Magnus: neutrino oscillation probabilities for any Hermitian Hamiltonian, any number of flavors, and any matter profile
课题
在连续非均匀介质（如地球、太阳、超新星激波及相对论喷流）中，空间变化的哈密顿量在不同位置不对易，传统多层阶梯近似误差衰减缓慢（仅按平板厚度二次方 收敛），而自适应连续 ODE 积分器在面对超大累积振荡相位 时计算代价急剧膨胀，且常用近似方法在短基线或强涨落湍流区域常会失效。
突破
数值求解器单步幺正性偏离低至 ，在保持几何精度的同时比 DOP853 自适应积分提速 30 至 4800 倍。对于超高能（）天体物理中微子穿过地核时，物质势远超真空项且在味空间严格对角，地核穿越造成的味成分变化小于 ；在超新星激波后湍流扰动 下，解析旋转波近似（RWA）仅在长传播区域（）有效，短距离（）必须使用高阶 Magnus 展开才能精确解析能级间跃迁。
理由
该工作开发了用于计算任意厄米哈密顿量、任意味数及任意介质剖面下中微子振荡概率的开源高阶几何积分求解器 Magnus，覆盖了超新星激波、相对论喷流与超高能天体物理中微子等高能瞬变天体物理核心场景中的味演化计算需求。
Magnus: neutrino oscillation probabilities for any Hermitian Hamiltonian, any number of flavors, and any matter profile
核心课题、机制与切入建议 Problem, Method & Entry
🎯 课题目标 在连续非均匀介质（如地球、太阳、超新星激波及相对论喷流）中
在连续非均匀介质（如地球、太阳、超新星激波及相对论喷流）中，空间变化的哈密顿量在不同位置不对易，传统多层阶梯近似误差衰减缓慢（仅按平板厚度二次方 收敛），而自适应连续 ODE 积分器在面对超大累积振荡相位 时计算代价急剧膨胀，且常用近似方法在短基线或强涨落湍流区域常会失效。
⚙️ 方法与机制 基于 Magnus 展开将时间有序演化算符表达为单矩阵指数 $\...
基于 Magnus 展开将时间有序演化算符表达为单矩阵指数 ，利用反厄米生成元截断天然保证严格幺正性；结合 Gauss-Legendre 配置求积法实现高达 4 至 8 阶收敛，并结合瞬时本征态绝热输运与非绝热共振区域的 Magnus 局部修补（hybrid engine）；对超长基线引入探测器相对能量展宽 诱导的干涉阻尼因子 计算相平均概率 ；针对伪狄拉克分裂态将质量态拆分为 ，并支持显式断点输入处理激波前沿。
💡 建议入口 4 The method
物理假设、图表与结果边界 Physical Models, Figures & Results
📐 物理假设
- 在伪狄拉克（pseudo-Dirac）模型中，分裂态假设由有效质量平方差分别为 与 的质量本征态叠加构成 。
📊 核心图表
🏁 作者报告的结果与边界 数值求解器单步幺正性偏离低至 $4\times 10^{-16}...
数值求解器单步幺正性偏离低至 ，在保持几何精度的同时比 DOP853 自适应积分提速 30 至 4800 倍。对于超高能（）天体物理中微子穿过地核时，物质势远超真空项且在味空间严格对角，地核穿越造成的味成分变化小于 ；在超新星激波后湍流扰动 下，解析旋转波近似（RWA）仅在长传播区域（）有效，短距离（）必须使用高阶 Magnus 展开才能精确解析能级间跃迁。
⚠️ 物理边界与局限
- 解析旋转波近似（RWA）忽略了耦合中快速振荡项，仅在跨越多倍振荡长度的长距离演化中准确，在单振荡波长尺度下误差显著。
学术脉络、进展比较与前置建议 Lineage, Context & Progress
🔗 组内学术脉络 无组内匹配
自动筛选未建立论文与组内 Work 的科学关系。
📈 进展比较
将李群几何积分技术（Magnus 展开）系统化引入多味中微子物理，消除了计算网格步长对极高振荡相位的依赖，并将计算复杂度转化为仅由物质密度剖面几何梯度的变化率决定；建立了统摄真空、PREM 地球、太阳模型、超新星激波、相对论喷流及包含 NSI/LIV/惰性/伪狄拉克等新物理哈密顿量的全流程统一算法与开源实现架构。
🔍 引用线索
- ... and 0.18 ms: “Each split state \nu_{j} is replaced by two mass states, (\nu_{j}\pm s_{j})/\sqrt{2}, where s_{j} is its sterile partner, with masses squared m_{j}^{2} and m_{j}^{2}+\delta m^{2}_{j}.”
- [0.332 0.323 0.345]: “Crossing the core changes the composition by at most , at 1 TeV, and by less than above 10 TeV: at these energies, the matter potential exceeds the vacuum term by orders of magnitude, and it is diagonal in flavor,”
- transition between levels 2 and 3: “The closed form relies on the rotating-wave approximation, which neglects a rapidly oscillating part of the coupling; that part averages out only over many oscillation lengths. Hence, it is accurate only in long regions”
证据审计与信源材料 覆盖：正文指定部分 展开核验 ▾
📑 实际阅读范围 正文指定部分
覆盖范围：已检查正文中支撑判断的指定章节；版本 arXiv:2610.07159v1
IntroductionMag$nu$s: neutrino oscillation probabilities for any Hermitian Hamiltonian, any number of flavors, and any matter profile1 Introduction2 SynopsisStandard oscillation parametersActive-sterile mixingExponentially falling density: 3e3 g/cm^3 at theorigin, scale height 10 km, 25 km baseline.Every mixing parameter left out defaults to zero.3 Neutrino oscillations: units, notation, conventions4 The method4.1 History and context4.2 Where the method applies and where it does not4.3 The Magnus expansion4.4 Truncation is exactly unitary4.5 Two quadrature rules for the Magnus expansion4.6 Slab composition4.7 Tolerance and adaptive refinement4.8 The matrix exponential4.9 Adiabatic transport with exact Magnus patches4.10 The phase-averaged probability4.10.1 The averaged probability4.10.2 Coherent blocks4.10.3 The phase average4.10.4 A varying Hamiltonian5 Code description5.1 Scope and interface5.2 The Mag$nu$s architecture5.3 The seven engines5.4 Example: one request, through the layers---- What the user writes: the whole request ----One energy, one chord, and two refinement keywords---- Inside Magnus: what the call runs through ----Top layer: the named wrapperSecond layer: builds H, tries its enginesThird layer: the scan over (energy, baseline)Base layer: the refinement ladder, one point5.5 Batching and parallelization5.6 Declaring the structure of a profile24 slabs of 250 km, alternating 2 and 8 g/cm^3The jumps declared: the ladder keeps its own grid,adds an edge at each of the 23 jumps, then refinesto the toleranceThe grid given in full, as [start, end] pairs: oneslab per layer, no refinement, a single evaluationA grid of 12 slabs, each straddling one jump: theuser owns this grid, and nothing repairs itP1 = P2 = 0.0240046, exact: the profile is constantwithin every slab. P3 = 0.0246158, off by 6e-45.7 Documentation and executable examples5.8 InstallationThe package:Plus what the notebooks need:Plus what the tests need:5.9 Code availability5.10 Code validation5.10.1 Against the derivation5.10.2 Against closed forms5.10.3 Against an independent solver5.10.4 Against itself, where meaningful5.10.5 Against a population5.10.6 Continuous integration5.11 Limitations5.11.1 Limitations of the method5.11.2 Limitations of the implementation6 Mag$nu$s usage and examples6.1 The full application programming interface (API)6.2 A minimal example in vacuum, step-by-stepNatural units: energies in eV, lengths in eV^-1Pme = 0.03127, Pmm = 0.39231, Pmt = 0.57643The amplitude is indexed the other way6.3 Choosing the oscillation parameters6.3.1 Three flavorsPme = 0.05166, against 0.03127 by defaultPme = 0.05579; ’LEM’ gives 0.05558Pme = 0.04998; ’freefluxesrsbl’gives 0.0455253 lines, e.g.,OSCPARAMSNUFIT10IO | NuFIT 1.0, IOThe releases the loader reads, and thecategories of each18 lines, e.g., NuFIT 2.1 [’LEM’, ’LID’]Pme = 0.05098; the other five stay attheir defaultsThe other five stay at their NuFIT 6.1 IOvalues. Pme = 0.01800, against0.05241 with the set’s own dCP6.3.2 Specifying angle conventionss12 reads 0.5557, 0.3088, 0.5892, 33.759and Pme = 0.03127 in all four cases6.3.3 Two flavorsP[0][1] = 0.013089The same point, angles in degrees6.3.4 Four flavorsPme = 0.02801, Pmm = 0.399586.3.5 Five flavorsPme = 0.03984, Pmm = 0.406846.3.6 Default non-standard parameters6.4 Batched callsP[-1] equals the single call at 1300 km6.5 Constant density’NuFIT 6.1’ is the default; re-loaded for clarity onlyThree flavors, one energy and one baseline0.0 g/cm3 Pme = 0.0037733.0 g/cm3 Pme = 0.0135898.0 g/cm3 Pme = 0.05049713.0 g/cm3 Pme = 0.113343Two flavors: one angle and one splittingP[0][1] = 0.005766Two hundred energies in one callP.shape = (200, 3, 3)6.6 Using wrappers vs. scenario calls vs. direct calls--- 1. With the named wrappersPme = 0.00377306 in vacuum, 0.01358909 inmatter, in 0.15 msPscan.shape = (200, 3, 3), in 0.13 ms--- 2. With a scenario functionThe same number, to 1e-16, in 3.6 ms--- 3. With a Hamiltonian built by handThe same two numbers, in 0.043 msThe same 200 points, in 5.8 ms6.6.1 Through a wrapperPme = 0.017126. Dropping epsem gives0.013589, the standard wrapper’s answer6.6.2 Through a scenario functionPme = 0.017126, the same number thewrapper above returns6.6.3 Through a direct callproj for three flavors is diag(1, 0, 0)1. mixing, 2. matter, 3. new physicsPme = 0.017126 again6.6.4 Which to use6.6.5 Returning the evolution operatorP is the matrix the call returns withoutthe keyword; U is complex and unitary.Both of these hold:the second denser than the first6.6.6 Asking for the averaged probabilityP[0, 0] = 0.5481, P[1, 0] = 0.2155: thefamiliar sumi |Uei|^2 |Umi|^2, from theeigenvectors alonePhaseAveragingWarning: the phase-averagedprobability depends on the energy spreadat 1 of 1 (energy, L) point(s) ...P[0, 0] = 0.9844, P[1, 1] = 0.9088;P[0, 0] = 0.8934: the decohered limit,now from the eigenvectors in matter6.7 The Hamiltonians6.7.1 Shipped HamiltoniansOver 200 energies: 3.7 ms... and 0.18 ms6.7.2 Building a new Hamiltonian from scratchPme = 0.021309Pme = 0.0048656.8 Pre-packaged plotting routines6.9 Probability vs. energy and distance0.082635, as in the listing0.071325, as in the listingOne mole of electrons per cubic centimeter, in thenatural units that magnus works inGive a constant profile as a single number, anda varying one as a function of the distancetraveled, returning the electron number densitythere. The distance arrives in eV^-1, likeevery length in magnusThe first keyword says that the functions returnan electron number density, not a mass densityLeft column: 5,000 distances at one energy. Thelist of distances goes where a single one wouldRight column: 3,000 energies at one distanceThe vacuum curves, for reference6.10 Layered matterThe slab edges are declared, so no slab straddles ajump; within each, one term of the expansion is exact6.11 Bi-probability plotDUNE: 2 GeV, 1300 km; a crust-like 3 g/cm^3With the shipped plotterWith matplotlib directly6.12 The Earth6.12.1 Neutrinos & antineutrinos, 2–5 flavorsTwo flavors is the (nue, numu) reduction of the 1-3sector, so it takes that one angle and splitting byname. Every wrapper above two flavors defaults tothe same NuFIT 6.1 set, so it needs none passed.The sterile pairs run three decades higher, where aneV-scale splitting places its resonance6.12.2 Probabilities between two locationspositive, south and west negativeTwo site names replace costhz and L: the chordbetween them is the baselineThe chord itself, in km, for a legend6.12.3 New physicsThree decades up is where an eV-scale splittingplaces its matter resonanceThe same call each time, a different extra termChange between 3nu and 3+1 survival probability6.12.4 Earth oscillogramsWith the shipped plotter, one call per panelOr with matplotlib directly, one axes per panel6.12.5 Underground sources and detectorsOn the surface, the baseline is given0.9115872 km down, the depth fixes the baseline0.911600Horizontal: no path at all at the surface1.0000002 km down it crosses 160 km of rock0.9975600.020930Under ice, as at IceCubeUnder rock, as at SNOLAB0.0213256.12.6 Using