{
  "model": "J1J2",
  "lattice": "square",
  "n_sites": 100,
  "boundary": "O",
  "params": {
    "J2": 0.2
  },
  "cluster": "10 x 10 square lattice, open boundaries in both directions: nearest- and next-nearest-neighbour bonds only inside the lattice (arXiv:2304.01996 Sec. 6.2: 'the challenging 2D J1-J2 Heisenberg model with open boundary condition', the setting of every energy table of the paper, Tables 1-4, and of its 4 x 4 to 12 x 12 transfer learning, Appendix D)",
  "instance_id": "J1J2/square_100_O_0.2",
  "rows": [
    {
      "energy": -199.7537,
      "sigma": null,
      "energy_variance": null,
      "dof": 100,
      "einf": 0,
      "v_score": null,
      "method": "DMRG",
      "ansatz": null,
      "method_detail": "",
      "method_as_published": "MPS (8)",
      "family": "tensor network",
      "bound_type": "variational",
      "bound_type_reason": "variational ansatz at a stated size; energy is an upper bound (assigned during source reading)",
      "reference": "Chen, Newhouse, Chen, Luo & Soljačić, ANTN: Bridging Autoregressive Neural Networks and Tensor Networks for Quantum Many-Body Simulation, Adv. Neural Inf. Process. Syst. 36 (NeurIPS 2023), 450-476, doi:10.52202/075280-0022, arXiv:2304.01996",
      "peer_reviewed": true,
      "source": "sweep-allresults-2026-11-11",
      "provenance": "primary",
      "verified": {
        "checked_on": "2026-11-11",
        "method": "source text of arXiv:2304.01996 read locally (arXiv HTML or pypdf layout text); value copied from the harvested cell and the text, no LLM transcription of numbers; single reading (2026-11-11)",
        "reported_as": "-1.997537",
        "note": "Table 2 (10 x 10), row 'MPS (8)', column J2 = 0.2; sources/2304.01996-neurips.txt line 423; MPS optimised by DMRG at this bond dimension (Table 2 caption), deterministic: no error bar; version read: the version of record, Adv. Neural Inf. Process. Syst. 36 (NeurIPS 2023), publisher PDF from proceedings.neurips.cc (text in sources/2304.01996-neurips.txt), and arXiv v3 (HTML, sources/2304.01996.html); the two print the same value in every energy cell of Tables 1-4. The older committed PDF text (sources/2304.01996.txt) interleaves Tables 1 and 2 and reads the 10 x 10 MPS (8) cell at J2 = 0.7 as -1.7682258; the PDF page and both versions above print -1.768225 | Eq. (8): H = J1 sum_<i,j> σ_i.σ_j + J2 sum_<<i,j>> σ_i.σ_j with Pauli matrices (Eq. 9), J1 = 1; the tables print 'Energy per site' of this H, the Pauli form, so E/N in the S.S convention is the printed value / 4 (RULES.md 5). Open boundaries (Sec. 6.2). Plausibility: the 10 x 10 cells at J2 = 0.5 (best -1.94030(9), -0.48508 in S.S units) lie above the periodic 10 x 10 records (about -0.4976), as an open lattice with fewer bonds per site must.",
        "secondary_of": null,
        "second_read": {
          "checked_on": "2026-11-11",
          "scope": "value",
          "method": "NeurIPS 2023 version of record (proceedings.neurips.cc file 01772a8b0420baec00c4d59fe2fbace6-Paper-Conference.pdf), pypdf plain and layout mode (sources/2304.01996-neurips.rp1-plain.txt); cross-checked against arXiv:2304.01996v3 PDF (sources/2304.01996v3.rp1-plain.txt) and the v3 HTML: all energy cells identical in both versions",
          "location": "Table 2 (10 x 10) line 429; Table 3 (10 x 10, Appendix C) line 1243, column J2 = 0.2",
          "reported_as": "-1.997537 (Energy per site, Pauli sigma.sigma form of Eq. (8), J1 = 1)",
          "conversion": "-1.997537 * 100 = -199.753700 (Pauli per site x N = stored Pauli total; S.S per site = -0.4993842)",
          "note": "L = 10 open square lattice (Sec. 6.2 'open boundary condition', the setting of all VMC tables), J1 = 1, J2 = 0.2 (column header), Pauli Eq. (8)-(9), 'Energy per site' column; bound_type variational (Sec. 6.2: 'based on the variational principle ... provides an upper bound'); MPS optimised with DMRG at the stated bond dimension (Table 2 caption); no error bar printed; no duplicate: data/J1J2/square_100_O_0.2.json holds only rows of this paper and batch",
          "source_entry": "RP1-arxiv-2304-01996#0 (qmbl-verify 2026-11-11, instances)"
        }
      },
      "compute": {
        "parameters": null,
        "gpu_hours": null,
        "device": null,
        "n_devices": null,
        "samples": null,
        "wall_clock": null,
        "cpu_core_hours": null,
        "bond_dimension": 8,
        "iterations": null,
        "reported_as": "method string: \"MPS (8)\"",
        "source": "the row's published method string (scripts/method_names.mjs, 2026-09-17)",
        "scope": "row",
        "confidence": "high",
        "note": "Only the numbers the method string states; nothing else about this run's cost is recorded."
