1
  2
  3
  4
  5
  6
  7
  8
  9
 10
 11
 12
 13
 14
 15
 16
 17
 18
 19
 20
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
use config;
use config::{CentralDifference, Config, Index3, InitialCondition};
use errors::*;
use indicatif::{ProgressBar, ProgressStyle};
use input;
use ndarray::{Array3, ArrayView3, ArrayViewMut3, Zip};
use ndarray_parallel::prelude::*;
use noisy_float::prelude::*;
use output;
use potential;
use potential::Potentials;
use slog::Logger;
use std::f64::MAX;

#[derive(Debug)]
/// Holds all computed observables for the current wavefunction.
pub struct Observables {
    /// Normalised total energy.
    pub energy: R64,
    /// A squared normalisation. Square root occurs later when needed to apply a complete
    /// normalisation condition. This needs to be separate as we include other adjustments
    /// from time to time.
    pub norm2: R64,
    /// The value of the potential at infinity. This is used to calculate the binding energy.
    pub v_infinity: R64,
    /// Coefficient of determination
    pub r2: R64,
}

/// Runs the calculation and holds long term (system time) wavefunction storage
pub fn run(config: &Config, log: &Logger, debug_level: usize) -> Result<()> {
    let potentials = potential::load_arrays(config, log)?;

    let mut w_store: Vec<Array3<R64>> = Vec::new();
    if config.wavenum > 0 {
        //We require wavefunctions from disk, even if initial condition is not `FromFile`
        //The wavenum = 0 case is handled later
        input::load_wavefunctions(config, log, &mut w_store)?;
    }

    info!(log, "Starting calculation");

    for wnum in config.wavenum..config.wavemax + 1 {
        solve(config, log, debug_level, &potentials, wnum, &mut w_store)?;
    }
    Ok(())
}

/// Runs the actual computation once system is setup and ready.
fn solve(
    config: &Config,
    log: &Logger,
    debug_level: usize,
    potentials: &Potentials,
    wnum: u8,
    w_store: &mut Vec<Array3<R64>>,
) -> Result<()> {
    // Initial conditions from config file if ground state,
    // but start from previously converged wfn if we're an excited state.
    let mut phi: Array3<R64> = if wnum > 0 {
        let num = &config.grid.size;
        let bb = config.central_difference.bb();
        let init_size: [usize; 3] = [
            num.x as usize + bb,
            num.y as usize + bb,
            num.z as usize + bb,
        ];
        // Try to load current wavefunction from disk.
        // If not, we start with the previously converged function
        if let Ok(wfn) = input::wavefunction(wnum, init_size, bb, &config.output.file_type, log) {
            info!(log, "Loaded (current) wavefunction {} from disk", wnum);
            // If people are lazy or forget, their input files may contaminate the run here.
            // For example: starting wfn = 0 with random gaussian, max = 3.
            // User has wavefunction_{0,1}.csv in `input` from old run.
            // System will generate the random ground state and calculate it fine,
            // then try to load the old excited state 1 from disk, which is most likely
            // something completely irrelevant. Throw a warning for this scenario.
            if config.init_condition != InitialCondition::FromFile && wnum > config.wavenum {
                warn!(
                    log,
                    "Loaded a higher order wavefunction from disk although Initial conditions are set to '{}'.",
                    config.init_condition
                );
            }
            wfn
        } else {
            info!(
                log,
                "Loaded wavefunction {} from memory as initial condition",
                wnum - 1
            );
            // If we fail to load due to a error or missing, we just start from the previous
            // stored wavefunction.
            // We have to clone here, otherwise we cannot borrow w_store for the GS routines.
            w_store[wnum as usize - 1].clone()
        }
    } else {
        //This sorts out loading from disk if we are on wavefunction 0.
        config::set_initial_conditions(config, log).chain_err(|| ErrorKind::SetInitialConditions)?
    };

    output::print_observable_header(wnum);