a custom Earth density profileFour shells for PREM’s ten, each at the mass-weightedmean density of the region it replacesThose four edges are PREM boundaries, so theircrossings come straight out of the helper6.13 Strategy report after a call[’certified’, ’declined’, ’engine’,’family’, ’hiddenfeature’, ’sampling’,’trace’]3227.2[(’hybrid’, ’the profile is notresolved at the probe scale’)]6.14 The SunThe table ships with Magnus. Its columns are readonly to plot the density (top panel of the figure)The solar spectrum, 0.1 to 20 MeVShared by the four calls: the model by name, fromthe center to its last row, nue to nue, the averageStandard three-flavor oscillationsThe same, with non-standard interactionsOne and two sterile states. nn/np is read fromthe table; passing rationumberneutronstoprotonsoverrides itP3 = 0.5449 at 0.1 MeV, 0.2993 at 20 MeVP = 0.666, 0.092, 0.259 in about 2 s6.14.1 Varying the production pointP = 0.389, against 0.375 from the center6.14.2 Probability averaging6.14.3 Comparing solar models6.14.4 Comparing against approximationsFold theta13 back in6.15 A supernova shock frontUnmarkedDiscontinuityWarning: theHamiltonian is discontinuous at the scaleof the grid this scan builds, and notbreakpoints were given. A slabstraddling a density jump degrades thequadrature no matter how many slabs areused; pass tbreakpoints at thediscontinuities.Both fronts, 0.07 km wide, centered on the contactdiscontinuity and on the forward shock: declarewhere each begins and ends4000 baselines from 10,200 to 80,000 km, at oneenergy: the cumulative engine answersOne and two sterile statesPhaseAveragingWarning: average=True on aprofile with discontinuities has no closedform, so the probability was propagatedacross an energy window of +/-10.0% andaveraged over 41 samples. This is theaverage over that window, not the L/E ->infinity limit, and it depends on thewindow: the largest standard error of themean here is 4.47e-02.Pass averagespread to set the half-widthof the window to the resolution of themeasurement, and averagensamples toreduce the error, which falls as theinverse square root of the number ofsamples.P = 0.56P = 0.5846.16 High-energy astrophysical neutrinosThe standard case at 1 Mpc and at 1 GpcLargest difference over all energiesand channels: nonePhaseAveragingWarning: depends onthe energy spread at 46 of 60 points[0.333 0.666 0.001]A source 100 Mpc away: far enough for every phaseto average away at every energy. The averagedepends on the baseline through the phasesStandard: the averaged matrix at each energyLIV: the n = 1 operator, aligned with the PMNSangles, sized to equal the vacuum term at 100 TeVNSI through the Earth: the flux arrives decohered,then crosses the Earth coherently, so the two legscompose as probability matrices. The ladder startsat 32 slabs: the matter term alone winds more thanpi across a coarser one3+1: one sterile state, sin^2 of both active-sterile angles 0.1, Dm41^2 = 1 eV^2Pseudo-Dirac: the second mass state is pairedwith a partner split by 1e-13 eV^2from 100 Mpc, as in the fifth panelsplittings of nu2, at 100 TeV and100 Mpc, from coherent to decohered[0.326 0.343 0.33 ] Coherent: standard[0.328 0.338 0.334] Between: warns[0.332 0.323 0.345] DecoheredInputs of the vacuum part: the PMNSmatrix and the three m^2 [eV^2]PREM along the chord through the core:at position l, VCC and the neutron-to-proton ratio of the local layerThe Hamiltonian, of energy and position:vacuum part plus matter partThrough the core, with the PREM layerboundaries declared as breakpointsThe averaged vacuum leg, then the Earth[0.332 0.323 0.345]6.17 Assorted long examples6.17.1 A cavity in the Earth’s crustAn empty crust needs no profile, only a numberWater, iron, a mineral deposit, a zone of faults:(rho, Ye, w) for each6.17.2 GeoneutrinosThe mantle chord: cosine of the zenithangle at the detector, and the depths ofthe production point and of Borexino [km]Length of the chord [km]Density along the chord [g/cm^3]: PREM,with its ocean layer replaced by rockRadius at distance l along the chordPhase-averaged survival probability inmatter, at electron fraction of the mantle0.551The same, in vacuum0.548Twelve energies per cycle of the pair split byDm31^2 on the longest chord, 7,300 kmBorexino, 1.4 km underground. A production pointenters as its depth and the cosine of its zenithangle at the detector; the call declares the PREMboundaries on the chord by itself. The oscillationparameters are the defaults, NuFIT 6.1.The local point, 10 km deep and 100 km away,run in reverse, from the detector (see text)6.17.3 Neutrino tomography of the SunThe chord above: b = 0.3 Rsun6.17.4 Astrophysical relativistic jetEvolve from the jet head (L0) to thesurface (L); also return the operatorProbabilities and evolution operatorWeights, w[i, alpha]The three envelopes, g/cm3, r in cmProduction points spread over one oscillationlength at the jet headDrawn: Psmooth[:, gd.NUE, gd.NUE], and the others6.17.5 Long-range interactions in the SunAverage over them, then scale by the 8Bspectrum and total flux, cm^-2 s^-1 MeV^-1The B16-GS98 model that ships with Magnus,on its tabulated radii and on to 20 RsunTwo mediator ranges, one coupling: g’^2 makes Vemua tenth of VCC at R0 when 1/m = RsunThe standard part from the shipped builders; thenew term is the only line written here6.17.6 Turbulent matter profileH at the mean density: vacuum plus matterEigenvalues w, matter eigenvectors Vm, andthe gap the mode has to matchCoupling of levels 2 and 3, and the lengthof complete conversion, in units of LOSCVCC follows the densityOne operator per slab, from 9 nodes atMagnus order 6Ordered product, earliest slab firstRotate to the matter basis, then read thetransition between levels 2 and 3Two flavors: the 1-3 sectorThree flavorsThe same call with a sterile state added6.18 Running a scan in parallelEvery point has its own baseline, so nobatched engine takes this scanengine takes it, at the default njobs = 1nothing; the call runs in one process6.19 Using Mag$nu$s from the command line7 Performance, precision, and reach7.1 Timing measurement protocol7.2 Slab width follows the profile, not the phase7.3 What a probability costs7.4 Precision, accuracy, and tolerance7.5 Accuracy and speed: smooth density profile7.6 Accuracy and speed: a supernova shock front7.7 Accuracy and speed: the Sun8 Comparison to other codes8.1 Relation to other oscillation codes8.2 Speed-accuracy comparison8.3 When to use Mag$nu$s9 ConclusionsDeclaration of generative AI and AI-assisted technologies in the manuscript preparation processAcknowledgmentsReferencesAppendix A Magnus expansion at any orderA.1 Computing Magnus termsA.2 Why the collocation orders are evenAppendix B Predefined oscillation-parameter setsAppendix C The tutorial notebooksAppendix D The test suiteInstructions for reporting errors
📦 信源依据与定位
- 📦 arXiv TeX 源码包 位置：Introduction; Mag$nu$s: neutrino oscillation probabilities for any Hermitian Hamiltonian, any number of flavors, and any matter profile; 1 Introduction; 2 Synopsis; Standard oscillation parameters; Active-sterile mixing; Exponentially falling density: 3e3 g/cm^3 at the; origin, scale height 10 km, 25 km baseline.; Every mixing parameter left out defaults to zero.; 3 Neutrino oscillations: units, notation, conventions; 4 The method; 4.1 History and context; 4.2 Where the method applies and where it does not; 4.3 The Magnus expansion; 4.4 Truncation is exactly unitary; 4.5 Two quadrature rules for the Magnus expansion; 4.6 Slab composition; 4.7 Tolerance and adaptive refinement; 4.8 The matrix exponential; 4.9 Adiabatic transport with exact Magnus patches; 4.10 The phase-averaged probability; 4.10.1 The averaged probability; 4.10.2 Coherent blocks; 4.10.3 The phase average; 4.10.4 A varying Hamiltonian; 5 Code description; 5.1 Scope and interface; 5.2 The Mag$nu$s architecture; 5.3 The seven engines; 5.4 Example: one request, through the layers; ---- What the user writes: the whole request ----; One energy, one chord, and two refinement keywords; ---- Inside Magnus: what the call runs through ----; Top layer: the named wrapper; Second layer: builds H, tries its engines; Third layer: the scan over (energy, baseline); Base layer: the refinement ladder, one point; 5.5 Batching and parallelization; 5.6 Declaring the structure of a profile; 24 slabs of 250 km, alternating 2 and 8 g/cm^3; The jumps declared: the ladder keeps its own grid,; adds an edge at each of the 23 jumps, then refines; to the tolerance; The grid given in full, as [start, end] pairs: one; slab per layer, no refinement, a single evaluation; A grid of 12 slabs, each straddling one jump: the; user owns this grid, and nothing repairs it; P1 = P2 = 0.0240046, exact: the profile is constant; within every slab. P3 = 0.0246158, off by 6e-4; 5.7 Documentation and executable examples; 5.8 Installation; The package:; Plus what the notebooks need:; Plus what the tests need:; 5.9 Code availability; 5.10 Code validation; 5.10.1 Against the derivation; 5.10.2 Against closed forms; 5.10.3 Against an independent solver; 5.10.4 Against itself, where meaningful; 5.10.5 Against a population; 5.10.6 Continuous integration; 5.11 Limitations; 5.11.1 Limitations of the method; 5.11.2 Limitations of the implementation; 6 Mag$nu$s usage and examples; 6.1 The full application programming interface (API); 6.2 A minimal example in vacuum, step-by-step; Natural units: energies in eV, lengths in eV^-1; Pme = 0.03127, Pmm = 0.39231, Pmt = 0.57643; The amplitude is indexed the other way; 6.3 Choosing the oscillation parameters; 6.3.1 Three flavors; Pme = 0.05166, against 0.03127 by default; Pme = 0.05579; ’LEM’ gives 0.05558; Pme = 0.04998; ’freefluxesrsbl’; gives 0.04552; 