      }
    },
    {
      "energy": -219.1048,
      "sigma": null,
      "energy_variance": null,
      "dof": 100,
      "einf": 0,
      "v_score": null,
      "method": "DMRG",
      "ansatz": null,
      "method_detail": "",
      "method_as_published": "MPS (70)",
      "family": "tensor network",
      "bound_type": "variational",
      "bound_type_reason": "variational ansatz at a stated size; energy is an upper bound (assigned during source reading)",
      "reference": "Chen, Newhouse, Chen, Luo & Soljačić, ANTN: Bridging Autoregressive Neural Networks and Tensor Networks for Quantum Many-Body Simulation, Adv. Neural Inf. Process. Syst. 36 (NeurIPS 2023), 450-476, doi:10.52202/075280-0022, arXiv:2304.01996",
      "peer_reviewed": true,
      "source": "sweep-allresults-2026-11-11",
      "provenance": "primary",
      "verified": {
        "checked_on": "2026-11-11",
        "method": "source text of arXiv:2304.01996 read locally (arXiv HTML or pypdf layout text); value copied from the harvested cell and the text, no LLM transcription of numbers; single reading (2026-11-11)",
        "reported_as": "-2.191048",
        "note": "Table 2 (10 x 10), row 'MPS (70)', column J2 = 0.2; sources/2304.01996-neurips.txt line 424; MPS optimised by DMRG at this bond dimension (Table 2 caption), deterministic: no error bar; version read: the version of record, Adv. Neural Inf. Process. Syst. 36 (NeurIPS 2023), publisher PDF from proceedings.neurips.cc (text in sources/2304.01996-neurips.txt), and arXiv v3 (HTML, sources/2304.01996.html); the two print the same value in every energy cell of Tables 1-4. The older committed PDF text (sources/2304.01996.txt) interleaves Tables 1 and 2 and reads the 10 x 10 MPS (8) cell at J2 = 0.7 as -1.7682258; the PDF page and both versions above print -1.768225 | Eq. (8): H = J1 sum_<i,j> σ_i.σ_j + J2 sum_<<i,j>> σ_i.σ_j with Pauli matrices (Eq. 9), J1 = 1; the tables print 'Energy per site' of this H, the Pauli form, so E/N in the S.S convention is the printed value / 4 (RULES.md 5). Open boundaries (Sec. 6.2). Plausibility: the 10 x 10 cells at J2 = 0.5 (best -1.94030(9), -0.48508 in S.S units) lie above the periodic 10 x 10 records (about -0.4976), as an open lattice with fewer bonds per site must.",
        "secondary_of": null,
        "second_read": {
          "checked_on": "2026-11-11",
          "scope": "value",
          "method": "NeurIPS 2023 version of record (proceedings.neurips.cc file 01772a8b0420baec00c4d59fe2fbace6-Paper-Conference.pdf), pypdf plain and layout mode (sources/2304.01996-neurips.rp1-plain.txt); cross-checked against arXiv:2304.01996v3 PDF (sources/2304.01996v3.rp1-plain.txt) and the v3 HTML: all energy cells identical in both versions",
          "location": "Table 2 (10 x 10) line 430; Table 3 (10 x 10, Appendix C) line 1244, column J2 = 0.2",
          "reported_as": "-2.191048 (Energy per site, Pauli sigma.sigma form of Eq. (8), J1 = 1)",
          "conversion": "-2.191048 * 100 = -219.104800 (Pauli per site x N = stored Pauli total; S.S per site = -0.5477620)",
          "note": "L = 10 open square lattice (Sec. 6.2 'open boundary condition', the setting of all VMC tables), J1 = 1, J2 = 0.2 (column header), Pauli Eq. (8)-(9), 'Energy per site' column; bound_type variational (Sec. 6.2: 'based on the variational principle ... provides an upper bound'); MPS optimised with DMRG at the stated bond dimension (Table 2 caption); no error bar printed; no duplicate: data/J1J2/square_100_O_0.2.json holds only rows of this paper and batch",
          "source_entry": "RP1-arxiv-2304-01996#1 (qmbl-verify 2026-11-11, instances)"
        }
      },
      "compute": {
        "parameters": null,
        "gpu_hours": null,
        "device": null,
        "n_devices": null,
        "samples": null,
        "wall_clock": null,
        "cpu_core_hours": null,
        "bond_dimension": 70,
        "iterations": null,
        "reported_as": "method string: \"MPS (70)\"",
        "source": "the row's published method string (scripts/method_names.mjs, 2026-09-17)",
        "scope": "row",
        "confidence": "high",
        "note": "Only the numbers the method string states; nothing else about this run's cost is recorded."