    let prog_bar = ProgressBar::new(100);
    if debug_level == 3 {
        //If debug_level is 4 or 5 we have info on the screen. Remove the progress bar
        let term_width = *output::TERMWIDTH;
        let bar_width = (term_width - 24).to_string();
        let mut bar_template = String::new();
        bar_template.push_str("{msg}\n\n[{elapsed_precise}] |{bar:");
        bar_template.push_str(&bar_width);
        bar_template.push_str(".cyan/blue}| {spinner:.green} ETA: {eta:>3}");
        prog_bar.set_style(
            ProgressStyle::default_bar()
                .template(&bar_template)
                .progress_chars("█▓░")
                .tick_chars("⣾⣽⣻⢿⡿⣟⣯⣷ "),
        );
        prog_bar.set_position(0);
    }

    let mut step = 0;
    let mut converged = false;
    let mut last_energy = r64(MAX); //std::f64::MAX
    let mut diff_old = MAX;
    loop {
        let observables = compute_observables(config, potentials, &phi);
        let norm_energy = observables.energy / observables.norm2;
        let tau = r64(step as f64) * config.grid.dt;
        normalise_wavefunction(&mut phi, observables.norm2);

        // Orthoganalise wavefunction
        if wnum > 0 {
            orthogonalise_wavefunction(wnum, &mut phi, w_store);
        }
        // Save partial if requested
        if config.output.snap_update.is_some() && step % config.output.snap_update.unwrap() == 0 {
            config::symmetrise_wavefunction(config, &mut phi);
            normalise_wavefunction(&mut phi, observables.norm2);
            let ext = config.central_difference.ext();
            let work = get_work_area(&phi, ext);
            info!(
                log,
                "Saving partially converged wavefunction {} to disk.", wnum
            );
            if let Err(err) = output::wavefunction(
                &work,
                wnum,
                false,
                &config.project_name,
                &config.output.file_type,
            ) {
                warn!(
                    log,
                    "Could not output partial wavefunction per snap_update request: {}", err
                );
            }
        }

        // Check convergence state and break loop if succesful
        let diff = (norm_energy - last_energy).abs();
        if diff < config.tolerance {
            if debug_level == 3 {
                prog_bar.finish_and_clear();
            }
            println!("{}", output::print_measurements(tau, diff, &observables));
            output::finalise_measurement(
                &observables,
                wnum,
                r64(config.grid.size.x as f64),
                &config.project_name,
                &config.output.file_type,
            )?;
            if config.output.snap_update.is_some() {
                info!(
                    log,
                    "Removing partially converged wavefunction {} from disk.", wnum
                );
                if let Err(err) =
                    output::remove_partial(wnum, &config.project_name, &config.output.file_type)
                {
                    warn!(
                        log,
                        "The temporary wavefunction_{}_partial{} file could not be removed from the output directory: {}",
                        wnum,
                        config.output.file_type.extentsion(),
                        err
                    );
                }
            }
            converged = true;
            break;
        } else {
            last_energy = norm_energy;
        }

        // Output status to screen
        if debug_level == 3 {
            if let Some(estimate) = eta(step, diff_old, diff.raw(), config) {
                let percent = (100. - (estimate
                    / ((step as f64 / config.output.screen_update as f64) + estimate)
                    * 100.))
                    .floor();
                if percent.is_finite() {
                    prog_bar.set_position(percent as u64);
                }
            }
            prog_bar.set_message(&output::print_measurements(tau, diff, &observables));
        }
        // Make sure we don't evolve too far if we are not allowed
        if config.max_steps.is_some() && step > config.max_steps.unwrap() {
            break;
        }

        // Evolve solution until next screen update
        evolve(wnum, config, potentials, &mut phi, w_store);

        // Ready next iteration
        diff_old = diff.raw();
        step += config.output.screen_update;
    }

    if config.output.save_wavefns {
        //NOTE: This wil save regardless of whether it is converged or not, so we
        //flag it if that's the case.
        info!(log, "Saving wavefunction {} to disk", wnum);
        let work = get_work_area(&phi, config.central_difference.ext());
        if let Err(err) = output::wavefunction(
            &work,
            wnum,
            converged,
            &config.project_name,
            &config.output.file_type,
        ) {
            warn!(log, "Could not write wavefunction to disk: {}", err);
        }
    }

    if converged {
        info!(log, "Caluculation Converged");
        w_store.push(phi); //Save state
        Ok(())
    } else {
        Err(ErrorKind::MaxStep.into())
    }
}