53 lines, e.g.,; OSCPARAMSNUFIT10IO | NuFIT 1.0, IO; The releases the loader reads, and the; categories of each; 18 lines, e.g., NuFIT 2.1 [’LEM’, ’LID’]; Pme = 0.05098; the other five stay at; their defaults; The other five stay at their NuFIT 6.1 IO; values. Pme = 0.01800, against; 0.05241 with the set’s own dCP; 6.3.2 Specifying angle conventions; s12 reads 0.5557, 0.3088, 0.5892, 33.759; and Pme = 0.03127 in all four cases; 6.3.3 Two flavors; P[0][1] = 0.013089; The same point, angles in degrees; 6.3.4 Four flavors; Pme = 0.02801, Pmm = 0.39958; 6.3.5 Five flavors; Pme = 0.03984, Pmm = 0.40684; 6.3.6 Default non-standard parameters; 6.4 Batched calls; P[-1] equals the single call at 1300 km; 6.5 Constant density; ’NuFIT 6.1’ is the default; re-loaded for clarity only; Three flavors, one energy and one baseline; 0.0 g/cm3 Pme = 0.003773; 3.0 g/cm3 Pme = 0.013589; 8.0 g/cm3 Pme = 0.050497; 13.0 g/cm3 Pme = 0.113343; Two flavors: one angle and one splitting; P[0][1] = 0.005766; Two hundred energies in one call; P.shape = (200, 3, 3); 6.6 Using wrappers vs. scenario calls vs. direct calls; --- 1. With the named wrappers; Pme = 0.00377306 in vacuum, 0.01358909 in; matter, in 0.15 ms; Pscan.shape = (200, 3, 3), in 0.13 ms; --- 2. With a scenario function; The same number, to 1e-16, in 3.6 ms; --- 3. With a Hamiltonian built by hand; The same two numbers, in 0.043 ms; The same 200 points, in 5.8 ms; 6.6.1 Through a wrapper; Pme = 0.017126. Dropping epsem gives; 0.013589, the standard wrapper’s answer; 6.6.2 Through a scenario function; Pme = 0.017126, the same number the; wrapper above returns; 6.6.3 Through a direct call; proj for three flavors is diag(1, 0, 0); 1. mixing, 2. matter, 3. new physics; Pme = 0.017126 again; 6.6.4 Which to use; 6.6.5 Returning the evolution operator; P is the matrix the call returns without; the keyword; U is complex and unitary.; Both of these hold:; the second denser than the first; 6.6.6 Asking for the averaged probability; P[0, 0] = 0.5481, P[1, 0] = 0.2155: the; familiar sumi |Uei|^2 |Umi|^2, from the; eigenvectors alone; PhaseAveragingWarning: the phase-averaged; probability depends on the energy spread; at 1 of 1 (energy, L) point(s) ...; P[0, 0] = 0.9844, P[1, 1] = 0.9088;; P[0, 0] = 0.8934: the decohered limit,; now from the eigenvectors in matter; 6.7 The Hamiltonians; 6.7.1 Shipped Hamiltonians; Over 200 energies: 3.7 ms; ... and 0.18 ms; 6.7.2 Building a new Hamiltonian from scratch; Pme = 0.021309; Pme = 0.004865; 6.8 Pre-packaged plotting routines; 6.9 Probability vs. energy and distance; 0.082635, as in the listing; 0.071325, as in the listing; One mole of electrons per cubic centimeter, in the; natural units that magnus works in; Give a constant profile as a single number, and; a varying one as a function of the distance; traveled, returning the electron number density; there. The distance arrives in eV^-1, like; every length in magnus; The first keyword says that the functions return; an electron number density, not a mass density; Left column: 5,000 distances at one energy. The; list of distances goes where a single one would; Right column: 3,000 energies at one distance; The vacuum curves, for reference; 6.10 Layered matter; The slab edges are declared, so no slab straddles a; jump; within each, one term of the expansion is exact; 6.11 Bi-probability plot; DUNE: 2 GeV, 1300 km; a crust-like 3 g/cm^3; With the shipped plotter; With matplotlib directly; 6.12 The Earth; 6.12.1 Neutrinos & antineutrinos, 2–5 flavors; Two flavors is the (nue, numu) reduction of the 1-3; sector, so it takes that one angle and splitting by; name. Every wrapper above two flavors defaults to; the same NuFIT 6.1 set, so it needs none passed.; The sterile pairs run three decades higher, where an; eV-scale splitting places its resonance; 6.12.2 Probabilities between two locations; positive, south and west negative; Two site names replace costhz and L: the chord; between them is the baseline; The chord itself, in km, for a legend; 6.12.3 New physics; Three decades up is where an eV-scale splitting; places its matter resonance; The same call each time, a different extra term; Change between 3nu and 3+1 survival probability; 6.12.4 Earth oscillograms; With the shipped plotter, one call per panel; Or with matplotlib directly, one axes per panel; 6.12.5 Underground sources and detectors; On the surface, the baseline is given; 0.911587; 2 km down, the depth fixes the baseline; 0.911600; Horizontal: no path at all at the surface; 1.000000; 2 km down it crosses 160 km of rock; 0.997560; 0.020930; Under ice, as at IceCube; Under rock, as at SNOLAB; 0.021325; 6.12.6 Using a custom Earth density profile; Four shells for PREM’s ten, each at the mass-weighted; mean density of the region it replaces; Those four edges are PREM boundaries, so their; crossings come straight out of the helper; 6.13 Strategy report after a call; [’certified’, ’declined’, ’engine’,; ’family’, ’hiddenfeature’, ’sampling’,; ’trace’]; 3227.2; [(’hybrid’, ’the profile is not; resolved at the probe scale’)]; 6.14 The Sun; The table ships with Magnus. Its columns are read; only to plot the density (top panel of the figure); The solar spectrum, 0.1 to 20 MeV; Shared by the four calls: the model by name, from; the center to its last row, nue to nue, the average; Standard three-flavor oscillations; The same, with non-standard interactions; One and two sterile states. nn/np is read from; the table; passing rationumberneutronstoprotons; overrides it; P3 = 0.5449 at 0.1 MeV, 0.2993 at 20 MeV; P = 0.666, 0.092, 0.259 in about 2 s; 6.14.1 Varying the production point; P = 0.389, against 0.375 from the center; 6.14.2 Probability averaging; 6.14.3 Comparing solar models; 6.14.4 Comparing against approximations; Fold theta13 back in; 6.15 A supernova shock front; UnmarkedDiscontinuityWarning: the; Hamiltonian is discontinuous at the scale; of the grid this scan builds, and no; tbreakpoints were given. A slab; straddling a density jump degrades the; quadrature no matter how many slabs are; used; pass tbreakpoints at the; discontinuities.; Both fronts, 0.07 km wide, centered on the contact; discontinuity and on the forward shock: declare; where each begins and ends; 4000 baselines from 10,200 to 80,000 km, at one; energy: the cumulative engine answers; One and two sterile states; PhaseAveragingWarning: average=True on a; profile with discontinuities has no closed; form, so the probability was propagated; across an energy window of +/-10.0% and; averaged over 41 samples. This is the; average over that window, not the L/E ->; infinity limit, and it depends on the; window: the largest standard error of the; mean here is 4.47e-02.; Pass averagespread to set the half-width; of the window to the resolution of the; measurement, and averagensamples to; reduce the error, which falls as the; inverse square root of the number of; samples.; P = 0.56; P = 0.584; 6.16 High-energy astrophysical neutrinos; The standard case at 1 Mpc and at 1 Gpc; Largest difference over all energies; and channels: none; PhaseAveragingWarning: depends on; the energy spread at 46 of 60 points; [0.333 0.666 0.001]; A source 100 Mpc away: far enough for every phase; to average away at every energy. The average; depends on the baseline through the phases; Standard: the averaged matrix at each energy; LIV: the n = 1 operator, aligned with the PMNS; angles, sized to equal the vacuum term at 100 TeV; NSI through the Earth: the flux arrives decohered,; then crosses the Earth coherently, so the two legs; compose as probability matrices. The ladder starts; at 32 slabs: the matter term alone winds more than; pi across a coarser one; 3+1: one sterile state, sin^2 of both active-; sterile angles 0.1, Dm41^2 = 1 eV^2; Pseudo-Dirac: the second mass state is paired; with a partner split by 1e-13 eV^2; from 100 Mpc, as in the fifth panel; splittings of nu2, at 100 TeV and; 100 Mpc, from coherent to decohered; [0.326 0.343 0.33 ] Coherent: standard; [0.328 0.338 0.334] Between: warns; [0.332 0.323 0.345] Decohered; Inputs of the vacuum part: the PMNS; matrix and the three m^2 [eV^2]; PREM along the chord through the core:; at position l, VCC and the neutron-; to-proton ratio of the local layer; The Hamiltonian, of energy and position:; vacuum part plus matter part; Through the core, with the PREM layer; boundaries declared as breakpoints; The averaged vacuum leg, then the Earth; [0.332 0.323 0.345]; 6.17 Assorted long examples; 6.17.1 A cavity in the Earth’s crust; An empty crust needs no profile, only a number; Water, iron, a mineral deposit, a zone of faults:; (rho, Ye, w) for each; 6.17.2 Geoneutrinos; The mantle chord: cosine of the zenith; angle at the detector, and the depths of; the production point and of Borexino [km]; Length of the chord [km]; Density along the chord [g/cm^3]: PREM,; with its ocean layer replaced by rock; Radius at distance l along the chord; Phase-averaged survival probability in; matter, at electron fraction of the mantle; 0.551; The same, in vacuum; 0.548; Twelve energies per cycle of the pair split by; Dm31^2 on the longest chord, 7,300 km; Borexino, 1.4 km underground. A production point; enters as its depth and the cosine of its zenith; angle at the detector; the call declares the PREM; boundaries on the chord by itself. The oscillation; parameters are the defaults, NuFIT 6.1.; The local point, 10 km deep and 100 km away,; run in reverse, from the detector (see text); 6.17.3 Neutrino tomography of the Sun; The chord above: b = 0.3 Rsun; 6.17.4 Astrophysical relativistic jet; Evolve from the jet head (L0) to the; surface (L); also return the operator; Probabilities and evolution