      }
    },
    {
      "energy": -225.5633,
      "sigma": null,
      "energy_variance": null,
      "dof": 100,
      "einf": 0,
      "v_score": null,
      "method": "DMRG",
      "ansatz": null,
      "method_detail": "",
      "method_as_published": "MPS (1024)",
      "family": "tensor network",
      "bound_type": "variational",
      "bound_type_reason": "variational ansatz at a stated size; energy is an upper bound (assigned during source reading)",
      "reference": "Chen, Newhouse, Chen, Luo & Soljačić, ANTN: Bridging Autoregressive Neural Networks and Tensor Networks for Quantum Many-Body Simulation, Adv. Neural Inf. Process. Syst. 36 (NeurIPS 2023), 450-476, doi:10.52202/075280-0022, arXiv:2304.01996",
      "peer_reviewed": true,
      "source": "sweep-allresults-2026-11-11",
      "provenance": "primary",
      "verified": {
        "checked_on": "2026-11-11",
        "method": "source text of arXiv:2304.01996 read locally (arXiv HTML or pypdf layout text); value copied from the harvested cell and the text, no LLM transcription of numbers; single reading (2026-11-11)",
        "reported_as": "-2.255633",
        "note": "Table 2 (10 x 10), row 'MPS (1024)', column J2 = 0.2; sources/2304.01996-neurips.txt line 425; MPS optimised by DMRG at this bond dimension (Table 2 caption), deterministic: no error bar; version read: the version of record, Adv. Neural Inf. Process. Syst. 36 (NeurIPS 2023), publisher PDF from proceedings.neurips.cc (text in sources/2304.01996-neurips.txt), and arXiv v3 (HTML, sources/2304.01996.html); the two print the same value in every energy cell of Tables 1-4. The older committed PDF text (sources/2304.01996.txt) interleaves Tables 1 and 2 and reads the 10 x 10 MPS (8) cell at J2 = 0.7 as -1.7682258; the PDF page and both versions above print -1.768225 | Eq. (8): H = J1 sum_<i,j> σ_i.σ_j + J2 sum_<<i,j>> σ_i.σ_j with Pauli matrices (Eq. 9), J1 = 1; the tables print 'Energy per site' of this H, the Pauli form, so E/N in the S.S convention is the printed value / 4 (RULES.md 5). Open boundaries (Sec. 6.2). Plausibility: the 10 x 10 cells at J2 = 0.5 (best -1.94030(9), -0.48508 in S.S units) lie above the periodic 10 x 10 records (about -0.4976), as an open lattice with fewer bonds per site must.",
        "secondary_of": null,
        "second_read": {
          "checked_on": "2026-11-11",
          "scope": "value",
          "method": "NeurIPS 2023 version of record (proceedings.neurips.cc file 01772a8b0420baec00c4d59fe2fbace6-Paper-Conference.pdf), pypdf plain and layout mode (sources/2304.01996-neurips.rp1-plain.txt); cross-checked against arXiv:2304.01996v3 PDF (sources/2304.01996v3.rp1-plain.txt) and the v3 HTML: all energy cells identical in both versions",
          "location": "Table 2 (10 x 10) line 431; Table 3 (10 x 10, Appendix C) line 1245, column J2 = 0.2",
          "reported_as": "-2.255633 (Energy per site, Pauli sigma.sigma form of Eq. (8), J1 = 1)",
          "conversion": "-2.255633 * 100 = -225.563300 (Pauli per site x N = stored Pauli total; S.S per site = -0.5639082)",
          "note": "L = 10 open square lattice (Sec. 6.2 'open boundary condition', the setting of all VMC tables), J1 = 1, J2 = 0.2 (column header), Pauli Eq. (8)-(9), 'Energy per site' column; bound_type variational (Sec. 6.2: 'based on the variational principle ... provides an upper bound'); MPS optimised with DMRG at the stated bond dimension (Table 2 caption); no error bar printed; no duplicate: data/J1J2/square_100_O_0.2.json holds only rows of this paper and batch",
          "source_entry": "RP1-arxiv-2304-01996#2 (qmbl-verify 2026-11-11, instances)"
        }
      },
      "compute": {
        "parameters": null,
        "gpu_hours": null,
        "device": null,
        "n_devices": null,
        "samples": null,
        "wall_clock": null,
        "cpu_core_hours": null,
        "bond_dimension": 1024,
        "iterations": null,
        "reported_as": "method string: \"MPS (1024)\"",
        "source": "the row's published method string (scripts/method_names.mjs, 2026-09-17)",
        "scope": "row",
        "confidence": "high",
        "note": "Only the numbers the method string states; nothing else about this run's cost is recorded."
      }
    },
    {
      "energy": -222.462,
      "sigma": 0.024,
      "energy_variance": null,
      "dof": 100,
      "einf": 0,
      "v_score": null,
      "method": "VMC",
      "ansatz": "PixelCNN",
      "method_detail": "7 layers, 48 hidden dimensions, no sign rule",
      "method_as_published": "PixelCNN (7-48)",
      "family": "CNN / ResNet",
      "bound_type": "variational",
      "bound_type_reason": "variational ansatz at a stated size; energy is an upper bound (assigned during source reading)",
      "reference": "Chen, Newhouse, Chen, Luo & Soljačić, ANTN: Bridging Autoregressive Neural Networks and Tensor Networks for Quantum Many-Body Simulation, Adv. Neural Inf. Process. Syst. 36 (NeurIPS 2023), 450-476, doi:10.52202/075280-0022, arXiv:2304.01996",
      "peer_reviewed": true,
      "source": "sweep-allresults-2026-11-11",
      "provenance": "primary",
      "verified": {
        "checked_on": "2026-11-11",
        "method": "source text of arXiv:2304.01996 read locally (arXiv HTML or pypdf layout text); value copied from the harvested cell and the text, no LLM transcription of numbers; single reading (2026-11-11)",
        "reported_as": "-2.22462(24)",
        "note": "Table 2 (10 x 10), row 'PixelCNN' (PixelCNN (7-48) in Tables 3-5), column J2 = 0.2; sources/2304.01996-neurips.txt line 426; version read: the version of record, Adv. Neural Inf. Process. Syst. 36 (NeurIPS 2023), publisher PDF from proceedings.neurips.cc (text in sources/2304.01996-neurips.txt), and arXiv v3 (HTML, sources/2304.01996.html); the two print the same value in every energy cell of Tables 1-4. The older committed PDF text (sources/2304.01996.txt) interleaves Tables 1 and 2 and reads the 10 x 10 MPS (8) cell at J2 = 0.7 as -1.7682258; the PDF page and both versions above print -1.768225 | Eq. (8): H = J1 sum_<i,j> σ_i.σ_j + J2 sum_<<i,j>> σ_i.σ_j with Pauli matrices (Eq. 9), J1 = 1; the tables print 'Energy per site' of this H, the Pauli form, so E/N in the S.S convention is the printed value / 4 (RULES.md 5). Open boundaries (Sec. 6.2). Plausibility: the 10 x 10 cells at J2 = 0.5 (best -1.94030(9), -0.48508 in S.S units) lie above the periodic 10 x 10 records (about -0.4976), as an open lattice with fewer bonds per site must.",