/// Estimates completion time for the convergence of the current wavefunction.
///
/// # Returns
///
/// An estimate of the number of `screen_update` cycles to go until convergence.
/// Uses an option as it may not be finite.
fn eta(step: u64, diff_old: f64, diff_new: f64, config: &Config) -> Option<f64> {
    //Convergenge is done in exponential time after a short stabilisation stage.
    //So we can use the point slope form of a linear equation to find an estimate
    //to hit the tolerance on a semilogy scale. y - y1 = m(x-x1); where here we
    //use (x1,y1) as (step,diff_new), y is tolerance, find m then solve for x
    //
    //We don't use noisy floats here since it's already handled and fails gracefully
    //when out of bounds.
    let x1 = step as f64;
    let y1 = diff_new.log10();
    let rise = y1 - diff_old.log10();
    let run = config.output.screen_update as f64;
    let m = rise / run;

    //Step at which we estimate reaching tolerance.
    let x = ((config.tolerance.log10().raw() - y1) / m) + x1;

    //Now to return an expectation
    // Initially, we obtain a -inf which needs to be treated, and we can't correctly estimate the runtime until
    // the unstable region has been crossed. Luckily, we can use a previous estimate to identify this region.
    // We'll handle the second issue outside though and just return the estimate here.
    if x.is_finite() {
        let estimate = ((x - x1) / run).floor();
        //This catch stops our percentage from going above 100% and making indicatif throw a memory error.
        if estimate > 0. {
            return Some(estimate);
        }
    }
    None
}

/// Computes observable values of the system, for example the energy
///
/// # Arguments
///
/// * `config` - Reference to the configuration struct.
/// * `potentials` - Reference to the Potentials struct.
/// * `phi` - Current, active wavefunction array.
///
/// # Returns
///
/// A struct containing the energy and normalisation condition of the system,
/// as well as the potential energy value at infinity (used for the binding energy calulation)
/// and the r² expectation value.
///
/// # Remarks
///
/// Previously each of the variables were calculated in their own function.
/// The current implementation seems to be much faster though...
fn compute_observables(config: &Config, potentials: &Potentials, phi: &Array3<R64>) -> Observables {
    let ext = config.central_difference.ext();
    let phi_work = get_work_area(phi, ext);
    let mut work = Array3::<R64>::zeros(phi_work.dim());

    let energy = {
        let v = get_work_area(&potentials.v, ext);