operator; Weights, w[i, alpha]; The three envelopes, g/cm3, r in cm; Production points spread over one oscillation; length at the jet head; Drawn: Psmooth[:, gd.NUE, gd.NUE], and the others; 6.17.5 Long-range interactions in the Sun; Average over them, then scale by the 8B; spectrum and total flux, cm^-2 s^-1 MeV^-1; The B16-GS98 model that ships with Magnus,; on its tabulated radii and on to 20 Rsun; Two mediator ranges, one coupling: g’^2 makes Vemu; a tenth of VCC at R0 when 1/m = Rsun; The standard part from the shipped builders; the; new term is the only line written here; 6.17.6 Turbulent matter profile; H at the mean density: vacuum plus matter; Eigenvalues w, matter eigenvectors Vm, and; the gap the mode has to match; Coupling of levels 2 and 3, and the length; of complete conversion, in units of LOSC; VCC follows the density; One operator per slab, from 9 nodes at; Magnus order 6; Ordered product, earliest slab first; Rotate to the matter basis, then read the; transition between levels 2 and 3; Two flavors: the 1-3 sector; Three flavors; The same call with a sterile state added; 6.18 Running a scan in parallel; Every point has its own baseline, so no; batched engine takes this scan; engine takes it, at the default njobs = 1; nothing; the call runs in one process; 6.19 Using Mag$nu$s from the command line; 7 Performance, precision, and reach; 7.1 Timing measurement protocol; 7.2 Slab width follows the profile, not the phase; 7.3 What a probability costs; 7.4 Precision, accuracy, and tolerance; 7.5 Accuracy and speed: smooth density profile; 7.6 Accuracy and speed: a supernova shock front; 7.7 Accuracy and speed: the Sun; 8 Comparison to other codes; 8.1 Relation to other oscillation codes; 8.2 Speed-accuracy comparison; 8.3 When to use Mag$nu$s; 9 Conclusions; Declaration of generative AI and AI-assisted technologies in the manuscript preparation process; Acknowledgments; References; Appendix A Magnus expansion at any order; A.1 Computing Magnus terms; A.2 Why the collocation orders are even; Appendix B Predefined oscillation-parameter sets; Appendix C The tutorial notebooks; Appendix D The test suite; Instructions for reporting errorsComplete, revision-matched arXiv TeX source package
sha256:acef7efe9a4338f4...
⏳ 未决核查待办
- 未进行独立同行评审或逐项复算。
- 源文件中的图像未进行视觉核对。
原始摘要与来源Original abstract and source
阅读范围与依据Coverage and evidence
实际阅读范围：已检查正文中支撑判断的指定章节；版本 arXiv:2610.07159v1；检查材料：Introduction、Mag$nu$s: neutrino oscillation probabilities for any Hermitian Hamiltonian, any number of flavors, and any matter profile、1 Introduction、2 Synopsis、Standard oscillation parameters、Active-sterile mixing、Exponentially falling density: 3e3 g/cm^3 at the、origin, scale height 10 km, 25 km baseline.、Every mixing parameter left out defaults to zero.、3 Neutrino oscillations: units, notation, conventions、4 The method、4.1 History and context、4.2 Where the method applies and where it does not、4.3 The Magnus expansion、4.4 Truncation is exactly unitary、4.5 Two quadrature rules for the Magnus expansion、4.6 Slab composition、4.7 Tolerance and adaptive refinement、4.8 The matrix exponential、4.9 Adiabatic transport with exact Magnus patches、4.10 The phase-averaged probability、4.10.1 The averaged probability、4.10.2 Coherent blocks、4.10.3 The phase average、4.10.4 A varying Hamiltonian、5 Code description、5.1 Scope and interface、5.2 The Mag$nu$s architecture、5.3 The seven engines、5.4 Example: one request, through the layers、---- What the user writes: the whole request ----、One energy, one chord, and two refinement keywords、---- Inside Magnus: what the call runs through ----、Top layer: the named wrapper、Second layer: builds H, tries its engines、Third layer: the scan over (energy, baseline)、Base layer: the refinement ladder, one point、5.5 Batching and parallelization、5.6 Declaring the structure of a profile、24 slabs of 250 km, alternating 2 and 8 g/cm^3、The jumps declared: the ladder keeps its own grid,、adds an edge at each of the 23 jumps, then refines、to the tolerance、The grid given in full, as [start, end] pairs: one、slab per layer, no refinement, a single evaluation、A grid of 12 slabs, each straddling one jump: the、user owns this grid, and nothing repairs it、P1 = P2 = 0.0240046, exact: the profile is constant、within every slab. P3 = 0.0246158, off by 6e-4、5.7 Documentation and executable examples、5.8 Installation、The package:、Plus what the notebooks need:、Plus what the tests need:、5.9 Code availability、5.10 Code validation、5.10.1 Against the derivation、5.10.2 Against closed forms、5.10.3 Against an independent solver、5.10.4 Against itself, where meaningful、5.10.5 Against a population、5.10.6 Continuous integration、5.11 Limitations、5.11.1 Limitations of the method、5.11.2 Limitations of the implementation、6 Mag$nu$s usage and examples、6.1 The full application programming interface (API)、6.2 A minimal example in vacuum, step-by-step、Natural units: energies in eV, lengths in eV^-1、Pme = 0.03127, Pmm = 0.39231, Pmt = 0.57643、The amplitude is indexed the other way、6.3 Choosing the oscillation parameters、6.3.1 Three flavors、Pme = 0.05166, against 0.03127 by default、Pme = 0.05579; ’LEM’ gives 0.05558、Pme = 0.04998; ’freefluxesrsbl’、gives 0.04552、53 lines, e.g.,、OSCPARAMSNUFIT10IO | NuFIT 1.0, IO、The releases the loader reads, and the、categories of each、18 lines, e.g., NuFIT 2.1 [’LEM’, ’LID’]、Pme = 0.05098; the other five stay at、their defaults、The other five stay at their NuFIT 6.1 IO、values. Pme = 0.01800, against、0.05241 with the set’s own dCP、6.3.2 Specifying angle conventions、s12 reads 0.5557, 0.3088, 0.5892, 33.759、and Pme = 0.03127 in all four cases、6.3.3 Two flavors、P[0][1] = 0.013089、The same point, angles in degrees、6.3.4 Four flavors、Pme = 0.02801, Pmm = 0.39958、6.3.5 Five flavors、Pme = 0.03984, Pmm = 0.40684、6.3.6 Default non-standard parameters、6.4 Batched calls、P[-1] equals the single call at 1300 km、6.5 Constant density、’NuFIT 6.1’ is the default; re-loaded for clarity only、Three flavors, one energy and one baseline、0.0 g/cm3 Pme = 0.003773、3.0 g/cm3 Pme = 0.013589、8.0 g/cm3 Pme = 0.050497、13.0 g/cm3 Pme = 0.113343、Two flavors: one angle and one splitting、P[0][1] = 0.005766、Two hundred energies in one call、P.shape = (200, 3, 3)、6.6 Using wrappers vs. scenario calls vs. direct calls、--- 1. With the named wrappers、Pme = 0.00377306 in vacuum, 0.01358909 in、matter, in 0.15 ms、Pscan.shape = (200, 3, 3), in 0.13 ms、--- 2. With a scenario function、The same number, to 1e-16, in 3.6 ms、--- 3. With a Hamiltonian built by hand、The same two numbers, in 0.043 ms、The same 200 points, in 5.8 ms、6.6.1 Through a wrapper、Pme = 0.017126. Dropping epsem gives、0.013589, the standard wrapper’s answer、6.6.2 Through a scenario function、Pme = 0.017126, the same number the、wrapper above returns、6.6.3 Through a direct call、proj for three flavors is diag(1, 0, 0)、1. mixing, 2. matter, 3. new physics、Pme = 0.017126 again、6.6.4 Which to use、6.6.5 Returning the evolution operator、P is the matrix the call returns without、the keyword; U is complex and unitary.、Both of these hold:、the second denser than the first、6.6.6 Asking for the averaged probability、P[0, 0] = 0.5481, P[1, 0] = 0.2155: the、familiar sumi |Uei|^2 |Umi|^2, from the、eigenvectors alone、PhaseAveragingWarning: the phase-averaged、probability depends on the energy spread、at 1 of 1 (energy, L) point(s) ...、P[0, 0] = 0.9844, P[1, 1] = 0.9088;、P[0, 0] = 0.8934: the decohered limit,、now from the eigenvectors in matter、6.7 The Hamiltonians、6.7.1 Shipped Hamiltonians、Over 200 energies: 3.7 ms、... and 0.18 ms、6.7.2 Building a new Hamiltonian from scratch、Pme = 0.021309、Pme = 0.004865、6.8 Pre-packaged plotting routines、6.9 Probability vs. energy and distance、0.082635, as in the listing、0.071325, as in the listing、One mole of electrons per cubic centimeter, in the、natural units that magnus works in、Give a constant profile as a single number, and、a varying one as a function of the distance、traveled, returning the electron number density、there. The distance arrives in eV^-1, like、every length in magnus、The first keyword says that the functions return、an electron number density, not a mass density、Left column: 5,000 distances at one energy. The、list of distances goes where a single one would、Right column: 3,000 energies at one distance、The vacuum curves, for reference、6.10 Layered matter、The slab edges are declared, so no slab straddles a、jump; within each, one term of the expansion is exact、6.11 Bi-probability plot、DUNE: 2 GeV, 1300 km; a crust-like 3 g/cm^3、With the shipped plotter、With matplotlib directly、6.12 The Earth、6.12.1 Neutrinos & antineutrinos, 2–5 flavors、Two flavors is the (nue, numu) reduction of the 1-3、sector, so it takes that one angle and splitting by、name. Every wrapper above two flavors defaults to、the same NuFIT 6.1 set, so it needs none passed.