        "secondary_of": null,
        "second_read": {
          "checked_on": "2026-11-11",
          "scope": "value",
          "method": "NeurIPS 2023 version of record (proceedings.neurips.cc file 01772a8b0420baec00c4d59fe2fbace6-Paper-Conference.pdf), pypdf plain and layout mode (sources/2304.01996-neurips.rp1-plain.txt); cross-checked against arXiv:2304.01996v3 PDF (sources/2304.01996v3.rp1-plain.txt) and the v3 HTML: all energy cells identical in both versions",
          "location": "Table 2 (10 x 10) line 432; Table 3 (10 x 10, Appendix C) line 1246, column J2 = 0.2",
          "reported_as": "-2.22462(24) (Energy per site, Pauli sigma.sigma form of Eq. (8), J1 = 1)",
          "conversion": "-2.22462 * 100 = -222.462000 (Pauli per site x N = stored Pauli total; S.S per site = -0.5561550); sigma 24e-5 * 100 = 0.024",
          "note": "L = 10 open square lattice (Sec. 6.2 'open boundary condition', the setting of all VMC tables), J1 = 1, J2 = 0.2 (column header), Pauli Eq. (8)-(9), 'Energy per site' column; bound_type variational (Sec. 6.2: 'based on the variational principle ... provides an upper bound'); no duplicate: data/J1J2/square_100_O_0.2.json holds only rows of this paper and batch",
          "source_entry": "RP1-arxiv-2304-01996#3 (qmbl-verify 2026-11-11, instances)"
        }
      }
    },
    {
      "energy": -226.034,
      "sigma": 0.006,
      "energy_variance": null,
      "dof": 100,
      "einf": 0,
      "v_score": null,
      "method": "VMC",
      "ansatz": "ANTN",
      "method_detail": "elementwise, PixelCNN 7 layers, 48 hidden dimensions, no sign rule",
      "method_as_published": "EW (7-48-8)",
      "family": "tensor network",
      "bound_type": "variational",
      "bound_type_reason": "variational ansatz at a stated size; energy is an upper bound (assigned during source reading)",
      "reference": "Chen, Newhouse, Chen, Luo & Soljačić, ANTN: Bridging Autoregressive Neural Networks and Tensor Networks for Quantum Many-Body Simulation, Adv. Neural Inf. Process. Syst. 36 (NeurIPS 2023), 450-476, doi:10.52202/075280-0022, arXiv:2304.01996",
      "peer_reviewed": true,
      "source": "sweep-allresults-2026-11-11",
      "provenance": "primary",
      "verified": {
        "checked_on": "2026-11-11",
        "method": "source text of arXiv:2304.01996 read locally (arXiv HTML or pypdf layout text); value copied from the harvested cell and the text, no LLM transcription of numbers; single reading (2026-11-11)",
        "reported_as": "-2.26034(6)",
        "note": "Table 2 (10 x 10), row 'Elementwise (8)' (EW (7-48-8) in Tables 3-5), column J2 = 0.2; sources/2304.01996-neurips.txt line 427; version read: the version of record, Adv. Neural Inf. Process. Syst. 36 (NeurIPS 2023), publisher PDF from proceedings.neurips.cc (text in sources/2304.01996-neurips.txt), and arXiv v3 (HTML, sources/2304.01996.html); the two print the same value in every energy cell of Tables 1-4. The older committed PDF text (sources/2304.01996.txt) interleaves Tables 1 and 2 and reads the 10 x 10 MPS (8) cell at J2 = 0.7 as -1.7682258; the PDF page and both versions above print -1.768225 | Eq. (8): H = J1 sum_<i,j> σ_i.σ_j + J2 sum_<<i,j>> σ_i.σ_j with Pauli matrices (Eq. 9), J1 = 1; the tables print 'Energy per site' of this H, the Pauli form, so E/N in the S.S convention is the printed value / 4 (RULES.md 5). Open boundaries (Sec. 6.2). Plausibility: the 10 x 10 cells at J2 = 0.5 (best -1.94030(9), -0.48508 in S.S units) lie above the periodic 10 x 10 records (about -0.4976), as an open lattice with fewer bonds per site must.",
        "secondary_of": null,
        "second_read": {
          "checked_on": "2026-11-11",
          "scope": "value",
          "method": "NeurIPS 2023 version of record (proceedings.neurips.cc file 01772a8b0420baec00c4d59fe2fbace6-Paper-Conference.pdf), pypdf plain and layout mode (sources/2304.01996-neurips.rp1-plain.txt); cross-checked against arXiv:2304.01996v3 PDF (sources/2304.01996v3.rp1-plain.txt) and the v3 HTML: all energy cells identical in both versions",
          "location": "Table 2 (10 x 10) line 433; Table 3 (10 x 10, Appendix C) line 1249, column J2 = 0.2",
          "reported_as": "-2.26034(6) (Energy per site, Pauli sigma.sigma form of Eq. (8), J1 = 1)",
          "conversion": "-2.26034 * 100 = -226.034000 (Pauli per site x N = stored Pauli total; S.S per site = -0.5650850); sigma 6e-5 * 100 = 0.006",
          "note": "L = 10 open square lattice (Sec. 6.2 'open boundary condition', the setting of all VMC tables), J1 = 1, J2 = 0.2 (column header), Pauli Eq. (8)-(9), 'Energy per site' column; bound_type variational (Sec. 6.2: 'based on the variational principle ... provides an upper bound'); record row: lowest energy printed for this instance in the paper; no other row on the instance (data file holds only this paper's rows); no duplicate: data/J1J2/square_100_O_0.2.json holds only rows of this paper and batch",
          "source_entry": "RP1-arxiv-2304-01996#4 (qmbl-verify 2026-11-11, instances)"
        }
      },
      "compute": {
        "parameters": null,
        "gpu_hours": null,
        "device": null,
        "n_devices": null,
        "samples": null,
        "wall_clock": null,
        "cpu_core_hours": null,
        "bond_dimension": 8,
        "iterations": null,
        "reported_as": "method string: \"EW (7-48-8)\"",
        "source": "the row's published method string (scripts/method_names.mjs, 2026-09-17)",
        "scope": "row",
        "confidence": "high",
        "note": "Only the numbers the method string states; nothing else about this run's cost is recorded."