        //TODO: We don't have any complex conjugation here.
        match config.central_difference {
            CentralDifference::ThreePoint => {
                let denominator = r64(2.) * config.grid.dn * config.grid.dn * config.mass;
                Zip::indexed(&mut work).and(v).and(phi_work).par_apply(
                    |(i, j, k), work, &v, &w| {
                        // Offset indexes as we are already in a slice
                        let lx = i as isize + 1;
                        let ly = j as isize + 1;
                        let lz = k as isize + 1;
                        let o = 1;
                        // get a slice which gives us our matrix of central difference points
                        let l = phi.slice(s![lx - 1..lx + 2, ly - 1..ly + 2, lz - 1..lz + 2]);
                        // l can now be indexed with local offset `o` and modifiers
                        *work = v * w * w
                            - w * (l[[o + 1, o, o]]
                                + l[[o - 1, o, o]]
                                + l[[o, o + 1, o]]
                                + l[[o, o - 1, o]]
                                + l[[o, o, o + 1]]
                                + l[[o, o, o - 1]] - r64(6.) * w)
                                / denominator;
                    },
                );
            }
            CentralDifference::FivePoint => {
                let denominator = r64(24.) * config.grid.dn * config.grid.dn * config.mass;
                Zip::indexed(&mut work).and(v).and(phi_work).par_apply(
                    |(i, j, k), work, &v, &w| {
                        // Offset indexes as we are already in a slice
                        let lx = i as isize + 2;
                        let ly = j as isize + 2;
                        let lz = k as isize + 2;
                        let o = 2;
                        let _16 = r64(16.);
                        // get a slice which gives us our matrix of central difference points
                        let l = phi.slice(s![lx - 2..lx + 3, ly - 2..ly + 3, lz - 2..lz + 3]);
                        // l can now be indexed with local offset `o` and modifiers
                        *work = v * w * w
                            - w * (-l[[o + 2, o, o]]
                                + _16 * l[[o + 1, o, o]]
                                + _16 * l[[o - 1, o, o]]
                                - l[[o - 2, o, o]]
                                - l[[o, o + 2, o]]
                                + _16 * l[[o, o + 1, o]]
                                + _16 * l[[o, o - 1, o]]
                                - l[[o, o - 2, o]]
                                - l[[o, o, o + 2]]
                                + _16 * l[[o, o, o + 1]]
                                + _16 * l[[o, o, o - 1]]
                                - l[[o, o, o - 2]]
                                - r64(90.) * w) / denominator;
                    },
                );
            }
            CentralDifference::SevenPoint => {
                let denominator = r64(360.) * config.grid.dn * config.grid.dn * config.mass;
                Zip::indexed(&mut work).and(v).and(phi_work).par_apply(
                    |(i, j, k), work, &v, &w| {
                        // Offset indexes as we are already in a slice
                        let lx = i as isize + 3;
                        let ly = j as isize + 3;
                        let lz = k as isize + 3;
                        let o = 3;
                        let _2 = r64(2.);
                        let _27 = r64(27.);
                        let _270 = r64(270.);
                        // get a slice which gives us our matrix of central difference points
                        let l = phi.slice(s![lx - 3..lx + 4, ly - 3..ly + 4, lz - 3..lz + 4]);
                        // l can now be indexed with local offset `o` and modifiers
                        *work = v * w * w
                            - w * (_2 * l[[o + 3, o, o]] - _27 * l[[o + 2, o, o]]
                                + _270 * l[[o + 1, o, o]]
                                + _270 * l[[o - 1, o, o]]
                                - _27 * l[[o - 2, o, o]]
                                + _2 * l[[o - 3, o, o]]
                                + _2 * l[[o, o + 3, o]]
                                - _27 * l[[o, o + 2, o]]
                                + _270 * l[[o, o + 1, o]]
                                + _270 * l[[o, o - 1, o]]
                                - _27 * l[[o, o - 2, o]]
                                + _2 * l[[o, o - 3, o]]
                                + _2 * l[[o, o, o + 3]]
                                - _27 * l[[o, o, o + 2]]
                                + _270 * l[[o, o, o + 1]]
                                + _270 * l[[o, o, o - 1]]
                                - _27 * l[[o, o, o - 2]]
                                + _2 * l[[o, o, o - 3]]
                                - r64(1470.) * w) / denominator;
                    },
                );
            }
        }
        // Sum result for total energy.
        r64(work.into_par_iter().map(|i| i.raw()).sum())
    };
    let norm2 = phi_work.into_par_iter().map(|&el| el * el).sum();
    let v_infinity = {
        match potentials.pot_sub {
            (Some(ref potsub), None) => {
                Zip::from(&mut work)
                    .and(phi_work)
                    .and(potsub.view())
                    .par_apply(|work, &w, &potsub| {
                        *work = w * w * potsub;
                    });
                work.into_par_iter().map(|i| i.raw()).sum()
            }
            (None, Some(potsub)) => {
                Zip::from(&mut work).and(phi_work).par_apply(|work, &w| {
                    *work = w * w * potsub;
                });
                work.into_par_iter().map(|i| i.raw()).sum()
            }
            _ => 0.,
        }
    };
    let r2 = {
        Zip::indexed(&mut work)
            .and(phi_work)
            .par_apply(|(i, j, k), work, &w| {
                let idx = Index3 { x: i, y: j, z: k };
                let r2 = potential::calculate_r2(&idx, &config.grid);
                *work = w * w * r2;
            });
        work.into_par_iter().map(|i| i.raw()).sum()
    };

    Observables {
        energy: energy,
        norm2: norm2,
        v_infinity: r64(v_infinity),
        r2: r64(r2),
    }
}

/// Calculate the normalisation condition of a wavefunction.
/// The square root portion of this calculation happens later as we sometimes require
/// just this condition.
///
/// # Arguments
///
/// * `w` - Current wavefunction array.
fn get_norm_squared(w: &ArrayView3<R64>) -> R64 {
    //NOTE: No complex conjugation due to all real input for now
    w.into_par_iter().map(|&el| el * el).sum()
}