、The sterile pairs run three decades higher, where an、eV-scale splitting places its resonance、6.12.2 Probabilities between two locations、positive, south and west negative、Two site names replace costhz and L: the chord、between them is the baseline、The chord itself, in km, for a legend、6.12.3 New physics、Three decades up is where an eV-scale splitting、places its matter resonance、The same call each time, a different extra term、Change between 3nu and 3+1 survival probability、6.12.4 Earth oscillograms、With the shipped plotter, one call per panel、Or with matplotlib directly, one axes per panel、6.12.5 Underground sources and detectors、On the surface, the baseline is given、0.911587、2 km down, the depth fixes the baseline、0.911600、Horizontal: no path at all at the surface、1.000000、2 km down it crosses 160 km of rock、0.997560、0.020930、Under ice, as at IceCube、Under rock, as at SNOLAB、0.021325、6.12.6 Using a custom Earth density profile、Four shells for PREM’s ten, each at the mass-weighted、mean density of the region it replaces、Those four edges are PREM boundaries, so their、crossings come straight out of the helper、6.13 Strategy report after a call、[’certified’, ’declined’, ’engine’,、’family’, ’hiddenfeature’, ’sampling’,、’trace’]、3227.2、[(’hybrid’, ’the profile is not、resolved at the probe scale’)]、6.14 The Sun、The table ships with Magnus. Its columns are read、only to plot the density (top panel of the figure)、The solar spectrum, 0.1 to 20 MeV、Shared by the four calls: the model by name, from、the center to its last row, nue to nue, the average、Standard three-flavor oscillations、The same, with non-standard interactions、One and two sterile states. nn/np is read from、the table; passing rationumberneutronstoprotons、overrides it、P3 = 0.5449 at 0.1 MeV, 0.2993 at 20 MeV、P = 0.666, 0.092, 0.259 in about 2 s、6.14.1 Varying the production point、P = 0.389, against 0.375 from the center、6.14.2 Probability averaging、6.14.3 Comparing solar models、6.14.4 Comparing against approximations、Fold theta13 back in、6.15 A supernova shock front、UnmarkedDiscontinuityWarning: the、Hamiltonian is discontinuous at the scale、of the grid this scan builds, and no、tbreakpoints were given. A slab、straddling a density jump degrades the、quadrature no matter how many slabs are、used; pass tbreakpoints at the、discontinuities.、Both fronts, 0.07 km wide, centered on the contact、discontinuity and on the forward shock: declare、where each begins and ends、4000 baselines from 10,200 to 80,000 km, at one、energy: the cumulative engine answers、One and two sterile states、PhaseAveragingWarning: average=True on a、profile with discontinuities has no closed、form, so the probability was propagated、across an energy window of +/-10.0% and、averaged over 41 samples. This is the、average over that window, not the L/E ->、infinity limit, and it depends on the、window: the largest standard error of the、mean here is 4.47e-02.、Pass averagespread to set the half-width、of the window to the resolution of the、measurement, and averagensamples to、reduce the error, which falls as the、inverse square root of the number of、samples.、P = 0.56、P = 0.584、6.16 High-energy astrophysical neutrinos、The standard case at 1 Mpc and at 1 Gpc、Largest difference over all energies、and channels: none、PhaseAveragingWarning: depends on、the energy spread at 46 of 60 points、[0.333 0.666 0.001]、A source 100 Mpc away: far enough for every phase、to average away at every energy. The average、depends on the baseline through the phases、Standard: the averaged matrix at each energy、LIV: the n = 1 operator, aligned with the PMNS、angles, sized to equal the vacuum term at 100 TeV、NSI through the Earth: the flux arrives decohered,、then crosses the Earth coherently, so the two legs、compose as probability matrices. The ladder starts、at 32 slabs: the matter term alone winds more than、pi across a coarser one、3+1: one sterile state, sin^2 of both active-、sterile angles 0.1, Dm41^2 = 1 eV^2、Pseudo-Dirac: the second mass state is paired、with a partner split by 1e-13 eV^2、from 100 Mpc, as in the fifth panel、splittings of nu2, at 100 TeV and、100 Mpc, from coherent to decohered、[0.326 0.343 0.33 ] Coherent: standard、[0.328 0.338 0.334] Between: warns、[0.332 0.323 0.345] Decohered、Inputs of the vacuum part: the PMNS、matrix and the three m^2 [eV^2]、PREM along the chord through the core:、at position l, VCC and the neutron-、to-proton ratio of the local layer、The Hamiltonian, of energy and position:、vacuum part plus matter part、Through the core, with the PREM layer、boundaries declared as breakpoints、The averaged vacuum leg, then the Earth、[0.332 0.323 0.345]、6.17 Assorted long examples、6.17.1 A cavity in the Earth’s crust、An empty crust needs no profile, only a number、Water, iron, a mineral deposit, a zone of faults:、(rho, Ye, w) for each、6.17.2 Geoneutrinos、The mantle chord: cosine of the zenith、angle at the detector, and the depths of、the production point and of Borexino [km]、Length of the chord [km]、Density along the chord [g/cm^3]: PREM,、with its ocean layer replaced by rock、Radius at distance l along the chord、Phase-averaged survival probability in、matter, at electron fraction of the mantle、0.551、The same, in vacuum、0.548、Twelve energies per cycle of the pair split by、Dm31^2 on the longest chord, 7,300 km、Borexino, 1.4 km underground. A production point、enters as its depth and the cosine of its zenith、angle at the detector; the call declares the PREM、boundaries on the chord by itself. The oscillation、parameters are the defaults, NuFIT 6.1.、The local point, 10 km deep and 100 km away,、run in reverse, from the detector (see text)、6.17.3 Neutrino tomography of the Sun、The chord above: b = 0.3 Rsun、6.17.4 Astrophysical relativistic jet、Evolve from the jet head (L0) to the、surface (L); also return the operator、Probabilities and evolution operator、Weights, w[i, alpha]、The three envelopes, g/cm3, r in cm、Production points spread over one oscillation、length at the jet head、Drawn: Psmooth[:, gd.NUE, gd.NUE], and the others、6.17.5 Long-range interactions in the Sun、Average over them, then scale by the 8B、spectrum and total flux, cm^-2 s^-1 MeV^-1、The B16-GS98 model that ships with Magnus,、on its tabulated radii and on to 20 Rsun、Two mediator ranges, one coupling: g’^2 makes Vemu、a tenth of VCC at R0 when 1/m = Rsun、The standard part from the shipped builders; the、new term is the only line written here、6.17.6 Turbulent matter profile、H at the mean density: vacuum plus matter、Eigenvalues w, matter eigenvectors Vm, and、the gap the mode has to match、Coupling of levels 2 and 3, and the length、of complete conversion, in units of LOSC、VCC follows the density、One operator per slab, from 9 nodes at、Magnus order 6、Ordered product, earliest slab first、Rotate to the matter basis, then read the、transition between levels 2 and 3、Two flavors: the 1-3 sector、Three flavors、The same call with a sterile state added、6.18 Running a scan in parallel、Every point has its own baseline, so no、batched engine takes this scan、engine takes it, at the default njobs = 1、nothing; the call runs in one process、6.19 Using Mag$nu$s from the command line、7 Performance, precision, and reach、7.1 Timing measurement protocol、7.2 Slab width follows the profile, not the phase、7.3 What a probability costs、7.4 Precision, accuracy, and tolerance、7.5 Accuracy and speed: smooth density profile、7.6 Accuracy and speed: a supernova shock front、7.7 Accuracy and speed: the Sun、8 Comparison to other codes、8.1 Relation to other oscillation codes、8.2 Speed-accuracy comparison、8.3 When to use Mag$nu$s、9 Conclusions、Declaration of generative AI and AI-assisted technologies in the manuscript preparation process、Acknowledgments、References、Appendix A Magnus expansion at any order、A.1 Computing Magnus terms、A.2 Why the collocation orders are even、Appendix B Predefined oscillation-parameter sets、Appendix C The tutorial notebooks、Appendix D The test suite、Instructions for reporting errors。
This guide uses the recorded body material; inspected sections: Introduction, Mag$nu$s: neutrino oscillation probabilities for any Hermitian Hamiltonian, any number of flavors, and any matter profile, 1 Introduction, 2 Synopsis, Standard oscillation parameters, Active-sterile mixing, Exponentially falling density: 3e3 g/cm^3 at the, origin, scale height 10 km, 25 km baseline., Every mixing parameter left out defaults to zero., 3 Neutrino oscillations: units, notation, conventions, 4 The method, 4.1 History and context, 4.2 Where the method applies and where it does not, 4.3 The Magnus expansion, 4.4 Truncation is exactly unitary, 4.5 Two quadrature rules for the Magnus expansion, 4.6 Slab composition, 4.7 Tolerance and adaptive refinement, 4.8 The matrix exponential, 4.9 Adiabatic transport with exact Magnus patches, 4.10 The phase-averaged probability, 4.10.1 The averaged probability, 4.10.2 Coherent blocks, 4.10.3 The phase average, 4.10.4 A varying Hamiltonian, 5 Code description, 5.1 Scope and interface, 5.2 The Mag$nu$s architecture, 5.3 The seven engines, 5.4 Example: one request, through the layers, ---- What the user writes: the whole request ----, One energy, one chord, and two refinement keywords, ---- Inside Magnus: what the call runs through ----, Top layer: the named wrapper, Second layer: builds H, tries its engines, Third layer: the scan over (energy, baseline), Base layer: the refinement ladder, one point, 5.5 Batching and parallelization, 5.6 Declaring the structure of a profile, 24 slabs of 250 km, alternating 2 and 8 g/cm^3, The jumps declared: the ladder keeps its own grid,, adds an edge at each of the 23 jumps, then refines, to the tolerance, The grid given in full, as [start, end] pairs: one, slab per layer, no refinement, a single evaluation, A grid of 12 slabs, each straddling one jump: the, user owns this grid, and nothing repairs it, P1 = P2 = 0.0240046, exact: the profile is constant, within every slab. P3 = 0.0246158, off by 6e-4, 5.7 Documentation and executable examples, 5.8 Installation, The package:, Plus what the notebooks need:, Plus what the tests need:, 5.9 Code availability, 5.10 Code validation, 5.10.1 Against the derivation, 5.10.2 Against closed forms, 5.10.3 Against an independent solver, 5.10.4 Against itself, where meaningful, 5.10.5 Against a population, 5.10.6 Continuous integration, 5.11 Limitations, 5.11.1 Limitations of the method, 5.11.2 Limitations of the implementation, 6 Mag$nu$s usage and examples, 6.1 The full application programming interface (API), 6.2 A minimal example in vacuum, step-by-step, Natural units: energies in eV, lengths in eV^-1, Pme = 0.03127, Pmm = 0.39231, Pmt = 0.57643, The