      }
    },
    {
      "energy": -225.755,
      "sigma": 0.008,
      "energy_variance": null,
      "dof": 100,
      "einf": 0,
      "v_score": null,
      "method": "VMC",
      "ansatz": "ANTN",
      "method_detail": "blockwise, PixelCNN 7 layers, 48 hidden dimensions, no sign rule",
      "method_as_published": "BW (7-48-70)",
      "family": "tensor network",
      "bound_type": "variational",
      "bound_type_reason": "variational ansatz at a stated size; energy is an upper bound (assigned during source reading)",
      "reference": "Chen, Newhouse, Chen, Luo & Soljačić, ANTN: Bridging Autoregressive Neural Networks and Tensor Networks for Quantum Many-Body Simulation, Adv. Neural Inf. Process. Syst. 36 (NeurIPS 2023), 450-476, doi:10.52202/075280-0022, arXiv:2304.01996",
      "peer_reviewed": true,
      "source": "sweep-allresults-2026-11-11",
      "provenance": "primary",
      "verified": {
        "checked_on": "2026-11-11",
        "method": "source text of arXiv:2304.01996 read locally (arXiv HTML or pypdf layout text); value copied from the harvested cell and the text, no LLM transcription of numbers; single reading (2026-11-11)",
        "reported_as": "-2.25755(8)",
        "note": "Table 2 (10 x 10), row 'Blockwise (70)' (BW (7-48-70) in Tables 3-5), column J2 = 0.2; sources/2304.01996-neurips.txt line 428; version read: the version of record, Adv. Neural Inf. Process. Syst. 36 (NeurIPS 2023), publisher PDF from proceedings.neurips.cc (text in sources/2304.01996-neurips.txt), and arXiv v3 (HTML, sources/2304.01996.html); the two print the same value in every energy cell of Tables 1-4. The older committed PDF text (sources/2304.01996.txt) interleaves Tables 1 and 2 and reads the 10 x 10 MPS (8) cell at J2 = 0.7 as -1.7682258; the PDF page and both versions above print -1.768225 | Eq. (8): H = J1 sum_<i,j> σ_i.σ_j + J2 sum_<<i,j>> σ_i.σ_j with Pauli matrices (Eq. 9), J1 = 1; the tables print 'Energy per site' of this H, the Pauli form, so E/N in the S.S convention is the printed value / 4 (RULES.md 5). Open boundaries (Sec. 6.2). Plausibility: the 10 x 10 cells at J2 = 0.5 (best -1.94030(9), -0.48508 in S.S units) lie above the periodic 10 x 10 records (about -0.4976), as an open lattice with fewer bonds per site must.",
        "secondary_of": null,
        "second_read": {
          "checked_on": "2026-11-11",
          "scope": "value",
          "method": "NeurIPS 2023 version of record (proceedings.neurips.cc file 01772a8b0420baec00c4d59fe2fbace6-Paper-Conference.pdf), pypdf plain and layout mode (sources/2304.01996-neurips.rp1-plain.txt); cross-checked against arXiv:2304.01996v3 PDF (sources/2304.01996v3.rp1-plain.txt) and the v3 HTML: all energy cells identical in both versions",
          "location": "Table 2 (10 x 10) line 434; Table 3 (10 x 10, Appendix C) line 1251, column J2 = 0.2",
          "reported_as": "-2.25755(8) (Energy per site, Pauli sigma.sigma form of Eq. (8), J1 = 1)",
          "conversion": "-2.25755 * 100 = -225.755000 (Pauli per site x N = stored Pauli total; S.S per site = -0.5643875); sigma 8e-5 * 100 = 0.008",
          "note": "L = 10 open square lattice (Sec. 6.2 'open boundary condition', the setting of all VMC tables), J1 = 1, J2 = 0.2 (column header), Pauli Eq. (8)-(9), 'Energy per site' column; bound_type variational (Sec. 6.2: 'based on the variational principle ... provides an upper bound'); no duplicate: data/J1J2/square_100_O_0.2.json holds only rows of this paper and batch",
          "source_entry": "RP1-arxiv-2304-01996#5 (qmbl-verify 2026-11-11, instances)"
        }
      },
      "compute": {
        "parameters": null,
        "gpu_hours": null,
        "device": null,
        "n_devices": null,
        "samples": null,
        "wall_clock": null,
        "cpu_core_hours": null,
        "bond_dimension": 70,
        "iterations": null,
        "reported_as": "method string: \"BW (7-48-70)\"",
        "source": "the row's published method string (scripts/method_names.mjs, 2026-09-17)",
        "scope": "row",
        "confidence": "high",
        "note": "Only the numbers the method string states; nothing else about this run's cost is recorded."