/// Normalisation of the wavefunction
///
/// # Arguments
///
/// * `w` - Wavefunction to normalise.
/// * `norm2` - The squared normalisation observable.
fn normalise_wavefunction(w: &mut Array3<R64>, norm2: R64) {
    let norm = norm2.sqrt();
    w.par_map_inplace(|el| *el /= norm);
}

/// Uses Gram Schmidt orthogonalisation to identify the next excited state's wavefunction, even if it's degenerate
///
/// # Arguments
///
/// * `wnum` - Current exited state value.
/// * `w` - Current, active wavefunction array.
/// * `w_store` - Vector of currently converged wavefunctions.
fn orthogonalise_wavefunction(wnum: u8, w: &mut Array3<R64>, w_store: &[Array3<R64>]) {
    for lower in w_store.iter().take(wnum as usize) {
        // This MUST be created inside the loop or else we throw nans.
        // I've tried a number of ways to treat this method as it's
        // pretty expensive. Here is the best one I've identified.
        let mut overlap = Array3::<R64>::zeros(w.dim());
        Zip::from(&mut overlap)
            .and(lower)
            .and(w.view())
            .par_apply(|overlap, &lower, &w| *overlap = lower * w);
        let overlap_sum: f64 = overlap.into_par_iter().map(|i| i.raw()).sum();
        Zip::from(w.view_mut())
            .and(lower)
            .par_apply(|w, &lower| *w -= lower * overlap_sum);
    }
}

/// Shortcut to getting a slice of the workable area of the current array.
/// In other words, the finite element only cells are removed
///
/// # Arguments
///
/// * `arr` - A reference to the array which requires slicing.
/// * `ext` - Extent of central difference limits. From `config.central_difference.ext()`.
///
/// # Returns
///
/// An array view containing only the workable area of the array.
pub fn get_work_area(arr: &Array3<R64>, ext: usize) -> ArrayView3<R64> {
    let dims = arr.dim();
    let exti = ext as isize;
    arr.slice(s![
        exti..dims.0 as isize - exti,
        exti..dims.1 as isize - exti,
        exti..dims.2 as isize - exti
    ])
}

/// Shortcut to getting a mutable slice of the workable area of the current array.
/// In other words, the finite element only cells are removed
///
/// # Arguments
///
/// * `arr` - A mutable reference to the array which requires slicing.
/// * `ext` - Extent of central difference limits. From `config.central_difference.ext()`.
///
/// # Returns
///
/// A mutable arrav view containing only the workable area of the array.
pub fn get_mut_work_area(arr: &mut Array3<R64>, ext: usize) -> ArrayViewMut3<R64> {
    let dims = arr.dim();
    let exti = ext as isize;
    arr.slice_mut(s![
        exti..dims.0 as isize - exti,
        exti..dims.1 as isize - exti,
        exti..dims.2 as isize - exti
    ])
}

/// Evolves the solution a number of `steps`
///
/// # Arguments
///
/// * `wnum` - Current exited state value.
/// * `config` - Reference to the configuration struct.
/// * `phi` - Current, active wavefunction array.
/// * `w_store` - Vector of currently converged wavefunctions.
fn evolve(
    wnum: u8,
    config: &Config,
    potentials: &Potentials,
    phi: &mut Array3<R64>,
    w_store: &[Array3<R64>],
) {
    //without mpi, this is just update interior (which is really updaterule if we dont need W)
    let bb = config.central_difference.bb();
    let ext = config.central_difference.ext();
    let mut work_dims = phi.dim();
    work_dims.0 -= bb;
    work_dims.1 -= bb;
    work_dims.2 -= bb;
    let pa = get_work_area(&potentials.a, ext);
    let pb = get_work_area(&potentials.b, ext);
    let mut work = Array3::<R64>::zeros(work_dims);
    let mut steps = 0;
    loop {
        {
            let w = get_work_area(phi, ext);