amplitude is indexed the other way, 6.3 Choosing the oscillation parameters, 6.3.1 Three flavors, Pme = 0.05166, against 0.03127 by default, Pme = 0.05579; ’LEM’ gives 0.05558, Pme = 0.04998; ’freefluxesrsbl’, gives 0.04552, 53 lines, e.g.,, OSCPARAMSNUFIT10IO | NuFIT 1.0, IO, The releases the loader reads, and the, categories of each, 18 lines, e.g., NuFIT 2.1 [’LEM’, ’LID’], Pme = 0.05098; the other five stay at, their defaults, The other five stay at their NuFIT 6.1 IO, values. Pme = 0.01800, against, 0.05241 with the set’s own dCP, 6.3.2 Specifying angle conventions, s12 reads 0.5557, 0.3088, 0.5892, 33.759, and Pme = 0.03127 in all four cases, 6.3.3 Two flavors, P[0][1] = 0.013089, The same point, angles in degrees, 6.3.4 Four flavors, Pme = 0.02801, Pmm = 0.39958, 6.3.5 Five flavors, Pme = 0.03984, Pmm = 0.40684, 6.3.6 Default non-standard parameters, 6.4 Batched calls, P[-1] equals the single call at 1300 km, 6.5 Constant density, ’NuFIT 6.1’ is the default; re-loaded for clarity only, Three flavors, one energy and one baseline, 0.0 g/cm3 Pme = 0.003773, 3.0 g/cm3 Pme = 0.013589, 8.0 g/cm3 Pme = 0.050497, 13.0 g/cm3 Pme = 0.113343, Two flavors: one angle and one splitting, P[0][1] = 0.005766, Two hundred energies in one call, P.shape = (200, 3, 3), 6.6 Using wrappers vs. scenario calls vs. direct calls, --- 1. With the named wrappers, Pme = 0.00377306 in vacuum, 0.01358909 in, matter, in 0.15 ms, Pscan.shape = (200, 3, 3), in 0.13 ms, --- 2. With a scenario function, The same number, to 1e-16, in 3.6 ms, --- 3. With a Hamiltonian built by hand, The same two numbers, in 0.043 ms, The same 200 points, in 5.8 ms, 6.6.1 Through a wrapper, Pme = 0.017126. Dropping epsem gives, 0.013589, the standard wrapper’s answer, 6.6.2 Through a scenario function, Pme = 0.017126, the same number the, wrapper above returns, 6.6.3 Through a direct call, proj for three flavors is diag(1, 0, 0), 1. mixing, 2. matter, 3. new physics, Pme = 0.017126 again, 6.6.4 Which to use, 6.6.5 Returning the evolution operator, P is the matrix the call returns without, the keyword; U is complex and unitary., Both of these hold:, the second denser than the first, 6.6.6 Asking for the averaged probability, P[0, 0] = 0.5481, P[1, 0] = 0.2155: the, familiar sumi |Uei|^2 |Umi|^2, from the, eigenvectors alone, PhaseAveragingWarning: the phase-averaged, probability depends on the energy spread, at 1 of 1 (energy, L) point(s) ..., P[0, 0] = 0.9844, P[1, 1] = 0.9088;, P[0, 0] = 0.8934: the decohered limit,, now from the eigenvectors in matter, 6.7 The Hamiltonians, 6.7.1 Shipped Hamiltonians, Over 200 energies: 3.7 ms, ... and 0.18 ms, 6.7.2 Building a new Hamiltonian from scratch, Pme = 0.021309, Pme = 0.004865, 6.8 Pre-packaged plotting routines, 6.9 Probability vs. energy and distance, 0.082635, as in the listing, 0.071325, as in the listing, One mole of electrons per cubic centimeter, in the, natural units that magnus works in, Give a constant profile as a single number, and, a varying one as a function of the distance, traveled, returning the electron number density, there. The distance arrives in eV^-1, like, every length in magnus, The first keyword says that the functions return, an electron number density, not a mass density, Left column: 5,000 distances at one energy. The, list of distances goes where a single one would, Right column: 3,000 energies at one distance, The vacuum curves, for reference, 6.10 Layered matter, The slab edges are declared, so no slab straddles a, jump; within each, one term of the expansion is exact, 6.11 Bi-probability plot, DUNE: 2 GeV, 1300 km; a crust-like 3 g/cm^3, With the shipped plotter, With matplotlib directly, 6.12 The Earth, 6.12.1 Neutrinos & antineutrinos, 2–5 flavors, Two flavors is the (nue, numu) reduction of the 1-3, sector, so it takes that one angle and splitting by, name. Every wrapper above two flavors defaults to, the same NuFIT 6.1 set, so it needs none passed., The sterile pairs run three decades higher, where an, eV-scale splitting places its resonance, 6.12.2 Probabilities between two locations, positive, south and west negative, Two site names replace costhz and L: the chord, between them is the baseline, The chord itself, in km, for a legend, 6.12.3 New physics, Three decades up is where an eV-scale splitting, places its matter resonance, The same call each time, a different extra term, Change between 3nu and 3+1 survival probability, 6.12.4 Earth oscillograms, With the shipped plotter, one call per panel, Or with matplotlib directly, one axes per panel, 6.12.5 Underground sources and detectors, On the surface, the baseline is given, 0.911587, 2 km down, the depth fixes the baseline, 0.911600, Horizontal: no path at all at the surface, 1.000000, 2 km down it crosses 160 km of rock, 0.997560, 0.020930, Under ice, as at IceCube, Under rock, as at SNOLAB, 0.021325, 6.12.6 Using a custom Earth density profile, Four shells for PREM’s ten, each at the mass-weighted, mean density of the region it replaces, Those four edges are PREM boundaries, so their, crossings come straight out of the helper, 6.13 Strategy report after a call, [’certified’, ’declined’, ’engine’,, ’family’, ’hiddenfeature’, ’sampling’,, ’trace’], 3227.2, [(’hybrid’, ’the profile is not, resolved at the probe scale’)], 6.14 The Sun, The table ships with Magnus. Its columns are read, only to plot the density (top panel of the figure), The solar spectrum, 0.1 to 20 MeV, Shared by the four calls: the model by name, from, the center to its last row, nue to nue, the average, Standard three-flavor oscillations, The same, with non-standard interactions, One and two sterile states. nn/np is read from, the table; passing rationumberneutronstoprotons, overrides it, P3 = 0.5449 at 0.1 MeV, 0.2993 at 20 MeV, P = 0.666, 0.092, 0.259 in about 2 s, 6.14.1 Varying the production point, P = 0.389, against 0.375 from the center, 6.14.2 Probability averaging, 6.14.3 Comparing solar models, 6.14.4 Comparing against approximations, Fold theta13 back in, 6.15 A supernova shock front, UnmarkedDiscontinuityWarning: the, Hamiltonian is discontinuous at the scale, of the grid this scan builds, and no, tbreakpoints were given. A slab, straddling a density jump degrades the, quadrature no matter how many slabs are, used; pass tbreakpoints at the, discontinuities., Both fronts, 0.07 km wide, centered on the contact, discontinuity and on the forward shock: declare, where each begins and ends, 4000 baselines from 10,200 to 80,000 km, at one, energy: the cumulative engine answers, One and two sterile states, PhaseAveragingWarning: average=True on a, profile with discontinuities has no closed, form, so the probability was propagated, across an energy window of +/-10.0% and, averaged over 41 samples. This is the, average over that window, not the L/E ->, infinity limit, and it depends on the, window: the largest standard error of the, mean here is 4.47e-02., Pass averagespread to set the half-width, of the window to the resolution of the, measurement, and averagensamples to, reduce the error, which falls as the, inverse square root of the number of, samples., P = 0.56, P = 0.584, 6.16 High-energy astrophysical neutrinos, The standard case at 1 Mpc and at 1 Gpc, Largest difference over all energies, and channels: none, PhaseAveragingWarning: depends on, the energy spread at 46 of 60 points, [0.333 0.666 0.001], A source 100 Mpc away: far enough for every phase, to average away at every energy. The average, depends on the baseline through the phases, Standard: the averaged matrix at each energy, LIV: the n = 1 operator, aligned with the PMNS, angles, sized to equal the vacuum term at 100 TeV, NSI through the Earth: the flux arrives decohered,, then crosses the Earth coherently, so the two legs, compose as probability matrices. The ladder starts, at 32 slabs: the matter term alone winds more than, pi across a coarser one, 3+1: one sterile state, sin^2 of both active-, sterile angles 0.1, Dm41^2 = 1 eV^2, Pseudo-Dirac: the second mass state is paired, with a partner split by 1e-13 eV^2, from 100 Mpc, as in the fifth panel, splittings of nu2, at 100 TeV and, 100 Mpc, from coherent to decohered, [0.326 0.343 0.33 ] Coherent: standard, [0.328 0.338 0.334] Between: warns, [0.332 0.323 0.345] Decohered, Inputs of the vacuum part: the PMNS, matrix and the three m^2 [eV^2], PREM along the chord through the core:, at position l, VCC and the neutron-, to-proton ratio of the local layer, The Hamiltonian, of energy and position:, vacuum part plus matter part, Through the core, with the PREM layer, boundaries declared as breakpoints, The averaged vacuum leg, then the Earth, [0.332 0.323 0.345], 6.17 Assorted long examples, 6.17.1 A cavity in the Earth’s crust, An empty crust needs no profile, only a number, Water, iron, a mineral deposit, a zone of faults:, (rho, Ye, w) for each, 6.17.2 Geoneutrinos, The mantle chord: cosine of the zenith, angle at the detector, and the depths of, the production point and of Borexino [km], Length of the chord [km], Density along the chord [g/cm^3]: PREM,, with its ocean layer replaced by rock, Radius at distance l along the chord, Phase-averaged survival probability in, matter, at electron fraction of the mantle, 0.551, The same, in vacuum, 0.548, Twelve energies per cycle of the pair split by, Dm31^2 on the longest chord, 7,300 km, Borexino, 1.4 km underground. A production point, enters as its depth and the cosine of its zenith, angle at the detector; the call declares the PREM, boundaries on the chord by itself. The oscillation, parameters are the defaults, NuFIT 6.1., The local point, 10 km deep and 100 km away,, run in reverse, from the detector (see text), 6.17.3 