      }
    },
    {
      "energy": -219.2,
      "sigma": 0.09,
      "energy_variance": null,
      "dof": 100,
      "einf": 0,
      "v_score": null,
      "method": "VMC",
      "ansatz": "PixelCNN",
      "method_detail": "7 layers, 48 hidden dimensions, pretrained on the MPS result, no sign rule",
      "method_as_published": "PixelCNN (P) (7-48)",
      "family": "CNN / ResNet",
      "bound_type": "variational",
      "bound_type_reason": "variational ansatz at a stated size; energy is an upper bound (assigned during source reading)",
      "reference": "Chen, Newhouse, Chen, Luo & Soljačić, ANTN: Bridging Autoregressive Neural Networks and Tensor Networks for Quantum Many-Body Simulation, Adv. Neural Inf. Process. Syst. 36 (NeurIPS 2023), 450-476, doi:10.52202/075280-0022, arXiv:2304.01996",
      "peer_reviewed": true,
      "source": "sweep-allresults-2026-11-11",
      "provenance": "primary",
      "verified": {
        "checked_on": "2026-11-11",
        "method": "source text of arXiv:2304.01996 read locally (arXiv HTML or pypdf layout text); value copied from the harvested cell and the text, no LLM transcription of numbers; single reading (2026-11-11)",
        "reported_as": "-2.1920(9)",
        "note": "Table 3 (10 x 10), row 'PixelCNN (P) (7-48)', column J2 = 0.2; sources/2304.01996-neurips.txt line 1228; version read: the version of record, Adv. Neural Inf. Process. Syst. 36 (NeurIPS 2023), publisher PDF from proceedings.neurips.cc (text in sources/2304.01996-neurips.txt), and arXiv v3 (HTML, sources/2304.01996.html); the two print the same value in every energy cell of Tables 1-4. The older committed PDF text (sources/2304.01996.txt) interleaves Tables 1 and 2 and reads the 10 x 10 MPS (8) cell at J2 = 0.7 as -1.7682258; the PDF page and both versions above print -1.768225 | Eq. (8): H = J1 sum_<i,j> σ_i.σ_j + J2 sum_<<i,j>> σ_i.σ_j with Pauli matrices (Eq. 9), J1 = 1; the tables print 'Energy per site' of this H, the Pauli form, so E/N in the S.S convention is the printed value / 4 (RULES.md 5). Open boundaries (Sec. 6.2). Plausibility: the 10 x 10 cells at J2 = 0.5 (best -1.94030(9), -0.48508 in S.S units) lie above the periodic 10 x 10 records (about -0.4976), as an open lattice with fewer bonds per site must.",
        "secondary_of": null,
        "second_read": {
          "checked_on": "2026-11-11",
          "scope": "value",
          "method": "NeurIPS 2023 version of record (proceedings.neurips.cc file 01772a8b0420baec00c4d59fe2fbace6-Paper-Conference.pdf), pypdf plain and layout mode (sources/2304.01996-neurips.rp1-plain.txt); cross-checked against arXiv:2304.01996v3 PDF (sources/2304.01996v3.rp1-plain.txt) and the v3 HTML: all energy cells identical in both versions",
          "location": "Table 3 (10 x 10, Appendix C) line 1247, column J2 = 0.2",
          "reported_as": "-2.1920(9) (Energy per site, Pauli sigma.sigma form of Eq. (8), J1 = 1)",
          "conversion": "-2.1920 * 100 = -219.200000 (Pauli per site x N = stored Pauli total; S.S per site = -0.5480000); sigma 9e-4 * 100 = 0.09",
          "note": "L = 10 open square lattice (Sec. 6.2 'open boundary condition', the setting of all VMC tables), J1 = 1, J2 = 0.2 (column header), Pauli Eq. (8)-(9), 'Energy per site' column; bound_type variational (Sec. 6.2: 'based on the variational principle ... provides an upper bound'); sign rule not printed for this Table 3 row; 'no sign rule' read from the paper's labelling, where (S) marks sign-rule runs (Table 1, Table 4 caption) and the Table 2 scan is 'without using the sign rule'; caption: pretraining 'using MPS result (with bond dimension the last number in parenthesis)', but the label carries no bond dimension: which MPS was used is not printed; no duplicate: data/J1J2/square_100_O_0.2.json holds only rows of this paper and batch",
          "source_entry": "RP1-arxiv-2304-01996#6 (qmbl-verify 2026-11-11, instances)"
        }
      }
    },
    {
      "energy": -223.581,
      "sigma": 0.014,
      "energy_variance": null,
      "dof": 100,
      "einf": 0,
      "v_score": null,
      "method": "VMC",
      "ansatz": "PixelCNN",
      "method_detail": "8 layers, 48 hidden dimensions, no sign rule",
      "method_as_published": "PixelCNN (8-48)",
      "family": "CNN / ResNet",
      "bound_type": "variational",
      "bound_type_reason": "variational ansatz at a stated size; energy is an upper bound (assigned during source reading)",
      "reference": "Chen, Newhouse, Chen, Luo & Soljačić, ANTN: Bridging Autoregressive Neural Networks and Tensor Networks for Quantum Many-Body Simulation, Adv. Neural Inf. Process. Syst. 36 (NeurIPS 2023), 450-476, doi:10.52202/075280-0022, arXiv:2304.01996",
