational quantum algorithm dizayni + numerical simulation + shartli optimization-complexity tahlili.

Haqiqiy quantum hardware: Ishlatilmagan / xabar qilinmagan.

Asosiy usul: System-only parametric variational layers bilan ancilla-assisted trainable CPTP dissipative channelsni birlashtirish.

Dissipative implementation: Parametric single-qubit rotations + fully connected system–ancilla \(i\mathrm{SWAP}\) couplings; har bir dissipative round uchun \(nm\) system–ancilla iSWAP interaction.

Ancilla amali: Har bir round oxirida partial trace va reset.

Vazifa 1: Dissipative VQE bilan ground-state search; objective final energy expectation value.

Vazifa 2: Dissipative state recovery; objective target-state fidelity.

VQE benchmark reference: Exact ground-state energies, boshqacha ko‘rsatilmagan bo‘lsa exact diagonalization bilan hisoblangan.

Noise modellari: Depolarizing, Bit-Flip va Amplitude Damping.

Target states: W state, uniform superposition state va dressed cluster state.

Muhim numerical natija: Ko‘rilgan target-state experimentlarda \(n=3\), \(m=3\), \(T=3\), \(lr=0.8\) va \(p=0.1\) ostida uch target uchun final fidelity 100 iteratsiya ichida %99dan yuqoriga chiqqan.

Resource natijasi: Moderate dissipative depth ko‘pchilik testlar uchun yetarli; manba \(T=3\) ni umumiy muvozanat nuqtasi sifatida ishlatadi va state-dependent ancilla saturation xabar qiladi.

Approximate ancilla saturation: W state \(m=2\), uniform superposition \(m=5\), dressed cluster \(m=3\).

B ilova natijasi: PL va trajectory-uniform flatnessni o‘z ichiga olgan qo‘shimcha farazlar ostida dissipative VQA uchun polinomik, ordinary VQA uchun eksponensial gradient-descent iteration-complexity xatti-harakati chiqariladi. Bu natija shartli va hardware speedup emas.

Moliyalashtirish: Quantum Science and Technology-National Science and Technology Major Project (2023ZD0300200); Beijing Natural Science Foundation Z250004; NSAF U2330201; National Natural Science Foundation of China 12361161602; Beijing Science and Technology Planning Project Z25110100810000; High-performance Computing Platform of Peking University.

Ma’lumotlarning mavjudligi: Taqdim etilgan PDFda alohida Data Availability statement yo‘q.

Code availability: Taqdim etilgan PDFda alohida Code Availability statement yo‘q.

Manfaatlar to‘qnashuvi: Taqdim etilgan PDFda alohida competing-interest/conflict-of-interest statement yo‘q; shu sababli manfaatlar to‘qnashuvi yo‘q deb faraz qilinmagan.

Mualliflar hissasi: Taqdim etilgan PDFda alohida CRediT yoki author-contribution statement yo‘q.

Reproducibility chegarasi: Manba muhim fizik va hyperparameter sozlamalarini xabar qilsa-da, asosiy numerical implementation uchun software framework, optimizer tafsilotlari va barcha seed/repetition protocol to‘liq berilmagan.

Manba ichidagi notatsiya izohi: Bit-Flip noise ko‘pchilik joyda BF deb ta’riflansa-da, bir jumlada BP qisqartmasi ko‘rinadi; bu yangi noise modeli sifatida talqin qilinmagan.


Ulashish:

Izohlar ko‘rib chiqilgandan keyin e’lon qilinadi. Izohingiz tasdiqlash jarayoniga yuboriladi va ma’qullangach ko‘rinadi.

Izoh qoldiring

E-pochta manzilingiz chop etilmaydi. Majburiy maydonlar * bilan belgilangan