            //TODO: We don't have any complex conjugation here.
            match config.central_difference {
                CentralDifference::ThreePoint => {
                    let denominator = r64(2.) * config.grid.dn * config.grid.dn * config.mass;
                    Zip::indexed(&mut work).and(pa).and(pb).and(w).par_apply(
                        |(i, j, k), work, &pa, &pb, &w| {
                            // Offset indexes as we are already in a slice
                            let lx = i as isize + 1;
                            let ly = j as isize + 1;
                            let lz = k as isize + 1;
                            let o = 1;
                            // get a slice which gives us our matrix of central difference points
                            let l = phi.slice(s![lx - 1..lx + 2, ly - 1..ly + 2, lz - 1..lz + 2]);
                            // l can now be indexed with local offset `o` and modifiers
                            *work = w * pa
                                + pb * config.grid.dt
                                    * (l[[o + 1, o, o]]
                                        + l[[o - 1, o, o]]
                                        + l[[o, o + 1, o]]
                                        + l[[o, o - 1, o]]
                                        + l[[o, o, o + 1]]
                                        + l[[o, o, o - 1]]
                                        - r64(6.) * w)
                                    / denominator;
                        },
                    );
                }
                CentralDifference::FivePoint => {
                    let denominator = r64(24.) * config.grid.dn * config.grid.dn * config.mass;
                    Zip::indexed(&mut work).and(pa).and(pb).and(w).par_apply(
                        |(i, j, k), work, &pa, &pb, &w| {
                            // Offset indexes as we are already in a slice
                            let lx = i as isize + 2;
                            let ly = j as isize + 2;
                            let lz = k as isize + 2;
                            let o = 2;
                            let _16 = r64(16.);
                            // get a slice which gives us our matrix of central difference points
                            let l = phi.slice(s![lx - 2..lx + 3, ly - 2..ly + 3, lz - 2..lz + 3]);
                            // l can now be indexed with local offset `o` and modifiers
                            *work = w * pa
                                + pb * config.grid.dt
                                    * (-l[[o + 2, o, o]]
                                        + _16 * l[[o + 1, o, o]]
                                        + _16 * l[[o - 1, o, o]]
                                        - l[[o - 2, o, o]]
                                        - l[[o, o + 2, o]]
                                        + _16 * l[[o, o + 1, o]]
                                        + _16 * l[[o, o - 1, o]]
                                        - l[[o, o - 2, o]]
                                        - l[[o, o, o + 2]]
                                        + _16 * l[[o, o, o + 1]]
                                        + _16 * l[[o, o, o - 1]]
                                        - l[[o, o, o - 2]]
                                        - r64(90.) * w)
                                    / denominator;
                        },
                    );
                }
                CentralDifference::SevenPoint => {
                    let denominator = r64(360.) * config.grid.dn * config.grid.dn * config.mass;
                    Zip::indexed(&mut work).and(pa).and(pb).and(w).par_apply(
                        |(i, j, k), work, &pa, &pb, &w| {
                            // Offset indexes as we are already in a slice
                            let lx = i as isize + 3;
                            let ly = j as isize + 3;
                            let lz = k as isize + 3;
                            let o = 3;
                            let _2 = r64(2.);
                            let _27 = r64(27.);
                            let _270 = r64(270.);
                            // get a slice which gives us our matrix of central difference points
                            let l = phi.slice(s![lx - 3..lx + 4, ly - 3..ly + 4, lz - 3..lz + 4]);
                            // l can now be indexed with local offset `o` and modifiers
                            *work = w * pa
                                + pb * config.grid.dt
                                    * (_2 * l[[o + 3, o, o]] - _27 * l[[o + 2, o, o]]
                                        + _270 * l[[o + 1, o, o]]
                                        + _270 * l[[o - 1, o, o]]
                                        - _27 * l[[o - 2, o, o]]
                                        + _2 * l[[o - 3, o, o]]
                                        + _2 * l[[o, o + 3, o]]
                                        - _27 * l[[o, o + 2, o]]
                                        + _270 * l[[o, o + 1, o]]
                                        + _270 * l[[o, o - 1, o]]
                                        - _27 * l[[o, o - 2, o]]
                                        + _2 * l[[o, o - 3, o]]
                                        + _2 * l[[o, o, o + 3]]
                                        - _27 * l[[o, o, o + 2]]
                                        + _270 * l[[o, o, o + 1]]
                                        + _270 * l[[o, o, o - 1]]
                                        - _27 * l[[o, o, o - 2]]
                                        + _2 * l[[o, o, o - 3]]
                                        - r64(1470.) * w)
                                    / denominator;
                        },
                    );
                }
            }
        }
        {
            let mut w_fill = get_mut_work_area(phi, ext);
            Zip::from(&mut w_fill)
                .and(&work)
                .par_apply(|w_fill, &work| {
                    *w_fill = work;
                });
        }
        if wnum > 0 {
            let norm2 = {
                let phi_work = get_work_area(phi, ext);
                get_norm_squared(&phi_work)
            };
            normalise_wavefunction(phi, norm2);
            orthogonalise_wavefunction(wnum, phi, w_store);
        }
        steps += 1;
        if steps >= config.output.screen_update {
            break;
        }
    }
}