      "peer_reviewed": true,
      "source": "sweep-allresults-2026-11-11",
      "provenance": "primary",
      "verified": {
        "checked_on": "2026-11-11",
        "method": "source text of arXiv:2304.01996 read locally (arXiv HTML or pypdf layout text); value copied from the harvested cell and the text, no LLM transcription of numbers; single reading (2026-11-11)",
        "reported_as": "-2.23581(14)",
        "note": "Table 3 (10 x 10), row 'PixelCNN (8-48)', column J2 = 0.2; sources/2304.01996-neurips.txt line 1229; version read: the version of record, Adv. Neural Inf. Process. Syst. 36 (NeurIPS 2023), publisher PDF from proceedings.neurips.cc (text in sources/2304.01996-neurips.txt), and arXiv v3 (HTML, sources/2304.01996.html); the two print the same value in every energy cell of Tables 1-4. The older committed PDF text (sources/2304.01996.txt) interleaves Tables 1 and 2 and reads the 10 x 10 MPS (8) cell at J2 = 0.7 as -1.7682258; the PDF page and both versions above print -1.768225 | Eq. (8): H = J1 sum_<i,j> σ_i.σ_j + J2 sum_<<i,j>> σ_i.σ_j with Pauli matrices (Eq. 9), J1 = 1; the tables print 'Energy per site' of this H, the Pauli form, so E/N in the S.S convention is the printed value / 4 (RULES.md 5). Open boundaries (Sec. 6.2). Plausibility: the 10 x 10 cells at J2 = 0.5 (best -1.94030(9), -0.48508 in S.S units) lie above the periodic 10 x 10 records (about -0.4976), as an open lattice with fewer bonds per site must.",
        "secondary_of": null,
        "second_read": {
          "checked_on": "2026-11-11",
          "scope": "value",
          "method": "NeurIPS 2023 version of record (proceedings.neurips.cc file 01772a8b0420baec00c4d59fe2fbace6-Paper-Conference.pdf), pypdf plain and layout mode (sources/2304.01996-neurips.rp1-plain.txt); cross-checked against arXiv:2304.01996v3 PDF (sources/2304.01996v3.rp1-plain.txt) and the v3 HTML: all energy cells identical in both versions",
          "location": "Table 3 (10 x 10, Appendix C) line 1248, column J2 = 0.2",
          "reported_as": "-2.23581(14) (Energy per site, Pauli sigma.sigma form of Eq. (8), J1 = 1)",
          "conversion": "-2.23581 * 100 = -223.581000 (Pauli per site x N = stored Pauli total; S.S per site = -0.5589525); sigma 14e-5 * 100 = 0.014",
          "note": "L = 10 open square lattice (Sec. 6.2 'open boundary condition', the setting of all VMC tables), J1 = 1, J2 = 0.2 (column header), Pauli Eq. (8)-(9), 'Energy per site' column; bound_type variational (Sec. 6.2: 'based on the variational principle ... provides an upper bound'); sign rule not printed for this Table 3 row; 'no sign rule' read from the paper's labelling, where (S) marks sign-rule runs (Table 1, Table 4 caption) and the Table 2 scan is 'without using the sign rule'; no duplicate: data/J1J2/square_100_O_0.2.json holds only rows of this paper and batch",
          "source_entry": "RP1-arxiv-2304-01996#7 (qmbl-verify 2026-11-11, instances)"
        }
      }
    },
    {
      "energy": -225.946,
      "sigma": 0.006,
      "energy_variance": null,
      "dof": 100,
      "einf": 0,
      "v_score": null,
      "method": "VMC",
      "ansatz": "ANTN",
      "method_detail": "elementwise, PixelCNN 7 layers, 48 hidden dimensions, MPS randomly initialised, no sign rule",
      "method_as_published": "EW (RI) (7-48-8)",
      "family": "tensor network",
      "bound_type": "variational",
      "bound_type_reason": "variational ansatz at a stated size; energy is an upper bound (assigned during source reading)",
      "reference": "Chen, Newhouse, Chen, Luo & Soljačić, ANTN: Bridging Autoregressive Neural Networks and Tensor Networks for Quantum Many-Body Simulation, Adv. Neural Inf. Process. Syst. 36 (NeurIPS 2023), 450-476, doi:10.52202/075280-0022, arXiv:2304.01996",
      "peer_reviewed": true,
      "source": "sweep-allresults-2026-11-11",
      "provenance": "primary",
      "verified": {
        "checked_on": "2026-11-11",
        "method": "source text of arXiv:2304.01996 read locally (arXiv HTML or pypdf layout text); value copied from the harvested cell and the text, no LLM transcription of numbers; single reading (2026-11-11)",
        "reported_as": "-2.25946(6)",