Neutrino tomography of the Sun, The chord above: b = 0.3 Rsun, 6.17.4 Astrophysical relativistic jet, Evolve from the jet head (L0) to the, surface (L); also return the operator, Probabilities and evolution operator, Weights, w[i, alpha], The three envelopes, g/cm3, r in cm, Production points spread over one oscillation, length at the jet head, Drawn: Psmooth[:, gd.NUE, gd.NUE], and the others, 6.17.5 Long-range interactions in the Sun, Average over them, then scale by the 8B, spectrum and total flux, cm^-2 s^-1 MeV^-1, The B16-GS98 model that ships with Magnus,, on its tabulated radii and on to 20 Rsun, Two mediator ranges, one coupling: g’^2 makes Vemu, a tenth of VCC at R0 when 1/m = Rsun, The standard part from the shipped builders; the, new term is the only line written here, 6.17.6 Turbulent matter profile, H at the mean density: vacuum plus matter, Eigenvalues w, matter eigenvectors Vm, and, the gap the mode has to match, Coupling of levels 2 and 3, and the length, of complete conversion, in units of LOSC, VCC follows the density, One operator per slab, from 9 nodes at, Magnus order 6, Ordered product, earliest slab first, Rotate to the matter basis, then read the, transition between levels 2 and 3, Two flavors: the 1-3 sector, Three flavors, The same call with a sterile state added, 6.18 Running a scan in parallel, Every point has its own baseline, so no, batched engine takes this scan, engine takes it, at the default njobs = 1, nothing; the call runs in one process, 6.19 Using Mag$nu$s from the command line, 7 Performance, precision, and reach, 7.1 Timing measurement protocol, 7.2 Slab width follows the profile, not the phase, 7.3 What a probability costs, 7.4 Precision, accuracy, and tolerance, 7.5 Accuracy and speed: smooth density profile, 7.6 Accuracy and speed: a supernova shock front, 7.7 Accuracy and speed: the Sun, 8 Comparison to other codes, 8.1 Relation to other oscillation codes, 8.2 Speed-accuracy comparison, 8.3 When to use Mag$nu$s, 9 Conclusions, Declaration of generative AI and AI-assisted technologies in the manuscript preparation process, Acknowledgments, References, Appendix A Magnus expansion at any order, A.1 Computing Magnus terms, A.2 Why the collocation orders are even, Appendix B Predefined oscillation-parameter sets, Appendix C The tutorial notebooks, Appendix D The test suite, Instructions for reporting errors.
- arXiv body material Introduction; Mag$nu$s: neutrino oscillation probabilities for any Hermitian Hamiltonian, any number of flavors, and any matter profile; 1 Introduction; 2 Synopsis; Standard oscillation parameters; Active-sterile mixing; Exponentially falling density: 3e3 g/cm^3 at the; origin, scale height 10 km, 25 km baseline.; Every mixing parameter left out defaults to zero.; 3 Neutrino oscillations: units, notation, conventions; 4 The method; 4.1 History and context; 4.2 Where the method applies and where it does not; 4.3 The Magnus expansion; 4.4 Truncation is exactly unitary; 4.5 Two quadrature rules for the Magnus expansion; 4.6 Slab composition; 4.7 Tolerance and adaptive refinement; 4.8 The matrix exponential; 4.9 Adiabatic transport with exact Magnus patches; 4.10 The phase-averaged probability; 4.10.1 The averaged probability; 4.10.2 Coherent blocks; 4.10.3 The phase average; 4.10.4 A varying Hamiltonian; 5 Code description; 5.1 Scope and interface; 5.2 The Mag$nu$s architecture; 5.3 The seven engines; 5.4 Example: one request, through the layers; ---- What the user writes: the whole request ----; One energy, one chord, and two refinement keywords; ---- Inside Magnus: what the call runs through ----; Top layer: the named wrapper; Second layer: builds H, tries its engines; Third layer: the scan over (energy, baseline); Base layer: the refinement ladder, one point; 5.5 Batching and parallelization; 5.6 Declaring the structure of a profile; 24 slabs of 250 km, alternating 2 and 8 g/cm^3; The jumps declared: the ladder keeps its own grid,; adds an edge at each of the 23 jumps, then refines; to the tolerance; The grid given in full, as [start, end] pairs: one; slab per layer, no refinement, a single evaluation; A grid of 12 slabs, each straddling one jump: the; user owns this grid, and nothing repairs it; P1 = P2 = 0.0240046, exact: the profile is constant; within every slab. P3 = 0.0246158, off by 6e-4; 5.7 Documentation and executable examples; 5.8 Installation; The package:; Plus what the notebooks need:; Plus what the tests need:; 5.9 Code availability; 5.10 Code validation; 5.10.1 Against the derivation; 5.10.2 Against closed forms; 5.10.3 Against an independent solver; 5.10.4 Against itself, where meaningful; 5.10.5 Against a population; 5.10.6 Continuous integration; 5.11 Limitations; 5.11.1 Limitations of the method; 5.11.2 Limitations of the implementation; 6 Mag$nu$s usage and examples; 6.1 The full application programming interface (API); 6.2 A minimal example in vacuum, step-by-step; Natural units: energies in eV, lengths in eV^-1; Pme = 0.03127, Pmm = 0.39231, Pmt = 0.57643; The amplitude is indexed the other way; 6.3 Choosing the oscillation parameters; 6.3.1 Three flavors; Pme = 0.05166, against 0.03127 by default; Pme = 0.05579; ’LEM’ gives 0.05558; Pme = 0.04998; ’freefluxesrsbl’; gives 0.04552; 53 lines, e.g.,; OSCPARAMSNUFIT10IO | NuFIT 1.0, IO; The releases the loader reads, and the; categories of each; 18 lines, e.g., NuFIT 2.1 [’LEM’, ’LID’]; Pme = 0.05098; the other five stay at; their defaults; The other five stay at their NuFIT 6.1 IO; values. Pme = 0.01800, against; 0.05241 with the set’s own dCP; 6.3.2 Specifying angle conventions; s12 reads 0.5557, 0.3088, 0.5892, 33.759; and Pme = 0.03127 in all four cases; 6.3.3 Two flavors; P[0][1] = 0.013089; The same point, angles in degrees; 6.3.4 Four flavors; Pme = 0.02801, Pmm = 0.39958; 6.3.5 Five flavors; Pme = 0.03984, Pmm = 0.40684; 6.3.6 Default non-standard parameters; 6.4 Batched calls; P[-1] equals the single call at 1300 km; 6.5 Constant density; ’NuFIT 6.1’ is the default; re-loaded for clarity only; Three flavors, one energy and one baseline; 0.0 g/cm3 Pme = 0.003773; 3.0 g/cm3 Pme = 0.013589; 8.0 g/cm3 Pme = 0.050497; 13.0 g/cm3 Pme = 0.113343; Two flavors: one angle and one splitting; P[0][1] = 0.005766; Two hundred energies in one call; P.shape = (200, 3, 3); 6.6 Using wrappers vs. scenario calls vs. direct calls; --- 1. With the named wrappers; Pme = 0.00377306 in vacuum, 0.01358909 in; matter, in 0.15 ms; Pscan.shape = (200, 3, 3), in 0.13 ms; --- 2. With a scenario function; The same number, to 1e-16, in 3.6 ms; --- 3. With a Hamiltonian built by hand; The same two numbers, in 0.043 ms; The same 200 points, in 5.8 ms; 6.6.1 Through a wrapper; Pme = 0.017126. Dropping epsem gives; 0.013589, the standard wrapper’s answer; 6.6.2 Through a scenario function; Pme = 0.017126, the same number the; wrapper above returns; 6.6.3 Through a direct call; proj for three flavors is diag(1, 0, 0); 1. mixing, 2. matter, 3. new physics; Pme = 0.017126 again; 6.6.4 Which to use; 6.6.5 Returning the evolution operator; P is the matrix the call returns without; the keyword; U is complex and unitary.; Both of these hold:; the second denser than the first; 6.6.6 Asking for the averaged probability; P[0, 0] = 0.5481, P[1, 0] = 0.2155: the; familiar sumi |Uei|^2 |Umi|^2, from the; eigenvectors alone; PhaseAveragingWarning: the phase-averaged; probability depends on the energy spread; at 1 of 1 (energy, L) point(s) ...; P[0, 0] = 0.9844, P[1, 1] = 0.9088;; P[0, 0] = 0.8934: the decohered limit,; now from the eigenvectors in matter; 6.7 The Hamiltonians; 6.7.1 Shipped Hamiltonians; Over 200 energies: 3.7 ms; ... and 0.18 ms; 6.7.2 Building a new Hamiltonian from scratch; Pme = 0.021309; Pme = 0.004865; 6.8 Pre-packaged plotting routines; 6.9 Probability vs. energy and distance; 0.082635, as in the listing; 0.071325, as in the listing; One mole of electrons per cubic centimeter, in the; natural units that magnus works in; Give a constant profile as a single number, and; a varying one as a function of the distance; traveled, returning the electron number density; there. The distance arrives in eV^-1, like; every length in magnus; The first keyword says that the functions return; an electron number density, not a mass density; Left column: 5,000 distances at one energy. The; list of distances goes where a single one would; Right column: 3,000 energies at one distance; The vacuum curves, for reference; 6.10 Layered matter; The slab edges are declared, so no slab straddles a; jump; within each, one term of the expansion is exact; 6.11 Bi-probability plot; DUNE: 2 GeV, 1300 km; a crust-like 3 g/cm^3; With the shipped plotter; With matplotlib directly; 6.12 The Earth; 6.12.1 Neutrinos & antineutrinos, 2–5 flavors; Two flavors is the (nue, numu) reduction of the 1-3; sector, so it takes that one angle and splitting by; name. Every wrapper above two flavors defaults to; the same NuFIT 6.1 set, so it needs none passed.; The sterile pairs run three decades higher, where an; eV-scale splitting places its resonance; 6.12.2 Probabilities between two locations; positive, south and west negative; Two site names replace costhz and L: the chord; between them is the baseline; The chord itself, in km, for a legend; 6.12.3 New physics; Three decades up is where an eV-scale splitting; places its matter resonance; The same call each time, a different extra term; Change between 3nu and 3+1 survival probability; 6.12.4 Earth oscillograms; With the shipped plotter, one call per panel; Or with matplotlib directly, one axes per panel; 6.12.5 Underground sources and detectors; On the surface, the baseline is given; 0.911587; 2 km down, the depth