#[cfg(test)]
mod tests {
    use super::*;

    macro_rules! assert_approx_eq {
        ($a:expr, $b:expr) => {{
            let eps = 1.0e-6;
            let (a, b) = (&$a, &$b);
            assert!(
                (*a - *b).abs() < eps,
                "assertion failed: `(left !== right)` \
                 (left: `{:?}`, right: `{:?}`, expect diff: `{:?}`, real diff: `{:?}`)",
                *a,
                *b,
                eps,
                (*a - *b).abs()
            );
        }};
        ($a:expr, $b:expr, $eps:expr) => {{
            let (a, b) = (&$a, &$b);
            assert!(
                (*a - *b).abs() < $eps,
                "assertion failed: `(left !== right)` \
                 (left: `{:?}`, right: `{:?}`, expect diff: `{:?}`, real diff: `{:?}`)",
                *a,
                *b,
                $eps,
                (*a - *b).abs()
            );
        }};
    }

    #[test]
    fn gram_schmidt() {
        let ground = Array3::<R64>::from_shape_fn((2, 2, 2), |(i, j, k)| r64((i + j + k) as f64));
        let w_store: Vec<Array3<R64>> = vec![ground];

        let mut test = Array3::<R64>::from_shape_fn((2, 2, 2), |(i, j, k)| {
            let (fi, fj, fk) = (i as f64, j as f64, k as f64);
            r64(-fi - fj - fk)
        });
        orthogonalise_wavefunction(1, &mut test, &w_store);

        let compare = Array3::<R64>::from_shape_vec(
            (2, 2, 2),
            vec![
                r64(0.),
                r64(23.),
                r64(23.),
                r64(46.),
                r64(23.),
                r64(46.),
                r64(46.),
                r64(69.),
            ],
        ).unwrap();
        assert!(compare.all_close(&test, r64(0.01)));
    }

    #[test]
    fn work_area() {
        let test = Array3::<R64>::zeros((5, 8, 7));
        let work = get_work_area(&test, 1);
        let dims = work.dim();
        assert_eq!(dims.0, 3);
        assert_eq!(dims.1, 6);
        assert_eq!(dims.2, 5);
    }

    #[test]
    fn mut_work_area() {
        let mut test = Array3::<R64>::zeros((5, 8, 7));
        let dims = {
            let mut work = get_mut_work_area(&mut test, 1);
            work.fill(r64(1.));
            work.dim()
        };

        let compare = Array3::<R64>::from_shape_fn((5, 8, 7), |(i, j, k)| {
            if (i == 0 || i == 4) || (j == 0 || j == 7) || (k == 0 || k == 6) {
                r64(0.)
            } else {
                r64(1.)
            }
        });
        assert_eq!(dims.0, 3);
        assert_eq!(dims.1, 6);
        assert_eq!(dims.2, 5);
        assert!(compare.all_close(&test, r64(0.01)));
    }

    #[test]
    fn norm2() {
        let test = Array3::<R64>::from_shape_fn((5, 8, 7), |(i, j, k)| r64((i * j * k) as f64));
        let work = get_work_area(&test, 1);
        let result = get_norm_squared(&work);
        assert_approx_eq!(result, r64(70070.));
    }

    #[test]
    fn wfn_normalise() {
        let normalised = Array3::<R64>::from_shape_fn((3, 2, 5), |(i, j, k)| {
            let norm = r64(1.1091);
            r64((i * j * k) as f64) / norm
        });

        let mut test = Array3::<R64>::from_shape_fn((3, 2, 5), |(i, j, k)| r64((i * j * k) as f64));
        normalise_wavefunction(&mut test, r64(1.23));

        assert!(test.all_close(&normalised, r64(0.01)));
    }
}