        "note": "Table 3 (10 x 10), row 'EW (RI) (7-48-8)', column J2 = 0.2; sources/2304.01996-neurips.txt line 1231; version read: the version of record, Adv. Neural Inf. Process. Syst. 36 (NeurIPS 2023), publisher PDF from proceedings.neurips.cc (text in sources/2304.01996-neurips.txt), and arXiv v3 (HTML, sources/2304.01996.html); the two print the same value in every energy cell of Tables 1-4. The older committed PDF text (sources/2304.01996.txt) interleaves Tables 1 and 2 and reads the 10 x 10 MPS (8) cell at J2 = 0.7 as -1.7682258; the PDF page and both versions above print -1.768225 | Eq. (8): H = J1 sum_<i,j> σ_i.σ_j + J2 sum_<<i,j>> σ_i.σ_j with Pauli matrices (Eq. 9), J1 = 1; the tables print 'Energy per site' of this H, the Pauli form, so E/N in the S.S convention is the printed value / 4 (RULES.md 5). Open boundaries (Sec. 6.2). Plausibility: the 10 x 10 cells at J2 = 0.5 (best -1.94030(9), -0.48508 in S.S units) lie above the periodic 10 x 10 records (about -0.4976), as an open lattice with fewer bonds per site must.",
        "secondary_of": null,
        "second_read": {
          "checked_on": "2026-11-11",
          "scope": "value",
          "method": "NeurIPS 2023 version of record (proceedings.neurips.cc file 01772a8b0420baec00c4d59fe2fbace6-Paper-Conference.pdf), pypdf plain and layout mode (sources/2304.01996-neurips.rp1-plain.txt); cross-checked against arXiv:2304.01996v3 PDF (sources/2304.01996v3.rp1-plain.txt) and the v3 HTML: all energy cells identical in both versions",
          "location": "Table 3 (10 x 10, Appendix C) line 1250, column J2 = 0.2",
          "reported_as": "-2.25946(6) (Energy per site, Pauli sigma.sigma form of Eq. (8), J1 = 1)",
          "conversion": "-2.25946 * 100 = -225.946000 (Pauli per site x N = stored Pauli total; S.S per site = -0.5648650); sigma 6e-5 * 100 = 0.006",
          "note": "L = 10 open square lattice (Sec. 6.2 'open boundary condition', the setting of all VMC tables), J1 = 1, J2 = 0.2 (column header), Pauli Eq. (8)-(9), 'Energy per site' column; bound_type variational (Sec. 6.2: 'based on the variational principle ... provides an upper bound'); sign rule not printed for this Table 3 row; 'no sign rule' read from the paper's labelling, where (S) marks sign-rule runs (Table 1, Table 4 caption) and the Table 2 scan is 'without using the sign rule'; no duplicate: data/J1J2/square_100_O_0.2.json holds only rows of this paper and batch",
          "source_entry": "RP1-arxiv-2304-01996#8 (qmbl-verify 2026-11-11, instances)"
        }
      },
      "compute": {
        "parameters": null,
        "gpu_hours": null,
        "device": null,
        "n_devices": null,
        "samples": null,
        "wall_clock": null,
        "cpu_core_hours": null,
        "bond_dimension": 8,
        "iterations": null,
        "reported_as": "method string: \"EW (RI) (7-48-8)\"",
        "source": "the row's published method string (scripts/method_names.mjs, 2026-09-17)",
        "scope": "row",
        "confidence": "high",
        "note": "Only the numbers the method string states; nothing else about this run's cost is recorded."
      }
    }
  ],
  "coverage": [
    {
      "checked_on": "2026-11-11",
      "method": "table harvest of cached arXiv sources (match_tables.mjs --all), every matched paper read for this instance",
      "screened": [
        "arXiv:2304.01996"
      ],
      "found": 1,
      "note": "Only papers whose tables matched the instance by size were read; not a citation screen. Energies from: arXiv:2304.01996."
    }
  ],
  "url": "https://qmbl.org/i/J1J2/square_100_O_0.2/",
  "per_site_divisor": 400,
  "per_site_label": "E/N (S.S)",
  "best": {
    "exact": null,
    "variational": {
      "energy": -226.034,
      "sigma": 0.006,
      "method": "VMC",
      "ansatz": "ANTN",
      "method_detail": "elementwise, PixelCNN 7 layers, 48 hidden dimensions, no sign rule",
      "reference": "Chen, Newhouse, Chen, Luo & Soljačić, ANTN: Bridging Autoregressive Neural Networks and Tensor Networks for Quantum Many-Body Simulation, Adv. Neural Inf. Process. Syst. 36 (NeurIPS 2023), 450-476, doi:10.52202/075280-0022, arXiv:2304.01996",
      "bound_type": "variational",
      "energy_per_site": -0.565085
    },
    "projected": null,
    "extrapolated": null,
    "approximate": null,
    "below_exact": []
  },
  "no_best_reason": null,
  "record": {
    "energy": -226.034,
    "sigma": 0.006,
    "method": "VMC",
    "ansatz": "ANTN",
    "method_detail": "elementwise, PixelCNN 7 layers, 48 hidden dimensions, no sign rule",
    "reference": "Chen, Newhouse, Chen, Luo & Soljačić, ANTN: Bridging Autoregressive Neural Networks and Tensor Networks for Quantum Many-Body Simulation, Adv. Neural Inf. Process. Syst. 36 (NeurIPS 2023), 450-476, doi:10.52202/075280-0022, arXiv:2304.01996",
    "bound_type": "variational",
    "energy_per_site": -0.565085
  },
  "no_record_reason": null
}