fixes the baseline; 0.911600; Horizontal: no path at all at the surface; 1.000000; 2 km down it crosses 160 km of rock; 0.997560; 0.020930; Under ice, as at IceCube; Under rock, as at SNOLAB; 0.021325; 6.12.6 Using a custom Earth density profile; Four shells for PREM’s ten, each at the mass-weighted; mean density of the region it replaces; Those four edges are PREM boundaries, so their; crossings come straight out of the helper; 6.13 Strategy report after a call; [’certified’, ’declined’, ’engine’,; ’family’, ’hiddenfeature’, ’sampling’,; ’trace’]; 3227.2; [(’hybrid’, ’the profile is not; resolved at the probe scale’)]; 6.14 The Sun; The table ships with Magnus. Its columns are read; only to plot the density (top panel of the figure); The solar spectrum, 0.1 to 20 MeV; Shared by the four calls: the model by name, from; the center to its last row, nue to nue, the average; Standard three-flavor oscillations; The same, with non-standard interactions; One and two sterile states. nn/np is read from; the table; passing rationumberneutronstoprotons; overrides it; P3 = 0.5449 at 0.1 MeV, 0.2993 at 20 MeV; P = 0.666, 0.092, 0.259 in about 2 s; 6.14.1 Varying the production point; P = 0.389, against 0.375 from the center; 6.14.2 Probability averaging; 6.14.3 Comparing solar models; 6.14.4 Comparing against approximations; Fold theta13 back in; 6.15 A supernova shock front; UnmarkedDiscontinuityWarning: the; Hamiltonian is discontinuous at the scale; of the grid this scan builds, and no; tbreakpoints were given. A slab; straddling a density jump degrades the; quadrature no matter how many slabs are; used; pass tbreakpoints at the; discontinuities.; Both fronts, 0.07 km wide, centered on the contact; discontinuity and on the forward shock: declare; where each begins and ends; 4000 baselines from 10,200 to 80,000 km, at one; energy: the cumulative engine answers; One and two sterile states; PhaseAveragingWarning: average=True on a; profile with discontinuities has no closed; form, so the probability was propagated; across an energy window of +/-10.0% and; averaged over 41 samples. This is the; average over that window, not the L/E ->; infinity limit, and it depends on the; window: the largest standard error of the; mean here is 4.47e-02.; Pass averagespread to set the half-width; of the window to the resolution of the; measurement, and averagensamples to; reduce the error, which falls as the; inverse square root of the number of; samples.; P = 0.56; P = 0.584; 6.16 High-energy astrophysical neutrinos; The standard case at 1 Mpc and at 1 Gpc; Largest difference over all energies; and channels: none; PhaseAveragingWarning: depends on; the energy spread at 46 of 60 points; [0.333 0.666 0.001]; A source 100 Mpc away: far enough for every phase; to average away at every energy. The average; depends on the baseline through the phases; Standard: the averaged matrix at each energy; LIV: the n = 1 operator, aligned with the PMNS; angles, sized to equal the vacuum term at 100 TeV; NSI through the Earth: the flux arrives decohered,; then crosses the Earth coherently, so the two legs; compose as probability matrices. The ladder starts; at 32 slabs: the matter term alone winds more than; pi across a coarser one; 3+1: one sterile state, sin^2 of both active-; sterile angles 0.1, Dm41^2 = 1 eV^2; Pseudo-Dirac: the second mass state is paired; with a partner split by 1e-13 eV^2; from 100 Mpc, as in the fifth panel; splittings of nu2, at 100 TeV and; 100 Mpc, from coherent to decohered; [0.326 0.343 0.33 ] Coherent: standard; [0.328 0.338 0.334] Between: warns; [0.332 0.323 0.345] Decohered; Inputs of the vacuum part: the PMNS; matrix and the three m^2 [eV^2]; PREM along the chord through the core:; at position l, VCC and the neutron-; to-proton ratio of the local layer; The Hamiltonian, of energy and position:; vacuum part plus matter part; Through the core, with the PREM layer; boundaries declared as breakpoints; The averaged vacuum leg, then the Earth; [0.332 0.323 0.345]; 6.17 Assorted long examples; 6.17.1 A cavity in the Earth’s crust; An empty crust needs no profile, only a number; Water, iron, a mineral deposit, a zone of faults:; (rho, Ye, w) for each; 6.17.2 Geoneutrinos; The mantle chord: cosine of the zenith; angle at the detector, and the depths of; the production point and of Borexino [km]; Length of the chord [km]; Density along the chord [g/cm^3]: PREM,; with its ocean layer replaced by rock; Radius at distance l along the chord; Phase-averaged survival probability in; matter, at electron fraction of the mantle; 0.551; The same, in vacuum; 0.548; Twelve energies per cycle of the pair split by; Dm31^2 on the longest chord, 7,300 km; Borexino, 1.4 km underground. A production point; enters as its depth and the cosine of its zenith; angle at the detector; the call declares the PREM; boundaries on the chord by itself. The oscillation; parameters are the defaults, NuFIT 6.1.; The local point, 10 km deep and 100 km away,; run in reverse, from the detector (see text); 6.17.3 Neutrino tomography of the Sun; The chord above: b = 0.3 Rsun; 6.17.4 Astrophysical relativistic jet; Evolve from the jet head (L0) to the; surface (L); also return the operator; Probabilities and evolution operator; Weights, w[i, alpha]; The three envelopes, g/cm3, r in cm; Production points spread over one oscillation; length at the jet head; Drawn: Psmooth[:, gd.NUE, gd.NUE], and the others; 6.17.5 Long-range interactions in the Sun; Average over them, then scale by the 8B; spectrum and total flux, cm^-2 s^-1 MeV^-1; The B16-GS98 model that ships with Magnus,; on its tabulated radii and on to 20 Rsun; Two mediator ranges, one coupling: g’^2 makes Vemu; a tenth of VCC at R0 when 1/m = Rsun; The standard part from the shipped builders; the; new term is the only line written here; 6.17.6 Turbulent matter profile; H at the mean density: vacuum plus matter; Eigenvalues w, matter eigenvectors Vm, and; the gap the mode has to match; Coupling of levels 2 and 3, and the length; of complete conversion, in units of LOSC; VCC follows the density; One operator per slab, from 9 nodes at; Magnus order 6; Ordered product, earliest slab first; Rotate to the matter basis, then read the; transition between levels 2 and 3; Two flavors: the 1-3 sector; Three flavors; The same call with a sterile state added; 6.18 Running a scan in parallel; Every point has its own baseline, so no; batched engine takes this scan; engine takes it, at the default njobs = 1; nothing; the call runs in one process; 6.19 Using Mag$nu$s from the command line; 7 Performance, precision, and reach; 7.1 Timing measurement protocol; 7.2 Slab width follows the profile, not the phase; 7.3 What a probability costs; 7.4 Precision, accuracy, and tolerance; 7.5 Accuracy and speed: smooth density profile; 7.6 Accuracy and speed: a supernova shock front; 7.7 Accuracy and speed: the Sun; 8 Comparison to other codes; 8.1 Relation to other oscillation codes; 8.2 Speed-accuracy comparison; 8.3 When to use Mag$nu$s; 9 Conclusions; Declaration of generative AI and AI-assisted technologies in the manuscript preparation process; Acknowledgments; References; Appendix A Magnus expansion at any order; A.1 Computing Magnus terms; A.2 Why the collocation orders are even; Appendix B Predefined oscillation-parameter sets; Appendix C The tutorial notebooks; Appendix D The test suite; Instructions for reporting errors; Complete, revision-matched arXiv TeX source package
Interpreting neutrino oscillation measurements requires computing the probability that a neutrino born with one flavor is detected with another. In vacuum and in matter of constant density, this probability has exact formulas. Where the density varies along the path, as in the Earth, the Sun, or a supernova, it does not. There, the usual approaches either replace the profile by constant-density steps, whose error falls only as the square of their width, or integrate the evolution equation with a general-purpose solver, which must shorten its steps to meet a tighter tolerance. Both become slow at high accuracy, yet an analysis needs the probability at many energies, directions, and parameter values. Each new experiment or model often requires writing and validating a new solver. We present Magnus, an open-source Python code that avoids these limitations. It uses the Magnus expansion, whose error falls as a high power of its step width and whose cost is set by how fast the density varies, not by how many oscillations the path holds. Magnus accepts any density profile, any number of flavors, and any Hermitian Hamiltonian, such as those of non-standard interactions, sterile neutrinos, Lorentz-invariance violation, and pseudo-Dirac neutrinos. Where a cheaper route exists, it takes it automatically. For example, where the density varies slowly, as in the Sun, it follows the instantaneous eigenstates and applies the expansion only at resonances. Besides oscillating probabilities, Magnus returns phase-averaged ones, which describe solar and astrophysical neutrinos. A typical probability takes a few milliseconds, and batched scans are one to two orders of magnitude cheaper per point. Among the public codes we compared, Magnus is the fastest at high accuracy on a varying profile. Magnus shifts the effort of studying a new experiment or model from writing a solver to specifying its physics.
这里保留 arXiv 提供的原始摘要；它是来源材料，不表示独立验证。
This is the original arXiv abstract used as source material; it does not represent independent verification.
打开 arXiv:2610.07159v1 原文页面Open arXiv:2610.07159v1 source page 在 NASA ADS 查看NASA ADS ↗
Click the button above to view full quantitative analysis and verbatim evidence in dialog.